CNS*2020 Online

Alireza Ghadimi, Luka Milosevic, Suneil Kalia, Mojgan Hodaie, Andres Lozano, William Hutchison, Milos Popovic, Milad Lankarany

Time:
Jul 20, 2020 09:00 PM Eastern Time (US and Canada)
Join Zoom Meeting
https://utoronto.zoom.us/j/93714822044



Background:
Short-term synaptic plasticity (STP) is the dynamic of change in the synaptic weight with respect to the presynaptic spiking activity. Recent studies showed that STP is involved in several brain processes. Several computational works have been done on inferring STP parameters [1-4]; however, these studies utilized in-vitro signals of intracellular recordings (except Ghanbari et al.) from rodents to create their models, which is not necessarily representative of in-vivo human brain dynamics. To this end, we developed a parameter inference method which estimates parameters of STP induced by DBS, verified by experimental data obtained from intracranial recordings of single-neuron activity in surgical patients.

Method:
To acquire spiking activity, two closely spaced microelectrodes were placed in the thalamic ventral intermediate nucleus (Vim) during awake DBS surgeries. One electrode was used to record single-unit spiking activity, while the other was used to deliver stimulation pulses at various frequencies (5Hz, 20Hz, 30Hz, 50Hz, 100Hz, 200Hz). The included data were collected from 15 patients [5]. The narrow stimulus pulses were removed, after which data were high-pass filtered to better isolate single unit activity. Using stimulus artifacts as triggers, we extracted the instantaneous firing rate of neurons in response to each DBS pulse of the 5Hz stimulation trains, and averaged over all inter-pulse intervals. Due to the low stimulation rate (i.e. 5Hz), no STP is induced in this data. The resultant waveform is equivalent to the impulse response. In order to mimic STP behavior, we used the Tsodyks-Markram phenomenological model, which generates the postsynaptic current according to the spiking history of the presynaptic neuron [6]. To reconstruct the firing rate induced by DBS, we give the pulse train of DBS as the input of the Tsodyks-Markram model. The model generates a postsynaptic current in response to DBS pulses. We use this response to make modulated pulse trains that represent the effect of STP by changing the amplitude of each pulse. The modulated pulse train is convolved with the impulse response of the neuron in order to make an estimation of the DBS-induced firing rate. The estimated firing rate is compared with the experimental instantaneous firing rates throughout the stimulation trains at each of the other stimulation frequencies. To achieve the true parameters of the Tsodyks-Markram model we should minimize the error between experimental and estimated firing rate. True parameters should be valid for all frequencies, therefore we define the error function as the average error of 30Hz, 50Hz, 100Hz, and 200Hz frequencies. To minimize the error, we employ Bayesian Adaptive Direct Search [7], which is a fast non-derivative optimization algorithm. The optimization algorithm should select parameters such that the output of the model most accurately represents the dynamic changes which occur to the synaptic weights induced by each individual successive stimulus pulse throughout individual stimulation trains.

Results:
The figure below (Fig. 1) shows the experimental data versus the model output generated by the parameter estimation algorithm. The estimated parameters show a good match for all frequencies, verifying the validity of this approach. Overall, the results suggest that this method can be used for the assessment of STP dynamics of in-vivo human neuronal recordings.