Brain-gut Microbiota Multimodal Predictive Model in Patients with Bipolar Depression

Caixi Xi,Ang Li,Jianbo Lai,Xiaojie Huang,Peifen Zhang,Su Yan,Mengfan Jiao,Huimin Huang,Shaohua Hu
DOI: https://doi.org/10.1016/j.jad.2022.11.026
IF: 6.533
2023-01-01
Journal of Affective Disorders
Abstract:BACKGROUND:The "microbiota-gut-brain axis" which bridges the brain and gut microbiota is involved in the pathological mechanisms of bipolar disorder (BD), but rare is known about the exact association patterns and the potential for clinical diagnosis and treatment outcome prediction.METHODS:At baseline, fecal samples and resting-state MRI data were collected from 103 BD depression patients and 39 healthy controls (HCs) for metagenomic sequencing and network-based functional connectivity (FC), grey matter volume (GMV) analyses. All patients then received 4-weeks quetiapine treatment and were further classified as responders and non-responders. Based on pre-treatment datasets, the correlation networks were established between gut microbiota and neuroimaging measures and the multimodal kernal combination support vector machine (SVM) classifiers were constructed to distinguish BD patients from HCs, and quetiapine responders from non-responders.RESULTS:The multi-modal pre-treatment characteristics of quetiapine responders, were closer to the HCs compared to non-responders. And the correlation network analyses found the substantial correlations existed in HC between the Anaerotruncus_ unclassified,Porphyromonas_asaccharolytica,Actinomyces_graevenitzii et al. and the functional connectomes involved default mode network (DMN),somatomotor (SM), visual, limbic and basal ganglia networks were disrupted in BD. Moreover, in terms of the multimodal classifier, it reached optimized area under curve (AUC-ROC) at 0.9517 when classified BD from HC, and also acquired 0.8292 discriminating quetiapine responders from non-responders, which consistently better than even using the best unique modality.LIMITATIONS:Lack post-treatment and external validation datasets; size of HCs is modest.CONCLUSIONS:Multi-modalities of combining pre-treatment gut microbiota with neuroimaging endophenotypes might be a superior approach for accurate diagnosis and quetiapine efficacy prediction in BD.
What problem does this paper attempt to address?