Ketamine improves short-term plasticity in depression by enhancing sensitivity to prediction errors

This EEG study (n=30) in patients with depression shows that prediction error sensitivity (a possible proxy for brain plasticity, lacking in this population) is improved by ketamine (30.8mg/70kg).

Abstract

“Major depressive disorder negatively impacts the sensitivity and adaptability of the brain’s predictive coding framework. The current electroencephalography study into the antidepressant properties of ketamine investigated the downstream effects of ketamine on predictive coding and short-term plasticity in thirty patients with depression using the auditory roving mismatch negativity (rMMN). The rMMN paradigm was run 3-4 h after a single 0.44 mg/kg intravenous dose of ketamine or active placebo (remifentanil infused to a target plasma concentration of 1.7 ng/mL) in order to measure the neural effects of ketamine in the period when an improvement in depressive symptoms emerges. Depression symptomatology was measured using the Montgomery-Asberg Depression Rating Scale (MADRS); 70% of patients demonstrated at least a 50% reduction their MADRS global score. Ketamine significantly increased the MMN and P3a event related potentials, directly contrasting literature demonstrating ketamine’s acute attenuation of the MMN. This effect was only reliable when all repetitions of the post-deviant tone were used. Dynamic causal modelling showed greater modulation of forward connectivity in response to a deviant tone between right primary auditory cortex and right inferior temporal cortex, which significantly correlated with antidepressant response to ketamine at 24 h. This is consistent with the hypothesis that ketamine increases sensitivity to unexpected sensory input and restores deficits in sensitivity to prediction error that are hypothesised to underlie depression. However, the lack of repetition suppression evident in the MMN evoked data compared to studies of healthy adults suggests that, at least within the short term, ketamine does not improve deficits in adaptive internal model calibration.”

Authors: Rachael L. Sumner, Rebecca McMillan, Meg J. Spriggs, Doug Campbell, Gemma Malpas, Elizabeth Maxwell, Carolyn Deng, John Hay, Rhys Ponton, Frederick Sundram & Suresh D. Muthukumaraswamy

Notes

This paper is included in our ‘Top 12 Articles on on Ketamine for Mental Health‘

Summary

Abstract

Major depressive disorder negatively impacts the brain’s predictive coding framework. Ketamine significantly increased the MMN and P3a event related potentials in patients with depression, directly contrasting literature demonstrating ketamine’s acute attenuating effects.

1. Introduction

Ketamine has proven to be a powerful tool for treating and understanding the neurobiology of depression. The effects of ketamine are sustained for about a week. Ketamine’s antidepressant properties are due to its alteration of downstream signalling cascades, which in turn enhances neural plasticity via Hebbian long-term potentiation (LTP) within 3 h. This is just one mechanism through which learning and memory occurs in the brain.

Predictive coding is a recursive form of plasticity characterised by feedback loops, whereby the brain forms a generative model based on an internal probabilistic model. Hebbian learning is intrinsic to this process, and leads to long-term memory via optimal matching between presynaptic predictions and postsynaptic prediction error.

Predictive coding is most frequently studied using mismatch paradigms, which involve repeated exposure to an unexpected sensory input. Repetition suppression is a neural mechanism underlying the brain’s active process of recalibrating its predictive generative model based on error signals received following an unexpected event. The predictive model updates the lower regions in the cortical hierarchy based on new sensory context information.

Theoretical accounts of major depressive disorder suggest the depressed brain is less sensitive to prediction error and/or miscalibration of model precision.

Studies using sensory mismatch paradigms have shown that the MMN and P3a response to deviant stimuli are reliable indices of aberrant cognitive functioning in a range of central nervous system disorders, including schizophrenia and bipolar depression. However, the findings are not consistent.

Ketamine can be expected to play a role mediating the neural mechanisms behind predictive coding, and repetition suppression in participants with MDD. However, there is a lack of research on the impact of ketamine on predictive coding in the time-frame that the antidepressant effects are beginning to emerge.

Ketamine’s pro-glutamatergic and pro-plasticity AMPA receptor-mediated effects are already evidenced by its enhancement of NMDA vere depression. Ketamine also has widespread effects on forward and backward projecting as well as intrinsic connectivity.

The current study will use DCM to investigate changes to the MMN, P3a, and repetition suppression in patients with MDD, and will determine whether ketamine improves short-term mechanisms of neural plasticity, alleviating deficits in incoming error processing and internal model calibration that may underlie depression.

2.1. Study design

This study used a randomised, double-blind, active placebo-controlled crossover design in 30 participants who met DSM-V criteria for MDD, and provided a summary of demographics in Table 1 and Supplementary Material.

Participants received racemic ketamine or remifentanil hydrochloride on one study visit, and the active placebo on the other. The order of treatment was randomised and counterbalanced.

Ketamine decreased depression symptoms in 70% of participants at 1 day post-infusion compared to the baseline (pre-infusion). There was no difference in baseline MADRS between responders and non-responders pre-ketamine.

The 1-day MADRS post ketamine was used as the primary outcome measure and for correlations.

2.2. Drug infusion

All interventions were administered by an anaesthetist. Ketamine was administered as a 0.25 mg/kg bolus, followed by a 0.25 mg/kg/hr infusion for 45 min, and remifentanil was administered as a 9 minute target-controlled infusion.

2.3. Roving mismatch negativity task

Participants engaged in a roving auditory oddball task using trains of one to eleven identical sinusoidal tones. The roving paradigm differs from a classic oddball task in that all trains had at least 5 subsequent repetitions, and an additional 1-5 tones occurred randomly but equally often.

Tone frequency varied within 500 and 800 Hz in random steps of integer multiples of 50 Hz, and participants were instructed to ignore the tones and focus on a visual distractor task.

2.6. Event-related potential analyses

The pre-processed data were averaged based on their position in the train of tones, collapsed across frequencies, and smoothed using a 6 6 6 FWHM Gaussian kernel. ERP analyses were performed using SPM12 and controlled for the multiple comparisons problem.

A 2 2 ANOVA was run to confirm the MMN had been elicited, and to test the effect of ketamine on the MMN. A 2 4 ANOVA was performed to assess tone repetition effects.

Main effects and interaction effects were considered significant at p 0.05 family-wise error corrected (FWE-c), and simple effects tests were conducted as appropriate.

2.7. Source space analysis

The Multiple Sparse Priors method was used to select sources within the peak time-window of interest for the MMN and P3a components, and a 2 x 2 ANOVA was run to test the effect of deviant and ketamine.

2.8. Dynamic causal modelling

DCM is used to explore the generative model of the effective connectivity between biophysically plausible and hypothesis driven cortical networks, using M/EEG evoked response data. It uses hierarchically organised extrinsic and intrinsic connections.

DCM was carried out using the standard ERP neural mass model to explore the likelihood of a parameter given the data in terms of both a main effect of deviant, and a main effect of ketamine. Six different model architectures were defined and 36 different combinations of these models were tested.

Statistical analysis was performed on the winning family using Random Effects (RFX) Bayesian Model family-level inference. The winning model was identified using the family exceedance probability (fxp), and the individual posterior parameter estimates were correlated with the antidepressant response (determined by percent change in MADRS from baseline to 24 h post-ketamine).

Wilcoxon-sign rank tests and Spearman’s rho correlations were used to test for significant changes in the modulation of posterior parameter estimates.

3. Results

The 2 2 ANOVA revealed that the MMN ERP peaked at 243 ms with a frontal distribution, and that the bilateral temporal peak at 233 ms and a fronto-central P3a component at 333 ms were significantly more positive for the deviant than standard tone.

The main effect of ketamine was not significant. A ketamine by deviant interaction was found at 225 ms, but the result was determined to be too small to interpret.

A 2 x 4 ANOVA was performed to investigate the effect of ketamine versus placebo on the difference wave of the deviant tone minus subsequent tones. Ketamine had a significant effect on the left lateralised frontal peak and the posterior temporal peak.

Post-hoc t-contrasts revealed that both negativities were significantly more negative in the ketamine session, while the temporal peak and P3a were more positive.

3.1. ERP and MADRS covariate analyses

A highly exploratory investigation of the effect of ketamine on the MMN and P3a response to deviant was conducted. The results revealed no significant relationship.

3.2. Source analysis

Source analysis in the 200 – 300 ms time-window revealed bilateral inferior temporal cortex (ITC) and a right inferior frontal gyrus source (IFG) that were used as nodes for the DCM. Ketamine had a main effect on ITC, with left fusiform gyrus showing significantly lower activation overall post ketamine.

3.3. DCM results

The FB model was found to have 4.77-fold greater evidence compared to the next best model for the effect of deviant tone, and also had very strong evidence for the effect of ketamine.

3.4. DCM and MADRS correlation

The connection from right A1 to ITC was significantly modulated by deviant tone and ketamine, and was negatively correlated with the magnitude of the antidepressant response to ketamine.

4. Discussion

The current study revealed that ketamine increases the amplitude of the roving MMN several hours post-infusion, indicating increased sensitivity to prediction error, and that greater modulation of right A1 to ITC connection strength underlying the effect of a deviant tone was significantly correlated to the antidepressant response to ketamine at 24 h.

Fig. 7 shows that greater reduction in MADRS score was correlated with higher modulation of connectivity in the right primary auditory cortex and right inferior temporal cortex in response to deviant tones.

The evoked response results of the roving MMN paradigm post-ketamine revealed a more negative MMN response to deviant tones, which was strengthened when the four subsequent tone repetitions were also analysed. There was no effect of ketamine on repetition suppression.

The winning model from the BMS of the DCMs included forward and backwards connections, and the greater modulation of the right A1 to ITC connection in response to the deviant tone was significantly related to a greater percent reduction in MADRS.

In the predictive coding framework, forward connection modulation represents a change in sensitivity to prediction error, and the MMN represents a failure to suppress the error response. The MMN and P3a ERPs may not represent the affected mechanism, or may be less reliably projected.

Depression is thought to lead to insensitivity to prediction error and/or miscalibration of model precision. Ketamine may act on the ‘locked in’ brain to restore sensitivity to external sensory input and prediction errors when the input is unexpected.

Participants in the current study demonstrated disordered repetition suppression that was not changed by ketamine. This may indicate that the generative model was not properly calibrated.

Ketamine did not appear to modify repetition suppression within the time-frame of the initial mood changes, but did appear to influence deviant stimulus processing by improving sensitivity to prediction errors.

There are several limitations to this study, including the fact that 20/30 of the participants were on antidepressant medication, and that there is no previous research testing ketamine and the MMN 3 – 4 h post-ketamine rather than acutely.

The decision to use data driven sources of deviant processing led to the use of ITC rather than the more commonly used STG for DCM. However, an identical DCM analysis using STG is provided in the Supplementary Material.

Ketamine increases predictive coding mechanisms but not repetition suppression, and may mediate an antidepressant response by increasing sensitivity to prediction errors from unexpected sensory input.

Role of funding source

Researchers from Auckland Medical Research Foundation, Rutherford Discovery Fellowship, Brain Research New Zealand and the Health Research Council of New Zealand funded the study.

CRediT authorship contribution statement

Rachael L. Sumner, Rebecca McMillan, Meg J. Spriggs, Doug Campbell, Gemma Malpas, Elizabeth Maxwell, Carolyn Deng, John Hay, Rhys Ponton, Frederick Sundram, Suresh D. Muthukumaraswamy contributed to this work.