The Acute Effects of the Atypical Dissociative Hallucinogen Salvinorin A on Functional Connectivity in the Human Brain

The investigators used a single-blind, placebo-controlled, within-subjects protocol. Each participant completed an unblinded practice session and a single-blinded scanning session at Johns Hopkins University. Participants were required to abstain from psychoactive drugs for 24 hours and provided a negative urine drug test before sessions; they were instructed to eat a low-fat breakfast and maintain usual caffeine intake. The inhalation procedure was practiced while participants lay supine with eyeshades and music; during the scanning session participants again wore eyeshades and MR-safe headphones. Placebo (hot air) and 15 μg/kg of inhaled SA were administered approximately 45 seconds after the start of two 20-minute functional scans, with placebo given in the first scan and SA in the second. Subjective strength ratings were collected during practice sessions but not during the scans. The extracted text implies a total sample of 12 participants (the classification leave-two-out scheme produced 66 combinations), all male and experienced hallucinogen users; the Methods do not clearly state the final sample size in a single sentence, but other procedural details are consistent with n = 12. Practice inhalations were repeated for some participants if necessary. Physiological measures such as electrocardiogram or respiration belts were not recorded; motion reduction and nuisance regression using principal components from white matter and cerebrospinal fluid were applied during preprocessing. For sFC, timeseries from 268 regions of the Shen atlas were correlated pairwise (35,778 edges), Fisher z-transformed, and analysed at whole-brain, within-network and between-network levels. Networks comprised eight Shen atlas networks (medial frontal, frontoparietal, default mode, subcortical-cerebellum, somatosensory-motor, medial visual, occipital pole, lateral visual). Edge-wise contrasts were thresholded using one-sample t-tests with a Bonferroni correction across all 35,778 edges for conservative inference. Dynamic functional connectivity was estimated using dynamic conditional correlations (DCC), a method that avoids arbitrary sliding-window artefacts; dFC was summarised as the variance of each edge's correlation timeseries. Entropic functional connectivity was calculated as an approximation to differential entropy from each edge's correlation timeseries using a histogram approach (results reported using 60 bins) to capture the number of qualitatively distinct connectivity states. Network-level measures averaged within- and between-network edges for each connectivity metric. To probe internal validity, the team used connectome-based classification via partial least squares discriminant analysis with leave-two-participant-out cross-validation: at each iteration models were trained on 10 participants and tested on the remaining two, classifying four conditions (placebo first half, placebo second half, SA first half, SA second half). Classification models were trained on whole-connectome sFC, dFC, and eFC separately and in combination, and separately on within-network and between-network connectomes. Time series were split into first and second halves because subjective drug strength decayed by about 50% by 10 minutes, so analyses used a 2 (drug) × 2 (time half) design.

Subjective effects during the practice session followed the expected rapid time course: peak strength at 1–2 minutes after inhalation, about a 50% reduction by 10 minutes, and largely subsiding by 15–20 minutes. This decay motivated splitting each 20-minute functional scan into first and second halves for neuroimaging analyses. Static functional connectivity: In the first half of scans, SA tended to reduce sFC within networks (7 of 8 networks showed numerical decreases) while increasing between-network sFC (21 of 28 network pairs showed numerical increases). The most robust attenuation was observed within the default mode network (DMN) during the first half. A moderate but non-significant correlation was reported between the change in DMN sFC from the first to second half of the SA scan and the area under the curve (AUC) of subjective drug strength over the first 20 minutes (r = -0.46, p = 0.132), suggesting those with longer subjective effects tended to have more sustained DMN attenuation, although this relationship did not reach significance. Applying a conservative Bonferroni threshold across all edges resulted predominantly in retained within-network (often bilateral homologous) connections; overall fewer static connections survived thresholding under SA than under placebo (first half: 350 vs 514 edges; second half: 405 vs 457), and more edges were significantly attenuated under SA than under placebo (first half: 161 vs 40; second half: 96 vs 16). Dynamic and entropic connectivity: Visual inspection and quantitative summaries indicated widespread reductions in dFC (variability of edge correlations) and widespread increases in eFC (entropy of edge correlations) under SA. Numerically, SA reduced dFC and increased eFC within and between most networks, but only the decreases in dFC during the first half of the scan survived corrections for multiple comparisons. Drug-by-time interactions for dFC changes were significant. Motion metrics correlated with dFC and eFC in some comparisons when motion correction was less aggressive; however, after more rigorous motion reduction these correlations were reduced and dFC effects strengthened. Changes in eFC were negatively correlated with motion, suggesting motion may have attenuated observed eFC increases. Connectome-based classification: Classifiers performed above chance (chance = 25%) and commonly confused temporally adjacent halves (placebo first vs second; SA first vs second) as expected. The first half of the SA scans was most distinct and most accurately classified, rarely mislabelled as placebo. Whole-brain models identified first-half SA at rates of 67% for sFC, 56% for dFC and 48% for eFC. Models trained on dFC tended to be more consistently accurate across conditions than sFC or eFC. Combining connectomes did not markedly improve classification. When training on within- or between-network connectomes, dFC models were substantially more accurate than sFC or eFC models for most network interactions. Notably, within-DMN sFC and eFC were among the most predictive static/entropic connectomes for identifying the first half of SA; dFC within the DMN also predicted SA and was not correlated with motion in those analyses.

Using fMRI, the investigators found that inhaled SA produces patterns of altered functional connectivity that resemble effects reported with classic psychedelics and some dissociative anaesthetics: a tendency to decrease sFC within resting-state networks, most prominently within the DMN, and to increase sFC between networks. Brainwide decreases in dFC under SA contrasted with some expectations about hallucinogen-induced increases in signal variability or entropy, but the authors note that prior studies have also observed decreases in variance with LSD and decreases in dFC with ketamine. SA tended to increase eFC across the brain, though those entropic changes did not survive strict multiple-comparison correction. Connectome-based classification highlighted the DMN as particularly informative for distinguishing SA from placebo, with within-DMN sFC and eFC among the better static/entropic predictors and dFC models showing strong predictive power. The authors emphasise that measures of variability (dFC) and entropy (eFC) capture different properties: an edge that swings between very strong and very weak states yields high variability but low entropy, whereas an edge fluctuating among several neighbouring strengths yields higher entropy without necessarily increasing variance. They suggest SA may reduce large abrupt shifts in connectivity while increasing the repertoire of distinct connectivity states. The paper acknowledges several limitations. A single-session, non-counterbalanced design may have allowed expectancy and order effects; physiological measures such as ECG or respiration were not recorded to directly control for cardiac or respiratory artefacts; participants were experienced hallucinogen users and the final sample comprised only males, limiting generalisability. The sample size was small (comparable to prior first-in-human psychedelic fMRI work), although scan durations were relatively long (20 minutes), which supports reliability of resting-state estimates. Analytically, the authors adopted conservative strategies—Bonferroni thresholding for edge-level contrasts, inclusion of all pairwise edges for network summaries, and internal validation via connectome classification—to avoid selective reporting. In terms of implications, the authors argue the similarity of SA effects to other classes of hallucinogens motivates direct comparative work to delineate shared versus unique neural mechanisms, and to explore structures such as the claustrum that are rich in receptors implicated across drug classes. They also note the potential translational relevance given preliminary therapeutic interest in κ-opioid agonists for mood and addiction disorders, while cautioning against strong reverse inferences about constructs like "ego dissolution" based solely on DMN changes given similar DMN effects with other psychoactive drugs and the heterogeneity of methods across studies.