The widely used electroencephalogram-based indices for depth-of-anesthesia monitoring assume that the same index value defines the same level of unconsciousness for all anesthetics. In contrast, we show that different anesthetics act at different molecular targets and neural circuits to produce distinct brain states that are readily visible in the electroencephalogram. We present a two-part review to educate anesthesiologists on use of the unprocessed electroencephalogram and its spectrogram to track the brain states of patients receiving anesthesia care. Here in part I, we review the biophysics of the electroencephalogram and the neurophysiology of the electroencephalogram signatures of three intravenous anesthetics: propofol, dexmedetomidine, and ketamine, and four inhaled anesthetics: sevoflurane, isoflurane, desflurane, and nitrous oxide. Later in part II, we discuss patient management using these electroencephalogram signatures. Use of these electroencephalogram signatures suggests a neurophysiologically based paradigm for brain state monitoring of patients receiving anesthesia care.
The authors review the neurophysiology of the electroencephalogram signatures of the anesthetics: propofol, dexmedetomidine, ketamine, sevoflurane, isoflurane, desflurane, and nitrous oxide. These signatures provide a neurophysiologically based paradigm for brain state monitoring of patients receiving anesthesia care.
The Electroencephalogram and Brain Monitoring under General Anesthesia
Almost 80 yr ago, Gibbs et al. demonstrated that systematic changes occur in the electroencephalogram and patient arousal level with increasing doses of ether or pentobarbital. They stated that “a practical application of these observations might be the use of electroencephalogram as a measure of the depth of anesthesia.”1 Several subsequent studies reported on the relation between electroencephalogram activity and the behavioral states of general anesthesia.2–6 Faulconer7 showed in 1949 that a regular progression of the electroencephalogram patterns correlated with the concentration of ether in arterial blood. Bart et al. and Findeiss et al. used the spectrum—the decomposition of the electroencephalogram signal into the power in its frequency components—to show that the electroencephalogram was organized into distinct oscillations at particular frequencies under general anesthesia.8,9 Bickford et al.10 introduced the compressed spectral array or spectrogram to display the electroencephalogram activity of anesthetized patients over time as a three-dimensional plot (power by frequency vs. time).11 Fleming and Smith12 devised the density-modulated or density spectral array, the two-dimensional plot of the spectrogram, for the same purpose.13 Levy14 later suggested using multiple electroencephalogram features to track anesthetic effects. Despite further documentation of systematic relations among anesthetic doses, electroencephalogram patterns, and patient arousal levels,4,15–20 use of the unprocessed electroencephalogram and the spectrogram to monitor the states of the brain under general anesthesia and sedation never became a standard practice in anesthesiology.
Instead, since the 1990s, depth of anesthesia has been tracked using indices computed from the electroencephalogram and displayed on brain monitoring devices.21–25 The indices have been developed by recording simultaneously the electroencephalogram and the behavioral responses to various anesthetic agents in patient cohorts.26 Some of the indices have been derived by using regression methods to relate selected electroencephalogram features to the behavioral responses.26–29 One index has been constructed by using classifier methods to derive a continuum of arousal levels ranging from awake to profound unconsciousness from visually categorized electroencephalogram recordings.30,31 Another index has been constructed by relating the entropy of the electroencephalogram signal—its degree of disorder—to the behavioral responses of the patients.32,33 The indices are computed from the electroencephalogram in near real time and displayed on the depth-of-anesthesia monitor as values scaled from 0 to 100, with low values indicating greater depth of anesthesia. The algorithms used in many of the current depth-of-anesthesia monitors to compute the indices are proprietary.
Although the electroencephalogram-based indices have been in use for approximately 20 yr, there are several reasons why they are not part of standard anesthesiology practice. First, use of electroencephalogram-based indices does not ensure that awareness under general anesthesia can be prevented.34,35 Second, these indices, which have been developed from adult patient cohorts, are less reliable in pediatric populations.36,37 Third, because the indices do not relate directly to the neurophysiology of how a specific anesthetic exerts its effects in the brain, they cannot give an accurate picture of the brain’s responses to the drugs. Finally, the indices assume that the same index value reflects the same level of unconsciousness for all anesthetics. This assumption is based on the observation that several anesthetics, both intravenous and inhaled agents, eventually induce slowing in the electroencephalogram oscillations at higher doses.1,4,22 The slower oscillations are assumed to indicate a more profound state of general anesthesia. Two anesthetics whose electroencephalogram responses frequently lead clinicians to doubt index readings are ketamine38,39 and nitrous oxide.40–42 These agents are commonly associated with faster electroencephalogram oscillations that tend to increase the value of the indices at clinically accepted doses. Higher index values cause concern as to whether the patients are unconscious. At the other extreme, dexmedetomidine can produce profound slow electroencephalogram oscillations43 and low index values consistent with the patient being profoundly unconscious. However, the patient can be easily aroused from what is a state of sedation rather than unconsciousness.43,44
These ambiguities in using electroencephalogram-based indices to define brain states under general anesthesia and sedation arise because different anesthetics act at different molecular targets and neural circuits to create different states of altered arousal45,46 and, as we show, different electroencephalogram signatures.43,47 The signatures are readily visible as oscillations in the unprocessed electroencephalogram and its spectrogram. We relate these oscillations to the actions of the anesthetics at specific molecular targets in specific neural circuits.
Therefore, we propose a new approach to brain monitoring of patients receiving general anesthesia or sedation: train anesthesiologists to recognize and interpret anesthetic-induced brain states defined by drug-specific neurophysiological signatures observable in the unprocessed electroencephalogram and the spectrogram. The new concept of defining the anesthetic state by using drug-specific electroencephalogram signatures that relate to molecular and neural circuit mechanisms of anesthetic action would allow anesthesiologists to make more detailed and accurate assessments than those based on electroencephalogram-based indices. The potential benefits of using the unprocessed electroencephalogram to monitor anesthetic states have been recently stated.48,49 Current brain monitors display the unprocessed electroencephalogram and the spectrogram.22,31,50
To define anesthetic states in terms of drug-specific electroencephalogram signatures that relate to molecular and neural circuit mechanisms of anesthetic action, we synthesize different sources and levels of information: (1) formal behavioral testing along with either simultaneous human electroencephalogram recordings or human intracranial recordings during anesthetic administration; (2) clinical observations of behavior along with simultaneous electroencephalogram recordings during anesthetic administration; (3) the neurophysiology and the molecular pharmacology of how the anesthetics act at specific molecular targets in specific circuits; (4) the neurophysiology of altered states of arousal such as non-rapid eye movement sleep, coma (medically induced, pathologic, or hypothermia induced), hallucinations, and paradoxical excitation; (5) time–frequency analyses of high-density electroencephalogram recordings; and (6) mathematical modeling of anesthetic actions in neural circuits.
We present this new education paradigm in two parts. Here, in part I, we review the basic neurophysiology of the electroencephalogram and the neurophysiology and the electroencephalogram signatures of three intravenous anesthetics: propofol, dexmedetomidine, and ketamine, and four inhaled anesthetics: sevoflurane, isoflurane, desflurane, and nitrous oxide. We explain, where possible, how the anesthetics act at specific receptors in specific neural circuits to produce the observed electroencephalogram signatures. In part II, we discuss how knowledge of the different electroencephalogram signatures may be used in patient management.
The Electroencephalogram: A Window into the Brain’s Oscillatory States
Coordinated action potentials, or spikes, transmitted and received by neurons, are one of the fundamental mechanisms through which information is exchanged in the brain and central nervous system (fig. 1A).51,52 Neuronal spiking activity generates extracellular electrical potentials,53 composed primarily of postsynaptic potentials and neuronal membrane hyperpolarization (fig. 1A).53,54 These extracellular potentials are often referred to as local field potentials. Populations of neurons often show oscillatory spiking and oscillatory local field potentials that are thought to play a primary role in coordinating and modulating communication within and among neural circuits.52 Local field potentials produced in the cortex can be measured at the scalp as the electroencephalogram (fig. 1B).
The organization of the pyramidal neurons in the cortex favors the production of large local field potentials because the dendrites of the pyramidal neurons run parallel with each other and perpendicular to the cortical surface (fig. 1C). This geometry creates a biophysical transmitting antenna that generates large extracellular currents whose potentials can be measured through the skull and scalp as the electroencephalogram.53,55,56 Subcortical regions, such as the thalamus (fig. 1D), produce much smaller potentials that are more difficult to detect at the scalp because the electric field decreases in strength as the square of the distance from its source.56 However, because cortical and subcortical structures are richly interconnected, scalp electroencephalogram patterns reflect the states of both cortical and subcortical structures.57 Thus, the electroencephalogram provides a window into the brain’s oscillatory states.
To show the unprocessed electroencephalogram recordings and their spectrograms computed with the same methods on comparable displays, we have taken examples from cases of patients receiving general anesthesia and sedation at our institution. These data were recorded following a protocol approved by the Massachusetts General Hospital Human Research Committee to construct a database of electroencephalogram of recordings of patients receiving general anesthesia and sedation. Electroencephalograms were recorded using the Sedline monitor (Masimo Corporation, USA). The Sedline electrode array records approximately at positions Fp1, Fp2, F7, and F8, with the reference approximately 1 cm above Fpz and the ground at Fpz. We required impedances less than 5kΩ in each channel. Our analyses use Fp1 as results were identical for Fp1 and Fp2. We also include analyses of human intracranial recordings60 and high-density electroencephalogram recordings20 from previous studies.
The spectrograms were computed using the multitaper method64,65 from the unprocessed electroencephalogram signals recorded at a sampling frequency of 250 Hz. Individual spectra were computed in 3-s windows with 0.5-s overlap between adjacent windows. Multitaper spectral estimates have near optimal statistical properties64,65 that substantially improve the clarity of spectral features. At present, no current electroencephalogram monitor displays multitaper spectra or spectrograms. We present multitaper spectra in this review to show the electroencephalogram signatures of different anesthetic drugs with the greatest clarity.
Time Domain and Spectral Measures of the Brain’s Sedative and Anesthetic States
Many of the changes that occur in the brain with changes in anesthetic states can be readily observed in unprocessed electroencephalogram recordings (fig. 2). Different behavioral and neurophysiological states induced by anesthetics are associated with different electroencephalogram waveforms. For example, figure 2 shows the electroencephalogram of the same patient in different states of propofol-induced sedation and unconsciousness.20,65 These include the awake state (fig. 2A), paradoxical excitation (fig. 2B), a sedative state (fig. 2C), the slow and alpha oscillation anesthetic state (fig. 2D), the slow oscillation anesthetic state (fig. 2E), burst suppression (fig. 2F), and the isoelectric state (fig. 2G). Visualization and analysis of the unprocessed electroencephalogram is a form of time domain analysis.64,65 This approach is commonly used in sleep medicine and sleep research to define sleep states66 and also in epileptology to characterize seizure states.67
Reading the frequencies and amplitudes from the unprocessed electroencephalogram in real time in the operating room is challenging. If the frequencies of the oscillatory components were known, then it would be possible to design specific filters to extract these components (fig. 3, A and B). The more practical and informative solution is to conduct a spectral analysis by computing the spectrum (fig. 3C) and the spectrogram (figs. 3D and 3E).64,65,69
For a given segment of electroencephalogram data, the spectrum (fig. 3C) provides a decomposition of the segment into its frequency components usually computed by Fourier methods.64,65 The advantage of the spectrum is that it shows the frequency decomposition of the electroencephalogram segment for all of the frequencies in a given range (fig. 3C) by plotting frequency on the x-axis and power on the y-axis. Power is commonly represented in decibels, defined as 10 times the log base 10 of the squared amplitude of a given electroencephalogram frequency component. Electroencephalogram power can differ by orders of magnitude across frequencies. Taking logarithms makes it easier to visualize on the same scale frequencies whose powers differ by orders of magnitude. The spectrum of a given data segment is thus a plot of power (10 log10 (amplitude)2) by frequency.
The frequency bands in the spectrum are named following a generally accepted convention (table 1). Changes in power in these bands can be used to track the changes in the brain’s anesthetic states. In the signal shown in figure 3A, the low-frequency oscillation has a period of approximately 1 cycle per second or 1 Hz (table 1, slow oscillation), whereas the period of the faster oscillation is at approximately 10 Hz (table 1, alpha oscillation). The spectrum (fig. 3C) also shows that this signal has power in the delta range (1 to 4 Hz) and little to no power beyond 12 Hz.
In addition to these conventional frequency bands, two other spectral features are commonly reported in electroencephalogram analyses in anesthesiology: the median frequency (fig. 3C, lower white curve) and the spectral edge frequency (fig. 3C, upper white curve). The median frequency is the frequency that divides the power in the spectrum in half70,71 (fig. 3C), whereas the spectral edge frequency is the frequency below which 95% of the spectral power is located.71 In other words, in the frequency range we use in our analyses of 0.1 to 30 Hz, half of the power in the spectrum lies below the median and 95% of the power lies below the spectral edge frequency. In figure 3C, the median frequency and spectral edge frequency are 3.4 and 15.9 Hz, respectively, where the frequency range is 0.1 to 30 Hz. The median frequency and the spectral edge are displayed on commercial monitors and are useful clinically for tracking whether spectrogram power is shifting to lower (lower median frequency and spectral edge) or higher (higher median frequency and spectral edge) frequencies. As we discuss, the interpretations of these shifts are anesthetic dependent.
The spectrum displays the power content by frequency for only a single segment of electroencephalogram data. Use of the spectrum in electroencephalogram analyses during anesthesia care requires computing it on successive data segments. Successive computation of the spectrum across contiguous, often overlapping, segments of data is termed the spectrogram64,65 (fig. 3D and 3E). The spectrogram makes it possible to display how the oscillations in the electroencephalogram change in time, with changes in the dosing of the anesthetics and/or the intensity of arousal-provoking stimuli. The spectrogram is a three-dimensional structure (fig. 3D). However, it is plotted in two dimensions by placing time on the x-axis, frequency on the y-axis, and power through color coding on the z-axis (fig. 3E). As discussed earlier, this two-dimensional plot of the spectrogram is termed the density spectral array (fig. 3E),12,13 whereas the three-dimensional plot of the spectrogram is termed the compressed spectral array (fig. 3D).10,22 We display the spectrogram as a density spectral array and refer to it as the spectrogram.
Spectral analyses make it easier to visualize frequency content, especially oscillations, and to detect subtle changes in frequency structure. Nevertheless, it is important to know both the time domain and spectral representations of a given behavioral or neurophysiological state induced by an anesthetic. We present both in our discussions in the following sections for commonly used intravenous and inhaled anesthetics.
Neurophysiology and Clinical Electrophysiology of Selected Intravenous Anesthetics
We review the neuropharmacology and clinical electrophysiology of propofol, dexmedetomidine, and ketamine. For each anesthetic, we discuss the putative mechanism through which its actions at specific molecular targets in specific neural circuits produce the electroencephalogram signatures and the behavioral changes associated with its anesthetic state.
Molecular and Neural Circuit Mechanisms of Propofol
Propofol, the most widely administered anesthetic agent, is used as an induction agent for sedation and maintenance of general anesthesia. Low-dose propofol is frequently administered either as small boluses or by infusion for surgeries and diagnostic procedures that require only sedation.
The molecular mechanism of propofol has been well characterized. Propofol binds postsynaptically to γ-aminobutyric acid type A (GABAA) receptors where it induces an inward chloride current that hyperpolarizes the postsynaptic neurons thus leading to inhibition.72,73 Because the drug is lipid soluble and GABAergic inhibitory interneurons are widely distributed throughout the cortex, thalamus, brainstem, and spinal cord, propofol induces changes in arousal through its actions at multiple sites (fig. 4). In the cortex, propofol induces inhibition by enhancing GABA-mediated inhibition of pyramidal neurons.73 Propofol decreases excitatory inputs from the thalamus to the cortex by enhancing GABAergic inhibition at the thalamic reticular nucleus, a network that provides important inhibitory control of thalamic output to the cortex. Because the thalamus and cortex are highly interconnected, the inhibitory effects of propofol lead not to inactivation of these circuits but rather to oscillations in the beta (figs. 2C and 5) and alpha (figs. 2D, 6 and 7) ranges. Propofol also enhances inhibition in the brainstem at the GABAergic projections from the preoptic area of the hypothalamus to the cholinergic, monoaminergic, and orexinergic arousal centers (fig. 4). Decreasing excitatory inputs from the thalamus and the brainstem to the cortex enhances hyperpolarization of cortical pyramidal neurons, an effect that favors the appearance of slow and delta oscillations on the electroencephalogram (figs. 5–7).20,60,75
The Electroencephalogram Signatures of Propofol Sedation and Paradoxical Excitation Are Beta-gamma Oscillations
The electroencephalogram patterns seen during sedation are organized, regular beta-gamma oscillations (figs. 2C and 5) and slow-delta oscillations (fig. 5). The amplitudes of these oscillations are larger than those of the gamma oscillations seen in the awake electroencephalogram (fig. 2A). When general anesthesia has been maintained by a propofol infusion, a similar beta oscillation pattern is visible in patients after extubation as they lie quietly before transfer to the postanesthesia care unit. A beta oscillation is also seen during paradoxical excitation (fig. 2B), the state of euphoria or dysphoria with movements, that can occur when patients are sedated. The state is termed paradoxical because a dose of propofol intended to sedate results in excitation. Two mechanisms have been proposed to explain propofol-induced paradoxical excitation. One involves GABAA1-mediated inhibition of inhibitory inputs from the globus pallidus to the thalamus leading to increased excitatory inputs from the thalamus to the cortex.46 This mechanism is also the one through which the sedative zolpidem is postulated to induce arousal in minimally conscious patients.76 The second mechanism, established in simulation studies, postulates that low-dose propofol induces transient blockade of slow potassium currents in cortical neurons.76
The Electroencephalogram Signatures of Propofol Are Slow-delta Oscillations on Induction
The electroencephalogram patterns observed during propofol general anesthesia depend critically on several factors, the most important of which is the rate of drug administration. When propofol is administered as a bolus for induction of general anesthesia, the electroencephalogram changes within 10 to 30 s from an awake pattern with high-frequency, low-amplitude gamma and beta oscillations (fig. 2A) to patterns of high-amplitude slow and delta oscillations (fig. 2E). The slow-delta and delta oscillations appear in the spectrogram as increased power between 0.1 to 5 Hz (fig. 6, A and B, between minutes 0 and 5) and in the time domain as high-amplitude oscillations (fig. 6C, minute 5.5, and fig. 6D, minute 7.1). The electroencephalogram change is dramatic as the slow-delta oscillation amplitudes can be 5 to 20 times larger than the amplitudes of the gamma and beta oscillations seen in the electroencephalogram of an awake patient.43
The appearance of slow and delta oscillations (figs. 2 and 6, C and D) within seconds of administering propofol for induction of general anesthesia coincides with loss of responsiveness, loss of the oculocephalic reflex, apnea, and atonia.46,74 These electroencephalogram signatures and the clinical signs are consistent with a rapid action of the anesthetic in the brainstem. After bolus administration, propofol quickly reaches the GABAergic inhibitory synapses emanating from the preoptic area of the hypothalamus onto the major arousal centers in the brainstem and hypothalamus (fig. 4). Action of the anesthetic at these synapses inhibits the excitatory arousal inputs from the brainstem, favoring hyperpolarization of the cortex, the appearance of slow-delta oscillations on the electroencephalogram with loss of consciousness (LOC).20,60,75 Loss of the oculocephalic reflex is consistent with the anesthetic acting at cranial nerve nuclei III, IV, and VI in the midbrain and pons.46,74 Apnea is most likely due to the drug’s inhibition of the ventral and dorsal respiratory centers in the medulla and pons,76 whereas the brainstem component of atonia is most likely due to inhibition of the pontine and medullary reticular nuclei.46
Administration of an additional propofol bolus, either before or after intubation, can result in either enhancement of the slow oscillation or the conversion of the slow oscillation into burst suppression (fig. 6B, minutes 7 to 14, and fig. 6D, minute 11.5). Burst suppression is a state of unconsciousness and profound brain inactivation in which the electroencephalogram shows periods of electrical activity alternating with periods of isoelectricity or electrical silence. In the spectrogram, burst suppression appears as vertical lines in the spectrogram (fig. 6B, minutes 7 to 14). When propofol is administered as an induction bolus, patients, particularly elderly patients, can enter burst suppression within seconds.68
As the effects of the bolus doses of propofol recede, the electroencephalogram evolves into slow-delta oscillation and alpha oscillation patterns (fig. 6A, minute 7 and fig. 6B, minute 15). The time of transition from the slow oscillations or burst suppression into a combined slow oscillation and alpha oscillation patterns depends on how profound the effect of the bolus dose was. Even if no additional propofol is administered after the first bolus, it can take several minutes for the transition to occur. The patient shown in figure 6A received a second bolus dose of 50 mg at minute 5. An infusion of propofol was started at a rate of 100 μg kg−1 min−1. Her slow-delta oscillations did not evolve into the slow-delta and alpha oscillations until minute 7. In contrast, the patient shown in figure 6B stayed in a state of burst suppression from minutes 7 to 14 after her two bolus doses of propofol. After the second bolus dose, she received no additional propofol or other anesthetics and transitioned into slow-delta and alpha oscillations (fig. 6B, minutes 15 to 20, and fig. 6D, minute 17). If after the bolus doses, propofol is administered as an infusion to maintain general anesthesia, the patient’s electroencephalogram will continue to show the slow-delta and alpha oscillation patterns in the surgical plane (fig. 6A, minutes 9 to 26, and fig. 6C, minute 24).
Slow-delta and Alpha Oscillations Are Markers of Propofol-induced Unconsciousness
Most brain function monitors used in anesthesiology record only a few channels of the electroencephalogram from the front of the head. An awake patient with eyes closed and a full set of scalp electroencephalogram electrodes will show the well-known eyes-closed alpha oscillations in the occipital areas (fig. 7A, awake baseline).79 Concomitant with the transition to LOC and the appearance of the slow and alpha oscillations is the phenomenon of anteriorization, in which the power in the alpha and beta bands of the electroencephalogram shifts from the occipital area to the front of the head (fig. 7A, LOC and unconsciousness).15,19,20
Although LOC due to propofol has a strong brainstem effect, maintenance of unconsciousness involves the brainstem and other brain centers. Studies of propofol-induced unconsciousness using a high-density (64-lead) electroencephalogram have shown that the alpha oscillations are highly coherent across the front of the head during unconsciousness (fig. 7A, unconsciousness).19,20 An explanation for this coherence is that propofol could be inducing an alpha oscillation within circuits linking the thalamus and the frontal cortex (fig. 7B).57 In contrast, the slow oscillations are not coherent.19,20 Studies of patients with intracranial electrodes also show that the slow oscillations induced by bolus administration of propofol are not coherent across the cortex (fig. 7C) and serve as a marker of phase-limited spiking activity in the cortex (fig. 7C).60 We postulate that the highly organized coherent alpha oscillations most likely prevent normal communications between the thalamus and cortex, whereas the incoherent slow oscillations represent an impediment to normal intracortical communications.20,57,60,80 Together, these two mechanisms likely contribute to the patient being unconscious when the slow and alpha oscillations are visible in the spectrogram of patients receiving propofol. These highly coherent alpha oscillations, incoherent slow oscillations, and anteriorization most certainly contribute to the loss of frontal-parietal effective connectivity80–82 that is also associated with propofol-induced LOC.
Electroencephalogram Signatures of Emergence from Propofol General Anesthesia
When the propofol infusion is discontinued, and the patient is allowed to emerge, the slow and alpha oscillations dissipate and are gradually replaced by higher-frequency beta and gamma oscillations that have lower amplitudes (fig. 6A, minutes 25 to 30, and fig. 6B, minutes 23 to 27).20 In the spectrogram, this gradual shift to higher-frequency beta and gamma oscillations appears as a “zipper opening” pattern (fig. 6A, minutes 25 to 30, and fig. 6B, minutes 23 to 27). The decrease in the amplitude is shown by the shift in the spectrum from red to yellow. In the unprocessed electroencephalogram, this change is marked by a gradual increase in frequency and decrease in amplitude of the oscillations. Simultaneously, there is dissipation of power in the slow and delta bands, which appears in the time domain as a flattening of the unprocessed electroencephalogram. There is also reversal of anteriorization with emergence (fig. 7A, return of consciousness and awake emergence) with loss of the coherent frontal alpha oscillations and return of the coherent occipital alpha oscillations.20 The return of high-frequency power in the electroencephalogram is consistent with return of normal cortical activity and indirectly with return of normal thalamic and brainstem activity. Brainstem function returns in an approximate caudal (medulla and lower pons) to rostral (upper pons and midbrain) manner.46,74 Breathing and gagging are controlled in the medulla and lower pons, whereas the corneal and oculocephalic reflexes are controlled in the upper pons and the midbrain.46,74 Return of brainstem, thalamic, and cortical activity is necessary to restore the awake state.
Burst Suppression and Medically Induced Coma
When delivered in a sufficiently high dose, several anesthetics, including propofol, the barbiturates, and the inhaled ether drugs, induce burst suppression (figs. 2F and 6, B and D, minute 11.5, fig. 8, A and B).83–86 Burst suppression is induced by hypothermia for surgeries requiring total circulatory arrest87 and by administering anesthetics in the intensive care unit for cerebral protection to treat intracranial hypertension or to treat status epilepticus.86,88–90 This latter state is termed a medically induced coma. If the patient is in burst suppression, then, as the dose of the anesthetic is increased, the length of the suppression periods between the bursts increases. The dose can be increased to the point at which the electroencephalogram is isoelectric (fig. 2G). Within the bursts, the electroencephalogram can, in some cases, maintain the brain dynamics that were present before the start of burst suppression.85,91,92 For example, if the patient is in a state of slow and alpha oscillations before the burst suppression, then the same pattern is present within the bursts (figs. 6B and 8A). In addition to deep general anesthesia, burst suppression is also observed in other conditions of profound brain inactivation including coma93 and in children with significantly compromised brain development.94 The observation that multiple different mechanisms induce burst suppression are consistent with a recently proposed neuronal metabolic mechanism of burst suppression.85 Transient increases and decreases in extracellular calcium, leading to synaptic disfacilitation, could also play a role in determining suppression duration.84
The burst suppression ratio or suppression ratio is a time domain measure used to track quantitatively the level of burst suppression. The burst suppression ratio is a number between 0 and 1, which measures the fraction of time in a given time interval that the electroencephalogram is suppressed.95,96 The burst suppression ratio is displayed on some brain function monitors and is one of the measures used in electroencephalogram-based indices to assess depth of anesthesia.97,98 A refinement of the burst suppression ratio, termed the burst suppression probability, has been recently developed (fig. 8C).100 The burst suppression probability is a measure of the instantaneous probability of the brain being in a state of suppression that can be reliably computed using state-space methods and used to track burst suppression in real time and to implement control systems for medical coma.99,100 A burst suppression probability of 0.5 means a 0.5 probability of being suppressed, whereas a burst suppression probability of 0.75 means a 0.75 probability of being suppressed.
Neural Circuit Mechanisms of Ketamine
Ketamine, an anesthetic adjunct and an analgesic, acts primarily by binding to N-methyl-d-aspartate (NMDA) receptors in the brain and spinal cord.45 Ketamine is a channel blocker so that to be effective, the channels have to be open.101 Because in general, the channels on inhibitory interneurons are more active than those on pyramidal neurons, ketamine at low-to-moderate doses has its primary effect on inhibitory interneurons (fig. 9A).102,103 By blocking inputs to inhibitory interneurons, ketamine allows downstream excitatory neurons to become disinhibited or more active.45,46 This is why cerebral metabolism increases with low doses of ketamine. Hallucinations, dissociative states, euphoria, and dysphoria are common with low-dose ketamine because brain regions, such as the cortex, hippocampus, and the amygdala, continue to communicate but with less modulation and control by the inhibitory interneurons. Information is processed without proper coordination in space and time.45,46 The hallucinatory effects are likely enhanced by disruption of dopaminergic neurotransmission in the prefrontal cortex due in part to increased glutamate activity at non-NMDA glutamate receptors.104 Analgesia is due in part to the action of ketamine on glutamate NMDA receptors at the dorsal root ganglia, the first synapse of the pain pathway in the spinal cord where glutamate is the primary neurotransmitter (fig. 9A).101 As the dose of ketamine is increased, the NMDA receptors on the excitatory glutamatergic neurons are also blocked and consciousness is lost.
The Electroencephalogram Signatures of Ketamine Sedation Are Beta and Gamma Oscillations
Given the preference of ketamine for NMDA receptors on inhibitory interneurons, whose inhibition results in increased cerebral metabolic rate, cerebral blood flow, and hallucinations,105–107 it is no surprise that ketamine is associated with an active electroencephalogram pattern. When ketamine is administered alone in low dose, the electroencephalogram shows fast oscillations (fig. 9, B and C) in the high beta, low gamma range between 25 and 32 Hz. The beta-gamma oscillation did not start until 2 min after the initial ketamine dose. Compared with propofol and dexmedetomidine (discussed in the section The Electroencephalogram Signatures of Dexmedetomidine Are Slow-delta Oscillations and Spindles), the ketamine slow oscillation is less regular (fig. 9C). The 61-yr old whose spectrogram is shown in figure 9B required sedation and analgesia for change of a vacuum dressing over a panniculectomy site. At minute 0, she received a 50-mg bolus of ketamine in two doses of 30 and 20 mg, 2 min apart. The dressing change started at minute 10. Five minutes after administering the 20-mg ketamine dose, the patient continued to breathe spontaneously and was unresponsive to verbal commands and the nociceptive stimulation from the procedure. Although the beta and gamma oscillations lasted for only 27 min, the sedative effect of being unresponsive to verbal commands persisted for several minutes after the beta-gamma oscillations had disappeared.
Dexmedetomidine is used as a sedative in the intensive care unit and as a sedative and anesthetic adjunct in the operating room. Compared with propofol, patients are easily arousable when sedated with dexmedetomidine, with minimal to no respiratory depression. Unlike propofol and the benzodiazepines, dexmedetomidine cannot also be used as a hypnotic agent.
Neural Circuit Mechanisms of Dexmedetomidine
Dexmedetomidine alters arousal primarily through its actions on presynaptic α2 adrenergic receptors on neurons projecting from the locus ceruleus. Binding of dexmedetomidine to the α2 receptors hyperpolarizes locus coeruleus neurons decreasing norepinephrine release.108–110 The behavioral effects of dexmedetomidine are consistent with this proposed mechanism of action.111 Hyperpolarization of locus ceruleus neurons results in loss of inhibitory inputs to the preoptic area of the hypothalamus (fig. 10A). The preoptic area sends GABAergic and galanergic inhibitory projections to the major arousals centers in the midbrain, pons, and hypothalamus (fig. 10A).45,46,112 Hence, loss of the inhibitory inputs from the locus ceruleus results in sedation due to activation of these inhibitory pathways from the preoptic area to the arousal centers. Activation of inhibitory inputs from the preoptic area is postulated to be an essential component of how nonrapid eye movement sleep is initiated.113,114 Sedation by dexmedetomidine is further enhanced due to blockage of presynaptic release of norepinephrine, leading to loss of excitatory inputs from the locus ceruleus to the basal forebrain, intralaminar nucleus of the thalamus, and cortex115 and decreased thalamocortical connectivity.116
The Electroencephalogram Signatures of Dexmedetomidine Are Slow-delta Oscillations and Spindles
The relation between the actions of dexmedetomidine in the preoptic area and the initiation of non-rapid eye movement sleep is important to appreciate because it helps explain the similarities in the electroencephalogram patterns between this anesthetic and those observed in nonrapid eye movement sleep. Dexmedetomidine administered as a low-dose infusion induces a level of sedation in which the patient responds to minimal auditory or tactile stimulation. The electroencephalogram shows a combination of slow-delta oscillations with spindles, which are 9 to 15 Hz oscillations that occur in bursts lasting 1 to 2 s (figs. 10B and 11, A and B).43,44 In the frequency domain, the dexmedetomidine spindles appear as streaks in the high alpha and low beta bands between 9 to 15 Hz (fig. 11A). The spindles occur in a similar frequency range as the alpha oscillations seen with propofol but have much less power than the alpha oscillations (figs. 6, A and B).43 Because the color scales on the spectrograms in figures 6, A and B, and 11A are the same, the plots provide an informative comparison of the amplitudes of the propofol alpha oscillations and the dexmedetomine spindles. The dexmedetomidine spindles closely resemble the spindles that define stage II nonrapid eye movement sleep.43,44
The slow-delta oscillations are apparent in the spectrogram as power from 0 to 4 Hz (fig. 11A). The 44-yr-old, 59-kg female patient, whose spectrogram and time domain traces are shown in figures 11, A and B, received a 1 μg/kg loading infusion of dexmedetomidine over 10 min, followed by a maintenance dexmedetomidine infusion of 0.65 μg kg−1 h−1—a rate in the intermediate to high end of the sedative range—for the creation of a left forearm arteriovenous fistula for dialysis. During the surgery, the patient was sedated, meaning that she responded to verbal queries from the anesthesiologist and moved in response to nociceptive stimulation from the surgery.
When the rate of the dexmedetomidine infusion is increased, spindles disappear and the amplitude of the slow-delta oscillations increases (fig. 11, C and D). This electroencephalogram pattern of slow-delta oscillations closely resembles nonrapid eye movement sleep stage III or slow-wave sleep.66 The slow-delta oscillations appear again as intense power in the slow oscillation band (fig. 11C). The power in the slow oscillation band for the higher dose of dexmedetomidine (fig. 11C) is considerably stronger than the slow oscillations for the lower dose of dexmedetomine (fig. 11A). The spectrogram is that of a 48-yr-old, 65-kg female patient who received dexmedetomidine as a loading infusion of 1 μg/kg over 10 min and a maintenance infusion of 0.85 μg kg−1 h−1 for placement of a left forearm arteriovenous fistula. The intense slow-delta oscillation persisted for the duration of the procedure. During the surgery, the patient was sedated, meaning that she was unresponsive to verbal queries from the anesthesiologist and but moved in response to changes in the level of nociceptive stimulation from the procedure.
Propofol frontal alpha oscillations (fig. 7A), dexmedetomidine spindles45 (fig. 11, A and B), and sleep spindles117 are all thought to be generated by thalamocortical loop mechanisms (figs. 7B and 10B). Although the propofol frontal alpha oscillations are highly coherent and continuous in time,19,20,57 the sleep and dexmedetomidine spindles are brief and episodic.43 Sleep spindles are a marker for nonrapid eye movement stage II sleep, which is a state of unconsciousness that is less profound than nonrapid eye movement stage III or slow-wave sleep.66 Dexmedetomidine spindles are consistent with a light state of sedation.43
Propofol-induced slow oscillations likely result from decreased excitatory inputs to the cortex due to propofol’s GABAA-mediated inhibition arousal centers in the midbrain, pons, and hypothalamus (fig. 4). Dexmedetomine-induced slow oscillations likely result from decreased excitatory inputs to the cortex due to dexmedetomidine’s disinhibition of the inhibitory circuits emanating from the preoptic area of the hypothalamus to the arousal centers along with decreased adrenergically-mediated excitatory inputs to the basal forebrain, the intralaminar nucleus of the thalamus, and to the cortex directly (fig. 10A). Propofol-induced slow oscillations gate brief periods of neuronal activity (fig. 7B).60 In contrast, sleep slow-wave oscillations are associated with brief interruptions in neuronal activity.118 Similarly, both slow oscillations and spindles under dexmedetomidine are smaller than their counterparts under propofol, which may reflect a lower level of disruption in neuronal activity under dexmedetomidine compared with propofol.43 Consequently, cortical and thalamocortical activities are likely to be more profoundly inhibited under propofol-induced unconsciousness compared with either slow-wave sleep or dexmedetomidine-induced sedation. This difference in cortical and thalamocortical activity suggests why patients can be aroused from sleep and dexmedetomidine-induced sedation but not from propofol-induced unconsciousness.43,60
Clinical Electrophysiology of the Inhaled Anesthetics
Neurophysiological Mechanisms of Inhaled of Anesthetic Action
The principal inhaled anesthetics are the ether derivatives sevoflurane, isoflurane, and desflurane. The inhaled agents are total anesthetics in that they can maintain all of the required behavioral and physiological characteristics of general anesthesia without adjunct. In practice, the inhaled anesthetics are rarely used alone but more commonly in conjunction with intravenous drugs, nitrous oxide, and muscle relaxants to provide balanced general anesthesia, which maximizes the desired effects while minimizing side effects. The inhaled agents are known to create their behavioral and physiological effects through binding at multiple targets in the brain and central nervous system including binding to GABAA receptors and enhancing GABAergic inhibition; blockade of two-pore potassium channels and hyperpolarizing-activated cyclic nucleotide-gated channels; and blocking glutamate release by binding to NMDA receptors.72 Studies of minimal alveolar concentration (MAC) have shown that the muscle relaxant effects of the inhaled anesthetics are the result of direct action at one or more of these receptors and channels in the spinal cord.119 This mechanism is also consistent with the observation that spinal cord motor-evoked potentials are difficult to record in patients receiving anesthetic concentrations of an ether anesthetic.120
Gibbs et al.1 showed that the electroencephalogram of patients receiving ether show slow-delta oscillations and alpha oscillations at surgical levels of general anesthesia. When sevoflurane is administered at sub-MAC concentrations to achieve surgical levels of general anesthesia, the electroencephalogram shows slow-delta oscillations and coherent alpha oscillations similar to those of propofol.47 This observation suggests that for sevoflurane, as for propofol, enhanced GABAergic inhibition may be its dominant mechanism of action.47 However, unlike propofol, sevoflurane also shows a small coherent theta oscillation,47 which may reflect one of its non-GABAergic mechanisms.
The Electroencephalogram Signatures of Sevoflurane, Isoflurane, and Desflurane Are Alpha, Slow-delta, and Theta Oscillations
At sub-MAC concentrations, sevoflurane shows strong alpha and slow-delta oscillations (fig. 12, A–D) that closely resemble those of propofol (fig. 6). As the concentration of sevoflurane is increased to MAC levels and above, a strong theta oscillation appears creating a distinctive pattern of evenly distributed power from the slow oscillation range up through the alpha range (fig. 12, A and C). This approximately equal power from the slow oscillation range through to the alpha range (fig. 12, A and C) is typical during maintenance with sevoflurane at or above MAC. The theta oscillation power appears to fill in between the slow-delta and alpha oscillation power. As the concentration of sevoflurane is decreased, the theta oscillations dissipate first. This is evident in figure 12A as the theta oscillations dissipate when the concentration of sevoflurane is lowered. On emergence, as in the case of propofol (fig. 6A, minute 27, and fig. 6B, minute 25), the alpha oscillations transition to lower amplitude beta and gamma oscillations (fig. 12A, minute 180). At the same time, the slow and delta oscillations dissipate. The loss of the alpha and slow-delta oscillation power again appears in the spectrogram as a zipper opening pattern.
Isoflurane (fig. 12, E and F) and desflurane (fig. 12, G and H) have similar patterns to sevoflurane (fig. 12, C and D). At sub-MAC concentrations, they also show strong alpha and slow-delta oscillations. When the concentration of isoflurane or desflurane is increased to MAC levels and above, a theta oscillation fills in between the delta and alpha bands (fig. 12E, minute 30, and fig. 12G, minute 28). As in the case of sevoflurane, on emergence from isoflurane and oxygen or desflurane and oxygen anesthesia, there is loss of the theta oscillation power, followed by dissipation of alpha and slow-delta oscillation power, and reappearance of the power in the beta and gamma bands (fig. 12E, minutes 78 to 80, and fig. 12G, minutes 85 to 90). These observations suggest that by analogy with sevoflurane, enhanced GABAergic inhibition is likely a primary but not the only mechanism through which isoflurane and desflurane induce their anesthetic states. This relation between MAC and theta oscillations is useful clinically. The appearance of theta oscillations indicates a more profound state of unconsciousness and immobility for an inhaled ether anesthetic.
High-amplitude Slow-delta Oscillations Mark the Transition to High-flow Nitrous Oxide
Nitrous oxide has been used as a sedative and as an anesthetic since the late 1800s.3,121 Today, it is most often used as an anesthetic adjunct because, unlike the ether anesthetics, nitrous oxide is not sufficiently potent by itself to produce general anesthesia. Attempts to administer sufficient doses of only nitrous oxide to produce unconsciousness predictably result in nausea and vomiting.41 Unlike the ether anesthetics, administering nitrous oxide with oxygen is not commonly thought to produce slow and alpha oscillations. Instead, nitrous oxide is associated with prominent beta and gamma40 oscillations and, possibly, with a relative decrease in power in the slow and delta oscillation bands.41
A common practice at our institution is to use an inhaled ether anesthetic with oxygen for maintenance of general anesthesia and to switch to a high concentration of nitrous oxide with oxygen with high total flow rates near the end of the case to hasten the patient’s emergence. In this situation, a distinctive electroencephalogram pattern appears with the transition to nitrous oxide (illustrated in fig. 13). In anticipation of emergence, a 34-yr-old patient was maintained on 0.5% isoflurane and 58% oxygen during the last minutes of a laparoscopic cholecystectomy. At minute 82, the anesthetic was switched to 0.2% isoflurane, 75% nitrous oxide, and 24% oxygen. With the switch, the total flow rate was increased from 3 to 7 l/min. Two minutes after the switch, the power in the slow delta and alpha bands begins to decline (fig. 13A, minute 84). Four minutes after the switch, a profound slow-delta oscillation appears in the electroencephalogram (fig. 13A, minute 86, and fig. 13B, minute 86). These slow-delta oscillations dominate with near loss of all power above 5 Hz for approximately 4 min (fig. 13, minutes 86 to 90, blue area in the spectrogram) before they evolve to the beta-gamma oscillations that are more commonly associated with nitrous oxide (fig. 13A, minute 90, and fig. 13B, minute 90). The appearance of the beta-gamma oscillations and the loss of the slow-delta oscillations take place from minutes 90 to 94. The patient was extubated at minute 110 (not shown).
The appearance of slow-delta oscillations we commonly observe in the switch from one of the ether anesthetics, halothane or propofol, to a high concentration (> 70%) nitrous oxide has been previously reported.122–124 The slow-delta oscillations tend to be transient,122–124 and the beta-gamma oscillations usually reported with nitrous oxide40,41 appear as the slow-delta oscillations dissipate. These slow-delta oscillations are noticeably different from the slow waves seen during propofol induction (fig. 6) and during deep dexmedetomidine sedation (fig. 11D) in that they are accompanied by a substantial reduction in spectral power at all frequencies above 10 Hz (fig. 13A, minutes 86 to 90). Although the mechanism of this slow oscillation is unknown, one possible explanation is the blockade of NMDA receptor–mediated excitatory inputs from the parabrachial nucleus and the median pontine reticular formation to the basal forebrain and to the thalamus.124,125 The beta-gamma oscillations may have a mechanism similar to that of ketamine (fig. 9B) whose increased beta-gamma activity depends critically on NMDA-mediated inactivation of inhibitory interneurons.45
Anesthesiologists administer selected combinations of drugs to create the anesthetic state best suited for the patient and the given surgical or diagnostic procedure. A growing body of evidence suggests that the behavioral effects of the anesthetics are due to neural oscillations induced by their actions at specific molecular targets in specific neural circuits. Anesthesia-induced oscillations are 5 to 20 times larger than normal brain oscillations, likely disrupt normal brain communication (fig. 1) and are readily visible in the unprocessed electroencephalogram (fig. 2) and its spectrogram (fig. 3). Therefore, for three widely used intravenous anesthetics—propofol, ketamine, and dexmedetomidine—we related the electroencephalogram signatures to the molecular targets and the neural circuits in the brain at which these drugs most likely act (figs. 4–11). For the inhaled ether-derived anesthetics such as sevoflurane, isoflurane, and desflurane, we observed that, with the exception of the theta oscillations that appear around 1 MAC and beyond, their electroencephalogram patterns during maintenance and emergence closely resemble those seen in propofol (fig. 12). Nitrous oxide is known to be associated with increased beta and gamma oscillations and likely decreased slow-delta oscillations. However, we demonstrated that nitrous oxide also produces profound slow-delta oscillations during the transition from an inhaled ether anesthetic (fig. 13).
In contrast to brain state monitoring based on the electroencephalogram-derived indices, which assumes that the same index value defines for any anesthetic the same level of unconsciousness, the unprocessed electroencephalogram and the spectrogram define a broader range of brain states. Therefore, under our paradigm, their use should enable a more nuanced characterization of brain states guided by mechanistic insights. Our paradigm for brain state monitoring differs also from that used by neurologists and clinical neurophysiologists. These clinicians commonly use unprocessed electroencephalogram patterns to identify seizures in the intensive care unit,67,126 to characterize evoked potentials,127 to detect ischemia during neurophysiological monitoring in the operating room,128 and to characterize sleep stages.129 Their paradigm relies on identifying patterns in the unprocessed electroencephalogram to diagnose seizures or ischemia irrespective of the anesthetic. To do so, clinical neurophysiologists frequently ask anesthesiologists to reduce or not use certain anesthetics, for example, not to use ether anesthetics during spinal cord monitoring, to facilitate the detection of clinically significant changes in unprocessed electroencephalogram patterns.120, 130 Hence, the neurologist’s paradigm does not require understanding the electroencephalogram signatures of the anesthetics.
Our paradigm synthesizes research from the last 80 yr. Gibbs et al.1 showed that under general anesthesia, electroencephalogram patterns changed with level of unconsciousness. Several investigators showed that different anesthetics have different electroencephalogram patterns and that these patterns were visible in the spectra of the electroencephalogram.10,12–14,22 In 1959, Martin et al.6 wrote, “The big question, of course, remains unanswered; namely, do the electroencephalographic patterns defined in the several classifications presented constitute a valid measure of depth of anesthesia. We believe that this question unanswerable until a precise definition of ‘depth of anesthesia’ is made.” Relating the electroencephalogram patterns to the mechanisms was not possible at the time because lipid solubility was the primary theory of anesthetic mechanisms. The concept that anesthetics bind to specific molecular targets was not promulgated until 1984,131 and the detailed neural circuit descriptions of how anesthetic actions at specific molecular targets could lead to altered states of arousal were not reported until more than 25 yr later.45,46 In recent years, more has been learned about the biophysics of the electroencephalogram53 and about how altered states of arousal induced by anesthetics relate to electroencephalogram activity.19,20,43,47,57,60 Hence, it is now possible to link electroencephalogram signatures to anesthetic state and the actions of the drugs at specific molecular targets and in specific neural circuits (fig. 14).
Today, the unprocessed electroencephalogram and the spectrogram are displayed on several brain monitors.31,50,132 Accurate spectral analyses are required to track accurately the anesthetic effects. For this reason, we computed our spectrograms using multitaper spectral methods. For a given data length, multitaper methods have been shown to be the optimal nonparametric spectral techniques in the sense of giving the spectral estimates with the highest resolution and the lowest variance.64,65,133 As a result, the multitaper methods make it easier to identify the spectral features of anesthetic-specific signatures.
For the information we have covered in part I to be useful in the management of patients receiving general anesthesia and sedation, it is important to describe how the electroencephalogram patterns change as different drugs are combined because this is the more common scenario in anesthesiology practice. The effects of combinations of anesthetics on the electroencephalogram will be the topic of part II. An animated version of portions of parts I and II are available at www.AnesthesiaEEG.com. In part III, we will review the neurological examination for anesthesiologists.
Supported by grants DP1-OD003646 (to Dr. Brown), DP2-OD006454 (to Dr. Purdon), and TR01-GM104948 (to Dr. Brown) from the National Institutes of Health, Bethesda, Maryland, and funds from the Department of Anesthesia, Critical Care, and Pain Medicine, Massachusetts General Hospital, Boston, Massachusetts.
Masimo, Irvine, California, has signed an agreement with Massachusetts General Hospital, Boston, Massachusetts, to license the signal processing algorithms developed by Drs. Brown and Purdon for analysis of the electroencephalogram to track the brain states of patients receiving general anesthesia and sedation for incorporation into their brain function monitors. The other authors declare no competing interests.