2026 ECNP Congress

PS04-4093

Leveraging large multi-cohort datasets to investigate the replicability of neuroimaging findings in depression

Janik Goltermann — L. Altegoer, D. Bedir, K. Berger, T. Kircher, N. Opel, U. Dannlowski · Charité Universitätsmedizin Berlin, Department of Psychiatry and Neuroscience- Campus Benjamin Franklin, Berlin, Germany

2026 ECNP Congress

Background

Concerns about the replicability of brain–psychopathology associations have intensified in mental health neuroimaging. Small effect sizes, limited statistical power, and high analytic flexibility challenge the credibility of reported findings [1–3]. Large-scale work suggests that robust and reproducible brain–behavior associations require substantially larger samples than typically used [4]. Despite this, systematic replication efforts remain rare, particularly for spatially distributed voxel-wise MRI findings, for which no consensus framework for defining replication success has been established.

Methods

Within the ReFiNe-MDD project (Replicability of Findings in Neuroimaging in Major Depressive Disorder), we assembled a large harmonized structural MRI database comprising >6,000 participants from three independent cohorts (MACS, MNC, BiDirect), including healthy controls (HC) and patients with major depressive disorder (MDD). The database integrates extensive phenotyping (>80 instruments) covering depressive symptoms, illness course, and environmental risk factors such as childhood maltreatment (CM). We used voxel-based morphometry to investigate gray matter associations and combined two approaches to investigate replicability: i) conceptual cross-cohort replications investigating spatial convergence of significant gray matter correlates of MDD and CM across the MACS, MNC and BiDirect cohort, and ii) a series of preregistered direct replications (i.e., copying original methods as closely as possible) of published voxel-wise findings of neural correlates of depression-related variables (e.g., diagnosis, comorbidities, course of disease, treatment effects, risk and protective factors). Successful replicability was defined using MRI-tailored spatial criteria, including peak voxel correspondence, spherical regions-of-interest, anatomically defined regions, and global spatial pattern overlap, as part of a novel MRI replication framework.

Results

MDD-related gray matter reductions were small (maximum partial R² = .01) but showed consistent spatial convergence across cohorts, particularly in insula, thalamus, and temporo-limbic regions. Cross-cohort overlap reached up to k = 10,217 voxels (p < .001), with convergence across all cohorts at liberal thresholds and in all pairwise cohort combinations at stringent significance thresholds (pFWE < .05). Notably, no evidence was found for replicable depression-associated alterations of the hippocampus. CM-related gray matter correlates also showed no evidence of replicability. This was consistent across over 100 analytic variations in a multiverse analysis framework. While the conduction of direct replications from findings from the literature is still ongoing, preliminary evidence indicates that replicability is limited. Across seven preregistered direct replications, only one (ECT pre–post effects) met multiple replication criteria. All other effects were substantially attenuated and spatially inconsistent relative to original reports.

Conclusions

Using a large harmonized database and a comprehensive replication framework, this study provides an empirical assessment of replicability in depression neuroimaging. Replicability is highly construct-dependent: while MDD-related effects are detectable and show spatial convergence across cohorts, CM-related associations and many published findings fail to replicate. MRI-tailored spatial criteria enable a more nuanced evaluation of replication success and highlight the need for systematic replication practices to establish robust and generalizable neurobiological findings.

References

1Button KS, Ioannidis JPA, Mokrysz C, et al. Power failure: Why small sample size undermines the reliability of neuroscience. Nat Rev Neurosci 2013; 14: 365–76. 2Poldrack RA, Baker CI, Durnez J, et al. Scanning the horizon: Towards transparent and reproducible neuroimaging research. Nat Rev Neurosci 2017; 18: 115–26. 3Botvinik-Nezer R, Holzmeister F, Camerer CF, et al. Variability in the analysis of a single neuroimaging dataset by many teams. Nature 2020; 582: 84–8. 4Marek S, Tervo-Clemmens B, Calabro FJ, et al. Reproducible brain-wide association studies require thousands of individuals. Nature 2022; 603: 654–60.