Modeling seizure networks in neuron-glia cultures using microelectrode arrays

Abstract

Epilepsy is a common neurological disorder, affecting over 65 million people worldwide. Unfortunately, despite resective surgery, over 30 $\%$ of patients with drug-resistant epilepsy continue to experience seizures. Retrospective studies considering connectivity using intracranial electrocorticography (ECoG) obtained during neuromonitoring have shown that treatment failure is likely driven by failure to consider critical components of the seizure network, an idea first formally introduced in 2002. However, current studies only capture snapshots in time, precluding the ability to consider seizure network development. Over the past few years, multiwell microelectrode arrays have been increasingly used to study neuronal networks in vitro. As such, we sought to develop a novel in vitro MEA seizure model to allow for study of seizure networks. Specifically, we used 4-aminopyridine (4-AP) to capture hyperexcitable activity, and then show increased network changes after 2 days of chronic treatment. We characterize network changes using functional connectivity measures and a novel technique using dimensionality reduction. We find that 4-AP successfully captures persistently elevated mean firing rate and significant changes in underlying connectivity patterns. We believe this affords a robust in vitro seizure model from which longitudinal network changes can be studied, laying groundwork for future studies exploring seizure network development.

Keywords: 4-aminopyridine; epilepsy; epilepsy model; functional connectivity; microelectrode arrays; network changes; neuronal networks; seizure networks.

FIGURE 1

Functional connectivity in human epilepsy patients. (A) Here, we show the schematic for how FC was computed in our cohort of DRE patients. Specifically, we considerd icEEG obtained during neuromonitoring for seizure localization. We selected for contacts recording from regions identified as SOZ, SP, or control, equidistant to the SOZ. We then computed PLC and compared across all SOZ-SP and SOZ-control contact pairs across all patients (left). (B) Here, we depict extreme cases of what PLC measures. Specifically, PLC captures how in-phase, or synchronized two signals are. High PLC values (top) mean signals are synchronized, and therefore suggests that the underlying recording sites may be functionally connected. On the contrary, low PLC values recording sites are likely unrelated to one another. (C) We average PLC computed for all SOZ-SP and SOZ-control contact pairs. We show that within a given contact pair, SOZ-SP contacts are more functionally connected than their respective SOZ-control pair (left). When we compare PLC across all patients, we find that SOZ-SP contact pairs are truly significantly more synchronized, and hence likely more strongly connected, compared to SOZ-control pairs (0.29 versus 0.25, t(2) = 6.72, p = 0.0215, paired t-test, right). Abbreviations: FC, functional connectivity; DRE, drug-resistant epilepsy; SOZ, seizure-onset zone; SP, seizure spread; icEEG, intracranial electrocorticography; PLC, phase-locking coherence.

FIGURE 2

Experimental workflow. Here, we show the experimental workflow, delineating 4-AP treatment protocol. Treatment begins Day 0, after MFR has stabilized across neuron-glia cultures (see Methods). MEA recordings are collected prior to each day’s 4-AP wash-in using the Maestro Pro MEA system (Axion Biosystems, Atlanta, GA). 6-well MEA plates are used, with three wells serving as untreated controls and remaining used for 4-AP treatment. Protocol was repeated across multiple biological replicates, with each 6-well MEA plate reflecting a single biological replicate. During a single day’s 4-AP treatment, treatment wells (red) were spiked with concentrated stock solution of 4-AP. Simultaneously, control wells (grey) were treated with equivalent volume of Neurobasal medium. Treatment period lasts for 30-min, after which all wells (control and 4-AP-treated) undergo full wash-out and media change with Neurobasal medium. Treatment is repeated for 2 days, at approximately the same time each day. As such, each MEA recording is effectively collected 24-h apart, allowing for quantifying chronic changes. Abbreviations: 4-AP, 4-aminopyridine; MFR, mean firing rate; MEA, microelectrode arrays.

FIGURE 3

4-aminopyridine successfully captures hyperexcitable activity in vitro. To assess if 4-AP could capture hyperexcitability in our neuron-glia cultures on MEAs, we considered spike rasters and MFR before and after chronic 4-AP treatment. (A) Here, we show spike rasters from a representative control well before 4-AP treatment protocol (left) and after (right). We observe no discernable changes in spiking density. (B) Here, we consider spike rasters before (left) and after (right) 4-AP treatment. Compared to control, we note significantly higher spiking density and more synchronous firing periods, consistent with expectations. (C) We show changes in MFR throughout 4-AP treatment protocol, normalizing to pre-treatment MFRs, across multiple technical (9) and biological (3) replicates. MFR is significantly moreso increased in 4-AP-treated wells compared to control after just 1 day of treatment (1.90 $\pm$ 0.10 versus 0.53 $\pm$ 0.04, t = 11.70, Cohen’s d = 0.73, p $<$ 0.0001, two-sample t-test). MFR elevation persisted after 2 days of 4-AP treatment as well (2.51 $\pm$ 0.14 versus 0.31 $\pm$ 0.05, t = 14.36, Cohen’s d = 0.89, p $<$ 0.0001, two-sample t-test). Data are represented as $\overline{x}\pm$ SEM. Abbreviations: 4-AP, 4-aminopyridine; MFR, mean firing rate.

FIGURE 4

Increased baseline functional connectivity after chronic 4-aminopyridine treatment. To show our model can be used to study network changes, we considered FC changes before and after chronic 4-AP treatment. We quantified FC considering pairwise Pearson’s correlation (Fisher z-transformed $\rho$ ) across binned spike trains across all unique microelectrode pairs (see Methods). We create network plots by plotting significant pairwise connections (Fisher z-transformed $\rho >$ 0.80). Stronger connections (i.e., edges) are delineated using warmer colors (red) lines that are thicker. Weaker connections (i.e., edges) are delineated using cooler colors (yellow) lines that are thinner. Microelectrodes (i.e., nodes) with more connections are delineated by blue with larger diameter. Below each network plot, histogram distributions of Pearson’s correlations (Fisher z-transformed $\rho$ ) are shown. (A) Here, we show changes in FC in a representative control well from before (left) experimental protocol and after (right). We observe that FC estimates increase slightly after 48-h, and maintain a unimodal distribution (blue histograms). (B) Here, we show changes in FC in a representative 4-AP well before (left) and after (right) chronic 4-AP treatment. We observe a discernable increase in the number of stronger pairwise connections, compared to control. Furthermore, FC increases significantly more in 4-AP-treated wells, compared to control ( $\mathrm{\Delta }\rho$ CI(95 $\%$ ) = [0.3827, 0.4155], t = 47.75, p $<$ 0.0001, two-sample t-test). Interestingly, FC correlation distributions become more bimodal after chronic 4-AP treatment as well. Abbreviations: FC, functional connectivity; 4-AP, 4-aminopyridine.

FIGURE 5

Temporal evolution of network connectivity mapped in low-dimension space. As network connectivity may be complex and change dynamically, we developed a method to track how a well’s network connectivity evolves over time by reducing high-dimensional connectivity information to a low-dimension embedding (see Methods). (A) Here, we show how network connectivity evolves across 3 days in representative control (blue) and 4-AP-treated (red) wells. Timepoint represented is delineated next to each point. We observe that after 2 days of chronic 4-AP treatment, baseline network connectivity occupies a distinct subspace, far from pre-treatment network connectivity. In contrast, we observe that network connectivity in a representative control well occupies a smaller subspace, closer to what is observed Day 0. (B) To quantify change in network connectivity over time, we compute the Euclidean distance of each point to Day 0. We find that after 2 days, compared to controls, 4-AP-treated wells show significant changes in network connectivity compared to pre-treatment (22.43 $\pm$ 3.55 versus 10.67 $\pm$ 5.74, t = 2.58, Cohen’s d = 1.43, p = 0.0298, two-sample t-test). Abbreviations: FC, functional connectivity; 4-AP, 4-aminopyridine.