Neurons in the Nonhuman Primate Amygdala and Dorsal Anterior Cingulate Cortex Signal Aversive Memory Formation under Sedation

We aimed to study the effect of anesthetics on stimulus valence, acquisition, and memory in vivo and to identify correlates in the medial prefrontal cortex–amygdala circuit using a nonhuman primate model and clinically relevant doses of anesthetics. The hypothesis posited that aversive valence and memory formation are maintained under anesthetics and that amygdala–medial prefrontal cortex activity is sufficiently resilient to anesthetics during acquisition to allow later retention.

To this end, two behaving nonhuman primates engaged in a classical tone–odor conditioning task. This is a useful translational model allowing for invasive neural recording under anesthetics. The paradigm traces respiratory responses and does not require conscious volition, making it suitable as an implicit measure of learning and memory in both anesthetized and awake conditions. We simultaneously recorded the responses of single neurons in the amygdala and dorsal anterior cingulate cortex and compared the behavioral responses to preceding and simultaneous neural responses.

We defined two a priori primary outcomes, behavioral and neural, to test the hypothesis of memory formation under anesthetics. Data recorded from a conditioning session were classified as suggesting one of two endpoints: (1) learning and memory (i.e., acquisition and retention), which included a statistically significant difference (P < 0.05) between habituation and retention OR a statistically significant difference between habituation and acquisition and no statistically significant difference between acquisition and retention; or (2) no response, which included behavioral and neural results not fulfilling the previous criteria and hence no statistically significant difference between habituation and acquisition and no statistically significant difference between habituation and retention.

No statistical power calculation was conducted before the study. The number of animals is an accepted standard in invasive nonhuman primate neurophysiology research. The design of the study did not allow for blinding of investigators during the experiment. However, recorded data were collected and analyzed only much later using preprogrammed automated software (MATLAB, MathWorks, USA). Therefore, the analyses and findings are blind to the experimental conditions.

We implanted male Macaca fascicularis (weight, 4 to 7 kg; age, 4 to 5 yr) with a recording chamber (30 × 30 mm) above the basolateral amygdalae and dorsal anterior cingulate cortices under deep anesthesia and aseptic conditions. All surgical and experimental procedures were approved and conducted in accordance with the regulations of the Weizmann Institute (Rehovot, Israel) Animal Care and Use Committee, following National Institutes of Health (Bethesda, Maryland) regulations and with American Association for Accreditation of Laboratory Animal Care (Frederick, Maryland) accreditation. Food, water, and enrichments (e.g., fruits and play instruments) were available ad libitum during the whole period, except for the 6 h before a recording session because of the required anesthesia. Sessions took place in the morning and were limited to 4 h for the experiment and required procedures. After completion of the experiments, the animals were discharged to a nonhuman primate rehabilitation center.

To ensure accurate recordings from the target anatomical structures, we performed magnetic resonance imaging scans before, during, and after the recording period using a 3-tesla magnetic resonance imaging scanner (MAGNETOM Trio, Siemens, Germany) with a circular polarized knee coil (Siemens; T1 weighted and three-dimensional gradient-echo pulse sequence with a repetition time of 2,500 ms, inversion time of 1,100 ms, echo time of 3.36 ms, 8° flip angle, and 2 averages). We extracted images in the sagittal plane, with a 192 × 192 matrix and 0.83- or 0.63-mm resolution. We used the first scan before surgery to align and refine anatomical maps for each individual animal (relative location of the amygdala, dorsal anterior cingulate cortex, and anatomical markers such as the interaural line and the anterior commissure). This scan guided the positioning of the chamber on the skull at surgery. After surgery, we performed another scan with two electrodes directed toward the amygdala and another two at the dorsal anterior cingulate cortex. Three observers reviewed the scan separately to inspect the images and calculate the amygdala anterior–posterior and lateral–medial borders relative to each of the electrode penetrations and the location at the dorsal anterior cingulate cortex. We calculated the depth of the two structures from the dura surface based on the magnetic resonance imaging at all penetration points. Clear anatomical markers and visual similarity were used to identify these structures based on magnetic resonance imaging images from a primate atlas.²⁵ 

The conditioned response previously described in this paradigm consists of an augmented respiratory response. Hence, we recorded a continuous airway pressure trace throughout the paradigm. We extracted and quantified the volume of each inspiration, peak pressure, and rise time.

Detailed descriptions of the odor delivery system (olfactometer) have been previously reported.²⁶  In brief, three hoses attach to a silicon nasal mask placed over the monkey’s nose. The first hose delivers air into the mask at a constant flow. When stimuli are commanded, silent vacuum solenoids divert away the clean air and allow odorized air into the mask. Importantly, we delivered the odorized air at the same pressure and flow as the clean air and commanded it from outside the room to guarantee that the monkey received no cues regarding odor delivery. The aversive odor is discharged for 1 s. The second hose evacuates air from the mask at an equal flow to that delivered into the mask, and quickly removes odors right after their release while maintaining stable pressure within the mask. The third hose is connected to two pressure sensors with different sensitivity ranges (0.25- and 1-inch H₂O pressure range; All Sensors, USA) that allow measurement of respiratory behavior with minimal time lag. To load air with odor, filtered air is flowed through a Teflon odor canister. We used real-time detection of spontaneous inhalation onsets to trigger tones and odors.

The recording chamber allowed simultaneous recording from the amygdala and dorsal anterior cingulate cortex of both hemispheres. During each recording session, we lowered four to eight microelectrodes, up to two at each structure (0.6 to 1.2 M glass/narylene-coated tungsten microelectrode [Alpha Omega Technologies, USA, or FHC, USA]). The monkey’s head was fixated, and we lowered the electrodes into the brain inside a metal guide (gauge 25xxtw, Cadence Design Systems, USA) using a head-tower and electrode positioning system (Alpha Omega Technologies). The guide penetrated and crossed the dura and stopped approximately 0.5 mm in the cortex. We then moved the electrodes independently further into the amygdala and dorsal anterior cingulate cortex, respectively (mapping sessions in each animal were performed moving slowly and identifying electrophysiologic markers of firing properties tracking the known anatomical pathway into the target structures). Electrode signals were preamplified, 0.3- to 6-KHz band pass–filtered, and sampled at 25 KHz (Alpha Lab Pro, Alpha Omega Technologies). We allowed 15 min for the tissue and signal to stabilize before starting the behavioral protocol. At the end of the recording period, we performed offline spike sorting for all sessions to improve neuron isolation (offline sorter, Plexon Inc., USA). We synchronized and recorded the behavioral paradigm and all variables using MATLAB software and Alpha Omega Technologies analog and digital recorders.

We used the Ramsay score²⁷  to assess the depth of anesthesia during the study sessions. The scale ranges from 1 to 5 and corresponds to the baseline state of the subject and the response to discrete stimuli: 1, agitated or restless or both; 2, alert and tranquil; 3, brisk response to a light glabellar tap or loud auditory stimulus; 4, sluggish response to a light glabellar tap or loud auditory stimulus; and 5, no response to a light glabellar tap or loud auditory stimulus.

After the injection of an anesthetic, we allowed 5 min for induction and clinical effect to take place. An investigator (human sedation and anesthesia experts, N.S. and E.K.) then briefly entered the recording chamber and scored the depth of sedation. Sedation scoring was not blinded but was performed before the beginning of the conditioning paradigm. However, the automated data collection and analysis were applied to all conditions equally and were therefore “blind” (unbiased) to all assessments of dose-related behavior and neural results.²⁸ 

After administration of a discrete drug dose, the monkeys engaged a classical conditioning task, a learning paradigm of tone–odor conditioning. The conditioned response described in this paradigm is increased breath effort in response to the auditory conditioned stimulus, which is locked to a breath onset in preparation for the aversive odor discharge, the unconditioned stimulus timed to the following breath onset. The unconditioned response is decreased breath volume in response to the noxious odor. We performed the experiment and recorded variables while the monkeys sat in a customized chair, in a dark, acoustically isolated room. We placed a mask over the monkey’s nose for respiratory measurements and odor delivery.

Experimental sessions consisted of habituation to tones, injection of anesthetics, and the acquisition of tone–odor associations followed by testing retention of learned associations. During habituation, an auditory conditioned stimulus was generated randomly at the range of 500 to 5,000 Hz. Conditioned stimuli differed from the conditioned stimuli of the previous session by 500 Hz and were not repeated in additional sessions during the following 2 weeks. Tones were pure sinus waves of 1,000-ms duration with 5-ms onset and offset ramps, generated with a standard computer and delivered through a speaker (Adam 5 studio monitor, ADAM Audio, Germany) located 40 cm behind and to the center of the animal at 75 dB. During habituation, the daily conditioned stimulus was presented 10 times (intertrial interval was approximately 60 s).

After habituation, we delivered anesthetics by intramuscular injection. We used ketamine, an NMDA receptor antagonist, and midazolam, a benzodiazepine GABA coagonist. We used four different doses of each drug (1, 2, 4, and 8 mg/kg, and 0.1, 0.2, 0.4, and 0.8 mg/kg for ketamine and midazolam, respectively), as well as a normal saline control condition. The dose range aimed to include light to deep sedative/anesthetic states according to estimates from human and primate studies.^29–31  In addition to existing data, we also evaluated the drug dose/response effects on three monkeys according to the Ramsay sedation scale.²⁷ 

During the experiment period, we randomized the order of testing sessions for drug and dose to avoid the potential bias of a consistent increase or decrease of doses or desensitization to a given agent throughout the study period. After injection, we allowed 5 min for induction and clinical effect to take place.

Acquisition followed by presenting 12 conditioned stimulus–unconditioned stimulus pairs randomly intermingled with five unpaired conditioned stimulus presentations (intertrial interval was approximately 60 s). For the aversive odor, we chose a 1:20 solution of propionic acid (Sigma-Aldrich, USA) diluted in mineral oil, a highly aversive agent to monkeys and humans that triggers both olfactory and trigeminal receptors. This specific solution has been previously tested in this behavioral paradigm.²⁴ 

We tested retention 45 min from the time of injection, after recovery, with 10 presentations of the conditioned stimulus (intertrial interval was approximately 60 s). This interval was chosen after preexperiment trials revealed it was the maximal time required for return to baseline behavior (Ramsay score²⁷  of 1 to 2) and is in line with similar reports.³¹ 

In this work, we applied anesthetics to a previously published experimental design and analysis paradigm.^17,24,26,32  Hence, we were able to preplan the analysis of data. We used histograms to verify the normality of distributions and similarity of variances.

We measured breath augmentation by quantifying peak inspiratory pressures, time to peak pressures (the inverse being inspiratory velocity), and inspiratory volumes during the first 40% of the respiratory cycle (the average inspiratory phase). These measures are commonly used to assess breath dynamics and responses.^26,32  Because anesthetics may significantly modulate respiration, we corrected the respiratory variables to prevent direct drug effects from confounding results. To correct for effects on the absolute inspiratory pressures, the pressure traces were z-scored using the prestimuli, intertrial interval as a reference baseline. To correct for possible direct drug effects on the respiratory rate, time points along the respiratory cycle were normalized to cycle length and reported as a fraction of the respiratory cycle. We evaluated differences in response to stimuli across habituation, acquisition, and retention according to peak inspiratory pressures and inspiratory velocity³²  using multivariate ANOVA for each session to enable independent evaluation of each animal, day, and session. We used two-tailed testing for all analysis.

Finally, we performed ANOVA and post hoc sensitivity analyses separately for behavioral responses and for neuronal responses and across the relevant factors (drug dose and behavioral phase) and using the entire set of experimental sessions. Namely, this was done independently of the a priori classification to memory positive/negative sessions as described above.

As a measure of anesthetic effect on respiratory rate, we sampled the prestimuli respiratory rates during the acquisition phase and compared them to the prestimuli respiratory rates during the habituation phase before anesthetics using the paired t test.

For each trial, we calculated the response to stimuli as the absolute difference in pre- and poststimulus firing rates divided by their sum,

where TR is the trial response, FR is the firing rate, Pre refers to prestimulus, and Post refers to poststimulus.

We chose a time window of 500 ms before and after stimuli. This is in line with previously described epochs of neural responses to fearful stimuli.³³  When sorting, there is an inherent tradeoff between sensitivity (in this case utilizing all “true” neurons) and specificity (in this case using only “true” neurons). We explicitly favored specificity. Previous works used a cutoff of 1 Hz average firing rates.³⁴  Given that anesthetics, especially midazolam, may (and did; see fig. 1C) lower baseline firing rates, we chose a cutoff of 0.5 Hz. We included only neurons that maintained a median firing rate of more than 0.5 Hz throughout the paradigm, during habituation, acquisition, and retention.

In a similar fashion to the behavioral analysis, for each cell, differences in response to stimuli across habituation, acquisition, and retention were evaluated using one-way ANOVA. We calculated the neuron strength of acquisition and retention as the mean response during the respective phases across trials.

To determine the anesthetic effect on neuronal firing rate, we sampled the prestimuli firing rates during the acquisition phase and compared them to the prestimuli firing rates during the habituation phase before anesthetics using the paired t test.

We calculated the behavioral and neuronal strengths of acquisition and retention as their mean responses during the respective phases across trials. We used the Pearson correlation to test their association.

We evaluated differences between results in multiple subgroups (e.g., doses, drugs, etc.) using one-way ANOVA. When comparing two discrete groups (e.g., outcome A vs. outcome B), we used the independent two-sample t test. We considered P values less than 0.05 statistically significant.

We injected two M. fascicularis monkeys (monkey S had 26 sessions, and monkey L had 42 sessions) with low (1 to 2 mg/kg ketamine or 0.1 to 0.2 mg/kg midazolam) or high (4 to 8 mg/kg ketamine or 0.4 to 0.8 mg/kg midazolam) doses of anesthetics. After injections, the monkeys underwent tone–odor (conditioned stimulus–unconditioned stimulus) aversive conditioning. We used the respiratory pattern to detect successful conditioning (conditioned response) as shown in previous studies.^24,26 

The Ramsay sedation scale²⁷  was measured 5 min after injection and showed a dose-dependent effect of drug on behavioral parameters (low vs. high dose mean ± SD, 3 ± 1, 4 ± 1, respectively, one-way ANOVA between drug conditions, degrees of freedom = 4, F = 23.4, P < 0.001; fig. 1A). Midazolam and ketamine produced similar effects on the depth of sedation as measured by the modified Ramsay sedation scale (mean ± SD, 4 ± 1, 4 ± 1, respectively, post hoc test, P = 0.952; fig. 1A). As expected, respiratory rates under anesthetics decreased (fig. 1B), absolute firing rate under midazolam slightly decreased, and absolute firing rates under ketamine slightly increased (fig. 1C).

In total, 68 sessions entered analysis, five sessions were aborted because of technical and veterinary issues or corrupt data, 119 amygdala and 132 dorsal anterior cingulate cortex neurons (150 neurons from monkey S and 101 neurons from monkey L) met inclusion criteria, and 39 amygdala and 42 dorsal anterior cingulate cortex neurons were excluded based on independent criteria before any further analysis was performed (see “Materials and Methods”).

The respiratory pattern measured by pressure sensors was used to identify conditioning: a change after the conditioned stimulus is indicative of successful conditioning (namely, a conditioned response). Each daily session began with habituation to a novel tone conditioned stimulus followed by drug injection. We allowed 5 min after injection for the drug effect to reach a stable sedative/anesthetic plane and assessed the depth of sedation using a modified Ramsay score (see “Materials and Methods”). Anesthetic induction was followed by an acquisition phase in which conditioned stimulus–unconditioned stimulus pairs were presented 12 times. The animals were then allowed to naturally recover from the drug, and once they were fully alert, but not before 45 min had passed, we presented the conditioned stimulus 10 times to test retention of the learned association (fig. 1D).

We found that conditioned responses during retention existed in 44% of all sessions under anesthetics, suggesting memory formation under both types of anesthetics and in all doses. Aversive memory formation occurred in 26 of 59 sessions under anesthetics (16 of 29 and 10 of 30, 5 of 30 and 21 of 29 for midazolam and ketamine, low and high doses, respectively; multivariate ANOVA; see examples and cumulative incidence in fig. 2, A and C). This proportion was not different from retention observed in control sessions after injection of normal saline (44.4%, chi-square test, P = 0.983, n = 9).

Anesthetics may impair motor responses, and hence acquisition of conditioned responses is usually examined just after recovery.²³  In the current case, the use of a tone–odor paradigm allowed us to overcome this challenge by measuring breathing responses, and so we turned to examine the behavioral dynamics of learning under anesthetics. We observed a gradual development of the conditioned response throughout acquisition and then a gradual decrease during retention, likely because of the unpaired conditioned stimulus presentations (namely, an extinction-like process).³⁵  Sessions that did not culminate in memory formation also displayed a gradual (albeit smaller) acquisition response. The behavioral response to the conditioned stimulus during acquisition under anesthetics was not statistically different between sessions with or without subsequent retention (independent two-sample t test, P = 0.253), suggesting that behavioral responses under anesthetics may not accurately predict later retention and memory formation (fig. 2B).

In sessions with subsequent retention (n = 26), the size of the conditioned response during acquisition was positively correlated with the size of the conditioned response during retention²⁴  (Pearson correlation, r = 0.62, P < 0.001). This was not the case in sessions when retention was not evident (n = 33, r = −0.06, P = 0.735; fig. 2D) or between habituation and retention (n = 26, r = 0.28 P = 0.156), suggesting an effect of learning. The correlation between the strength of acquisition under anesthetics and the strength of the subsequent retention provides additional support for the link between the two and further suggests that memory strength is also maintained.³⁵ 

We further compared behavioral responses across the entire set of experimental sessions and independent of the a priori behavioral classification to memory-positive or -negative sessions. To do so, we compared respiratory behavioral responses (normalized inspiratory velocities) across phase and drug dose and found both factors to be significant (two-way ANOVA for factors drug dose × phase, mean ± SD: high-dose ketamine, 0.111 ± 0.013; low-dose ketamine, 0.085 ± 0.012; high-dose midazolam, 0.147 ± 0.013; low-dose midazolam, 0.101 ± 0.013; saline, 0.075 ± 0.017; and for phase: habituation, 0.0791 ± 0.011; acquisition, 0.123 ± 0.011; and retention, 0.110 ± 0.011; F[4,188] = 3.919, F[2,188] = 4.286, F[8,188] = 0.758; and P = 0.004, 0.015, and 0.639 for drug dose, phase, and interaction, respectively). In addition, comparing the respiratory response across drug conditions did not yield a significant difference (one-way ANOVA, mean ± SD: saline, 1.501 ± 0.168; low-dose ketamine, 1.462 ± 0.126; low-dose midazolam, 1.497 ± 0.135; high-dose ketamine, 1.284 ± 0.135; high-dose midazolam, 1.663 ± 0.130; F[4,63] = 1.032; P = 0.400).

We then compared behavioral responses for high versus low dose across drug choice and did not find any significant interaction or main effect (fig. 2E; two-way ANOVA, drug × dose, mean ± SD: ketamine, 1.373 ± 0.097; midazolam, 1.580 ± 0.098; high doses, 1.474 ± 0.098; low doses, 1.479 ± 0.097; F[1,55] = 2.237, 0.002, 1.554 and P = 0.140, 0.966, 0.218 for drug, dose, and interaction, respectively).

In summary, we found behavioral evidence of learning and aversive memory formation under anesthetics with both types of drugs from minimal sedation to deep anesthetic–sedative states. Although ketamine and midazolam modulate different systems (excitatory vs. inhibitory), we observed retention after recovery with both drugs.

To investigate the neural dynamics of memory formation under sedation, we recorded the activity of single neurons in the amygdala and dorsal anterior cingulate cortex simultaneously (n = 101 and 121, respectively). Responses of single neurons were measured as the absolute difference in firing rates between before and after stimuli and divided by their sum (response index). We found that responses during acquisition were higher in sessions with behavioral conditioned responses, namely, sessions exhibiting memory formation under anesthetics (sessions with conditioned responses vs. sessions with no conditioned responses, independent two-sample t test: n = 101, P = 0.021 for the amygdala; and n = 121, P = 0.012 for dorsal anterior cingulate cortex). Dorsal anterior cingulate cortex neurons, but not amygdala neurons, exhibited this difference also during retention (independent two-sample t test: P = 0.003 for the dorsal anterior cingulate cortex and P = 0.096 for the amygdala; fig. 3A).

To explore correlates of learning and memory in single cells, we divided the responses into two possible outcomes, similar to the behavioral data: no change in responses to the conditioned stimulus or a development of statistically significant response during acquisition that extended into retention (fig. 3B). Similar to behavior, both these outcomes were present under anesthetics with both drugs, in all doses, and across the sedation–anesthesia continuum (one-way ANOVA over habituation, acquisition, and retention). Overall, 21% of amygdala neurons and 21% of dorsal anterior cingulate cortex neurons showed changes in firing rates during acquisition under anesthetics and the following retention (binomial test, P_amygdala < 0.001 and P_{dorsal anterior cingulate cortex} = 0.006; see example and cumulative incidence in fig. 3, B and C).

To further test whether acquisition responses extend to the retention after recovery from anesthesia, we tested the correlation between neural responses in acquisition and retention. Single-neuron responses in the amygdala and dorsal anterior cingulate cortex were positively correlated between acquisition and retention (amygdala, n = 101, Pearson r = 0.51, P < 0.001; dorsal anterior cingulate cortex, n = 121, Pearson r = 0.32, P < 0.001; fig. 3D). This correlation does not seem to stem from a baseline response to the conditioned stimulus, because the correlation between habituation and retention responses to the conditioned stimulus was much weaker (amygdala, r = 0.16, P = 0.103; dorsal anterior cingulate cortex, r = 0.19, P = 0.035) and increased in the amygdala during conditioning (Fisher transformation: amygdala, P = 0.005; dorsal anterior cingulate cortex, P = 0.280), suggesting an effect of learning.

Neural responses during acquisition were correlated with behavioral retention responses, but only in the amygdala (amygdala, n = 101, Pearson r = 0.22, P = 0.026; dorsal anterior cingulate cortex, n = 121, r = −0.13, P = 0.154; fig. 3E). This result provides important evidence for the link between amygdala activity during acquisition under anesthetics and the memory as tested during awake retention.

Because amygdala and dorsal anterior cingulate cortex synchrony in response to the conditioned stimulus has been implicated in aversive memory formation in awake conditions,²⁴  we turned to look at the signal correlation between all possible pairs of neurons within the amygdala–dorsal anterior cingulate cortex circuit under anesthetics (excluding pairs recorded on the same electrode). For each pair, we measured the correlation of trial responses (i.e., the firing rate change induced by the stimuli; see “Materials and Methods”). Indeed, during acquisition, we found correlated activity in dorsal anterior cingulate cortex–dorsal anterior cingulate cortex pairs (n = 142, Pearson r = 0.3, P < 0.001), amygdala–amygdala pairs (n = 81, Pearson r = 0.25, P = 0.023), and amygdala–dorsal anterior cingulate cortex pairs (n = 221, Pearson r = 0.26, P < 0.001; fig. 3F).

Finally, we also compared neural responses across the entire set of experimental sessions and independent of the a priori behavioral classification to memory-positive or -negative sessions. To do so, we compared neural responses from all trials across phase and drug dose. Trial responses of amygdala neurons were significant for the phase and the interaction but not for the drug dose (mean ± SD: drug dose conditions: high-dose ketamine, 0.474 ± 0.022; low-dose ketamine, 0.475 ± 0.020; high-dose midazolam, 0.535 ± 0.027; low-dose midazolam, 0.473 ± 0.021; saline, 0.537 ± 0.026; and phase conditions: habituation, 0.450 ± 0.018; acquisition, 0.545 ± 0.018; retention, 0.502 ± 0.018; F[4,342] = 1.822, F[2,342] = 6.479, F[8,342] = 3.815; and P = 0.124, 0.001, 0.001 for drug dose, phase, and interaction).

Responses of dorsal anterior cingulate cortex neurons were significantly different for the interaction and drug dose but not for the phase (mean ± SD: drug dose conditions: high-dose ketamine, 0.475 ± 0.020; low-dose ketamine, 0.464 ± 0.020; high-dose midazolam, 0.537 ± 0.018; low-dose midazolam, 0.520 ± 0.025; saline, 0.532 ± 0.034; and phase conditions: habituation, 0.489 ± 0.019; acquisition, 0.516 ± 0.019; retention, 0.511 ± 0.019; F[4,342] = 2.508, F[2,342] = 0.576, F[8,342] = 3.064 and P = 0.041, 0.562, 0.002 for drug dose, phase, and interaction).

In summary, we found a population of single neurons in both regions exhibiting progressive changes in response to the conditioned stimulus and corresponding behavioral evidence of memory formation. Amygdala and dorsal anterior cingulate cortex responses were elevated during acquisition, when memory was successfully formed under anesthetics, and synchrony between neuron pairs within the amygdala–dorsal anterior cingulate cortex circuit was maintained.

Anesthetics may affect learning and memory by interfering with the acquisition process itself and/or by attenuating the valence of the aversive stimulus. To distinguish between these two options, we tested whether the anesthetics modulated the response to the unconditioned stimulus (aversive odor) and whether this modulation affected memory formation.

The stereotypical unconditioned response to this aversive odor paradigm is a reduction of inhale volume once the odor is encountered (see cumulative incidence and examples in fig. 4, A and B). Importantly, this response did not differ between the different anesthetic conditions (agents and doses, one-way ANOVA, P = 0.137; fig. 4D, top left), and yet we found that inhale volumes in response to the aversive odor were 9.5% lower in sessions with no retention when compared to sessions with successful retention (independent two-sample t test, P = 0.010; fig. 4D, top right). Nevertheless, this was not the case in high-dose (deeper sedation) sessions (namely, no difference in unconditioned response between sessions with retention and sessions without retention), suggesting that under deep sedation, the behavioral response fails to predict subsequent memory. This can result from differences in perception and/or different degrees of arousal.

Notably, we did not observe any habituation of the unconditioned response under anesthetics, and in contrast, there was an escalating trend toward the end of acquisition in sessions with retention (paired t test, P = 0.0803). Moreover, responses at the end of acquisition were more robust in sessions with successful retention than those without successful retention (independent two-sample t test, P = 0.008). This dynamic may stem from unconditioned stimulus facilitation³⁶  that in turn leads to stronger association and more reliable memory formation.

We evaluated single neuron responses to the aversive odor (the unconditioned stimulus; see examples in fig. 4C). Neural responses to the unconditioned stimulus did not differ between anesthetics and saline controls (awake vs. anesthetics, P = 0.118 and P = 0.236 for the amygdala and dorsal anterior cingulate cortex, respectively). However, neural responses to the unconditioned stimulus under midazolam were higher than under ketamine (independent two-sample t test, P < 0.001 and P < 0.001 for the amygdala and dorsal anterior cingulate cortex, respectively). As expected, neural responses to the unconditioned stimulus were more robust in sessions culminating in successful retention than those that did not (independent two-sample t test, P < 0.001 and P = 0.002 for the amygdala and dorsal anterior cingulate cortex, respectively; fig. 4C). To conclude, we found evidence in both the behavioral response and the neural response that the aversive valence is preserved under anesthetics in both agents and contributes to memory formation.

In this study, we demonstrated implicit aversive memory formation occurring under different anesthetic states in nonhuman primates using two commonly used agents that leverage two distinct mechanisms: GABA and NMDA transmission. We also noticed a maintained representation of aversive valence despite anesthetics administration, suggesting that anesthetics directly affect memory formation.

We recorded changes in neural responses in the amygdala and dorsal anterior cingulate cortex that varied by the different behavioral outcomes and correlated with aversive stimulus association and successful memory formation. The large proportion of neurons showing these changes supports the assumption that this is indeed a manifestation of memory formation.

NMDA and GABA communication have an essential and extensive role in memory formation, including in the dorsal anterior cingulate cortex and amygdala. Targeting these two structures and these two mechanisms offered a natural hypothesis.

A large body of rodent studies has suggested that anesthetics negatively affect acquisition and retention of learned associations.²³  However, a number of studies did suggest simple associative learning to be possible under anesthetics.^37,38  Indeed, we found that although retention was successful in approximately 44% of sessions and occurred at all anesthetic depths, it was not uniform. This allowed us to compare different aspects of successful versus aborted memory formation under anesthetics.

We found the trajectories of learning and memory under anesthetics to be similar to those observed in awake animals. We found incremental acquisition slopes under anesthetics followed by decrement responses of extinction once awake and found the behavioral response during acquisition under anesthetics correlated with the strength of retention after it. This suggests that learning and memory under anesthetics follow similar rules to pharmacologically naïve conditions and that the function of structures and circuits serving these processes remains conserved despite the presence of anesthetics.

The amygdala is considered sufficient for encoding simple aversive associations.³⁹  It gathers and associates diverse sensory inputs.¹⁶  This suggests that regardless of global neural deficits induced by anesthetics and more specifically hippocampal deficits,¹⁵  amygdala function under anesthetics may suffice for associative memory formation. Previous studies have shown that shielding the amygdala from anesthetics by local injection of antagonists enables acquisition under anesthetics.⁴⁰ 

The medial prefrontal cortex, more specifically the dorsal anterior cingulate cortex, forms a tight circuit with the amygdala and is thought to appraise and regulate its acquired inputs and associations.²⁰  Although there are several studies of the amygdala, only a few studies have focused on medial prefrontal cortex function under anesthetics.⁴¹  These were inconclusive and demonstrated cases in which the activity decreased,⁴²  whereas in other cases it was maintained.⁴¹ 

Amygdala and dorsal anterior cingulate cortex responses to the conditioned stimulus during acquisition correlated with their response to the conditioned stimulus during retention testing after recovery. Furthermore, amygdala acquisition responses under anesthetics correlated with behavior after recovery. Elevated and incremental amygdala and dorsal anterior cingulate cortex responses to the conditioned stimulus during acquisition under anesthetics correlated with the behavioral trajectory and heralded successful memory retention after recovery. This suggests that maintained amygdala and dorsal anterior cingulate cortex function is both possible and necessary for acquisition under anesthetics and that their activity under anesthetics may serve to predict future memory. It is noteworthy that unlike amygdala neurons, which only responded preferentially during acquisition, in retention-positive sessions, dorsal anterior cingulate cortex neurons showed elevated responses to the conditioned stimulus during retention testing after recovery as well, which may hint at an extended role for the dorsal anterior cingulate cortex in consolidation and retrieval.

When assessing simultaneous activity in pairs of neurons, a correlation in interamygdala, interdorsal anterior cingulate cortex, and amygdala–dorsal anterior cingulate cortex responses emerged. This is in line with previous findings of the role of amygdala–dorsal anterior cingulate cortex synchrony in aversive learning²⁴  and suggests a functioning dorsal anterior cingulate cortex–amygdala circuit under anesthetics.

When anesthetics affect memory formation, a seminal question is whether the effect stems from a change in the emotional state, a diminished integration of aversive stimuli, or a diminished integration and association of the environment presented. These options are not mutually exclusive. Although it is well accepted that primary representations of stimuli persist under anesthetics,⁴³  the level of stimuli integration remains an open question.⁴⁴  Eloquently stated, stimuli are often assumed to be “received but not perceived.”⁴⁵  Our results suggest that under the continuum of sedation and anesthesia, stimuli not only reach primary cortices but also go further upstream and are integrated by secondary association cortices and nuclei.

An accurate attribution of salience by amygdala neurons to the aversive stimulus is required to drive learning and memory and for the transfer of salience to the conditioned stimulus.⁴⁶  Studies that assess the direct effect of anesthetics on aversive valence are relatively few and often contradictory.^47,48  We chose an olfactory stimulus, a sensory modality well suited for dissecting both valence and intensity.⁴⁹  The aversive nature of the chosen odor is based on previous studies^17,24,26  in awake animals. Although we did not compare aversive and rewarding (e.g., appetitive) stimuli, these results suggest this known aversive stimulus maintains its salience under anesthetics.

This study has several limitations. Although deep sedation was achieved, both the use of a single agent in each session as well as the maximal doses used (especially for ketamine) may limit generalization to general anesthesia, which almost always utilizes multiple agents simultaneously and often at higher equivalent doses. We chose ketamine and midazolam for their different mechanisms and their ubiquitous clinical use in a variety of settings, yet the results may not apply to other agents.

Additionally, we found a relatively high incidence of memory formation, especially when compared to the extremely low incidence of explicit memory in human studies, and we therefore assume that the memory observed here is implicit. This is also likely given the nature of the stimuli and responses, olfaction and respiration. Nevertheless, this study is not suitable, nor was it intended, to probe explicit/episodic memory, and we therefore cannot rule out independent or overlapping explicit memory formation.

It is also worth noting that olfaction holds some unique neuronal properties. It does not require thalamic relay and utilizes cranial nerve (olfactory and trigeminal) transmission. Here, we chose an aversive odor paradigm for two main reasons: first, it allowed us to explore an aversive experience under anesthetics that is not strictly nociceptive; and second, it enabled us an implicit direct measure of learning and memory across all sedation states by using respiratory responses.

Finally, we found similar proportions of sessions with memory, both when comparing proportions directly between anesthesia and saline control and when comparing the behavioral responses of retention themselves. Although this is an interesting finding, it is a null result (not a statistically significant difference), and we therefore cannot conclude that memory is indeed similarly formed in these different conditions. Here, we only used this separation to test for the neural correlates of this memory. Further studies with adequate statistical power are required to test the relationship between implicit memory occurrence and anesthetic drug and/or dose.

These results suggest that implicit memory formation under anesthetics is likely in a clinical setting. Intact aversive valence precedes implicit memory formation, as reflected in behavioral responses and more robustly in neural responses to the unconditioned stimulus, suggesting that it is sufficient to drive memory formation. The patterns we observed are similar to those found in conditioning studies in awake animals, suggesting that implicit aversive memory may be resilient to anesthetics. A major strength of this study lies in pairing behavior and invasive electrophysiologic recordings of a nonhuman primate brain under increasing depths of sedation by commonly used anesthetics, mechanisms, and doses.⁵⁰  This improves the translatability and generalizability of these results and may help to bridge the gap between the methodologic and ethical limitations of human studies and the limitations of rodent studies caveated by evolutionary distance.

In summary, these findings suggest that under various degrees of sedation and during multiple anesthetic states, implicit aversive memory formation and the integration of stimuli persist. Acquisition and retention of aversive information seem to follow similar rules and engage the same structures and mechanisms as those described in awake animals. Moreover, activity patterns in the amygdala–dorsal anterior cingulate cortex circuit under anesthetics predict these phenomena.

The authors thank Mr. Yosef Shohat for animal care and behavioral consultation; Dr. Yoav Kfir, Dr. Rita Perets, Dr. Yarden Cohen, and Dr. Raviv Pryluk for scientific and technical advice; and Dr. Edna Furman-Haran and Ms. Fanny Attar for magnetic resonance imaging procedures (all members of the Department of Neurobiology, Weizmann Institute of Science, Rehovot, Israel). Dr. Samuel thanks Prof. Karl Skorecki (Faculty of Medicine, Bar Ilan University, Safed, Israel) for unwavering support and wise advice.

Supported by Israel Science Foundation (Jerusalem, Israel) grant No. 2352/19 and ERC-2016-CoG (Brussels, Belgium) grant No. 724910 (to Dr. Paz) and a grant from the United States–Israel Binational Science Foundation (Jerusalem, Israel; to Dr. Raz).

Dr. Raz serves as a consultant to Medtronic (Dublin, Ireland). The other authors declare no competing interests.