Personalized brain circuit scores identify clinically distinct biotypes in depression and anxiety

Depression and anxiety disorders represent a significant global health burden, hindered by substantial etiological and phenotypic heterogeneity. Current diagnostic systems often fail to capture the complex interplay of neurobiological dysfunctions underlying these conditions, leading to suboptimal treatment outcomes. A significant proportion of patients do not respond to first-line treatments, highlighting the urgent need for a precision medicine approach. This approach requires standardized, personalized, and clinically interpretable metrics that quantify neurobiological dysfunctions within a neuroscientific framework. Existing biotyping efforts often rely solely on task-free fMRI or employ unsupervised methods with numerous features, potentially leading to overfitting. This study proposes a novel approach using a combination of task-free and task-evoked fMRI data to create personalized brain circuit scores and define clinically distinct biotypes in depression and anxiety, validated across multiple treatments.

Previous research has attempted to characterize biotypes in depression and anxiety using task-free fMRI, identifying patterns of aberrant connectivity in frontostriatal and limbic networks and within the default mode network. These studies have demonstrated associations between specific connectivity patterns and treatment response. However, these studies often lack the inclusion of task-evoked fMRI data, which offers valuable insight into the dynamic activity of brain circuits during cognitive and emotional processing. The current study addresses this gap by integrating both task-free and task-evoked data to achieve a more comprehensive understanding of brain circuit dysfunction.

This study utilized data from four independent studies: iSPOT-D, RAD, HCP-DES, and ENGAGE, encompassing a large sample (n=801) of unmedicated patients with depression and anxiety, and a subset (n=250) who participated in randomized controlled trials. The Stanford Et Cere Image Processing System was used to quantify task-free and task-evoked brain circuit function for each participant, yielding 41 measures across six circuits (default mode, salience, attention, negative affect, positive affect, and cognitive control). A hierarchical clustering algorithm was then applied to the personalized regional circuit scores (expressed as standard deviation units relative to a healthy control group) to identify distinct biotypes. The optimal number of clusters was determined using multiple validation procedures, including the elbow method, silhouette index testing (against null distributions and permutations), cross-validation, and comparison to a theoretical framework of circuit dysfunction. Biotype validation involved assessing differences in symptom severity, performance on cognitive tests, and treatment response across the identified biotypes using Mann-Whitney U-tests, Chi-square tests, and effect sizes.

The analyses revealed six distinct biotypes characterized by unique profiles of task-free and task-evoked brain circuit dysfunction. These biotypes were named based on the circuits and features distinguishing them from healthy controls and each other (e.g., D_CS_CA_C representing hyperconnectivity in the default mode, salience, and attention circuits). The six biotypes demonstrated significant differences in: 1. **Symptom severity**: Biotypes showed distinct patterns of symptom severity, including rumination, anhedonia, anxious arousal, and threat dysregulation. 2. **Behavioral performance**: Biotypes varied in performance on cognitive tasks assessing sustained attention, executive function, and emotional processing. 3. **Treatment response**: Biotypes exhibited differential responses to three types of antidepressants (escitalopram, sertraline, venlafaxine) and a behavioral intervention (I-CARE) compared to usual care. For example, the D_CS_CA_C biotype showed a better response to I-CARE, while the A_C biotype had a worse response. Biotype C_A showed a better response to venlafaxine. The findings demonstrated that the biotypes transcended traditional diagnostic boundaries, with the exception of a higher proportion of major depressive disorder in the A_C biotype and a lower proportion in the D_XS_XA_XN_XP_XC_X biotype. Furthermore, the combination of task-free and task-evoked features provided superior biotyping performance compared to using only task-free data or alternative feature sets from previous studies.

This study's findings significantly advance our understanding of the heterogeneity of depression and anxiety. The identification of six distinct biotypes, characterized by specific patterns of brain circuit dysfunction and validated across multiple clinical domains, supports the move towards a precision medicine approach in psychiatry. The inclusion of both task-free and task-evoked fMRI data proved crucial for achieving a more nuanced and comprehensive characterization of these biotypes, revealing dynamic patterns of brain activity that were not captured by previous studies using task-free data alone. The findings highlight the importance of considering multiple aspects of brain function when evaluating depression and anxiety, and the potential for these biotypes to inform personalized treatment strategies.

This study introduces a novel, theory-driven approach to biotyping in depression and anxiety using personalized brain circuit scores derived from a combination of task-free and task-evoked fMRI data. The identification and validation of six distinct biotypes, each with unique symptom profiles, behavioral characteristics, and treatment responses, offers a promising avenue for advancing precision clinical care in psychiatry. Future research should focus on validating these findings in larger, independent datasets and employing these biotypes to prospectively guide treatment selection. Further investigation is needed to explore the underlying mechanisms driving these biotypes and to develop more refined clinical measures to better capture the subtle differences between them.

While this study utilizes a large sample size and incorporates data from multiple studies, some limitations warrant consideration. The relatively small number of participants in some treatment subgroups may limit the generalizability of the treatment response findings. The use of specific fMRI tasks might also limit the generalizability of the findings to other contexts. Future studies should explore the replicability of these findings using different fMRI tasks and larger clinical trial samples. The effect sizes for symptom differences between biotypes were mostly small, highlighting the need for consistent measures and finer-grained clinical assessments.