Antidepressants and Placebo Rewire the Brain Within Two Weeks, Machine Learning Study Reveals

One of the most frustrating realities in psychiatry is that antidepressants take weeks to work, and even then only for some patients. Clinicians prescribe a pill, wait, and hope, adjusting course through trial and error when the first attempt fails. A new study published in Nature Mental Health offers a way to see what is happening in the brain long before symptoms shift, and it suggests that the earliest neural consequences of treatment are both broader and more surprising than previously assumed. By tracking functional connectivity across nearly 400 patients with major depressive disorder, a research team led by Xiaoyu Tong and Yu Zhang of Stanford University School of Medicine has identified brain changes that appear within just one to two weeks of starting treatment, some shared by almost everyone who takes a pill and others that separate a true drug effect from the power of expectation.

The research drew on two of the largest biomarker studies in depression research: EMBARC, the Establishing Moderators and Biosignatures of Antidepressant Response in Clinical Care trial conducted in the United States, and CAN-BIND-1, the Canadian Biomarker Integration Network in Depression. Together, the cohorts included 386 patients aged 18 to 65, with 257 women and 129 men. Participants were randomly assigned to receive the selective serotonin reuptake inhibitors sertraline or escitalopram, or a placebo, with 123 patients on sertraline, 138 on escitalopram and 125 on placebo. Crucially, each patient underwent resting-state functional MRI scans both before treatment and again after one to two weeks of medication, allowing the researchers to compute how thousands of pairwise connectivity relationships between brain regions changed as treatment began.

Functional connectivity, measured as the temporal correlation of spontaneous blood-oxygen-level-dependent signal between brain areas, provides a window into the brain’s intrinsic organization without requiring patients to perform any task. The team faced a familiar obstacle in this kind of data: the signal-to-noise ratio of individual connectivity edges is low, and naive machine learning models tend to latch onto features that do not replicate. To address this, the researchers developed an innovative analytical strategy combining predictive and contrastive machine learning frameworks, constrained to connectivity changes with sufficient reliability. The contrastive component allowed them to statistically disentangle effects that are common to all treated patients from effects specific to drug or to placebo, a decomposition that has rarely been attempted at this scale.

The first major finding was a universal signature. Regardless of whether patients received sertraline, escitalopram or placebo, and regardless of whether their symptoms ultimately improved, nearly all medicated patients showed increased connectivity within a system linking visual cortex, the precuneus and the thalamus. The precuneus, a hub of the brain’s default mode network implicated in self-referential thought and memory, and the thalamus, the brain’s central relay station, are both known to interact with visual processing regions. Earlier work has shown that placebo treatment can alter primary visual cortex activity and connectivity, and this new result places those observations within a much larger and more systematic framework. The consistency of this visual-precuneus-thalamus change across two independent cohorts suggests it reflects a generalizable response to entering pharmacological treatment for depression, not a drug-specific mechanism.

By contrast, the neural correlates of placebo-related symptom improvement were centered elsewhere. The study found that striatal and attention networks mediated the placebo-driven reduction in depressive symptoms. The striatum, a subcortical structure central to reward processing and motivational learning, has repeatedly been implicated in placebo phenomena, including a well-known finding that reward-related ventral striatal activity distinguishes sertraline responders from placebo responders. The involvement of attention networks aligns with the psychological literature on expectation, which holds that placebo effects arise when anticipation and attention reorient the brain’s evaluative machinery. In practical terms, patients whose early connectivity shifts in these networks were pronounced were the ones whose symptoms improved most from expectation and context, independent of any pharmacological action.

The true drug effects were narrower and more selective than many researchers expected. Connectivity changes specific to sertraline and escitalopram converged on the amygdala, the midcingulate cortex, the orbitofrontal cortex and the cerebellum. These are regions with wellestablished roles in emotional regulation and depression: the amygdala generates threat and salience responses, the orbitofrontal cortex evaluates reward and punishment, the midcingulate cortex integrates motivation and control, and the cerebellum, long dismissed as purely motor, is increasingly recognized as a participant in cognitive and affective circuitry, with altered cerebellar-cerebral connectivity reliably distinguishing patients with depression. Notably, these drug-specific changes appeared in only a subset of the patients actually taking antidepressants. Pharmacological treatment, in other words, leaves a detectable early neural fingerprint in some brains but not in others, a neural reflection of the heterogeneity that has always frustrated clinicians.

The most clinically consequential finding emerged from that heterogeneity. When patients on sertraline did not show the drug-specific amygdala, midcingulate, orbitofrontal or cerebellar connectivity changes, their responses could be predicted using a model trained on placebo response signatures. This implies that a substantial share of what looks like antidepressant response in the clinic may actually be placebo response occurring in medicated patients. Given that systematic reviews of antidepressant trials have long documented substantial and growing placebo response rates, this study provides a mechanistic account of why: expectation engages striatal and attentional circuitry, and patients who are predisposed to engage that circuitry will improve whether or not the drug’s molecular mechanisms take hold in their emotional-regulation circuits.

Methodologically, the study’s strength lies in its cross-cohort validation. Patterns identified in one cohort were tested in the other, and the universal visual-precuneus-thalamus changes, the placebo mediators and the drug-specific effects all replicated across the EMBARC and CAN-BIND-1 datasets, which used different scanners, sites and clinical protocols. The analysis pipeline itself was rigorous, employing established preprocessing tools including fMRIPrep, boundary-based registration, ICA-based motion artifact removal and standard strategies to control the spurious correlations introduced by subject head motion. The predictive models were evaluated with cross-validated Pearson correlations between predicted and observed symptom change, and hyperparameters governing sparsity were tuned to avoid overfitting. The code was released publicly through Code Ocean, and the EMBARC data are available through the National Institute of Mental Health Data Archive, inviting independent scrutiny.

The implications for patient care are considerable. Today, deciding whether an antidepressant is working typically requires six to eight weeks of observation, and roughly half of patients discontinue treatment early, often because of side effects before any benefit arrives. If early connectivity changes measured after one week could be incorporated into an interactive treatment optimization framework, clinicians could potentially distinguish, within days, patients whose brains are responding to the drug’s pharmacology from patients whose trajectory depends on placebo-related circuitry, who might instead benefit from psychotherapy, neuromodulation or placebo-enhanced care strategies. This aligns with a broader movement in psychiatry toward biosignature-guided treatment, including prior work from overlapping research groups demonstrating that electroencephalographic signatures and structure-function covariation patterns can predict antidepressant response.

Important caveats remain. The study examined only two serotonergic medications over a brief window, and it remains unknown whether the same signatures generalize to other antidepressant classes, to longer treatment durations, or to adolescent and older populations. Resting-state fMRI measures indirect hemodynamic activity rather than neural firing, and even carefully denoised connectivity estimates carry residual uncertainty at the individual level. The authors also note that drug-specific connectivity changes were present in a subset rather than all medicated responders, so the absence of such changes does not guarantee nonresponse. Still, by systematically partitioning early brain changes into universal, placebo-mediated and drug-specific components, the study delivers what the field has lacked: a mechanistic map of how treatment for depression begins in the brain, weeks before the patient reports feeling better. It transforms the placebo from a statistical nuisance into a defined neural process, and it moves precision psychiatry a tangible step closer to the clinic.

Subject of Research: Early treatment-induced changes in brain functional connectivity in major depressive disorder following antidepressant or placebo administration

Article Title: Early brain functional connectivity changes induced by antidepressants and placebo

Article References: Tong, X., Fonzo, G. A., Carlisle, N. B., Xie, H., Berdichevsky, Y., Keller, C. J., Oathes, D. J., Nemeroff, C. B., Lin, F. V., & Zhang, Y. (2026). Early brain functional connectivity changes induced by antidepressants and placebo. Nature Mental Health. https://doi.org/10.1038/s44220-026-00729-y

Image Credits: AI Generated

DOI: 10.1038/s44220-026-00729-y

Keywords: major depressive disorder, functional connectivity, antidepressants, placebo effect, sertraline, escitalopram, machine learning, resting-state fMRI, EMBARC, CAN-BIND, biomarkers, precision psychiatry