Long-term brain-on-chip: Multielectrode array recordings in 3D neural cell cultures

This study presents sensing of network bursts in a three-dimensional (3D) cell culture system consisting of a microbioreactor and a multielectrode array (MEA), i.e., brain-on-chip, to interpret neural network dynamics in a label-free manner. While our initial results reported an increased single spiking activity already over the course of days 7, 14, and 21 in vitro , the advanced long-term analysis of the data set (including a last timepoint at day 79) here demonstrates a proof-of-principle for following bursting patterns upon maturation of the network in the microbioreactor as an add-on device for a commercial MEA recording system. These quantitative electrophysiological findings, including mean bursting rate, mean burst duration, and network burst dynamics, confirm a 2D to 3D transition in coherence with the literature.


I. INTRODUCTION
Human three-dimensional (3D) cell cultures are exquisitely positioned to imitate the complex in vivo cell-cell and cell-tissue behavior. Compared to two-dimensional (2D) cultures, the 3D approach improves the expression of differentiated functions and tissue organization. 3D cell culture models can, therefore, fill the gap between animal and 2D cell culture studies and harness the potential to unlock complex content-rich experiments based on human cell sources and reduce the extensive use of animals in (bio)medical studies. 1 Furthermore, by means of 3D cell cultures and their analysis, these systems can be enabled to decipher biological functions, including disease mechanisms, of the human body which are not detectable in 2D culture formats or animal models. [2][3][4] Despite all these benefits, quantitative data collection of 3D cell cultures is challenging. They often need advanced confocal (fluorescence) microscopy with enhanced spatial resolution in all directions, 5 while potentially content-rich optical techniques can be very time-consuming and expert-labor intensive, 6 and additionally, fluorescence-based characterization techniques are mostly performed as an endpoint method, i.e., being toxic and static. 7 Therefore, there is a need for label-free in situ assays. This requirement makes electrical measurement techniques a method of choice for readouts in 3D cell cultures. 6 The development of impedancebased sensors 7 and advances in multielectrode arrays (MEAs) 8 for electrophysiological recording techniques in human induced pluripotent stem cell (hiPSC)-derived 3D cultures of cortical neurons are the foundation for an effective brain-on-chip (BoC) technology. 9 Generally, MEAs have planar electrodes on glass surfaces and require very close vicinity of the neurons for high signal-to-noise ratio data collection. Nevertheless, this well-known arrangement of electrodes can also act as a transducer for the signals generated in the spatially extended 3D network to retrieve biologically meaningful results. Previously in our group, we introduced such a concept of signal collection using a MEMS add-on, i.e., a microbioreactor, which can be reversibly assembled on a MEA plate to maintain a 3D cell culture. This 3D culture system consists of a circular cell culture chamber, surrounded by a ring-shaped microfluidic channel. 10,11 Cell culture chamber and microfluidic channel are separated by a porous barrier. As a preliminary result, we reported that hiPSC-derived cortical neurons can be cultured up to day 21 by using this system. It was also shown that spontaneous spikes, having 100 μV amplitudes, were present already at day 7. Moreover, burst formation, consisting of higher amplitude (400 μV) spikes, was observed at days 14 and 21. 11 In this research paper, we completed the analysis of the data set, received from this unique experiment with cultured hiPSC-derived cortical neurons on MEA, in a quantitative manner up to 79 days-in vitro (DIV) as an example of illustrating the value of MEMS technology in this emerging application field of BoC technology. By means of the encircling microfluidic channel and the porous cell culture wall from polyethersulfone (PES), neurons did not lose their contact with 2D electrodes of MEA since nutrients were provided in 3D from the top and the side walls. There are a few studies which present long-term culturing of neuronal cell culture in 2D (Table I). However, for 3D, the culture durations are shorter than 2 months (Table I). To the best of our knowledge, this analysis of data, received by brain-on-chip-based sensing of network bursts in a 3D neuronal cell culture system, gave us an opportunity to obtain the widest electrophysiological data repertoire belonging to hiPSCs-derived cortical neurons in 3D cell culture established in the literature so far.

A. Integration of microbioreactor and MEA
Our microbioreactor was fabricated by using soft lithography of polydimethylsiloxane (PDMS) in a mechanically machined poly (methyl methacrylate) mold and porous PES cylindrical membranes. Details of this fabrication are presented in Ref. 11. Figure 1 depicts our brain-on-chip 3D cell culture system as a schematic in parts (a) and an assembled fashion (b), which had been previously introduced in Ref. 11. In brief, the operational procedure of the system assembly steps can be described as follows: a 2D cell layer is formed first. Next, cells inside a hydrogel are seeded into the porous PES tube, i.e., the culture chamber [ Fig. 1(c)]. Cell medium is loaded into the microfluidic channel (volume = 120 μl), encircling the PES tube (wall thickness <500 μm), to provide nutrients to cells and waste removal via the side wall [ Fig. 1(c)] and subsequently cell culture medium is topped up in the cell culture chamber from the top, too. MEA electrodes acquire in situ and label-free response of the 3D cultured networks. Two different experimental groups were prepared in the cell experiments [ Fig. 1(c)] by applying the same procedures for surface coating and cell seeding. One group was a 3D cell culture in a microbioreactor on MEA and the other was a 3D cell culture, prepared in a confined area on MEA, provided only by a plain PDMS ring without microfluidic exchange. The latter serves as a culture control to assess the effect of applying a surrounding microfluidic channel in such 3D cultures.

B. hiPSCs-derived cortical neuron cultures
hiPSC-derived cortical neurons (Axol Bioscience, Cambridge, UK) were differentiated and cultured based on the description of the manufacturer, 18 which should result in a pure population of cortical neurons. In brief, neural progenitor cells were thawed and subsequently seeded into a microbioreactor integrated with a MEA, coated with ReadySet and SureBond (Axol Bioscience, UK), consecutively, before cell seeding. After 30 min, cells mixed with a hydrogel, i.e., Corning Matrigel matrix (Life Sciences, USA), were added on top of them to obtain 3D cell culture and culture medium was topped up. Details of this procedure are described already in Ref. 11.

C. Spike, burst, and network burst (NB) detection and analysis
Measurements were carried out via a 2100MEA system (Multi Channel Systems GmbH, Germany) by using 120 electrodes MEA plate, having a plastic ring with an inner diameter of 26.5 mm on it (120MEA200/30 iR-Ti-pr). Cells were kept within neural differentiation medium (Axol Bioscience, UK) for 9 days and subsequently in neural maintenance XF medium (Axol Bioscience, UK) for the  29 The first parameter to understand firing behavior of a neuronal cell culture is spikes. Spontaneous spikes are the action potential discharges without applying stimulation. 19 For spike detection, voltage threshold was chosen as five times V noise-rms . 20,21 Bursts occur when spikes become more organized. 22 The level of organization in spikes to define a burst is determined with respect to features of spikes, including interspike intervals. Network bursts   22 There should be a threshold like minimum spontaneous channels.
In this study, we defined it as the half of total number of bursting channels. For burst and NB recognition, the parameters are presented in Table II. 23

D. Statistical analysis
More than four measurements at each DIV were taken consecutively for 1 min. Data were presented as average (avg) ± standard deviation (SD). Statistical correlation between the averages of these measurements at different DIVs was investigated by one-way ANOVA technique using Fisher's Least Significant Difference test. p values of <0.05 were considered statistically different. Origin Pro 2019 (OriginLab) was used for these analyses.

III. RESULTS AND DISCUSSION
3D cell culture recordings were performed up to day 79 in the microbioreactor/MEA (group 1) and day 29 in PDMS ring/MEA (group 2). In supplementary video S1 in the supplementary material, 29 an example measurement with microbioreactor at day 35 is presented. After 29 DIV, the PDMS ring culture rapidly degraded or lost connection with the electrodes and was discarded. The microbioreactor culture was still producing signals after day 79 but was discarded shortly thereafter for practical experimental reasons. Spike and burst numbers, average burst duration, NB occurrence and duration, and in-burst spike percentages were investigated for these two cell cultures to make neuronal network analysis, quantitatively. Spikes, bursts, and NBs were detected at 6 DIV for both groups although burst amount (4 ± 4 bursts/min) was significantly low and stochastic in group 1 and relatively higher (14 ± 9 bursts/ min) but randomly distributed in 5 min for group 2. All three types of negative spikes (mono-, bi-, and tri-phasic) were observed as early as day 6, indicating good coupling between the different parts of neurons, such as soma and axon, and MEA electrodes. 24 Measurements were taken as 1 min durations day-by-day. Minute-to-minute fluctuations occurred in neuron dynamics of both groups (for whole sets of data, please see Secs. II A and II C in the supplementary material). 29 Fluctuating behavior was expected for mature neuronal networks after 3-week in vitro. 22 Therefore, comparisons between recordings at different DIVs were carried out by ANOVA one-way analysis at p = 0.05 level in which average values per minute were considered.
For group 1 (microbioreactor), the average of spike numbers increased up to day 19 with significant differences (at p = 0.05 level) between day 6, 7, 8, 9, 10,12, 14, and 19 [ Fig. 2(a)]. It started to decrease at day 21. There was no meaningful variation in spike number up to day 27. At day 27, an abrupt spike number reduction occurred although there was no significant difference in the active electrode (AE) and bursting electrode (BE) numbers for day 25, 26, and 27 (10 ± 1, 8 ± 2, and 9 ± 2 for AEs and 7 ± 2, 5 ± 1, and 4 ± 2 for BE, respectively). This oscillating behavior, i.e., increase and decrease in spike number, was observed between days 35 and 51 and between days 51 and 72. These tendencies were followed by burst numbers, also [ Fig. 2(b)]. Although there was no specific variation in active and bursting electrode numbers (AE and BE number graphs with respect to DIVs are presented in Sec. II D in  the supplementary material), 29 this behavior of cells can be interpreted as neurons started to remodel their environment and changed their position by migration into the third dimension within the Matrigel. Overall, the increase in spike number and high order patterns of bursts indicated neuronal maturity in 3D cell culture. 25 Based on Frega et al., the number of spikes included in bursts (i.e., in-burst spikes) were divided by total number of spikes and the obtained value was named in-burst spike ratio. 16 Up to the end of week 3, there was a significant increase to 85% in the mean value of in-burst spike ratio (at p = 0.05 level, n = 1) [Fig. 2(c)]. This is coherent with the maturation level of 2D cell culture. According to Frega et al., spikes form clusters other than being in random format after 3 weeks in 2D cell culture. These clusters are grouped as bursts, showing maturation in neurons. 16 However, at day 22, there was a sharp decrease in in-burst spike ratio of cells up to 59% in microbioreactor and after this point no statistically significant increase occurred in in-burst spike ratio up to day 33 [ Fig. 2(c)]. This in-burst spike ratio (60% as average) showed a transition to 3D cell culture. 16 The largest minute-to-minute fluctuation occurred in spike number at day 22. This can also be a clue for 3D transition of cells. Additionally, supporting random spike ratio tendency as evidence for 3D transition, average burst number for 1 min increased up to 86 ± 8 bursts/min at 19 DIV and significantly (p < 0.05) lower number of bursts (40 ± 30 bursts/min) were observed at day 22 [ Fig. 2(b)]. There is no conclusive evidence in the literature about a significant increase in mean burst duration when 3D transition occurs, yet. According to our analysis, there was an increase in mean burst durations while neuronal culture became mature up to end of week 3 although this increase was not statistically significant [ Fig. 2(d)]. Hyvärinen et al. presented the burst numbers (per minute) of 2D hiPSCs-derived cortical neurons in comparison with rat cortical neurons. According to their results, the average burst number was increased for hiPSCs in 2D up to 33 bursts/min while with the same duration in cell culture, rat cortical neurons reached 15 bursts/min at maximum. 12 Frega et al. obtained 20 bursts/min for 2D cultures of hippocampal neurons, dissociated from E18 Sprague Dawley rats, and 5 bursts/min for 3D cultures of the same cells. 16 Our findings indicate that for hiPSCs-derived cortical neurons, burst amount increased up to 86 bursts/min in first 3 weeks during which 2D to 3D transitions did not occur. After this point, burst amount decreased to 40 bursts/min and then 15 bursts/min. The burst amounts in our cell culture were greater than both 2D and 3D cell culture findings in the literature. The burst durations were also higher than that of presented in Frega et al. 16 These differences in burst numbers and durations can be potentially caused by the differences between astrocyte ratio of these cell cultures since the existence of astrocytes can vary neuronal activity due to the feedback and feedforward signals between them and neurons or more speculative by the difference in donors that the cells have been induced from Refs. 26 and 27. Figure 2(e) presents average network burst (NB) numbers per 1 min. There is no data presented for day 7 and 72 because no NB formation was observed at these days. As early as day 6, NBs were formed. After day 21, average NB numbers decreased and did not exceed 4. This finding is in-line with the in-burst spike ratio and average burst numbers by endorsing 3D transition in cell culture. 16 It might not be interpreted as the death of cells since the occurrence of spikes, bursts, and NBs up to the end of day 79 provides sufficient evidence that the cell culture was healthy throughout the entire analysis time. Variable NB duration promotes the evidences in the literature at which it was claimed that the duration of NB in the 3D cell culture can be either close to the ones in 2D or longer than that of 2D [ Fig. 2(f )]. 28 As evidence for 3D cell culture, fluorescent cell images from different heights of cell culture inside microbioreactor are presented in Fig. 3 from bottom (a) to top (d) of cell culture. The z-stack (obtained in confocal microscope, Leica TCS, SP5X) video from which these images are taken is presented as supplementary video in the supplementary material. 29 For group 2 (PDMS ring), significant decrease in spike number [ Fig. 4(a)] and burst number [ Fig. 4 Fig. 4(c)]. Figure 4(d) presents average burst durations, which were similar with the durations obtained in Ref. 12 for longer period in 2D hiPSCs culture and were also longer than indicated in Ref. 16 for the same period in 3D cell culture of hippocampal neurons, dissociated from E18 Sprague Dawley rats. Figures 4(e) and 4(f ) show the NB number for 1 min and average NB durations, respectively. No values were presented for days 22 and 29 since no NB formation was observed in these days. There was no significant day-to-day variation between NB numbers and durations for group 2. If group 1 and group 2 are compared for 4 weeks in vitro, NB numbers are higher and NB durations are longer in group 1 than those of group 2. This observation offers us an argument for the importance of the microbioreactor design, which had an encircling microfluidic channel providing shear-free exchange of nutrients and waste in and out of the 3D cell culture chamber. Hence, we hypothesis that by means of this property, cells did not lose their contact with 2D MEA electrodes and data collection could be achieved for longer times in 3D cell culture as ever reported before in the literature for such type of culture arrangements.

IV. SUMMARY AND CONCLUSIONS
The collection of network activities by electrophysiological recordings provides us a strong clue for arranging proper microenvironments and maintenance of hiPSC-derived cortical neuron cultures utilizing, for example, an encircling microfluidic channel in such cultures. While cells demonstrated their transition to a 3D configuration after two to three weeks in vitro in both, the simple PDMS ring confined as well as the microbioreactor culture on MEA plates, our brain-on-chip 3D cell culture system applying a porous side wall in the culture chamber with an encircling flow channel can be successfully used for applications studying healthy or pathophysiological stem cell-derived neural microtissues aiming for a long-term study in a label-free and noninvasive manner. For this purpose, the efforts in proposing new research with a focus on repeatability and reliability analysis of this system with technical and biological replicas for such long culture durations in vitro are deemed justified.