Functional magnetic resonance imaging offers a compelling, new perspective on altered brain function but is sparsely used in studies of anesthetic effect. To examine effects on verbal memory encoding, the authors imaged human brain response to auditory word stimulation using functional magnetic resonance imaging at different concentrations of an agent not previously studied, and tested memory after recovery.
Six male volunteers were studied breathing 0.0, 2.0, and 1.0% end-tidal sevoflurane (awake, deep, and light states, respectively) via laryngeal mask. In each condition, they heard 15 two-syllable English nouns via closed headphones. Each word was repeated 15 times (1/s), followed by 15 s of rest. Blood oxygenation level-dependent brain activations during blocks of stimulation versus rest were assessed with a 3-T Siemens Trio scanner and a 20-voxel spatial extent threshold. Memory was tested approximately 1.5 h after recovery with an auditory recognition task (chance performance = 33% correct).
Scans showed widespread activations (P < 0.005, uncorrected) in the awake state, including bilateral superior temporal, frontal, and parietal cortex, right occipital cortex, bilateral thalamus, striatum, hippocampus, and cerebellum; more limited activations in the light state (bilateral superior temporal gyrus, right thalamus, bilateral parietal cortex, left frontal cortex, and right occipital cortex); and no significant auditory-related activation in the deep state. During recognition testing, subjects correctly selected 77 +/- 12% of words presented while they were awake as "old," versus 32 +/- 15 and 42 +/- 8% (P < 0.01) correct for the light and deep stages, respectively.
Sevoflurane induces dose-dependent suppression of auditory blood oxygenation level-dependent signals, which likely limits the ability of words to be processed during anesthesia and compromises memory.
TO understand the mechanisms of anesthetic action requires various levels of analysis.1Although molecular and cellular studies help to elucidate the chemical properties of agents, they do not necessarily explain how brain structures are affected and cognitive function is compromised. Clinical endpoints, such as the ablation of consciousness and memory,2are traditionally addressed with behavioral measures (subject movement, performance on memory tests)3–6and, more recently, in combination with physiologic measures such as the electroencephalogram.7–9Advances in neuroimaging take these studies one step further as positron emission tomography and functional magnetic resonance imaging (fMRI) offer a unique opportunity to visualize changes in cerebral functionality, with excellent spatial resolution. In effect, these techniques bring us closer to the neural basis of anesthesia-induced unconsciousness and amnesia.
Using positron emission tomography, Alkire et al. 10–14have shown that the thalamus plays a key role in anesthesia-induced unconsciousness, and Veselis et al. 15–17have highlighted elements of anesthesia-induced amnesia involving the right prefrontal and parietal cortices. Although fMRI has several advantages over positron emission tomography,18including superior temporal and spatial resolution, fMRI studies of central anesthetic effect in humans are rare. Examined were the effect of isoflurane on visual19and noxious or tactile stimulation20and the effect of propofol on auditory language processing.21All three studies demonstrated dose-dependent decreases in stimulus-related activation of appropriate cortical (i.e. , visual, somatosensory, auditory) areas, with some residual activation at low (sedative) concentrations. None addressed effects on memory.
To explore mechanisms of anesthesia-induced amnesia, we examined the effect of increasing concentrations of sevoflurane on brain activity in response to auditory stimulation using blood oxygenation level–dependent (BOLD) fMRI and two-syllable English words. After recovery from anesthesia, memory for presented stimuli was tested using an auditory recognition task. We hypothesized that anesthesia suppresses brain activity, and thereby memory function, in a dose-dependent fashion.
Materials and Methods
Six volunteers were included (mean age, 23 yr; range, 22–24 yr) and were paid for their participation in this study, which was approved by Emory University’s Institutional Review Board (Atlanta, Georgia). Eligible for participation were right-handed, nonsmoking male volunteers aged between 21 and 35 yr who were fluent in English and weighed less than 115 kg. Exclusion criteria (ad verbatim ) were known (anesthetic) drug allergies; history of or current illicit drug abuse (per urine sample); psychoactive drug treatment; history of head trauma resulting in loss of consciousness; history of or current neurologic, cardiovascular, or respiratory disease; claustrophobia; a hearing deficit; or a memory disorder. Criteria were checked through personal interview and confirmed during an orientation session with the treating physician. Candidates signed informed consent before illicit-drug screening. After having been found eligible for participation, volunteers agreed to a study time (early morning, late morning, or early afternoon).
Subjects arrived nil by mouth (for 8 h) at the study site and were studied under three conditions (within-subjects factor): an awake condition (0.0% end-tidal concentration sevoflurane [Etsevo]), a “light” anesthesia concentration (1.0% Etsevo, approximately minimum alveolar concentration [MAC]awake22,23), and a “deep” anesthesia concentration (2.0% Etsevo, approximately 1 MAC in this age group24). The order of concentration was not counterbalanced because of the small sample size, and started with 0.0%, followed by 2.0% and 1.0% Etsevofor all. At each level, a word list was presented, and concordant imaging data were collected (see next two sections). After data collection in the awake state and before induction of anesthesia, volunteers gargled with 4% viscous lidocaine. Anesthesia was then induced with sevoflurane in oxygen, using the single-breath technique. A laryngeal mask airway was then secured, after which Etsevowas held constant at 2% for 15 min before stimulus presentation to allow for effect site equilibration. After data collection at this level, Etsevowas adjusted to 1% and held constant for 15 min before stimulus presentation. Volunteers breathed spontaneously throughout the anesthetic procedure, and oxygen flow was not adjusted after induction. Sevoflurane was discontinued after data collection at the 1% level, and volunteers were allowed to recover. Standard anesthetic monitoring was used throughout the experimental (fMRI scanning) procedure.
Volunteers were presented with three lists of words over the course of the experiment, a different list at each level of anesthesia. The assignment of lists to anesthetic levels was counterbalanced. Each list contained 15 two-syllable English words drawn from the Toronto Noun Pool, available on-line from the Computational Memory Lab,**which contains more than 400 nouns and their lexical properties, such as word frequency (a common index of familiarity). Because word familiarity affects recognition memory performance, words were primarily matched on their lexical frequency and were selected from the lower ranges because recognition memory for low-frequency words is higher than for high-frequency words. In addition, words with similar meaning were avoided by checking free-association norms,25available at Appendix A on-line.††Based on these criteria, 135 words (mean Kucera and Francis word frequency = 40 ± 16 per million) were selected: 45 served as stimuli during the scanning phase of the experiment (“targets”), and the remaining 90 served as filler items (“distractors”) during the retrieval phase. The appropriate sound files were downloaded from the Internet and uploaded to a laptop computer (iBook; Apple Computer, Inc., Cupertino, CA) with appropriate software26for auditory presentation of stimuli.
To study auditory perception and memory function at each level of anesthesia, the 45 target words were divided into three lists (A, B, and C), matched for word frequency. Volunteers listened to one list at each level of anesthesia, the order varying randomly between subjects. Within lists, the order of words was fixed. Each word was presented 15 times in succession, for 1 s each presentation, after which 15 s of rest (no word presentation) followed. In this way, BOLD signals during alternating periods of auditory word stimulation and rest were obtained. Subjects were asked to listen passively to the words during the awake state but were not given specific instructions to memorize the words. They were aware that their memory would be tested, as outlined in the consent form. Words were presented via headphones connected to the scanner’s audio system and the laptop in the control room. The headphones completely covered the subject’s ears and also served to help shield the subject from scanner noise. The auditory stimuli were presented at the highest possible amplified computer volume without producing sound distortion to make the words audible against the background scanner noise. When the scanner was started, a pulse simultaneously triggered the series of auditory word presentations, thereby time locking the BOLD signal to task-related stimulation.
Scanning Specifics and Procedure
All magnetic resonance imaging scans were obtained on a 3-T Siemens Trio scanner (Siemens Medical Solutions, Malvern, PA). Each volunteer session started with a high-resolution three-dimensional anatomical scan to assess brain anatomy. These were standard 1 × 1 × 1-mm isotropic voxel, axial T1-weighted anatomical scans. Volunteers then listened to the first list, in which repeated words alternated with periods of rest while functional scans were acquired as described above (total scan duration, 7.5 min). Gradient recalled echoplanar imaging to assess task-related BOLD activations was acquired using pulse sequence parameters TR/TE of 2,500/30 ms, 34 slices, and 3 × 3 × 3-mm isotropic voxels. After the first series of scans, volunteers were taken out of the scanner for induction of anesthesia. The second series of scans (deep anesthesia) started with an anatomical scan to correct for changes in position since the preceding scans, followed by the functional scans. For the third series (light anesthesia), the functional scans were conducted, but an anatomical scan was only required if it was determined that the subject moved significantly. This occurred for one subject.
To test volunteers’ memory for presented words, an auditory recognition task consisting of 45 items was created. Each item contained three words, one of which had been played previously during the experiment (i.e. , a word from list A, B, or C). The target was uploaded in Audacity, a shareware audio editor,‡‡together with two distractors to create a three-word audio sequence. Within items, words were matched on word frequency to counteract familiarity effects other than those induced by the experiment. Across items, targets were evenly dispersed to occur in the first, second, or third place in the sequence. After recovery from anesthesia, volunteers listened to all 45 test items in a randomly determined but fixed order. The three words composing each item were presented sequentially with 1 s between words, after which the series was repeated. Subjects then decided which word had been played during the scanning phase (three-alternative forced choice), and indicated their confidence (sure vs. not sure). To index memory function, the number of correctly recognized words was calculated for each condition. Because every test item contained one target and two distractors, chance performance equaled 33% correct for each condition.
Neuroimaging data were preprocessed and analyzed using Statistical Parametric Mapping27,28version 99.§§All functional echoplanar-imaging volumes were realigned to the first volume and resliced. Each volume was then normalized to a standard echoplanar-imaging template volume using 3 × 3 × 3-mm voxels in Talairach space.29Images were subsequently smoothed with a 6-mm isotropic gaussian kernel, proportionally scaled, and band-pass filtered in the temporal domain: The low-pass filter corresponded to the frequency of the canonical hemodynamic response function as implemented in the software package; the high-pass filter was 150 s. Condition effects were estimated at each voxel according to the general linear model, and regionally specific effects were compared using linear contrasts, implementing a fixed-effects statistical model. Because the neural response to repeated stimuli typically showed decreases in specific brain regions relative to single presentations, we predicted that some brain regions (e.g. , those more involved in processing word identification or meaning) would show an initial activation at the beginning of each block of 15 repeated words but would show a rapid decrement across repetitions. Therefore, we assessed two different types of functional responses: a rapidly habituating component (transient responses) sensitive to responses that decreased quickly across each block of repeated words, and a sustained component that was sensitive to responses that were maintained across the block (e.g. , regions sensitive to auditory stimulation that did not exhibit significant habituation). These components were modeled using the “mean and exponential decay” basis set in the software package, in which the diminishing exponential decay term corresponds to the transient activations and the mean term corresponds to the sustained activations (fig. 1). Each term was convolved with a canonical hemodynamic response function to account for the effects of hemodynamic delay on the BOLD signal. Contrasts between conditions produced statistical parametric maps of the t statistic at each voxel, which were subsequently transformed to the Z distribution. We thresholded these summary statistical maps at a voxel-wise intensity threshold of P < 0.005 (uncorrected for multiple comparisons) with a spatial extent threshold of 20 contiguous voxels (i.e. , a group of voxels had to be at least 20 voxels large to appear as an activated cluster). This threshold is based on previous studies of cortical activations to sensory stimuli (which correspond to a probability of approximately P < 0.05). For visualization purposes, these activation maps were overlaid on a high-resolution structural T1-weighted image.
Memory performance was assessed by comparing the number of correct recognition responses (i.e. , selection of a word presented earlier during scanning) across the three anesthesia conditions and against chance performance (= 0.33 relative frequency). Analysis of variance was used to analyze condition effects; three one-sample t tests determined the reliability of deviations from chance. For these data, P < 0.05 was considered statistically significant.
After the awake state, anesthesia was administered for 1:30 ± 0:30 h (range, 1:15–2:15 h). One patient moved when the scanner sequence was started at 1.0% Etsevo. His light concentration was thereupon increased to 1.3%, which rendered him motionless after 15 min of effect-site equilibration and for the remainder of the scanning procedure. Removal of this subject’s data did not significantly affect the finding reported below.
All volunteers maintained normal physiologic parameters during anesthetic administration. End-tidal carbon dioxide varied overall between 30 and 45 mmHg. Variation within individuals was limited (average SD = 1.87 mmHg, maximum recorded absolute difference = 8 mmHg), and across conditions, end-tidal carbon dioxide, blood pressure, and heart rate changed as indicated in table 1. Memory testing took place 1:27 ± 0:26 h (range, 1:00–2:15 h) after recovery.
Tables 2 and 3specify the areas of habituating and sustained brain activations, respectively, in response to repeated auditory word stimulation across the three levels of anesthesia. In the awake state, magnetic resonance scans showed widespread activations during auditory word presentation (P < 0.005), including bilateral superior temporal gyri, thalamus, striatum, and frontal and parietal cortices (fig. 2). During light anesthesia (1% Etsevo), we observed more limited activation involving the bilateral superior temporal gyrus, right thalamus, bilateral parietal cortex, left frontal cortex, and right occipital cortex. No significant task-related activation was observed during deep (2% Etsevo) anesthesia at this threshold.
Overall, the number of activated voxels (P < 0.005) during 1% Etsevowas reduced by approximately 90% compared with awake and was completely suppressed during 2% Etsevo. A comparison of brain activations in the awake versus light anesthesia states (table 4) identified largely similar and overlapping areas with the awake state alone, illustrating the suppressive effect of 1% end-tidal sevoflurane. Furthermore, all of the activations listed for the light state in tables 2 and 3were significantly different from those in the deep state at our statistical threshold.
When neural responses that were sustained across 15 presentations of the same word were contrasted with neural responses that rapidly habituated with repetition, a clear pattern emerged. Sustained responses during the light anesthetic state were restricted to auditory processing regions in the bilateral superior and middle temporal gyri (table 3). Transient, habituating responses, by contrast, were found in the bilateral superior temporal gyri as well as in a more extensive network of frontal, parietal, and occipital regions (table 2), suggesting that aspects of auditory word processing beyond low-level auditory processing were recruited (e.g. , word identification and meaning).
One and a half hours after recovery, volunteers recognized on average (± SD) 77 ± 12% of words presented while awake, as opposed to 32 ± 15 and 42 ± 8% for the light and deep anesthetic stages, respectively (fig. 2). Recognition performance declined significantly between the awake and anesthetized conditions (P < 0.01) but was still slightly above chance for words presented during 2% Etsevo(P < 0.05). No significant memory effect was observed for words presented during the last phase of the experiment, at 1% Etsevo. Confidence ratings of recognition decisions (fig. 3) indicated that volunteers believed they were merely guessing as far as words presented during anesthesia were concerned.
Two main findings emerged in this study of central anesthetic action in human male volunteers: On one hand, BOLD activations in response to auditory stimulation were suppressed by sevoflurane in a dose-dependent manner. Recognition memory performance, on the other hand, did not show a dose-dependent decline. Both results are discussed in this section, along with recommendations for future studies.
The BOLD activations that were observed in response to auditory stimulation during the awake and light anesthetic stages of the experiment are in agreement with the localization of auditory processing as identified by fMRI, with the thalamus and superior temporal gyrus centrally involved with primary auditory perception.30Our results complement those reported by Heinke et al. ,21who studied the effect of propofol on brain response to auditory sentences. In particular, we are able to confirm that ongoing temporal lobe activations occur at steady state concentrations of anesthesia, in this case 1% Etsevo. The temporal lobe BOLD response observed by Heinke et al. 21dissipated over the course of a slow (10-min) infusion period with 3 mg/kg propofol, whereas we observed a BOLD response consistently after 15 min of equilibration, when functional data were acquired. The level of sedation (1% Etsevo) was sufficiently deep to keep all but one volunteer from moving when the loud echoplanar-imaging sequence started. Similarly, volunteers in the study by Heinke et al. did not maintain motor response over the course of the sedative infusion. These findings support the notion that residual auditory processing is a reliable phenomenon during light anesthesia.
In addition to mediating auditory perception, the thalamus is increasingly recognized for its role in anesthesia-induced unconsciousness.13,14,16It is interesting to note in this respect that thalamic activation was still present at 1% Etsevo. A formal assessment of sedative effect was not attempted in this study because of limited access to subjects during scanning, but thalamic functionality has been observed during non–rapid eye movement sleep31and beyond the point where anesthetics impair episodic memory function16and ablate brain response to tactile stimulation.20The striatum, on the other hand, plays a major role in voluntary motor response and is implicated in the development of habits and probabilistic responses.32In the absence of a distinct motor task, the striatal activations in the awake condition aligns better with the notion that part of the striatum supports executive function,33a key element of working memory.34Working memory is also implicated by the frontal and parietal activations we observed during both the awake and light anesthesia conditions, with the left prefrontal cortex being specifically associated with verbal memory encoding.35In general, our imaging results support the notion that anesthesia suppresses brain metabolism in a dose-dependent fashion, with some regional variation.
Although the average response during 15 presentations of the same word indicated that stimuli were perceived during light anesthesia (i.e. , activation of the auditory cortex), a more extensive network of transient, habituating responses suggested that aspects of auditory word processing beyond low-level auditory processing were recruited as well. The fact that these more extensive responses rapidly habituated is consistent with the conclusion that after repeated presentations of the same word, some aspects of lexical processing (i.e. , word identification and processing of meaning) had already been completed.
Based on these particular observations and the partially preserved BOLD response during light anesthesia in general, we anticipated a memory effect for presented material. Evidence of preserved memory function has been found at concentrations lower than 2.0% Etsevo. Maye and Smith36observed implicit (unconscious) but not explicit (conscious) memory for answers to general knowledge questions presented at steady state 0.3% Etsevo(supplemented with remifentanil). Renna et al. 37demonstrated a priming effect for words presented at 1.2% Etsevobut, notably, not at higher concentrations (1.5 and 2.0% Etsevo). Also at higher concentrations (1 MAC), Aceto et al. 38did not demonstrate evidence of memory function. These findings are in accordance with the BOLD responses we observed and present here. However, they are not in agreement with our memory test findings, as discussed next.
The awake activations, in particular those involving the (para)hippocampal, frontal, and parietal areas, are well in line with the substantial memory effect observed for this level. However, although the BOLD signal and the studies outlined above supported our contention that memory function was partially preserved at 1% Etsevo, we found no reliable memory for words presented at this level: Volunteers could not discriminate between old and new items. However, because all had first been exposed to 2% Etsevo, carryover effects of the deep level cannot be excluded. Although pharmacologic carryover effects were supposedly minimized by the 15-min equilibration period, the lagging effect of deep anesthesia on cognitive functionality at subsequent lighter levels is unknown. Suppression of memory function may be expected, however, based on the anterograde amnesic effect that many anesthetics induce. Also, the duration of anesthesia was longest in the 1% condition, it being the last of three successive conditions. Equally important are effects of continued workload (e.g. , fatigue), which may have been substantial in the 1% condition. To elucidate these matters would require a larger number of subjects with order of anesthetic concentration counterbalanced. Counterbalancing of anesthetic doses may pose practical challenges, however, such as establishing an airway under light rather than deep sedation.
The memory effect for words presented at 2% Etsevo, on the other hand, is hard to explain with the lack of auditory activation as measured by BOLD. In addition, this finding is inconsistent with the memory studies outlined above, which report no evidence of memory function at 2% Etsevo. Therefore, it is possible that the above-chance recognition performance is a type I error in which the null hypothesis (i.e. , no memory function) should be accepted but for statistical reasons is not. Although such errors are unlikely (P < 0.05), future studies will help to establish the reliability of current findings.
Recent interest has focused on identifying brain activity during initial processing of stimuli (memory encoding) that predicts whether stimuli will be remembered successfully on a subsequent memory test.35,39Such studies are typically conducted with awake subjects and use large numbers of stimuli. In studies during anesthesia, by contrast, the number of items is typically small because processing resources are limited and easily attenuated. This small number poses a constraint on retrieval-based comparisons. Therefore, in future studies, investigators may want to use more stimuli that are preferably simpler in nature (e.g. , sounds as opposed to words) to maintain a reasonably low cognitive load. Simple, in particular novel, sound stimuli can induce cognitively meaningful brain responses, as auditory evoked-potential studies demonstrate. Conversely, it may be fruitful to rely on physiologic (magnetic resonance imaging, positron emission tomography, electroencephalography) rather than behavioral outcome measures to index preserved function.
Finally, decreases in the BOLD signal may arise from both suppressed neuronal activity and hemodynamic changes, such as reduced cerebral blood flow or volume. We did not measure regional changes in cerebral blood flow or volume. Therefore, the net effect of sevoflurane on neuronal metabolism in the activations identified in this study remains to be determined. Subanesthetic concentrations of sevoflurane may increase cerebral blood flow in gray and white matter, in particular in occipital areas, while leaving cerebral blood volume largely unaffected.40On the other hand, 1.7% sevoflurane may reduce cerebral blood flow by as much as 50% depending on the region of interest.41Under these circumstances, the cerebellum is deeply affected. There is evidence to suggest that the flow–metabolism coupling is maintained at 1.7% sevoflurane,42suggesting that measured changes in the BOLD signal do reflect changes in neuronal activity.
We conclude that sevoflurane markedly suppresses brain response to repeated auditory word stimulation, as indicated by BOLD fMRI. A distributed network of transient, rapidly habituating responses suggested that word processing beyond low-level auditory processing was recruited during 1% Etsevo. No such responses were identified at 2% Etsevo, nor were any sustained responses observed at the selected threshold. Memory function was markedly compromised under anesthesia, as evidenced by recognition test performance after recovery, but did not necessarily parallel observed BOLD responses. Statistical and practical issues may account for this controversy.
The authors thank David I. Bauman (Anesthetist, Department of Anesthesiology, Emory University School of Medicine, Atlanta, Georgia) for assistance in administering anesthesia and Robert L. Smith III (Research Technologist, Department of Biomedical Engineering, Emory University School of Medicine/Georgia Institute of Technology, Atlanta, Georgia) for assistance in magnetic resonance imaging data acquisition.