ABSTRACT Despite the prevalence and public health significance of depression, up to 40% of depressed adolescents do not respond to first-line antidepressants (i.e., serotonin selective reuptake inhibitors [SSRIs]). Adolescents with treatment non-response (TNR) are at high risk for physical and mental health difficulties associated with ineffectively treated depression, including cardiovascular disease and suicide. Thus, identifying the neurobiological mechanisms that underlie TNR in adolescents is a critical step toward optimizing treatment plans for those who do not respond to first-line treatments. In this context, sustained threat to social stressors, as measured by elevated inflammatory profiles to stressful stimuli, has been shown to drive the onset and maintenance of depression among adolescents and is associated with TNR. The mechanisms by which elevated inflammation impact the brain in depressed adolescents, however, are unclear. To address these gaps in our knowledge, we will test our central hypothesis that excessive glutamate (Glu) in depression-related corticolimbic circuits—including the anterior cingulate cortex, ventromedial prefrontal cortex, amygdala, and hippocampus—is a critical mediator between peripheral inflammation and TNR in depressed adolescents. Specifically, we will conduct a prospective 18-month study of 160 unmedicated treatment-seeking depressed adolescents (ages 14-18) using state-of-the-art multimodal neuroimaging data at 7 Tesla. At Time 1 (prior to SSRI treatment) and Time 2 (after an open-label 12-week SSRI trial), we will assess peripheral measures of pro-inflammatory cytokines and glutamate in corticolimbic circuits before and after a well-validated adolescent- version of the Trier Social Stress Test (TSST). We also will use a well-validated fMRI task designed to probe behavioral and neural responses to negative peer evaluation, a salient form of social threat for adolescents. At Time 1, we will test if TSST induces increases in inflammation and glutamate in corticolimbic circuits in unmedicated adolescents with depression. At Time 2, we will use machine learning methods to identify multi- level predictors of TNR based on behavioral, inflammatory, and neural indicators of sustained threat to social stress; we will also test whether glutamate in corticolimbic circuits mediates the association between baseline levels of inflammation and TNR. Finally, we will continue to clinically assess depression symptoms and collect information on social stressors (e.g., context, severity, duration) every 3 months for 15 months following Time 2 (i.e., from Time 3 to Time 7), which will enable us to use functional clustering analyses to identify subgroups of adolescents on the basis of depression trajectories (e.g., persistent depression, gradual remission, etc), and identify predictors of these subgroups and other related clinical outcomes (e.g., remission status), while accounting for the effects of TNR status and any change...