Universal interface will help construct living neural networks
American researchers have developed a scalable and inexpensive platform with flexible settings to interface with living neurons and their networks in cell culture and process the resulting data. It is based on photolithography, standard electrophysiology terminals, open-configuration hardware and open-source software. A report of the work has been published in the journal Advanced Science.
The study of the architecture and operation of neuronal networks is based on recording the electrical impulses they emit. Electrophysiological studies of the brains of living organisms play a fundamental role in neuroscience, but their detailed interpretation is difficult due to the complexity of the nervous system's structure. In vitro cultures of living neurons allow us to study their interactions at different levels - single cells, small networks, more complex tissue samples and organoids. On the other hand, the creation of artificial systems from organized populations of neurons with defined properties, geometry and input-output capabilities through electronic platforms is of interest for the development of biological processors and computing systems based on them. Key to such interfaces are biocompatible microelectrode arrays (microelectrode array, MEA) - complex and expensive devices that allow simultaneous recording of electrical signals from different neurons.
Employees of the University of Illinois at Urbana-Champaign and Indiana University in Bloomington, led by Mattia Gazzola (Mattia Gazzola) set out to create a universal and affordable platform for experiments with live neural networks in vitro. The body of the recording device laser-cut from two acrylic plates and connected them with a pivoting hinge on one side and a latch on the other, putting the whole structure on vibration-isolating pads made of soft rubber.
The bottom plate has four sliding guides and 3D printed stops for centering and fixing MEA-chips of different sizes and configurations on which neuronal substrates are incubated (the chips are made by photolithography in a standard clean room). A printed circuit board with a matrix of soldered spring contacts is fixed on top, which align with the MEA electrode substrates when the device is closed. On the outer part of this board are one or more connectors for Intan electrophysiologic electrodes, which are connected via SPI cable to the Open Ephys data acquisition terminal.
The recorder boards allow connection of up to four SPI cables supporting up to 128 digital data streams each, i.e. they allow simultaneous recording of up to 512 channels. Accordingly, MEAs with 59, 128, 256 and 512 pins have been designed for use depending on the application and resources. The system also supports electrical stimulation of neurons using the Stimjim stimulator via spring contacts and optical stimulation with 465-nanometer LEDs and Doric lenses lasers via optical fibers connected at the top and bottom of the MEA chip. For short- and long-term recording, the device is placed in incubators of various sizes and functionalities.
A single 24-hour experiment with 512 channels of 30-kilohertz recording produces about four terabytes of data. To process them, the authors developed a cloud computing system. Its core is a Python-based pipelined computing platform that maximizes user-side customization flexibility. The server side contains key functionality for analysis, including caching, data distribution, filtering, spike and volley recognition, principal component analysis, criticality analysis, and a variety of visualization modules. The software supports high-performance computing systems and integrates with various data processing packages such as H5py, Aim UI, Jupyter server, PyInform/IDTxl, Globus APIs, NeuralEnsemble, Kilosort and scikit-learn.
The platform was successfully tested on two-dimensional mouse neuron cultures, rat brain slices and three-dimensional bioengineered neural tissue organoids. In addition, the researchers modified it for compatibility with a fluorescence microscope to record neuronal activity using calcium ion current indicators simultaneously with electrophysiological observations.
To demonstrate the unit's portability, durability, and reliability, it was transported disassembled nearly 300 kilometers from Urbana-Champaign to Bloomington, reassembled in half an hour, and reproduced the results of previous tests with adequate accuracy. In total, the authors obtained about a thousand hours of recordings over the course of a year during the tests. The cost of the platform was about $2,500 for the 59-channel version and $12,000 for the 512-channel version, which, according to the researchers, is 10 to 25 times cheaper than systems available on the market. The development was named Mind in Vitro (MiV, "mind in a test tube") - so the authors emphasize their goal in the future to use it for computing with live neurons. They have made the hardware configuration and software available to the public.