Effects of ketamine on brain function during metacognition of episodic memory

Metacognition — the capacity to reflect on and monitor one's own cognitive states — is a core aspect of conscious processing and is commonly measured by trial-by-trial confidence judgments about first-order perceptual or memory decisions. While signal-detection-theory approaches such as the meta-d' framework allow quantification of metacognitive sensitivity (meta-d') independently of response bias and primary-task performance (d'), very little is known about the neurochemical substrates that support metacognitive processes. Previous pharmacological work has implicated monoamines in metacognition, and antagonism of the NMDA glutamate receptor with compounds such as ketamine produces robust alterations of conscious experience and impairs episodic memory; however, the specific role of glutamatergic transmission in metacognition has not been directly tested in humans. Lehmann and colleagues addressed this gap in a double-blind, placebo-controlled fMRI challenge study. They tested whether a subanesthetic, psychotomimetic ketamine infusion (target plasma level 100 ng/ml) affects metacognitive sensitivity, metacognitive bias, and their neural correlates during an episodic memory task that manipulated depth of processing (deep vs shallow encoding). Confidence ratings and BOLD activity were collected during two study phases designed to dissociate effects of ketamine on encoding versus retrieval, and subjective altered states were quantified with the 5D-ASC questionnaire. The primary hypothesis was that ketamine would alter metacognitive sensitivity and associated brain activation during metacognitive confidence ratings in the episodic memory paradigm.

Participants were 53 healthy, non-smoking, right-handed young adults (age mean 23.47, SD 3.24; 29 female) screened for psychiatric, neurological, medical and pregnancy contraindications. Exclusion criteria included prior ketamine experience, psychiatric or neurological disorder, positive drug test, and extreme BMI. Ethical approval was obtained and behavioural data and materials were made available as supplementary materials or on request. A between-subjects, double-blind, randomised placebo-controlled design was used. Twenty-four participants received racemic ketamine by bolus plus continuous infusion to maintain a target plasma level of 100 ng/ml; 29 received saline. An unblinded anaesthesiologist prepared the infusion and monitored physiological parameters. The study comprised two study–test phases. In Study Phase I, participants performed an encoding task outside the scanner before infusion, with a recognition retrieval test inside the scanner during infusion. In Study Phase II, a second encoding task was performed in the scanner during continued infusion; retrieval for those items took place roughly 60 minutes after the infusion ended and participants had left the scanner. The 5D-ASC questionnaire was completed immediately after leaving the scanner. The episodic memory task manipulated level of processing (LoP): half of items were encoded deeply (pleasantness judgement) and half shallowly (syllable parity). Study Phase I used 120 encoded words (60 deep, 60 shallow) and a recognition retrieval run of 180 items (120 old, 60 new) during infusion. During retrieval participants made Type 1 old/new decisions and on two-thirds of trials provided Type 2 confidence ratings on a 6-point scale ('Report' trials); the remaining trials required moving a cursor to a highlighted number as a non-metacognitive control ('Follow' trials). Study Phase II comprised encoding of 100 novel words (50 deep/50 shallow) during infusion and a later retrieval test (150 items: 100 old, 50 new) without a Follow condition so confidence was reported on every trial. Behavioural analyses used SDT-based measures: Type 1 sensitivity (d'), metacognitive sensitivity (meta-d') computed via a maximum-likelihood implementation of Maniscalco and Lau's type2 SDT code, and metacognitive efficiency (meta-d'/d'). Metacognitive bias (mean confidence minus performance) was also examined. A hierarchical Bayesian extension (HMeta-d) estimated group-level log(meta-d'/d') and highest-density intervals (HDIs) to assess group differences in efficiency. Correlations between 5D-ASC and task measures were explored with Bonferroni correction. Standard frequentist tests (t-tests, mixed ANOVAs) assessed LoP and drug effects; assumption checks were performed as reported. fMRI was acquired on a 1.5 T scanner. Preprocessing included realignment, coregistration, segmentation, normalization to MNI space, resampling to 2×2×2 mm and spatial smoothing (8 mm FWHM). First-level models modelled stimulus onsets and set durations to reaction times for Type 1 responses and to the time from scale onset to first movement for Type 2 Report trials; realignment parameters were covariates. Because incorrect trials were too few for many participants, incorrect responses were pooled into a residual regressor, deviating from the preregistration. Regressors included Deep, Shallow, New and Incorrect for Type 1, each with separate Report and Follow second-order regressors (total eight second-order regressors) and an exploratory parametric modulation of Report trials by confidence. Second-level full-factorial models tested Drug (between-subject) × Word Type (within) and Rating Type (Report/Follow) effects. Whole-brain inference used a voxelwise height threshold P < 0.001 and cluster-level family-wise error correction at P < 0.05. Four participants' Study I and three Study II fMRI datasets were excluded for failed normalization, leaving 49 participants (23 ketamine, 26 placebo) for Study I fMRI and 50 (23 ketamine, 27 placebo) for Study II. Behavioural analyses included all 53 participants.

Subjective effects: Ketamine produced robust alterations of conscious experience. Participants who received ketamine scored higher on the 5D-ASC global index and on all individual scales (5D-ASC global: t(23.7)=4.69, P < 0.001, d=1.35), confirming a pronounced psychotomimetic effect of the infusion. Study Phase I — behaviour: The LoP manipulation worked strongly: deeply encoded items were recognised better than shallowly encoded items (Encoding Level main effect: F(1,51)=241.44, P < 0.001, partial η2=0.83). Ketamine did not produce a significant main effect on Type 1 sensitivity (d'); retrieval performance and reaction times were unaffected by drug. In contrast, ketamine affected metacognitive measures: metacognitive sensitivity (meta-d') was reduced under ketamine during retrieval, and metacognitive bias was increased, indicating overconfidence in the ketamine group (metacognitive bias: t(51)=2.15, P=0.037, d=0.59). Metacognitive efficiency (meta-d'/d') did not show a statistically significant group difference in the primary frequentist analyses; hierarchical Bayesian estimation yielded an estimated group difference in meta-d'/d' of mean 0.23 with a 95% HDI of −0.04 to 0.58, indicating uncertainty about a true effect on efficiency. Exploratory correlations found no significant relationships between the 5D-ASC global score and Type 1 or Type 2 measures in either study phase (all P > 0.006). Study Phase I — fMRI: Contrasting Report versus Follow trials revealed greater BOLD in posterior visual regions (right calcarine, lingual gyrus, bilateral cuneus, superior occipital gyrus) and a left posterior medial frontal cortex (pMFC) cluster. The reverse contrast (Follow>Report) engaged typical default-mode network regions (angular gyrus, precuneus, PCC, superior frontal and parahippocampal cortex). Comparing drug groups during second-order ratings (Report and Follow combined), ketamine produced increased BOLD relative to placebo in multiple posterior clusters: right superior parietal lobule (SPL) extending to supramarginal and inferior parietal lobule/angular gyrus, left calcarine gyrus, right lingual gyrus, left IPL, and a left-hemispheric lingual–fusiform cluster. A confidence-parametric modulation of Report trials produced a similar ketamine>placebo pattern (bilateral lingual, fusiform and calcarine gyri and right SPL). There were no significant drug effects for the reverse contrasts and no interactions reported; Type 1 BOLD did not differ between drug groups. Study Phase II — behaviour and fMRI: Encoding during ongoing infusion showed no significant ketamine-related differences in BOLD activity (no cluster-level effects). Behaviourally, a strong LoP effect persisted (deep > shallow: F(1,51)=273.94, P < 0.001, partial η2=0.85), and there was no significant effect of drug on Type 1 sensitivity for items encoded during infusion. Metacognitive bias remained elevated under ketamine even when retrieval occurred after infusion had ceased, indicating persistent overconfidence; however, no reliable drug effect on meta-d' or meta-d'/d' was detected in this phase. Reaction times were unaffected across phases. fMRI analyses for Study Phase II included 50 participants after exclusions.

Lehmann and colleagues interpret their findings as evidence that NMDA-receptor antagonism with ketamine perturbs metacognitive monitoring without producing a general degradation of first-order episodic memory performance or processing speed. Behaviourally, ketamine administered during retrieval reduced metacognitive sensitivity (meta-d') and increased metacognitive bias (overconfidence), while leaving Type 1 recognition accuracy and reaction times largely intact. The lack of a clear group difference in metacognitive efficiency (meta-d'/d') in both frequentist and hierarchical Bayesian analyses tempers conclusions about an isolated impairment of metacognitive computation independent of primary-task performance. At the neural level, ketamine produced an up-regulation of activity in posterior cortical regions — superior parietal, inferior parietal, calcarine, lingual and fusiform gyri — during second-order rating periods, and these effects were not specific to genuine confidence formation since they were seen in both Report and Follow conditions. The authors suggest two non-mutually exclusive mechanisms: (1) ketamine may increase noise or reduce the reliability of sensory evidence feeding into metacognitive computations, prompting greater posterior cortical recruitment as participants attempt to process distorted input; and (2) ketamine's alteration of phenomenal conscious experience (confirmed by the 5D-ASC) may introduce vivid, imaginative or hallucinatory-like content that perturbs second-order judgements and shifts confidence criteria, producing overconfidence. The posterior localisation of ketamine-related BOLD increases is noted as consistent with proposals that sensory/posterior cortices form a 'hot zone' for phenomenal aspects of consciousness. The authors caution that the observed posterior BOLD increases occurred for both Report and Follow trials and therefore may reflect altered processing of the rating scale or sustained imaginative activity rather than core confidence computations. They also acknowledge the absence of correlation between subjective 5D-ASC scores and metacognitive changes, limiting direct linkage between the phenomenological state and objective metacognitive impairment. Limitations discussed include the between-subjects design (which avoids expectancy effects but cannot fully equate individual differences), potential habituation to the ketamine state over the infusion, possible inference about subsequent memory testing during Study Phase II encoding, the necessity to include only correctly retrieved trials in fMRI analyses due to too few errors, and limited power to detect some group differences. The authors call for further research using advanced computational models (for example to distinguish increased sensory noise from variable confidence criteria) and causal modelling of fMRI data to clarify network dynamics underlying the observed effects. Overall, they conclude that while ketamine impacts metacognitive bias and sensitivity and up-regulates posterior cortical activity during second-order ratings, these effects are neither strong nor specific enough to attribute metacognitive function solely to glutamatergic mechanisms.

Lehmann and colleagues conclude that acute NMDA-receptor antagonism with ketamine affects metacognitive processing: it increases metacognitive bias (overconfidence) and is associated with reduced metacognitive sensitivity during retrieval, alongside nonspecific increases in posterior cortical activity during second-order rating periods. At the same time, ketamine did not produce clear changes in metacognitive efficiency or first-order recognition accuracy. The authors emphasise that their results suggest a modulatory role for glutamatergic neurotransmission in metacognition but are not sufficient to attribute metacognition exclusively to the glutamate system; further experimental and modelling work is needed to delineate the underlying mechanisms.