Final Report Summary - NEUROCMOS (Seamless Integration of Neurons with CMOS Microelectronics)
The NeuroCMOS project was targeted at a seamless integration of state-of–the-art microelectronics with living neuronal cells in order advance the understanding of neuronal behavior. The project included (i) the development of a multifunctional microelectronics chip platform in complementary metal oxide semiconductor (CMOS) technology, which was then used to enable (ii) key neurobiological and neuromedical research on network dynamics and plasticity of rodent neuronal networks and visual encoding in retinae. Moreover, the project included (iii) the concurrent development of algorithms and models to efficiently process and harness the obtained high-resolution electrophysiological data.
During the project, we were able to develop 2 new generations of high-density CMOS-based microelectrode arrays (HD-MEAs) and a real-time feedback scheme. We were able to use HD-MEAs to identify individual presynaptic neurons and their contributions to postsynaptic potentials (PSPs), the propagation and signal characteristics of action potentials along single axons, and to do extensive ganglion cell characterizations in retinal preparations. We could show the involvement of a specific inhibitory neuron type in the pathophysiology of a neurological disease, congenital nystagmus. Moreover, we progressed in handling the huge datasets produced by the devices, in implementing dedicated and automated spike sorting and data evaluation methods, and to decipher the role of noise and signal correlations in neuronal coding.
Technology
After a first design of a bidirectional microelectronics-based high-density microelectrode array system featuring a sensing area of 3.85 x 2.10 mm2, including 26’400 electrodes at a pitch of 17.5 μm and 1024 recording channels in 2013, we designed, fabricated and successfully tested a next-generation chip, which is a truly multi-functional design chip design. The chip system included an array of 59’700 electrodes at 13.5 micron pitch in an active area of 2.4 x 4.5 mm2, along with 2048 action potential voltage recording and 32 current recording units, 32 local field potential recording units, 32 cyclovoltammetry and neurotransmitter measurement units, 64 impedance measurement units, and 16 voltage and 16 current stimulation units. The low circuitry noise along with the high spatial electrode resolution allows for recording at subcellular resolution from neurons and single axons.
Applications
We applied our electrode arrays systems to study plasticity in neuronal networks, signal characteristics and propagation in axons and details of visual coding in retinal preparations. Any neuron lying on our array could be recorded at high spatio-temporal resolution and simultaneously precisely be stimulated with little artifact. Using these features, we were able to identify individual presynaptic neurons and their contributions to postsynaptic potentials (PSPs), including inhibitory and excitatory synaptic inputs during spontaneous activity. We developed strategies to electrically identify any neuron in the network, while subcellular spatial resolution recording of extracellular action potential (AP) traces enabled their assignment to the different neuronal compartments. In collaboration with the Roska group, we simultaneously recorded from all four types of direction-selective ganglion cells in the retina by using HD-MEAs and decoded their concerted activity. Moreover, we could show the involvement of a specific inhibitory neuron type in the pathophysiology of a neurological retinal disease, nystagmus. Finally, we elucidated details of axonal signal propagation characteristics in microtunnels, and the frequency-dependence of axonal signaling.
Data Analysis and Algorithms
HD-MEAs allow for simultaneous recording of extracellular activity of a large number of neurons with every neuron being detected by multiple electrodes. The neuronal spiking events have to be assigned to individual neurons through “spike sorting”. We assessed the performance of independent-component analysis and found that the performance strongly depends on neuronal density and spike amplitude. Additionally, we developed computational complexity optimization techniques, which facilitate the use of spike-sorting algorithms in large multi-electrode-based recording systems. A real-time, low-latency, high-performance hardware architecture of the algorithm was developed that is capable of processing the activity of hundreds of neurons simultaneously. Another approach to spike sorting included Bayes optimal template matching and a combination of Fisher discriminant analysis with optimal filtering. Finally, we were able to demonstrate the advantages of being able to record comprehensive data sets from large numbers of different and identified ganglion cells in order to decipher the role of noise and signal correlations in neuronal coding. We showed that ‘‘noise’’ in neuronal signals comes with a particular structure, which emerges from circuit properties and which counteracts the harmful effect of variability.
During the project, we were able to develop 2 new generations of high-density CMOS-based microelectrode arrays (HD-MEAs) and a real-time feedback scheme. We were able to use HD-MEAs to identify individual presynaptic neurons and their contributions to postsynaptic potentials (PSPs), the propagation and signal characteristics of action potentials along single axons, and to do extensive ganglion cell characterizations in retinal preparations. We could show the involvement of a specific inhibitory neuron type in the pathophysiology of a neurological disease, congenital nystagmus. Moreover, we progressed in handling the huge datasets produced by the devices, in implementing dedicated and automated spike sorting and data evaluation methods, and to decipher the role of noise and signal correlations in neuronal coding.
Technology
After a first design of a bidirectional microelectronics-based high-density microelectrode array system featuring a sensing area of 3.85 x 2.10 mm2, including 26’400 electrodes at a pitch of 17.5 μm and 1024 recording channels in 2013, we designed, fabricated and successfully tested a next-generation chip, which is a truly multi-functional design chip design. The chip system included an array of 59’700 electrodes at 13.5 micron pitch in an active area of 2.4 x 4.5 mm2, along with 2048 action potential voltage recording and 32 current recording units, 32 local field potential recording units, 32 cyclovoltammetry and neurotransmitter measurement units, 64 impedance measurement units, and 16 voltage and 16 current stimulation units. The low circuitry noise along with the high spatial electrode resolution allows for recording at subcellular resolution from neurons and single axons.
Applications
We applied our electrode arrays systems to study plasticity in neuronal networks, signal characteristics and propagation in axons and details of visual coding in retinal preparations. Any neuron lying on our array could be recorded at high spatio-temporal resolution and simultaneously precisely be stimulated with little artifact. Using these features, we were able to identify individual presynaptic neurons and their contributions to postsynaptic potentials (PSPs), including inhibitory and excitatory synaptic inputs during spontaneous activity. We developed strategies to electrically identify any neuron in the network, while subcellular spatial resolution recording of extracellular action potential (AP) traces enabled their assignment to the different neuronal compartments. In collaboration with the Roska group, we simultaneously recorded from all four types of direction-selective ganglion cells in the retina by using HD-MEAs and decoded their concerted activity. Moreover, we could show the involvement of a specific inhibitory neuron type in the pathophysiology of a neurological retinal disease, nystagmus. Finally, we elucidated details of axonal signal propagation characteristics in microtunnels, and the frequency-dependence of axonal signaling.
Data Analysis and Algorithms
HD-MEAs allow for simultaneous recording of extracellular activity of a large number of neurons with every neuron being detected by multiple electrodes. The neuronal spiking events have to be assigned to individual neurons through “spike sorting”. We assessed the performance of independent-component analysis and found that the performance strongly depends on neuronal density and spike amplitude. Additionally, we developed computational complexity optimization techniques, which facilitate the use of spike-sorting algorithms in large multi-electrode-based recording systems. A real-time, low-latency, high-performance hardware architecture of the algorithm was developed that is capable of processing the activity of hundreds of neurons simultaneously. Another approach to spike sorting included Bayes optimal template matching and a combination of Fisher discriminant analysis with optimal filtering. Finally, we were able to demonstrate the advantages of being able to record comprehensive data sets from large numbers of different and identified ganglion cells in order to decipher the role of noise and signal correlations in neuronal coding. We showed that ‘‘noise’’ in neuronal signals comes with a particular structure, which emerges from circuit properties and which counteracts the harmful effect of variability.