Propofol and Midazolam Inhibit Conscious Memory Processes Very Soon after Encoding: An Event-related Potential Study of Familiarity and Recollection in Volunteers

Healthy volunteers were recruited by Internet advertisement (CraigsList**) and word of mouth. The study was approved by the Institutional Review Board of Memorial Sloan-Kettering Cancer Center, New York, New York. Volunteers gave their informed consent and were paid $200 for their participation.

Seventy-nine healthy volunteers were recruited; 12 withdrew their consent to participate, thus 67 volunteers participated. Of these, 55 are included in the current study. In the remaining 12 volunteers, 6 did not remain awake or failed to perform the memory task adequately during drug administration (1 thiopental, 1 propofol, 1 midazolam, 3 dexmedetomidine), and the electroencephalogram was not recorded in 6 voluneers for technical reasons (1 placebo, 2 thiopental, 2 propofol, and 1 dexmedetomidine participant).

Inclusion criteria included evidence of good health, normal vision, age between 18–50, and being right-handed, of normal weight for height, a high school graduate, and fluent in English.

Exclusion criteria were evidence of medical, neurologic or psychiatric disease, currently taking central nervous system active medication, visual impairment, body mass index ≥ 30, history of alcohol or drug abuse, and pregnancy. Recruitment was followed by a telephone interview and in-person orientation session, where volunteers were given a brief physical exam and trained on the experimental tasks. Volunteers who did not perform the experimental task adequately were excluded from study. Volunteers were screened for handedness using the Edinburgh Handedness Inventory.26 

Participants were randomized to one of five experimental groups who received placebo (n = 11), thiopental (n = 11), propofol (n = 10), midazolam (n = 12), or dexmedetomidine (n = 11). The first target dose of drug was predicted to produce 40% memory impairment (40AmnDose), estimated from a previous study.25The first set of CRTs was then administered, with stimuli delivered in 3 equal blocks of approximately 7 min duration each, with 5 min of rest between blocks (table 1). After the low-dose condition, a somewhat higher dose was given, estimated to provide 60% memory impairment (60AmnDose), and the CRT was repeated with new stimuli. Drug infusion was then discontinued, and repeated recognition testing was performed over the subsequent 3.75 h, the detailed results of which will be reported in a companion manuscript. Only the final recognition task at 225 min (end of day) is reported in the current manuscript, with data analyses focused on results from the CRTs, when drug was present.

After verification of nil per os  (nothing by mouth) status, electroencephalogram electrodes were applied using collodion adhesive. A 20-g IV was started using D5½NS running at approximately 125 ml/h. Including electrode application and removal, the session lasted from 8:00 am until about 4:30 pm. Volunteers were provided a light lunch after the end of drug administration.

Participants were informed that they would view a series of pictures in which some pictures would be repeated. Each picture was shown twice during the CRT, with recognition either 6 or 27 s (but not both) after first presentation. The CRT was designed for maximal efficiency, and almost all pictures were shown twice with a small number of additional neutral pictures acting as filler items. Participants categorized the stimulus as either new (first presentation), or a repeat of a previously seen image (old) by pressing a mouse button. Reaction times (RT) were collected on each trial, with ones less than 100 ms being excluded.

Recognition tasks contained an equal number of new and old pictures that had been previously presented during one of the CRTs. Thus, each picture was shown twice on the CRT and a third time during recognition testing after drug discontinuation.

Nineteen channels of the electroencephalogram signal (plus vertical and horizontal electrooculography and left and right mastoid) from 1–75 Hz were sampled at 1 kHz using Scan Synamps amplifiers in the standard 10/20 configuration with impedances of less than 5 Kohms (Compumedics Neuroscan, Charlotte, NC). A ground electrode was placed on the forehead. All leads were wrapped and shielded to decrease 60 Hz noise. After accrual of approximately one third of participants, a large 2 × 4–inch disposable electrode (Viasys Healthcare, Madison WI) was placed on the left shoulder to provide additional grounding to decrease 60 Hz noise. During acquisition, electroencephalogram data were referenced to the left mastoid, but were rereferenced to linked mastoids for ERP analysis.

In offline processing, electroencephalogram was epoched to 2500 ms, with a prestimulus baseline of 199 ms. Vertical and horizontal electrooculography artifacts were reduced using the Neuroscan regression algorithm (Compumedics Neuroscan Scan 4.3, Charlotte, NC). Epoched files were visually inspected, and artifacts were removed (after automated rejection of amplitudes exceeding ±75 μV). Baseline correction, rereferencing, and averaging over the three CRT tasks presented at each dose level was then performed.

For each participant, electrodes with fewer than 10 valid sweeps were eliminated from the analysis. Most commonly, invalid sweeps represented an incorrect recognition response but could also include excessive artifact. In all participants in the current study, recordings from the Cz and Pz electrodes were of good technical quality. ERP waveforms were filtered with a 20-point smoothing algorithm (equivalent to a low-pass filter of 6.9 Hz) for graphical presentation only. Unfiltered values were used in statistical analyses.

Pictures used on the CRT were selected from the International Affective Picture System.,27with all tasks counterbalanced for valence and arousal criteria. Pictures were presented in color in full screen mode, displayed for 1.5 s, with a 3 s interstimulus interval. A total of 140 unique pictures were presented at each dose level, divided over 3 blocks in counterbalanced order. Subjects were seated in a semireclining posture on a gurney, about 2.5 feet from the screen of a CRT or (after 33 participants) a liquid crystal display monitor.

Drugs were infused using STANPUMP,††which administers a continuous infusion of drug adjusted for body weight and age using the appropriate pharmacokinetic models (thiopental, Shafer; propofol, Schnider; midazolam, Greenblatt; dexmedetomidine, Markku Anttila). Two sequential drug concentrations used were: 1.5 and 3 μg/ml thiopental, 0.45 and 0.9 μg/ml propofol, 20 and 35 ng/ml midazolam, and 0.2 and 0.4 ng/ml dexmedetomidine. Placebo solutions were matched in appearance to active drugs (thiopental, multivitamin solution; propofol, intralipid; midazolam and dexmedetomidine, normal saline), and the corresponding STANPUMP parameters were used for infusion. Testing was started after predicted effect site concentrations were within 5% of the target concentration.

Results of memory tasks are expressed as percent correct recognitions, excluding trials where no response was made. Only correct recognitions were used in producing the average ERP waveforms.

Note that the CRT was divided into three segments at each dose level to avoid fatigue effects of one long CRT. The ERPs from these three segments were analyzed together at each dose level. ERPs were averaged by stimulus type (new [first presentation] and old [second presentation]) and by interval (6 and 27 s). Figure 2validates Cz and Pz electrodes as measuring familiarity and recollection memory processes, respectively. Average ERP amplitudes at the Cz and Pz electrodes for old (red and green waveforms) and new (blue waveforms) pictures over the 300- to 400-ms and 400- to 600-ms time windows, respectively, were calculated. Similarly, the area between old and new ERP waveforms was determined as the sum of differences for each millisecond (data were collected at 1 kHz) over the time window of interest (uV × sec). At Pz, the areas are therefore about twice as large as at Cz because the time window at Pz is 200 ms.

Data analyses were performed using linear mixed-effects models.28Bagiella and colleagues summarized the main advantages of mixed-effects models over repeated-measures ANOVA in psychophysiology research.29An important advantage is that casewise deletion of missing observations is no longer necessary, which allows the experimenters to analyze all available data with higher statistical power by comparison. Mixed-effects models also better handle the correlation structures of repeated measures nested within participants, which circumvents the need to check for heteroscedascity and sphericity assumptions and make statistical adjustments thereafter. Moreover, there is no need to use the quasi-F statistic to test random effects associated with experimental materials.30The mixed effects model has the ability to simultaneously include both categorical and continuous data.31 

For the current study, the factors of drug (placebo, thiopental, propofol, midazolam, dexmedetomidine), dose (40AmnDose, 60AmnDose), interval (6 or 27 s), and stimulus type (old, new) were incorporated as fixed factors. Participant (ID1-55) was used as a random effect in the model. To calculate the unbiased P  values in testing experimental effects,30we used the Markov-Chain Monte Carlo method in the statistical computer language R.‡‡Specifically, the lme4 package in R was used to implement the mixed-effects models analyses.§§The Markov-Chain Monte Carlo P  values were calculated using pvals.fnc and aovlmer.fnc functions available in the languageR package.∥∥The Markov-Chain Monte Carlo simulations were repeated 10,000 times to reach a stable burn in. P  values were based on t  statistics when model coefficient estimates were compared with the referent (e.g. , placebo group) or the F statistic when fixed factors in the model were tested. Confidence intervals (95%) of model coefficient estimates were obtained from Markov-Chain Monte Carlo and provided whenever appropriate. As an estimate of effect size, coefficients from the model fit (for the effect of drug in comparison with placebo) were divided by the SE for the given effect.

Fifty-five cases were available for analysis (men, 36; women, 19; age, 29.1 ± 8.1 yr; weight, 71.5 kg ± 12.4 kg; body mass index, 24.5 ± 5.7; 51% Caucasian; 22% African-American; 16% Asian; and 11% Hispanic). All but two individuals continued their education beyond high school, with 47.3% being college graduates and a further 18.1% having postgraduate education. As tested by one-way ANOVA, groups did not differ on any variable (table 2).

Testing was performed to assess whether drugs were administered in sufficient concentration to produce a measurable effect on behavioral responses (factor of drug), while still allowing adequate performance on the CRT (recognition memory). The similarity of drug effect was tested by excluding the placebo group. The factor of interval (recognition 6 or 27 s after presentation) was used to contrast working and long-term memory function, and the factor of stimulus type (old, new) was used to contrast reaction to pictures recognized from memory versus  new pictures. The difference between correct recognitions on the CRT and delayed recognition at the end of the study day (end-of-day recognition) measured the degree of forgetting.

Impairment was equal among active drugs, except for a greater effect of midazolam.

With the placebo group included, there was a significant effect of drug on reaction times (F(4,432) = 2.88; P = 0.002), with midazolam having RTs 187 ms (range, 70.1–301.2 ms) slower than placebo (t = 3.2; P = 0.001; fig. 3A; table 3). Excluding placebo, there was little or no difference among drugs on RTs (F(3,345) = 1.96; P = 0.12). Thus all drugs slowed processing to a similar extent.##As dose increased, reaction times were slower, with the 60AmnDose slowing RTs by 76.7 ms (range, 59.1–94.8 ms) compared with the 40AmnDose (placebo group included, F(1,432) = 69.69; P < 0.001; effect size 8.3).

A false alarm (FA) occurred when a new picture was incorrectly recognized as being old. Increased FA rates may represent changes in attention or a change in bias. In other words, an increase in FA rate may represent a lower confidence level in being correct when the participant was uncertain. Overall, drugs increased FA rates by 6.2% (range, 1.5–11.0%) when compared to the placebo group (effect of drug F(4,104) = 3.07; P = 0.02), except possibly thiopental (t = 1.95; P = 0.054 vs.  placebo; fig. 3 B). As with reaction times, there was a large effect of dose, increasing FA rates by 6.5% (range, 4.8–8.3%) (F(1,104) = 60.79; P < 0.001; effect size 7.8). Excluding placebo, there was no drug by dose interaction (the amount of increase in FA rate as dose increased was the same for each drug; F(3,80) = 1.04; P = 0.38). Thus, all drugs had similar effects on FA rates among themselves.

Recognitions of pictures from working memory (6 s after presentation) was 22 ms (range, 3.8–39.9 ms) faster than recognition after 27 s (F(1,432) = 5.83; P = 0.002; fig. 3A). This indicates that at 6 s, the picture was in working memory and thus easier to match than at 27 s, when the picture would be gone from working memory, requiring a search in long-term memory. There was no drug by interval interaction (F(4,428) = 0.44, NS), i.e. , there was no difference with any drug from placebo in the relationship of RT from working versus  long-term memory.

In contrast to reaction times, there was no effect of interval on recognition memory (F(1,213) = 0.48, NS; fig. 3C). Thus, participants recognized pictures in working and long-term memory equally well in the presence of drug, despite the presence of sedation as measured by overall increased reaction times.

Overall, participants recognized old pictures 108.8 ms (range, 91.3–127.6 ms) faster than new ones (F(1,432) = 140.35; P < 0.001; effect size 11.8), indicative of a repetition effect.32This indicates that it was easier to match a picture already in memory (i.e. , the recognized old picture) than to search for one that was not there. There was a drug by stimulus type (old, new) interaction (F(4,428) = 3.22; P = 0.01). Pairwise testing revealed that thiopental increased the difference between these categories of stimuli by 69.3 ms (range, 12.8–126.3 ms) more than in the placebo group (t = 2.43; P = 0.02). In other words, there was no indication that the presence of drug delayed retrieval of pictures in memory in relation to new presentations.

Correct recognitions of pictures on the CRT revealed similar effects for the factor of dose as for reaction times (fig. 3Cshows the inverse effect of panel A). There were 6.8% (range, 5.13–8.42%) fewer recognitions as dose increased from 40AmnDose to 60AmnDose (F(1,213) = 66.6; P < 0.001; effect size 8.17 compared with the reaction time effect size 8.35). The interaction of drug by dose was significant for all drugs (F(4,209) = 9.34; P < 0.001; all drugs P < 0.02 compared with placebo). In other words, all drugs decreased recognitions on the CRT as dose increased, in contrast to the placebo group, which maintained performance on the second set of CRTs similar to the first. As memory function was the primary behavioral response of interest, data from both doses were analyzed separately.

At the lower 40AmnDose, there was a significant effect of drug (F(4,104) = 2.66; P = 0.04), with participants receiving midazolam having 9.8% (range, 3.1–16.3%) less success in recognition than placebo (t =–3.02; P = 0.003). Excluding placebo, there was similarity of drug effect (F(3,83) = 2.07; P = 0.11), though this level of significance suggested that midazolam may have had 7.6% (range, 0.6–15.2%) lower recognitions than propofol (t = 2.10; P = 0.04) and 7.5% (range, 0.2–14.5%) fewer recognitions than dexmedetomidine (t = 2.13; P = 0.04).

At the higher 60AmnDose, drug impaired memory compared with placebo (overall, factor of drug F(4,104) = 5.60; P < 0.001). Dexmedetomidine was not different from placebo, but other drugs had significantly lower recognitions (midazolam: 23.8% [range, 13.4–33.8%], t =–4.63, P < 0.001; propofol: 11.4% [range, 0.5–22.1%], t =–2.10, P = 0.04; thiopental: 11.1% [range, 0.6–21.6%], t =–21.2, P = 0.04 less than placebo). The midazolam group had lower recognition scores than the other drugs (excluding placebo, factor of drug F(3,83) = 3.04; P = 0.03; all drugs P < 0.04 compared with midazolam).

The effect of midazolam on correct recognitions on the CRT (effect size 4.25) was larger than the increase in reaction times (effect size 3.23), demonstrating a greater effect on memory than processing speed, i.e. , conscious amnesia.

After completion of the CRTs, drug infusion was stopped, and forgetting of newly learned information was measured by the difference from recognitions on the CRT to the end of the study day (end-of-day recognition).***

Pictures encoded at the higher dose were forgotten 4.4% (range, 0.8–8.2%) more than at the lower dose (factor of dose, F(1,213) = 5.45; P = 0.02), demonstrating that forgetting was a drug-related effect. The effect size of 2.34 for the factor of dose on forgetting is smaller than for its effect on CRT recognitions, 8.17. This indicates that forgetting was normalized to a certain degree among different starting points, i.e. , any memory effect present on the CRT was discounted in the forgetting measure.

There was significant forgetting in the placebo group alone (t = 6.14; P < 0.001), with a loss of memory for 33.9% (SD 18.3%) of pictures initially recognized on the CRT. All drugs had a greater forgetting rate than placebo (F(4,213) = 5.14; P < 0.001; fig. 3D). This difference ranged from 14.7% (range, 3.2–27.1%) greater than placebo for dexmedetomidine (t = 2.47; P = 0.01) to 23.4% (range, 11.1–35.2%) for propofol (t = 3.85; P < 0.001).

As opposed to recognitions on the CRT, midazolam was no different than other drugs in terms of forgetting (excluding placebo, factor of drug F(3,120) = 0.98) Thus, after the discontinuation of drug infusion, all active drugs had similar effect on the rate of forgetting. This measure normalized different starting points, including the difference of the midazolam group with other active drugs.

Note that separate forgetting rates were available for pictures recognized from either working memory (6 s) or long-term memory (27 s after presentation) on the CRT. Pictures recognized from long-term memory were forgotten 5.6% (range, 2.1–9.4%) less than overall forgetting (F(1,213) = 8.72; P = 0.004; effect size 2.95). Thus, if a participant could recognize a picture at 27 s, then they were more likely to recognize it at the end of the study day despite a large degree of forgetting. Recall that there was no effect of interval on correct recognitions during the CRT (fig. 3C). Thus, proportionally more pictures were forgotten from working than long-term memory. In other words, pictures encoded into long-term memory represented stronger memories. Excluding placebo, the effect of interval was larger (F(1,170) = 10.01; effect size 3.16), indicating that this differential forgetting effect for pictures recognized from working memory was exaggerated in the presence of drug. There was no interaction with drug by interval; in other words, all drugs had the same effect (F(4,209) = 1.74; P = 0.1). If one accepts that a significance of P = 0.1 was suggestive of an effect, then this was due to thiopental (t =–2.40; P = 0.02 compared with placebo). In other words, the relatively good retention of stimuli in long-term memory on the CRT may have been exaggerated for thiopental (compare third and fourth bars in thiopental cluster; fig. 3D). Note also that thiopental decreased the time for retrieval of these items from memory in comparison to other drugs (factor of stimulus type (old,new), see bottom of p. 301). These results point to thiopental lacking a conscious amnesic effect on memory, as represented by the other active drugs.

A more positive recognition ERP to a second stimulus presentation documents that this stimulus is present either in working memory (6 s) or long-term memory (27 s) during performance of the CRT. The factor of stimulus type (old, new) was tested to determine the presence of conscious memories. The difference between old/new waveforms (i.e. , area between waveforms) is a quantification of conscious memory processes.15,16Note that recognitions from working or long-term memory could involve the use of familiarity, recollection, or both memory processes. ERP measures of these processes were analyzed separately. The effect of drug (placebo, thiopental, propofol, midazolam, and dexmedetomidine) and interval (6 or 27 s) and their interactions were tested to determine the effect of different drugs on familiarity and recollection processes used in recognition of pictures from working and long-term memory. ERPs were obtained only for correct categorizations, and thus represent memory processes supporting successful recognition.

The difference between old and new ERP waveforms was highly significant, as shown by the slope of the line connecting average amplitudes in the associated time windows for new (first) and old (second) picture presentations on the CRT (fig. 4; table 4). For recollection, measured at the Pz electrode (400–600 ms), there was a 2.67-μV (range, 2.28–3.02 μV) more positive potential waveform for the second (old) presentation of the picture (F(1,423) = 200.53; P < 0.001; effect size 14.2), i.e. , the slopes of the lines in figure 4are ascending for all drugs in all conditions.

As with recollection, the old/new difference was robustly present for familiarity measured at the Cz electrode (time window, 300–400 ms), having a 2.24-μV (range, 1.73–2.74 μV) more positive potential for old pictures (F(1,420) = 77.86; P < 0.001; effect size 8.8) (table 5).

Thus, ERP waveforms confirmed the presence of memories in conscious or episodic memory, indicating that the CRT was successful in encoding memories.

The functioning of working memory was measured by the area between old and new waveforms to correctly recognized pictures 6 s after presentation (table 6; fig. 5, comparisons being made among values on the left side of the graphs).

There was no effect of dose either on familiarity (at Cz, F(1,101) = 0.33, NS) or recollection (at Pz, F(1,102) = 0.34, NS); areas at each electrode were collapsed across dose. There was no effect of drug on area associated with working memory for either familiarity (Cz: F = 2.06 [df 4,101]; P = 0.09) or recollection (Pz: F = 1.25 [df 4,102]; P = 0.29). If one accepts that P = 0.09 is suggestive of an effect of drug on familiarity, then this resulted from a larger area at Cz for propofol (t = 1.56 vs.  placebo; P = 0.12). These are key results; they showed that working memory in participants receiving drug functioned equivalently to placebo participants, with no decrement at higher doses of drug, corroborating behavioral results. Thus, any sedation effect was only evident by increased reaction times and errors, namely false alarm rates and less correct recognition. When recognition was correct, as was true for a majority of pictures, working memory processes supporting recognition were intact.

Recollection of a memory in a context of time and place is the memory process most visibly affected by midazolam and propofol. In this section, the impact of the study drugs on ERP measures of recollection, which occur in the parietal region 400–600 ms after picture presentation, is presented. The impact of drugs on the familiarity processes is discussed in the next section.

There was a shift of ERP amplitudes to less positive potentials as the interval of recognition increased from 6 to 27 s (fig. 4). ERP waveforms became less positive by 0.41 μV (range, 0.04–0.78 μV) (F(1,423) = 4.53; P = 0.03; effect size 2.2), with no interaction with drug for this effect (F(4,420) = 0.42). In other words, all drugs shifted ERP potentials to less positive values equally when recognition occurred from long-term as compared to working memory. This shift may represent a decrease in the efficacy of recognition processes from long-term memory, even though successful recognition occurred at the same rate as from working memory (though with longer reaction times). Recall that ERPs were obtained only to correctly recognized pictures; these ERP changes are independent of error rates. We will interpret this finding in terms of the pedestal effect in the discussion.

At the Pz electrode, there were shifts in ERP amplitudes to less positive values as drug dose increased, with ERP amplitudes being 0.75 μV (range, 0.39–1.13 μV) less positive (effect of dose: F(1,423) = 15.21, P < 0.001; effect size 3.9; dose increases from panels A and C to B and D in fig. 4). This relationship was different among drugs, and there was a significant drug by dose interaction (F(4,420) = 6.81; P < 0.001). Propofol was the only drug to reduce ERP potentials by 3.50 μV (range, 2.06–4.92 μV) as dose increased (t =–4.81; P < 0.001; propofol is highlighted in fig. 4). A similar but smaller effect may have been present for midazolam (t =–1.90; P = 0.06). A shift in ERP potentials towards less positive values is shown in figure 4by the position of the lines connecting new and old amplitudes moving towards the x-axis (amplitude = 0) as propofol, and possibly midazolam, doses increased.

A quantifiable measure of episodic memory processing is the area between old and new ERP waveforms.15,16This area can be approximated by the slope of the lines in figure 4, with 0 slope indicative of the absence of an old/new effect (e.g. , propofol in panel D of fig. 4has a smaller slope than in panel A, and thus a smaller area between waveforms). Area values are displayed in figure 5B. Though there were dose-related effects on ERP amplitudes (fig. 4), there was no dose-related effect on the areas between waveforms at Pz (F = 1.35; P = 0.25). Therefore, data were collapsed across doses.

There was a decrease of approximately 30% in area between waveforms when pictures were recognized 27 s after presentation compared to 6 s (F(1,209) = 8.34; P = 0.004; effect size 2.9). Thus, the size of the effect of interval on area was slightly larger than on the shift in ERP amplitudes to less positive voltages, whose effect size was 2.2.

Importantly, there was a suggestion of a drug by interval interaction, i.e. , the decrease in area when recognition occurs from long-term memory versus  working memory may have been different among groups (F(4,205) = 1.86; P = 0.12). Using pairwise comparisons for individual drugs with placebo, a highly significant drug by interval interaction was found for propofol (F(1,79) = 9.30; P = 0.003; effect size 3.1), with weaker evidence present for midazolam (F(1,83) = 2.91; P = 0.09; effect size 1.7). In other words, the decrease in area from 6-s to 27-s recognitions seemed to be larger for propofol than for placebo, as represented by the slopes of the lines for individual drugs in figure 5B.†††

The undifferentiated recognition response of familiarity occurs 300–400 ms after item presentation, maximal at the Cz electrode (fig. 2). This recognition represents one without full context; in other words, the stimulus is familiar, but details about when and where it was acquired are not retrieved at this early time point.

Similar to effects on recollection, there was a shift towards more negative ERP amplitudes in the presence of drug (factor of drug F(4,420) = 3.13; P = 0.01), with propofol being responsible for this effect. It produced 5.1-μV (range, 1.73–8.21 μV) more negative amplitudes (t =–3.10 vs.  placebo; P = 0.002; fig. 6). The voltages at Cz became 0.75-μV (range, 0.23–1.24 μV) more negative when recognition was from long-term memory compared with working memory (F(1,420) = 8.71; P = 0.003; effect size 3.0). Different from the recollection response at Pz, familiarity ERP amplitudes at Cz were not affected by dose (F(1,420) = 1.19; P = 0.28).

Similar to recollection processes, there was only a marginal effect of increased dose on the area between waveforms (F(1,207) = 3.40; P = 0.07), with no drug by dose interaction. In other words, any effect of dose was equivalent among drugs (drug by dose interaction, F(4,203) = 0.49, NS). Thus, areas were collapsed across dose.

There likely was a significant decrease in area between the old/new waveforms by approximately 25% when recognition was from long-term memory as opposed to working memory (F(1,207) = 3.76; P = 0.054;fig. 5A).‡‡‡As can be seen by the differences in slopes of the lines in figure 5A, there were differences between drugs on the effect of interval, with an interaction of drug by interval (F(4,203) = 2.69; P = 0.03). Propofol (t =–2.70; P = 0.008) and midazolam (t =–2.34; P = 0.02) were different from placebo. In other words, there was a larger decrease in area as recognition interval increased for propofol and midazolam than for other drugs or placebo, reflecting the suggestion of a similar effect present at Pz.

Dexmedetomidine might have decreased the areas between waveforms for familiarity at Cz in a dose-related fashion (fig. 6). There was a suggestion of an effect of drug (F(4,207) = 1.96; P = 0.10); in comparison with placebo, dexmedetomidine did have a smaller area (t =–2.29; P = 0.02).§§§There was no effect of interval on familiarity with dexmedetomidine, whereas weak evidence for this may have been present for recollection (fig. 5B; t =–1.36; P = 0.17 compared with placebo).

In summary, propofol and midazolam decreased areas between old and new recognition ERP waveforms by some 50% when a picture was recognized correctly from long-term memory 27 s after presentation during the CRT. There was no effect on recognitions from working memory. Both drugs affected familiarity and recollection processes used for recognition from long-term memory to a similar extent. There was a suggestion that dexmedetomidine had different effects on familiarity and recollection processes, with familiarity processes impacted more. Dexmedetomidine was more similar to propofol and midazolam in its effect on recollection processes. Propofol shifted both recollection and familiarity ERP waveforms to more negative values, with a dose effect on recollection but not familiarity processes. The effect size for propofol was larger at the Pz than at Cz electrode, and thus the power to detect the effects of increasing propofol dose may have been better at Pz.

As revealed by ERPs, successful recognition of items on the CRT was supported by memory processes that were affected in different ways by the presence of the study drugs. These differences occurred during recognition of pictures from long-term memory, as no drug affected working memory processes. Though processing was slowed by the presence of drug in a dose-related fashion, access to items in working memory continued to be consistently faster than items in long-term memory, and ERP waveforms were preserved at 6-s recognitions. Propofol and midazolam, drugs possessing conscious amnesic properties, impaired long-term memory processes as measured by the reduced area between ERP waveforms 27 s after encoding. As the size of this area relates to memory strength, this indicates that memory was weaker in the presence of propofol and midazolam, even at this early time.15,16Thiopental had no effect on ERPs, despite memory impairment being present at the end of the day. Thus, propofol and midazolam affected memory processes much sooner than previously appreciated. Two previous studies have shown decreased area between old/new waveforms when benzodiazepines were given but at longer time intervals than in the current study.33,34 

Differentiation of drug effects on memory function was revealed more readily by ERP measures than by behavioral responses. No differential memory impairment of propofol in comparison with thiopental was apparent, yet robust decreases in areas between ERP waveforms were present, with ERP potentials being shifted to smaller amplitudes. The presence of conscious amnesia with midazolam was difficult to ascertain by behavioral measures. Recognition from working memory was no better than from long-term memory (no effect of interval; fig. 3C). It could be argued that any decrement in long-term memory on the CRT was a function of a sedation effect on working memory. By these measures, midazolam’s effects on memory could not be differentiated from thiopental. Conscious amnesia from midazolam was only evident in the greater effect of dose on recognition memory as opposed to reaction time. In contrast to these subtle behavioral changes, differences in recognition ERPs from long-term memory were similar to the effects of propofol. Thus, ERP measures documented the influence of propofol and midazolam on memory at drug concentrations just below or at those producing conscious amnesia. ERP changes seemed to presage changes in behavioral performance that would become apparent if the dose were increased or time passed. Though propofol had the same rate of forgetting as thiopental in the current study, there may have been a differential effect present sometime before the end of day recognitions. In a recent imaging study, we showed that the effect of propofol on recognition memory became apparent between 18 and 42 min after encoding.35 

Once begun, the effects of propofol and midazolam on memory continue after discontinuation of the drug after completion of the CRTs. Forgetting of memories encoded in the presence of drug continued at a higher rate than placebo for all drugs. A previous study has shown an increased rate of forgetting for propofol compared with thiopental.4In the current study, forgetting was determined only for pictures proven to be present in memory by correct recognition on the CRT; this is different from the previous study, in which overall recognition rates were measured. A companion manuscript will describe in more detail the memory decay in the period between the CRT and final delayed recognition, where the same memory end-point was reached for all drugs. The current study provides ERP evidence that, for propofol and midazolam, the stage for memory loss is set almost immediately after a memory is formed.

Changed ERP recognition waveforms so soon after encoding raise the question of whether encoding was affected, and this was the reason why memory was lost so readily. One difficulty is to define the exact time point at which encoding occurs.36The current study demonstrated normal functioning of working memory at 6 s. Propofol and midazolam affected memory processes within 30 s, which may point to an effect on encoding into long-term memory. However, it should be noted that ERPs were averaged only from correct responses. Thus changes in ERP waveforms were uncontaminated by responses that were incorrect or missed. In other words, ERPs represent memory processes underlying successful recognition in the presence of drug. In this sense, these memories are encoded as they are available for retrieval, but the memory processes supporting successful recognition were impaired. We have previously shown that encoding is not inhibited to a major degree by propofol over a time period of approximately 12 min, as measured by changes in cerebral blood flow.35There were relatively fewer participants in that study; therefore, the sensitivity to detect effects may not have been as good as the current study. Due to the difficulty in separating encoding from recognition processes,37an alternative approach would be to consider the effect of a drug on memory in terms of its influence on a particular portion of the cascade of processes put into play once an item is in memory, i.e. , consolidation.38–41The effect of propofol or midazolam on some of these yet to be defined processes at an early time point sets the stage for loss rather than retention of a memory.

ERPs are excellent tools for isolating a particular temporal window of drug action on memory. A repeated observation that no drug has been demonstrated to have a retrograde amnesic effect in humans supports the notion that propofol and midazolam inhibit memory processes very early after the acquisition of information, in contrast to observations in animals.42–44In other words, if the critical effect of drug on memory was at a significantly later time in memory consolidation, then memories “in the pipeline” would be lost, as occurs, for example, when hippocampal function is lost due to hypoxia.45,46The period of retrograde amnesia is defined by the time between acquisition of a memory and the critical memory process affected by the drug. If the critical effect is earlier in the consolidation process, then retrograde amnesia would be of smaller magnitude. This seems to be the case in humans receiving midazolam or propofol.

How can the larger effects on ERPs be reconciled with smaller effects on behavior? In other words, why is recognition successful when recognition memory ERPs were impaired? Even though there was a decrease in area between waveforms during recognition, the area was not zero. In other words, there continued to be a difference between old and new waveforms. To critically test this, the difference between these waveforms for propofol and midazolam at the higher dose for recognition from long-term memory was tested (fig. 4D). Indeed, there was a robust difference (F(1,39) = 5.11; P = 0.03; effect size 2.3), with no effect of drug (F(1,39) = 2.87; P = 0.10). Thus, recognition was supported by the remnants of the old/new effect.

Propofol had an additional effect on ERPs from midazolam, producing dose-related shifts of ERP potentials to smaller amplitudes. As the relative dosing of midazolam was greater than propofol, any tendency of midazolam to shift ERP amplitudes was clearly less, as shown in figure 5. Smaller ERP potentials could represent decreased confidence of decisions, decreased strength of memory trace, or less discriminability of items.21One informative way to conceptualize memory-related ERP potentials is to consider recognition processes as resting on a pedestal, the ERP waveform to new items, which is larger or smaller under certain conditions. The memory processes brought into play by neural generators activated by recognition sum their potentials to cause a deflection from this pedestal, i.e. , the old/new effect. Thus, shifts in ERP potentials may be as important as decreases in the displacement between waveforms. For example, subsequent recognition success in the elderly was related to the size of the pedestal at encoding.47Thus, propofol had two different actions on long-term memory processes, whereas midazolam seemed to have one.

Dexmedetomidine, an α-2 adrenoceptor agonist drug, produces a downstream effect on γ-aminobutyric acid receptors, a different mechanism of action from the other study drugs.24This difference seemed to be echoed in the ERPs. Though definitive conclusions cannot be drawn from the current study, the data are suggestive. One problem with the current study design was that multiple drugs were studied simultaneously, which decreased the power to detect true effects of any given drug due to the number of comparisons required. Behavioral data reveal very little effect of dexmedetomidine in comparison with placebo, even at the higher dose (fig. 3). During conduct of the study, however, we noted substantial sedation in these participants. As others have noted, the transition from sedation to functioning well was quite rapid, even in the presence of constant doses of dexmedetomidine.24This seems to be reflected in the behavioral data of the current study. Despite high cognitive functioning, suggestive results indicated that dexmedetomidine affected ERPs differently from the other drugs. The familiarity response for dexmedetomidine may have been preferentially suppressed in comparison with the recollection, as shown by a difference in slopes at Cz and Pz in figure 5and smaller area in figure 6. As mentioned, thiopental did not seem to affect recognition ERPs in any way, even though it produced significant effects on memory at the end of the study day. In other words, the sedation effect of thiopental increased reaction times and the error rate, but normal memory processes were used for recognition of items in long-term memory.

In summary, the effects of propofol and midazolam on long-term memory began within 30 s of encoding. Both drugs impaired familiarity and recollection processes in recognition from long-term memory at this early time. Propofol produced a shift in ERPs to smaller amplitudes, an effect absent for midazolam. Thus, despite the great similarity of conscious amnesic effects and actions on γ-aminobutyric acid receptors, it appears that these drugs affected memory by somewhat different mechanisms. In the past, we have used the term drug-induced amnesia to describe the sedation independent effects of these drugs on memory. However, amnesia is a nebulous term, and we wish to differentiate temporarily induced drug effects on memory from organic amnesias, including postoperative cognitive dysfunction. As pointed out in the introduction, all anesthetic drugs will produce amnesia but, as demonstrated in the current study, via  different mechanisms. The first signs of accelerated memory decay induced by propofol and midazolam were inhibition of recognition ERPs from long-term memory. This effect on ERPs differentiated this mechanism of action from thiopental, which also produced accelerated memory decay compared with placebo. Thus, conscious amnesia produced by propofol and midazolam is characterized by early electrophysiologic changes, which presage accelerated memory decay. Dexmedetomidine may have still different actions on memory, with greater effects on familiarity than recollection processes. Thiopental had no discernable effect on recognition ERPs, even though significant impairment of memory occurred. Thus, to fully understand how anesthetic drugs impair memory, measures of memory function other than behavioral ones are needed. The current study demonstrated the utility of memory-related ERPs to fulfill this goal. In addition, as ERPs seemed to presage behavioral effects, the possibility exists to develop an electroencephalogram-based measure of memory function during light levels of anesthesia.

The authors thank Steven Shafer, M.D., Professor of Anesthesiology, Columbia University, New York, New York, for the use of STANPUMP software (http://anesthesia.stanford.edu/pkpd). The authors also thank Elizabeth Murphy, B.A., Research Study Assistant II, Department of Anesthesiology, Memorial Sloan-Kettering Cancer Center, New York, New York, for help in recruitment and data collection. They also thank Elizabeth Weiner, B.S., Research Study Assistant II, Memorial Sloan Kettering Cancer Center, for invaluable assistance in preparation of this manuscript.