Mapping Pharmacologically-induced Functional Reorganisation onto the Brain’s Neurotransmitter Landscape

The authors assembled cortical PET maps for 18 receptors and transporters (examples include D1, D2, DAT; NET; multiple serotonin receptors and transporter subtypes; acetylcholine receptors and VAChT; mGluR5; GABA-A; H3; CB1; MOR) and parcellated these maps to 100 cortical regions using the Schaefer atlas. Resting-state fMRI datasets were collected or re-used for seven pharmacological interventions: two independent propofol datasets (Western University and Cambridge), sevoflurane, ketamine, LSD, ayahuasca, modafinil, and a methylphenidate dataset (the latter noted as coming from traumatic brain injury patients rather than healthy controls). Sample sizes varied by dataset (examples from the extracted text: Western propofol n=16 after exclusions; Cambridge propofol retained 15; sevoflurane n=20; ketamine n=20; LSD n=15; ayahuasca n=9). For each subject and condition, BOLD timeseries were parcellated into the 100 regions, pairwise correlations computed, negative edges removed, and regional node strength (functional connectivity, FC) calculated. Drug effects were summarised as the within-subject change in regional FC strength between baseline and drug conditions, then averaged across subjects to yield one 100-region vector per pharmacological intervention. Preprocessing and denoising were standardised across datasets: anatomical CompCor (aCompCor) was used to regress out components from white matter and CSF, six motion parameters and their derivatives, and outlier volumes (scrubbing); denoised timeseries were detrended and band-pass filtered (0.008–0.09 Hz). The authors deliberately did not apply global signal regression, citing its impact on correlation structure and prior findings that the global signal carries information about states of consciousness. PET-derived receptor maps and the drug-induced FC change vectors were entered into a multivariate Partial Least Squares (PLS) correlation analysis to identify latent variables maximising covariance between neurotransmitter distributions (predictors, 100×18) and pharmacological FC changes (responses, 100×d). Statistical significance of PLS components was assessed against autocorrelation-preserving spin-based null models, and robustness was checked with a distance-dependent cross-validation procedure. To evaluate regional pairwise relationships, the investigators computed region-by-region matrices of pharmacological co-susceptibility (correlating drug-effect vectors across drugs) and neurotransmitter co-expression (correlating receptor/transporter profiles across PET maps). To mitigate spatial autocorrelation effects, an exponential trend with Euclidean distance was regressed out from these matrices prior to some correlation tests. Finally, the study compared drug/receptor score maps to several canonical cortical hierarchies—intracortical myelination (T1w/T2w), evolutionary cortical expansion, AHBA PC1 (principal component of gene expression), NeuroSynth PC1 (task-activation component), the principal gradient of functional connectivity, cerebral blood flow, and a redundancy-to-synergy information gradient—and also constructed a region-by-region ‘‘disorder co-susceptibility’’ matrix from ENIGMA cortical abnormality maps (11 disorders) for cross-modal comparisons. Diffusion map embedding was applied to fused pharmacological and disease co-susceptibility matrices to extract principal joint gradients.

At the pairwise level, regions that co-express similar neurotransmitter receptor and transporter profiles also tended to show similar susceptibility to drug-induced FC changes. Specifically, after regressing out the exponential relationship with Euclidean distance, pharmacological co-susceptibility correlated with neurotransmitter profile similarity with Spearman's rho = 0.26 (p < 0.001), indicating a modest but statistically robust association. The PLS multivariate analysis identified two statistically significant latent variables linking neurotransmitter distributions and pharmacologically-induced FC reorganisation. Both components were significant against spin-based nulls and cross-validated using a distance-dependent method, yielding out-of-sample correlations of r = 0.40 for PLS1 and r = 0.59 for PLS2 (both p < 0.001 versus spin-based null distributions). For PLS1, drug loadings separated anaesthetics from psychedelics and cognitive enhancers: anaesthetics loaded at one end of the axis, while psychedelics and cognitive enhancers loaded at the opposite end. Neurotransmitter loadings for PLS1 separated transporters (positive end: acetylcholine and noradrenaline transporters, with serotonin and dopamine transporters nearby) from receptors (negative end), producing a receptor-versus-transporter division that aligned with the drug grouping. PLS2 distinguished monoaminergic systems (dopamine and serotonin, except 5-HT1b) from other neurotransmitters in its neurotransmitter loadings, and its drug scores produced a clear dorsal–ventral cortical separation; however, drug loadings on PLS2 were less cleanly interpretable, with propofol appearing at both ends. Spatially, PLS1 score maps corresponded to known intrinsic resting-state networks, separating visual and somatomotor cortices from transmodal association regions. PLS1 (for both neurotransmitter and drug scores) correlated significantly with multiple canonical cortical hierarchies: intracortical myelin (T1w/T2w), the AHBA PC1 transcriptomic gradient (with the extraction noting the myelin and AHBA PC1 maps are reversed relative to others), NeuroSynth PC1, the principal functional connectivity gradient, cerebral blood flow, and the redundancy–synergy information gradient. When analysing individual drug patterns, nearly all drugs except methylphenidate showed significant correlations with one or more hierarchical gradients; anaesthetics consistently displayed similar patterns opposite to those of modafinil and psychedelics (ketamine, LSD, ayahuasca). PLS2 produced a dorsal–ventral score separation that did not significantly correlate with the canonical hierarchies tested. Extending beyond pharmacology, the authors constructed a region-by-region co-susceptibility matrix for 11 ENIGMA disorders and found that regional co-susceptibility to pharmacological perturbation correlated with co-susceptibility to disorder-related cortical abnormalities: Spearman's rho = 0.29 (p < 0.001) after regressing out Euclidean distance. Non-linear dimensionality reduction (diffusion map embedding) applied to fused pharmacological and disease co-susceptibility revealed two principal joint gradients: the first resembled the principal unimodal–transmodal functional gradient (similar to PLS1), and the second separated dorsal prefrontal from temporal regions (reminiscent of PLS2); together these two gradients explained nearly half of the variance in regional co-susceptibility. When diffusion embedding was applied to the pharmacological co-susceptibility matrix alone, the first two gradients aligned with previously reported functional connectivity gradients (unimodal–transmodal and a visual–somatomotor axis).

The investigators interpret their findings as evidence that psychoactive drugs operate not through single receptors alone but via coordinated engagement of multiple neurotransmitter systems, and that these multi-receptor contributions are topographically organised across the cortex. They emphasise the primary division uncovered by PLS1 between transporters and receptors, which maps onto a pharmacological contrast between anaesthetics and psychedelics/cognitive enhancers. Luppi and colleagues note that this receptor-versus-transporter axis accords with known drug actions and with prior reports that anaesthetics tend to reduce complexity and strengthen structure–function coupling, whereas psychedelics tend to increase complexity and loosen such coupling. A further key interpretation is that drug-induced functional reorganisation is organised along established cortical hierarchies: transmodal association cortices are especially implicated, plausibly because they exhibit high receptor diversity, elevated excitability, richer long-range connectivity, and relatively higher cerebral blood flow—all factors that could increase both drug availability and the likelihood that local perturbations reverberate widely. The authors propose that these neuroanatomical and molecular properties help explain why mind-altering drugs with powerful subjective effects disproportionately involve higher-order association regions. The discussion also highlights the observed relationship between pharmacological cosusceptibility and disease-related cortical vulnerability. The authors argue that regions structurally most vulnerable across a range of neuropsychiatric disorders are also those most susceptible to functional reorganisation by drugs, suggesting potential utility for mapping drug effects onto disorder-specific vulnerabilities. They frame the study as a first, data-driven step towards identifying candidate neurotransmitter targets for therapeutic intervention and for generating empirically testable hypotheses about which neurotransmitters underlie macroscale drug effects. The authors acknowledge multiple limitations and uncertainties that temper causal interpretation. The datasets are heterogeneous and not exhaustive: some drugs and radioligands are missing, PET maps and fMRI datasets come from different cohorts, and cortical coverage predominates over subcortical regions; one pharmacological dataset (methylphenidate) derived from TBI patients. Analytically, the results are correlational and based on linear models that assume independence despite the brain's complex, nonlinear connectivity; the authors therefore rely on conservative spin-based nulls, cross-validation, and triangulation to bolster robustness. They recommend future experimental manipulations and biologically informed whole-brain computational models to test causal hypotheses and to extend analyses to subcortex, brainstem, and cerebellum. Overall, the study provides a replicable computational framework linking PET-derived chemoarchitecture to macroscale functional consequences of pharmacological perturbation, while cautioning that experimental validation will be required to establish mechanistic causality.

The paper concludes that mapping regional changes in fMRI functional connectivity induced by a set of potent psychoactive drugs onto PET-derived maps of 18 neurotransmitter receptors and transporters reveals manifold, hierarchical patterns: multiple neurotransmitter systems contribute to drug effects; anaesthetics and psychedelics/cognitive enhancers exhibit largely opposite associations that align with cortical hierarchical gradients; and regional cosusceptibility to pharmacological perturbation recapitulates cosusceptibility to disorder-related structural abnormalities. The authors propose that this computational workflow could support data-driven prediction of the brain effects of candidate drugs and help identify neurotransmitter targets for therapeutic intervention, while noting that these results constitute a first, correlational step needing experimental follow-up.