Ketamine induces a robust whole-brain connectivity pattern that can be differentially modulated by drugs of different mechanism and clinical profile

Ketamine, an N-methyl-D-aspartate receptor (NMDAR) antagonist, induces transient glutamatergic dysfunction and acutely produces psychotomimetic effects in healthy volunteers. Earlier neuroimaging work has documented robust ketamine effects on the blood oxygen level-dependent (BOLD) signal, typically analysed in terms of amplitude, and pre-treatment with compounds that reduce glutamate release has been shown to attenuate those amplitude changes. However, ketamine also affects functional connectivity across the brain, and it remains uncertain whether connectivity changes are driven primarily by direct NMDAR blockade or by downstream glutamatergic mechanisms. Joules and colleagues set out to characterise whole-brain task-free functional connectivity changes induced by an acute sub-anaesthetic ketamine infusion and to test whether two pharmacological probes with different mechanisms—lamotrigine (a presynaptic inhibitor of glutamate release via sodium channel modulation) and risperidone (an atypical antipsychotic with 5-HT2A antagonism and reported NMDAR potentiation)—differentially modulate those connectivity effects. Using a multivariate pattern-recognition approach applied to graph-theory centrality metrics, the study aims to identify ketamine-induced connectivity patterns and to determine whether pre-treatment with lamotrigine or risperidone attenuates or alters these patterns, thereby informing mechanistic interpretation.

The investigators conducted a double-blind, placebo-controlled, cross-over study in 20 initially recruited right-handed male volunteers; after withdrawals, 16 participants (age range 20–37, mean 25.8, SD 5.7) completed all sessions. Each participant attended four sessions separated by at least 10 days. Oral pre-treatments were lamotrigine 300 mg, risperidone 2 mg, or placebo (ascorbic acid, given in two sessions). Following oral dosing at times chosen to coincide with maximal plasma exposure, an intravenous infusion of either racemic ketamine (target plasma level 75 ngml-1) or saline was administered. The ketamine dosing protocol consisted of a bolus of approximately 0.12 mgkg-1 in the first minute followed by a pseudocontinuous infusion of ~0.31 mgkg-1 h-1; plasma samples for ketamine, risperidone and lamotrigine concentrations were taken at specified intervals. Functional MRI data were acquired on a 3T scanner, collecting 450 task-free EPI volumes over 15 min (TR=2000 ms). The infusion began 5 min into the scan. Pre-processing included slice-timing correction, realignment, co-registration to structural images, spatial normalisation, smoothing (8 mm FWHM), tissue segmentation, nuisance regression (mean CSF, mean white matter and six motion parameters) and band-pass filtering (0.01–0.1 Hz). The time series for each session were divided into pre- and post-infusion conditions defined as the first and last 150 volumes (5 min each). For network analysis, the pre-processed data were parcellated using the automated anatomical labelling (AAL) atlas into 116 regions; node time series were the mean voxel time course per region. To account for temporal dynamics of resting-state networks, a tapered Gaussian moving window of length 120 volumes (240 s) was applied in increments equivalent to 2 s, yielding eight windowed time series per condition. Pairwise linear correlations between nodes produced adjacency matrices per window. Degree centrality (weighted degree) was computed for each node and normalised by network density to quantify regional connectedness. Multivariate analysis employed Gaussian process classification (GPC) for binary contrasts and ordinal regression using Gaussian processes (ORGP) for rank-ordered analyses. Feature sets comprised subject-wise mean-centred degree centrality measures across all windows. Classifier performance was assessed with leave-one-out cross-validation (LOOCV); statistical significance was estimated by permutation testing with Nperm=1000 and Bonferroni correction applied across multiple contrasts. For significant models, multivariate maps (g-maps) visualised node weights driving classification; univariate paired t-tests on node differences were performed qualitatively without formal multiple comparison correction.

Ketamine produced a robust and classifiable change in whole-brain degree centrality (DC) compared with control conditions. Binary classification separating the post-infusion ketamine condition from the pre-infusion placebo condition (PLA+KET) achieved 81.25% accuracy (p<0.003), and ketamine versus saline achieved 87.5% accuracy (p<0.001). By contrast, classifiers could not reliably separate the two pre-infusion placebo sessions (Acc=68.75%, p>0.05) nor the pre- versus post-infusion saline session (Acc=62.5%, p>0.05), supporting within-session and between-session stability in the absence of active pharmacology. The spatial pattern driving ketamine classification reflected a shift from cortical to subcortical and cerebellar dominance. Nodes with increased DC in the ketamine state (positive weights) were concentrated in basal ganglia and cerebellar regions, while nodes with decreased DC (negative weights) were primarily cortical, notably occipital, temporal, medial temporal and frontal areas. Post hoc univariate tests confirmed these directions of change across nodes. Risperidone pre-treatment modulated the ketamine-induced DC pattern. Classification distinguished the risperidone-plus-ketamine (RIS+KET) condition from placebo pre-treated ketamine (PLA+KET) with 81.25% accuracy (p<0.001), indicating risperidone altered the ketamine effect. RIS+KET was also discriminable from saline (Acc=81.25%, reported p>0.05 in the extracted text), and direct comparison of lamotrigine and risperidone post-ketamine states showed strong discrimination (lam+ketamine v ris+ketamine, Acc=93.75%, p<0.001). Univariate contrasts suggested risperidone increased DC in frontal and temporal cortices and decreased DC in basal ganglia, occipital and parietal regions relative to placebo-pre-treated ketamine. The classifier could not reliably separate pre- and post-infusion for the risperidone session (risperidone v ris+ketamine, Acc=62.5%, p>0.05), and when a GPC trained on saline and ketamine was used to classify RIS+KET samples, RIS+KET was variably confused with both classes. Lamotrigine pre-treatment did not show evidence of modulating ketamine-induced DC patterns. Classification could not distinguish lamotrigine from placebo (Acc=32.125%, p>0.05) nor ketamine from lamotrigine-plus-ketamine (ketamine v lam+ketamine, Acc=43.75%, p>0.05). Within-session ketamine effects remained detectable for both placebo and lamotrigine pre-treated sessions (placebo pre-ket v ketamine Acc=81.5%, p<0.001; lamotrigine pre v lam+ketamine Acc=87.5%, p=0.01). Direct lamotrigine v risperidone comparisons were not significant (Acc=62.5%, p>0.05). Ketamine plasma concentrations did not show a significant correlation with subject-level classifier posterior probabilities (no significant correlation at p<0.05). Ordinal regression analyses did not reveal a clear attenuative ordinal trend for either pre-treatment. In one configuration saline was well identified but ketamine and lamotrigine-plus-ketamine were confused; in the other, saline and ketamine were distinguished while risperidone-plus-ketamine was confused with both classes.

Joules and colleagues interpret their findings as showing that a sub-anaesthetic ketamine infusion reliably reconfigures whole-brain functional connectivity, shifting from a cortically centred organisation toward increased centrality in subcortical structures and the cerebellum. The authors note that this multivariate centrality change complements prior reports of increased ‘‘global connectivity’’ under ketamine and expands those observations by mapping a coherent spatial pattern of increased subcortical and decreased cortical hubness. Mechanistically, the investigators propose that NMDAR antagonism underlies the observed connectivity reorganisation. They discuss a plausible microcircuit mechanism in which ketamine disinhibits parvalbumin-positive GABAergic interneurons, impairing pyramidal synchrony and reducing cortical hub connectivity; the thalamus may contribute to disrupted cortical synchrony and thus appear with increased centrality. The pattern of connectivity alteration is linked by the authors to ketamine's acute experiential effects—perceptual distortion, cognitive disorganisation and anhedonia—via affected parietal, visual, fronto-parietal and striatal/ventromedial nodes. The differential effects of the two pre-treatments are emphasised. Lamotrigine, despite attenuating ketamine-evoked BOLD amplitude in prior work and being administered at a clinically effective 300 mg dose here, did not alter the ketamine-induced DC pattern; the authors infer that lamotrigine’s presynaptic reduction of glutamate release does not account for the connectivity reorganisation. In contrast, risperidone substantially modulated ketamine-induced DC, producing a DC pattern distinct from both saline and ketamine. The authors suggest risperidone's combined pharmacology—5-HT2A antagonism reducing downstream glutamate and reported NMDAR potentiation—may oppose or interact with ketamine’s action, and that NMDAR-related mechanisms are a likely contributor to the DC effects. They acknowledge that serotonergic and dopaminergic receptor actions could also play a role and indicate that studies with more selective ligands would be required to disentangle receptor-specific contributions. Methodological strengths and caveats are discussed. The multivariate graph-theory plus machine-learning approach demonstrated stability in the absence of pharmacological challenge, supporting interpretation of drug effects. Limitations include the focus on a single connectivity metric (degree centrality) and temporal scale, insensitivity to phase-delayed or non-linear interregional relationships, and the inability of centrality measures to localise changes in specific connections. The authors also note that low ketamine doses produced sparse subjective effects, limiting correlation of connectivity changes with phenomenology, and that broader sets of connectivity measures and alternative windowing scales would be informative in future work. Overall, the authors conclude that the connectivity findings favour a primary role for NMDAR blockade in generating the observed reorganisation and that multivariate connectivity analyses yield mechanistic information distinct from amplitude-based BOLD measures, with implications for interpreting pharmacological manipulations of brain networks.

The study demonstrates a robust acute connectivity signature of sub-anaesthetic ketamine characterised by reduced cortical hub connectivity and increased degree centrality in basal ganglia and cerebellar nodes. Pattern-recognition analysis showed no spurious changes in the absence of pharmacological stimulus. Pre-treatment with risperidone, but not lamotrigine, markedly modulated the ketamine-induced hub changes, a result the authors attribute to the distinct mechanisms of the two compounds and which suggests the connectivity alterations are more likely driven by NMDAR blockade and possible serotonergic modulation rather than by presynaptic suppression of glutamate release alone. The authors emphasise that connectivity analyses offer complementary mechanistic insight to conventional BOLD amplitude measures and can inform on neural-circuit correlates of specific pharmacological mechanisms.