Establishing Multimodal Brain Biomarkers for Treatment Selection in Depression

Our prior published studies found that sertraline outcome was predicted by biomarkers, primarily in frontoparietal control (FPCN), default mode (DMN), and ventral attention networks (VAN), from different single-modality neuroimaging data, including fMRI activation during emotional conflict regulation, resting EEG power signature, and resting fMRI connectivity. Our recent study with resting EEG connectivity defined two sertraline-predictive subtypes that showed convergent validity between EEG and fMRI. Therefore, the investigators hypothesize that modality-specific regional activity/connections in these networks exhibit similar subject-wise covariation that will jointly predict sertraline treatment response more precisely than either modality alone. However, the derivation of multimodal biomarkers remains highly challenging and underexplored for treatment selection in depression. The overall objective of this proposal is to establish multimodal brain biomarkers using data-driven analytics for treatment selection in depression. With multimodal data from EMBARC, a large publicly-available dataset, the investigators will devise advanced machine learning models to probe brain biomarkers jointly from multiple feature modalities including resting connectivity, task fMRI activation, and EEG band power. An independent cohort will be collected at Dell Medical School, The University of Texas at Austin (UT Austin) with methodology matching that utilized in the EMBARC study to produce a new sample of participants with independent data to validate these biomarker findings. To this end, the investigators will utilize an integrative analysis of both fMRI and EEG to: 1) identify moderators of sertraline versus placebo response in MDD; 2) quantify brain signatures that predict antidepressant treatment outcome; 3) recruit 50 depressed patients, non-invasively assess brain function with a combination of task and resting state fMRI and EEG prior to treatment initiation, administer/prescribe the common antidepressant medication, sertraline (tradename Zoloft), track symptom response over time, and utilize these data as an independent new cohort to optimize and validate brain biomarkers.

Aim 1: Identify multimodal brain moderators of antidepressant treatment effects of sertraline medication in major depressive disorder (MDD) using an existing, publicly-available database (EMBARC). Task 1.1: Using the EMBARC dataset, the investigators will design a canonical correlation analysis-based method to integrate fMRI and EEG to extract region-wise combined brain features. Task 1.2: The investigators will use linear mixed-effect models in a full intent-to-treat framework with the combined features to identify moderators of sertraline versus placebo. Task 1.3: The investigators will compare the statistical strength of the moderators identified between using multimodal features and each single-modality features, reveal neurobiological mechanisms underlying the multimodal moderators, and interpret their associations with depression-relevant clinical symptoms. This aim is crucial to establish novel multimodal brain moderators that can guide treatment selection in depression from a group-level perspective.

Aim 2: Quantify multimodal brain signatures that predict individual antidepressant treatment response to sertraline treatment in MDD using an existing, publicly-available database. Task 2.1: The Investigators will characterize multimodal brain signatures predictive of individual treatment outcomes using a machine learning tool that incorporates predictive modeling into our well-established latent space model with subtype guidance defined in our prior study. Task 2.2: The Investigators will compare multimodal signatures with single-modality and non-biological factors to confirm their efficacy in treatment outcome prediction and demonstrate their transferability to using EEG alone in clinical practice. Task 2.3: The investigators will investigate which brain regions/connections of different modalities are most critical to delineating the multimodal signatures by examining their associations with single-modality features, and interpret the neurocircuitry models in depression underlying the treatment response. These tasks are necessary for developing multimodal signatures of individual responses to antidepressant treatment for personalized medicine.

Aim 3: Optimize and validate multimodal brain biomarkers using new data collected in an independent, open-label clinical trial of sertraline treatment for MDD. Task 3.1: The investigators will recruit 50 individuals with MDD as an independent cohort and will adopt the EMBARC protocol design, including fMRI and EEG assessments at baseline followed by sertraline prescribed to these patients with clinical assessment of outcomes over 8 weeks. Task 3.2: The investigators will refine the multimodal brain signatures using an adaptive optimization strategy with the samples collected each year. Task 3.3: The investigators will perform an extensive validation using the new cohort to verify the multimodal biomarkers discovered in Aims 1-2. This aim will optimize our biomarker findings and provide strong evidence for their generalizability and reproducibility.

This project will establish informative multimodal biomarkers that can moderate clinical effects and predict individual responses to sertraline treatment, thereby providing a promising new avenue towards one of the first implementations in psychiatry of an objective test to inform treatment selection decisions. Our central hypothesis is that modality-specific regional brain activity/connections in FPCN, DMN, and VAN exhibit similar subject-wise covariation that can jointly predict sertraline treatment response more precisely than either modality alone.