Major depressive disorder: advancing treatment; missing translation – Nature Mental Health

Major depressive disorder (MDD) is among the most extensively characterized psychiatric conditions, yet clinical progress in this area has remained modest. Despite decades of research, pharmacological innovation in depression has been limited, and although psychological and other non-pharmacological interventions have advanced, this has not translatsqed into substantial gains in overall treatment outcomes, with only around half of patients responding to first-line care.

Credit: Marina Spence

This persistent gap reflects a deeper challenge. Depression is defined not by a single biological signature but instead by a range of symptoms, including low mood, irritability, cognitive impairment, sleep disturbance and reduced experience of pleasure, which vary substantially — both across individuals and across interindividual episodes.

Within this heterogeneity, certain symptoms stand out both for their prevalence and for their clinical importance. Anhedonia (reduced experience of pleasure) is one such feature. Nearly 70% of depressed individuals report experiencing anhedonia, which is closely linked to poorer outcomes yet still lacks targeted treatments.

Synthesizing two decades of work dedicated to understanding and measuring this core symptom of anhedonia in depression, the July 2026 issue of Nature Mental Health features an overarching Review by Diego Pizzagalli. Across species and experimental contexts, disturbances in reward learning can be quantified through the use of objective paradigms such as the probabilistic reward task, providing a rare example of a translationally consistent marker in depression research.

Approaches that probe the reward system have yielded important insights into the neurobiological mechanisms underlying anhedonia and have shown promise in the prediction of clinical outcomes, including treatment response and illness trajectory. Even so, their incorporation into routine clinical practice has been limited. As Pizzagalli notes, “After two decades of progress, the main barrier is no longer our ability to measure reward dysfunction — it is our ability to integrate those measures into clinical decision-making. The field has developed increasingly robust cross-species assays and demonstrated meaningful links with course and outcome; we now need to move to standardized implementation and prospective validation, particularly in the context of supporting personalized treatment and identification of individuals at risk for depression. Success in these areas will have the greatest impact on individuals and society.”

One step toward addressing this gap is provided by a prospective, biomarker-guided clinical trial by Zhukovsky et al., featured in this issue — among the first to test these principles in real-world antidepressant selection. As Pizzagalli, the study’s lead, reflects, “As we (and our patients) know, antidepressant treatment has been following a trial-and-error approach; building on prior biomarker studies, we felt it was imperative to perform a prospective study in which such biomarkers were used to guide treatment selection at the outset.” Incorporating reward-processing biobehavioral markers, cognitive control and neural connectivity as predictors of response to sertraline and bupropion, the authors then applied this model in an independent, prospective clinical trial to treat individuals with MDD. The study shows that such measures can prospectively stratify patients with MDD according to their likelihood of responding to treatment and provides support for the clinical relevance of biologically informed approaches.

Also included in this issue is an Analysis by Yan and colleagues that takes a complementary, population-level approach to understanding MDD. Drawing on harmonized neuroimaging data from around 12,000 individuals across 64 cohorts, the study identifies consistent reductions in cortical thickness in regions implicated in emotion regulation and sensory processing. This large-scale effort advances the definition of shared neurobiological features of depression across diverse populations. However, it also illustrates the complexity of the disorder, including the distributed and overlapping regions involved, which can be observed in both individuals with MDD and control participants. These findings deepen understanding of the underlying neurobiology, but translating such group-level patterns into tools that inform treatment decisions at the individual level remains challenging. The authors suggest that structural measures may be less useful as diagnostic tools and instead hold more promise as predictors of relapse, recurrence and treatment response.

Even when treatment is successful initially, many patients experience relapse after remission, and the course of MDD is often marked by recurrence. This persistent instability highlights the need for tools that can anticipate clinical trajectories and guide preventive care. A Comment by Winter and colleagues in this issue examines the specific issue of relapse prediction in depression — an area that has generated a growing number of models but little impact on routine care. Despite technical advances, predictive tools remain largely confined to research settings. The authors suggest that this reflects not only methodological limitations but also a misalignment of priorities, such as focusing too heavily on improving accuracy, often without equal attention to external validation, feasibility and clinical usefulness. Even models that perform well under controlled conditions may be difficult to implement or may not meaningfully improve decision-making. This reinforces a broader pattern across the field, in which increasingly sophisticated approaches have yet to deliver consistent benefits for patients.

Across levels of analysis — from behavioral assays of reward learning to biomarker-guided clinical trials to large-scale population studies — depression is increasingly yielding measurable and reproducible signals. Stratification is becoming feasible, and early attempts to guide treatment are beginning to produce meaningful gains. These advances, however, have yet to translate into comparable developments in clinical practice or patient outcomes. The central challenge has shifted: it is no longer simply about identifying biological signals but is about embedding them into clinical workflows that are scalable, interpretable and useful at the individual level. Closing this gap will require more than refinement of existing tools. It will depend on a sustained focus on implementation, ensuring that advances in measurement and prediction are converted into decisions that alter the course of illness and, ultimately, improve patients’ lives.