SUMMARY: Spatiotemporal neural patterns depend on the physical structure of neural circuits. Neural plasticity can thus be associated with changes in the circuit structure. For example, newborn neurons migrate toward existing, already matured, neural networks in order to participate in neural computation. In the present study, we have conducted two experiments to investigate how neural migration is associated with the development of neural activity in primary dissociated cultures of neuronal cells. In Experiment 1, using a mature culture, a high‐density CMOS microelectrode array was used to continuously monitor neural migration and activity for more than two weeks. Consequently, we found that even in mature neuronal cultures neurons moved 2.0 ± 1.0 μm a day and that the moving distance was negatively correlated with their firing rate, suggesting that neurons featuring low firing rates tend to migrate actively. In Experiment 2 using a co‐culture of mature and immature neurons, we found that immature neurons moved more actively than matured neurons to achieve functional connections to other neurons. These findings suggest that neurons with low firing rates as well as newborn neurons actively migrate in order to establish their connections and function in a neuronal network.
CMOS; dissociated cultured neuron; cell migration; microelectrode array; neural activity
Neural activity patterns depend on the configuration of neural networks [
There are reports of neuron migration, in particular, the migration of newborn neurons [
In order to resolve the above problems, we conceived a method of measuring dissociated culture neurons and analyzing functional networks in dissociated cultures using a high‐density CMOS MEA, as shown in Fig. [NaN] (a) [
To the best of our knowledge, there are no reports of long‐term, individual cell‐level experiments intended to investigate the relation between network configurations and neural activity patterns. We investigated the migration of individual neurons in a network using primary dissociated cultures, and also clarified the relationship between the migration of neurons and their activity. Specifically, we considered two issues:
Variation of activity with migration distance (using a high‐density CMOS array).
Migration and functional connections of immature neurons (with relatively low firing rates) and mature neurons (with relatively high firing rates) co‐cultured in the same dish.
All neurons used in the experiments were extracted from embryos of Wistar rats on the 18th day of gestation. After treating the cerebral cortex with 0.25% trypsin, the neurons were dissociated by pipetting and were seeded on a high‐density CMOS array and a culture dish. In both cases, the seeding surface was coated with polyethyleneimine and laminin. After seeding, the neurons were deposited for 30 minutes, then cultured overnight in 0.5 ml seeding medium. On the next day, 0.5 ml normal broth was added. The seeding medium was prepared by adding 10% horse serum, 0.5 mM GlutaMax (Life Technologies), and 2% B‐27 to Neurobasal (Life Technologies). The normal broth was prepared by adding 10% horse serum, 1 mM sodium pyruvate, and 0.5 mM GlutaMax (Life Technologies) to D‐MEM (Life Technologies). The neurons were cultured in an incubator at 37°C, 5% CO
In this experiment, we measured the migration distance and activity of neurons in a mature neural network using a high‐density CMOS array. The results were used to investigate the relation between the migration and activity of neurons. Two samples were examined in this experiment, as shown in Table [NaN] . The table describes the conditions of culturing and measurement.
Test conditions of Experiment 1
Dish # Dish 1‐1 Dish 1‐2 Plating density 16,000 cells/20 μl 14,000 cells/20 μl Measurement day DIV 26 DIV 26 DIV 28 DIV 28 DIV 34 DIV 30 DIV 36 DIV 34 DIV 40 DIV 42 DIV 42
Dissociated culture systems with 14,000 or 16,000 cells seeded per 20 μl on the high‐density CMOS array shown in Fig. [NaN] were continuously measured in the normal broth. The broth was replaced every week. The action potential was recorded using the open‐source software MEABench [
The average amplitude of the action potential on each electrode was used to build action potential maps showing the spatial distribution. The neuron locations were identified after smoothing with a Gaussian filter. The kernel size of the Gaussian filter was 9 × 9 pixels and the standard deviation of the Gaussian distribution was set to σ=1. In the smoothed action potential maps, locations meeting the following conditions were recognized as cell bodies, and their spatial extremal values were recognized as body cell centers.
The action potential takes its negative extremal value.
This action potential is measured on three adjacent electrodes.
Condition (ii) was imposed to prevent misdetection of neurites as cell bodies.
We calculated the migration distance of the neuron bodies by comparing their estimation locations on every observation day. As shown in Fig. [NaN] , the peak distributions of consecutive days were superimposed and new the location of a neuron body was determined as that with the shortest Euclidean distance from the initial location. This Euclidean distance was used as the migration distance. The migration distances of every neuron body found in this way were compared with the numbers of firings and action potential waveforms to estimate the correlation. The Pearson linear coefficient of correlation was used as a measure.
When measurements were completed, immune antibody staining was applied to the samples in order to verify the results of cell body estimation. We used MAP2 antibodies (MAB3418, Millipore Corp.) that specifically bound to microtubule‐associated protein 2 contained in the neurons. First, the cultured neurons were fixed for 15 minutes in 4% paraformaldehyde and permeabilized in 0.25% Triton X‐100 PBS solution. Then, after blocking for 30 minutes in 4% Block Ace (DS Pharma Biomedical), the samples were incubated for 2 hours at 37°C in mouse monoclonal MAP2 antibody at a concentration of 1:200. They were then incubated for 30 minutes at 37°C in Alexa 488 anti‐mouse IgG antibody at a concentration of 1:200.
In this experiment, we investigated the relation between the migration and activity of neurons by co‐culturing of immature and mature neurons.
We used an H‐shaped separator (Culture‐Insert, Ibidi) with the walls partly removed, as shown in Fig. [NaN] . The bottom surface of the H‐separator was cemented to the culture dish. First, neurons were seeded in one well and cultured for 34 days. Then new neurons were seeded in the other well and cultured for 4 days, after which the H‐separator was removed. Thus, we observed the cell migration and functional connections in the partition area (thickness: 500 ±50 μm).
Mature and immature neutrons were seeded in separate wells at a density of 20,000 cells per 20 μl. In this experiment, three samples (Dish‐2‐1‐1, Dish 2‐1‐2, and Dish 2‐1‐3) were used as shown in Table [NaN] . We compared photo‐contrast microscope images (Fig. [NaN] (a)) of the samples taken immediately after removal of the H‐separator and after 2 weeks. The images were binarized (Fig. [NaN] (b)) in order to count the neurons. The rectangle in the middle of Fig. [NaN] (b) shows the partition area of the H‐separator. The area was divided into three equal regions (ROI: region of interest), and the neurons were counted in each region. The migration of immature and mature cells was evaluated from the rate of change of the number of cells in each ROI. Specifically, we found the rate of increase of the percentage of neurons contained in a certain ROI, and used it as the neuron increase rate.
Test conditions of Experiment 2
Dish 2‐1‐1, Dish 2‐1‐2, Dish 2‐1‐3 Well 1 (old) Well 2 (young) The first observation (immediately after removal of H‐shaped separator) DIV 38 DIV 4 The second observation (2 weeks after removal of H‐shaped separator) DIV 52 DIV 18
We used calcium imaging to check whether neuron groups of different age formed functional connections via the partition area when the H‐separator was removed in the co‐cultured samples. Specifically, we used 10 μM Fluo‐4 (fluorescent calcium indicator). This reagent has a relatively high dissociation constant of 345 nM and is often used for the imaging of rat neocortical neurons [
Monte Carlo simulations [
Using the action potentials on the electrodes of the high‐density CMOS array, action potential maps were built as shown in Fig. [NaN] (a)(i). The maps were then smoothed and the neurons were localized as shown in Fig. [NaN] (a)(ii). The superposition of an action potential map and immunostaining image is shown in Fig. [NaN] (b). As can be seen in the close‐up view in Fig. [NaN] (c)(i), the cell body activity is obtained from multiple adjoining electrodes. In contrast, when the action potential is not obtained from multiple adjoining electrodes, as shown in Fig. [NaN] (c)(ii), there is a high probability of neurites, and such cases were discarded in the analysis. Thus, we investigated the changes in cell body locations identified from the action potential maps. Neuron locations observed on different days are shown in Fig. [NaN] . These results indicate that dissociated cultured neurons migrate not only during the development period but also after maturity. Over 16 days of observation, the average daily distance of neuronal body migration was 2.0±1.0 μm.
We next investigated the relation between the total migration distance of individual neurons during the observation period and the firing rate on the initial observation day, as shown in Fig. [NaN] . The coefficients of correlation for the two test samples were –0.44 and –0.15, and a significant negative correlation was confirmed (t‐test: p<0.05).
However, we may assume that neurons can move easily when they do not strongly adhere to the electrode surface of the high‐density CMOS array. In addition, we may assume that the action potentials of the neurons, and therefore the firing rates are underestimated. Thus, we examined the relation between the total migration distance of individual neurons during the observation period and the action potential on the last measurement day (Fig. [NaN] ). The correlation coefficients for the two test samples were –0.076 and –0.13, and no significant correlation was confirmed (t‐test: p>0.05). Therefore, the adhesiveness of the cells does not affect their migration distance.
In order to investigate the relation between the migration of neurons and their maturity, we observed neurons migrating toward the former separator area in co‐culture systems including immature and mature neurons. Figure [NaN] shows the neuron increase rates in every ROI immediately after the removal of the H‐separator and after two weeks. As can be seen from the diagram, a higher increase rate was obtained in the ROI on the immature neuron side than on the mature neuron side.
We then employed calcium imaging to measure the activity of neuron groups in the co‐culture systems and to examine the functional connections via the correlation in activity between arbitrary cell pairs. Thus, functional connections were confirmed in neuron groups of different age, as shown in Fig. [NaN] (a). In order to verify the analytical results, the partition was cut with a scalpel; the functional connections then disappeared, as shown in Fig. [NaN] (b).
As can be concluded from these results, the immature neuron group migrates toward the mature neuron group to form functional connections.
Experiment 1 showed that some neurons migrate even in a mature network, and that there is a significant correlation between the migration distance and the firing rate. This correlation is not caused by the adhesion of neurons to the electrode surface of the high‐density CMOS array. In addition, in Experiment 2, immature neurons migrated over longer distances than mature neurons. Usually immature neurons have higher activity rates than mature ones. Therefore, the results of Experiment 2 agree with Experiment 1 as regards the correlation between neuron migration and activity. Functional connections between immature and mature neurons were also confirmed. Therefore, we may conclude that immature neurons migrate and form connections with mature neural networks, thus playing an important role in network development.
The adhesion of cells to the electrode surface can be interpreted in two ways in these experiments. First, weak adhesion contributes to cell migration and low action potentials are observed. Second, the S/N ratio drops as a result, and the firing rates are underestimated. If these interpretations are correct, then the action potential of migrating neurons cannot be measured properly, and a significant negative correlation should be observed between the migration distance and the absolute value of the action potential. However, no such significant correlation was observed in our experiments (Fig. [NaN] ). Therefore, we may infer that the activities of migrating neurons were correctly observed at nearly the same S/N rate regardless of the migration distance.
However, these experiments are not sufficient to clarify a causal relationship between neuron activity and migration. The following assumptions can be made about such a causal relationship. First, neurons with high firing rates are connected to many neighboring neurons, and migration is suppressed by the tension of the neurites, while in contrast, neurons with low firing rates have few if any connections with their neighbors and can therefore migrate easily. Second, we may consider nervous system homeostasis. That is, if neurons act within a network so as to keep the firing rates above a certain level, then they should migrate to locations where the firing rate can be increased. In order to verify these possibilities, it is necessary to quantify the migration distance when the firing rates of the entire well are adjusted by varying the composition of the culture broth, and when the firing rates are forced to increase by stimulating the neurons.
There are also experimental issues requiring verification and improvement. In Experiment 1, the neuron bodies were localized by using electrical activity patterns, but the validity of this kind of localization is yet to be proved. Actually, the positions of the extremal values on the action potential maps do not necessarily correspond to cell body centers. Therefore, the reliability of experimental results must be improved by reducing the intervals between observations, and by examining action potential maps when neurons are labeled with fluorescent proteins.
Experiment 2 showed that neurons with different numbers of days in culture migrate and make functional connections; however, the relationship between migration and changes in neural activities was not investigated. In addition, the cells observed in this experiment were not only neurons but also glial cells. Therefore, the cell migration characteristics of immature neurons and mature neurons (with a low firing rate) could not be compared accurately in this experiment. In order to resolve these problems, an experimental system for co‐culture on a high‐density CMOS array must be developed.
We investigated the migration of individual neurons in a network using dissociated cultures, and reached the following conclusions about the relationship between the migration of neurons and their activity.
(i) Using a high‐density CMOS array, we investigated the relationship between the migration density and activity of mature neurons. The average daily distance of neuron migration was found to be 2.0 ± 1.0 μm. We also confirmed the existence of a negative correlation between the migration distance and the activity change: that is, neurons with low firing rates migrated vigorously.
(ii) Immature and mature neurons were co‐cultured using an H‐shaped separator. We found that the immature neurons migrated more vigorously than the mature neurons. This kind of migration is assumed to play an important role in the making of functional connections between immature and mature neurons.
These results indicate that neurons with low firing rates migrate vigorously in order to acquire increasingly important roles in a network.
This study was performed in the framework of the “Creation of Future Machine Technologies” collaborative program of Denso Corporation and the University of Tokyo. The study was also supported by a Grant‐in‐Aid for Scientific Research (23680050).
Graph: High‐density CMOS microelectrode array.
Graph: Illustration of quantification of migration distance of neurons.
Graph: H‐shaped separator for co‐culture of different DIV neurons.
Graph: Localization of cell bodies under phase‐contrast microscope.
Graph: Localization of cell bodies based on neural signals.
Graph: Migration trajectories of neurons on CMOS array. (Gray levels correspond to DIV: lightest, DIV 26; darkest, DIV 42.)
Graph: Correlation between migration distance and firing rates of neurons on CMOS array. (Results from 2 independent dishes (Dish 1‐1 and 1‐2) are shown.)
Graph: Action potential amplitude with respect to migration distance as measured on CMOS array (no correlation observed).
Graph: Estimation of cell migration. (Numbers of cells were counted twice, 2 weeks apart, at 3 ROIs: on old/young culture sides and in center region).
Graph: Functional connectivity between immature and mature neurons as determined by Ca 2 + imaging. (Center of H‐shaped separator, i.e., partition, is indicated by broken line.)
By SATORU OKAWA; TAKESHI MITA; DOUGLAS BAKKUM; URS FREY; ANDREAS HIERLEMANN; RYOHEI KANZAKI and HIROKAZU TAKAHASHI
Satoru Okawa (non‐member) received a bachelor's degree in mechanoinformatics from the University of Tokyo in 2013. His student research dealt with dissociated neuron culture systems. He subsequently joined Talknote, Inc. He is now involved in the development and operation of social collaboration and communication tool services.
Takeshi Mita (non‐member) received a bachelor's degree in mechanical control systems from Shibaura Institute of Technology in 2003, and in 2003–2008 was engaged in the development of test systems for CCD image sensors at the Sony Corporation Atsugi Technology Center. He completed the M.E. program in mechanoinformatics at the University of Tokyo (Graduate School of Information Science and Technology) in 2011 and is now in the doctoral program at the same university. Research interests include information coding in neural networks.
Douglas Bakkum (non‐member) graduated from the M.E. program in bioengineering at Georgia Institute of Technology and Emory University in 2008 and became a researcher at the Research Center for Advanced Science and Technology, University of Tokyo (JSPS Special Fellow). Since 2010 he has been a Swiss National Science Foundation Ambizione research scientist at ETH Zurich. Research interests include MEA‐based investigation of cortical neuron development, mechanisms of memory, learning and creativity.
Urs Frey (non‐member) completed the doctoral program in physical electronics at ETH Zurich in 2008. He holds a D. Eng degree. He was engaged in research on CMOS microelectrode arrays at the Bioenginee ring Laboratory, ETH Zurich, in 2008–09 and in the design of mixed‐signal circuits for nonvolatile memory devices at IBM Research—Zurich in 2009–10. Since 2011, he has been an international researcher at RIKEN Quantitative Biology Center. Research interests include CMOS‐driven bioelectronics and biosensors.
Andreas Hierlemann (non‐member) completed the doctoral program in physics at Eberhard‐Karls‐Universität Tübingen. From 1997 to 1998 he was a postdoctoral research scientist at Texas A&M University and Sandia National Laboratories. Then he moved to ETH Zurich in 1999. He was an associate professor at ETH Zurich (Microsensorics) in 2004, professor at the Department of Biosystems Science and Engineering in 2008. Research interests include measurement of neurons, cardiomyocytes, and other biomaterials using CMOS‐based microdevices, microfluidic technologies for cell manipulation and examination of cell characteristics.
Ryohei Kanzaki (non‐member) completed the doctoral program in biology at the University of Tsukuba in 1986. He holds a D.Sc. degree. He was a postdoctoral research scientist in neurobiology at the University of Arizona in 1987 and research associate at the University of Tsukuba in 1991, subsequently lecturer and associate professor and professor 2003. He was a professor of mechanoinformatics at the University of Tokyo (Graduate School of Information Science and Technology) in 2004 and professor at Research Center for Advanced Science and Technology in 2006. He had received JSPCB Yoshida Memorial Award in 2008. Research interestsinclude neurobehavioral science, biomechanical hybrid systems. He has membership from ISN, ZSJ, JSPCB and others.
Hirokazu Takahashi (member) completed the doctoral program in industrial mechanical engineering at the University of Tokyo (Graduate School of Engineering) in 2003. He holds a D.Eng. degree. He had joined the faculty of the University as a research associate in industrial mechanical engineering (Graduate School of Engineering); lecturer in mechanoinformatics at the Graduate School of Information Science and Technology 2004; lecturer at the Research Center for Advanced Science and Technology since 2006. JST PRESTO researcher (decoding and control of brain information) 2008–2012. Research interests include welfare engineering, sensory substitution devices, auditory physiology and other fields at the interface of medicine and engineering. Membership: JSMBE and others.