Neuroimaging correlates and predictors of response to repeated-dose intravenous ketamine in PTSD: preliminary evidence

Abstract

Promising initial data indicate that the glutamate N-methyl-D-aspartate (NMDA) receptor antagonist ketamine may be beneficial in post-traumatic stress disorder (PTSD). Here, we explore the neural correlates of ketamine-related changes in PTSD symptoms, using a rich battery of functional imaging data (two emotion-processing tasks and one task-free scan), collected from a subset of participants of a randomized clinical trial of repeated-dose intravenous ketamine vs midazolam (total N = 21). In a pre-registered analysis, we tested whether changes in an a priori set of imaging measures from a target neural circuit were predictive of improvement in PTSD symptoms, using leave-one-out cross-validated elastic-net regression models (regions of interest in the target circuit consisted of the dorsal and rostral anterior cingulate cortex, ventromedial prefrontal cortex, anterior hippocampus, anterior insula, and amygdala). Improvements in PTSD severity were associated with increased functional connectivity between the ventromedial prefrontal cortex (vmPFC) and amygdala during emotional face-viewing (change score retained in model with minimum predictive error in left-out subjects, standardized regression coefficient [β] = 2.90). This effect was stronger in participants who received ketamine compared to midazolam (interaction β = 0.86), and persisted following inclusion of concomitant change in depressive symptoms in the analysis model (β = 0.69). Improvement following ketamine was also predicted by decreased dorsal anterior cingulate activity during emotional conflict regulation, and increased task-free connectivity between the vmPFC and anterior insula (βs = -2.82, 0.60). Exploratory follow-up analysis via dynamic causal modelling revealed that whilst improvement in PTSD symptoms following either drug was associated with decreased excitatory modulation of amygdala→vmPFC connectivity during emotional face-viewing, increased top-down inhibition of the amygdala by the vmPFC was only observed in participants who improved under ketamine. Individuals with low prefrontal inhibition of amygdala responses to faces at baseline also showed greater improvements following ketamine treatment. These preliminary findings suggest that, specifically under ketamine, improvements in PTSD symptoms are accompanied by normalization of hypofrontal control over amygdala responses to social signals of threat.

© 2021. The Author(s), under exclusive licence to American College of Neuropsychopharmacology.

Figures

Fig. 1. Description of study procedures, clinical measures, and functional imaging measures.

a Timeline of study procedures. Upon recruitment to the trial, participants were randomly allocated to a drug condition (ketamine vs midazolam) by the trial pharmacist in a 1:1 ratio. Participants received three intravenous infusions per week for two weeks of either 0.045 mg/kg midazolam or 0.5 mg/kg ketamine. The baseline (pre-infusion) imaging session took place prior to administration of any drugs. The second (post-infusion) scan was the day after the 4th or 5th drug infusion for 2/3 participants, and within 48 h of the 6th infusion for 1/3 participants. b Baseline (pre-infusion 1) and outcome visit clinical measures for all study participants. CAPS-5, Clinician-Administered PTSD Scale for DSM-5; MADRS, Montgomery-Åsberg Depression Rating Scale. Both measures probed symptom levels over the past week and were administered by blinded raters, who were not present during drug infusion sessions. Raincloud plots were generated using [92]. c Depiction of the target circuit (regions of interests [ROIs; coloured overlays], and connections between ROIs [grey arrows]) from which functional imaging measures were extracted. dACC, dorsal anterior cingulate cortex; rACC, rostral anterior cingulate cortex, vmPFC, ventromedial prefrontal cortex; aHC, anterior hippocampus; a.insula, anterior insula. d Top panels, the three functional imaging measures collected during each scan session: (1) the emotional face-processing task; (2) the emotional conflict regulation (face Stroop) task; (3) 12-min task-free (‘resting state’) scan. Bottom panels, imaging measures extracted from each functional run. Single ROIs represent mean univariate BOLD signal, pairs of ROIs represent generalized psychophysiological interaction (gPPI) functional connectivity estimates, and the multivariate dissimilarity metric was cross-validated linear discriminant contrast values.

Fig. 2. Neuroimaging correlates of PTSD symptom change.

a Standardized regression coefficient (beta) values for the elastic net model with minimum predictive error for change in PTSD symptoms over the course of treatment in left-out subjects. Non-zero coefficients represent measures with predictive value for change in PTSD severity. All imaging measures represent change scores (post-infusion scan minus pre-infusion scan). Grey shading highlights interaction terms between imaging measures and received drug identity (ketamine vs midazolam). rsfMRI resting-state/task-free fMRI scan, CADSS Clinician-Administered Dissociative States Scale, BPRS, four items from the Brief Psychiatric Rating Scale probing psychotomimetic symptoms, YMRS a single item from the Young Mania Rating Scale indexing elevated mood, PRISE Patient Rated Inventory of Side-Effects (total score calculated by summing across all somatic domains), CAPS-5 Clinician-Administered PTSD Scale for DSM-5, utox thc urine toxicology results for presence of THC. b Increased connectivity between the vmPFC and amygdala (AMG) during emotional face-viewing was predictive of improvement in total PTSD symptom severity across all participants, but the effect was stronger in individuals who received ketamine. Decreased dACC BOLD during emotional conflict regulation (c), and increased resting (task-free) vmPFC-anterior insula functional connectivity (d), were only associated with improvement in CAPS-5 total score in individuals who received ketamine. For generalized psychophysiological interaction (gPPI) and task-free functional connectivity estimates, values represent Fisher-transformed correlation coefficients. For regional BOLD signal, values are in arbitrary units. For ease of interpretation, bivariate plots represent ‘raw’ imaging measure values (i.e., calculated prior to regression against average within-scan framewise displacement).

Fig. 3. Dynamic causal modelling of task-modulated effective connectivity during the emotional face-processing task.

a Parametric Empirical Bayes (PEB) analysis of pre-infusion scan data for all participants revealed that, at baseline, the onset of emotional faces shifted the amygdala→vmPFC connection towards an excitation—with only weak evidence for a shift towards top-down inhibition of the AMG by the vmPFC. b PEB analysis of pre-post changes in connectivity related to improvement in PTSD symptoms revealed that across all participants, improvement in overall PTSD severity (CAPS-5 total score) was associated with both increased vmPFC inhibition of the amygdala, and decreased amygdala excitation of the vmPFC, during emotional face-viewing. c When participants were divided by drug condition, the relationship between PTSD symptom improvement and increased top-down inhibition of the amygdala by the vmPFC during emotional stimuli was only evident in participants who received ketamine, a difference confirmed by passing both models to a 3rd level PEB analysis (more negative parameter estimate for the vmPFC→amygdala connection in the ketamine group, posterior probability = 0.98). For all panels, pointed arrowheads represent connections between regions, and circular arrowheads represent modulation of those connections by the experimental condition of interest (emotional faces). Values are rates of change constants in Hz. Insets depict PEB posterior parameter estimates for modulation of connectivity by the listed effect of interest: 1, vmPFC→amygdala; 2, amygdala→vmPFC; error bars represent 90% confidence intervals.

Fig. 4. Baseline predictors of PTSD symptom change.

a Standardized regression coefficient (beta) values for the elastic net model with minimum predictive error for change in PTSD symptoms over the course of treatment in left-out subjects. Imaging measures represent estimates extracted from the baseline (pre-infusion) session. Grey shading highlights interaction terms between imaging measures and received drug identity (ketamine vs midazolam). fHX AUD, family history of alcohol use disorder in first degree relatives; MOS-SS, Medical Outcomes Study Social Support Survey total score; cogState, composite score for executive function derived from the cogState neurocognitive test battery. Lower rACC BOLD during emotional face-viewing (b) and increased baseline representational distance between fearful and neutral faces in the rACC (c) were retained in the minimum error model as baseline predictors of PTSD symptom improvement across all study participants. d Decreased baseline connectivity between the vmPFC and amygdala (AMG) during emotional face-viewing was predictive of symptom improvement across all participants, with evidence of a stronger effect in individuals who went on to receive ketamine (interaction with drug retained in the model with minimum predictive error). Regional BOLD signal values are in arbitrary units. Representational dissimilarity was evaluated using cross-validated linear discriminant contrast estimates, equivalent to the Mahalanobis distance between patterns of response evoked by each kind of stimulus. For generalized psychophysiological interaction (gPPI) functional connectivity estimates, values represent Fisher-transformed correlation coefficients. Bivariate plots represent ‘raw’ imaging measure values (i.e., calculated prior to regression against average within-scan framewise displacement).