Automatic optimization of conductance-based models to reproduce the dynamics of the claustrum neurons
Razvan Gamanut*1, Carlos Enrique Gutierrez2, Kenji Doya1
1Neural Computation Unit, Okinawa Institute of Science and Technology Graduate University, Onna, Japan 2SoftBank, Tokyo, Japan
*Email: razvan.gamanut@oist.jp
Introduction The claustrum, a structure having extensive connectivity with the rest of the brain and being involved in attention and many other high-cognitive processes, is still one of the least understood parts of the mammalian nervous system. One of the biggest obstacles, in addition to its thin anatomical structure, is its highly complex dynamics: despite its high connectivity, the claustrum is surprisingly quiet, with baseline activity below 1 Hz, and bursts of activities up to 4-10 Hz with a short lifetime. In order to build a detailed network model of the claustrum that would reveal its principles of functioning, in this work we focused on the dynamic characteristics of single claustrum neurons. Specifically, we used a single compartment, conductance-based model to reproduce the dynamics of claustrum neurons from experiments. The model captured the adaptive firing of the neurons, that we argue is essential for the claustrum function.
Methods We chose two types of excitatory and two types of inhibitory claustrum neurons, analyzed in slice experiments [1]. We recovered the voltage responses of their membranes to different step currents from the publication, together with the measured resting potentials. From the responses of the neurons to the hyperpolarizing step currents we calculated their membrane capacitance and the leak conductance. We used Brian 2 [2] to build Hodgkin-Huxley-type models with the sodium and potassium channels [3] that had previously been fitted to mammalian cortical neurons [4]. To account for the adapting firing, we added a noninactivating muscarinic potassium channel [5]. We optimized fifteen parameters describing the maximum conductances of the channels, the offsets of the voltages, the reversal potentials of the ions and the transition rates of the channels by the nondominated sorting genetic algorithm 3 (NSGA3) [6]. We set the objectives of the algorithm such that the model neurons would reproduce the features of the biological neurons: i. the number of actions potentials, ii. timing of the first spike, and iii. interspike intervals. Results We ran the NSGA3 algorithm for up to 100 generations, multiple times. Solutions that resulted in closely fulfilling the objectives started to appear in the early generations. By monitoring the process through visual inspections of the results we selected the sets of parameters that best reproduced the behaviour of the biological neurons, particularly their adaptive firing during the long current stimulation. Discussion The adaptive firing of the claustrum neurons should strongly account for the known complex behaviour of the claustrum. Therefore, our neuron models can further be integrated in a mesoscale spiking network of the claustrum interacting with the cortex, to make genuine descriptions of the claustrum dynamics and its functions.
JSPS Postdoctoral Fellowships PE20032, P23383, and Kakenhi 23KF0286
1. https://doi.org:10.1007/s00429-018-1731-0 2. https://doi.org:10.7554/eLife.47314 3. https://doi.org:10.1113/jphysiol.1952.sp004764 4. https://doi.org:citeulike-article-id:134404 5. Yamada, W. M., Koch, C. & Adams, P. R. in Methods in neuronal modeling: From synapses to networks 97-133 (MIT Press, 1989). 6. https://doi.org:10.1109/TEVC.2013.2281535
