LSD modulates effective connectivity and neural adaptation mechanisms in an auditory oddball paradigm

Serotonergic psychedelics such as LSD act primarily at 5-HT2A receptors and are increasingly used to investigate the neural mechanisms of normal waking consciousness and psychopathology. Prior work suggests overlap between acute psychedelic states and early psychosis, with shared alterations in predictive coding mechanisms. The auditory oddball paradigm, which elicits a Mismatch Negativity (MMN) response to infrequent ‘deviant’ tones among repeated ‘standard’ tones, offers a well-established probe of perceptual learning and the balance between top-down predictions and bottom-up sensory signals. Timmermann and colleagues set out to characterise how LSD alters perceptual learning and effective connectivity within a fronto‑temporal network during an auditory oddball task recorded with magnetoencephalography (MEG). Using Dynamic Causal Modelling (DCM), the study tested whether LSD primarily modulates backward (top-down) extrinsic connections, intrinsic connectivity in primary auditory cortex (A1), or both, based on prior evidence that 5-HT2A receptors are located on neurons involved in backward projections and on earlier reports of psychedelic effects on effective connectivity.

Design and participants: Twenty healthy volunteers (16 male, 4 female) were recruited and screened for mental and physical health; exclusion criteria included diagnosed psychiatric illness, first‑degree family history of psychosis, recent psychedelic use, problematic alcohol use and medical contraindications. Each participant attended two sessions at least 14 days apart, receiving intravenous LSD on one day and placebo on the other in a counterbalanced, single‑blind design. The extracted text reports the LSD dose as "75 mg in a 10 ml solution administered over a period of 2 min"; this value is presented as extracted and the text does not clarify units further. Subjective effects emerged within 5–15 min, peaked around 60–90 min and plateaued for roughly 4 h. The auditory oddball task and MEG were performed at about 220 min post‑infusion. Stimuli and MEG acquisition: The paradigm was a replication of the ‘Optimum‑1’ mismatch negativity design. Standard harmonic tones (500, 1000, 1500 Hz; 75 ms) constituted 50% of trials, and five types of deviants (each 10% probability) altered frequency, duration, intensity, perceived direction or contained a brief gap. Stimulus‑onset asynchrony was 300 ms and 3015 trials were presented across three 5‑minute blocks. Whole‑head MEG was recorded with a 275‑channel CTF radial gradiometer system sampled at 1200 Hz (0–300 Hz bandpass), with additional reference channels and concurrent ECG and eye‑tracking. After artefact rejection the final analysed sample comprised 14 participants (one withdrew and five were excluded for excessive artefacts). Preprocessing and sensor analyses: MEG data were bandpass filtered at 0.5–35 Hz, visually inspected for gross artefacts, epoched −60 to 300 ms around stimulus onset, downsampled to 300 Hz and subject to semi‑automatic rejection and sensor interpolation. Robust averaging produced event‑related fields (ERFs) for standard and deviant tones under placebo and LSD. Sensor‑space ERFs were converted to scalp images (0–300 ms), smoothed and entered into a two‑factor ANOVA (Surprise: standard vs deviant; Drug: LSD vs placebo). Cluster‑level family‑wise error (FWE) correction (P < 0.05) with an uncorrected peak threshold of P < 0.001 was applied; results at a cluster‑forming threshold of P < 0.005 are also reported. Source reconstruction and DCM: Individual MRIs were used to compute leadfields and group source inversion employed the Multiple Sparse Priors method over a time window selected from the sensor ANOVA. Sources showing the main effect of Surprise were taken forward to DCM. The DCM network comprised five regions: bilateral primary auditory cortex (A1), bilateral superior temporal gyrus (STG) and right inferior frontal gyrus (IFG); subcortical input entered bilaterally at A1. The ERP neural mass model in DCM was used, modelling intrinsic connections among inhibitory interneurons, spiny stellate cells and pyramidal cells, and extrinsic forward/backward/lateral connections. DCMs used a 0–250 ms post‑stimulus window (with post‑onset at 64 ms) and sensor data were reduced to 8 principal dimensions for computational tractability. Model space and statistics: Two sets of model comparisons were pre‑specified. First, six models (combinations of forward/backward/both extrinsic modulations with presence/absence of intrinsic A1 modulation) tested the effects of deviant versus standard under placebo. Second, 36 models (all pairwise combinations of the six patterns for main effects of Surprise and Drug) tested how both deviance and LSD modulated connectivity. Models were inverted under a Variational Bayesian scheme; random‑effects Bayesian Model Selection (BMS) provided protected exceedance probabilities (pxp) and Bayesian Omnibus Risk (P0). After BMS, posterior connectivity parameter estimates from the winning model were subjected to classical t‑tests (Holm‑Bonferroni corrected) to test which connections were modulated by Surprise and by Drug.

Participants and trials: Of 20 recruited participants, 14 remained for analysis after attrition and artefact exclusions. Average numbers of trials retained were reported: LSD condition — 1178.07 standard trials (SD 117.32) and 234.57 deviant trials (SD 25.95); placebo condition — 1126.64 standard (SD 135.72) and 221.93 deviant (SD 32.43). Sensor‑level ERFs: A main effect of Surprise (standards vs deviants) produced widespread scalp negativity with four clusters peaking at approximately 50, 77, 137 and 143 ms post‑stimulus (all cluster‑level pFWE < 0.0002). Crucially, an interaction of Drug × Surprise was found in a predominantly right‑lateralised cluster spanning occipital, parietal and temporal sensors in the 57–127 ms window (peak at 67 ms; cluster‑level pFWE = 0.0075). Post‑hoc tests within this interaction mask indicated that under placebo the expected mismatch response (deviant > standard) was present as two components: an early component peaking at ~67 ms (right occipital) and a later component peaking at ~127 ms (right temporal). In contrast, the LSD condition showed a greatly attenuated mismatch response, with significance only in a small late cluster (123–127 ms). Direct contrasts of deviant tones (LSD vs placebo) revealed two large clusters of reduced deviant‑evoked negativity under LSD: an early effect at 60–93 ms peaking at 80 ms (right occipital; T(13) = 7.01; cluster pFWE = 9e‑4) and a later effect at 110–127 ms peaking at 117 ms (cluster pFWE = 0.005). Additionally, standard tones yielded a stronger negative component under LSD compared with placebo in a 103–110 ms window peaking at 107 ms (cluster pFWE = 0.01). Analyses using other deviant types produced marginal interaction effects for gap and duration deviants, with the gap deviant showing a replication of the pattern using a somewhat less conservative threshold. Source identification: Source reconstruction for the main effect of Surprise (50–220 ms window) implicated bilateral STG (left and right) and a right‑lateral cluster extending from precentral gyrus into right IFG; statistics for these source peaks were reported uncorrected (left STG F = 5.49, p = 0.024; right STG F = 6.00, p = 0.019; right IFG cluster F ≈ 3.91, p = 0.054 uncorrected). DCM model selection — placebo: Comparing six DCMs for the placebo condition yielded split evidence between two models: one with backward plus intrinsic A1 modulation (pxp = 0.37) and the full model with forward, backward and intrinsic modulation (pxp = 0.39). The combined pxp of 0.76 indicated these two models best explained the placebo data; Bayesian Omnibus Risk P0 = 0.0361. DCM model selection — LSD effect and parameter inference: Across the 36 models testing Surprise and Drug effects, the winning model included backward extrinsic and intrinsic A1 modulation for both Surprise and Drug. The protected exceedance probability for this model was 0.20, double that of the second best model (0.10); for most other models pxp ≤ 0.01 and Bayesian Omnibus Risk P0 = 0.2954. Classical inference on parameters from the winning model showed that deviant tones (main effect of Surprise) produced significant increases in intrinsic connectivity bilaterally in A1 (left: t(13) = 3.17, p = 0.03; right: t(13) = 8.24, p < 0.001). By contrast, the main effect of Drug (LSD versus placebo) produced a significant decrease in intrinsic connectivity in left A1 (t(13) = −3.27, p = 0.03) and a marginal decrease in right A1 (t(13) = −2.66, p = 0.07). For backward extrinsic connections, LSD produced a marginal decrease in connectivity from right IFG to right STG (t(13) = −2.37, p = 0.10) and a marginal increase in backward connectivity from left STG to left A1 (t(13) = 2.13, p = 0.10). An additional set of models that substituted interaction terms for main effects had lower model evidence and were not pursued further.

Timmermann and colleagues interpret the findings as indicating that LSD causes a focal dampening of the canonical mismatch response while enhancing an earlier negative response to standard tones. The combined sensor‑level and DCM results point to LSD‑related reductions in intrinsic connectivity within primary auditory cortex (A1) and marginal modulations of backward top‑down extrinsic connections within a right‑lateralised fronto‑temporal network. Under a predictive coding account, intrinsic A1 coupling is related to sensory memory formation and the precision of prediction errors; the authors argue that reduced intrinsic connectivity under LSD implies a weakened memory trace or adaptation for standard tones and a diminished precision afforded to prediction errors. Attenuation of backward connections, meanwhile, suggests reduced top‑down predictions arriving at sensory cortex. Together, these effects could impair recursive perceptual learning and increase uncertainty or ‘entropy’ in sensory inference. The authors relate their results to prior work on psychosis and anaesthetic/dissociative models. Reduced MMN amplitude and decreased intrinsic auditory connectivity have been documented in schizophrenia, and ketamine (an NMDA antagonist) has been shown to disrupt bottom‑up propagation of prediction error. The present pattern—chiefly modulation of top‑down connectivity and intrinsic auditory processing—aligns with the idea that serotonergic psychedelics may model aspects of psychosis dominated by positive symptoms, while NMDA antagonists may better model negative or different symptom clusters. The discussion also highlights an apparent overlap between findings here and studies of reduced conscious level, where diminished frontal‑to‑temporal connectivity has been reported; the authors caution that reduced top‑down connectivity may index altered sensitivity to the environment rather than a simple marker of conscious level. Limitations acknowledged by the study team include the relatively posterior localisation of the IFG source compared with prior MMN studies—a result that may reflect MEG sensitivity constraints, inter‑run sensor variability or the short inter‑stimulus interval used. Another caveat is timing: the oddball paradigm was administered at ~220 min post‑infusion, after the reported peak effects of LSD, which could have increased inter‑individual variability in neurophysiological and subjective effects. The authors note that their findings are specific to the oddball paradigm and may not generalise to spontaneous, task‑independent brain activity. Finally, the authors recommend future work to dissect which aspects of psychosis are modelled by different pharmacological manipulations and to include comparator drugs or states to clarify specificity. They emphasise the importance of rigorous, targeted designs to inform both mechanistic understanding and potential therapeutic implications of psychedelics.

The study concludes that LSD administration reduces the magnitude of the mismatch response in an auditory oddball paradigm and alters neural adaptation mechanisms at the scalp. Modelled effective connectivity changes indicate decreased intrinsic connectivity in primary auditory cortex and attenuated backward extrinsic connectivity within a fronto‑temporal network. These effects are interpreted as impairing local adaptation required for perceptual learning and reducing top‑down suppression of prediction error during auditory processing. The authors note partial correspondence with findings from ketamine models and schizophrenia, as well as overlap with states of reduced consciousness, and suggest that these similarities may reflect a shared alteration in sensitivity to environmental stimuli. They call for further work using targeted comparisons and specific symptom‑level hypotheses to resolve which mechanistic features are shared or distinct across these states.