Introduction

Elucidating the connectivity pattern and dynamic interaction mechanisms among distinct regions of the nervous system constitutes a fundamental challenge in understanding the brain’s information processing mechanism1,2,3. Growing evidence suggests that the brain executes complex functional tasks via the coordinated activity of distributed neural circuits rather than isolated local firing patterns. Core cognitive and behavioral functions, such as sensory integration, decision-making, and learning and memory, depend on the interplay of transmission, integration, and reconfiguration of information in distributed functional areas, and not on independent computations carried out in individual brain areas4,5. During this process, synchronized population firing, phase coordination of populations, and functional connectivity’s dynamic reconfigurations have been identified as critical mechanisms for high-efficiency information encoding and transmitting in neuronal networks6,7.

In recent years, in vitro cultured neuronal networks have provided reproducible experimental models for investigating cross-regional information processing, synaptic plasticity, and the functions of neuronal networks8,9,10. Specifically, using microelectrode arrays (MEA), researchers can achieve long-term recording and electrical stimulation, thereby enabling the observation of neuronal network activity characteristics at multiple scale8,9. However, conventional MEA typically cultivate large-scale neuronal populations on a single chip surface, rather than enabling compartmentalized and spatially isolated neuronal cultures10,11,12, which limits its application to explore the mechanisms of cross-regional neuronal network interactions13,14. To address these challenges, researchers have developed diverse compartmentalized culture architectures that rely on microfluidic technology. Through the connection of microchannels, they constructed spatially independent but functionally interacting neuronal populations on the same chip. These approaches depend on the spontaneous growth of axons to connect compartments to establish synaptic connections across compartments, providing an essential means for investigating the developmental processes and functional connectivity of neuronal networks15,16. However, this ‘natural connectivity’ approach suffers from high stochasticity in connectivity strength, temporal order, and topological patterns, posing significant hurdles to the precise manipulation of information exchanges across distinct neuronal networks.

Most importantly, while basic functional connectivity within compartmentalized networks has been studied, there is a lack of systematic and quantitative investigation into their responses at the levels of synchrony, functional coupling, and plasticity after the introduction of physical electrical interconnections. Exploring these dynamics is of significant importance in elucidating the network-level reconfiguration mechanisms in response to artificial modulation as well as for developing controllable in vitro models of neuronal information processing.

To address these issues, this study designed a CMNC integrated with a MEA and a specialized compartmentalized microchamber structure. In terms of hardware architecture, we employed digital electrophysiological interface chips from Intan Technologies, widely validated for their reliability in neuronal signal recording and stimulation17,18,19,20,21. Moreover, this study innovatively introduces an analog switch matrix to achieve interconnection control between electrode sites, thereby supporting cross-regional electrical interconnection.

Utilizing this microsystem, this study recorded electrophysiological signals from the DCNNs at three stages: control, interconnection, and post-disconnection. Functional connectivity was quantitatively assessed using the Spike Timing Tiling Coefficient (STTC)22,23, Pearson Correlation Coefficient (PCC)24,25 and Phase Locking Value (PLV)26. These metrics analyze the impact of electrical interconnection on network coupling across three dimensions: spike timing synchrony, firing activity correlation and phase coherence27. Furthermore, we constructed a dynamical model based on Wilson–Cowan type equations28 to elucidate the observed experimental phenomena hereafter referred to as the Electrical-Interconnection Wilson–Cowan Model (EI-WCM). A coupling parameter K was introduced into the model to represent the regulatory influence of the physical pathways on network integration29,30,31.

In conclusion, this integrated microsystem and dynamical modeling approach provide a stable and controllable experimental and theoretical platform for investigating functional connectivity, synergetic interactions and plasticity in in vitro neuronal networks. Furthermore, it offers innovative tools and insights for the advancement of brain-computer interface (BCI) and neural computing.

Materials and methods

Architecture of the microsystem and design of the hardware circuitry

The microsystem architecture (Fig. 1a) comprises two primary subsystems: a front-end CMNC for spatially isolating neuronal cultures, and a back-end control unit facilitating programmable interconnection and high-fidelity signal recording. The programmable electrical interconnection module is created with an AD75019 analog crosspoint switch (Analog Devices, Wilmington, MA, USA) and the recording module is based on the RHD2164 chip (Intan Technologies, Los Angeles, CA, USA). All hardware resources are controlled by an Artix-7 FPGA (Xilinx Inc., San Jose, CA, USA), which also controls communication with the host PC. The power management unit is based on a multi-stage voltage regulation system, which provides stable supply rails for each of the system’s function sub-modules. The unit has step-down DC–DC converters (LM2596SX-5.0 and TPS65130, Texas Instruments, USA) integrated with low-dropout (LDO) linear regulators (AMS1117, Advanced Monolithic Systems, USA and TPS7A30, Texas Instruments, USA), delivering accurate and noise-immune operating potentials.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Microsystem design schematic of the experimental platform. a Overall microsystem architecture. b Schematic diagram of the integrated microsystem. c Vertical cross-sectional view of the internal platform layers. d FPGA-based control logic and data flow

The microsystem is physically composed of a vertical three-layer hierarchical structure (Fig. 1b, c): CMNC comprises the top layer; the programmable electrical interconnection module acts as the middle layer; and the bottom layer consists of FPGA, signal recording, and power management modules. The circuit board design was completed using Altium Designer 20.2.6 (Altium LLC, San Diego, California, USA), and the corresponding printed circuit board (PCB) was fabricated. A six-layer PCB stack-up with dedicated ground and power planes was employed to ensure signal integrity and minimize electromagnetic interference (EMI) in sensitive electrophysiological recordings32,33. Figure S1a and Fig. S1b show pictures of the fabricated PCB, respectively. The electronic modules in the system are connected via flexible printed circuits and the assembly is housed inside a customized aluminum enclosure. This design provides excellent EMI shielding and suppresses surrounding noise to allow for accurate signal fidelity for sensitive electrophysiological recordings.

The system control logic is implemented on the Vivado 2020.2 (Xilinx Inc., San Jose, California, USA) environment using Verilog HDL and SystemVerilog (Fig. 1d). A unified system clock serves as the timing reference, and data from different frequency domains are managed via asynchronous FIFOs to effectively prevent data loss. Within the interconnection module, the system sends control commands through a UART interface to selectively enable microelectrode connections, achieving programmable electrical interconnection control. For the signal recording module, the FPGA acts as the master controller, configuring the RHD2164 chip through a differential SPI interface and sending real-time polling recording commands. The recorded data are amplified, filtered, and digitized, then written into the FIFO, packaged frame by frame, and transmitted to the host PC via USB 2.0.

Design, fabrication, and integration of the CMNC

The CMNC developed in this study comprises a 256-channel MEA with a 64-compartment microchamber structure.

The design of the 256-channel MEA was based on our previous work34, the array consists of 256 recording sites arranged into four square sub-arrays, each configured as an 8 × 8 regular grid. The detailed fabrication process of the MEA is provided in Supplementary Material S2, as illustrated in Fig. S2. To enhance electrode performance, the electrode sites were coated with a PtNPs/PEDOT:PSS bilayer composite34,35,36,37,38. Electrode modification methods are presented in supplementary material S3 and Figs S3–S4.

A microchamber structure consisting of 64 square compartments, matching the layout of the aforementioned 256-channel MEA, was designed and fabricated. Each square compartment has a size of 1000 × 1000 μm, and the corners of each compartment are rounded with a radius of 200 μm. The overall size of the structure is 12,400 × 12,400 μm (as shown in Fig. S5a). Each microchamber precisely covers four microelectrode sites on the MEA, and adjacent chambers are physically separated by predesigned Polydimethylsiloxane (PDMS) microstructures, effectively preventing the intermixing of neurons from different regions during culture and ensuring the independence and controllability of each compartmentalized neuronal network.

The fabrication process of the microchamber structures is illustrated in Fig. 2a. The specific process flow includes photolithography and development of alignment mark pattern, deposition and patterning of Pt alignment mark, photolithography of the microchamber structure, development of the microchamber mold structures, PDMS spin-coating and curing, release of the PDMS layer and the detailed steps and parameters are provided in Supplementary Material S2. The resulting microchamber structures are shown in Fig. 2a-VI. For integration, the microchamber structures were precisely aligned with the surface-modified MEA (Fig. 2b-I) under a microscope and temporarily fixed by adsorption with the assistance of 75% ethanol (Fig. 2-II). Subsequently, a liquid PDMS layer was used as an adhesive to position and secure the neuronal culture ring at the outer edge of the microchamber structures (Fig. 2b-III), completing the integration of the CMNC. Figure S5b presents the physical image of the CMNC integrated with MEA and microchamber structures.

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Fabrication and Integration of the CMNC. a Fabrication process of the compartment microchamber structures: (I) Photolithography and development of alignment mark patterns; (II) Deposition and patterning of Pt alignment marks; (III) Photolithography of the microchamber structures; (IV) Development of the microchamber mold structures; (V) PDMS spin-coating and curing; (VI) Release of the PDMS layer. b Integration workflow of the CMNC: (I) The surface-modified MEA substrate; (II) Ethanol-assisted alignment of the PDMS microchambers onto the MEA surface; (III) Bonding of the neuronal culture ring; (IV) Seeding of hippocampal neurons. c A complete schematic diagram with a 64-compartment layout. d Magnified schematic view of the four microelectrode sites within a single compartment. e Scanning electron microscopy (SEM) images of the fabricated microelectrode sites

Cell culture

Following integration, primary hippocampal neurons were seeded into the compartments, as illustrated in Fig. 2b-IV. In vitro hippocampal neuronal networks can generate stable spontaneous synchronized bursting activity, making them an important experimental model for studying neuronal network synchronization, plasticity mechanisms, and neural dynamics39,40.

Primary hippocampal neurons were cultured according to previously established protocols34,41. This experimental plan was reviewed and approved by the Ethics Committee of the Institute of Aerospace Information Innovation of the Chinese Academy of Sciences. All the researchers involved in the experiment held valid licenses for laboratory animal handling in Beijing. Pregnant ICR mice at 15.5 days of gestation were euthanized using the cervical dislocation method and their embryos were removed and placed in pre-cooled Hank’s balanced salt solution buffer. Hippocampal tissue was isolated from embryonic brains, minced, and then digested in Dulbecco’s modified Eagle medium containing papain and DNase for 15 min. Following digestion, the cell suspension was centrifuged at 100 × g for 5 min and resuspended in NeurobasalTM Plus medium. The neurons were inoculated at a density of 2 × 10³ cells/mm² into each chamber of the neural chip, and cultured in a 5% CO₂, 95% humidity 37 °C incubator. After 24 h of inoculation, Ara-C (10 μM) was added to the culture medium to inhibit the excessive proliferation of glial cells. The culture medium was replaced every 3 days. Under these culture conditions, the resulting neuronal networks are predominantly composed of excitatory glutamatergic neurons, with a relatively low proportion of inhibitory interneurons42,43.

Signal recording and process of the DCNNs electrical interconnection experiment

After culturing hippocampal dissociated neurons in the CMNC for 14 days, we performed DCNNs electrical interconnection experiments using the custom-built microsystem, while simultaneously recording neuronal spike activity at a 30 kHz sampling rate. The experimental procedure is shown in Fig. 4a. The experiment can be divided into three stages, lasting a total of 25 min. The first 0–5 min are for the control phase, during which the spontaneous activity of neurons in each chamber are recorded, the next 5–10 min are for the experimental break. From 10 to 15 min, electrical interconnections are established between the DCNNs, as shown in Fig. 4b. Five sets of interconnection paths were established, namely S1–S2, S3–S4, S5–S6, S7–S8, and S9–S10. Each pair of paths corresponds to an independent DCNNs group. Each DCNNs group consists of two compartmentalized neuronal networks cultured on the CMNC and electrically interconnected to form an independent dual-chamber neuronal network system. Therefore, a total of five independent DCNNs experimental samples were included in the subsequent statistical analyses (n = 5). The next 15–20 min are for the experimental break; from 20 to 25 min, the discharge activities are recorded after the interconnections are disconnected.

For the raw data recorded by the microsystem at a sampling rate of 30 kHz, post-processing was first performed using Python 3.12(Python Software Foundation, Wilmington, DE, USA). The original binary data streams acquired by the underlying hardware were converted into .NEV and .NS6 formats to facilitate subsequent analysis.

During spike signal processing, the built-in fourth-order Butterworth digital band-pass filter in Offline Sorter v3 (Plexon Inc., Dallas, Texas, USA) was applied to the raw signals, with a passband of 300–3000 Hz, to effectively isolate neuronal action potential components. Subsequently, spike detection was performed using the software’s built-in peak detection algorithm based on raw voltage thresholding. The detection threshold was determined using the median absolute deviation (MAD) method to estimate the background noise standard deviation (σ) and was set to −4.5σ44. After spike detection, spike waveforms were subjected to principal component analysis (PCA) for feature extraction, followed by classification of neuronal units using the Valley Seek automatic clustering algorithm. For LFPs, the raw neural signals were processed using custom data analysis scripts written in Python 3.12. First, a fourth-order Butterworth digital band-pass filter was applied to retain the 1–300 Hz frequency band, effectively removing low-frequency baseline drift and high-frequency spike components, thereby extracting the LFPs. The filtered signals were then downsampled to 1 kHz.

Results

Electrochemical characterization of microelectrode sites

Electrochemical Impedance Spectroscopy revealed a drastic reduction in impedance across the broad frequency spectrum (10 Hz–10 MHz) following surface modification. Correspondingly, the phase lag was significantly optimized (Fig. 3a, b). At 1 kHz—the standard reference frequency for neural recording—the impedance dropped from 897.41 ± 164.68 kΩ for bare electrodes to 20.28 ± 1.47 kΩ following PtNPs modification, and further reached 11.51 ± 1.06 kΩ after the PtNPs/PEDOT:PSS composite coating (Fig. 3c). At the same time, the phase angle enhanced from –72.65 ± 6.26°to –52.60 ± 1.86° with PtNPs and was further improved to –34.99 ± 2.42° with the composite modification. The results demonstrate that electrode modification significantly enhances the electrochemical performance, thus providing a reliable foundation for high-quality neuronal signal recording.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

Electrochemical characterization of MEA. a Impedance magnitude response across the frequency range of 10 Hz to 1 MHz for bare, PtNPs-modified, and PtNPs/PEDOT:PSS-coated electrodes (n = 11). b Corresponding phase response (n = 11). c Statistical comparison of impedance and phase angle at 1 kHz (n = 11, p < 0.001). Statistical significance was assessed using Student’s t-test (*p < 0.05, **p < 0.01, ***p < 0.001)

Analysis of firing activity in DCNNs

The firing activities of the DCNNs were recorded across the three experimental phases: control, interconnection, and post-disconnection. Figure 4c displays the spike activity, while Fig. 4d illustrates the LFPs.

Fig. 4
Fig. 4The alternative text for this image may have been generated using AI.
Full size image

Experimental procedure and neuronal activity recording. a Schematic illustration of the overall experimental procedure. b Schematic diagram of the electrical interconnection between DCNNs(n = 5). c Representative spike activity recorded from different DCNNs(n = 5). d LFPs signals recorded from different DNCCs(n = 5)

To quantitatively assess the effects of electrical interconnection on the functional connectivity, coordinated activity, and plasticity of DCNNs, this study employed three complementary synchrony metrics: STTC, PCC, and PLV. These metrics characterize network synchrony from three distinct dimensions—spike timing synchrony, firing activity correlation and phase coherence—providing a more comprehensive evaluation of network dynamic changes. Among these, the STTC is used to evaluate the temporal consistency of spike trains within individual neurons or neuronal populations. STTC is calculated based on the relative timing between discrete spike events rather than the absolute firing rate, making it robust to changes in firing frequency and effectively reflecting the precise temporal synchrony of neuronal spiking23. In contrast, PCC measures the linear correlation of firing activity between different neuronal populations over time. By comparing the overall trends of neuronal firing across populations, PCC captures network-level coordinated changes, revealing functional coupling or cooperative effects between neuronal groups45,46. Additionally, PLV is employed to assess the phase synchrony of LFPs across different neuronal populations. Unlike spike-based metrics, PLV focuses on the consistency of signal phase rather than amplitude, allowing characterization of oscillatory-level phase coherence in the network and reflecting the degree of coordination in population neural activity at the level of network dynamics26.

Figure 5a illustrates the STTC variations of five DCNNs groups across the three experimental phases (control, interconnection, and post-disconnection), while Fig. 5b presents the corresponding summary statistics. Notably, STTC increased significantly during the interconnection phase compared to the control baseline (Δ = 0.69), demonstrating that electrical interconnection robustly promotes temporal synchrony and functional coupling of spike activity between different DCNNs. In addition, the PCC analysis revealed a mean increase of Δ = 0.43 during the interconnection phase (Fig. 5c, d), indicating stronger linear correlations and enhanced coordinated firing activity across neuronal networks.

Fig. 5
Fig. 5The alternative text for this image may have been generated using AI.
Full size image

Synchronization analysis in DCNNs. a Changes in STTC across the three experimental phases (n = 5). b Statistical box plots of changes to STTC. c Changes in PCC across the three experimental phases (n = 5). d Statistical box plots of changes to PCC. e Changes in PLV across the three experimental phases (n = 5). f Statistical box plots of changes to PLV. Significance was calculated using a one-sided Wilcoxon signed-rank test t: *p < 0.05

Furthermore, phase synchronization analysis based on PLV demonstrated a significant increase in phase coherence of local field potentials between different DCNNs during the interconnection phase, with a mean change of Δ = 0.31 (Fig. 5e, f). This result indicates that electrical interconnection not only enhanced spike-level synchrony but also promoted phase alignment at the network oscillatory level, reflecting strengthened dynamic coupling between networks.

Following the disconnection of the interconnection pathways, all synchrony metrics exhibited a partial decrease; however, their values remained elevated relative to the control level, indicating a residual effect. Specifically, compared with the baseline, STTC, PCC, and PLV increased by Δ = 0.11, Δ = 0.15, and Δ = 0.17, respectively. These findings suggest that even after the removal of electrical interconnections, enhanced functional connectivity and coordinated activity between networks were partially retained, potentially reflecting plasticity or persistent modulation of network states induced by electrical interconnection.

Construction of the dynamical model

To provide a mechanistic understanding of the observed synchronization and residual effects, we formulated the EI-WCM(see Fig. 6a), a phenomenological model derived from Wilson-Cowan mean-field dynamics28. In this model, the internal dynamics of both neuronal networks incorporates spontaneous excited (E) and inhibition (I) activities, while the interactions between the two regions are mediated by a coupling parameter K. Specifically, K = 0 reflects the absence of electrical interconnection (control), whil K = 1 indicates electrical interconnection pathways are present.

Fig. 6
Fig. 6The alternative text for this image may have been generated using AI.
Full size image

Construction and validation of the EI-WCM. a Schematic diagram of the EI-WCM. b Phase space trajectory (K = 0). c Phase space trajectory (K = 1). d Response of STTC and PLV to the coupling strength wcoup

For each region (\(i,j\in \{\mathrm{1,2}\}\) and \(i\ne j\)), the dynamics of the mean firing rates for the excitatory and inhibitory activities Ei(t) and Ii(t) are governed by the following system of differential equations:

$${{\rm{\tau }}}_{E}\frac{d{E}_{i}}{{dt}}=-{E}_{i}+{S}_{E}\left({w}_{{EE}}{E}_{i}-{w}_{{EI}}{I}_{i}+{P}_{i}\left(t\right)+\mathop{\underbrace{K\cdot {w}_{{coup}}{E}_{j}}}\limits_{\text{Coupling Term}}\right)$$
(1)
$${{\rm{\tau }}}_{I}\frac{d{I}_{i}}{{dt}}=-{I}_{i}+{S}_{I}\left({w}_{{IE}}{E}_{i}-{w}_{{II}}{I}_{i}+{Q}_{i}\right)$$
(2)

Where:

τEτI: Standard time constant28;

wEE: the synaptic weight of self-excitation within the excitatory population;

wEI: the synaptic strength of local inhibition from the inhibitory population;

wIE: the synaptic weight of local excitation from the excitatory population to the inhibitory population;

wII:the synaptic weight of self-inhibition within the inhibitory population;

\(K\cdot {w}_{{coup}}\): the introduced inter-region coupling term.

Here, S(x) is the sigmoidal activation function, defined as28,47:

$$S(x)=\frac{1}{1+{e}^{-\alpha (x-\theta )}}$$
(3)

Here, Pi(t) represents a linear combination of independent noise \({\xi }_{i}(t)\) for each region and common noise \({\xi }_{c}(t)\).

$$\,{P}_{i}(t)={P}_{{base},i}+\sigma [(1-\rho ){\xi }_{i}\left(t\right)+\rho {\xi }_{c}\left(t\right)]$$
(4)

ρ is the proportion of common input, and σ is the noise amplitude. In this model, a small external drive mismatch \(({P}_{1}\ne {P}_{2})\), so that the two regions exhibit different intrinsic oscillation frequencies in the absence of coupling.

Validation of the dynamical model

To explain the observed experimental phenomena in a semi-quantitative way using the EI-WCM, we created phase-plane trajectories using the normalized firing rates for the two regions (E1 and E2) as state variables. In Fig. 6b, with K = 0, the trajectories’ distributions are scattered, which mathematically corresponds to the system having a high degree of freedom, indicating that the two regions are dynamically independent. In contrast, with an electrical interconnection (K = 1) in Fig. 6c, the trajectories converged rapidly to a low-dimensional attractor near the diagonal. Upon observing these geometric characteristics, it was noted that the coupling caused strong phase synchronization and the trajectories did not perfectly overlap the diagonal, thus preserving the local independence of each individual neuronal network.

Furthermore, the parameter response characteristics of the EI-WCM were validated by quantitatively evaluating the responses of synchronization metrics to variations in the coupling strength wcoup under the condition K = 1 (STTC and PLV in Fig. 6d, see Supplementary Material S4 for the full calculation methods). The coupling strength, wcoup, was continuously varied in the range [0, 4] in increments of 0.25. For each value of the parameter, 20 independent simulations were performed. Figure 6d shows the average value and the shaded area represents the standard deviation. We observed that the synchronization metrics exhibit a typical nonlinear sigmoidal response to wcoup. Beyond a specific coupling threshold, the synchronization coefficients begin to increase but remain strictly less than 1. This incomplete saturation phenomenon confirms that the model construction aligns with biological interpretations; specifically, that upon the establishment of electrical interconnection, the networks maintain their distinct activities rather than reaching a state of total synchronization. Regarding the interpretation of the residual effect, it can be understood as the persistence of the coupling strength wcoup after the two neuronal networks have interacted. In this state, the connectivity parameter K effectively remains at an intermediate value between 0 and 1. All model parameter values used in these simulations are listed in Supplementary Table S1.

Discussion

This research implemented a custom-built integrated microsystem that enables compartmentalized neuronal culture, programmable electrical interconnection between DCNNs, and simultaneous multichannel electrophysiological recording. Experimental results demonstrated that the introduction of electrical interconnection led to significant increases in multiple synchronization metrics, including STTC, PCC, and PLV, across DCNNs These results demonstrate that artificial electrical coupling, which is fundamentally distinct from natural synaptic connectivity, can robustly enhance spike timing synchrony, firing activity correlation and phase coherence between spatially separated neuronal networks. Importantly, this suggests that physical electrical interconnection may serve as an alternative functional coupling modality for modulating coordinated neural dynamics.

Intriguingly, although all synchronization metrics partially receded after disconnection, they persisted at levels significantly above the control baseline, underscoring a sustained residual effect. This observation suggests that electrical interconnection not only induces an immediate enhancement of network synchrony but may also trigger a transient alteration in the functional state of the neuronal networks. Such residual synchrony suggests that the neuronal network exhibit a certain degree of plastic response to the electrical interconnection intervention, potentially reflecting short-term functional reorganization or sustained modulation of the network’s dynamical state. Potential short-term plasticity mechanisms may include local changes in extracellular ion concentrations induced by electrical interconnection, or short-term synaptic potentiation. These processes may enhance neuronal firing synchrony and strengthen network coupling. In addition, electrical interconnection may induce a reorganization of oscillatory synchrony at the network level, thereby exerting a lasting influence on the dynamical state of collective neuronal activity. However, the specific biological mechanisms underlying these phenomena remain to be verified through more systematic experimental investigations in future studies.

It should be clarified that the “electrical interconnection” established via the AD75019 switch matrix refers to a direct electrical link between DCNNs, allowing the activity of one compartment to be directly transmitted to another. This interconnection is regarded as an externally applied and controllable synchronous stimulus, designed to couple two originally isolated neuronal populations into a shared pattern of electrical activity. The electrical interconnection is not intended to mimic or replace natural synaptic connections; rather, it represents an artificially constructed physical coupling used to investigate its modulatory effects on neuronal network dynamics.

At the same time, from a physical perspective, the process for establishing the electrical interconnection inevitably introduces a certain degree of signal mixing. Therefore, the observed increases in STTC, PCC, and PLV during the interconnection phase should be understood as resulting from a combination of genuine biological functional coordination and artifacts from electronic coupling. In this context, the interconnection phase should be regarded as an induced stimulation process rather than a definitive measure for quantitatively assessing biological synchrony. In contrast, the persistent enhancement of synchrony observed after disconnecting the interconnection serves as the key indicator of electrical interconnection–induced functional plasticity.

Furthermore, to provide a dynamical explanation of the synchronization improvement mechanism induced by the programmable electrical interconnection present in our experiments, we developed the EI-WCM. The coupling term introduced in this model effectively accounts for the enhanced information transfer between DCNNs, qualitatively reproducing the synchronization enhancement phenomena observed experimentally. The development of this dynamical model establishes a novel theoretical framework for investigating the plasticity inherent in neuronal networks.

Despite these advances, several limitations should be acknowledged. First, the EI-WCM represents a simplified phenomenological description, and its correspondence to the biophysical mechanisms at the single-neuron level remains to be further explored. This model is intended to provide a semi-quantitative, conceptual framework for illustrating how artificial electrical interconnection can reshape the overall dynamics of neuronal networks, rather than to infer the specific mechanisms of individual neurons or the underlying cellular composition. Future work could employ systematic immunolabeling and cell-type identification methods (e.g., GABA, glutamate, MAP2, NeuN) to quantitatively characterize the cellular composition of the cultured neuronal networks, thereby providing a more in-depth validation of the quantitative relationship between model parameters and specific cellular composition. Second, the current microsystem is limited in terms of recording channel count and functional extensibility. Future work will focus on increasing the number of recording channels and integrating electrical stimulation capabilities, enabling the construction of more complex network architectures and systematic investigation of neuroplasticity mechanisms under the combined influence of electrical interconnection and electrical stimulation.

Conclusion

In summary, we developed a complete integrated microsystem that enables programmable electrical interconnection and multichannel neuronal signal recording for DCNNs. Experimental results demonstrate that electrical interconnection significantly enhances firing synchrony and phase coherence between DCNNs, and that a sustained residual effect persists following disconnection, indicating that electrical interconnection may induce alterations in the functional state of neuronal networks and thereby modulate network plasticity.

Furthermore, the proposed EI-WCM dynamical model provides a theoretical explanation for the mechanisms underlying electrically induced synchronization enhancement and residual effects, offering insight into how artificial electrical coupling modulates coordinated network dynamics. Together, these results establish an integrated experimental and theoretical framework for investigating functional connectivity, cooperative dynamics, and plasticity in in vitro neuronal networks, and lay the groundwork for the future studies in brain-computer interface, neural rehabilitation, and synthetic biological intelligence.