- Email: [email protected]
Induced functional connectivity in hippocampal cultures using Hebbian electrical stimulation
Neurocomputing 151 (2015) 4–10
Contents lists available at ScienceDirect
Neurocomputing journal homepage: www.elsevier.com/locate/neucom
Induced functional connectivity in hippocampal cultures using Hebbian electrical stimulation J.M. Ferrández a,b,n, V. Lorente b, F. de la Paz c, E. Fernández a a
Instituto de Bioingeniería, Universidad Miguel Hernández, Alicante, Spain Departamento de Electrónica, Tecnología de Computadores y Proyectos, Universidad Politécnica de Cartagena, Spain c Departamento de Inteligencia Artificial, UNED, Spain b
art ic l e i nf o
a b s t r a c t
Article history: Received 19 May 2014 Accepted 23 September 2014 Available online 31 October 2014
In this article a bio-hybrid system based on neural cultured is described and the learning processes for programming this biological neuroprocessor are revised. Different authors proposed many different learning techniques for managing neural plasticity, however it is necessary to provide a formal methodology for verifying this induced plasticity and validating this bio-hybrid programming paradigm. We used different electrical stimulation protocols, such as low-frequency current stimulation and tetanic voltage stimulation, on dissociated cultures of hippocampal cells to study how neuronal cultures could be trained with this kind of stimulations. We show that persistent and synchronous stimulation of adjacent electrodes may be used for creating adjacent physical or logical connections in the connectivity graph following Hebb's Law. & 2014 Elsevier B.V. All rights reserved.
Keywords: Hebbian Law Cultured neural network Induced plasticity Learning
1. Introduction Biological brains use millions of biological processors, with dynamic structure, slow commutations compared with silicon circuits [1,12], with low power consumption and unsupervised learning. In our previous works we used silicon models [2,3] of neurophysiological studies of visual systems [4] for building neuromimetic platforms. The use of dissociated cortical neurons cultured onto MEAs represents a useful experimental model to characterize both the spontaneous behavior of neuronal populations and their activity in response to electrical and pharmacological changes, and permit the construction of real biological platforms. Learning is a natural process that needs the creation and modulation of sets of associations between stimuli and responses. Many different stimulation protocols have been used to induced changes in the electrophysiological activity of neural cultures looking for achieve learning [11,7] and low-frequency stimulation has brought good results to researchers enhancing bursting activity in cortical cultures [13,14]. Hebbian learning describes a basic mechanism for synaptic plasticity wherein an increase in synaptic efficacy arises from the presynaptic cell's repeated and persistent stimulation of the postsynaptic cell. The theory is commonly evoked to explain some types of associative learning in which simultaneous activation of cells leads to pronounced increases in synaptic strength. Basically the efficiency of a synaptic connection is increased when presynaptic activity is
synchronous with post-synaptic activity. In this work, we use this kind of stimulation to create adjacent physical or logical connections in the connectivity graphs using Hebb's Law. In previous articles, we used a specific low-frequency current stimulation on dissociated cultures of hippocampal cells to study how neuronal cultures could be trained with this kind of stimulation [8,9]. We showed that persistent and synchronous stimulation of adjacent electrodes may be used for creating adjacent physical or logical connections in the connectivity graph following Hebb's Law. In later experiments, we have used different parameters for this stimulation and tetanic voltage stimulation to check if those connections can be created stimulating with different configurations. The results provided in this article show that low-frequency stimulation and tetanic stimulation can create adjacent connections with different amplitude values modifying the electrical characteristics of the acquired neural responses. The outline of the article is as follows. We present the methods for addressing Hebbian Learning through electrical stimulation. The following section shows the results obtained using specific stimulation with our experimental setup on hippocampal cultures in order to train them. We conclude by discussing some crucial aspects of the research and the remaining challenges.
2. Methods 2.1. Cell culture preparation
n
Corresponding author. E-mail address: [email protected] (J.M. Ferrández).
http://dx.doi.org/10.1016/j.neucom.2014.09.069 0925-2312/& 2014 Elsevier B.V. All rights reserved.
Dissociated cultures of hippocampal CA1-CA3 neurons were prepared from E17.5 sibling embryos (Fig. 1). During the extraction
J.M. Ferrández et al. / Neurocomputing 151 (2015) 4–10
of the hippocampus a small amount of cortical tissue will have inevitably also been included. Tissue was kept in 2 ml of HBSS. 10 mg/ml of trypsin was added to the medium and placed in a 37 1C water bath for 13 min for subsequent dissociation. The tissue was then transferred to a 15 ml falcon containing 4 ml of NB/FBS and triturated using combination of fine pore fire polished Pasteur pipettes (Volac). Cells were then transferred onto 12 well plates (Corning Incorporated) containing glass coverslips (Thermo Scientific). The coverslips were pre-treated overnight with PDL (50 mg/ml), a synthetic molecule used as a coating to enhance cell attachment. The PDL was then aspirated away and the coverslips washed twice with PBS. This was then followed by a final coating of laminin (50 μg/ml), a protein found in the extracellular matrix, to further help anchor the dissociated hippocampal cells. The cells were maintained in a mixture of 500 ml NB/B27 (promotes neural growth) and 500 ml NB/FBS (promotes glial growth), each supplemented with Glutamax and Pen/Strep (dilution 1/100). Glutamax improves cells viability and growth while preventing build up of ammonia and Pen/Strep helps to prevent any infections. Cell density for each coverslip was roughly 200000 cells. Cells were kept in an incubator at 37 1C in 6% CO2. 2.2. Experimental setup Microelectrode arrays (Multichannel systems, MCS) consisted of 60 TiN/SiN planar round electrodes (200 μm electrode spacing, 30 μm electrode diameter) arrange in a 8 8 grid were used. The activity of all cultures was recorded using a MEA60 System (MCS). After 1200X amplification, signals were sampled at 10 kHz and acquired through the data acquisition card and MCRack software
5
(MCS). Electrical stimuli were delivered through a two-channel stimulator (MCS STG1002) to each pair of electrodes. 2.3. Experimental protocol A total of 24 cultures was used in five experiments of 2–3 weeks duration. In every experiment 4–5 cultures were stimulated with a specific electrical stimulation protocol. In experiments E1 to E3 a low frequency current stimulation with different parameters for each experiment was used. Experiments E4 to E5 used a more aggressive stimulation called Tetanization. Experiments were started when neural cultures had 14 days in Vitro (DIV) and were carried out during 2–3 weeks. Table 1 summarizes the experiments and the stimulation parameters applied to the cultures. In experiment1 (E1), cultures 48 to 52 were stimulated during 16 days with trains of five biphasic pulses cathodic-first (50 μA peak, 100 μs phase, 50 ms Interval), that were delivered every 3 s for 10 minutes. In experiment2 (E2), cultures 68 to 72 were stimulated during 10 days with trains of five biphasic pulses cathodic-first (60 μA peak, 100 μs phase, 50 ms Interval), that were delivered every 3 s for 8 minutes. Experiment3 (E3), cultures 73– 77 were stimulated during 11 days with trains of five biphasic pulses cathodic-first (40 μA peak, 100 μs phase, 50 ms Interval), that were delivered every 3 s for 8 minutes. Tetanization stimulation was applied for experiments four and five. In experiment4 (E4), cultures 78 to 82 were stimulated during 10 days with trains of 20 biphasic pulses cathodic-first (600 mV peak, 400 μs phase, 50 ms Interval), that were delivered every 3 s for 10 minutes. In experiment5 (E5), cultures 73 to 76 were stimulated during 13 days with trains of 20 biphasic pulses cathodic-first (400 mV peak, 400 μs phase, 50 ms Interval), that were delivered every 3 s for 10 minutes. In every experiment, the electrophysiological activity of the cultures was previously analyzed and connectivity diagrams based on cross-correlation were obtained for each culture. Two pairs of electrodes with an acceptable spiking activity and no logical connections between them were selected for stimulation. The following days every stimulation session would follow these steps: 1. Spontaneous activity was recorded for 2 min after a recovery period. 2. Cultures were then stimulated through the two pairs of electrodes using the corresponding stimulation protocol. 3. Spontaneous activity was recorded for 2 min after the stimulation.
2.4. Analysis performed
Fig. 1. Hippocampal CA1-CA3 culture (21 DIV) on a microelectrode array.
We observed the spontaneous activity of the cultures before and after the stimulation experiments, as well as their evoked response to the applied stimulus. Extensive burst analysis, poststimulus time histograms and functional connectivity were the main analysis performed to the registered data. Functional connectivity [5,6] captures patterns of deviations from statistical independence between distributed neurons units,
Table 1 Experiments, stimulation durations in days, cultures identifiers, and stimulation parameters. Experiment
Duration
Cultures ID
Stimulation Parameters
E1 E2 E3 E4 E5
16 10 11 10 13
48–52 68–72 73–77 78–82 83–86
Low freq.: 5x(50 mA, 100, 50, 100, 20 Hz), 3 s ITI, 100 Low freq.: 5x(7 60 mA, 20 Hz), 3 s ITI, 80 Low freq.: 5x(7 40 mA, 20 Hz), 3 s ITI, 80 Tetani: 20x(600 mV, 400 ms, 600 mV, 400 ms, 50 ms IPI), 6 s ITI, 100 Tetani: 20x(400 mV, 400 ms, 400 mV, 400 ms, 50 ms IPI), 6 s ITI, 100
6
J.M. Ferrández et al. / Neurocomputing 151 (2015) 4–10
measuring their correlation/covariance, spectral coherence, or phase locking. Functional connectivity is often evaluated among all the elements of a system, regardless whether these elements are connected by direct structural links; moreover, it is highly timedependent (hundreds of milliseconds) and model-free, and it measures statistical interdependence (e.g. mutual information) without explicit reference to causal effects. Correlation and information theory-based methods are used to estimate the functional connectivity of in vitro neural networks: Cross-correlation, Mutual Information, Transfer Entropy, and Joint Entropy. Such methods need to be applied to each possible pair of electrodes, which shows spontaneous electrophysiological activity. For each pair of neurons, the connectivity method provides an estimation of the connection strength (one for each direction). The connection strength is supposed to be proportional to the value yielded by the method. Thus, each method is associated to a matrix, the Connectivity Matrix (CM), whose elements (X, Y) correspond to the estimated connection strength between neuron X and Y. High and low values in the CM are expected to correspond to strong and weak connections. By using such approach, inhibitory connections could not be detected because they would be mixed with small connection values. However, non-zero CM values were also obtained when no apparent causal effects were evident, or no direct connections were present among the considered neurons. In our experiments, connectivity maps offered a visualization of the connectivity changes that occur in the culture. Connectivity maps were generated using the connectivity matrix (CM) obtained after applying the analysis and Cross-Correlation or Mutual Information. By setting thresholds in the CM, it is possible to filter out some small values that may correspond to noise or very weak connections. In consequence, these maps show the strongest synaptic pathways, and can be used for visualizing the neural weights dynamics, and validate the achieved learning. Spike parameters, such as height and width of positive and negative voltages, as the evolution of the spiking activity were also subject of research. We were very interested in analyzing how the different stimulations could affect the electrical properties of the cultures.
3. Results Low-frequency current stimulation and tetanic stimulation had both an impact on the electrophysiological responses of the cultures, as previous studies had reported [7,10]. Raster plots showed that all of the stimulations provided induce changes in the firing frequency of the cultures. We can observe some kind of reorganization in the firing activity, from a nearly random continuous spiking activity to a more discrete bursting spiking activity. After the third week in vitro, this bursting activity becomes more frequent and robust and this effect is much more evident than during the first weeks (Fig. 2). First experiments showed that this effect takes effect initially with low-frequency stimulation, however from our last experiments it may be concluded that both stimulations have a frequency impact on the spiking activity of the culture. It may be concluded that this effect is more evident using low-frequency neural stimulation. The change on the spiking activity of the cultures can also be seen analytically by observing the instantaneous firing frequencies (Fig. 3) and the interspike intervals of the neural cultures. Instantaneous firing frequency graphs shows that stimulated electrodes start firing in more separated period of times after stimulation but each firing period last longer. In addition, interspike intervals show the same results observed in the spiking periods but also it can be seen analytically the ISI decrease both in value and dispersion. Both effects are related to the neural
Fig. 2. Raster plots extracted from cultures of experiments E2 and E5. (a) (21DIV) and (b) (32DIV) belong to ID68 from E2, (c) (23DIV) and (d) (30DIV) belong to ID86 from E5. Each figure is divided in two graphs, which show the spiking activity of the culture before and after stimulation. Raster plots show a change in the spiking activity, changing from a uniform activity before stimulation to a more concentrated activity after stimulation. This result is emphasized after the third week in vitro due to maturing occurred in the cultures.
J.M. Ferrández et al. / Neurocomputing 151 (2015) 4–10
stimulation, which modulates the firing capabilities of the cultures. Connectivity diagrams based on cross-correlation between electrodes showed some kind of connections reorganization after stimulations, concentrating them in a few electrodes. Furthermore, adjacent physical or logical connections in the connectivity graph following Hebb's law appeared in some pairs of stimulated electrodes (Fig. 4). Electrodes with created connections between them can distinctly be detected with the instantaneous firing frequencies graphs. Fig. 3 showed two pair of stimulated electrodes (12, 21 and 38, 47) before and after the stimulation session. The firing periods of the electrodes from the second pair follow exactly each other, whereas the firing periods of the first pair of electrodes do not match. Furthermore, the electrodes of the second pair change both the firing periods after stimulation. This features indicates that there exists a strong connection between them. Functional connectivity analyses are based on cross-correlation of the selected electrodes. Cross-correlation provides the conditioned probability of an action potential in the target electrode at
7
time t0 þt, given that an action potential has been detected in source electrode at time t. Fig. 5A shows the cross-correlation evolution in time of electrodes 31 and 42 in experiment E1. It can be observed that a persistent and continuous stimulation over two initially unpaired electrodes induces an increase in their correlation indexes, with values over 0.3. Seven days of stimulation were needed for modifying the functional connectivity of the electrodes, thus for creating a logical connection between both of them, as seen in Fig. 5B. When neurons are cultured for 30 days (30 days In Vitro (DIV)), a degradation process becomes more evident, with a decrease in the correlation indexes and the loss of the created functional connections. Connectivity maps represent the functional connectivity of the neural cultures showing only the best three output logical connections per electrode in terms of strongest correlation. The creation of a new logical connection implies that this new value is between the three best values compared with the rest of the electrodes. Functional connections may be bi-directional or not, as in this case. Taking electrode 31 as reference and electrode-42 as target, there
Fig. 3. Instantaneous firing frequencies of stimulated electrodes 12–21 and 38–47 in the culture ID68 (22DIV) from E2 before (a) and after (b) stimulation. A change in the spiking periods can clearly be seen, with less active frequencies and with longer periods. This culture only created a connection between the second pair of electrodes (38, 47), which had an impact on the instantaneous firing frequencies. Connected electrodes fire in the same firing periods, whereas not connected electrodes had no firing relation.
Fig. 4. Connectivity graphs based on cross-correlation between electrodes. The graph belongs to the culture ID48 (E1) at 25 DIV. Pair of electrodes 31, 42 and 52, 53 were stimulated with low-frequency current stimulation with 50 mA biphasic pulses. (a) No logical connections were observed before stimulation. (b) A connection (red arrow) between electrodes 31 and 42 has appeared.
8
J.M. Ferrández et al. / Neurocomputing 151 (2015) 4–10
Fig. 6. Cross correlation temporal evolution between electrode-42 and electrode31 in culture 48 (experiment E1).
Fig. 5. A. Cross correlation temporal evolution of electrodes 31 and 42 in culture 48 (experiment E1). Y-axis represents cross-correlation probability and X-axis represents each registered data, labeled with date acquired. Red dots represents the moment in which logical connections appeared from origin electrode to source electrode. B. Culture 48 connectivity map.
exists a positive evolution on the correlation that induces the origin of logical connections. However the inverse- correlation values are very low (maximum probability is below 0.012), denoting that no logical output connection is created in electrode 42, and that the induced connection paths are oriented, Fig. 6. Analyzing spike parameters such as peaks heights and widths and number of spikes has lead us to an important result. Both types of stimulation, low frequency and tetanic stimulation, produced a reactivation of neurons over time, which led to the creation of adjacent physical or logical connections in the connectivity graph following Hebb's Law. Fig. 7 shows the mean values of spike parameters of stimulated electrodes for each register number. Fig. 7A and B belong to the culture ID48 from E1, stimulated with low frequency. We can observe in these figures that each electrode presented a good evolution of its parameters until register number 20. Negative peak height decreased, positive peak height increased, width of negative peak slightly increased, and number of spikes increased. This positive evolution of the parameters was due to the activation of more new neurons near the electrodes, which produced a higher value of the spikes parameters. Fig. 7C and D belong to the culture ID86 from E5, stimulated with tetanic stimulation. The first electrode stimulated showed a positive evolution of its parameters until register number 10, whereas the second electrode became nearly stable until register number 12. In this case, the neural activation took place in the first electrodes, which lead to the creation of few connections over time. The changes in the spikes parameters can also be seen clearly
in Fig. 8, which shows the spikes waveform from cultures ID68 and ID86 at the stimulated electrode 38 and 62, respectively. The figure presents the spikes waveform at early stages pre-stimulation (blue) and late stages post-stimulation (green). The acquired neural spikes increase their number and become bigger and more consistent (less variant slope), due to the activation of more neural aggregates around the target electrode with more mature and permanent electrical responses, which induces a more solid and synchronous activation related with paired electrode. All E1 cultures created a connection between the paired stimulated electrodes, whereas 60% of the cultures of E3 and E4, and only 20% of the cultures of E2 and E5 showed that connection. In some cases, the connection was intermittent, lasting one to several days. In others, a persistent connection was created. Finally, some cultures did not create any kind of connections. In this way, Hebbian tetanization created ad-hoc permanent or transient logical connections by modifying the efficiency of the paths between the selected electrodes. We speculate that unconnected electrode cultures may be caused by a not-homogeneous culture growth between the electrodes or by the neurobiological properties of the connections as still need to be confirmed using histological techniques in future works. In this case, using low-frequency current stimulation with 50 μA biphasic pulses provided the best results for creating connections following Hebb's law. Finally it can be concluded that low-frequency stimulation is better in terms of neural response stability, culture duration, and connection-oriented processes than tetanization that evokes a more aggressive process in the hippocampal culture.
4. Conclusions Learning in biological neural cultures is a challenging task. Different authors have proposed different methods for inducing a desired and controlled plasticity over the biological neural structure. Low-frequency stimulation and tetanization have brought good results to researchers enhancing bursting activity in cortical cultures. In this article, we have shown that using these kind of stimulations it is possible to create adjacent physical or logical connections in the connectivity graph following Hebb's Law and such connections induce changes in the electrophysiological response of the cells in the culture, which can be observed in the different analysis performed. Furthermore, low-frequency stimulation induces permanent changes in most experiments using different values of current amplitude and stimulation patterns. Persistent and synchronous stimulation of relevant adjacent electrodes may be used for strengthening the efficiency of their connectivity graph. These processes may be used for imposing a desired behavior over the network dynamics. In this work, different stimulation procedures are described in order
J.M. Ferrández et al. / Neurocomputing 151 (2015) 4–10
9
Fig. 7. For each electrode, Mean Negative Peak Height, Mean Positive Peak Height and Spike Number are represented. A) and B) graphs are for Electrode 31 and 42 from culture ID48. C) and D) graphs are from Electrode 62 and 72 from culture ID86.
10
J.M. Ferrández et al. / Neurocomputing 151 (2015) 4–10
Fig. 8. Spikes waveform from cultures of ID68 and ID86 of experiments E2 and E5, respectively. (a) Spikes waveforms from electrode 38 at 21 DIV (blue) and 32 DIV (green). (b) Spikes waveforms from electrode 62 at 23 DIV (blue) and 30 DIV (green).
to achieve the desired plasticity over the neural cultures, shaping in this way the functional connectivity of the neural culture by modifying their electrical neural responses acquired in the induced connection process. In future works, we will try to refine what are the stimulation optimal parameters for inducing persistent changes in the cultured network or testing how chemicals affect the learning temporal parameters of the experimentation, analyzing the forgetting terms.
José Manuel Ferrández Vicente was born in Elche, Spain. He received the M.Sc. degree in Computer Science in 1995, and the Ph.D. degree in 1998, all of them from the Universidad Politécnica de Madrid, Spain. He is currently Associate Professor at the Department of Electronics, Computer Technology and Projects at the Universidad Politécnica de Cartagena, Head of the Electronic Design and Signal Processing Research Group, and Internationalization Vice-rector at the same University. He is the coordinator of the Spanish and the Iberoamerican Network on Natural and Artificial Computation. His research interests include bioinspired processing, neuromorphic engineering, and low vision prosthesis.
Acknowledgments This work is being funded by grant 12361/FPI/09 from Séneca Foundation, Science and Technology Agency from the region of Murcia and by the project 2010V/PUNED/0011 from Universidad Nacional de Educación a Distancia. References [1] J.A. Anderson, E. Rosenfeld, Neurocomputing: Foundations of Research, MIT Press, Cambridge, MA, 1988. [2] Toledo F.J., Martinez J.J, Ferrandez J.M., FPGA-based platform for image and video processing embedded systems. Programmable Logic, 2007. SPL'07. 2007 3rd Southern Conference on, 171–176, 2007. [3] J.J. Martinez, J.J. Toledo, J.M. Ferrandez, Implementation of a discrete cellular neuron model (DT-CNN) architecture on FPGA, Proc. SPIE 5839 (2005) 332–340, http://dx.doi.org/10.1117/12.608118. [4] J.M. Ferrández, J.A. Bolea, J. Ammermüller, R.A. Normann, E. Fernández, A neural network approach for the analysis of multineural recordings in retinal ganglion cells, in: J. Mira, J.V. Sanchez-Andres (Eds.), Lecture Notes in Computer Science, vol. 1607, Engineering Applications of Bio-Inspired Artificial Neural Networks, 1999, pp. 289-298. [5] Bologna, L.L., Nieus, T., Tedesco, M., Chiappalone, M., Benfenati, F. Lowfrequency stimulation enhances burst activity in cortical cultures during development, Neuroscience 165 (2010) 692–704. [6] V. Braitenberg, Vehicles: Experiments in Synthetic Psychology, MIT Press, Cambridge, MA, 1986. [7] Z.S. Chao, D.J. Bakkum, S.M. Potter, Region-specific network plasticity in simulated and living cortical networks: comparison of the center of activity trajectory (cat) with other statistics, J. Neural. Eng. 4 (3) (2007) 294–308. [8] A.N. Ide, A. Andruska, M. Boehler, B.C. Wheeler, G.J., Brewer, Chronic network stimulation enhances evoked action potentials, J. Neural. Eng. 7 (1) (2010) (February), 16008, http://dx.doi.org/10.1088/1741-2560/7/1/016008. [9] J. Stegenga, J. le Feber, E. Marani, W.L.C. Rutten, Analysis of cultured neuronal networks using intraburst firing characteristics. IEEE transactions on biomedical engineering, 55 (4) (2008) 1382–1390. [10] Y. Jimbo, H.P. Robinson, A. Kawana, Strengthening of synchronized activity by tetanic stimulation in cortical cultures: application of planar electrode arrays, IEEE T. Biomed. Eng. 45 (11) (1998) 1297–1304. [11] L. Landweber, L. Kari, The evolution of cellular computing: Nature's solution to a computational problem, Biosystems 52 (1/3) (1999) 3–13. [12] The FACETS Project, http://facets.kip.uni-heidelberg.de/, University of Heidelberg (2006). [13] M.E. Ruaro, P. Bonifazi, V. Torre, Toward the neurocomputer: image processing and pattern recognition with neuronal cultures, IEEE T. Biomed. Eng. 52 (3) (2005) 371–383. [14] D.A. Wagenaar, J. Pine, S.M. Potter, Searching for plasticity in dissociated cortical cultures on multi-electrode arrays, J. Negative Res. Biomed. 5 (16) (2006), http:/dx.doi.org/10.1186/1477-5751-5-16.
Víctor Lorente Sánchez received the Telecommunications degree from the Polytechnic University of Cartagena, Spain, in 2008 and the Ph.D. Degree in Communications and Information Technology in 2014. In the last years he has been studying the computing capabilities of human neurol cultured cells to define stimulation patters able to modulate the neural activity in response to external stimuli for controlling an autonomous robot.
Félix de la Paz López received his B.Sc and M.Sc degrees in Physics (1995) from Univ. Complutense de Madrid, Spain, and his Ph.D degree in Physics (2003) from Univ. Nacional de Educación a Distancia, Madrid, Spain. He is associate professor at the Artificial Intelligence Department of Univ. Nacional de Educación a Distancia, Spain. His main research interests include navigation and representation tasks in mobile robots.
Eduardo Fernández Jover received the M.D. degree from the University of Alicante, Spain, in 1986 and the Ph.D. degree in Neurosciences in 1990. He is currently Associate Professor at the University Miguel Hernández, Spain, and Director of the Artificial Vision Laboratory at the Bioengineering Institute of the University Miguel Hernández, Spain. In the last years he has been using histological as well as electrophysiological techniques to understand how mammalian retinal cells and the circuitry within the retina can manage and code visual information. He is actively working on the development of a visual neuroprosthesis for the profoundly blind.
Contents lists available at ScienceDirect
Neurocomputing journal homepage: www.elsevier.com/locate/neucom
Induced functional connectivity in hippocampal cultures using Hebbian electrical stimulation J.M. Ferrández a,b,n, V. Lorente b, F. de la Paz c, E. Fernández a a
Instituto de Bioingeniería, Universidad Miguel Hernández, Alicante, Spain Departamento de Electrónica, Tecnología de Computadores y Proyectos, Universidad Politécnica de Cartagena, Spain c Departamento de Inteligencia Artificial, UNED, Spain b
art ic l e i nf o
a b s t r a c t
Article history: Received 19 May 2014 Accepted 23 September 2014 Available online 31 October 2014
In this article a bio-hybrid system based on neural cultured is described and the learning processes for programming this biological neuroprocessor are revised. Different authors proposed many different learning techniques for managing neural plasticity, however it is necessary to provide a formal methodology for verifying this induced plasticity and validating this bio-hybrid programming paradigm. We used different electrical stimulation protocols, such as low-frequency current stimulation and tetanic voltage stimulation, on dissociated cultures of hippocampal cells to study how neuronal cultures could be trained with this kind of stimulations. We show that persistent and synchronous stimulation of adjacent electrodes may be used for creating adjacent physical or logical connections in the connectivity graph following Hebb's Law. & 2014 Elsevier B.V. All rights reserved.
Keywords: Hebbian Law Cultured neural network Induced plasticity Learning
1. Introduction Biological brains use millions of biological processors, with dynamic structure, slow commutations compared with silicon circuits [1,12], with low power consumption and unsupervised learning. In our previous works we used silicon models [2,3] of neurophysiological studies of visual systems [4] for building neuromimetic platforms. The use of dissociated cortical neurons cultured onto MEAs represents a useful experimental model to characterize both the spontaneous behavior of neuronal populations and their activity in response to electrical and pharmacological changes, and permit the construction of real biological platforms. Learning is a natural process that needs the creation and modulation of sets of associations between stimuli and responses. Many different stimulation protocols have been used to induced changes in the electrophysiological activity of neural cultures looking for achieve learning [11,7] and low-frequency stimulation has brought good results to researchers enhancing bursting activity in cortical cultures [13,14]. Hebbian learning describes a basic mechanism for synaptic plasticity wherein an increase in synaptic efficacy arises from the presynaptic cell's repeated and persistent stimulation of the postsynaptic cell. The theory is commonly evoked to explain some types of associative learning in which simultaneous activation of cells leads to pronounced increases in synaptic strength. Basically the efficiency of a synaptic connection is increased when presynaptic activity is
synchronous with post-synaptic activity. In this work, we use this kind of stimulation to create adjacent physical or logical connections in the connectivity graphs using Hebb's Law. In previous articles, we used a specific low-frequency current stimulation on dissociated cultures of hippocampal cells to study how neuronal cultures could be trained with this kind of stimulation [8,9]. We showed that persistent and synchronous stimulation of adjacent electrodes may be used for creating adjacent physical or logical connections in the connectivity graph following Hebb's Law. In later experiments, we have used different parameters for this stimulation and tetanic voltage stimulation to check if those connections can be created stimulating with different configurations. The results provided in this article show that low-frequency stimulation and tetanic stimulation can create adjacent connections with different amplitude values modifying the electrical characteristics of the acquired neural responses. The outline of the article is as follows. We present the methods for addressing Hebbian Learning through electrical stimulation. The following section shows the results obtained using specific stimulation with our experimental setup on hippocampal cultures in order to train them. We conclude by discussing some crucial aspects of the research and the remaining challenges.
2. Methods 2.1. Cell culture preparation
n
Corresponding author. E-mail address: [email protected] (J.M. Ferrández).
http://dx.doi.org/10.1016/j.neucom.2014.09.069 0925-2312/& 2014 Elsevier B.V. All rights reserved.
Dissociated cultures of hippocampal CA1-CA3 neurons were prepared from E17.5 sibling embryos (Fig. 1). During the extraction
J.M. Ferrández et al. / Neurocomputing 151 (2015) 4–10
of the hippocampus a small amount of cortical tissue will have inevitably also been included. Tissue was kept in 2 ml of HBSS. 10 mg/ml of trypsin was added to the medium and placed in a 37 1C water bath for 13 min for subsequent dissociation. The tissue was then transferred to a 15 ml falcon containing 4 ml of NB/FBS and triturated using combination of fine pore fire polished Pasteur pipettes (Volac). Cells were then transferred onto 12 well plates (Corning Incorporated) containing glass coverslips (Thermo Scientific). The coverslips were pre-treated overnight with PDL (50 mg/ml), a synthetic molecule used as a coating to enhance cell attachment. The PDL was then aspirated away and the coverslips washed twice with PBS. This was then followed by a final coating of laminin (50 μg/ml), a protein found in the extracellular matrix, to further help anchor the dissociated hippocampal cells. The cells were maintained in a mixture of 500 ml NB/B27 (promotes neural growth) and 500 ml NB/FBS (promotes glial growth), each supplemented with Glutamax and Pen/Strep (dilution 1/100). Glutamax improves cells viability and growth while preventing build up of ammonia and Pen/Strep helps to prevent any infections. Cell density for each coverslip was roughly 200000 cells. Cells were kept in an incubator at 37 1C in 6% CO2. 2.2. Experimental setup Microelectrode arrays (Multichannel systems, MCS) consisted of 60 TiN/SiN planar round electrodes (200 μm electrode spacing, 30 μm electrode diameter) arrange in a 8 8 grid were used. The activity of all cultures was recorded using a MEA60 System (MCS). After 1200X amplification, signals were sampled at 10 kHz and acquired through the data acquisition card and MCRack software
5
(MCS). Electrical stimuli were delivered through a two-channel stimulator (MCS STG1002) to each pair of electrodes. 2.3. Experimental protocol A total of 24 cultures was used in five experiments of 2–3 weeks duration. In every experiment 4–5 cultures were stimulated with a specific electrical stimulation protocol. In experiments E1 to E3 a low frequency current stimulation with different parameters for each experiment was used. Experiments E4 to E5 used a more aggressive stimulation called Tetanization. Experiments were started when neural cultures had 14 days in Vitro (DIV) and were carried out during 2–3 weeks. Table 1 summarizes the experiments and the stimulation parameters applied to the cultures. In experiment1 (E1), cultures 48 to 52 were stimulated during 16 days with trains of five biphasic pulses cathodic-first (50 μA peak, 100 μs phase, 50 ms Interval), that were delivered every 3 s for 10 minutes. In experiment2 (E2), cultures 68 to 72 were stimulated during 10 days with trains of five biphasic pulses cathodic-first (60 μA peak, 100 μs phase, 50 ms Interval), that were delivered every 3 s for 8 minutes. Experiment3 (E3), cultures 73– 77 were stimulated during 11 days with trains of five biphasic pulses cathodic-first (40 μA peak, 100 μs phase, 50 ms Interval), that were delivered every 3 s for 8 minutes. Tetanization stimulation was applied for experiments four and five. In experiment4 (E4), cultures 78 to 82 were stimulated during 10 days with trains of 20 biphasic pulses cathodic-first (600 mV peak, 400 μs phase, 50 ms Interval), that were delivered every 3 s for 10 minutes. In experiment5 (E5), cultures 73 to 76 were stimulated during 13 days with trains of 20 biphasic pulses cathodic-first (400 mV peak, 400 μs phase, 50 ms Interval), that were delivered every 3 s for 10 minutes. In every experiment, the electrophysiological activity of the cultures was previously analyzed and connectivity diagrams based on cross-correlation were obtained for each culture. Two pairs of electrodes with an acceptable spiking activity and no logical connections between them were selected for stimulation. The following days every stimulation session would follow these steps: 1. Spontaneous activity was recorded for 2 min after a recovery period. 2. Cultures were then stimulated through the two pairs of electrodes using the corresponding stimulation protocol. 3. Spontaneous activity was recorded for 2 min after the stimulation.
2.4. Analysis performed
Fig. 1. Hippocampal CA1-CA3 culture (21 DIV) on a microelectrode array.
We observed the spontaneous activity of the cultures before and after the stimulation experiments, as well as their evoked response to the applied stimulus. Extensive burst analysis, poststimulus time histograms and functional connectivity were the main analysis performed to the registered data. Functional connectivity [5,6] captures patterns of deviations from statistical independence between distributed neurons units,
Table 1 Experiments, stimulation durations in days, cultures identifiers, and stimulation parameters. Experiment
Duration
Cultures ID
Stimulation Parameters
E1 E2 E3 E4 E5
16 10 11 10 13
48–52 68–72 73–77 78–82 83–86
Low freq.: 5x(50 mA, 100, 50, 100, 20 Hz), 3 s ITI, 100 Low freq.: 5x(7 60 mA, 20 Hz), 3 s ITI, 80 Low freq.: 5x(7 40 mA, 20 Hz), 3 s ITI, 80 Tetani: 20x(600 mV, 400 ms, 600 mV, 400 ms, 50 ms IPI), 6 s ITI, 100 Tetani: 20x(400 mV, 400 ms, 400 mV, 400 ms, 50 ms IPI), 6 s ITI, 100
6
J.M. Ferrández et al. / Neurocomputing 151 (2015) 4–10
measuring their correlation/covariance, spectral coherence, or phase locking. Functional connectivity is often evaluated among all the elements of a system, regardless whether these elements are connected by direct structural links; moreover, it is highly timedependent (hundreds of milliseconds) and model-free, and it measures statistical interdependence (e.g. mutual information) without explicit reference to causal effects. Correlation and information theory-based methods are used to estimate the functional connectivity of in vitro neural networks: Cross-correlation, Mutual Information, Transfer Entropy, and Joint Entropy. Such methods need to be applied to each possible pair of electrodes, which shows spontaneous electrophysiological activity. For each pair of neurons, the connectivity method provides an estimation of the connection strength (one for each direction). The connection strength is supposed to be proportional to the value yielded by the method. Thus, each method is associated to a matrix, the Connectivity Matrix (CM), whose elements (X, Y) correspond to the estimated connection strength between neuron X and Y. High and low values in the CM are expected to correspond to strong and weak connections. By using such approach, inhibitory connections could not be detected because they would be mixed with small connection values. However, non-zero CM values were also obtained when no apparent causal effects were evident, or no direct connections were present among the considered neurons. In our experiments, connectivity maps offered a visualization of the connectivity changes that occur in the culture. Connectivity maps were generated using the connectivity matrix (CM) obtained after applying the analysis and Cross-Correlation or Mutual Information. By setting thresholds in the CM, it is possible to filter out some small values that may correspond to noise or very weak connections. In consequence, these maps show the strongest synaptic pathways, and can be used for visualizing the neural weights dynamics, and validate the achieved learning. Spike parameters, such as height and width of positive and negative voltages, as the evolution of the spiking activity were also subject of research. We were very interested in analyzing how the different stimulations could affect the electrical properties of the cultures.
3. Results Low-frequency current stimulation and tetanic stimulation had both an impact on the electrophysiological responses of the cultures, as previous studies had reported [7,10]. Raster plots showed that all of the stimulations provided induce changes in the firing frequency of the cultures. We can observe some kind of reorganization in the firing activity, from a nearly random continuous spiking activity to a more discrete bursting spiking activity. After the third week in vitro, this bursting activity becomes more frequent and robust and this effect is much more evident than during the first weeks (Fig. 2). First experiments showed that this effect takes effect initially with low-frequency stimulation, however from our last experiments it may be concluded that both stimulations have a frequency impact on the spiking activity of the culture. It may be concluded that this effect is more evident using low-frequency neural stimulation. The change on the spiking activity of the cultures can also be seen analytically by observing the instantaneous firing frequencies (Fig. 3) and the interspike intervals of the neural cultures. Instantaneous firing frequency graphs shows that stimulated electrodes start firing in more separated period of times after stimulation but each firing period last longer. In addition, interspike intervals show the same results observed in the spiking periods but also it can be seen analytically the ISI decrease both in value and dispersion. Both effects are related to the neural
Fig. 2. Raster plots extracted from cultures of experiments E2 and E5. (a) (21DIV) and (b) (32DIV) belong to ID68 from E2, (c) (23DIV) and (d) (30DIV) belong to ID86 from E5. Each figure is divided in two graphs, which show the spiking activity of the culture before and after stimulation. Raster plots show a change in the spiking activity, changing from a uniform activity before stimulation to a more concentrated activity after stimulation. This result is emphasized after the third week in vitro due to maturing occurred in the cultures.
J.M. Ferrández et al. / Neurocomputing 151 (2015) 4–10
stimulation, which modulates the firing capabilities of the cultures. Connectivity diagrams based on cross-correlation between electrodes showed some kind of connections reorganization after stimulations, concentrating them in a few electrodes. Furthermore, adjacent physical or logical connections in the connectivity graph following Hebb's law appeared in some pairs of stimulated electrodes (Fig. 4). Electrodes with created connections between them can distinctly be detected with the instantaneous firing frequencies graphs. Fig. 3 showed two pair of stimulated electrodes (12, 21 and 38, 47) before and after the stimulation session. The firing periods of the electrodes from the second pair follow exactly each other, whereas the firing periods of the first pair of electrodes do not match. Furthermore, the electrodes of the second pair change both the firing periods after stimulation. This features indicates that there exists a strong connection between them. Functional connectivity analyses are based on cross-correlation of the selected electrodes. Cross-correlation provides the conditioned probability of an action potential in the target electrode at
7
time t0 þt, given that an action potential has been detected in source electrode at time t. Fig. 5A shows the cross-correlation evolution in time of electrodes 31 and 42 in experiment E1. It can be observed that a persistent and continuous stimulation over two initially unpaired electrodes induces an increase in their correlation indexes, with values over 0.3. Seven days of stimulation were needed for modifying the functional connectivity of the electrodes, thus for creating a logical connection between both of them, as seen in Fig. 5B. When neurons are cultured for 30 days (30 days In Vitro (DIV)), a degradation process becomes more evident, with a decrease in the correlation indexes and the loss of the created functional connections. Connectivity maps represent the functional connectivity of the neural cultures showing only the best three output logical connections per electrode in terms of strongest correlation. The creation of a new logical connection implies that this new value is between the three best values compared with the rest of the electrodes. Functional connections may be bi-directional or not, as in this case. Taking electrode 31 as reference and electrode-42 as target, there
Fig. 3. Instantaneous firing frequencies of stimulated electrodes 12–21 and 38–47 in the culture ID68 (22DIV) from E2 before (a) and after (b) stimulation. A change in the spiking periods can clearly be seen, with less active frequencies and with longer periods. This culture only created a connection between the second pair of electrodes (38, 47), which had an impact on the instantaneous firing frequencies. Connected electrodes fire in the same firing periods, whereas not connected electrodes had no firing relation.
Fig. 4. Connectivity graphs based on cross-correlation between electrodes. The graph belongs to the culture ID48 (E1) at 25 DIV. Pair of electrodes 31, 42 and 52, 53 were stimulated with low-frequency current stimulation with 50 mA biphasic pulses. (a) No logical connections were observed before stimulation. (b) A connection (red arrow) between electrodes 31 and 42 has appeared.
8
J.M. Ferrández et al. / Neurocomputing 151 (2015) 4–10
Fig. 6. Cross correlation temporal evolution between electrode-42 and electrode31 in culture 48 (experiment E1).
Fig. 5. A. Cross correlation temporal evolution of electrodes 31 and 42 in culture 48 (experiment E1). Y-axis represents cross-correlation probability and X-axis represents each registered data, labeled with date acquired. Red dots represents the moment in which logical connections appeared from origin electrode to source electrode. B. Culture 48 connectivity map.
exists a positive evolution on the correlation that induces the origin of logical connections. However the inverse- correlation values are very low (maximum probability is below 0.012), denoting that no logical output connection is created in electrode 42, and that the induced connection paths are oriented, Fig. 6. Analyzing spike parameters such as peaks heights and widths and number of spikes has lead us to an important result. Both types of stimulation, low frequency and tetanic stimulation, produced a reactivation of neurons over time, which led to the creation of adjacent physical or logical connections in the connectivity graph following Hebb's Law. Fig. 7 shows the mean values of spike parameters of stimulated electrodes for each register number. Fig. 7A and B belong to the culture ID48 from E1, stimulated with low frequency. We can observe in these figures that each electrode presented a good evolution of its parameters until register number 20. Negative peak height decreased, positive peak height increased, width of negative peak slightly increased, and number of spikes increased. This positive evolution of the parameters was due to the activation of more new neurons near the electrodes, which produced a higher value of the spikes parameters. Fig. 7C and D belong to the culture ID86 from E5, stimulated with tetanic stimulation. The first electrode stimulated showed a positive evolution of its parameters until register number 10, whereas the second electrode became nearly stable until register number 12. In this case, the neural activation took place in the first electrodes, which lead to the creation of few connections over time. The changes in the spikes parameters can also be seen clearly
in Fig. 8, which shows the spikes waveform from cultures ID68 and ID86 at the stimulated electrode 38 and 62, respectively. The figure presents the spikes waveform at early stages pre-stimulation (blue) and late stages post-stimulation (green). The acquired neural spikes increase their number and become bigger and more consistent (less variant slope), due to the activation of more neural aggregates around the target electrode with more mature and permanent electrical responses, which induces a more solid and synchronous activation related with paired electrode. All E1 cultures created a connection between the paired stimulated electrodes, whereas 60% of the cultures of E3 and E4, and only 20% of the cultures of E2 and E5 showed that connection. In some cases, the connection was intermittent, lasting one to several days. In others, a persistent connection was created. Finally, some cultures did not create any kind of connections. In this way, Hebbian tetanization created ad-hoc permanent or transient logical connections by modifying the efficiency of the paths between the selected electrodes. We speculate that unconnected electrode cultures may be caused by a not-homogeneous culture growth between the electrodes or by the neurobiological properties of the connections as still need to be confirmed using histological techniques in future works. In this case, using low-frequency current stimulation with 50 μA biphasic pulses provided the best results for creating connections following Hebb's law. Finally it can be concluded that low-frequency stimulation is better in terms of neural response stability, culture duration, and connection-oriented processes than tetanization that evokes a more aggressive process in the hippocampal culture.
4. Conclusions Learning in biological neural cultures is a challenging task. Different authors have proposed different methods for inducing a desired and controlled plasticity over the biological neural structure. Low-frequency stimulation and tetanization have brought good results to researchers enhancing bursting activity in cortical cultures. In this article, we have shown that using these kind of stimulations it is possible to create adjacent physical or logical connections in the connectivity graph following Hebb's Law and such connections induce changes in the electrophysiological response of the cells in the culture, which can be observed in the different analysis performed. Furthermore, low-frequency stimulation induces permanent changes in most experiments using different values of current amplitude and stimulation patterns. Persistent and synchronous stimulation of relevant adjacent electrodes may be used for strengthening the efficiency of their connectivity graph. These processes may be used for imposing a desired behavior over the network dynamics. In this work, different stimulation procedures are described in order
J.M. Ferrández et al. / Neurocomputing 151 (2015) 4–10
9
Fig. 7. For each electrode, Mean Negative Peak Height, Mean Positive Peak Height and Spike Number are represented. A) and B) graphs are for Electrode 31 and 42 from culture ID48. C) and D) graphs are from Electrode 62 and 72 from culture ID86.
10
J.M. Ferrández et al. / Neurocomputing 151 (2015) 4–10
Fig. 8. Spikes waveform from cultures of ID68 and ID86 of experiments E2 and E5, respectively. (a) Spikes waveforms from electrode 38 at 21 DIV (blue) and 32 DIV (green). (b) Spikes waveforms from electrode 62 at 23 DIV (blue) and 30 DIV (green).
to achieve the desired plasticity over the neural cultures, shaping in this way the functional connectivity of the neural culture by modifying their electrical neural responses acquired in the induced connection process. In future works, we will try to refine what are the stimulation optimal parameters for inducing persistent changes in the cultured network or testing how chemicals affect the learning temporal parameters of the experimentation, analyzing the forgetting terms.
José Manuel Ferrández Vicente was born in Elche, Spain. He received the M.Sc. degree in Computer Science in 1995, and the Ph.D. degree in 1998, all of them from the Universidad Politécnica de Madrid, Spain. He is currently Associate Professor at the Department of Electronics, Computer Technology and Projects at the Universidad Politécnica de Cartagena, Head of the Electronic Design and Signal Processing Research Group, and Internationalization Vice-rector at the same University. He is the coordinator of the Spanish and the Iberoamerican Network on Natural and Artificial Computation. His research interests include bioinspired processing, neuromorphic engineering, and low vision prosthesis.
Acknowledgments This work is being funded by grant 12361/FPI/09 from Séneca Foundation, Science and Technology Agency from the region of Murcia and by the project 2010V/PUNED/0011 from Universidad Nacional de Educación a Distancia. References [1] J.A. Anderson, E. Rosenfeld, Neurocomputing: Foundations of Research, MIT Press, Cambridge, MA, 1988. [2] Toledo F.J., Martinez J.J, Ferrandez J.M., FPGA-based platform for image and video processing embedded systems. Programmable Logic, 2007. SPL'07. 2007 3rd Southern Conference on, 171–176, 2007. [3] J.J. Martinez, J.J. Toledo, J.M. Ferrandez, Implementation of a discrete cellular neuron model (DT-CNN) architecture on FPGA, Proc. SPIE 5839 (2005) 332–340, http://dx.doi.org/10.1117/12.608118. [4] J.M. Ferrández, J.A. Bolea, J. Ammermüller, R.A. Normann, E. Fernández, A neural network approach for the analysis of multineural recordings in retinal ganglion cells, in: J. Mira, J.V. Sanchez-Andres (Eds.), Lecture Notes in Computer Science, vol. 1607, Engineering Applications of Bio-Inspired Artificial Neural Networks, 1999, pp. 289-298. [5] Bologna, L.L., Nieus, T., Tedesco, M., Chiappalone, M., Benfenati, F. Lowfrequency stimulation enhances burst activity in cortical cultures during development, Neuroscience 165 (2010) 692–704. [6] V. Braitenberg, Vehicles: Experiments in Synthetic Psychology, MIT Press, Cambridge, MA, 1986. [7] Z.S. Chao, D.J. Bakkum, S.M. Potter, Region-specific network plasticity in simulated and living cortical networks: comparison of the center of activity trajectory (cat) with other statistics, J. Neural. Eng. 4 (3) (2007) 294–308. [8] A.N. Ide, A. Andruska, M. Boehler, B.C. Wheeler, G.J., Brewer, Chronic network stimulation enhances evoked action potentials, J. Neural. Eng. 7 (1) (2010) (February), 16008, http://dx.doi.org/10.1088/1741-2560/7/1/016008. [9] J. Stegenga, J. le Feber, E. Marani, W.L.C. Rutten, Analysis of cultured neuronal networks using intraburst firing characteristics. IEEE transactions on biomedical engineering, 55 (4) (2008) 1382–1390. [10] Y. Jimbo, H.P. Robinson, A. Kawana, Strengthening of synchronized activity by tetanic stimulation in cortical cultures: application of planar electrode arrays, IEEE T. Biomed. Eng. 45 (11) (1998) 1297–1304. [11] L. Landweber, L. Kari, The evolution of cellular computing: Nature's solution to a computational problem, Biosystems 52 (1/3) (1999) 3–13. [12] The FACETS Project, http://facets.kip.uni-heidelberg.de/, University of Heidelberg (2006). [13] M.E. Ruaro, P. Bonifazi, V. Torre, Toward the neurocomputer: image processing and pattern recognition with neuronal cultures, IEEE T. Biomed. Eng. 52 (3) (2005) 371–383. [14] D.A. Wagenaar, J. Pine, S.M. Potter, Searching for plasticity in dissociated cortical cultures on multi-electrode arrays, J. Negative Res. Biomed. 5 (16) (2006), http:/dx.doi.org/10.1186/1477-5751-5-16.
Víctor Lorente Sánchez received the Telecommunications degree from the Polytechnic University of Cartagena, Spain, in 2008 and the Ph.D. Degree in Communications and Information Technology in 2014. In the last years he has been studying the computing capabilities of human neurol cultured cells to define stimulation patters able to modulate the neural activity in response to external stimuli for controlling an autonomous robot.
Félix de la Paz López received his B.Sc and M.Sc degrees in Physics (1995) from Univ. Complutense de Madrid, Spain, and his Ph.D degree in Physics (2003) from Univ. Nacional de Educación a Distancia, Madrid, Spain. He is associate professor at the Artificial Intelligence Department of Univ. Nacional de Educación a Distancia, Spain. His main research interests include navigation and representation tasks in mobile robots.
Eduardo Fernández Jover received the M.D. degree from the University of Alicante, Spain, in 1986 and the Ph.D. degree in Neurosciences in 1990. He is currently Associate Professor at the University Miguel Hernández, Spain, and Director of the Artificial Vision Laboratory at the Bioengineering Institute of the University Miguel Hernández, Spain. In the last years he has been using histological as well as electrophysiological techniques to understand how mammalian retinal cells and the circuitry within the retina can manage and code visual information. He is actively working on the development of a visual neuroprosthesis for the profoundly blind.