Open access peer-reviewed chapter - ONLINE FIRST

Neurons on Microelectrode Arrays and In Vitro Electrophysiological Data Analysis

Written By

Andrey Vinogradov, Laura Ylä-Outinen, Susanna Narkilahti and Emre Kapucu

Submitted: 24 April 2025 Reviewed: 25 April 2025 Published: 01 June 2025

DOI: 10.5772/intechopen.1010742

Emerging Technologies in Computational Cognitive Neuroscience IntechOpen
Emerging Technologies in Computational Cognitive Neuroscience Edited by Vijayalakshmi Kakulapati

From the Edited Volume

Emerging Technologies in Computational Cognitive Neuroscience [Working Title]

Prof. Vijayalakshmi Kakulapati

Chapter metrics overview

42 Chapter Downloads

View Full Metrics

Abstract

Microelectrode array (MEA) technology has been used in functional neuronal network studies for decades, enabling the analysis of extracellular neuronal activity in both in vivo and in vitro. The proper interpretation of MEA recordings relies on effective neuronal signal processing and analysis. A key challenge in the field is accurate detection of extracellular neuronal activity (spikes) and identifying their occurrence in dense temporal clusters known as bursts. These bursts can be recorded by individual electrodes (single-channel bursts) or across multiple MEA electrodes (network bursts), adding substantial complexity to the analysis. This chapter provides a comprehensive overview of MEA signal analysis techniques, including methods for spike detection, burst identification, and the assessment of functional connectivity within neuronal cultures. We also include the studies involving the multi-compartment MEAs in the context of network activity and connectivity assessment, since such MEA setups offer a spatiotemporal depth in neural network interactions in vitro, facilitating more thorough insights into neurodevelopment and disease modeling. By integrating various analysis methods, the chapter offers a framework for understanding the complex dynamics of neuronal networks and their functional connectivity, making it a valuable resource for researchers using MEA technology for analyzing neuronal electrophysiology.

Keywords

  • burst
  • EAP
  • extracellular action potential
  • extracellular electrophysiology
  • functional connectivity
  • MEA
  • microelectrode array
  • network burst
  • neuron
  • signal analysis
  • spike

1. Introduction

The neurons belong to electrically excitable cells, as their membrane is capable of rapid impulse conduction: a transient voltage perturbation—the action potential, also called a spike. The brain’s remarkably complex function arises from the electrical activity of billions of neurons. The brain encodes information through the frequencies and timing patterns of action potentials generated by neuronal populations, which is rather difficult to study non-invasively [1, 2, 3].

The electrical properties of neuronal activity allow for distant acquisition through microelectrodes [4, 5]. However, accessing the micro- and mesoscale electrophysiological dynamics of the human brain in vivo presents significant challenges. On the other hand, these dynamics are effectively studied using animal models and in vitro human pluripotent stem cell-derived neuronal cultures by utilizing microelectrode arrays (MEAs) to record extracellular neuronal electrophysiological activity in both in vivo and in vitro settings [4, 5, 6, 7].

Unlike intracellular recordings, MEAs capture activity from multiple neurons across several locations simultaneously [4, 5, 8, 9, 10, 11], offering insights into network function and connectivity, which are crucial properties of neuronal networks. The obtained information about electrophysiological activity and functional interactions of the neurons serves as a proxy measure of synaptic transmissions [11]. Moreover, MEAs offer a non-invasive method to observe the electrical activity of the neuronal cells, thus allowing sequential or time-lapse follow-up of the same network. This approach is especially useful in developmental studies as well as mid- or long-term exposure experiments [3, 12, 13].

In vivo studies have the advantage of investigating neuronal networks’ dynamics in their actual physiological environment; however, in vitro approaches offer a more controlled setting for analyzing interactions from the subcellular to the network level. Moreover, the in vitro application of MEAs is particularly valuable due to their non-invasive nature compared to the patch clamp technique. These capabilities facilitate the study of developmental processes, neurodegeneration, and long-term pharmacological effects [13, 14, 15]. A key advantage of in vitro functional research using MEAs is the ability to efficiently integrate pharmacological treatments, electrical stimulation, fluorescent labeling, and targeted optogenetic activation to enhance experimental outcomes [4, 11, 16].

MEA analysis of in vitro neuronal cultures, ranging in complexity from single to multiple cell types and from 2D to 3D network structures, is a powerful tool for studying neuronal dynamics by means of electrophysiology. Electrophysiological recordings of neuronal populations with MEAs rely on electrodes that vary in size, density, and spatial distribution. Standard in vitro MEAs consist of multiple electrodes embedded at the bottom of the culture wells, where neuronal cells are plated. This setup enables mesoscale electrophysiological recordings, allowing researchers to analyze neuronal activity at the population level. Each electrode detects superimposed signals from nearby neurons [4]. Two examples of the standard commercially available MEA plates produced by Axion Biosystems are shown in Figure 1A. Recently, so-called high-density (HD) MEAs have also been utilized widely. By reducing the electrode size and increasing the density, hence obtaining the ability to record thousands of electrodes, HD MEAs are advantageous for measuring subcellular neuronal activities. It is an important achievement toward microscale electrophysiology with the possibility to capture axonal signaling and spatially track the signal propagation [20, 21].

Figure 1.

A. Schemes of well and electrode alignment of Axion MEA well plates. CytoView MEA 12 has 64 electrodes per well, and CytoView MEA 48 has 16 electrodes per well. Modified from Ref. [17]. B. Custom MEMO MEA plate components and scheme of its three compartments with microtunnels and electrode layout. Modified from Ref. [18]. C. (Left) Immunofluorescent images show neuronal axons penetrating the microtunnels. (Right) Zoomed schematic of a central part of the MEMO MEA plate. The black dots represent the electrodes, and the gray lines between compartments depict the microtunnels designed to enable axonal connections between three neuronal cultures. Each compartment has 24 electrodes. An additional eight electrodes per compartment are located near the entries of the microtunnels. Modified from Ref. [19].

In vitro electrophysiological experiments require stable and controlled environmental conditions to ensure neuronal viability and reproducible signal acquisition. This necessity influences both the design and practical use of MEA systems. Notably, neuronal cultures must remain in an incubator to maintain stable temperature, humidity, and gas levels essential for their survival. Consequently, MEA plates are typically designed to be portable, while the remaining hardware components are integrated into an MEA headstage. The headstage is connected to a computer that runs MEA recording software. During recording sessions, the MEA plate is temporarily positioned on the headstage [4, 22]. As an example, the components of the MEA2100-System produced by Multi Channel Systems MCS GmbH [23] are shown in Figure 2A. The proper temperature and gas environment can be maintained with integrated commercial solutions [24, 25] or by in-house developed designs [2627]. Some of the most recent commercial solutions have the option to operate the system continuously in an incubator [28].

Figure 2.

A. Components of MEA2100-System. Modified from Ref. [23]. B. The block scheme illustrates the main contributors to the noise of the MEA system. Modified from Ref. [4].

Novel culturing paradigms also impose another state-of-the-art subtype of in vitro MEA devices, multicompartmental MEAs. Such MEAs customized by different labs allow the study of the interactions between separated neuronal networks [18, 29, 30, 31, 32, 33, 34, 35, 36, 37], in some cases representing interconnectivities between different brain regions [36] or innervation to other cell types [38] as an advancement over cocultures located in the same compartments [39]. Some commercial solutions also exist for multi-compartmentalized MEAs [40]. An example of the custom three-compartment microfluidic MEMO MEA plate [18, 34, 41] is depicted in Figure 1BC. The MEMO MEA plates are compatible with the MEA2100-System produced by Multi Channel Systems MCS GmbH.

This chapter focuses on the emerging use of in vitro neuronal cultures on MEAs and methods for analyzing their recorded signals. Precise analysis of recorded electrophysiological data is fundamental for research on neuronal development, brain injury, disease modeling, and pharmacological studies. Gaining insights into neuronal interactions will help deepen our understanding of neural circuit function in the brain.

Advertisement

2. Microelectrode array signals

When an electrode is positioned close enough to a spiking neuron, the recorded extracellular voltage signal typically exhibits a distinct and easily recognizable action potential signature. Since extracellular electrodes do not disrupt neural activity, these recordings offer an accurate and impartial representation of the activity within intact neural circuits [11]. Electrophysiological research with in vitro MEAs typically involves culturing cells in MEA plates on top of the electrode grid to enable repeated electrophysiological recordings. These plates contain one or more wells with embedded electrodes at the bottom, onto which neuronal cells are seeded. The wells are filled with a cell maintenance medium that provides essential nutrients and chemicals to sustain cell viability [4]. This setup supports functional neuronal cultures in either 2D [7] or 3D [42, 43] configurations, where cells are distributed across the surface of the recording microelectrodes, which capture extracellular electric potential fluctuations [44]. Each microelectrode captures variations in the extracellular electric field within its local environment. Typically, these electrodes register activity from sources located within a range of up to 100 μm [22, 45]; however, the signal amplitude diminishes as the distance between the source and the electrode increases [46]. Additionally, an electrode’s sensitivity is influenced by multiple factors, including the properties of the surrounding medium [47], the electrode material, and the characteristics of other hardware components of the MEA system [4].

The voltage signal detected by an MEA electrode results from the linear summation of contributions from current sources, with each contribution weighed inversely to its distance from the electrode. Calculating the voltage signal at the electrode necessitates two key assumptions: (1) neurons can be modeled as a spatial distribution of ion currents and sinks, approximated as monopolar point sources, and (2) the surrounding medium is homogeneous, isotropic, and unbounded. Consequently, the potential Ve (V) at the electrode location, influenced by N point sources, is determined using the following equation:

Ve=14πσnNInrnE1

where In is the amplitude (A) of a point source n, located at distance r (m) from the electrode; σ stands for the medium conductivity (S/m). The point source approximation for ion currents and sinks assumes that the volumes of their source regions are significantly smaller than their distances from the recording electrode, enabling them to be represented as points rather than complex shapes [4, 44, 48, 49, 50].

The recorded MEA signal incorporates multiple noise sources that degrade the recorded signal. Non-biological noise sources include 1/f or 1/f2 electrode-electrolyte interface noise and thermal noise. Moreover, digital recordings of the voltage signal and electrical stimulation require a connection between the MEA plate electrodes and various hardware, including amplifiers, filters, a digitizer, data transmission cables, a stimulator, and power lines. All these components of MEA systems contribute further to device noise. Signal distortions may also result from phase noise and lossy data compression methods [4, 51, 52, 53]. A block scheme of noise contributors in MEA recordings is shown in Figure 2B.

The biological components of the signal include the spikes and less prominent local field potential (LFP) fluctuations [5]. The LFP component contains hard-to-detect extracellular electric field changes, which originate from subthreshold transmembrane activities of neurons and distant spiking activity. The LFP resides in the lower part of the spectrum of the signal. However, a part of the LFP component is usually considered as biological noise. In such cases, the desired signal and the noise have overlapping spectral characteristics [4, 50, 52, 54, 55, 56]. Although the LFP component is often challenging to analyze, it can still provide valuable insights. In vivo MEA recordings from the human primary motor cortex [57] and monkey motor cortices [58, 59] have demonstrated that the LFP carries rich movement-related information, which can be extracted even in the absence of detectable spike activity. Additionally, LFP analysis can reveal new insights into postsynaptic potentials (PSPs) in neuronal cultures [22].

In contrast, spikes buried in the MEA signal serve as precise indicators of neuronal action potentials. The initial step in processing a raw MEA signal typically involves pre-filtering to distinguish between the LFP and spiking components. The spike detection typically begins with the band-pass filtering within a frequency range of approximately 300–3000 Hz, whereas LFP analysis involves low-pass filtering with cutoff frequencies below 300 Hz [5, 56, 60, 61, 62].

Next, we will consider spike detection and consecutive analyses based on the spiking activity.

Advertisement

3. Microelectrode array signal analysis

Typically, the MEA signal analysis includes the detection of neuronal spikes in pre-filtered data, followed by the identification of spike-based features, such as firing rate (Section 3.1) or single-channel bursts and network bursts (NBs; Section 3.2). These analysis steps are visualized in Figure 3. Moreover, additional investigation of functional connectivity measures is often applied to map the functional topology of the networks [63, 64, 65] (Section 3.3). The functional connectivity evaluation methods may utilize both raw MEA data and the detected spikes [63, 64, 65]. Each analysis step is associated with the derivation of numerous feature-specific output parameters, which are then used for scientific inference.

Figure 3.

A. Exemplary signal recorded by a single MEA electrode. The pre-filtered waveform is shown in gray; detected spikes are highlighted with red circles; and periods identified as single-channel bursts are marked by black lines superimposed on the middle of the waveform. B. The raster plot demonstrates a segment of data recorded from 16 electrodes simultaneously. The spike train from each electrode is represented as a row of colored bins in time. Each row corresponds to the spike train of a separate electrode. The rows from all electrodes are stacked on top of each other. The bin color encodes firing rate within the bin: white indicates 0 Hz, and black indicates 40 Hz. Green and red lines denote the start and end of detected NBs, respectively. The data were acquired with the Axion CytoView MEA 48 plate, which has 16 electrodes per well. Modified from Ref. [19].

3.1 Spike detection and sorting

A single neuronal excitation, the extracellular action potential or the spike, is a key feature to extract from the pre-filtered signal of each electrode in an MEA [66], and it can be characterized by two main features: an amplitude peak and a biphasic shape [11]. These two characteristics are essential for designing spike detection algorithms. The segment of the signal induced by a spike in extracellular recordings typically lasts about 3 ms [11].

The typical shape of an extracellularly acquired action potential is considered biphasic. The action potential begins with a rapid influx of Na+ ions, forming a current sink that generates a prominent negative deflection in the extracellular action potential. This is followed by a slower efflux of K+ ions, creating a current source and leading to a smaller positive deflection [4]. However, the final signal shape captured by an electrode tends to depend on its position relative to the neuron’s morphology: with large negative spikes recorded at the perisomatic area and positive spikes at the dendritic area, generated by balancing outward return current as the net membrane current summed over the entire neuron is always zero [4, 45, 67, 68]. Thus, it is rather difficult to conclude which polarity of the spike waveform should be considered characteristic.

One of the simplest spike detection methods involves thresholding based on the amplitude peaks: either negative or both negative and positive. Thresholds can be either static, defined by the user, or adaptive, computed from the data [69, 70]. Static thresholding struggles with the inherent variability in MEA data since noise levels can differ across channels and change over time. To overcome this limitation, adaptive thresholding techniques were introduced. The adaptive threshold is typically set in relation to the noise level, which is estimated from segments of the raw signal that are free of spikes [11]. The most common adaptive method, known as absolute amplitude-based thresholding (ABS), calculates the threshold by estimating the noise’s standard deviation (SD), multiplied by a threshold calculation factor (TCF). This estimation uses the median of the absolute values of the signal to prevent excessively high thresholds [69]. The block diagram of the ABS spike detection method is shown in Figure 4 below.

Figure 4.

Block diagram of the ABS algorithm.

The performance of this algorithm still decreases significantly when the signal-to-noise ratio (SNR) is low [46]. Despite this, the amplitude-based thresholding algorithms remain widely used in both in vivo and in vitro MEA recordings [60, 62, 71, 72, 73, 74, 75, 76]. Another amplitude-based method is Precision Timing Spike Detection (PTSD), which detects spikes by comparing the difference between the local maximum and minimum (RMM) within a predefined time window denoted as a peak lifetime period (PLP). If the peak-to-peak amplitude exceeds the Spike Detection Differential Threshold (SDDT), the spike is identified. The SDDT is determined by multiplying the signal’s SD by a specific TCF [77, 78]. This method captures both primary spike features: the transient amplitude peak and biphasic shape. The block diagram of the PTSD algorithm is provided in Figure 5 below.

Figure 5.

Block diagram of the PTSD algorithm.

A different category of spike detection techniques focuses on transforming the signal into a specialized variable, known as the transformation variable or indicator signal, which is subsequently subjected to thresholding. This variable is designed to emphasize the key characteristics of the spikes and separate them from the noise. For instance, the Multiresolution Teager Energy Operator algorithm (MTEO) utilizes the Teager Energy Operator (TEO) concept, also known as Nonlinear Energy Operator (NEO), modified with a resolution parameter to detect various possible widths of action potential amplitude peaks. The discrete TEO is computed by finding the difference between the squared signal value at a given time point and the product of its adjacent samples. The algorithm highlights the transient amplitude changes in the signal. The corresponding equation is provided below:

Ψxn=x2nxnkxn+k,E2

where xn is the input signal, and k is the resolution parameter to reflect the width of the spikes to detect. The authors proposed utilizing multiple k values to generate several transformation variables, which were subsequently smoothed and processed through a maximum filter, which picked the maximum value from the three smoothed TEO outputs for every time instance, and thresholded it [54, 79]. The block diagram of the MTEO algorithm is presented in Figure 6 below.

Figure 6.

Block diagram of the MTEO algorithm. Modified from Ref. [54].

On the other hand, the transformation variable can be derived through various types of wavelet decomposition applied to the input signal. A wavelet is the probing function of finite duration, which can be described as a short-lived waveform or localized pulse [80, 81]. Wavelet basis functions typically exhibit biphasic, triphasic, or multiphasic waveforms with compact support, making them similar to action potentials. These functions are translated or slid over the signal to localize the time instants where the signal resembles the wavelet shape or its dilated versions in the time domain, though the exact waveform of the extracellular action potential is not known. The obtained output exhibits local maxima at points where the input signal closely matches the shape of the wavelet function [82, 83]. The wavelet decomposition is usually implemented with a pair of wavelets: a mother wavelet, also called the wavelet function, and a father wavelet, also called the scaling function. The mother wavelet is designed to capture a detailed (or high-frequency) profile of the input, while the father wavelet emphasizes the low-frequency approximation of the input signal. The resulting coefficients are denoted as detail coefficients and approximation coefficients, respectively. This multiresolution approach allows for efficient time-frequency analysis, particularly useful for non-stationary signals like MEA data [84]. A classical multilevel discrete wavelet transform (DWT) shifts to the next decomposition level, also called a scale, by downsampling the obtained coefficients and decomposing them with the same mother and father wavelets. Due to the downsampling, the classical DWT is not a time-invariant transform, which means that the DWT of a signal and the DWT of the same signal shifted in time do not differ in the shift of the coefficients only [46, 84]. A stationary wavelet transform is a time-invariant version of the DWT. Instead of downsampling the obtained coefficients, it moves to the next decomposition level by upsampling the mother and father wavelets [46, 82, 84].

Thus, in spike detection, wavelet-based methods are specifically designed to spotlight the distinctive shapes of spikes within an input signal. The previously published Stationary Wavelet Transform (SWT)-based spike detection algorithm involves denoising wavelet detail coefficients using MAD-based thresholding of each sub-band [82]. Since the signal’s energy is concentrated within a limited number of coefficients while noise is distributed across many coefficients, such thresholding helps eliminate low-energy time-scale points across all scales [82]. The block diagram of the SWT algorithm is presented in Figure 7 below.

Figure 7.

Block diagram of the SWT algorithm. Modified from Ref. [82].

Another technique, the Wavelet-based Teager Energy Operator algorithm (WTEO), combines multilevel DWT using only the father wavelet with the application of the TEO to the resulting approximation coefficients of each sub-band [83]. The TEO outputs are combined with the maximum filter and then thresholded using an SD-based threshold [46, 83]. The block diagram of the WTEO algorithm is presented in Figure 8 below.

Figure 8.

Block diagram of the WTEO algorithm. Adopted from Refs. [46, 83].

Building on a similar principle, the Stationary Wavelet Transform-based Teager Energy Operator method (SWTTEO) utilizes the stationary wavelet transform instead of traditional wavelet decomposition [46]. The block diagram of the SWTTEO algorithm is presented in Figure 9 below.

Figure 9.

Block diagram of the SWTTEO algorithm. Modified from Ref. [46].

However, wavelet-based methods rely on the selection of a suitable wavelet basis—a pair of wavelets whose shape should roughly match that of the spikes, despite the actual spike morphology being unknown. Several alternative advanced methods were developed to avoid using such predefined spike shape characteristics. In those, the spike templates were derived from the MEA signal using higher-order statistical measures and subsequently convolved with the raw signal for spike detection [85, 86]. Alternatively, some recent approaches employed supervised machine learning (ML) techniques for spike detection, such as pre-trained convolutional neural networks (CNNs). These models were trained using a combination of hand-curated datasets and synthetic data, which significantly enhanced detection accuracy. Another strategy involves conducting a preliminary analysis to establish a “ground truth” using conventional spike detection methods, which is then used to train the model [87, 88, 89].

The spike detection algorithms are typically assessed on synthetic datasets crafted to mimic real MEA recordings. However, evaluations using different synthetic datasets often yield markedly different performance ratings for the same algorithms [46, 66, 86].

In certain experimental setups, the spike detection is followed by spike sorting to further analyze the detected spikes. For example, analyzing single-neuron activity requires accurately assigning each detected spike to its originating neuron [11]. This process involves extracting spike events as signal segments spanning over several tens of samples corresponding to the putative spike duration. These segments are then projected into a lower-dimensional space using techniques such as principal component analysis (PCA) or shape-based metrics such as peak-to-peak amplitude or spike width [90, 91]. Clustering algorithms are then applied to sort the spikes. Alternatively, the extracted waveforms can be decomposed using wavelet transforms, with selected wavelet coefficients serving as inputs for further clustering [69, 92]. The challenges imposed on spike sorting techniques arise from various features of MEA recordings, such as the interference of overlapping waveforms, spike shape changes during bursts, drift of neurons relative to their initial positions on the electrodes, and the unknown number of neurons within the electrode’s receptive field [11]. In conventional MEAs, many neurons usually contribute to spikes at the electrode, while in HD MEA setups, a single spike is clearly reflected in several electrode signals [11].

The general logic of the spike detection is rather consistent across standard MEAs, HD MEAs, and multicompartmental MEA setups. However, the choice of a method, its parameterization, and the interpretation of results should reflect the specific characteristics of the MEA plate used, such as electrode size and density, cell type-related activity features, and other aspects of the experimental design.

3.2 Bursts

Bursts reflect network dynamics and are typically characterized by the simultaneous occurrence of frequent neuronal activity. Bursts imply compressed information transfer between neurons, consisting of a series of tightly packed single spikes. Bursts are recorded from single or several electrodes of an MEA and defined as single-channel bursts and NBs, respectively [11, 93, 94]. Although bursting is a key feature commonly used to evaluate neuronal network activity, a clear and universally accepted formal definition of a burst is still lacking in the field after more than 50 years of MEA research. As a result, burst detection strategies within the scientific community remain diverse and are often tailored to specific data at hand [11, 95, 96].

Bursts are believed to play a role in long-term potentiation and other processes related to synaptic transmission. This bursting activity contributes to a consistent and information-rich mode of neuronal communication [11, 93, 97, 98, 99]. In vitro bursts are considered to represent information packages similar to those observed in vivo neuronal networks [100]. Importantly, network bursts are characterized by their broader propagation within the network—they travel from cell to cell and across different regions of the neuronal network. The appearance of NBs in MEA data indicates the functional maturation of the recorded neuronal populations. NBs can be used to study the interconnectivity of the networks as well as connectivity velocity. Moreover, they represent the desired activity state for pharmacological experiments [18].

The lack of definition of bursts, together with unavailable ground truth MEA data, imposes serious constraints on the performance evaluation of the single-channel burst and NB detection algorithms. The efficiency assessment of the algorithms is usually achieved with synthetic data sets and visual performance estimation [11, 94, 99, 101].

3.2.1 Single-channel burst detection methods

Single-channel burst firing determines distinct developmental patterns in neuronal populations during MEA recordings [7, 98]. Numerous algorithms have been developed for detecting single-channel bursts [99], with many relying on interspike interval (ISI) characteristics within the recorded spike trains. The ISI-based approach may be further constrained with an additional criterion for a minimum number of spikes per burst to avoid detecting very brief single-channel bursts. The ISI threshold used for ISI-based detection can either be a fixed value [98] or be determined from the distribution of ISIs observed in the recorded signal [93, 101].

For example, the adaptive single-channel burst detection algorithm introduced by Pasquale and colleagues [93] constructed the logarithmic histogram of all ISIs of an input spike train. Then all peaks of the histogram were identified. The set of peaks was checked to contain the putative intra-burst peak: a maximum peak with an ISI value below the predefined intra-burst peak threshold. This peak was considered to represent the ISIs of intra-burst spikes. If no such peak was found, the data was marked as non-bursting. If the intra-burst peak was present, the algorithm evaluated the local minima between this peak and all other peaks that corresponded to longer ISI values. The evaluation criterion for the candidate minima was denoted as a void parameter, which reflected the quality of the peak separation and spanned from 0 when the peaks were not separated to 1 when the peaks were perfectly separated. The ISI value of the candidate minimum, which void parameter exceeded a predefined void threshold, was then selected as an ISI threshold, ISIth, for the single-channel burst detection. If the previously mentioned intra-burst peak threshold appeared lower than the obtained ISIth, both thresholds were used for detecting burst cores and extending their boundaries, correspondingly [93]. An example of the logarithmic histogram with highlighted peaks, intra-burst peak threshold, and derived ISI threshold ISIth is depicted in Figure 10A.

Figure 10.

Two ISI-based single-channel burst detection methods. A. (Left) Logarithmic distribution of ISI with the red star-marked ISI threshold ISIth that was selected at the local minima based on the peaks’ separation criterion. The corresponding peaks are marked with the black star symbols. An intra-burst peak threshold of 100 ms is depicted with the red dashed line. The horizontal axis of the histogram is logarithmic. (Right) Detection outcome: an exemplary spike train with the red spikes classified as belonging to the single-channel burst and the blue spikes marked as the out-burst spikes. The spike train is presented as a point process in time. Reproduced from Ref. [93]. B. Definition of the ISI threshold using the maximum value of the CMA curve of the ISI histogram. The maximum value of CMAm was reached at the bin with ISI value xm, and the ISI threshold xt was defined as the ISI value of the bin that corresponded to the αCMAm value. The defined ISI threshold for burst detection is marked with the red dashed vertical line. The ISI histogram is depicted with gray bars. The axes of the histogram are linear. Reproduced from Ref. [101].

Alternatively, the algorithm created by Kapucu and colleagues [101] computed a non-logarithmic histogram of ISIs and then derived the cumulative moving average (CMA) value for each bin. The maximum CMA value, CMAm, was then used to identify the bin whose CMA value was closest to the product αCMAm, where α denoted the tolerance parameter to identify the burst-related spikes. The ISI value xt of that bin was then selected as an ISI threshold for the single-channel burst detection. The tolerance parameter α may vary from 0 to 1 [101]. The non-logarithmic ISI histogram with the overlayed CMA curve and the defined ISI threshold is depicted in Figure 10B. Both described algorithms process the ISI histogram to determine the ISI threshold that is assumed to separate the in-burst and out-burst firing regimes most efficiently. The data with the developed bursting activity tend to produce histograms with an evident peak corresponding to shorter ISI values on the left side of the histogram.

On the other hand, there exists a family of so-called surprise-based methods for single-channel burst detection. These algorithms use statistical analysis to distinguish periods of bursting from baseline neuronal firing. Initially, a suitable assumption about the distribution of baseline spike firing times is developed. The bursting regimes are expected to produce evident deviation from this distribution, which is used to identify the single-channel bursts [96, 102, 103].

The spike timing-based detection of the single-channel bursts may, in principle, be applied to any MEA setup. However, the correct implementation of a particular algorithm relies on the general understanding of the origins of the recorded data. For example, with the standard commercial MEAs, it is impossible to infer directly whether the recorded single-channel bursting originates from one bursting neuron or from several active cells near the electrode [11]. Only the sorting of the in-burst spikes may provide an answer in this case.

3.2.2 Network burst detection methods

NB activity is commonly linked to network maturation, with its peak generally aligning with the period of highest synaptic density within a network [11, 93, 94, 104, 105, 106]. Numerous established methods for detecting NBs have been proposed [93, 94, 106, 107, 108, 109]. Based on their underlying detection principles, these methods can be categorized into ISI-based approaches [93, 94, 108], population firing rate (PFR)-based approaches [106, 107], and hybrid methods [109]. PFR integrates the time-varying binned spike firing rates from all involved MEA electrodes into a common firing rate signal.

Some of the listed techniques extrapolate single-channel burst detection results to the network level [93, 108]. Pasquale and colleagues [93] extended their single-channel burst detection logic to the NB detection level. After the single-channel bursts were detected in the signals of all involved electrodes, they pooled the timings of all these bursts into the cumulative burst event train. The logarithmic inter-burst interval (IBI) histogram was constructed using that. The histogram was used to identify the IBI threshold and to detect NBs within the cumulative burst event train. Finally, the criterion for minimum participating channels was applied to the detected NBs to preserve only the NBs with contributions from multiple electrodes [93]. Similarly, Välkki and colleagues [108] identified the single-channel bursts among MEA electrodes using the algorithm by Kapucu and colleagues [101] described above and evaluated their synchronization. The total NB signal was obtained by summing bursting behaviors over channels, which was then thresholded in accordance with the minimum participating channels criterion [108].

Others treat the NB detection as an independent process [94, 106, 107, 109]. For example, Bakkum and colleagues [94] developed a logarithmic ISIN histogram-based method. First, the spike trains from all involved MEA electrodes were combined and sorted into a joint spike train to calculate the logarithmic ISIN histogram of ISIs between N consecutive spikes. Again, the histogram exhibiting two separate peaks was assumed to represent two different firing regimes: in this case, the spikes that belong to NBs and the spikes outside the network firing regime, respectively. Therefore, the threshold for detecting NBs was selected at the minimum point between these peaks in the histogram [94]. Hedrich and colleagues [107] presented the PFR average-based method. The spike trains from all involved MEA electrodes were summed into 5 ms time bins to calculate the PFR, which was then smoothed by a Gaussian kernel with a 100 ms SD. The intervals with a PFR exceeding the slowly varying 1-second PFR average were selected as NB candidates. The candidates were accepted if they fit three criteria: their peak firing rate exceeded three SDs of the PFR, their peak firing rate exceeded 10% of the average of the top five peaks, and at least three electrodes contributed. At the last step, neighboring NBs with IBIs less than 200 ms were merged [107].

Matsuda and colleagues [109] combined ISI-thresholding with a form of PFR-thresholding. In their four-step method, the spike trains from all electrodes were pooled into a joint spike train, where the spikes separated by ISIs less than 4 ms were assigned to NB candidates. NBs with less than 20 spikes were discarded. Neighboring NBs with IBI less than 60 ms were merged. At the last step, only the NBs that accommodate more than 3000 spikes were considered valid [109]. The method involved a number of fixed parameters, which usually limit the adaptivity of the approach to highly variable MEA data.

The detection of NB activity provides an advanced measure of neuronal network functioning. Figure 11 shows the gradual drop in NB counts in a kainic acid (KA)-exposed culture compared to its baseline activity. The data was acquired with an Axion CytoView MEA 48 plate, which has 16 electrodes per well.

Figure 11.

Detected NB counts at baseline and after KA exposure. The raster plots of KA-induced perturbation of NB activity. The upper plot shows baseline spiking activity in a well with detected NBs. The green and red vertical lines represent the starting and ending times of NBs, respectively. The lower plot demonstrates disruption of the NBs into spread tonic firing. No NBs were detected. Modified from Ref. [110].

3.2.3 Detection of network bursts in multi-compartment setups

Application of the NB detection methods to multi-compartment MEA setups possesses additional challenges. While the listed methods are well-documented for standard MEA setups, they are not directly applicable to the data acquired with multi-compartment MEAs. The architecture of multi-compartment MEAs involves multiple potential synchronization levels, requiring specialized analysis tools to fully exploit the data they generate. Without such tailored analysis techniques, the advantages of multi-compartment MEAs over conventional single-network setups may be lost. Overall, the signal analysis for such systems should account for both the spatial separation and the interactions between the interconnected networks to ensure accurate assessment [41].

Previous studies involving multi-compartment MEA data analysis often considered synchronous bursting as a key feature [30, 31, 32, 33, 41]. Typically, these studies first identified local NBs within individual compartments using conventional methods, followed by evaluating the temporal alignment of these NBs within the entire compartmentalized network [30, 31, 32, 41]. For example, a recent work by Vinogradov and colleagues [41] defined three levels of synchronization within a three-compartment network of neurons (Figure 1BC) recorded with the MEMO MEA plate. The initial detection of NBs was performed separately in each compartment using a custom algorithm developed based on a previously published method [94], which was extended with additional parameters to reflect local network synchrony [41]. The locally detected NBs were then evaluated for temporal alignment to capture their global, circuitry-level synchronization [41]. Beyond the localized NBs occurring within each of the three individual compartments of the device, two additional, more complex synchronization patterns were identified and analyzed. An intermediate circuitry burst (ICB) was characterized as simultaneous activity in two out of three compartments. Furthermore, a circuitry burst (CB) refers to synchronized bursting that encompasses the entire network across all three compartments [41]. The analysis demonstrated that a local network within a single compartment of a MEMO MEA plate could not only generate local NBs but also contribute to both ICBs and CBs observed at the global network levels, as illustrated by the network in compartment B (Figure 12).

Figure 12.

Examples of NB, ICB, and CB detection. The detections are visualized on a combined raster plot of spiking activity in three compartments within a 150-second segment. The data was recorded with the MEMO MEA plate. The green and red vertical lines represent the starting and ending times of detected local NBs, respectively. The purple polygons at the top depict the time intervals of detected CBs involving all three compartments; the orange polygons indicate ICBs with two-compartment synchronization; and the yellow polygons indicate local NBs in a single (B) compartment only. The length of the polygons corresponds to the duration of the corresponding synchronization patterns. Reproduced with modification from Ref. [41].

3.3 Methods for functional connectivity analysis

Beyond analyzing synchronous bursting, various methods have been developed to assess functional connectivity and information propagation in neuronal cultures. One such approach is the spike time tiling coefficient (STTC), which measures the proportion of spike co-occurrences between electrode pairs to determine pairwise correlations in their spiking activity [64, 111]. Similarly, functional connectivity maps can be constructed by analyzing correlations of simultaneous spikes across electrodes [35]. Cross-correlation-based methods have also been employed on compartment-wise firing rates within the network to infer the direction of activity propagation by evaluating the latency of peak correlations [30] or implemented straightforwardly between electrode pairs [31]. An additional alternative approach, known as correlated spectral entropy (CorSE), evaluates functional connectivity by analyzing temporal changes in the power spectra of raw MEA signals. The method involves segmenting the signals into windows, calculating spectral entropies for each window, and then determining connectivity through cross-correlation of the resulting entropy signals [63].

In a work by Hyvärinen and colleagues [112], the rat cortical and human pluripotent stem cell (hPSC)-derived cultures were compared in terms of functional connectivity development using both the CorSE and the STTC methods. The data was acquired with Axion CytoView MEA 12 plates with 64 electrodes per well. Both algorithms indicated earlier onset of functionally interconnected networks in the rodent data [112]. Figure 13 provides the connectivity maps obtained using the CorSE algorithm and shows the dynamics of the average CorSE and STTC values for both culture types over the experimental timeline.

Figure 13.

Functional connectivity development analysis. A. (Up) Functional connectivity maps from an hPSC-derived network over 10 minutes of recording between days 21 and 42 on MEA and (Down) from a rat cortical network between days 14 and 35 on MEA. Each map presents the electrode layout of the Axion CytoView MEA 12 plate as an 8 × 8 array with detected functional connections depicted as lines. An arbitrary CorSE value of 0.7 was used as the connectivity strength threshold for the line plotting. B. Functional connectivity development of the hPSC-derived and rat cortical networks described with the average CorSE values over time. C. The same as described by the average STTC values. The average values were calculated from all electrode pairs for each well. Both the hPSC-derived and rat cortical network data consist of n = 12 networks per group; data are shown as Tukey box plots. A Mann-Whitney U-test was performed to compare the two groups at each time point. Statistical significances are marked as *p < 0.05, **p < 0.01, and ***p < 0.001. Adapted from Ref. [112].

In addition to the correlation-based methods, various event-based and information theory-based approaches have been developed to estimate functional connectivity and causal interactions. Event Synchrony (ES), introduced by Quiroga and colleagues, quantifies synchronization based on the timing proximity of discrete events such as spikes. It assigns synchrony scores based on the coincidence of spike times within a local temporal window, providing an intuitive, symmetric measure of synchrony between two spike trains [113].

Mutual Information (MI) is a widely used metric that quantifies statistical dependencies between signals by comparing the joint probability distribution of events to their marginal distributions [65, 114]. MI has been particularly useful for assessing nonlinear relationships in spike train data [65, 115].

Extending the concept of MI, Transfer Entropy (TE) accounts for the directionality of information flow by incorporating time-lagged dependencies, thus estimating the extent to which past activity in one signal predicts future activity in another [116, 117]. Since TE is asymmetric and captures both linear and nonlinear interactions, making it suitable for analyzing directed functional connectivity [117].

Similarly, Granger causality [118], or its extended models for the frequency domain and multivariate signals [119], evaluates directional influences by relying on autoregressive modeling. It assumes that if a causal relationship exists, the past values of one time series will significantly contribute to the prediction of another. Several works in the field applied the Granger causality method for MEA data [31, 120, 121].

Each of these approaches complements classical correlation-based techniques and can enhance functional network mapping, particularly when directionality and information flow are of interest.

Functional connectivity analysis methods are particularly valuable in the context of multi-compartment MEA systems. Several illustrative examples are listed here. Le Feber and colleagues [30] applied cross-correlation analysis to compartment-specific firing rates within defined time bins, using the latency of peak correlations to infer the direction of activity propagation within a custom two-compartment MEA platform [30]. In another custom two-compartment MEA configuration, DeMarse and colleagues [31] estimated functional connectivity by computing cross-correlation and conditional Granger causality metrics from spike train data across electrode pairs [31]. Likewise, Van de Wijdeven and colleagues [35] generated functional connectivity maps by analyzing the correlations of simultaneous spikes detected across electrodes [35]. The CorSE method was used for evaluation of KA-exposure effects on functional connectivity in a three-compartment setup using previously mentioned MEMO MEA plates [41].

Advertisement

4. Conclusions

MEA technology remains a powerful tool for investigating the functional dynamics of neuronal networks. Precise analysis of electrophysiological signals is crucial for in vitro MEA studies focused on development, disease modeling, and drug testing. In this chapter, we introduced basic steps of neuronal data analysis and provided the corresponding analysis methods. This chapter has reviewed the core methods and current challenges in MEA signal analysis, including the accurate detection of spikes and the identification of bursts, both at single-channel and network levels, and functional connectivity mapping.

By presenting a comprehensive overview of established and emerging analysis strategies, this chapter aims to support researchers in making informed choices when designing experiments and interpreting MEA data. Integrating these analytical approaches enables a deeper understanding of neuronal network behavior and promotes the advancement of in vitro models in neuroscience research.

Advertisement

Acknowledgments

The work was supported by the Facility of Electrophysiological Measurements (Faculty of Medicine and Health Technology, Tampere University). The authors also thank Biocenter Finland for the support of the Electrophysiological Measurements facility. This work was supported by the Research Council of Finland and EU H2020-FETOpen (PRIME project).

Advertisement

Conflict of interest

The authors declare no conflict of interest.

References

  1. 1. Cooper DC. The significance of action potential bursting in the brain reward circuit. Neurochemistry International. 2002;41:333-340. DOI: 10.1016/S0197-0186(02)00068-2
  2. 2. Emmenegger V, Obien MEJ, Franke F, Hierlemann A. Technologies to study action potential propagation with a focus on HD-MEAs. Frontiers in Cellular Neuroscience. 2019;13:159. DOI: 10.3389/fncel.2019.00159
  3. 3. Ylä-Outinen L, Tanskanen JMA, Kapucu FE, Hyysalo A, Hyttinen JAK, Narkilahti S. Advances in human stem cell-derived neuronal cell culturing and analysis. In: Chiappalone M, Pasquale V, Frega M, editors. Vitro Neuronal Networks: From Culturing Methods to Neuro-Technological Applications. Cham: Springer International Publishing; 2019. pp. 299-329. DOI: 10.1007/978-3-030-11135-9_13
  4. 4. Obien MEJ, Deligkaris K, Bullmann T, Bakkum DJ, Frey U. Revealing neuronal function through microelectrode array recordings. Frontiers in Neuroscience. 2015;8:423. DOI: 10.3389/fnins.2014.00423
  5. 5. Pelkonen A, Pistono C, Klecki P, Gómez-Budia M, Dougalis A, Konttinen H, et al. Functional characterization of human pluripotent stem cell-derived models of the brain with microelectrode arrays. Cells. 2022;11:106. DOI: 10.3390/cells11010106
  6. 6. Henze DA, Borhegyi Z, Csicsvari J, Mamiya A, Harris KD, Buzsáki G. Intracellular features predicted by extracellular recordings in the hippocampus In vivo. Journal of Neurophysiology. 2000;84:390-400. DOI: 10.1152/jn.2000.84.1.390
  7. 7. Heikkilä TJ, Ylä-Outinen L, Tanskanen JMA, Lappalainen RS, Skottman H, Suuronen R, et al. Human embryonic stem cell-derived neuronal cells form spontaneously active neuronal networks in vitro. Experimental Neurology. 2009;218:109-116. DOI: 10.1016/j.expneurol.2009.04.011
  8. 8. Thomas CA, Springer PA, Loeb GE, Berwald-Netter Y, Okun LM. A miniature microelectrode array to monitor the bioelectric activity of cultured cells. Experimental Cell Research. 1972;74:61-66. DOI: 10.1016/0014-4827(72)90481-8
  9. 9. Gross GW. Simultaneous single unit recording in vitro with a photoetched laser deinsulated gold multimicroelectrode surface. IEEE Transactions on Biomedical Engineering. 1979;BME-26:273-279. DOI: 10.1109/TBME.1979.326402
  10. 10. Pine J. Recording action potentials from cultured neurons with extracellular microcircuit electrodes. Journal of Neuroscience Methods. 1980;2:19-31. DOI: 10.1016/0165-0270(80)90042-4
  11. 11. Michela C, Valentina P, Monica F. In Vitro Neuronal Networks: From Culturing Methods to Neuro-Technological Applications. 1st ed. Cham: Springer International Publishing; 2019. DOI: 10.1007/978-3-030-11135-9
  12. 12. Shimba K, Sakai K, Iida S, Kotani K, Jimbo Y. Long-term developmental process of the human cortex revealed In vitro by axon-targeted recording using a microtunnel-augmented microelectrode Array. IEEE Transactions on Biomedical Engineering. 2019;66:2538-2545. DOI: 10.1109/TBME.2019.2891310
  13. 13. Frank CL, Brown JP, Wallace K, Mundy WR, Shafer TJ. From the cover: Developmental neurotoxicants disrupt activity in cortical networks on microelectrode arrays: Results of screening 86 compounds during neural network formation. Toxicological Sciences. 2017;160:121-135. DOI: 10.1093/toxsci/kfx169
  14. 14. Seidel D, Jahnke H-G, Englich B, Girard M, Robitzki AA. In vitro field potential monitoring on a multi-microelectrode array for the electrophysiological long-term screening of neural stem cell maturation. Analyst. 2017;142:1929-1937. DOI: 10.1039/C6AN02713J
  15. 15. Schulte S, Gries M, Christmann A, Schäfer K-H. Using multielectrode arrays to investigate neurodegenerative effects of the amyloid-beta peptide. Bioelectronic Medicine. 2021;7:15. DOI: 10.1186/s42234-021-00078-4
  16. 16. Zhang J, Laiwalla F, Kim JA, Urabe H, Van Wagenen R, Song Y-K, et al. Integrated device for optical stimulation and spatiotemporal electrical recording of neural activity in light-sensitized brain tissue. Journal of Neural Engineering. 2009;6:055007. DOI: 10.1088/1741-2560/6/5/055007
  17. 17. Kapucu FE, Vinogradov A, Hyvärinen T, Ylä-Outinen L, Narkilahti S. Comparative microelectrode array data of the functional development of hPSC-derived and rat neuronal networks. Scientific Data. 2022;9:120. DOI: 10.1038/s41597-022-01242-4
  18. 18. Pelkonen A, Mzezewa R, Sukki L, Ryynänen T, Kreutzer J, Hyvärinen T, et al. A modular brain-on-a-chip for modelling epileptic seizures with functionally connected human neuronal networks. Biosensors and Bioelectronics. 2020;168:112553. DOI: 10.1016/j.bios.2020.112553
  19. 19. Vinogradov A, Kapucu E, Narkilahti S. An analysis pipeline for signaling of human stem cell derived neurons in a microfluidics platform with an embedded microelectrode array. In: 2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC); Oct 31 - Nov 4, 2021. Mexico: IEEE (Institute of Electrical and Electronics Engineers); 2021
  20. 20. Müller J, Ballini M, Livi P, Chen Y, Radivojevic M, Shadmani A, et al. High-resolution CMOS MEA platform to study neurons at subcellular, cellular, and network levels. Lab on a Chip. 2015;15:2767-2780. DOI: 10.1039/C5LC00133A
  21. 21. Radivojevic M, Franke F, Altermatt M, Müller J, Hierlemann A, Bakkum DJ. Tracking individual action potentials throughout mammalian axonal arbors. eLife. 2017;6:e30198. DOI: 10.7554/eLife.30198
  22. 22. Stett A, Egert U, Guenther E, Hofmann F, Meyer T, Nisch W, et al. Biological application of microelectrode arrays in drug discovery and basic research. Analytical and Bioanalytical Chemistry. 2003;377:486-495. DOI: 10.1007/s00216-003-2149-x
  23. 23. MEA2100-systems | www.multichannelsystems.com n.d. Available from: https://www.multichannelsystems.com/products/mea2100-systems [Accessed: April 22, 2025]
  24. 24. Maestro edge MEA and live-cell analysis system | Axion Biosystems n.d. Available from: https://www.axionbiosystems.com/products/mea/maestro-edge [Accessed: April 22, 2025]
  25. 25. MaxTwo HD multiwell microelectrode array. MaxWell Biosystems n.d. Available from: https://www.mxwbio.com/products/maxtwo-multiwell-microelectrode-array/ [Accessed: April 22, 2025]
  26. 26. Kreutzer J, Ylä-Outinen L, Kärnä P, Kaarela T, Mikkonen J, Skottman H, et al. Structured PDMS chambers for enhanced human neuronal cell activity on MEA platforms. Journal of Bionic Engineering. 2012;9:1-10. DOI: 10.1016/S1672-6529(11)60091-7
  27. 27. Metsälä O, Kreutzer J, Högel H, Miikkulainen P, Kallio P, Jaakkola PM. Transportable system enabling multiple irradiation studies under simultaneous hypoxia in vitro. Radiation Oncology. 2018;13:220. DOI: 10.1186/s13014-018-1169-9
  28. 28. MEA2100-mini-systems | www.multichannelsystems.com n.d. Available from: https://www.multichannelsystems.com/products/mea2100-mini-systems [Accessed: April 22, 2025]
  29. 29. Taylor AM, Rhee SW, Tu CH, Cribbs DH, Cotman CW, Jeon NL. Microfluidic multicompartment device for neuroscience research. Langmuir. 2003;19:1551-1556. DOI: 10.1021/la026417v
  30. 30. le Feber J, Postma W, de Weerd E, Weusthof M, Rutten WLC. Barbed channels enhance unidirectional connectivity between neuronal networks cultured on multi electrode arrays. Frontiers in Neuroscience. 2015;9:412. DOI: 10.3389/fnins.2015.00412
  31. 31. DeMarse TB, Pan L, Alagapan S, Brewer GJ, Wheeler BC. Feed-forward propagation of temporal and rate information between cortical populations during coherent activation in engineered In vitro networks. Frontiers in Neural Circuits. 2016;10:32. DOI: 10.3389/fncir.2016.00032
  32. 32. Gladkov A, Pigareva Y, Kutyina D, Kolpakov V, Bukatin A, Mukhina I, et al. Design of cultured neuron networks in vitro with predefined connectivity using asymmetric microfluidic channels. Scientific Reports. 2017;7:15625. DOI: 10.1038/s41598-017-15506-2
  33. 33. Forró C, Thompson-Steckel G, Weaver S, Weydert S, Ihle S, Dermutz H, et al. Modular microstructure design to build neuronal networks of defined functional connectivity. Biosensors and Bioelectronics. 2018;122:75-87. DOI: 10.1016/j.bios.2018.08.075
  34. 34. Kapucu FE, Tujula I, Kulta O, Sukki L, Ryynänen T, Gram H, et al. Human tripartite cortical network model for temporal assessment of alpha-synuclein aggregation and propagation in Parkinson’s disease. Npj Parkinsons Diseases. 2024;10:1-20. DOI: 10.1038/s41531-024-00750-x
  35. 35. van de Wijdeven R, Ramstad OH, Valderhaug VD, Köllensperger P, Sandvig A, Sandvig I, et al. A novel lab-on-chip platform enabling axotomy and neuromodulation in a multi-nodal network. Biosensors and Bioelectronics. 2019;140:111329. DOI: 10.1016/j.bios.2019.111329
  36. 36. Brofiga M, Pisano M, Tedesco M, Boccaccio A, Massobrio P. Functional inhibitory connections modulate the electrophysiological activity patterns of cortical-hippocampal ensembles. Cerebral Cortex. 2022;32:1866-1881. DOI: 10.1093/cercor/bhab318
  37. 37. Pigareva Y, Gladkov A, Kolpakov V, Bukatin A, Li S, Kazantsev VB, et al. Microfluidic Bi-layer platform to study functional interaction between Co-cultured neural networks with unidirectional synaptic connectivity. Micromachines. 2023;14:835. DOI: 10.3390/mi14040835
  38. 38. Duc P, Vignes M, Hugon G, Sebban A, Carnac G, Malyshev E, et al. Human neuromuscular junction on micro-structured microfluidic devices implemented with a custom micro electrode array (MEA). Lab on a Chip. 2021;21:4223-4236. DOI: 10.1039/D1LC00497B
  39. 39. Takayama Y, Kushige H, Akagi Y, Suzuki Y, Kumagai Y, Kida YS. Selective induction of human autonomic neurons enables precise control of cardiomyocyte beating. Scientific Reports. 2020;10:9464. DOI: 10.1038/s41598-020-66303-3
  40. 40. Gabriel-Segard T, Rontard J, Miny L, Dubuisson L, Batut A, Debis D, et al. Proof-of-concept human organ-on-chip study: First step of platform to assess neuro-immunological communication involved in inflammatory bowel diseases. International Journal of Molecular Sciences. 2023;24:10568. DOI: 10.3390/ijms241310568
  41. 41. Vinogradov A, Kapucu EF, Narkilahti S. Exploring kainic acid-induced alterations in circular tripartite networks with advanced analysis tools. eNeuro. 2024;11(7):ENEURO.0035-24.2024. DOI: 10.1523/ENEURO.0035-24.2024
  42. 42. Frega M, Tedesco M, Massobrio P, Pesce M, Martinoia S. Network dynamics of 3D engineered neuronal cultures: A new experimental model for in-vitro electrophysiology. Scientific Reports. 2014;4:5489. DOI: 10.1038/srep05489
  43. 43. Smith I, Silveirinha V, Stein JL, de la Torre-Ubieta L, Farrimond JA, Williamson EM, et al. Human neural stem cell-derived cultures in three-dimensional substrates form spontaneously functional neuronal networks. Journal of Tissue Engineering and Regenerative Medicine. 2017;11:1022-1033. DOI: 10.1002/term.2001
  44. 44. Obien MEJ, Hierlemann A, Frey U. Factors affecting blind localization of a glass micropipette using a high-density microelectrode array. Proceedings of IEEE Sensors; Baltimore, MD, USA. 2013:1-4. DOI: 10.1109/ICSENS.2013.6688361
  45. 45. Egert U, Heck D, Aertsen A. Two-dimensional monitoring of spiking networks in acute brain slices. Experimental Brain Research. 2002;142:268-274. DOI: 10.1007/s00221-001-0932-5
  46. 46. Lieb F, Stark H-G, Thielemann C. A stationary wavelet transform and a time-frequency based spike detection algorithm for extracellular recorded data. Journal of Neural Engineering. 2017;14:036013. DOI: 10.1088/1741-2552/aa654b
  47. 47. Pettersen KH, Einevoll GT. Amplitude variability and extracellular low-pass filtering of neuronal spikes. Biophysical Journal. 2008;94:784-802. DOI: 10.1529/biophysj.107.111179
  48. 48. Nicholson C, Freeman JA. Theory of current source-density analysis and determination of conductivity tensor for anuran cerebellum. Journal of Neurophysiology. 1975;38:356-368. DOI: 10.1152/jn.1975.38.2.356
  49. 49. Nunez PL, Srinivasan R. Electric Fields of the Brain: The Neurophysics of EEG. New York, USA: Oxford University Press; 2006. DOI: 10.1093/acprof:oso/9780195050387.001.0001
  50. 50. Buzsáki G, Anastassiou CA, Koch C. The origin of extracellular fields and currents-EEG, ECoG, LFP and spikes. Nature Reviews Neuroscience. 2012;13:407-420. DOI: 10.1038/nrn3241
  51. 51. Mirfakhraei K, Horch K. Classification of action potentials in multiunit intrafascicular recordings using neural network pattern-recognition techniques. IEEE Transactions on Biomedical Engineering. 1994;41(1):89-91. DOI: 10.1109/10.277276
  52. 52. Kim KH, Kim SJ. Neural spike sorting under nearly 0-dB signal-to-noise ratio using nonlinear energy operator and artificial neural-network classifier. IEEE Transactions on Biomedical Engineering. 2000;47(10):1406-1411. DOI: 10.1109/10.871415
  53. 53. Jones IL, Livi P, Lewandowska MK, Fiscella M, Roscic B, Hierlemann A. The potential of microelectrode arrays and microelectronics for biomedical research and diagnostics. Analytical and Bioanalytical Chemistry. 2011;399:2313-2329. DOI: 10.1007/s00216-010-3968-1
  54. 54. Choi JH, Jung HK, Kim T. A new action potential detector using the MTEO and its effects on spike sorting systems at low signal-to-noise ratios. IEEE Transactions on Biomedical Engineering. 2006;53(4):738-746. DOI: 10.1109/TBME.2006.870239
  55. 55. Belitski A, Gretton A, Magri C, Murayama Y, Montemurro MA, Logothetis NK, et al. Low-frequency local field potentials and spikes in primary visual cortex convey independent visual information. The Journal of Neuroscience. 2008;28:5696-5709. DOI: 10.1523/JNEUROSCI.0009-08.2008
  56. 56. Quian QR. What is the real shape of extracellular spikes? Journal of Neuroscience Methods. 2009;177:194-198. DOI: 10.1016/j.jneumeth.2008.09.033
  57. 57. Lin J, Lai D, Wan Z, Feng L, Zhu J, Zhang J, et al. Representation and decoding of bilateral arm motor imagery using unilateral cerebral LFP signals. Frontiers in Human Neuroscience. 2023;17:1168017. DOI: 10.3389/fnhum.2023.1168017
  58. 58. Heldman DA, Wang W, Chan SS, Moran DW. Local field potential spectral tuning in motor cortex during reaching. IEEE Transactions on Neural Systems and Rehabilitation Engineering. 2006;14:180-183. DOI: 10.1109/TNSRE.2006.875549
  59. 59. Wang D, Zhang Q, Li Y, Wang Y, Zhu J, Zhang S, et al. Long-term decoding stability of local field potentials from silicon arrays in primate motor cortex during a 2D center out task. Journal of Neural Engineering. 2014;11:036009. DOI: 10.1088/1741-2560/11/3/036009
  60. 60. Wang Y, Song Y, Dai Y, Li X, Xie J, Luo J, et al. The burst of electrophysiological signals in the suprachiasmatic nucleus of mouse during the arousal detected by microelectrode arrays. Frontiers in Bioengineering and Biotechnology. 2022;10:970726. DOI: 10.3389/fbioe.2022.970726
  61. 61. Xu S, Deng Y, Luo J, Liu Y, He E, Yang Y, et al. A neural sensor with a nanocomposite interface for the study of spike characteristics of hippocampal neurons under learning training. Biosensors (Basel). 2022;12:546. DOI: 10.3390/bios12070546
  62. 62. Xu Z, Mo F, Yang G, Fan P, Wang Y, Lu B, et al. Grid cell remapping under three-dimensional object and social landmarks detected by implantable microelectrode arrays for the medial entorhinal cortex. Microsystems & Nanoengineering. 2022;8:1-11. DOI: 10.1038/s41378-022-00436-5
  63. 63. Kapucu FE, Välkki I, Mikkonen JE, Leone C, Lenk K, Tanskanen JMA, et al. Spectral entropy based neuronal network synchronization analysis based on microelectrode array measurements. Frontiers in Computational Neuroscience. 2016;10:112. DOI: 10.3389/fncom.2016.00112
  64. 64. Cutts CS, Eglen SJ. Detecting pairwise correlations in spike trains: An objective comparison of methods and application to the study of retinal waves. The Journal of Neuroscience. 2014;34:14288-14303. DOI: 10.1523/JNEUROSCI.2767-14.2014
  65. 65. Garofalo M, Nieus T, Massobrio P, Martinoia S. Evaluation of the performance of information theory-based methods and cross-correlation to estimate the functional connectivity in cortical networks. PLoS One. 2009;4:e6482. DOI: 10.1371/journal.pone.0006482
  66. 66. Di Florio M, Iyer V, Rajhans A, Buccelli S, Chiappalone M. Model-based online implementation of spike detection algorithms for neuroengineering applications. In: The 44th Annual International Conference of the IEEE Engineering in Medicine and Biology; 11-15 July 2022. Glasgow, United Kingdom: IEEE (Institute of Electrical and Electronics Engineers); 2022. pp. 736-739. DOI: 10.1109/EMBC48229.2022.9871444
  67. 67. Gold C, Henze DA, Koch C, Buzsáki G. On the origin of the extracellular action potential waveform: A modeling study. Journal of Neurophysiology. 2006;95:3113-3128. DOI: 10.1152/jn.00979.2005
  68. 68. Someck S, Levi A, Sloin HE, Spivak L, Gattegno R, Stark E. Positive and biphasic extracellular waveforms correspond to return currents and axonal spikes. Communications Biology. 2023;6:1-15. DOI: 10.1038/s42003-023-05328-6
  69. 69. Quiroga RQ, Nadasdy Z, Ben-Shaul Y. Unsupervised spike detection and sorting with wavelets and superparamagnetic clustering. Neural Computation. 2004;16:1661-1687. DOI: 10.1162/089976604774201631
  70. 70. Wertenbroek R, Thoma Y, Mor FM, Grassi S, Heuschkel MO, Roux A, et al. SpikeOnChip: A custom embedded platform for neuronal activity recording and analysis. IEEE Transactions on Biomedical Circuits and Systems. 2021;15:743-755. DOI: 10.1109/TBCAS.2021.3097833
  71. 71. Liu M, Chen C, Gao K, Gao F, Qin C, Yuan Q, et al. Neuronal network-based biomimetic chip for long-term detection of olfactory dysfunction model in early-stage Alzheimer’s disease. Biosensors and Bioelectronics. 2022;216:114619. DOI: 10.1016/j.bios.2022.114619
  72. 72. Chen R, Vakilna YS, Lassers SB, Tang WC, Brewer G. Hippocampal network axons respond to patterned theta burst stimulation with lower activity of initially higher spike train similarity from EC to DG and later similarity of axons from CA1 to EC. Journal of Neural Engineering. 2023;20:056004. DOI: 10.1088/1741-2552/acf68a
  73. 73. Yang Y, Deng Y, Xu S, Liu Y, Liang W, Zhang K, et al. PPy/SWCNTs-modified microelectrode Array for learning and memory model construction through electrical stimulation and detection of In vitro hippocampal neuronal network. ACS Applied Bio Materials. 2023;6:3414-3422. DOI: 10.1021/acsabm.3c00105
  74. 74. Zhang A, Mandeville ET, Xu L, Stary CM, Lo EH, Lieber CM. Ultraflexible endovascular probes for brain recording through micrometer-scale vasculature. Science. 2023;381:306-312. DOI: 10.1126/science.adh3916
  75. 75. Qin C, Yuan Q, Liu M, Zhuang L, Xu L, Wang P. Biohybrid tongue based on hypothalamic neuronal network-on-a-chip for real-time blood glucose sensing and assessment. Biosensors and Bioelectronics. 2024;244:115784. DOI: 10.1016/j.bios.2023.115784
  76. 76. Zhu Y, Jing L, Hu R, Mo F, Jia Q, Yang G, et al. High-throughput microelectrode arrays for precise functional localization of the globus pallidus internus. Cyborg and Bionic Systems. 2024;5:0123. DOI: 10.34133/cbsystems.0123
  77. 77. Maccione A, Gandolfo M, Massobrio P, Novellino A, Martinoia S, Chiappalone M. A novel algorithm for precise identification of spikes in extracellularly recorded neuronal signals. Journal of Neuroscience Methods. 2009;177:241-249. DOI: 10.1016/j.jneumeth.2008.09.026
  78. 78. Emery BA, Khanzada S, Hu X, Klütsch D, Amin H. Author spotlight: Advancing large-scale neural dynamics through HD-MEA technology. Journal of Visualized Experiments (JoVE). 2024;205:e66473. DOI: 10.3791/66473
  79. 79. Choi JH, Kim T. Neural action potential detector using multi-resolution TEO. Electronics Letters. 2002;38:541-543. DOI: 10.1049/el:20020386
  80. 80. Mallat SG. A theory for multiresolution signal decomposition: The wavelet representation. IEEE Transactions on Pattern Analysis and Machine Intelligence. 1989;11:674-693. DOI: 10.1109/34.192463
  81. 81. Daubechies I. Ten Lectures on Wavelets. Philadelphia, Pennsylvania, USA: SIAM (Society for Industrial and Applied Mathematics); 1992
  82. 82. Shalchyan V, Jensen W, Farina D. Spike detection and clustering with unsupervised wavelet optimization in extracellular neural recordings. IEEE Transactions on Biomedical Engineering. 2012;59:2576-2585. DOI: 10.1109/TBME.2012.2204991
  83. 83. Nabar N, Rajgopal K. A wavelet based Teager energy operator for spike detection in microelectrode array recordings. In: TENCON 2009-2009 IEEE Region 10 Conference; 23-26 January 2009. Singapore: IEEE (Institute of Electrical and Electronics Engineers); 2009. pp. 1-6. DOI: 10.1109/TENCON.2009.5395915
  84. 84. Semmlow JL, Griffel B. Biosignal and Biomedical Image Processing. 3rd ed. Boca Raton, Florida, USA: CRC Press; 2014
  85. 85. Shahid S, Walker J, Smith LS. A new spike detection algorithm for extracellular neural recordings. IEEE Transactions on Biomedical Engineering. 2010;57:853-866. DOI: 10.1109/TBME.2009.2026734
  86. 86. Natora M, Franke F, Obermayer K. Spike detection in extracellular recordings by hybrid blind beamforming. In: 2010 Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC’10. Buenos Aires, Argentina: IEEE (Institute of Electrical and Electronics Engineers); 2010. DOI: 10.1109/IEMBS.2010.5626483
  87. 87. Lee JH, Carlson DE, Shokri Razaghi H, Yao W, Goetz GA, Hagen E, et al. YASS: Yet another spike sorter. In: 31st Conference on Neural Information Processing Systems (Advances in Neural Information Processing Systems 30, NIPS 2017). Long Beach, California, USA: Curran Associates, Inc.; 2017
  88. 88. Rácz M, Liber C, Németh E, Fiáth R, Rokai J, Harmati I, et al. Spike detection and sorting with deep learning. Journal of Neural Engineering. 2020;17:016038. DOI: 10.1088/1741-2552/ab4896
  89. 89. Molen T v d, Lim M, Bartram J, Cheng Z, Robbins A, Parks DF, et al. RT-Sort: An action potential propagation-based algorithm for real time spike detection and sorting with millisecond latencies. PLOS One. 2024;19(12):e0312438. DOI: 10.1101/2024.04.08.588620
  90. 90. Abeles M, Goldstein MH. Multispike train analysis. Proceedings of the IEEE. 1977;65:762-773. DOI: 10.1109/PROC.1977.10559
  91. 91. Lewicki MS. A review of methods for spike sorting: The detection and classification of neural action potentials. Network. 1998;9:R53-R78
  92. 92. Chaure FJ, Rey HG, Quian QR. A novel and fully automatic spike-sorting implementation with variable number of features. Journal of Neurophysiology. 2018;120:1859-1871. DOI: 10.1152/jn.00339.2018
  93. 93. Pasquale V, Martinoia S, Chiappalone M. A self-adapting approach for the detection of bursts and network bursts in neuronal cultures. Journal of Computational Neuroscience. 2010;29:213-229. DOI: 10.1007/s10827-009-0175-1
  94. 94. Bakkum D, Radivojevic M, Frey U, Franke F, Hierlemann A, Takahashi H. Parameters for burst detection. Frontiers in Computational Neuroscience. 2014;7:193. DOI: 10.3389/fncom.2013.00193
  95. 95. Cocatre-Zilgien JH, Delcomyn F. Identification of bursts in spike trains. Journal of Neuroscience Methods. 1992;41:19-30. DOI: 10.1016/0165-0270(92)90120-3
  96. 96. Gourévitch B, Eggermont JJ. A nonparametric approach for detection of bursts in spike trains. Journal of Neuroscience Methods. 2007;160:349-358. DOI: 10.1016/j.jneumeth.2006.09.024
  97. 97. Lisman JE. Bursts as a unit of neural information: Making unreliable synapses reliable. Trends in Neurosciences. 1997;20:38-43. DOI: 10.1016/S0166-2236(96)10070-9
  98. 98. Chiappalone M, Novellino A, Vajda I, Vato A, Martinoia S, van Pelt J. Burst detection algorithms for the analysis of spatio-temporal patterns in cortical networks of neurons. Neurocomputing. 2005;65-66:653-662. DOI: 10.1016/j.neucom.2004.10.094
  99. 99. Cotterill E, Charlesworth P, Thomas CW, Paulsen O, Eglen SJ. A comparison of computational methods for detecting bursts in neuronal spike trains and their application to human stem cell-derived neuronal networks. Journal of Neurophysiology. 2016;116:306-321. DOI: 10.1152/jn.00093.2016
  100. 100. Madhavan R, Chao ZC, Potter SM. Plasticity of recurring spatiotemporal activity patterns in cortical networks. Physical Biology. 2007;4:181-193. DOI: 10.1088/1478-3975/4/3/005
  101. 101. Kapucu FE, Tanskanen J, Mikkonen J, Ylä-Outinen L, Narkilahti S, Hyttinen J. Burst analysis tool for developing neuronal networks exhibiting highly varying action potential dynamics. Frontiers in Computational Neuroscience. 2012;6:38. DOI: 10.3389/fncom.2012.00038
  102. 102. Legendy CR, Salcman M. Bursts and recurrences of bursts in the spike trains of spontaneously active striate cortex neurons. Journal of Neurophysiology. 1985;53:926-939. DOI: 10.1152/jn.1985.53.4.926
  103. 103. Ko D, Wilson CJ, Lobb CJ, Paladini CA. Detection of bursts and pauses in spike trains. Journal of Neuroscience Methods. 2012;211:145-158. DOI: 10.1016/j.jneumeth.2012.08.013
  104. 104. Kamioka H, Maeda E, Jimbo Y, Robinson HPC, Kawana A. Spontaneous periodic synchronized bursting during formation of mature patterns of connections in cortical cultures. Neuroscience Letters. 1996;206:109-112. DOI: 10.1016/S0304-3940(96)12448-4
  105. 105. Van Pelt J, Corner MA, Wolters PS, Rutten WLC, Ramakers GJA. Longterm stability and developmental changes in spontaneous network burst firing patterns in dissociated rat cerebral cortex cell cultures on multielectrode arrays. Neuroscience Letters. 2004;361:86-89. DOI: 10.1016/j.neulet.2003.12.062
  106. 106. Wagenaar DA, Pine J, Potter SM. An extremely rich repertoire of bursting patterns during the development of cortical cultures. BMC Neuroscience. 2006;7:11. DOI: 10.1186/1471-2202-7-11
  107. 107. Hedrich UBS, Liautard C, Kirschenbaum D, Pofahl M, Lavigne J, Liu Y, et al. Impaired action potential initiation in GABAergic interneurons causes hyperexcitable networks in an epileptic mouse model carrying a human NaV1.1 mutation. The Journal of Neuroscience. 2014;34:14874-14889. DOI: 10.1523/JNEUROSCI.0721-14.2014
  108. 108. Välkki IA, Lenk K, Mikkonen JE, Kapucu FE, Hyttinen JAK. Network-wide adaptive burst detection depicts neuronal activity with improved accuracy. Frontiers in Computational Neuroscience. 2017;11:40. DOI: 10.3389/fncom.2017.00040
  109. 109. Matsuda N, Odawara A, Katoh H, Okuyama N, Yokoi R, Suzuki I. Detection of synchronized burst firing in cultured human induced pluripotent stem cell-derived neurons using a 4-step method. Biochemical and Biophysical Research Communications. 2018;497:612-618. DOI: 10.1016/j.bbrc.2018.02.117
  110. 110. Mzezewa R, Lotila J, Kiiski H, Vinogradov A, Kapucu FE, Peltola J, et al. A kainic acid-induced seizure model in human pluripotent stem cell-derived cortical neurons for studying the role of IL-6 in the functional activity. Stem Cell Research. 2022;60:102665. DOI: 10.1016/j.scr.2022.102665
  111. 111. Gelfman S, Wang Q, Lu Y-F, Hall D, Bostick CD, Dhindsa R, et al. meaRtools: An R package for the analysis of neuronal networks recorded on microelectrode arrays. PLoS Computational Biology. 2018;14:e1006506. DOI: 10.1371/journal.pcbi.1006506
  112. 112. Hyvärinen T, Hyysalo A, Kapucu FE, Aarnos L, Vinogradov A, Eglen SJ, et al. Functional characterization of human pluripotent stem cell-derived cortical networks differentiated on laminin-521 substrate: Comparison to rat cortical cultures. Scientific Reports. 2019;9:17125. DOI: 10.1038/s41598-019-53647-8
  113. 113. Quian Quiroga R, Kreuz T, Grassberger P. Event synchronization: A simple and fast method to measure synchronicity and time delay patterns. Physical Review E. 2002;66:041904. DOI: 10.1103/PhysRevE.66.041904
  114. 114. Gray RM. Entropy and Information Theory. Boston, MA: Springer US; 2011. DOI: 10.1007/978-1-4419-7970-4
  115. 115. Salazar RF, Dotson NM, Bressler SL, Gray CM. Content-specific fronto-parietal synchronization during visual working memory. Science. 2012;338(6110):1097-1100. DOI: 10.1126/science.1224000
  116. 116. Schreiber T. Measuring information transfer. Physical Review Letters. 2000;85:461-464. DOI: 10.1103/PhysRevLett.85.461
  117. 117. Ito S, Hansen ME, Heiland R, Lumsdaine A, Litke AM, Beggs JM. Extending transfer entropy improves identification of effective connectivity in a spiking cortical network model. PLoS One. 2011;6:e27431. DOI: 10.1371/journal.pone.0027431
  118. 118. Granger CWJ. Investigating causal relations by econometric models and cross-spectral methods. Econometrica. 1969;37:424-438. DOI: 10.2307/1912791
  119. 119. Baccalá LA, Sameshima K. Partial directed coherence: A new concept in neural structure determination. Biological Cybernetics. 2001;84:463-474. DOI: 10.1007/PL00007990
  120. 120. Kispersky T, Gutierrez GJ, Marder E. Functional connectivity in a rhythmic inhibitory circuit using Granger causality. Neural Systems & Circuits. 2011;1:9. DOI: 10.1186/2042-1001-1-9
  121. 121. Cadotte AJ, DeMarse TB, He P, Ding M. Causal measures of structure and plasticity in simulated and living neural networks. PLoS One. 2008;3:e3355. DOI: 10.1371/journal.pone.0003355

Written By

Andrey Vinogradov, Laura Ylä-Outinen, Susanna Narkilahti and Emre Kapucu

Submitted: 24 April 2025 Reviewed: 25 April 2025 Published: 01 June 2025