Microfluidic Modeling of Macrophage‐Induced Cardiac Inflammation Using NF‐κB Reporter Cardiomyocytes

ABSTRACT

Inflammation is a key driver of cardiac disease progression, including heart failure and myocardial infarction, with macrophage‐cardiomyocyte interactions playing a central role in shaping the local inflammatory microenvironment. Prolonged exposure to pro‐inflammatory cytokines, many produced by macrophages, induces cardiomyocyte stress, dysfunction, and pathological tissue remodeling. However, the spatial and temporal dynamics of this crosstalk remain poorly understood, limiting the development of targeted therapies. To address this gap, we developed an in vitro cardiac inflammation model that enables controlled investigation of macrophage‐cardiomyocyte interactions. Cardiomyocytes were transduced with an NF‐κB‐responsive fluorescent reporter to allow real‐time monitoring of inflammatory signaling. MATLAB‐based simulations were used to optimize medium delivery and cell densities by modeling TNFα diffusion. Using a multiplex microfluidic co‐culture chip, we identified the TNFα threshold of 50 ng mL^{− 1} with a 6‐h exposure as sufficient to activate cardiomyocyte NF‐κB signaling. Both macrophage‐conditioned medium and direct co‐culture significantly increased NF‐κB activation, indicating enhanced cardiomyocyte inflammatory stress. Dexamethasone markedly suppressed NF‐κB activation when administered at onset of co‐culture to both cell types, whereas IL‐10 showed limited efficacy. Overall, this cardiac inflammation model provides a versatile platform for dissecting immune‐cardiac crosstalk and supports future studies of multicellular crosstalk, spatiotemporal signaling dynamics, and therapeutic screening.

A cardiac inflammation model was developed to enable real‐time analysis of macrophage‐cardiomyocyte interactions using NF‐κB‐reporter cardiomyocytes and a microfluidic co‐culture system. A MATLAB‐based model was implemented to optimize cell density and medium resupply based on TNFα diffusion kinetics. Activation thresholds and stress responses in NF‐κB‐reporter cardiomyocytes were identified, supporting the investigation of inflammatory signaling dynamics and therapeutic screening.

1. Introduction

The immune system plays a pivotal role in the onset, progression, and resolution of cardiovascular diseases. Following cardiac injury, immune cells rapidly migrate to the myocardium, where they help clear necrotic tissue and initiate repair. However, when this inflammatory response becomes prolonged or dysregulated, it can damage healthy tissue, exacerbate myocardial injury, and impair cardiac recovery [1]. Activated leukocytes release reactive oxygen species and proteolytic enzymes that can inadvertently harm viable cardiac cells [2], and elevated levels of pro‐inflammatory cytokines such as TNFα further intensify this damage [3]. These observations highlight the need to evaluate how immune cells shape the local inflammatory milieu and influence cardiomyocyte behavior, as this interaction fundamentally affects both the severity of myocardial injury and the trajectory of cardiovascular disease.

Crosstalk between macrophages and cardiomyocytes is a key driver of adverse cardiac inflammation. After cardiac injury or stress, activated macrophages secrete pro‐inflammatory cytokines that amplify inflammatory signaling in nearby cardiomyocytes. Otherwise, healthy cardiomyocytes near the injury site can become stressed, releasing damage‐associated molecular patterns (DAMPs) and additional cytokines [4, 5, 6]. This heightened cytokine activity establishes a feed‐forward inflammatory loop that intensifies local inflammation, disrupts tissue repair, and contributes to progressive myocardial damage and functional decline [7]. Understanding these intercellular interactions underscores the importance of tailoring therapeutic strategies to the disease context. Determining whether to target immune and cardiac cells concurrently, rather than focusing on a single dominant inflammatory single cell type, requires experimental models capable of dissecting this pathological feedback cycle.

A central regulator of this signaling network is nuclear factor kappa B (NF‐κB), a transcription factor that regulates inflammatory responses in both immune and cardiac cells [8]. NF‐κB controls the expression of cytokines, chemokines, and adhesion molecules that govern immune cell recruitment, activation, and survival [9]. In cardiomyocytes, chronic NF‐κB activation drives pathological stress signaling, impairs contractile function, and promotes cell death [10, 11, 12]. Its broad influence makes NF‐κB a valuable indicator of inflammatory status and a key target for evaluating anti‐inflammatory interventions. To explore strategies for modulating this pathway, we examined the anti‐inflammatory potential of IL‐10 [13, 14], an endogenous immunoregulatory cytokine, and dexamethasone [15, 16, 17, 18], a synthetic glucocorticoid, in suppressing NF‐κB signaling during cardiac inflammation. Although both agents have been studied in isolated cell types, their effects in co‐culture systems where macrophages and cardiomyocytes experience inflammation simultaneously remain insufficiently characterized [19].

Traditional cell culture system lacks the spatial control and temporal resolution needed to capture the dynamic nature of intercellular communication, leading to signal dilution and reduced physiological relevance. Microfluidic platforms address these limitations by enabling precise compartmentalization of cell types, controlled diffusion of soluble factors, and real‐time monitoring of cellular responses. Prior studies have used microfluidic co‐culture systems to investigate immune‐cardiac interactions, demonstrating improved physiological relevance [20, 21]. Some have specifically examined macrophage‐cardiomyocyte crosstalk in cardiac inflammation [22, 23], providing early insights but leaving the temporal dynamics of macrophage‐driven NF‐κB activation in cardiomyocytes insufficiently defined.

In this study, we employed a multiplex microfluidic co‐culture platform that supports controlled paracrine communication between macrophages and cardiomyocytes while minimizing signal dilution. Cardiomyocytes were transduced with an NF‐κB‐responsive fluorescent reporter to enable real‐time tracking of inflammatory signaling. We first developed a MATLAB‐based simulation to model macrophage‐derived TNFα production and its diffusion into the cardiomyocyte chamber, guiding optimization of medium delivery and cell‐seeding densities. Using this system, we performed dose‐response and pulse‐chase experiments to determine the TNFα concentration and exposure duration required to activate NF‐κB in cardiomyocytes. Both activated macrophage‐conditioned medium and direct co‐culture induced NF‐κB activation, demonstrating the sensitivity of cardiomyocytes to macrophage‐secreted inflammatory cues. To assess potential therapeutic interventions, we tested the anti‐inflammatory agents IL‐10 and dexamethasone. IL‐10 produced minimal reduction in NF‐κB activation in the presence of TNFα, whereas dexamethasone significantly suppressed NF‐κB activation when applied at the onset of co‐culture. Together, these findings illuminate the dynamic interplay between macrophages and cardiomyocytes and demonstrate the potential of this platform for evaluating anti‐inflammatory therapies in a co‐culture environment.

2. Results

2.1. Microfluidic Chip and Experiment Design for Cardiac Inflammation Co‐culture Study

During cardiac inflammation, macrophages release inflammatory cytokines that activate other immune cells, induce stress in nearby cardiomyocytes, and contribute to cardiomyocyte death. In this study, we used a multiplexed co‐culture chip to model paracrine interactions between cardiomyocytes and macrophages under inflammatory conditions. This previously described PDMS co‐culture chip contains 32 independently addressable chamber units (Figure 1A) [24]. Each unit includes one waste‐rinsing channel and four identical chambers that can function as either cell culture chambers or medium reservoirs (Figure 1B). Push‐down valves located at the inlets, outlets, and between the chambers regulate rising and chamber isolation. The co‐culture chip interfaces with a custom microfluidic controller and pressurized source, allowing user operation through a graphical interface [25]. Polarized M1 macrophages and naïve cardiomyocytes were seeded into adjacent chambers. After cell attachment, valves were opened to allow medium diffusion into the cell chambers, then closed between the medium and cell chambers while remaining open between the two cell chambers to enable paracrine communication (Figure 1C). Fresh medium flowed through the supply chambers in cycles of approximately 30–40 min. Periodic valve actuation between the two cell populations prevented cell migration throughout the 24‐h experiment.

FIGURE 1.

Microfluidic co‐culture chip and method schematics. (A) Stitched bright‐field image of the microfluidic co‐culture chip, demonstrating multiplexity with four food‐dye colors. The chip consists of 32 repeated units and has dimensions of approximately 60 mm × 20 mm. Scale bar: 5 mm. (B) Each unit contains four identical chambers and a waste rinse channel. In the co‐culture setup, chambers A and D serve as medium supply chambers, diffusing fresh medium into cell chambers B and C. Chamber connections are regulated by push‐down valves. Scale bar: 500 µm. (C) Schematic of macrophage‐cardiomyocyte co‐culture.

2.2. MATLAB Simulation of TNFα Secretion and Diffusion Dynamics

Pro‐inflammatory M1 macrophages secrete factors that worsen cardiovascular disease progression and induce inflammatory stress in nearby cardiomyocytes. Among these factors, tumor necrosis factor‐alpha (TNFα) plays a central role. Understanding the spatial and temporal distribution of TNFα is essential for determining its impact on recipient cells. To characterize these dynamics and ensure proper medium delivery to both cell populations, we developed a MATLAB model simulating TNFα secretion and diffusion from the macrophage chamber into the cardiomyocyte chamber. The model incorporates the actual chip chamber geometry, the TNFα diffusion coefficient (1.093 × 10⁻⁵ mm² s⁻¹) [26], the experimentally measured macrophage TNFα secretion rate per cell (3.7294 × 10⁻⁷ ng h⁻¹ cell⁻¹), the number of macrophages plated, macrophage growth rate, and the medium resupply frequency. It predicts TNFα concentration over time and space within the chamber unit, generating 2D maps (x‐y coordinates at specific time points) or 3D maps (concentration‐time‐y coordinates). These maps visualize TNFα gradients using color scales (Figure 2A,B). Concentration profiles along the y‐axis were plotted to track temporal changes (Figure 2C). As macrophages secreted TNFα, its concentration rose rapidly in the macrophage chamber and diffused into the adjacent cardiomyocyte chambers. Medium resupply every 30 min maintained nutrient availability and removed residual TNFα from the side chambers. TNFα concentrations were monitored at multiple positions within the chip (Figure 2D). Notably, TNFα levels in the cardiomyocyte chamber increased rapidly during co‐culture and were not substantially diluted by medium resupply.

FIGURE 2.

MATLAB simulation of TNFα secretion by macrophages (MΦ) and diffusion to the cardiomyocyte (CM) chamber. (A) Simulated spatiotemporal changes in TNFα concentration within a single microfluidic chip unit, visualized as a 2D heatmap over 24 h, with concentrations represented by a color scale. (B) A 3D surface plot illustrating TNFα concentration over time along the chip's y‐axis, with concentrations averaged across the x‐axis. (C) TNFα concentration profiles at specific locations within the chip, plotted across selected time points (indicated by shaded regions, which represent the locations of the chambers, with corresponding locations labeled below the graph). (D) TNFα concentration profiles over time, plotted for different locations along the y‐axis at the midpoints of all four chambers.

We also evaluated parameters influencing TNFα distribution, including macrophage seeding density (Figure S1A) and medium resupply frequency (Figure S1B). Macrophage density (20, 40, 80, 120, or 180 cells per chamber; 120 cells ∼50% confluence) strongly affected TNFα secretion rates and resulting concentrations in the cardiomyocyte chamber. In contrast, varying the medium resupply interval (10–120 min) had minimal effect on TNFα profile. Prior studies indicate that TNFα concentrations of 1–100 ng mL⁻¹ induce cardiomyocyte stress [27, 28]. Guided by these findings, we selected a macrophage density of 100–120 cells for subsequent experiments. A medium resupply frequency of 30–40 min provided adequate nutrients while rapidly achieving macrophage‐derived TNFα concentrations of 15–20 ng mL⁻¹ in the cardiomyocyte chamber. These conditions were used for all the following studies.

To experimentally validate the TNFα simulation model, we performed a microfluidic culture experiment using pro‐inflammatory M1a RAW 264.7 macrophages. Because real‐time tracking of TNFα diffusion was technically challenging and sample volume were limited, we adapted our approach to quantify total TNFα secretion across all chambers. Each of the four chambers in the chip was seeded with M1a macrophages, and conditioned medium was collected from the outlets every 30 min. Samples were pooled into four time intervals (0–11 h, 12–27 h, 28–49 h, and 50–73 h), and analyzed by ELISA (Figure S6D). The MATLAB model was configured to replicate this setup, using identical cell numbers, an estimated secretion rate from the initial time window, and the measured macrophage growth rate measured in the chip. The resulting 3D surface and 2D concentration maps showed a progressive increase in TNFα levels over time, despite the 30‐min medium exchange that reset the concentrations to zero (Figure S6E–G). The experimentally measured TNFα profiles exhibited trends consistent with the simulated predictions under equivalent conditions (Figure S6D–G). Both datasets showed a rapid rise in TNFα over the 73‐h culture period, indicating that the model captures the temporal dynamics of TNFα production. However, simulated concentrations were higher than the experimental measurements across all time points. Overall, these results demonstrate that the model captures TNFα dynamics in the microfluidic chip, supporting its use for co‐culture simulations.

2.3. Development and Characterization of NF‐κB Reporter H9c2 Cells

To monitor cardiac inflammation, we generated a stable H9c2 cardiomyoblast line expressing a fluorescent NF‐κB reporter (mKate2) via lentiviral transduction. The DNA construct, obtained from Addgene [29], included NF‐κB response elements, the mKate2 gene, and a nuclear localization signal (Figure 3A). In the absence of inflammatory stimuli, the reporter remains off. Upon TNFα stimulation, TNFα receptor activation leads to IκB degradation, releasing NF‐κB to translocate into the nucleus and bind response elements (Figure 3A) [30]. This triggers transcription of the mKate2 gene, resulting in nuclear‐localized fluorescence.

FIGURE 3.

Characterization of NF‐κB Reporter H9c2 Cell Response to TNFα Stimulation. (A) NF‐κB reporter H9c2 cells were generated using lentiviral transduction. In the absence of inflammatory stimuli, the NF‐κB reporter remains inactive. Upon TNFα or other inflammatory stimuli, NF‐κB translocates to the nucleus, binds to the promoter in the reporter construct, leading to the accumulation of mKate2 (red) fluorescent protein in the nucleus, which can be observed under a fluorescence microscope. (B) Representative overlay images of NF‐κB reporter H9c2 cells seeded in a microfluidic chip, either without stimulation (left) or with 50 ng mL‐1 TNFα (right). Blue: nuclei (DAPI), Yellow: mKate2 fluorescence, Gray: phase contrast. Scale bar: 100 µm. (C,D) Dose‐response analysis of TNFα in the chip (0.001–100 ng mL⁻¹, with 0 ng mL⁻¹ as control): (C) Percentage of mKate2‐positive cells and (D) fluorescence intensity per positive cell. Data represents three independent experiments with four technical replicates per condition. ^* p < 0.05, ^** p < 0.01. (E,F) Pulse‐chase TNFα experiments in the chip (100 ng mL⁻¹ applied for 10 min to 24 h, followed by TNFα removal and imaging at 24 h): Percentage of mKate2‐positive cells and fluorescence intensity per cell. Data represents three independent experiments with four technical replicates per condition. ^* p < 0.05. All data are presented as mean ± SD, and statistical analysis was performed using one‐way ANOVA with Dunnett's post hoc test.

Following puromycin selection, we tested reporter activation in response to TNFα. Stimulation with 100 ng mL⁻¹ TNFα for 10 h induced strong fluorescence, indicating NF‐κB activation and cellular stress (Video S1). To quantify activation, we measured both the percentage of positive cells and the average fluorescence intensity per positive cell, providing insights into the proportion of inflamed cells and the extent of inflammation. We then exposed reporter cells to a range of TNFα concentrations, from physiologically levels (0.001 ng mL⁻¹) [31] to 100 ng mL⁻¹. In conventional culture, number of activated cells increased significantly at 10 ng mL⁻¹ and above (Figure S2A,B), while fluorescence per positive cell remained stable (Figure S2C). In the microfluidic chip, however, a higher activation threshold of 50 ng mL⁻¹ was observed (Figure 3B,C), with fluorescence intensity increasing significantly at or above this level (Figure 3D). To ensure consistent activation, we selected 100 ng mL⁻¹ for subsequent experiments.

To determine the minimum TNFα exposure time required for NF‐κB activation and to identify potential intervention windows, we conducted a pulse‐chase experiment. Reporter cells were stimulated with 100 ng mL⁻¹ TNFα for durations ranging from 10 min to 24 h. After stimulation, TNFα was replaced with fresh medium, and fluorescence was tracked over a 24‐h period. This experiment was performed in both the microfluidic chip (Figure 3E,F) and conventional culture (Figure S3A–C), yielding similar results. The number of activated cells increased with longer TNFα exposure, with a significant rise observed after 6 h (Figure 3E). Fluorescence intensity per positive cell also increased gradually, reaching statistically significance at 24 h (Figure 3F).

2.4. RAW 264.7 Macrophage Polarization and M1‐conditioned Medium‐induced Inflammation and Cell Death

Macrophages are key mediators of the innate immune response, playing a central role in pathogen detection, phagocytosis, and elimination, as well as in coordinating inflammation and tissue repair. RAW 264.7 cells are a widely used macrophage model capable of polarizing into M1 (pro‐inflammatory) or M2 (anti‐inflammatory) phenotypes. While macrophages are broadly classified into these two categories, they can be further subdivided into specific subtypes. For example, M1 macrophages can differentiate into M1a (classically activated) or M1b (innately activated) subtypes. M1a polarization is induced by microbial lipopolysaccharide (LPS) combined with interferon‐gamma (IFNγ), whereas M1b polarization is triggered by LPS alone (Figure S4A). Although both M1a and M1b macrophages produce pro‐inflammatory mediators such as IL‐1β, IL‐6, and TNFα, only M1a cells exhibit enhanced phagocytic activity [32, 33]. In this study, we optimized M1 polarization protocols based on established methods from the literature [22, 34, 35].

RAW 264.7 macrophages were treated with LPS alone or in combination with varying concentrations of IFNγ (1–250 ng mL⁻¹) using high‐throughput microfluidic chips. M1 polarization efficiency was assessed by measuring the expression of M1 marker iNOS. Unstimulated (naïve) RAW 264.7 cells showed low iNOS expression (Figure S4B), and LPS alone also induced minimal expression (data not shown). When LPS was fixed at 200 ng mL⁻¹, increasing IFNγ concentrations led to a higher percentage of iNOS‐positive cells (Figure S4B,C). At IFNγ levels of 5 ng mL⁻¹ or more, iNOS‐positive cells reached 70%–80%. Additionally, IFNγ treatment caused a twofold increase in cell size (Figure S4D) and reduced proliferation (data not shown).

Due to the substantial reduction in cell numbers following LPS and IFNγ co‐treatment, we used M1b macrophages (LPS only) for conditioned medium experiments to ensure consistency with M0 (unstimulated) controls. TNFα levels in conditioned medium were quantified by ELISA and normalized to cell number. M1b macrophages secreted significantly more TNFα than M0 cells, while M1a macrophages produced approximately fivefold higher TNFα per cell than M0 controls (Figure S5A). M0 macrophages maintained a relatively high basal TNFα level, consistent with a previous report [36]. Based on these findings, M1a macrophages polarized with 200 ng mL⁻¹ LPS and 25 ng mL⁻¹ IFNγ were selected for further experiments due to their robust and consistent activation profile.

We next quantified TNFα secretion by M1a macrophages in microfluidic chips over four‐time intervals (0–11, 12–27, 28–49, and 50–73 h) and observed a gradual accumulation of TNFα (Figure S6A). Cell counts were taken at each time point to monitor proliferation (Figure S6B). When normalized to both time and cell number, TNFα secretion increased notably after 28 h (Figure S6C). Compared to dish cultures (Figure S5B), the TNFα secretion rate in the chip during the 12–27 h window was approximately 17 times higher.

ELISA analysis of H9c2 cardiomyocyte‐conditioned media showed no detectable TNFα under either unstimulated conditions or after 24 h of exposure to 100 ng mL⁻¹ TNFα (Figure S5C), suggesting that cardiomyocytes do not contribute to TNFα levels in our conditioned‐medium experiments. We also evaluated M2 polarization using various IL‐4 and IL‐10 combinations (Figure S7A,B). Expression of the M2 marker Arg1 was used to assess polarization efficiency. The combined IL‐4 and IL‐10 treatment yielded a maximum of 26% Arg1‐positive cells, indicating modest M2 polarization (Figure S7B).

To assess the effect of the macrophage secretome on cardiomyocytes, we treated H9c2 NF‐κB reporter cells with conditioned medium from macrophages, enabling analysis of unidirectional signaling via secreted factors without direct cell–cell contact. RAW 264.7 macrophages were stimulated with 1 µg mL⁻¹ LPS for 6 h to induce the M1b phenotype [22], while unstimulated M0 macrophages served as controls. After LPS removal and washing, macrophages were cultured in serum‐free medium, and conditioned medium was collected after 24 h. Conditioned medium from M1b or M0 macrophages was applied either undiluted or diluted to 50% onto H9c2 NF‐κB reporter cells to assess inflammatory response, or onto H9c2 cells to evaluate cell death (Figure S8A). Inflammatory assays showed that both undiluted and 50% diluted conditioned medium from M1b macrophages significantly activated NF‐κB signaling pathway (Figure S8B). Specifically, we observed a more than 7‐fold increase in NF‐κB activation compared to controls, while conditioned medium from M0 macrophages induced minimal activation (Figure S8D). These findings indicate that M1b macrophages secrete sufficient inflammatory cytokines, including TNFα, to robustly activate NF‐κB signaling, even when diluted. Furthermore, H9c2 cells exhibited significant cell death only when treated with undiluted M1b conditioned medium, as shown by live/dead staining (Figure S8C–E). This suggests that lower cytokine concentrations are sufficient to trigger inflammation, whereas higher concentrations, especially under serum deprivation, lead to cardiomyocyte death.

Building upon our earlier findings with TNFα alone, we next used the microfluidic chip to examine the temporal dynamics of inflammation induced by the full M1 macrophage secretome via one‐way paracrine signaling. Conditioned medium from LPS‐induced M1b macrophages was diluted to 50% using culture medium containing 20% fetal bovine serum (FBS), resulting in a final serum concentration of 10%. This differed from the earlier serum‐free dish experiments, as preliminary tests showed that 10% FBS was essential for maintaining cell viability in the chip, where limited medium volumes led to significant cell death under serum‐free conditions (data not shown). Conditioned medium was delivered into the cell chamber via pulsed perfusion, allowing precise control over treatment duration (Figure 4A). Prior to the experiments, food dye was used to measure chamber filling times across different numbers of parallel units and perfusion pressures (0.1–4.8 psi) (Figure S9A). As perfusion pressure increased, fill time per chamber decreased, although filling multiple chambers in parallel took longer overall (Figure S9B). Flow rates were calculated from time‐lapse imaging and chamber dimensions under each pressure condition (Figure S9C). Conditioned medium was perfused over H9c2 NF‐κB reporter cells at a low flow rate (0.6 psi for four chambers in parallel, ∼15 nL min⁻¹), with periodic refreshment until the treatment period ended, followed by washing and imaging (Figure 4A). As treatment duration increased, the percentage of NF‐κB positive H9c2 cells also increased, with significant activation observed as early as 1 h (Figure 4B). After 3 h, reporter‐positive cells showed markedly elevated fluorescence intensity (Figure 4C). These results demonstrate that H9c2 cells respond more sensitively to the cytokine‐rich M1b secretome than to TNFα alone, as evidenced by faster NF‐κB activation.

FIGURE 4.

Pulse‐chase delivery of LPS M1 macrophage‐conditioned medium to H9c2 NF‐κB reporter cells in a microfluidic chip. (A) Schematic representation of treatment or medium delivery within a single unit of the chip. Valves between chambers remained closed throughout the experiment, and delivery was achieved via slow perfusion. After overnight cell attachment, 50% diluted LPS M1b conditioned medium (CM) was applied starting at t = 0 and pulsed at designated time points, followed by washing with fresh culture medium and a chase period until the 24‐h endpoint. Fluorescence images were then acquired at 24‐h endpoint and analyzed. (B) Percentage of mKate2‐positive cells and (C) fluorescence intensity per positive cell. Data represents three independent biological replicates, each with 3–4 technical replicates per condition. All data presented as mean ± SD, and statistical analysis was performed using one‐way ANOVA with Dunnett's post hoc test; ^* p < 0.05, ^** p < 0.01, ^*** p < 0.001, ^**** p < 0.0001.

2.5. M1 Macrophage Co‐culture in Microfluidic Chip Induces NF‐κB Activation in H9c2 Cells

Pro‐inflammatory M1 macrophages release cytokines that initiate and regulate the injury response following an ischemic event in the heart. During prolonged inflammation, these cytokines impose stress on otherwise healthy cardiomyocytes located in the border zone of the injury. In response, cardiomyocytes exposed to chronic inflammatory signals begin secreting additional cytokines that further activate macrophages, creating a positive feedback loop that promotes cardiomyocyte apoptosis, worsens tissue damage, and accelerates disease progression. To model this pathological crosstalk in vitro, we established a co‐culture system of macrophages and cardiomyocytes to study inflammation‐induced cardiac stress. The microfluidic chip's high‐throughput design enabled parallel testing of multiple co‐culture conditions on a single platform. H9c2 NF‐κB reporter cells were cultured either alone in standard medium (negative control) or with TNFα supplementation (positive control for reporter activation). For co‐culture conditions, reporter cells were paired with either unstimulated (M0) or classically activated (M1a) macrophages. M1a macrophages were pre‐polarized in standard culture dishes before being seeded into chambers adjacent to H9c2 cells within the microfluidic chip. After cell seeding and attachment, microvalves were periodically actuated to enable paracrine signaling between the two cell populations (Figure 1C). Reporter activity was monitored using time‐lapse fluorescence imaging over a 24 h, with images captured every 2 h. The proportion of NF‐κB‐positive cells was quantified relative to the total number of nuclei at the final time point. Endpoint fluorescence imaging revealed strong NF‐κB activation in H9c2 cells co‐cultured with either M0 or M1a macrophages (Figure 5A,B). No reporter activation was observed in H9c2 cells cultured alone without treatment or co‐culture. Time‐lapse analysis showed a robust increase in NF‐κB activity beginning around 6 h after co‐culture initiation, reaching a plateau by approximately 12 h (Figure 5C), with individual replicates shown in Figure S10.

FIGURE 5.

M0 and M1 RAW 264.7 macrophage coculture with NF‐κB reporter H9c2 cells induces robust activation of the inflammation reporter. (A) Representative overlay images of NF‐κB reporter H9c2 cells at 24 h following culture with medium, 100 ng mL⁻¹ TNFα, M0, or M1 coculture. Blue: nuclei (DAPI); yellow: mKate2 fluorescence; gray: phase contrast. Scale bar: 100 µm. (B) Mean percentage (±SD) of mKate2‐positive cells after 24 h of coculture. (C) Time‐lapse analysis of the ratio of mKate2‐positive cells (mean ±SD) during coculture. Data represents three independent biological experiments with 3–4 technical replicates per condition. ^* p < 0.05, ^** p < 0.01, ^*** p < 0.001, ^**** p < 0.0001, non‐significant comparisons are unlabeled. Statistical analysis: (B) one‐way ANOVA with post hoc Tukey's test; (C) two‐way ANOVA with post hoc Dunnett's test.

2.6. Dexamethasone Reduces Inflammation Induced by M1a Macrophage Co‐culture

To evaluate the microfluidic co‐culture chip as a platform for screening potential anti‐inflammatory compounds, we tested the effects of interleukin‐10 (IL‐10) and dexamethasone on NF‐κB reporter activity. IL‐10 is an anti‐inflammatory cytokine secreted by immune cells such as M2 macrophages, and has been shown to inhibit NF‐κB activation by blocking IKK‐mediated phosphorylation and preventing NF‐κB nuclear translocation [14, 37]. This suppression of inflammatory signaling is associated with reduced cardiomyocyte apoptosis in vitro [13, 38]. While IL‐10 has been shown to attenuate TNFα‐induced NF‐κB activation in primary rat cardiomyocytes, its effects in H9c2 cardiomyoblasts remain uncharacterized. Dexamethasone, a synthetic glucocorticoid widely used clinically for its powerful anti‐inflammatory properties, suppresses NF‐κB signaling by inducing IκB synthesis [16], and has also been shown to protect against cardiomyocyte apoptosis [15]. As an initial assessment, we tested both compounds in conventional plate cultures to determine effective concentration ranges. IL‐10 was applied at 10, 100, and 500 ng mL⁻¹, within previously reported ranges [39], and co‐administered with 100 ng mL⁻¹ TNFα. After 24 h, NF‐κB reporter activity was measured. IL‐10 co‐treatment did not significantly reduce NF‐κB activation compared to TNFα treatment alone (Figure S11A). In contrast, dexamethasone markedly suppressed NF‐κB activation, reducing the proportion of NF‐κB‐positive cells to a fifth of what it was with TNFα treatment alone (Figure S11B). When administered 6 h after TNFα stimulation, dexamethasone maintained NF‐κB activity at levels similar to those observed following TNFα withdrawal at the same time point (Figure S11B). To assess dexamethasone's effect in the presence of macrophage‐derived inflammatory signals, we treated H9c2 reporter cells with conditioned medium from LPS‐stimulated M1b macrophages, either undiluted (100%) or diluted to 50%. Co‐treatment with dexamethasone significantly reduced NF‐κB activation under both conditions, with a stronger effect at 50% conditioned medium. This suggests that higher levels of inflammatory secretome may partially diminish the drug's efficacy (Figure S11C).

Dexamethasone also acts directly on macrophages to reduce pro‐inflammatory cytokines secretion [40, 41]. Using our microfluidic co‐culture chip, we next investigated its effects on individual cell populations and under co‐culture conditions where both cardiomyoblasts and M1a‐polarized macrophages were exposed to the drug. Based on our previous pulse‐chase experiments, which identified potential rescue windows before stable NF‐κB activation, we tested dexamethasone administration at various time points within the 0–6 h interval. The chip's multiplexing capability enabled simultaneous testing of all conditions in a single experiment (Figure S12).

We established co‐culture of M1a‐polarized macrophages and H9c2 NF‐κB reporter cells separated by a microvalve. Dexamethasone (1 µm) was applied at the start of co‐culture either to both chambers or selectively to H9c2 or RAW 264.7 cells. NF‐κB activation was fully suppressed only when both cell types received dexamethasone (Figure 6B). In contrast, treating either H9c2 or RAW cells alone led to partial suppression, with a slightly greater effect observed when H9c2 cells were treated. Endpoint analysis at 24 h confirmed that significant NF‐κB suppression occurred only with dual‐cell treatment compared to untreated M1a co‐cultures (Figure 6A).

FIGURE 6.

Effect of dexamethasone (Dex) on NF‐κB activity in M1 cocultured H9c2 reporter cells. (A,B) Dex (1 µm) was delivered via medium diffusion to the H9c2 chamber, the RAW 264.7 chamber, or both, as illustrated in Figure 1. Mean percentage (±SD) of mKate2‐positive cells at 24 h. Time‐lapse analysis of the averaged mKate2‐positive cell ratio over time during coculture and Dex treatment. (C,D) Dex (1 µm) was applied to both chambers at different time points (t = 0, 3, and 6 h) after coculture initiation. Mean percentage (±SD) of mKate2‐positive cells at 24 h, and time‐lapse analysis of averaged mKate2‐positive cell ratio corresponding to each treatment. Panels A‐B and C‐D share the same control, TNFα, M1 coculture, and M1 + Dex groups, but are presented separately for clarity; full data are shown in Figure S12. Panels B and D are shown without standard deviation bars to improve data visualization; the full datasets with SD bars are provided in Figure S12D. Data represents three independent experiments with 3–4 technical replicates per condition. ^* p < 0.05, ^** p < 0.01, ^*** p < 0.001, ^**** p < 0.0001, non‐significant comparisons are unlabeled. Statistical tests: (A, C) one‐way ANOVA with Tukey's post hoc test; (B, D) two‐way ANOVA with Dunnett's post hoc test.

To further explore the timing of dexamethasone's anti‐inflammatory effects, we introduced the drug to both cell types at delayed time points (0, 3, or 6 h). NF‐κB activation was effectively suppressed only when dexamethasone was added at the start of coculture (0 h). Delayed addition at 3 or 6 h failed to reduce reporter activity (Figure 6D). At 24 h, the proportion of NF‐κB‐positive cells in the 3 and 6 h treatment groups was comparable to the untreated co‐culture, indicating that key inflammatory signaling events occur early during co‐culture (Figure 6C). Collectively, these findings demonstrate the chip's ability to capture dynamic anti‐inflammatory responses and highlight its potential as a microphysiological platform for future drug‐screening applications.

3. Discussion

Understanding the dynamics of inflammation in cardiac disease progression is essential for uncovering underlying mechanisms and developing physiologically relevant models. Interactions between macrophages and cardiomyocytes contribute to tissue maintenance through transient immune cells residency. However, these same interactions can also initiate a feedforward cytokine cascade that elevates pro‐inflammatory signaling and leads to damage in neighboring healthy cardiomyocytes when sustained over time. Developing a model that captures this multicellular crosstalk provides a valuable platform to investigate how different signaling parameters influence cell fate decisions.

In this study, we established a cardiac inflammation co‐culture model to examine macrophage‐cardiomyocyte interactions. We employed a fluorescent NF‐κB reporter to monitor real‐time activation of inflammatory signaling in cardiomyocytes. Using TNFα as a pro‐inflammatory stimulus, we identified the concentration and exposure duration required to activate the NF‐κB reporter. We also observed NF‐κB expression in cardiomyocytes treated with M1 macrophage‐conditioned medium and in direct co‐culture, indicating that the inflammatory microenvironment induces cardiomyocyte stress responses. To assess the model's utility for therapeutic screening, we tested two anti‐inflammatory agents, IL‐10 and dexamethasone. IL‐10 co‐administration with TNFα failed to suppress NF‐κB activation, whereas dexamethasone applied at the onset of co‐culture significantly suppressed reporter activity. These results demonstrate the model's ability to capture dynamic immune‐cardiac signaling and support its potential use as a functional platform for evaluating anti‐inflammatory interventions.

Paracrine signaling is a central mechanism in inflammation, as immune cells release cytokines such as TNFα and IL‐1β that influence cardiomyocytes, fibroblasts, and endothelial cells. TNFα is a key inflammatory marker and is commonly associated with chronic inflammation and cardiac diseases progression [42]. In this study, we used TNFα as a surrogate for the immune cell secretome in a MATLAB‐based simulation to model cytokine concentration dynamics. The model focused on TNFα secretion and diffusion kinetics within a microfluidic chip and was used to optimize macrophage seeding density and medium delivery timing. We validated the model by comparing simulated TNFα concentrations with ELISA measurements, which showed similar temporal trends but a consistent two‐ to three‐fold difference in magnitude. This discrepancy may be a result of simplified assumptions in the computational model, including approximations in cytokine production, transport, and degradation, as well as from experimental limitations associated with sampling small fluid volumes in the chip. Future refinements could incorporate cardiomyocyte‐specific TNFα uptake kinetics based on experimental data.

Although we did not detect measurable TNFα in conditioned media from H9c2 cells exposed to TNFα for 24 h, previous studies have shown that cardiomyocytes can secrete TNFα and other cytokines such as IL‐6 and IL‐1β under stress conditions such as hypoxia [43] or volume overload [44]. These cytokines are typically present at low levels under baseline conditions, but their concentrations in the confined volume of the microfluidic chip may contribute to evaluated basal NF‐κB activity compared to conventional culture. While our study focused on TNFα as a representative marker of macrophage‐driven inflammation, future work will expand this approach to include additional cytokines to better characterize multicellular inflammatory signaling. Although our data suggest that cardiomyocyte‐derived TNFα play a limited role under the conditions tested, we acknowledge that low levels of TNFα may still exert biological effects that are currently below the detection threshold. Furthermore, the accumulation of other cardiomyocyte‐derived cytokines could contribute to baseline activation and should be considered in future models, especially under more severe or prolonged stress. In addition to TNFα, activated M1 macrophages secrete a variety of pro‐inflammatory cytokines, including IL‐6, IL‐1β, and MCP‐1 [22, 45]. These mediators act through distinct receptors but converge on the NF‐κB pathway, collectively amplifying inflammation in target cells [9]. In our experiments, we focused on TNFα as a representative macrophage‐derived cytokine and found that conditioned medium from M1 macrophages triggered robust NF‐κB activation in cardiomyocytes within one hour. This response reached levels comparable to those observed after 24 h of direct TNFα treatment. Although we cannot directly confirm the involvement of other cytokines, the rapid and sustained NF‐κB activation suggests that multiple inflammatory mediators may be contributing. Future studies incorporating cytokine profiling and combinatorial perturbation experiments will be important for clarifying the individual and synergistic roles of these cytokines in cardiomyocyte inflammatory signaling.

The NF‐κB signaling pathway plays a central role in sensing the inflammatory microenvironment and initiating transcriptional responses. In the context of cardiac inflammation, understanding NF‐κB activation is particularly important, as its dysregulation contributes to disease progression. NF‐κB activation has been linked to pro‐inflammatory, pro‐fibrotic, and pro‐apoptotic responses in cardiomyocytes, which exacerbate pathological cardiac remodeling and dysfunction [12]. Reporter‐based assays using fluorescent or luminescent genes have proven to be powerful tools for directly monitoring NF‐κB transcriptional activity. These systems offer real‐time, quantitative insights into NF‐κB dynamics and are widely used in inflammation research [46, 47, 48]. Stable NF‐κB reporter cell lines have been developed to study various biological processes, including immune responses in platelet precursors and neuroinflammation [46, 47]. In this study, we engineered an H9c2 cardiomyocyte cell line expressing an mKate2 fluorescent reporter under the control of an NF‐κB‐responsive promoter. Upon NF‐κB activation, mKate2 is synthesized and translocated to the nucleus, enabling direct visualization and quantification of pathway activation through colocalization with nuclear staining (e.g., Hoechst) in live cells. Nuclear localization enhances the accuracy of image‐based analysis, and the relatively long half‐life of mKate2, approximately 32.9 h in human cells [49], supports sustained detection of NF‐κB activity. However, this prolonged stability limits temporal resolution and may obscure rapid or transient signaling events. To address this limitation, future studies could incorporate fluorescent proteins with shorter half‐lives, such as d2EGFP (half‐life 3 h) [50], to better capture the kinetics of NF‐κB activation and pharmacological inhibition. In contrast, reporter systems like p65‐DSRed fusion proteins track real‐time subcellular localization of the NF‐κB complex and provide immediate readouts of nuclear translocation following stimulation [48, 51]. While localization‐based reporters are well‐suited for studying early signaling events, the mKate2 system offers the advantage of reflecting downstream transcriptional activation, providing a more integrated and sustained measure of NF‐κB pathway engagement.

Microfluidic platforms offer unique advantages for precisely controlling medium composition and delivery rate, which in turn modulate the local concentration and timing of signaling molecules. Culture volume significantly influences both stem cell and adult cell fate decisions, as well as the dynamics of autocrine and paracrine signaling [52, 53]. Our microfluidic chip employs culture volumes in the nanoliter range, approximately 100 times smaller than that of conventional dish‐based cultures. Medium is delivered through pulsatile diffusion from adjacent chambers, mimicking physiological transport via blood vessels and reducing the dilution of paracrine signals. In dose‐response and pulse‐chase experiments, NF‐κB reporter activation in the chip followed trends similar to those observed in conventional cultures. However, baseline NF‐κB activation under control conditions was consistently higher in the microfluidic chip. Several factors may contribute to this observation, including the accumulation of endogenous signaling molecules and metabolites within the confined culture volume. Environmental factors such as shear stress and adhesion were previously evaluated and found to have minimal impact [24]. ELISA quantification showed that the apparent TNFα secretion rate of M1a macrophages in the chip was approximately 17‐fold higher than in dish cultures, confirming that the confined environment amplifies cytokine accumulation.

Conditioned‐medium experiments revealed that M0 macrophages induced only minimal NF‐κB activation in H9c2 cardiomyocytes, whereas coculture in the chip produced a much stronger response. ELISA measurements indicated that M0 macrophages secrete a relatively high basal level of TNFα, which under confined conditions likely reaches concentrations sufficient to activate NF‐κB in neighboring cardiomyocytes. In contrast, cytokines in conventional cultures are rapidly diluted, resulting in subthreshold exposure. The conditioned‐medium assays also used M1b macrophages, which had higher cell numbers and secreted more TNFα, while the co‐culture experiments involved M1a macrophages that released less TNFα but expressed higher levels of iNOS. These subtype‐specific differences in cytokine output, combined with the enhanced cytokine retention in the microfluidic chip, likely explain the distinct NF‐κB activation patterns observed between conditioned‐medium and co‐culture experiments.

Cardiac macrophages make up approximately 6%–8% of non‐cardiomyocytes in the heart, and their density increases during chronic inflammation [54]. The microfluidic platform presented here offers new opportunities to model this dynamic cellular landscape with spatiotemporal precision. Modifying chamber geometry to create an anisotropic layout would allow control over the macrophage‐to‐cardiomyocyte ratio, enabling physiologically relevant crosstalk that mimics homeostatic conditions. The integrated microvalve system provides temporal control over intercellular signaling, making it possible to simulate transient inflammatory interactions observed during disease progression. Beyond static co‐culture, the platform can be adapted to study cellular migration, such as macrophage recruitment and directed movement in response to inflammatory cues. It also supports more complex multicellular configurations, including tri‐cultures in which two cell types engage in paracrine signaling while a third contributes soluble mediators via conditioned medium. Furthermore, cells can be recovered directly from the device for endpoint analyses such as PCR and multi‐omics profiling [51, 55], enabling deeper mechanistic insights into transcriptional and signaling changes driven by dynamic macrophage‐cardiomyocyte interactions.

In our cardiac inflammation model, we utilized the murine macrophage cell line RAW 264.7 and the rat cardiomyocyte cell line H9c2 to investigate immune‐cardiac cellular interactions. This cross‐species pairing is widely adopted and provides a practical, reproducible in vitro system for dissecting inflammatory signaling dynamics [22, 23]. However, potential species‐specific differences in cytokine signaling and receptor expression should be taken into account when interpreting the results. RAW 264.7 macrophages retain key features of macrophage physiology [56, 57], although they secrete detectable levels of TNFα even under basal conditions [36]. In our experiments, we observed a high M1 polarization efficiency of approximately 80%, while the M2 polarization remained lower at around 26%. The limited efficiency of M2 differentiation and the scarcity of M2 cell lines present challenges for studying the specific effects of M2 macrophages on cardiomyocytes without access to primary cells. Enrichment of M2 populations through flow cytometry or the use of primary macrophages could improve the precision of such analyses. H9c2 cells offer a convenient in vitro cardiomyocyte model due to their cardiac origin and ease of maintenance, and they have been widely used in pharmacological and physiological studies [58]. Together, the RAW 264.7 and H9c2 pairing represents a reproducible and accessible proof‐of‐concept model for demonstrating the functionality and versatility of the microfluidic co‐culture platform. Nonetheless, immortalized lines cannot fully replicate human physiology or the complexity of disease. Incorporating primary or human‐induced pluripotent stem cell (iPSC)‐derived macrophages and cardiomyocytes in future studies would enhance the physiological and translational relevance of this model [59, 60] iPSC‐derived cells would also enable the development of patient‐specific cardiac inflammation models, allowing the use of macrophages and cardiomyocytes from the same individual. This approach would support comparison between autologous and allogenic cells, facilitating studies of interindividual variability and personalized therapeutic responses. Within this context, the current system provides a strong foundation for future iPSC‐based and humanized co‐culture platforms aimed at investigating the molecular and functional consequences of immune‐cardiac interactions.

Cardiac inflammation arises through distinct cellular mechanisms depending on the underlying disease context, with each scenario involving unique intercellular crosstalk. In infectious myocarditis, inflammation is primarily driven by immune cells and is characterized by infiltration of innate and adaptive immune populations, such as macrophages and T cells, which interact with resident cardiac cells to amplify immune responses [61]. In contrast, ischemic injury, such as myocardial infarction, involves both immune and cardiac cells. Damaged cardiomyocytes release DAMPs that activate immune cells and provoke inflammation [62, 63]. In cancer therapy‐induced cardiotoxicity, particularly from agents like anthracyclines, inflammation may originate from cardiomyocyte themselves. In these cases, stressed or dying heart cells alter the local immune environment through the release of mitochondrial or nuclear components [64]. Understanding these distinct inflammatory contexts is critical for developing targeted therapies. For example, suppressing immune activation in infectious myocarditis, modulating both immune and cardiac signals following ischemia, or protecting cardiomyocytes during cancer therapy could help tailor anti‐inflammatory strategies to specific disease mechanisms. In our study, we first evaluated the anti‐inflammatory effects of IL‐10 and dexamethasone in H9c2 cardiomyocytes co‐treated with TNFα. IL‐10 failed to suppress NF‐κB activation, whereas dexamethasone showed a strong inhibitory effect. Based on these findings, we further investigated dexamethasone under more physiologically relevant inflammatory conditions. Specifically, we assessed its effects in cardiomyocytes exposed to conditioned medium from M1‐polarized macrophages and in direct co‐culture with M1 macrophages. In both settings, dexamethasone significantly reduced NF‐κB activation, highlighting its potent and direct anti‐inflammatory action in cardiomyocytes. A potential explanation for IL‐10's limited efficacy is that its anti‐inflammatory function primarily targets immune cells, such as macrophages, by inhibiting cytokine production through the JAK1/STAT3 pathway [65]. In contrast, dexamethasone acts more broadly through glucocorticoid receptor signaling, directly suppressing NF‐κB activity in both immune and non‐immune cells, including cardiomyocytes [15, 66, 67]. Notably, previous studies have shown that dexamethasone pretreatment confers myocardial protection by attenuating inflammatory signaling and preserving cardiomyocyte viability under inflammatory conditions [17, 68]. This broader mechanism of action likely accounts for the stronger inhibitory effect of dexamethasone observed in our co‐culture model. Together, these findings demonstrate the platform's ability to resolve drug‐specific anti‐inflammatory mechanisms and establish a foundation for its future development into a quantitative drug‐screening platform.

4. Conclusion

In conclusion, this study underscores the critical role of macrophage‐cardiomyocyte interaction in cardiac inflammation, with particular emphasis on the spatial and temporal dynamics of pro‐inflammatory signaling. By establishing an in vitro cardiac inflammation model that integrates an NF‐κB‐responsive fluorescent reporter with computational mass transport simulation, we identified key thresholds for inflammatory signaling and demonstrated how macrophage‐derived factors contribute to cardiomyocyte stress responses. Evaluation of anti‐inflammatory interventions revealed that dexamethasone effectively suppressed NF‐κB activation when administered early, highlighting the importance of precise temporal control in modulating inflammatory pathways. Overall, this platform serves as a versatile tool for mechanistic investigations of immune‐cardiac crosstalk and holds promise for screening and optimizing targeted anti‐inflammatory therapies.

5. Experimental Section

5.1. Cell Culture of RAW 264.7 Macrophages and H9c2 Cardiomyoblasts

RAW 264.7 macrophage cells (ATCC Cat# TIB‐71, RRID:CVCL_0493, kindly provided by Andrei Maiseyeu) and H9c2 cardiomyoblasts (ATCC Cat# CRL‐1446, RRID:CVCL_0286, kindly provided by Ozan Akkus) were used in this study. Both RAW264.7 and H9c2 cells have been characterized through phenotypic analysis. RAW264.7 cells were polarized into M1 and M2 phenotypes and identified using immunostaining markers. H9c2 cells were matured and characterized by cardiac troponin T expression (data not shown). RAW 264.7, H9c2, and NF‐κB reporter H9c2 cells were cultured in sterile conditions using medium, consisting of high‐glucose DMEM (Gibco, 11995) supplemented with 10% fetal bovine serum (FBS, Gibco, 10082147) and 1% penicillin/streptomycin (Gibco, 15140122), and maintained in a humidified incubator at 37°C with 5% CO₂. H9c2 and NF‐κB reporter H9c2 cells were passaged upon reaching 70%–90% confluence using 0.05% Trypsin‐EDTA (Thermo Fisher Scientific, 25300054). RAW 264.7 cells were subcultured by detaching them with a cell scraper and transferring to a new T75 flask upon reaching 70%–80% confluence. All RAW 264.7 experiments used cells below passage 25, and NF‐κB reporter H9c2 experiments used cells within 15 passages following lentiviral transduction.

5.2. Lentiviral Transduction

The NFкB reporter plasmid was purchased from Addgene (Gift from Timothy Lu, Addgene plasmid #82024) [29]. Bacteria containing plasmid were expanded and selected using 100 µg mL⁻¹ carbenicillin. Plasmid DNA was extracted using ZymoPURE II Plasmid Midiprep Kit (Zymo Research, D4200). HEK293Ta cells were transfected with NFκB constructs and the Lenti‐Pac HIV mix (GeneCopoeia, LT001‐01) in Opti‐MEM culture medium (Gibco, 31985070) and EndoFectin (GeneCopoeia, LT001‐03). Opti‐MEM medium was replaced with fresh culture medium (5% FBS High Glucose‐DMEM) and TiterBoost reagent (GeneCopoeia, LT001‐04) after 8–14 h, and viral supernatant was collected after 36 h and again after another 24 h. The collected supernatant was then centrifuged at 500X g for 10 min and pass through 0.45 µm polyethersulfone low protein‐binding filters to get rid of cell debris. H9c2 were plated at 1 × 10⁴ cells per cm² and transduced as previously described [69, 70] with a multiplicity of infection of 1. Transduced cells were purified with puromycin (InvivoGen, Ant‐pr‐1) for 72 h at concentrations of 0.5 µg mL⁻¹.

5.3. Microfluidic Coculture Chip Fabrication

The microfluidic coculture chips were fabricated using a two‐layer soft lithography technique as previously described in our publications [24, 71]. Briefly, polydimethylsiloxane (PDMS) and curing agent (Ellsworth Adhesives, 4019862 Dow Sylgard 184) were mixed at a 5:1 weight ratio for the control layer and a 20:1 ratio for the flow layer. The control layer PDMS was baked at 80°C for 30 min, then cut with a razor blade, and control layer inlets were punched using a 1 mm biopsy punch. The flow layer PDMS was spin‐coated onto the wafer at 1000 rpm for 60 s, left to reflow for 20 min, and then partially cured at 80°C for 6 min. After precise alignment using a stereo microscope, the control layer PDMS was bonded to the flow layer, and the assembly was baked at 80°C for an additional 30 min. The assembled chip was removed from the wafer, and flow layer inputs were punched. Plastic slides (Ibidi, 108112) were treated with oxygen plasma (30 W, 30 s), then coated with 1% v/v (3‐aminopropyl) triethoxysilane (APTES, Sigma–Aldrich, 440140) in deionized water for 20 min before bonding to PDMS assembled chips that had been treated separately with oxygen plasma (20 W, 30 s). Finally, the bonded chips were baked for 1 h and stored for at least two weeks prior to use to prevent permanent valve collapse.

5.4. Microfluidic Cell Culture Experiment

All microfluidic cell culture experiments were conducted following the procedures previously described [24, 71] Briefly, the microfluidic chip was degassed with sterile 1× PBS and sterilized by rinsing the waste channel and chambers with 70% ethanol. To prevent unwanted cell adhesion outside the chambers, the waste channel was then treated with 0.2% (w/v) Pluronic F‐127 (Sigma, P2443) in 1× PBS for 30 min. The H9c2 cell‐loading chambers were coated with 30 µg mL⁻¹ fibronectin (Sigma, FC010) in 1X PBS for 40 min, while the RAW macrophage chambers remained uncoated. The chip was incubated for at least 2 h at 37°C with 5% CO₂ before cell loading. H9c2 and RAW cells were concentrated to a density of 6–10 million cells per mL in high‐glucose DMEM supplemented with 16% (v/v) OptiPrep (Sigma, D1556), loaded into 24G PTFE tubing, and seeded into the respective cell chamber locations. H9c2 and RAW cells were supplied with culture medium every 30–60 min via slow perfusion or diffusion from side chambers (Chamber A or Chamber D), as illustrated in Figure 1C. Medium resupply and treatment were carried out using pre‐defined routines programmed in the graphical user interface and executed by an automated microfluidics control system [25]. An example of the cell culture sequence using diffusion or perfusion is provided in Note S1.

5.5. NF‐кB Reporter H9c2 Cells TNFα Dose Response

In Dish: NF‐κB reporter H9c2 cells were seeded in 96‐well plates at 80 000 cells/well and supplied with culture medium. After 18–24 h of cell attachment, half of the medium in each well was collected, mixed with equal volume of fresh culture medium, and supplemented with TNFα to achieve final concentrations of 0.001, 0.01, 0.1, 1, 10, 50, and 100 ng mL⁻¹. Cells were treated with TNFα for 24 h, followed by staining with 1 µg mL⁻¹ Hoechst 33342 for 10–20 min. Fluorescent images were acquired at 10× magnification from five random locations per well (Three biological replicates, each with three wells per condition, 15 images total per condition).

In Chip: NF‐κB reporter H9c2 cells were evenly seeded into fibronectin‐coated chambers at 6–8 million cells per mL density (approximately 40–70 cells per chamber) and allowed to adhere overnight with medium supplied via pulsed diffusion (Figure 1). Serial dilutions of TNFα (0.001–100 ng mL⁻¹) were prepared and loaded into the chip's fluid inputs. Cells were treated with TNFα medium via pulsed diffusion every 60 min for 24 h, followed by counterstaining with 1 µg mL⁻¹ Hoechst 33342. Fluorescent images were taken in the mKate2 (NF‐κB reporter) and DAPI channels. Experiments were conducted with three biological replicates, each with at least four technical replicates.

5.6. NF‐кB Reporter H9c2 cells Pulse‐chase Experiment

In Dish: NF‐κB reporter H9c2 cells were seeded in 48‐well plates at 30 000 cells per well and allowed to attach overnight. The next day, half of the medium in each well was replaced with fresh culture medium supplemented with TNFα at a final concentration of 100 ng mL⁻¹. Cells were treated for various durations (10 min, 30 min, 1, 3, 6, 12, and 24 h). After each treatment, cells were washed with sterile 1X PBS, replaced with fresh culture medium, and fluorescent images were captured at five random positions per well 24 h after the stimulus onset.

In Chip: NF‐κB reporter H9c2 cells were seeded into the microfluidic chip cell culture chambers as described earlier. TNFα (100 ng mL⁻¹ in culture medium) was slowly perfused into the cell culture chambers for durations complementing the In Dish experiment (10 min to 24 h). After 24 h, cells were counterstained with 1 µg mL⁻¹ Hoechst 33342 for 10–20 min and imaged using a fluorescence microscope.

5.7. NF‐кB Reporter H9c2 Cells Image Analysis

NF‐κB reporter H9c2 cell image analysis was conducted using CellProfiler (version 4.2.5) and MATLAB R2024b, following a previously described protocol [24]. Briefly, individual cells were identified by applying a threshold to the Hoechst‐stained nuclear channel. A mask generated from the Hoechst channel was used to measure the mKate2 intensity for individual cells as well as the background intensity. Cells were classified as positive if their mKate2 intensity exceeded the threshold defined as two standard deviations above the mean background intensity. The percentage of positive cells and positive cell mean intensity were then calculated in the custom MATLAB code.

5.8. RAW 264.7 Macrophages M1 and M2 Polarization

RAW 264.7 macrophages were polarized into M1 (pro‐inflammatory) and M2 (anti‐inflammatory) phenotypes based on established protocols [34, 56]. For M1 polarization, RAW 264.7 macrophages were seeded overnight and subsequently treated with either 1 µg mL⁻¹ lipopolysaccharide (LPS; Sigma–Aldrich, L2880) for 6 h or a combination of 25 ng mL⁻¹ IFNγ (PeproTech, 315‐05‐20 µg) and 200 ng mL⁻¹ LPS for 48 h. For M2 polarization, cells were treated with 40 ng mL⁻¹ IL‐4 (PeproTech 214‐14‐20UG) and 40 ng mL⁻¹ IL‐10 (Sigma–Aldrich H7541) for 48 h. Polarization was confirmed by immunostaining for phenotype‐specific markers: inducible nitric oxide synthase (iNOS) for M1 macrophages and arginase‐1 (Arg1) for M2 macrophages.

5.9. RAW 264.7 Macrophage Conditioned Medium Collection

RAW 264.7 macrophages were seeded into T75 flasks at a density of 2 million cells per flask and allowed to adhere overnight. The following day, the culture medium was replaced with either fresh culture medium for naïve M0 macrophages or medium supplemented with 1 µg mL⁻¹ LPS to induce polarization for 6 h. After the 6‐h incubation, the medium was removed, and the cells were washed with sterile, pre‐warmed 1X PBS. Subsequently, 10 mL of pure high‐glucose DMEM was added, and the cells were incubated for an additional 24 h. The conditioned medium was then collected, centrifuged at 1000 g for 10 min, and filtered through a 0.45 µm PES filter. The filtered medium was stored at −80°C for further use.

5.10. ELISA for TNFα Quantification from Conditioned Medium

RAW 264.7 macrophages were seeded at a density of 2 × 10⁶ cells per T75 flask and treated the following day to induce M1a (48 h with 200 ng/mL LPS and 25 ng/mL IFNγ) or M1b (6 h with 1 µg/mL LPS) phenotypes, or maintained as M0 controls, as described previously. After treatment, the reagents were removed, cells were washed with 1× PBS, and incubated for 24 h in serum‐free high‐glucose DMEM (HG‐DMEM). Conditioned medium was collected, diluted 1:5, and TNFα levels were quantified using the LEGEND MAX Mouse TNF‐α ELISA Kit (BioLegend, 430907). TNFα concentrations were normalized to cell counts measured at the time of medium collection (Figure S5A). The TNFα secretion rate was calculated based on the conditioned medium volume (10 mL), collection duration (24 h), and the corresponding cell number (Figure S5B). H9c2 cells were seeded at a density of 1.5 × 10⁶ cells per T75 flask and treated the following day with 100 ng/mL TNFα or fresh medium (control) for 24 h. Cells were then washed with 1× PBS and incubated for an additional 24 h in serum‐free HG‐DMEM. The resulting conditioned medium was analyzed undiluted using the LEGEND MAX Rat TNF‐α ELISA Kit (BioLegend, 438207) (Figure S5C). ELISA measurements were performed on medium collected from three biological replicates.

5.11. M1a TNFα Quantification from Microfluidic Chip

M1a macrophages were seeded into all 16 units (chambers A‐D) of the microfluidic chip and allowed to attach overnight. After seeding, normal culture medium (10% FBS, 1% P/S in HG‐DMEM) was slowly perfused into each chamber every 30 min. Each perfusion cycle required an excess volume to fully displace previously collected medium, resulting in an estimated dilution ratio of 1:13, calculated from food‐dye‐based measurements of chamber filling time (Figure S9). Conditioned medium was collected over four intervals: 0–11, 12–27, 28–49, and 50–73 h. Cell numbers were estimated from manual counts of time‐lapse images at four representative chamber locations, or at the final timepoint (73 h), using Hoechst staining to assess proliferation (Figure S6B). TNFα concentrations in conditioned medium were quantified at each interval using the LEGEND MAX Mouse TNF‐α ELISA Kit (BioLegend, 430907) according to the manufacturer's instructions (Figure S6D). TNFα secretion per cell was calculated based on the estimated cell numbers at each time point (Figure S6A), and secretion rate for each time interval was calculated from the conditioned medium volume, collection duration, and corresponding cell count (Figure S6C).

5.12. Microfluidic Chip Coculture Experiment

NF‐κB reporter H9c2 cells were seeded into Chamber B of the microfluidic chip and allowed to adhere overnight as described in the previous section. For co‐culture conditions, either naïve M0 or pro‐inflammatory M1 RAW 264.7 macrophages were seeded into the adjacent Chamber C, depending on the experimental group (M0 co‐culture or M1 co‐culture group). Following cell seeding, normal culture medium or medium supplemented with 100 ng mL⁻¹ TNFα was introduced into the chambers via pulsed diffusion. To enable cellular crosstalk, the microfluidic valve separating the chambers was periodically opened for 22 min every 36 min, allowing the secretome to diffuse between the cell populations, as depicted in Figure 1. During the remaining 14 min of each cycle, fresh medium was supplied through pulsed diffusion to maintain optimal culture conditions. For the dexamethasone treatment experiment, a culture medium containing 1 µm dexamethasone was supplied to the designated medium chamber at the specified treatment time. Fluorescent time‐lapse images were acquired using an Olympus IX81 microscope at 2‐h intervals over a 24‐h period to monitor changes in NF‐κB reporter mKate2 expression. The number of mKate2‐positive cells in each time frame was analyzed from the time‐lapse images using CellProfiler. The percentage of positive cells was calculated by dividing the number of mKate2‐positive cells in each frame by the total cell count, determined from Hoechst‐stained nuclei in the final image at 24 h. The averaged reporter activation trend was derived from three biological replicates, with each biological replicate containing 3–4 technical replicates per condition.

5.13. TNFα Secretion and Diffusion Model

To assess the impact of pulse medium supply on paracrine signaling and estimate TNFα concentration changes in the cardiomyocyte chamber, we developed a 2D MATLAB model simulating TNFα secretion and diffusion from macrophages. The MATLAB live script (version R2022b), with detailed explanations, is provided as File S1. The model incorporates the geometry of a single chip unit, consisting of four equally sized chambers (A‐D, as shown in Figure 1) connected by sieve valves that can be opened or closed. Cardiomyocytes and macrophages were located in the two central chambers (B and C, respectively), while medium were periodically perfused through the side chambers (A and D), allowing diffusion‐based medium exchange with the cell culture chambers. Assumptions in the model include the following: (1) cardiomyocyte TNFα consumption is not considered; (2) TNFα secretion from macrophages is constant. The TNFα secretion‐diffusion equation describing changes in TNFα concentration over time is given as follows:

$$\frac{\partial C}{\partial t}=D\left(\frac{{\partial }^2}{\partial x^2}+\frac{{\partial }^2}{\partial y^2}\right)C+P_0{N}_0{e}^{\mu t}$$ (1)

where C represents the TNFα concentration, D is the diffusion coefficient of TNFα approximated as 1.093 × 10⁻⁵ mm² s⁻¹ [26], and P₀ is the TNFα secretion rate per cell. The TNFα secretion rate of M1a macrophages over 24 h of culture was quantified via ELISA and approximated as 3.7294 × 10⁻⁷ ng h⁻¹ cell⁻¹ in the macrophage cell chamber (Figure S6C). N₀ represents the initial cell seeding number, µ is the cell growth rate, approximated as 0.0131 h⁻¹ based on experimental cell proliferation data (Figure S6B), and P was set to 0 in all the other chambers. To simulate periodic medium resupply in chamber A and D, the TNFα concentration was reset to 0 ng mL⁻¹ (indicating no TNFα in culture medium) at designated resupply interval.

An additional TNFα MATLAB model was constructed in MATLAB to replicate the experimental setup used for ELISA validation (Figure S6E–G). In this configuration, all four chambers (A‐D) were seeded with M1a macrophages, and the conditioned medium was collected every 30 min, resetting the TNFα concentration to zero after each cycle. The model incorporated the experimentally measured cell numbers, medium resupply frequency, and incubation duration to simulate TNFα concentration dynamics within the chip. Model validation was performed by comparing simulated TNFα concentrations with experimentally measured values under identical conditions.

5.14. Immunocytochemistry and Immunofluorescence

RAW 264.7 macrophages were fixed with 4% paraformaldehyde (PFA) for 10–20 min at room temperature. Following fixation, the cells were permeabilized with 0.1% Triton X‐100 (TX100) solution for 20 min and then blocked with 5% goat serum in 0.1% TX100 PBS solution (staining buffer) for 20 min. M1 or M2 polarized macrophages were incubated overnight at 4°C with either a mouse monoclonal iNOS primary antibody (1:400 dilution in staining buffer; Santa Cruz Biotechnology, sc‐7271) for M1 macrophages or a mouse monoclonal ARG1 primary antibody (1:400 dilution in staining buffer; Santa Cruz Biotechnology, sc‐271430) for M2 macrophages. After washing with permeabilization buffer, the cells were incubated with goat anti‐mouse Alexa Fluor 488 secondary antibody (1:200 dilution in staining buffer; Thermo Fisher Scientific, A‐11001) for 1 h at room temperature. Nuclear staining was performed with DAPI (2 µg mL⁻¹), followed by washing with 1X PBS. Imaging was carried out using an inverted Olympus IX81 microscope.

5.15. Live/Dead Staining

H9c2 cells were seeded in 48‐well plates at a density of 30 000 cells per well (27 000 cells cm^{− 2}) and maintained in culture medium, with three technical replicates per condition. Experiments were repeated three times. After 18–24 h of attachment, cells were treated with either high‐glucose DMEM medium (control) or conditioned medium from naïve M0 or LPS‐M1 cells at different dilution ratios (50% or undiluted 100%). Following 24 h of treatment, the medium was removed, and cells were incubated with 2 µm Ethidium Homodimer‐1 (EthD‐1) and 4 µm Calcein‐AM for 20–30 min using the Live/Dead Viability/Cytotoxicity Kit (Invitrogen, L3224), followed by a 10‐min incubation with 2 µg mL⁻¹ Hoechst. Fluorescent images were captured at five random positions per well, and the number of dead cells (EthD1‐positive) and total cells (Hoechst‐positive) was quantified using a CellProfiler pipeline. The percentage of dead cells was calculated as the ratio of EthD1‐positive cells to total Hoechst‐stained nuclei.

5.16. IL‐10 and Dexamethasone Anti‐inflammatory Experiment

NF‐κB reporter H9c2 cells were seeded in a 96‐well plate at a density of 8000 cells per well and allowed to attach overnight. The following day, half of the culture medium from each well was collected and mixed with an equal volume of fresh culture medium. Depending on the experimental condition, the mixed medium was either returned to the cells without supplementation, or supplemented with one of the following: 100 ng mL⁻¹ TNFα; a combination of 100 ng mL⁻¹ TNFα and 1 µm dexamethasone (ThermoFisher, 230302500); or 100 ng mL⁻¹ TNFα combined with varying concentrations of mouse IL‐10 (10, 100, or 500 ng mL⁻¹; PeproTech, AF‐210‐10‐10UG). For the time‐course treatment groups, the medium containing 100 ng mL⁻¹ TNFα was collected from the wells, supplemented with 1 µm dexamethasone, and returned to the wells. After 24 h of stimulation, the treatments were removed, and 1 µg mL⁻¹ Hoechst was added to stain the nuclei. The cells were then imaged at 10× magnification using an Olympus IX81 microscope.

5.17. Statistical Analysis

Statistical analyses were performed using GraphPad Prism software (version 8.0.1). For dose‐response and pulse‐chase experiments, group comparisons were conducted using one‐way analysis of variance (ANOVA) followed by Dunnett's post hoc test, with comparisons made against the control group. For comparisons involving the anti‐inflammatory drugs IL‐10 and dexamethasone, one‐way ANOVA followed by Tukey's post hoc test was used to assess differences among all groups. For time‐lapse analyses, data at each time point were compared to the corresponding control using two‐way ANOVA followed by Dunnett's post hoc test. A p‐value of < 0.05 was considered statistically significant.

Funding

R01 NS121374 (S.S.), NIH/NIBIB EB021911 (R.S.), T32 5T32EB007509, MSTP 5T32GM007250, 5TL1TR002549 (D.W.), CWRU‐ENGAGE (X.Q.), NIH/NIDDK R01DK128463 (P.R.), JDRF 3‐SRA‐2022‐1193‐S‐B (P.R.), 3‐SRA‐2024‐1552‐S‐B (P.R.), and VA BLRD I01 BX005941(P.R.).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File 1: adhm70961‐sup‐0001‐SuppMat.docx.

Supporting File 2: adhm70961‐sup‐0002‐SuppMat.docx.

Data Availability Statement

The data that support the findings of this study are available in the supplementary material of this article.