Ketamine induces multiple individually distinct whole-brain functional connectivity signatures

Earlier whole-brain studies indicate that ketamine produces robust, brain-wide alterations in functional connectivity, including increased global brain connectivity (GBC) in prefrontal cortex and shifts from cortical- to subcortically-centred states. However, those group-average approaches can obscure meaningful inter-individual differences, and prior findings have been inconsistent across timepoints and samples. The authors note that data-driven multivariate methods such as principal component analysis (PCA) can reveal both group-level and individual-level axes of variation and therefore are well suited to test whether ketamine's effects are uni-dimensional (a single uniform pattern) or multi-dimensional (multiple distinct patterns across individuals).

Moujaes and colleagues conducted a single-blind, within-subjects pharmacological neuroimaging study in N=40 healthy adults who completed two scanning sessions on the same day: a placebo (saline) session first and a ketamine session second (an initial bolus 0.23 mg/kg followed by continuous infusion 0.58 mg/kg/hr). Participants were screened for general health, absence of psychiatric or neurological disorders, and other standard MRI and safety criteria; the sample size was selected a priori to provide >86% power for an estimated medium effect (Cohen's d=0.5). Subjective measures (PANSS, CADSS, BDI) and an in-scanner spatial working memory task were collected; participants completed subjective scales 180 minutes after drug administration. Blood samples taken after the resting-state run yielded a mean serum ketamine concentration of 470 nmol/L. The investigators note that participants correctly identified the ketamine infusion 100% of the time. Resting-state functional MRI was acquired on a Siemens 3T scanner with HCP-aligned parameters (multiband EPI, TR=700 ms, 400 volumes, 4.67 min per resting run). The analysis used surface/volume 'grayordinate' CIFTI space. Data were preprocessed with the HCP minimal pipeline via QuNex, including motion correction, registration to MNI-152, FreeSurfer segmentation, conversion to grayordinate space, scrubbing of motion-affected frames (FD>0.5 mm or RMS>1.6×median), high-pass filtering (0.008 Hz), and nuisance regression including global signal regression (GSR); the authors also report parallel analyses without GSR to address that controversy. Neural data were reduced at multiple spatial scales: dense grayordinates (91,281), a 718-parcel functional parcellation (CAB-NP), 12 networks, and five subcortical regions. Global brain connectivity (GBC) was computed as the mean Fisher z-transformed Pearson correlation of each parcel/grayordinate with all others, yielding subject-wise Δ (ketamine − placebo) GBC maps. The primary neural analyses performed a PCA on the 718-parcel Δ GBC maps across participants to identify principal neural gradients; significance was assessed with permutation testing. Effective dimensionality of Δ GBC effects for ketamine (N=40), LSD (N=24), and psilocybin (N=23) was estimated using the participation ratio of covariance eigenvalues, with resampling/jackknifing to equalise sample size to 22 and group comparisons via one-way ANOVA and post-hoc t-tests. To relate neuroimaging effects to putative molecular targets, the authors correlated cortical maps of Δ GBC (and PCA-derived neural components) with cortical gene expression maps for somatostatin (SST) and parvalbumin (PVALB) derived from the Allen Human Brain Atlas (mapped to 180 HCP cortical parcels). Statistical significance of gene–imaging correlations was assessed by comparing observed correlations to a null distribution generated from 100,000 surrogate maps that preserve the spatial autocorrelation of the target map (BrainSMASH), with FDR correction. Behavioural data (31 PANSS items plus the cognition score) were submitted to a separate PCA across the 40 participants; behavioral PCs surviving permutation testing (5000 permutations) were then regressed onto parcel-wise Δ GBC via mass-univariate models (TFCE and permutation testing) to produce neuro-behavioural maps. Additional comparisons included regressing conventional PANSS three- and five-factor subscales and comparing ketamine results to LSD and psilocybin.

Principal component analysis of the parcel-wise Δ GBC maps for ketamine revealed five significant bi-directional neural components that together explained 42.1% of variance; PC1 and PC2 each accounted for >10% of variance and were the main focus. PC1 and PC2 differentiated association versus sensory systems: some individuals showed increased GBC in association networks (for example default mode) with decreased GBC in sensory cortices, while others showed the opposite pattern. Comparing substances, ketamine produced significantly higher effective dimensionality of Δ GBC than LSD and psilocybin (no difference between LSD and psilocybin), indicating a more complex, multi-dimensional neural effect for ketamine. The authors also note motion did not drive dimensionality differences. A mean-based (uni-dimensional) Δ GBC map was moderately correlated with neural PC1 but correlated weakly with PC2–PC5; PC1 captured more variation than the mean and the two maps were not linearly related in their negative values. Mapping neural components to cortical gene expression, PC1 Δ GBC correlated significantly with SST (r=0.47, p<0.001, 100,000 permutations) and negatively with PVALB (r=-0.22, p=0.025). In contrast, the mean Δ GBC map did not show significant correlations with SST or PVALB. PC3–PC5 also showed associations with SST/PVALB in supplemental analyses. Behavioural PCA on the 31 Δ measures (PANSS items + cognition) produced two significant bi-directional behavioural components that together explained approximately 41.4% of variance (PC1 ≈29%, PC2 ≈12%). Behavioural PC1 loaded on negative-symptom-like items (blunted affect, social withdrawal, poor attention) plus some general and disorganisation items, whereas PC2 loaded on a mixture including poor rapport, grandiosity, hallucinations/delusions, uncooperativeness, and cognition. Regressing the behavioural PCs onto parcel-wise Δ GBC produced two neuro-behavioural maps (neuro-behavioural PC1 and PC2). Behavioural PC1 scores were highly correlated with neuro-behavioural PC1 scores across individuals (r=0.70, p<0.001), indicating preserved participant rank order when behaviour is related to neural change. Neuro-behavioural PC1 showed a positive correlation with the SST cortical gene-expression map (r=0.29, p=0.009) and a non-significant negative correlation with PVALB (r=-0.14, p=0.089). At the single-subject level, 35/40 participants exhibited distinct SST/PVALB association patterns when each individual's Δ GBC map was correlated with SST and PVALB maps: 23 participants (≈58%) showed a positive SST and negative PVALB pattern, while 12 showed the reverse pattern. Ordering participants by their neuro-behavioural PC1 score revealed that negative neuro-behavioural PC1 scores broadly corresponded to negative SST and positive PVALB associations, and positive neuro-behavioural PC1 scores corresponded to the opposite pattern. Additional analyses showed that regressions of PANSS three- and five-factor subscales produced broadly comparable neural maps to the behavioural PCs, with neuro-behavioural PC2 correlating more strongly with a five-factor Positive scale (r=0.73) than with the three-factor Positive scale (r=0.42).

Moujaes and colleagues interpret their findings as evidence that acute ketamine induces multi-dimensional neural and behavioural effects with substantial inter-individual variability, rather than a single uniform response. The presence of five bi-directional neural principal components, and two behavioural components, suggests ketamine's acute impact is heterogeneous across people and brain systems. The investigators argue that a multi-dimensional analytic approach better aligns ketamine's system-level imaging effects with its putative molecular mechanisms: PCA-derived neural gradients (not the mean Δ GBC) showed robust associations with cortical SST and PVALB gene-expression patterns, consistent with hypotheses that ketamine acts via interneuron-sensitive pathways. The authors propose several mechanistic interpretations consistent with their data: different principal components may reflect distinct microcircuit targets (for example SST versus PVALB interneuron pathways), differential action at NMDA receptor subunits with varying affinities, or complementary inhibitory and excitatory dimensions that are less prominent with serotonergic psychedelics. This multiplicity of microcircuit-level targets could explain why ketamine showed higher effective dimensionality than LSD and psilocybin in their comparisons. Importantly, the neuro-behavioural framework produced maps that are interpretable at the single-subject level: an individual's behavioural PC1 score largely predicts their position on the corresponding neuro-behavioural neural gradient and whether their Δ GBC map aligns more with SST or PVALB cortical expression. The authors acknowledge several limitations they consider important for interpreting the findings. Because participants identified ketamine reliably, the single-blind design and lack of an active placebo may introduce expectancy effects. Scans were conducted on the same day without counterbalancing owing to residual ketamine effects, and imaging was limited to peak acute effects rather than the delayed antidepressant window (24–72 hr). Physiological variables such as heart rate and blood pressure were not regressed out, which could affect resting-state connectivity; the use of GBC as a summary measure may omit some behaviourally relevant connectivity details; and larger samples are required to unpack higher-order PCs (PC3–PC5) and to test demographic moderators. Finally, the authors stress that the next step is to test this multi-dimensional framework in clinical populations to evaluate whether the observed individual differences predict therapeutic response.

The study demonstrates robust inter-individual variability in both neural and behavioural responses to acute ketamine, which is better captured by a multi-dimensional analytic framework than by mean-based approaches. That framework links individual behavioural gradients to distributed neural connectivity changes and to cortical gene-expression patterns for SST and PVALB, and it is resolvable at the single-subject level. The authors propose that this multi-dimensional mapping could ultimately support predictions of individual treatment response, pending validation in patient samples and further methodological refinements.