Fig 1.
Common inhalation anesthetics have similar effects on synaptic receptors.
Experimental findings show similar effects across inhalation anesthetics on synaptic receptors [29,30,38,92]. Binding to inhibitory GABAA receptors is commonly potentiated while NMDA receptor activity is commonly inhibited with the magnitude of effect varying between anesthetics. Activation of muscarinic acetylcholine receptors and AMPA receptors is inhibited by isoflurane and sevoflurane while desflurane has a biphasic effect and null effect on muscarinic acetylcholine and AMPA receptors, respectively.
Table 1.
Parameter optimization for simulated anesthetic concentrations when performed on 10 different network realizations.
A/B-Series describe optimal values determined by the differential evolution algorithm fitting network connectivity parameters obtained when repeating the optimization for 10 total networks. Optimization includes A-Series, when ACh effects are assumed constant and B-Series, when ACh effects are allowed to change with anesthetic concentration. The scaling factors Px scale the effects of synaptic conductances mediated by the x receptor (x = NMDA, GABA and AMPA). A1-A4/B1-B4 denote optimal parameter sets fit to experimental recordings at varying anesthetic concentrations (0%, 2%, 4%, 6% desflurane, respectively). PAMPA is only fit for the 0% anesthetic case A1/B1. Error displayed is SEM.
Table 2.
Parameter values for best fit of simulated anesthesia and cholinergic reversal.
Parameters from lowest cost fit (cost averaged across anesthetic levels) used to simulate anesthetic effects and cholinergic reversal. A/B-Series describe optimal values of best fit determined by the differential evolution algorithm for network connectivity parameters obtained when ACh effects are assumed constant (i.e., gKs is constant; A-Series) and when ACh effects are allowed to change with anesthetic concentration (B-Series). Px denotes scaled changes in synaptic conductance’s mediated by the x receptor (x = NMDA, GABA and AMPA) as described in Table 1. AR/BR-Series represent simulated anesthetic reversal, obtained by increasing ACh effects (decreasing gKs from A4/B4 levels) while keeping all other parameters constant.
Fig 2.
Changes in anesthesia level lead to transitions from high frequency asynchronous to low frequency synchronous spiking patterns.
A). Raster plots of experimentally recorded neuronal activity in response to changes in desflurane levels. For higher concentrations of desflurane (6%), oscillatory synchronous network activity can be seen in spiking dynamics. For lower levels of anesthetic, oscillations are not apparent and asynchronous activity dominates. B) Raster plots for simulated anesthetic effects in optimized model networks for constant gKs (A-Series) and the simulated ACh-induced reversal of anesthetic effects (A-Series reversal). C) Raster plots for simulated anesthetic effects in optimized networks with changing gKs (B-Series) and its reversal (B Series reversal). In both B) and C), simulated anesthetic reversal shows reinstatement of asynchronous from synchronous spiking patterns. Simulation results based on best fit parameters (lowest cost optimization when averaged across anesthesia levels).
Fig 3.
Firing rate distributions for different levels of anesthetic concentration.
A) Changes in experimentally recorded firing rate distributions under increasing desflurane concentration (0, 2, 4, and 6%) show increased right skewness for the awake state in comparison to anesthetic states. The bins were normalized by the total number of spikes relative to the awake case (0%). B) and C) Firing rate distributions in optimized networks for A- (B) and B- (C) series parameter sets. Simulated networks show similar trends in frequency distributions when compared to experiment. The predicted ACh-induced reversal shows reinstatement of the right skew. The bins were normalized by the total number of spikes relative to the awake case A1/B1. Upper/Lower bound show histogram standard error. Results based on lowest cost fit parameters.
Fig 4.
Characterization of anesthetic effects on network dynamics and their simulated ACh reversal for best fit.
Measures of network dynamics computed from experimental data and optimized model networks as a function of anesthetic concentration and simulated reversal level: A) Average spike rate B). Mean Phase coherence C) Complexity C(X) D) Integration I(X). A1-AR4/B1-BR4 (x-axis) denote simulated anesthetic concentration levels and reversal states obtained in optimized networks with corresponding parameters listed in Table 2. Black line denotes simulations with A-series parameter sets (gKs constant) and pink line denotes simulations with B-series parameter sets (changing gKs). Blue line (with corresponding axis labels on the top) denotes measures computed from experimental spiking data at different desflurane concentrations. Stars denote significance between initial anesthetic/reversal and subsequent simulations. All calculations were made for 6s intervals and then averaged over 5 intervals. Error bars are +/-SEM based on 10 network realizations. Results from lowest cost fit parameters.
Fig 5.
Characterization of anesthetic effects on network connectivity and their simulated ACh reversal.
Measures of network connectivity computed from experimental data and optimized model networks as a function of anesthetic concentration and simulated reversal level: A) network excitatory connectivity strength, B) network inhibitory connectivity strength, C) network excitatory connectivity probability, D) network inhibitory connectivity probability. A1-AR4/B1-BR4 (x-axis) denote simulated anesthetic concentration levels and reversal states obtained in optimized networks with corresponding parameters listed in Table 2. Blue line (with corresponding axis labels on the top) denotes measures computed from experimental data, black (pink) line denotes measures computed from A-series (B-series) network simulations. In these measures, the presence of a significant connection was determined through cross correlogram analysis as described in Methods section. Stars denote significance between initial anesthetic/reversal and subsequent simulations. Error bars of +/- SEM based on 10 network realizations for best fit optimization.
Fig 6.
Effects of anesthetic concentration and simulated ACh-induced reversal on relative profiles of neuronal firing frequency.
Each panel depicts the firing frequency of each neuron in a given anesthetic/reversal state (x-axis) compared to its firing frequency in the non-anesthetic condition (0% desflurane or A1) (y-axis) A) Units recorded in the experimental data; B,C) Neurons in A-series optimized networks and reversal.
Fig 7.
Effects of anesthetic concentration and ACh induced reversal on similarity between cellular functional connectivity.
Cosine similarity Z-Score was computed for every pairwise functional connection between neurons. A) Experimental functional connectivity was computed between the highest firing neuron for each electrode with similarity computed across different levels of anesthesia. B,C) Functional network similarity computed for simulated anesthesia and reversal. Z-Scores were computed comparing the Network similarity to mean and standard deviation of similarities for distributions randomly jittered +/- 5 milliseconds. Each is averaged over ten runs.
Fig 8.
Network structure is populated by lognormal distributed random connection strengths.
A). Synaptic strengths in model networks varied according to a lognormal distribution with a minority of connections being mediated by strong synaptic strengths, while weak synaptic strengths constitute majority of connections B) Simulated network consists of 200 inhibitory and 800 excitatory cells connected randomly with 10% probability. Connection color reflects the log of synaptic strength. C, D) Postsynaptic potential time courses in response to synaptic currents mediated by different receptors. Excitatory currents are modeled with both AMPA and NMDA mediated currents. Bottom panel shows timing of presynaptic spikes, for simplicity both inhibitory and excitatory presynaptic neurons are shown with the same spike times.
Fig 9.
Binned spike patterns for complexity and integration measures.
To compute entropy metrics complexity (C(X)) and integration (I(X)), spike trains were binned in 1 ms bins. H(X) in Eqs (16)/(17) is computed according to unique patterns associated with column vectors (red vectors) while H(Xi) is the entropy associated with a single neuron spike train (blue vector).
Fig 10.
Cross Correlogram computes coincident spike relations by summing relative spike times of reference and comparison neurons.
. A-D). Cross correlograms between example pairs of “reference” and “comparison” cells, centered at spike times of the “reference” cell, from the experimental recordings (left column) and simulated networks (right column). Significance bands were computed from a jittered data set of “comparison” cell spike times (gray line = mean of jittered data set, red line = excitatory significance, blue line = inhibitory significance, see text). A-B) Example cross correlograms showing significant excitatory connections between cell pairs. C, D) Example cross correlograms showing significant inhibitory connections between cell pairs.
Fig 11.
Parameter search fine-tuned through Differential Evolution algorithm.
A). Evolutionary algorithm procedure, differential evolution, was used to optimize model parameters. For each generation, 10 agents (parameter sets) with the highest cost function from the population of 30, were chosen for replacement. Algorithm was repeated until stopping criteria of 100 generations without change in lowest cost function value across the population was met. B,C) Example optimization cost of lowest cost parameters across the population for each generation in the A-series (B) and B-series (C). Population A1/B1, A2/B2, A3/B3 and A4/B4 were optimized to experimental data from the 0%, 2%, 4% and 6% anesthetic cases, respectively. The optimizations for A1 and B1 were identical. In the A-series (A2-A4), PNDMA, PGABA, were optimized and in the B-Series (B2-B4), PNDMA, PGABA, gKs were varied.