Illustrating the biological complexity of neuronal networks
“The brain is the last and grandest biological frontier, the most complex thing we have yet discovered in our universe. It contains hundreds of billions of cells interlinked through trillions of connections. the brain boggles the mind”, 1992, James D. Watson, Nobel prize in Physiology or Medicine 1962.
Since the seminal work of Ramon y Caral, neurons have been identified as the prime component of nervous systems. Neuroscience has since described how and when neurons are recruited either by external sensory stimulus, or when the animal exhibits different behaviors. Neurons have the particular capacity to actively transmit signals by action potential (AP), which is produced when the summation of the signals received on its dendrite tree meets certain thresholds at the soma. Interestingly, while growing within a population of neurons, the increasing number of excitatory synapses drive this population into avalanches of activity spreading into the network, which locally can recruit individual neurons.
The objective of this project is to illustrate the complexity of such neuronal population activity focusing on these ambiguous interactions between neuronal firing creating avalanches versus avalanches recruiting neurons.
In neuroscience in vivo, electrophysiological software provides real-time representation of the recordings of 4 wires glued together: the tetrodes. Such quadruple recording of the same AP allows to perform a sort of triangulation of the actual neurons (called neuronal clustering) which attributes each recorded AP to individual neurons univocally. This task is partially supervised, usually through user-based comparison of the amplitude of each AP recorded in the four different wires. Once single neurons are identified, the whole population’s activity could be then represented in a more topological manners or being correlated to either sensory stimuli or a specific behavior.
Surprisingly single neuron identification is still not performed in Neuroscience in vitro, probably because the recording obtained on Multi-Electrodes-Arrays (MEA) are singled-recording sites thus not providing four independent recordings at the same location as tetrodes. Nevertheless, several lines of classification have been proposed classifying either neuronal action potential (Chiappalone, Martionoia 2006) or directly rhythms themselves (Gullo Wanke 2010-2012, DeMarse Potter 2005). The focus of this project is to illustrate action potentials, in real-time; the following stretchable objective will be to try unsupervised classifications for single neuron identifications which would give similar understanding, as obtained in-vivo, for possible topographical networks descriptions in vitro.
II Existing examples/ codes
The software’s market which monitor neuronal activity is large and just a chosen few are here described:
- In vivo, belonging to the lower price range, Neuronexus and Open-Ephys represent AP in real time in each channels, therefore illustrating the dimension in which the neuronal identification (“clustering”) will be performed. Real-time options for stimulations can be implemented. Open-Ephys is an open-source software with available scripts (https://github.com/open-ephys).
- In vitro, the market is mostly occupied by Multi-channels system. Their software first handle their amplifier, providing both recording and stimulation options. Mc-Rack (©Multichannels) represents simultaneously, in real time, up to 254 recording channels. This real time representation includes time and size scaling features. Mc-Rack provides several extra-options to the user, such as filtering (high middle low band), pausing, rewinding, enlarging single channels… Such handling could be included in the project as secondary stretchable objectives. As recently published, hardware based calculus can allow real time processing of biological results, as for example single neuron monitoring with the “neuronal response clamp” (Wallach Marom 2009).
- In addition, the open-source community for brain interfacing also contains a few illustration software examples. Back-Yard brain is specialized in insect electrophysiology and other basic electrophysiological signal (https://github.com/BackyardBrains); their system also work on cell phone through their Application. Open-Brain Computer Interface more specialized into EEG monitoring (http://docs.openbci.com/Getting%20Started/00-Welcome or https://github.com/OpenBCI).
III List of possible functions and representations
This project will first grow from an existing dataset in culture and a Matlab (©Matlab) suite that was created by Raphael Tinarrage back in 2013 and 2014, improved by Anton François in 2015 and recently debugged by Jérémie Sibille in 2017. This suite includes 1- extraction of action potentials (filtering, extraction, selection) 2- concatenation and cutting recordings, 3- different non-supervised classification of action potentials (Mixmod toolbox, http://www.mixmod.org/), 4- representation of AP classes, 5- quantification of bursts activations (avalanches), 6- classification of these bursts (Mixmod) 7- representation of both classified bursts together with classified AP. Additional in vivo functions could be added to this list (Barthos Buzsaki 2004) but should be treated after the previous functions, providing another stretchable objective. To guarantee a deliverable outcome within the timeline of the summer school, a good evaluation of the steps in the prototyping of a live illustration software should be initially settled on the basis of the student team composition, and the wishes they have for one or the other direction.
The second main objective of this project would be to actually reproduce the previous functions in a different language than Matlab. This decision should be made by the team if sufficient progress is done on this first objective and depending on the knowledge of the student’s team members. Potential outcomes from the In-vitro-Open-Amplifier will probably influence the choice of the language or other particular format-related development. Aiming at real-time analysis -machine language based- could open the door to real-time monitoring of single neurons within a population (Zrenner Marom 2010, Wallach Marom 2011).
IV Timeline - Releasable prototype
The timeline for this project is dictated by the time needed for the student to reach the first aim. Using the Matlab’s suite and the corresponding dataset to create a real-time illustration of some of the features calculated post-hoc is the first expected prototype. From there any secondary stretchable objective can be pursue and will be based on the student’s motivations. Any connections with the In-vitro-Open-Amplifier will be highly recommended.
V Future Directions
The future of this project lies in the growing of a fully developed open-source software, intended for educational purposes. The execution of this full scale project includes the following aims:
1- To provide the user with a synchronous multiple representation of: action potentials, burst dynamics, and as much as possible, single AP vs single neuron within this burst;
2- To store the obtained results from real-time analysis in a dynamic database including: the timing action potential observation, the potential neuronal identification, the burst size shape and classes.
3- To implement stimulatory feedback onto the culture based on the outcomes of the results obtained in the database handling.
The core constrains is to keep a real-time representation, calculus or stimulations so any addition into the existing project should be carefully manufacture in such a way it doesn’t slow down the rest.
The In-vitro-Open-Amplifier is a twin project to this software. Both these projects would constitute complete open-source system that could provide an alternative open-source material for research or educational purposes. Such cheap material will pave the way for educational neuroscientific games.