THE rapid pace of advancement in understanding anesthetic mechanisms of action—from the molecular level to macroscopic measures of human brain function—have coalesced this month in a paper using molecular level anesthetic effects to drive a mathematical model of cortical neuron responses.1This model nicely replicates human electroencephalograph patterns produced by volatile anesthetics, even when mixed with intravenous agents. We hope that this advance from the Steyn-Ross group will encourage further use of computational models in studies of anesthetic mechanisms and cortical brain states.
A wide range of neural network models have been developed to simulate brain function, and some models have been particularly successful in modeling cortical level processing in the visual and auditory systems. Many of these models consist of large-scale networks with complex structures that require a great deal of computing time using large-scale computers, and often the output is difficult to interpret.2–5In contrast, mean-field models using coarse-grained time equations were created to approximate neuronal function, to minimize the complexity of cortical simulations, and to reduce computational time.6These models allow more tractable simulations of specific features of a network, and the output can be as simple as a simulated electroencephalogram signal. These reduced models, like the one used by the Steyn-Ross group, seem to offer powerful and directed simulations of specific aspects of cortical function, including neural responses to pharmacological manipulation.
The “mean-field” model used by the Steyn-Ross group was based on realistic neuronal parameters—reasonable values for resting membrane potentials, synaptic connections among inhibitory and excitatory cortical neurons, as well as anesthetic concentration-dependent effects on molecular and cellular drug targets. The output from this model is an electroencephalogram-like signal that appears to be similar to those seen in normal humans.7When the model is anesthetized with volatile anesthetics, a concentration-dependent profile of effects are seen, and this study specifically addressed seizure-like effects produced by the ethers isoflurane and especially enflurane. The model was anesthetized by changing neuronal input parameters based on data from quantitative measures of anesthetic effects on γ-aminobutyric acid and N -methyl-d-aspartate synapses. With only these limited changes, the model was able to reproduce accurately ether-induced seizure electroencephalogram responses and even predicted the greater efficacy of enflurane versus isoflurane for burst activity observed during anesthesia. Thus, this model has confirmed earlier hypotheses about enflurane’s propensity for producing seizures that were based on neurophysiological findings.8,9
Of course, the scope of the model is limited, and there is room for improvement in the output behavior. The electroencephalogram signals generated by the model agree in general with brain recordings, but lack subtle features evident in real signals.7,10,11For example, at low anesthetic concentrations, the model correctly produces an increased amplitude of electroencephalogram responses, but fails to replicate some of the frequency slowing that occurs in actual recordings of anesthetic effects. At higher concentrations, the seizure electroencephalogram patterns generated by the model are highly simplified and stereotypic compared with complex patterns of burst-suppression and seizure discharge seen in real electroencephalogram recordings. However, the robustness with which the model reproduces major anesthetic-induced electroencephalogram transitions is encouraging. No doubt, some of the more subtle anesthetic effects on output will emerge when additional anesthetic sites of action are included in the model.
What additional anesthetic sites of action could we add to improve this model? If you ask 10 anesthesia researchers this question, you will likely receive more than 20 sites that need to be considered; however, there are three key neurophysiological processes that, if included in the model, would help to generate more realistic anesthetic-induced electroencephalogram output responses. First, a mechanism to trigger bursts using the well-documented buildup of excitatory synaptic inputs should be addressed.10,11This could include dynamic changes in inhibitory interactions among inhibitory interneurons, which lead to disinhibition of cortical circuits. Anesthetic-enhanced γ-aminobutyric acid inhibition of connected pairs of inhibitory interneurons12may help drive the barrages of excitatory postsynaptic potentials that initiate synchronized burst discharges in cortical neurons. A second important element to emphasize further in the model is the prolongation of both γ-aminobutyric acid receptor type A slow and fast inhibitory postsynaptic potential, because these have been shown to contribute to a slowing of electroencephalogram frequencies.13This would result in the large amplitude slow wave (δ 0.5–3.0 Hz) activity that is produced by many anesthetics—a much stronger effect than the model output currently exhibits. Finally, the present model needs to incorporate shunting of both excitatory and inhibitory synaptic inputs throughout the cortical circuitry. At present, inhibition is modeled as a change in membrane potential (hyperpolarization); however, a good deal of enhanced inhibition (i.e. , via tonic γ-aminobutyric acid-mediated chloride14and two pore-gated potassium conductances15) involves membrane resistance decreases that shunt dendritic synaptic inputs with little change in membrane potential. Shunting conductances seem to play important roles in anesthetic inhibition and for terminating burst discharges in cortical neurons.10,11Many other important sites of anesthetic action have not yet been incorporated into the mean-field model, such as presynaptic effects to increase γ-aminobutyric acid or to depress glutamate release from nerve terminals,16–18and effects on ascending modulatory systems19,20and subcortical inputs21to the cortical model will need to be explored. Thus, we can expect considerable improvement in the near term from including anesthetic effects on targets that have already been well characterized, to say nothing of improvements to the model that will result from the inclusion of yet undiscovered, perhaps even more important, anesthetic effects on cortical neurons.
A major advantage of this mean-field model is that it provides tools needed to incorporate new data into a mathematical framework that can be used to test hypotheses about anesthetic actions on electroencephalogram responses. The true test of its usefulness will come when predictions based on the model guide experimentalists to fruitful avenues of research. Care must be exercised with this, and all models in their early and incomplete stages, because, in the words of pioneering anesthesiologist John Severinghaus, as relayed to us by Ted Eger II: “To every problem there is a solution; neat, plausible and wrong.” To ensure that we have the right solution from the mean-field model, it will be necessary to compare findings using more complex, biologically realistic multisite anesthetic actions.
*Stanford Neuroscience Program and the Department of Anesthesia, Stanford University School of Medicine, Palo Alto, California. maciver@stanford.edu
