Exploring Potential Biomarkers For Major Depressive Disorder Based On Multiple Omics And Machine Learning Analysis

Part 1 Identifying plasma biomarkers with high specificity for major depressive disorder: a multi-level proteomics studyBackground: There are currently no objective diagnostic biomarkers for major depressive disorder(MDD)due to the biological complexity of the disorder.The existence of blood-based biomarkers with high specificity would be convenient for the clinical diagnosis of MDD.Objective: To identify potential plasma biomarkers with high specificity for the diagnosis of MDD.Methods: A comprehensive plasma proteomic analysis was conducted in a highly homogeneous cohort [7 drug-na(?)ve MDD patients and 7 healthy controls(HCs)],with bioinformatics analysis combined with machine learning used to screen candidate proteins.Verification of reproducibility and specificity was conducted in independent cohorts [60 HCs and 74 MDD,42 schizophrenia(SZ)and 39 bipolar I disorder(BD-I)drug-na(?)ve patients].Furthermore,verification of consistency was accomplished by proteomic analysis of postmortem brain tissue from 16 MDD patients and 16 HCs.Results: The plasma proteomic analysis revealed 64 significantly changed proteins in MDD patients(fold change > 1.5 or < 0.67;P < 0.05),and among them,five candidate proteins were selected by bioinformatics analyses and a machine learning algorithm,including C-reactive protein(CRP),antithrombin III(ATIII),orosomucoid 2(ORM2),interalpha-trypsin inhibitor heavy chain 4(ITIH4),and vitamin D-binding protein(VDB).Compared with 60 HCs,levels of CRP,ATIII,ITIH4,and VDB were significantly higher in 74 MDD patients,both in the discovery cohort and independent replication cohort.In comparison with 42 SZ or 39 BD-I patients,two proteins(VDB and ITIH4)were significantly elevated only in 74 MDD patients.In addition,increased VDB and ITIH4 were observed consistently in both plasma and postmortem dorsolateral prefrontal cortex tissues of 16 MDD patients.Furthermore,a panel consisting of all four plasma proteins was able to distinguish MDD patients from HCs or SZ or BD-I patients with the highest accuracy.Conclusions: Plasma ITIH4 and VDB may be potential plasma biomarkers of MDD with high specificity.The four protein panel is more suitable as a potential clinical diagnostic marker for MDD.Part II Circular RNAs as potential blood biomarkers for the diagnosis and treatment of major depressive disorderBackground: Circular RNAs(circ RNAs)are expressed abundantly in the brain and are implicated in the pathophysiology of neuropsychiatric disease.However,the potential clinical value of circ RNAs in major depressive disorder(MDD)remains unclear.Objective: To identify and validate differentially expressed circ RNAs in MDD patients and to evaluate their potential as disease diagnostic biomarkers and novel therapeutic targets of MDD.Methods: RNA sequencing was conducted in whole-blood samples in a discovery set [7 highly homogeneous MDD patients and 7 matched healthy controls(HCs)].The differential expression of circ RNAs was primarily verified in original samples(the discovery set).Subsequently,the expression of candidate circ RNA indicators was secondly verified in an independent validation set(53 MDD patients and 52 HCs),and correlations of circ RNA expression with neuropsychological assessment,levels of serum brain-derived neurotrophic factor(BDNF)and the neuroimaging indicator were analyzed.The interventional study was conducted to assess the potential effect of the repetitive transcranial magnetic stimulation(r TMS)antidepressive treatment on the circ RNA expression.Results: In the discovery set,88 blood circ RNAs had significant difference between MDD and HC groups.15 circ RNAs with the |log2(fold change)| > 1 were selected for validation in original samples(the discovery set),and among them,4 circ RNAs showed significantly different between two groups(all P < 0.05)and were further verified in the independent validation set.In the independent validation set,compared with 52 HCs,significantly decreased circ FKBP8 levels(P = 0.001)and significantly elevated circ MBNL1 levels(P = 0.009)were observed in 53 MDD patients.Meanwhile,the expression of circ MBNL1 was negatively correlated with 24-item Hamilton Depression Scale(HAMD-24)scores in 53 MDD patients.Additionally,the mediation analysis indicated that circ MBNL1 affected HAMD-24 scores in 53 MDD patients through a mediator,serum brain-derived neurotrophic factor.Furthermore,in 53 MDD patients,the amplitude of low-frequency fluctuations in the right orbital part middle frontal gyrus was positively correlated with the expression of circ FKBP8(r = 0.359,P = 0.008)and circ MBNL1(r = 0.331,P = 0.015).In addition,the interventional study of 53 MDD patients demonstrated that antidepressive treatment partly increased circ FKBP8 expression and the change in expression of circ FKBP8 was predictive of further reduced HAMD-24 scoresConclusions: Whole-blood circ FKBP8 and circ MBNL1 may be potential biomarkers for the diagnosis of MDD,respectively,and circ FKBP8 may show great potential for the antidepressive treatment.Part III Multivariate machine learning analyses in identification of major depressive disorder using resting-state functional connectivity: a multi-central studyBackground: Diagnosis of major depressive disorder(MDD)using resting-state functional connectivity(rs-FC)data faces many challenges,such as the high dimensionality,small samples,and individual difference.Objective: To assess the potential clinical value of rs-FC and identify the optimal multivariate machine learning(ML)model for the diagnosis of MDD.Methods: Based on the rs-FC data,a progressive three-step ML analysis was performed,including six different ML algorithms and two dimension reduction methods,to investigate the classification performance of ML model in a multi-central,large samples dataset(1021 MDD patients and 1100 normal controls [NCs]).Furthermore,the linear least square fitted the regression model was used to assess the relationships between rs-FC features of the classification model and the severity of clinical symptoms in MDD patients.Results: Among used ML methods,the rs-FC model constructed by the e Xtreme Gradient Boosting(XGBoost)method showed the optimal classification performance for distinguishing MDD patients from NCs at the individual level(accuracy = 0.728,sensitivity = 0.720,specificity = 0.739,area under the curve = 0.831).Meanwhile,identified rs-FCs by the XGBoost model were primarily distributed within and between default mode network,limbic network,and visual network.More importantly,the individual Hamilton Depression Scale-17 items scores of MDD patients can be accurately predicted using rs-FC features identified by the XGBoost model(adjusted R2 = 0.180,root mean squared error = 0.946).Conclusions: The XGBoost model using rs-FCs showed the optimal classification performance between MDD patients and HCs,with the good generalization and neuroscientifical interpretability.