Biomarker-Guided Antidepressant Selection for Treatment-Resistant Depression

Major depressive disorder is a leading cause of disability. Two FDA-approved treatments—and ketamine and accelerated repetitive transcranial magnetic stimulation targeting the left dorsolateral prefrontal cortex (DLPFC)—have emerged as highly effective alternatives to first-line antidepressants, capable of delivering

rapid antidepressant effects even for treatment-resistant depression (TRD) cases. Clinicians who are choosing between ketamine and rTMS usually rely on a trial-and-error approach that can take months. Biomarkers for informing this key clinical decision point have the potential to transform the management of TRD by rapidly

matching individual patients to the treatment most likely to benefit them. Leveraging recent technical advances and building on an extensive foundation of preliminary data establishing the feasibility of predicting accelerated rTMS and ketamine treatment outcomes, we propose a two-phase project aimed at developing, optimizing, and

testing a new approach to selecting the most effective treatment for individuals with TRD. Individual differences in treatment outcomes can be understood in part by considering the antidepressant mechanisms of action underlying these very different treatments and how they interact with neurobiological heterogeneity in TRD.

Our central hypothesis is that individual differences in antidepressant responses to rTMS and ketamine are due in part to heterogeneity in the neurobiology of depression. We hypothesize that whereas connectivity deficits involving the prefrontal cortex are a marker of TRD patients who are likely to respond to ketamine-

induced synaptogenesis, intact connectivity between the DLPFC stimulation site and downstream targets in the anterior cingulate and insula will be associated with enhanced rTMS responses. In the UG3 phase, we will use state-of-the-art machine learning methods to optimize statistical classifiers (“neuroimaging biomarkers”) for

predicting antidepressant responses in individual patients. We will enhance model performance by a) refining a subtyping procedure we developed; b) incorporating precision functional mapping of network topology; and c) optimizing for robust and reproducible results in held-out data. In parallel, we will validate our biomarker

approach in a prospective pilot study, validating the reliability, acceptability, and feasibility of fMRI biomarkers. In the UH3 phase, we evaluate the efficacy of this approach in a prospective clinical trial, randomizing to receive DLPFC-rTMS or ketamine, informed by our biomarkers, and evaluate their utility for supporting new,

highly scalable models for predicting outcomes without the need for fMRI data in a subset of individuals.