Resilience and Brain Changes in Long-Term Ayahuasca Users: Insights From Psychometric and fMRI Pattern Recognition

Ramos and colleagues situate their study within a growing literature on ayahuasca, an Amazonian psychedelic brew containing DMT and beta-carbolines, that has shown acute changes in perception, emotion and cognition and biochemical markers of neuroplasticity such as increased BDNF and sigma-1 receptor expression. Previous work has documented acute modulation of emotional circuitry (for example, default mode network and limbic regions including amygdala and insula), and recent neuroimaging studies of other psychedelics have used machine learning to reveal drug-related connectivity signatures and to predict individual clinical or behavioural traits. Against this background, the study aims to evaluate whether long-term, regular ayahuasca use is associated with lasting differences in emotional brain processing and psychological resilience when users are not acutely under the substance. Specifically, the investigators test whether experienced ayahuasca users differ from non-user controls on psychometric measures (notably the ER-89 resilience scale) and whether whole-brain fMRI patterns during an implicit emotional task can discriminate groups or predict individual resilience scores using multivariate pattern recognition methods.

This was a retrospective, cross-sectional case–control fMRI study of 38 male participants (19 long-term ayahuasca users, 19 non-user controls). Participants were recruited from two urban ayahuasca communities in Rio de Janeiro (users) and from university advertisements (controls). Inclusion criteria required male sex and age >18 years; for the ayahuasca group, lifetime experience of at least 36 ingestions within a one-year period. Exclusion criteria included history of psychiatric or neurological disease, use of medications or substances affecting the nervous or vascular system, MRI contraindications and recent major stress; ayahuasca users were excluded if they had ingested ayahuasca within one week of scanning. Four initially recruited users were excluded for MRI/task/movement issues, yielding the final N = 38 with age-matched groups (means ~31.5 and 33.1 years). Prior to scanning participants completed a battery of self-report instruments: PANAS (positive/negative affect), ER-89 (resilience trait), Beck Anxiety Inventory (BAI), Beck Depression Inventory (BDI) and the State–Trait Anxiety Inventory (STAI). During fMRI, subjects performed an implicit emotion processing task: a gender discrimination task on faces expressing neutral or aversive emotions (disgust and fear) presented in a block design across two runs. Faces were presented for 2 s each with short fixation intervals; total functional task time per run was 4.3 minutes and the full MRI session included structural T1 images. Imaging was acquired at 1.5 Tesla using gradient-echo planar T2*-weighted sequences. Preprocessing and first-level analyses used SPM12: motion and slice-timing correction, spatial smoothing (10 mm FWHM), high-pass temporal filtering (0.01 Hz) and normalization to MNI space. A GLM model created a regressor for aversive stimuli (fear and disgust) and included movement parameters as covariates. A custom mask excluded NaN voxels prior to pattern analysis. Machine learning analyses were conducted in PRoNTo using whole-brain voxels and AAL3 anatomical segmentation. For classification the primary algorithm was Multiple Kernel Learning (MKL) with k = 10 cross-validation (one subject per class held out per fold); Support Vector Machine (SVM) and Gaussian Process Classifier (GPC) were also applied with similar cross-validation. Permutation testing assessed significance (MKL n = 100 permutations; SVM/GPC n = 1000). For regression, Gaussian Process Regression (GPR) was used to predict psychometric scores from voxel patterns using a leave-one-subject-out cross-validation and 1000 permutations. Model performance metrics included accuracy, class-specific accuracy, AUC, correlation (r), R2 and mean square error (MSE). Group comparisons of psychometric scores used Student's t-test or Mann–Whitney U depending on normality.

On psychometric measures, the ayahuasca group had higher resilience scores on the ER-89 (mean = 43.89, SD = 5.64) than controls (mean = 39.05, SD = 5.34); the authors describe this difference as significant. No group differences were observed for PANAS positive affect (p = 0.72) or negative affect (p = 0.56), STAI trait (p = 0.30) or state (p = 0.17), Beck Anxiety Inventory (p = 0.35) or Beck Depression Inventory (p = 0.69). In whole-brain pattern classification, the MKL model (AAL3 atlas, k = 10) achieved a total accuracy of 75% and an AUC of 0.68. Class-specific accuracies were 65% for the ayahuasca-user class (reported p = 0.10) and 85% for the non-user class (reported as significant). The MKL weight map identified a set of regions that most contributed to group discrimination: parietal cortex, medial frontal cortex, limbic regions including amygdala, insula and anterior cingulate, occipital cortex, and basal ganglia including thalamic (and raphe) nuclei. The authors caution that weight maps indicate relative importance for classification and do not directly equate to local activation magnitudes. For regression, the GPR model relating whole-brain patterns to individual ER-89 resilience scores produced a correlation r = 0.69 (R2 = 0.50). Reported error metrics included MSE = 35.72% (p = 0.15) and a normalized MSE of 1.13 (p = 0.09) under permutation testing, which were not below the α = 0.05 threshold in the extracted text. The authors therefore report that brain patterns could predict resilience scores numerically (r = 0.69) but the permutation p-values provided do not meet the conventional significance threshold.

Ramos and colleagues interpret their findings as evidence that regular, long-term ayahuasca users exhibit a higher trait of psychological resilience and distinct patterns of emotional brain reactivity even when not acutely under the compound. The authors link the higher resilience to previously reported increases in openness and to studies showing no long‑term mental health deterioration among experienced users. They further argue that the ability of a regression model to recover individual resilience scores from fMRI patterns, coupled with group discrimination by MKL, suggests a neural signature of resilience and potential brain plasticity associated with chronic ayahuasca use. The discussion places emphasis on the anatomical regions that contributed most to the classification model. The authors highlight the amygdala, insula and anterior cingulate as central to emotional processing and previously implicated in acute ayahuasca effects, and they note medial frontal contributions related to executive and top‑down control over emotion. Parietal and occipital contributions are discussed in relation to sensorimotor integration, attention to emotional information and sensitivity to facial expressions; the basal ganglia and thalamic nuclei are described as part of limbic–cortical integration relevant to aversive social cues. They caution that MKL weights index discriminative importance rather than direct task-evoked activation. Several limitations are acknowledged. The sample was small and restricted to male participants, limiting generalisability and the power to detect univariate group differences using GLM. The authors note that small sample size constrained cross-validation options and recommend larger, better-controlled studies to permit region-specific analyses and stronger validation of machine learning models. Finally, they state that further work is needed to clarify short- and long-term effects of ayahuasca on brain function and behaviour and to explore causality.

The authors conclude that experienced regular users of ayahuasca show different emotional brain reactivity compared with non-users when scanned off-drug, alongside higher resilience scores. They propose that chronic ayahuasca use may modulate brain responses to aversive stimuli in a pattern associated with resilience and altered defensive responses, although they underscore the need for larger and more definitive studies to confirm and extend these observations.