Advertisement

Individualized Functional Connectome Identified Replicable Biomarkers for Dysphoric Symptoms in First-Episode Medication-Naïve Patients With Major Depressive Disorder

  • Youjin Zhao
    Affiliations
    Huaxi MR Research Center, Department of Radiology, West China Hospital of Sichuan University, Chengdu, Sichuan, China

    Functional and Molecular Imaging Key Laboratory of Sichuan Province, West China Hospital of Sichuan University, Chengdu, Sichuan, China

    Research Unit of Psychoradiology, Chinese Academy of Medical Sciences, Chengdu, Sichuan, China
    Search for articles by this author
  • Louisa Dahmani
    Affiliations
    Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Charlestown, Massachusetts
    Search for articles by this author
  • Meiling Li
    Affiliations
    Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Charlestown, Massachusetts
    Search for articles by this author
  • Yongbo Hu
    Affiliations
    Huaxi MR Research Center, Department of Radiology, West China Hospital of Sichuan University, Chengdu, Sichuan, China

    Functional and Molecular Imaging Key Laboratory of Sichuan Province, West China Hospital of Sichuan University, Chengdu, Sichuan, China

    Research Unit of Psychoradiology, Chinese Academy of Medical Sciences, Chengdu, Sichuan, China
    Search for articles by this author
  • Jianxun Ren
    Affiliations
    Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Charlestown, Massachusetts
    Search for articles by this author
  • Su Lui
    Affiliations
    Huaxi MR Research Center, Department of Radiology, West China Hospital of Sichuan University, Chengdu, Sichuan, China

    Functional and Molecular Imaging Key Laboratory of Sichuan Province, West China Hospital of Sichuan University, Chengdu, Sichuan, China

    Research Unit of Psychoradiology, Chinese Academy of Medical Sciences, Chengdu, Sichuan, China
    Search for articles by this author
  • Danhong Wang
    Affiliations
    Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Charlestown, Massachusetts
    Search for articles by this author
  • Weihong Kuang
    Affiliations
    Department of Psychiatry, West China Hospital of Sichuan University, Chengdu, Sichuan, China
    Search for articles by this author
  • Qiyong Gong
    Correspondence
    Qiyong Gong, M.D., Ph.D.
    Affiliations
    Huaxi MR Research Center, Department of Radiology, West China Hospital of Sichuan University, Chengdu, Sichuan, China

    Department of Radiology, West China Xiamen Hospital of Sichuan University, Xiamen, Fujian, China
    Search for articles by this author
  • Hesheng Liu
    Correspondence
    Address correspondence to Hesheng Liu, Ph.D.
    Affiliations
    Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Charlestown, Massachusetts

    Department of Neuroscience, Medical University of South Carolina, Charleston, South Carolina
    Search for articles by this author
Open AccessPublished:January 03, 2022DOI:https://doi.org/10.1016/j.bpsc.2021.12.010

      Abstract

      Background

      Major depressive disorder (MDD) is a heterogeneous syndrome and can be conceptualized as a mixture of dimensional abnormalities across several specific brain circuits. The neural underpinnings of different symptom dimensions in MDD are not well understood. We aimed to identify robust, generalizable, functional connectivity (FC)–based biomarkers for different symptom dimensions in MDD using individualized functional connectomes.

      Methods

      Patterns of FC associated with symptom severity were identified using a novel, individualized, functional network parcellation analysis in conjunction with hierarchical clustering. Dimension-specific prediction models were trained to estimate symptom severity in first-episode medication-naïve patients (discovery dataset, n = 95) and replicated in an independent validation dataset (n = 94). The correlation between FC changes and symptom changes was further explored in a treatment dataset (n = 55).

      Results

      Two distinct symptom clusters previously identified in patients with MDD, namely dysphoric and anxiosomatic clusters, were robustly replicated in our data. A connectivity biomarker associated with dysphoric symptoms was identified, which mainly involved the default, dorsal attention, and limbic networks. Critically, this brain-symptom association was confirmed in the validation dataset. Moreover, the marker also tracked dysphoric symptom improvement following a 2-week antidepressant treatment. For comparison, we repeated our analyses using a nonindividualized approach and failed to identify replicable brain-symptom biomarkers. Further quantitative analysis indicated that the generalizability of the connectivity-symptom association was hampered when functional regions were not localized in individuals.

      Conclusions

      This work reveals robust, replicable FC biomarkers for dysphoric symptoms in MDD, demonstrates the advantage of individual-oriented approach, and emphasizes the importance of independent validation in psychiatric neuroimaging analysis.

      Keywords

      Major depressive disorder (MDD) is a heterogeneous syndrome and has been increasingly conceptualized as a disconnection syndrome (
      • Korgaonkar M.S.
      • Goldstein-Piekarski A.N.
      • Fornito A.
      • Williams L.M.
      Intrinsic connectomes are a predictive biomarker of remission in major depressive disorder.
      ). Despite the impressive growth of neuroimaging studies, the underlying neural mechanisms of MDD remain obscure. The National Institute of Mental Health Research Domain Criteria project has developed a framework for psychiatric disorder research based on behavioral dimensions and neurobiological measures (
      • Cuthbert B.N.
      The RDoC framework: Facilitating transition from ICD/DSM to dimensional approaches that integrate neuroscience and psychopathology.
      ). Accordingly, the symptoms of patients with MDD can be considered as a mixture of dimensional abnormalities across several specific brain circuits. This is in keeping with a recent transcranial magnetic stimulation study, which found that dysphoric and anxiosomatic symptoms of patients with MDD were alleviated following stimulation of two distinct neuroanatomical treatment targets (
      • Siddiqi S.H.
      • Taylor S.F.
      • Cooke D.
      • Pascual-Leone A.
      • George M.S.
      • Fox M.D.
      Distinct symptom-specific treatment targets for circuit-based neuromodulation.
      ).
      Neuroimaging studies typically make group-level inferences, rendering it difficult to assess individual patients or develop personalized treatment plans. Thus, studies using machine learning methods have begun with a broad search for predictive models to make inferences at the individual level (
      • Arbabshirani M.R.
      • Plis S.
      • Sui J.
      • Calhoun V.D.
      Single subject prediction of brain disorders in neuroimaging: Promises and pitfalls.
      ,
      • Lebois L.A.M.
      • Li M.
      • Baker J.T.
      • Wolff J.D.
      • Wang D.
      • Lambros A.M.
      • et al.
      Large-scale functional brain network architecture changes associated with trauma-related dissociation.
      ,
      • Wang D.
      • Tian Y.
      • Li M.
      • Dahmani L.
      • Wei Q.
      • Bai T.
      • et al.
      Functional connectivity underpinnings of electroconvulsive therapy-induced memory impairments in patients with depression.
      ,
      • Brennan B.P.
      • Wang D.
      • Li M.
      • Perriello C.
      • Ren J.
      • Elias J.A.
      • et al.
      Use of an individual-level approach to identify cortical connectivity biomarkers in obsessive-compulsive disorder.
      ,
      • Liu H.
      • Liu W.J.
      • Wang D.
      • Dahmani L.
      Individual-specific analysis for psychoradiology.
      ,
      • Wang D.
      • Li M.
      • Wang M.
      • Schoeppe F.
      • Ren J.
      • Chen H.
      • et al.
      Individual-specific functional connectivity markers track dimensional and categorical features of psychotic illness [published correction appears in Mol Psychiatry 2020; 25:2200].
      ). However, most models are trained and tested on a single dataset, which can lead to an overestimation of performance due to overfitting, resulting in poor generalizability (
      • Woo C.W.
      • Chang L.J.
      • Lindquist M.A.
      • Wager T.D.
      Building better biomarkers: Brain models in translational neuroimaging.
      ). Similarly, studies with small sample sizes often yield overoptimistic results (
      • Scarpazza C.
      • Ha M.
      • Baecker L.
      • Garcia-Dias R.
      • Pinaya W.H.L.
      • Vieira S.
      • Mechelli A.
      Translating research findings into clinical practice: A systematic and critical review of neuroimaging-based clinical tools for brain disorders.
      ). Ideally, studies should seek to train models with larger sample sizes and assess the generalizability of the findings using previously unseen datasets.
      In addition, most neuroimaging studies have relied on applying group-level templates when assessing the functional organization of an individual’s brain. Accumulating evidence demonstrates that individuals vary in the layout of their brain’s functional networks, especially in the association cortices (
      • Gordon E.M.
      • Laumann T.O.
      • Gilmore A.W.
      • Newbold D.J.
      • Greene D.J.
      • Berg J.J.
      • et al.
      Precision functional mapping of individual human brains.
      ,
      • Kong R.
      • Li J.
      • Orban C.
      • Sabuncu M.R.
      • Liu H.
      • Schaefer A.
      • et al.
      Spatial topography of individual-specific cortical networks predicts human cognition, personality, and emotion [published correction appears in Cereb Cortex 2021; 31:3974].
      ,
      • Braga R.M.
      • Buckner R.L.
      Parallel interdigitated distributed networks within the individual estimated by intrinsic functional connectivity.
      ,
      • Gratton C.
      • Laumann T.O.
      • Nielsen A.N.
      • Greene D.J.
      • Gordon E.M.
      • Gilmore A.W.
      • et al.
      Functional brain networks are dominated by stable group and individual factors, not cognitive or daily variation.
      ,
      • Mueller S.
      • Wang D.
      • Fox M.D.
      • Yeo B.T.T.
      • Sepulcre J.
      • Sabuncu M.R.
      • et al.
      Individual variability in functional connectivity architecture of the human brain.
      ,
      • Wang D.
      • Buckner R.L.
      • Fox M.D.
      • Holt D.J.
      • Holmes A.J.
      • Stoecklein S.
      • et al.
      Parcellating cortical functional networks in individuals.
      ,
      • Sylvester C.M.
      • Yu Q.
      • Srivastava A.B.
      • Marek S.
      • Zheng A.
      • Alexopoulos D.
      • et al.
      Individual-specific functional connectivity of the amygdala: A substrate for precision psychiatry.
      ). Important brain network features that are evident within individuals can be absent from group-level atlases (
      • Kong R.
      • Li J.
      • Orban C.
      • Sabuncu M.R.
      • Liu H.
      • Schaefer A.
      • et al.
      Spatial topography of individual-specific cortical networks predicts human cognition, personality, and emotion [published correction appears in Cereb Cortex 2021; 31:3974].
      ,
      • Braga R.M.
      • Buckner R.L.
      Parallel interdigitated distributed networks within the individual estimated by intrinsic functional connectivity.
      ), presumably owing to the spatial blurring inherent in group-averaged data (
      • Buckner R.L.
      • DiNicola L.M.
      The brain’s default network: Updated anatomy, physiology and evolving insights.
      ). Moreover, brain-behavior relationships are stronger when brain networks are defined using individual features than when solely defined based on group-level templates (
      • Brennan B.P.
      • Wang D.
      • Li M.
      • Perriello C.
      • Ren J.
      • Elias J.A.
      • et al.
      Use of an individual-level approach to identify cortical connectivity biomarkers in obsessive-compulsive disorder.
      ,
      • Kong R.
      • Li J.
      • Orban C.
      • Sabuncu M.R.
      • Liu H.
      • Schaefer A.
      • et al.
      Spatial topography of individual-specific cortical networks predicts human cognition, personality, and emotion [published correction appears in Cereb Cortex 2021; 31:3974].
      ,
      • Li M.
      • Wang D.
      • Ren J.
      • Langs G.
      • Stoecklein S.
      • Brennan B.P.
      • et al.
      Performing group-level functional image analyses based on homologous functional regions mapped in individuals.
      ). To meet the challenge of advancing precision medicine, our group has developed individual-oriented approaches in resting-state functional connectivity (FC) to investigate functional network organization at the individual level, which holds great promise in clinical research (
      • Wang D.
      • Tian Y.
      • Li M.
      • Dahmani L.
      • Wei Q.
      • Bai T.
      • et al.
      Functional connectivity underpinnings of electroconvulsive therapy-induced memory impairments in patients with depression.
      ,
      • Wang D.
      • Buckner R.L.
      • Fox M.D.
      • Holt D.J.
      • Holmes A.J.
      • Stoecklein S.
      • et al.
      Parcellating cortical functional networks in individuals.
      ).
      Although there are a considerable number of neuroimaging studies that have aimed to find biomarkers that reflect the severity of clinical symptoms or cognitive decline in patients, the most important problem in this field is the substantial heterogeneity across findings and, relatedly, the lack of replicability of these neuroimaging results. Hence, the aim of our study was to find reliable biomarkers for distinct symptom dimensions in a relatively large sample of medication-naïve patients with first-episode MDD. We first mapped the fine-grained functional regions in each participant and constructed individualized functional connectomes to represent the participants’ network organization. Next, in a discovery dataset (n = 95), 28 MDD symptom measures were clustered according to their relationship with brain connectivity, i.e., symptoms that appeared to have shared connectivity underpinnings were grouped together, yielding several symptom clusters. We then trained machine learning models to predict the symptom cluster scores in the discovery dataset and tested the model in a previously unseen dataset (n = 94) to assess its generalizability. We further explored the relationship between FC changes and symptom changes following a 2-week antidepressant treatment. Finally, we performed parallel analyses using a group-level template as a benchmark to assess whether the individualized approach facilitated the discovery of neural biomarkers.

      Methods and Materials

      Participants

      Patients with DSM-IV–diagnosed MDD (N = 194) (
      • First M.
      • Spitzer R.
      • Gibbon M.
      • Williams J.
      Structured Clinical Interview for DSM-IV Axis I Disorders (SCID).
      ) were recruited from the Mental Health Centre of West China Hospital from November 2010 to November 2018. Inclusion criteria were 1) age 18 to approximately 65 years, 2) first episode, 3) did not receive psychiatric medication treatment earlier, 4) total Hamilton Depression Rating Scale scores >7, and 5) did not exhibit psychiatric comorbidities, such as anxiety disorders. Data from 189 participants were retained after quality control of the magnetic resonance imaging (MRI) data and divided into two matched groups: a discovery dataset (n = 95) and a validation dataset (n = 94). In addition, we assessed a group of patients (n = 55) before and after 2 weeks of antidepressant treatment (see Supplemental Methods for more details on participant information). Of the 55 patients, 41 patients were from the first-episode medication-naïve group, and the remaining 14 patients were recruited additionally (see Table 1 for demographic and clinical characteristics and Table S1 for details on psychopharmacology). In this experiment, we evaluated brain changes after a treatment of only 2 weeks. There is evidence that early response could be a clinical turning point in antidepressant treatment. A meta-analysis of 41 single- or double-blind clinical trials showed that improvement in the first 2 weeks of treatment is a predictor of stable response and remission in patients with MDD (
      • Szegedi A.
      • Jansen W.T.
      • van Willigenburg A.P.P.
      • van der Meulen E.
      • Stassen H.H.
      • Thase M.E.
      Early improvement in the first 2 weeks as a predictor of treatment outcome in patients with major depressive disorder: A meta-analysis including 6562 patients.
      ). Knowing whether a treatment is effective early on could increase the speed and rate of overall treatment response (
      • Widge A.S.
      • Bilge M.T.
      • Montana R.
      • Chang W.
      • Rodriguez C.I.
      • Deckersbach T.
      • et al.
      Electroencephalographic biomarkers for treatment response prediction in major depressive illness: A meta-analysis.
      ,
      • Spies M.
      • Kraus C.
      • Geissberger N.
      • Auer B.
      • Klöbl M.
      • Tik M.
      • et al.
      Default mode network deactivation during emotion processing predicts early antidepressant response.
      ,
      • Pei C.
      • Sun Y.
      • Zhu J.
      • Wang X.
      • Zhang Y.
      • Zhang S.
      • et al.
      Ensemble learning for early-response prediction of antidepressant treatment in major depressive disorder.
      ). This study was approved by the ethics committee of Sichuan University, China. Written informed consent was obtained from each participant.
      Table 1Demographic and Clinical Characteristics of Patients With MDD at Baseline
      CharacteristicDiscovery Dataset, n = 95Validation Dataset, n = 94Follow-up Dataset, n = 55F2 (df)p Value
      p values were obtained using the analysis of variance except where noted.
      Sex, F/M, n
      p values were obtained using the χ2 test.
      57/3857/3730/250.595 (2).743
      Age, Years31.568 (11.027)32.053 (11.375)30.818 (10.513)0.217 (2,241).805
      Education, Years13.800 (2.938)13.585 (3.384)13.418 (3.083)0.272 (2,241).762
      Duration, Weeks30.629 (30.193)37.955 (70.823)39.745 (66.424)0.551 (2,241).577
      HAMD24.358 (6.567)24.511 (5.570)23.164 (7.338)0.855 (2,241).427
      HAMA23.021 (9.604)22.638 (8.561)23.127 (9.655)0.063 (2,241).939
      FD0.104 (0.054)0.102 (0.059)0.089 (0.043)1.405 (2,241).247
      Handedness, Right/Left, n
      p values were obtained using the χ2 test.
      91/3
      Data for 1 subject were missing.
      92/255/05.670 (8).684
      Medications, n
      Multiple entries per patient possible.
       SSRI44
       SNRI7
       BZD38
       Antipsychotics21
       Tricyclic antidepressants3
       Mood stabilizers1
       NaSSA3
       5-HT1A receptor agonist2
       Others2
       NA4
      Values are presented as mean (SD) unless indicated otherwise.
      BZD, benzodiazepine; F, female; FD, framewise displacement; HAMA, Hamilton Anxiety Rating Scale; HAMD, Hamilton Depression Rating Scale; M, male; MDD, major depressive disorder; NA, not available; NaSSA, noradrenergic and specific serotonergic antidepressant; SNRI, serotonin-norepinephrine reuptake inhibitor; SSRI, selective serotonin reuptake inhibitor.
      a p values were obtained using the analysis of variance except where noted.
      b p values were obtained using the χ2 test.
      c Data for 1 subject were missing.
      d Multiple entries per patient possible.

      MRI Data Acquisition and Preprocessing

      High-resolution T1 and resting-state functional MRI (fMRI) data were acquired from each participant using a 3.0T MR scanner (Siemens Trio). MRI data were processed using procedures previously described (
      • Buckner R.L.
      • Krienen F.M.
      • Castellanos A.
      • Diaz J.C.
      • Yeo B.T.T.
      The organization of the human cerebellum estimated by intrinsic functional connectivity.
      ,
      • Yeo B.T.T.
      • Krienen F.M.
      • Sepulcre J.
      • Sabuncu M.R.
      • Lashkari D.
      • Hollinshead M.
      • et al.
      The organization of the human cerebral cortex estimated by intrinsic functional connectivity.
      ), including discarding the first four volumes, slice timing correction, motion correction, bandpass filtering (0.01–0.1 Hz), motion regression, whole-brain signal regression, and ventricular and white matter regression (see Supplemental Methods).

      Constructing Individualized Functional Connectomes

      First, a group-level parcellation with 92 functional clusters was created using data from the Genomic Superstruct Project (n = 1000) (
      • Holmes A.J.
      • Hollinshead M.O.
      • O’Keefe T.M.
      • Petrov V.I.
      • Fariello G.R.
      • Wald L.L.
      • et al.
      Brain Genomics Superstruct Project initial data release with structural, functional, and behavioral measures.
      ) (see Supplemental Methods). Second, individual-level parcellations of each lobe (prefrontal, temporal, parietal, occipital, and sensorimotor cortices) were performed by applying an iterative clustering approach similar to our previous whole-brain parcellation strategy (
      • Wang D.
      • Li M.
      • Wang M.
      • Schoeppe F.
      • Ren J.
      • Chen H.
      • et al.
      Individual-specific functional connectivity markers track dimensional and categorical features of psychotic illness [published correction appears in Mol Psychiatry 2020; 25:2200].
      ,
      • Wang D.
      • Buckner R.L.
      • Fox M.D.
      • Holt D.J.
      • Holmes A.J.
      • Stoecklein S.
      • et al.
      Parcellating cortical functional networks in individuals.
      ). Briefly, the group-level functional atlas was projected onto each individual’s cerebral cortex, after which the boundaries of the functional clusters were gradually refined using an iterative approach with a weighting strategy to minimize the impact of the group-level functional atlas and maximize the usage of the individuals’ data. Finally, an FC matrix (92 × 92) was constructed for each individual by calculating connectivity among the 92 functional regions.

      Clustering of Symptoms

      Correlations were computed between 28 symptom items (see Supplemental Methods) and the individualized FC matrices across subjects after controlling for age, sex, education, and head motion. Then, the 28 FC-symptom correlation matrices were converted to a normal distribution using Fisher’s r to z transform. We adopted the method by Siddiqi et al. (
      • Siddiqi S.H.
      • Taylor S.F.
      • Cooke D.
      • Pascual-Leone A.
      • George M.S.
      • Fox M.D.
      Distinct symptom-specific treatment targets for circuit-based neuromodulation.
      ) to classify FC-symptom correlation matrices into clusters (
      • Murtagh F.
      • Legendre P.
      Ward’s hierarchical agglomerative clustering method: Which algorithms implement Ward’s criterion?.
      ), i.e., symptoms that had shared connectivity substrates were clustered together. The optimal cluster solution was determined according to reliability, as measured by the Dice coefficient (see Supplemental Methods).

      Estimating Symptom Scores Using Connectivity

      We constructed symptom prediction models using data from the discovery dataset. To determine the model parameters, we performed the following analyses:
      • 1.
        For each symptom cluster that included multiple symptom measures, FC-symptom correlation matrices were averaged across all symptom items, yielding a mean FC-symptom matrix.
      • 2.
        We then performed nonparametric permutation tests to identify connections that were reliably associated with the symptom cluster. Specifically, if the symptom cluster involved n symptom items, we then randomly selected n FC-symptom matrices from the 28 FC-symptom matrices and calculated the mean matrix. This randomization was repeated 1000 times. For each functional connection, a permutation p value was then calculated as the percentage of permutation tests that yielded a higher absolute FC-symptom correlation value than the value obtained in step 1. Connections with p < .001 were kept as reliable features.
      • 3.
        Correlations between the feature connections were identified in step 2 and symptom cluster scores were used as weights. Feature connections and their weights were extracted, yielding a prediction model for the symptom.
      We tested the prediction models on the validation dataset. Specifically, for each subject, we first obtained the individualized functional connectome and then calculated the weighted sum of FC values according to the symptom prediction model defined in the discovery dataset. The weighted sum was considered the predicted symptom score of the subject. Finally, correlations between the predicted and the actual symptom cluster score were calculated, with age, sex, education, and head motion as covariates. These tests were performed without prespecified hypotheses; thus, Bonferroni adjustments were applied.
      To quantify the contributions of different functional networks to symptom estimation, the 92 functional regions were grouped into the seven canonical large-scale functional networks: the visual network, sensorimotor network, dorsal attention network (DAN), salience network, limbic (LMB) network, frontoparietal control network, and default network (DN) (
      • Yeo B.T.T.
      • Krienen F.M.
      • Sepulcre J.
      • Sabuncu M.R.
      • Lashkari D.
      • Hollinshead M.
      • et al.
      The organization of the human cerebral cortex estimated by intrinsic functional connectivity.
      ). For each network, within-network and between-network weights were estimated (see Supplemental Methods). We further calculated the correlation between the average FC changes and cluster symptom changes following the 2-week antidepressant treatment, with age, sex, education, baseline head motion, and follow-up head motion as covariates.

      Control Analyses

      As a comparison, we generated FC matrices of 92 functional regions as defined by the group-level network atlas and repeated the prediction analyses. Correlations between feature connections and measured symptom cluster scores based on individualized versus atlas-based data were compared (two-sample t test, p < .05).
      To examine the impact of anatomical characteristics on the symptom biomarkers, we calculated cortical thickness, sulcus depth, cortical volume, cortical area, and curvature values and used these as covariates in the prediction model (see Supplemental Methods). Finally, to determine whether our results were affected by head motion, correlations between head motion and measured and estimated symptom scores were also calculated.

      Results

      Two Symptom Clusters, Dysphoric and Anxiosomatic, Can Be Reliably Identified

      To parse neural mechanisms underlying distinct symptom dimensions, 28 symptom measures were clustered according to their relationship with brain connectivity among 92 subject-specific brain regions (
      • Murtagh F.
      • Legendre P.
      Ward’s hierarchical agglomerative clustering method: Which algorithms implement Ward’s criterion?.
      ). A two-cluster solution exhibited the highest test-retest reliability, as measured by the Dice coefficient (Figure 1; Figure S1 and Table S2). One cluster involved nine symptoms, including depressed mood, decreased interest, and suicidality; the other cluster involved 19 symptoms, including somatic anxiety, somatic general, anxious mood, and insomnia. These two clusters were largely consistent with previous findings by Siddiqi et al. (
      • Siddiqi S.H.
      • Taylor S.F.
      • Cooke D.
      • Pascual-Leone A.
      • George M.S.
      • Fox M.D.
      Distinct symptom-specific treatment targets for circuit-based neuromodulation.
      ) and were referred to as dysphoric and anxiosomatic symptom clusters, respectively.
      Figure thumbnail gr1
      Figure 1Hierarchical clustering of functional connectivity (FC)–symptom matrices. For each subject, an individualized FC matrix was created by calculating the correlation between the time series of 92 individualized functional regions generated using a fine-grained functional parcellation. For all subjects, 28 FC-symptom matrices were created by correlating their individualized FC matrices with their scores on 28 items assessing various major depressive disorder symptoms. The distance matrix between each pair of FC-symptom matrices was calculated and used for Ward’s hierarchical clustering analysis. In the clustering procedure, FC-symptom matrices were grouped according to similarity between pairs of matrices (shorter line length in the dendrogram indicates greater similarity). The names of symptom items are colored according to their cluster assignment. HAMA, Hamilton Anxiety Rating Scale; HAMD, Hamilton Depression Rating Scale.

      Biomarkers for Dysphoric Symptoms Are Reliable

      To determine whether the individually specified FC tracked symptom severity, we trained the models to estimate the symptom cluster scores in the discovery dataset (n = 95). We found that dysphoric and anxiosomatic symptoms were associated with a set of functional connections, with significant correlations between estimated and observed scores (dysphoric symptoms: r = 0.559, p < .001; anxiosomatic symptoms: r = 0.621, p < .001) (Figure S2A, B). We then examined these associations in the validation dataset (n = 94) and found a significant correlation between the estimated and observed scores for dysphoric symptoms (r = 0.345, p < .001) (Figure 2A and Figure S2C) but not for anxiosomatic symptoms (r = −0.040, p = .708) (Figure S2D).
      Figure thumbnail gr2
      Figure 2The prediction model derived from the discovery dataset can predict the dysphoric symptom severity in the validation dataset. (A) Correlation (r = 0.345, p < .001) between the dysphoric symptom scores that were estimated by individualized functional connections and the scores that were actually observed in the 94 patients of the validation dataset. (B) The top 20 connections that contributed the most to estimation of dysphoric symptom scores are shown. (C) A total of 26 functional regions that are involved in these top 20 connections are shown. These nodes are color-coded according to the seven canonical networks. DAN, dorsal attention network; DN, default network; FPN, frontoparietal control network; LMB, limbic network; MOT, sensorimotor network; SAL, salience network; VIS, visual network.
      Thus, only the dysphoric symptom model, which involved 80 functional connections, was robust in the previously unseen data. The connections contributing to dysphoric symptom estimation were mainly between-network connections involving the DN, DAN, and LMB networks (Figure S3). The top 20 connections that contributed to the estimation involved a set of brain regions, including the orbitofrontal cortex (OFC), parahippocampal cortex, dorsomedial prefrontal cortex, anterior cingulate cortex (ACC), posterior cingulate cortex, premotor cortex, and superior parietal lobule (SPL) (Figure 2B, C).

      FC Changes Track With Dysphoric Symptom Changes Following Treatment

      After a 2-week antidepressant treatment, both the dysphoric and anxiosomatic symptoms were alleviated (Table S3). Focusing on the connections that were selected as features in the prediction model, we sought to determine whether changes in these connections following treatment could track the alleviation in dysphoric symptoms. We found a significant correlation between FC changes and dysphoric symptom changes (r = 0.301, p = .033) (Figure 3), indicating the relevance of these connections in symptom alleviation. However, there was no significant correlation between FC changes and anxiosomatic symptom changes (r = 0.210, p = .144) (Figure S4).
      Figure thumbnail gr3
      Figure 3Functional connectivity (FC) changes among individually specified functional clusters track with dysphoric symptom alleviation after a 2-week antidepressant treatment (r = 0.301, p = .033).

      An Atlas-Based Functional Connectome Model Failed to Predict Symptom Severity

      Our prediction model was based on FC measured among subject-specific functional regions. As a comparison, we repeated our analyses using the 92 functional clusters extracted from the group-level functional network atlas. As in our initial analysis, the 28 symptoms could be clustered into dysphoric and anxiosomatic symptoms according to symptom-FC correlations (Table S2). The prediction models constructed from atlas-based FC were able to estimate dysphoric symptoms (r = 0.556, p < .001) and anxiosomatic symptoms (r = 0.575, p < .001) in the discovery dataset. However, the models showed weak generalizability and could not be confirmed in the validation dataset after Bonferroni correction (dysphoric symptoms: r = 0.233, p = .027; anxiosomatic symptoms: r = −0.090, p = .399) (Figure 4A, B).
      Figure thumbnail gr4
      Figure 4The prediction models based on atlas-based functional connectomes were unable to estimate (A) dysphoric and (B) anxiosomatic symptom scores in the validation dataset. (C) In the discovery dataset, the individualized and atlas-based connections showed no significant difference in their correlations with dysphoric symptom scores (t = 0.212, p = .833, Hedges’ g = 0.033). However, in the validation dataset, the atlas-based connections showed weaker correlations with dysphoric symptoms when compared with individualized connections (t = 2.132, p = .035, Hedges’ g = 0.330). Error bars denote standard deviations. ∗ indicates a statistically significant difference (p < .05).
      In the discovery dataset, the individualized and atlas-based connections showed no significant differences in the correlations between feature connections and measured dysphoric symptoms (t164 = 0.212, p = .833, Hedges’ g = 0.033) (Figure 4C). However, in the validation dataset, these individualized connections showed significantly higher correlations to dysphoric symptoms when compared with atlas-based connections (t164 = 2.132, p = .035, Hedges’ g = 0.330) (Figure 4C), indicating that brain-symptom associations identified in a dataset cannot be easily generalized to an independent dataset if the functional features are not individually specified.
      To ensure that various confounding factors did not significantly affect our results, we repeated our analyses with nine factors as covariates: age, sex, education, head motion, and five anatomical measurements (cortical thickness, sulcus depth, cortical volume, cortical area, and curvature). We also calculated the correlations between head motion and symptom scores. The prediction results did not change after adding these covariates (Table S4), and both the measured symptom scores and individualized biomarker scores were uncorrelated with head motion (p > .1) in both the discovery and validation datasets (Figures S5 and S6).

      Discussion

      This study investigated the functional network underpinnings of different symptom dimensions in medication-naïve patients with first-episode MDD. To properly account for the interindividual variability in functional brain organization, we used a novel individual-oriented approach to examine functional connectomes. Consistent with a previous report, we were able to identify two distinct symptom clusters for MDD, i.e., dysphoric and anxiosomatic symptom clusters. We found a robust imaging biomarker for dysphoric symptoms, while anxiosomatic symptom estimation was unreliable. Our data suggest that MDD-related dysphoric symptoms are associated with altered connectivity among broad networks, mainly involving the DN, DAN, and LMB networks. In addition, FC changes in these networks tracked dysphoric symptom alleviation following a 2-week antidepressant treatment. Critically, a parallel analysis with atlas-based functional connectomes, which ignored interindividual variability in FC, failed to identify replicable biomarkers for either symptom dimension.

      Cortical Connectivity Biomarkers of Dysphoric Symptoms

      The dysphoric symptom cluster included depressed mood, decreased interest or pleasure (i.e., anhedonia), and suicidality, which are often referred to as core symptoms of depression (
      • Kennedy S.H.
      Core symptoms of major depressive disorder: Relevance to diagnosis and treatment.
      ). The most predictive connections for dysphoric symptoms involved several regions, including the OFC, parahippocampal cortex, dorsomedial prefrontal cortex, ACC, posterior cingulate cortex, premotor cortex, and SPL. The OFC and ACC are two critical regions in emotion regulation and cognition (
      • Heilbronner S.R.
      • Hayden B.Y.
      Dorsal anterior cingulate cortex: A bottom-up view.
      ,
      • Wikenheiser A.M.
      • Schoenbaum G.
      Over the river, through the woods: Cognitive maps in the hippocampus and orbitofrontal cortex.
      ,
      • Rudebeck P.H.
      • Saunders R.C.
      • Prescott A.T.
      • Chau L.S.
      • Murray E.A.
      Prefrontal mechanisms of behavioral flexibility, emotion regulation and value updating.
      ,
      • Rushworth M.F.S.
      • Behrens T.E.J.
      Choice, uncertainty and value in prefrontal and cingulate cortex.
      ) and have shown a variety of alterations in gross morphology, neuronal structure, function, connectivity, and neurochemistry in MDD [for a review, see (
      • Treadway M.T.
      • Zald D.H.
      Reconsidering anhedonia in depression: Lessons from translational neuroscience.
      )]. Across a variety of reward processing task studies, anhedonia has been associated with hypoactivation of the OFC and ACC in patients with depression (
      • Treadway M.T.
      • Zald D.H.
      Reconsidering anhedonia in depression: Lessons from translational neuroscience.
      ,
      • Keedwell P.A.
      • Andrew C.
      • Williams S.C.R.
      • Brammer M.J.
      • Phillips M.L.
      The neural correlates of anhedonia in major depressive disorder.
      ,
      • Borsini A.
      • Wallis A.S.
      • Zunszain P.
      • Pariante C.M.
      • Kempton M.J.
      Characterizing anhedonia: A systematic review of neuroimaging across the subtypes of reward processing deficits in depression.
      ,
      • Williams L.M.
      Precision psychiatry: A neural circuit taxonomy for depression and anxiety.
      ). Structural and functional alterations in the OFC and ACC have also been found in suicidal patients (
      • Zhang H.
      • Chen Z.
      • Jia Z.
      • Gong Q.
      Dysfunction of neural circuitry in depressive patients with suicidal behaviors: A review of structural and functional neuroimaging studies.
      ,
      • Monkul E.S.
      • Hatch J.P.
      • Nicoletti M.A.
      • Spence S.
      • Brambilla P.
      • Lacerda A.L.T.
      • et al.
      Fronto-limbic brain structures in suicidal and non-suicidal female patients with major depressive disorder.
      ,
      • Jollant F.
      • Lawrence N.S.
      • Giampietro V.
      • Brammer M.J.
      • Fullana M.A.
      • Drapier D.
      • et al.
      Orbitofrontal cortex response to angry faces in men with histories of suicide attempts.
      ), such as smaller OFC gray matter volume (
      • Monkul E.S.
      • Hatch J.P.
      • Nicoletti M.A.
      • Spence S.
      • Brambilla P.
      • Lacerda A.L.T.
      • et al.
      Fronto-limbic brain structures in suicidal and non-suicidal female patients with major depressive disorder.
      ) and greater activity in OFC and ACC in response to angry and happy versus neutral faces (
      • Jollant F.
      • Lawrence N.S.
      • Giampietro V.
      • Brammer M.J.
      • Fullana M.A.
      • Drapier D.
      • et al.
      Orbitofrontal cortex response to angry faces in men with histories of suicide attempts.
      ). In addition, the ACC, dorsomedial prefrontal cortex, posterior cingulate cortex, and parahippocampal cortex are implicated in self-referential functions (
      • Zhou H.X.
      • Chen X.
      • Shen Y.Q.
      • Li L.
      • Chen N.X.
      • Zhu Z.C.
      • et al.
      Rumination and the default mode network: Meta-analysis of brain imaging studies and implications for depression.
      ), which have been linked to intensification of depressed mood (
      • Stroebe M.
      • Boelen P.A.
      • van den Hout M.
      • Stroebe W.
      • Salemink E.
      • van den Bout J.
      Ruminative coping as avoidance: A reinterpretation of its function in adjustment to bereavement.
      ). Neural responses to happy faces in the premotor cortex were found to be negatively correlated with anhedonia in patients with depression (
      • Henderson S.E.
      • Vallejo A.I.
      • Ely B.A.
      • Kang G.
      • Krain Roy A.
      • Pine D.S.
      • et al.
      The neural correlates of emotional face-processing in adolescent depression: A dimensional approach focusing on anhedonia and illness severity.
      ). Patients with MDD also showed reduced gray matter volume in the SPL (
      • Lu S.
      • Xu R.
      • Cao J.
      • Yin Y.
      • Gao W.
      • Wang D.
      • et al.
      The left dorsolateral prefrontal cortex volume is reduced in adults reporting childhood trauma independent of depression diagnosis.
      ) and increased FC between ACC and SPL (
      • Rolls E.T.
      • Cheng W.
      • Gong W.
      • Qiu J.
      • Zhou C.
      • Zhang J.
      • et al.
      Functional connectivity of the anterior cingulate cortex in depression and in health.
      ). Taken together, dysfunctions within and between the structures mentioned above are associated with disturbances in emotional regulation and cognition in depression, lending credence to their relevance in the prediction of dysphoric symptoms in patients with MDD.
      It is worth noting that the core regions involved in our dysphoric symptom model (Figure 2C) are not exactly the same as the dysphoric symptom circuit found by Siddiqi et al. (
      • Siddiqi S.H.
      • Taylor S.F.
      • Cooke D.
      • Pascual-Leone A.
      • George M.S.
      • Fox M.D.
      Distinct symptom-specific treatment targets for circuit-based neuromodulation.
      ). Approximately 70% of our core regions (i.e., DAN, salience network, sensorimotor network, visual network, and frontoparietal control network) are located in the positive areas in Siddiqi’s circuit, and the remaining 30% of regions (i.e., DN and LMB network) are in their negative areas. However, our results are not contradictory. Siddiqi’s circuit is a general pattern of symptom-specific treatment target (i.e., transcranial magnetic stimulation in the positive areas could alleviate the corresponding symptoms), but it does not pinpoint specific connection changes related to symptom alleviation. Symptom-related connections may involve regions outside Siddiqi’s circuit through anticorrelations. Our work aimed to identify specific connections that are reliably associated with the symptom cluster, and our dysphoric symptom model is based on 80 specific functional connections (Figure S7). In addition, our study and the study by Siddiqi et al. used different sample characteristics: for example, Siddiqi et al. included treatment-resistant patients with depression, while we only included medication-naïve individuals. Such discrepancies may also account for the differences in our results. In contrast, our anxiosomatic symptom model was relatively unreliable. There are at least two possible explanations for this. The first is that we used a high standard for reliability; we required that our models be replicated with a previously unseen dataset. The second potential explanation is that the patients with MDD in our study did not exhibit psychiatric comorbidities, such as anxiety disorders; therefore, they had an overrepresentation of dysphoric symptoms compared with anxiosomatic symptoms (t = 9.961, p < .001) (Figure S8). Further research using a transdiagnostic approach is warranted.

      Abnormal Connectivity Between Functional Networks in MDD

      From the network perspective, we found that connections among the DN, DAN, and LMB network (Figure S3) contributed heavily to the prediction of dysphoric symptom severity, suggesting that dysfunctional interactions between multiple higher-order association networks are associated with dysphoric symptoms. The DN is responsible for internal states (
      • Raichle M.E.
      The brain’s default mode network.
      ,
      • Dixon M.L.
      • De La Vega A.
      • Mills C.
      • Andrews-Hanna J.
      • Spreng R.N.
      • Cole M.W.
      • Christoff K.
      Heterogeneity within the frontoparietal control network and its relationship to the default and dorsal attention networks [published correction appears in Proc Natl Acad Sci U S A 2018; 115:E3068].
      ,
      • Buckner R.L.
      • Andrews-Hanna J.R.
      • Schacter D.L.
      The brain’s default network: Anatomy, function, and relevance to disease.
      ) and has been associated with higher levels of maladaptive rumination about depressive thoughts in patients with MDD (
      • Williams L.M.
      Precision psychiatry: A neural circuit taxonomy for depression and anxiety.
      ,
      • Yan C.G.
      • Chen X.
      • Li L.
      • Castellanos F.X.
      • Bai T.J.
      • Bo Q.J.
      • et al.
      Reduced default mode network functional connectivity in patients with recurrent major depressive disorder.
      ). In contrast, the DAN is involved in external awareness and top-down attention processing (
      • Dixon M.L.
      • De La Vega A.
      • Mills C.
      • Andrews-Hanna J.
      • Spreng R.N.
      • Cole M.W.
      • Christoff K.
      Heterogeneity within the frontoparietal control network and its relationship to the default and dorsal attention networks [published correction appears in Proc Natl Acad Sci U S A 2018; 115:E3068].
      ). Neuroanatomical and functional alterations in the DAN have been shown to be related to some aspects of emotional, visceral, and autonomic dysregulation in patients with MDD (
      • Sheline Y.I.
      • Price J.L.
      • Yan Z.
      • Mintun M.A.
      Resting-state functional MRI in depression unmasks increased connectivity between networks via the dorsal nexus.
      ,
      • Zhao Y.
      • Chen L.
      • Zhang W.
      • Xiao Y.
      • Shah C.
      • Zhu H.
      • et al.
      Gray matter abnormalities in non-comorbid medication-naive patients with major depressive disorder or social anxiety disorder.
      ). The LMB network plays a crucial role in the encoding and retrieval of emotional memory (
      • de Voogd L.D.
      • Fernández G.
      • Hermans E.J.
      Awake reactivation of emotional memory traces through hippocampal-neocortical interactions.
      ) as well as in emotion regulation (
      • Groenewold N.A.
      • Opmeer E.M.
      • de Jonge P.
      • Aleman A.
      • Costafreda S.G.
      Emotional valence modulates brain functional abnormalities in depression: Evidence from a meta-analysis of fMRI studies.
      ). Limbic structures have frequently been implicated in the onset and course of depression (
      • Price J.L.
      • Drevets W.C.
      Neural circuits underlying the pathophysiology of mood disorders.
      ,
      • Barber A.D.
      • Sarpal D.K.
      • John M.
      • Fales C.L.
      • Mostofsky S.H.
      • Malhotra A.K.
      • et al.
      Age-normative pathways of striatal connectivity related to clinical symptoms in the general population.
      ). Focusing on the symptom-related connections, we found that average FC changes over 2 weeks of antidepressant treatment were positively correlated with alleviation of dysphoric symptoms. These results imply that these feature connections are reliably associated with dysphoric symptoms because they are associated with baseline dysphoria in addition to tracking the changes in symptom severity over time. Taken together, abnormal interactions among the DN, DAN, and LMB network could be associated with disruptions in the balance between internal states, external awareness, and emotion regulation, which may reflect biases toward ruminative thoughts at the cost of attending to the external world (
      • Kaiser R.H.
      • Andrews-Hanna J.R.
      • Wager T.D.
      • Pizzagalli D.A.
      Large-scale network dysfunction in major depressive disorder: A meta-analysis of resting-state functional connectivity.
      ) and potentially cause dysphoric symptoms in MDD.

      Interindividual Differences in the Functional Organization of the Brain Are Critical

      To examine the generalizability of our findings on brain-symptom associations, we showed that FC defined using individually specified functional regions was predictive of dysphoric symptom severity in both the discovery and validation datasets. However, the prediction models using connectivity derived from a group-level atlas obscured the biological signals and failed to predict any dimension-specific symptom severity in previously unseen datasets (Figure 4A, B). Further analyses indicated that these atlas-based connections were less correlated with symptom scores when compared with individualized connections in previously unseen data (Figure 4C). Thus, applying a group-level atlas to define functional regions within individual subjects, which is the standard procedure in most fMRI studies, likely yields results that are unstable or that may not be replicated in independent datasets.

      Limitations

      First, our brain-symptom association analyses were based on rating scales that are inherently subjective. Our hypothesis is that there are some biological bases for certain symptom dimensions, although the symptom measures in these dimensions are subjective and coarse. This study identified a replicable FC marker for dysphoric symptoms, suggesting that there are some objective brain signatures associated with the subjective rating of symptoms. Should a more objective symptom measure be developed in the future, we will be able to improve the prediction model and identify biologically more meaningful markers. Second, a reliable technique for mapping subcortical functional networks in individual participants is not yet available, precluding our ability to investigate subcortical connectivity in this study. However, we plan to adapt our individual-level approach to examine corticostriatal connectivity in future studies, given the pivotal role of these circuits in psychiatric disorders (
      • Janes A.C.
      • Zegel M.
      • Ohashi K.
      • Betts J.
      • Molokotos E.
      • Olson D.
      • et al.
      Nicotine normalizes cortico-striatal connectivity in non-smoking individuals with major depressive disorder.
      ). Third, our analyses did not include healthy control subjects because symptom scores have little to no variation within healthy subjects, and therefore including them may strongly bias the symptom estimation models. Fourth, to maximize the potential benefits to patients, we did not include a placebo control for the 2-week antidepressant treatment group; thus, these early effects may include a placebo component. Fifth, although our individualized parcellation has been partially validated by electrical cortical stimulation, the current gold standard for preoperative functional mapping, and task fMRI data, the accuracy of individualized parcellation would inevitably be influenced by the quality of resting-state fMRI data.

      Conclusions

      Using individualized functional connectomes, we identified replicable imaging biomarkers for dysphoric symptomatology in medication-naïve patients with first-episode MDD. Changes in these connections tracked with improvements in dysphoric symptom severity after treatment. Taken together, this study highlights the importance of accounting for interindividual variability in functional brain organization, which facilitates the discovery of robust, replicable neural biomarkers in comparison with typical atlas-based approaches.

      Acknowledgments and Disclosures

      This study was supported by the National Natural Science Foundation of China (Grant Nos. 81790652 [to HL], 81790650 [to HL], 81621003 [to QG] , 81820108018 [to QG] , 81761128023 [to QG], 82027808 [to QG], and 82001795 [to YZ] ), the National Institutes of Health/National Institute of Mental Health (Grant Nos. 1R01DC017991 [to HL], R01MH045573 [to HL], P50DA046373 [to HL], 5K01MH111802 [to DW], and R01MH112189-01 [to QG]), China Postdoctoral Science Foundation (Grant No. 2020M673245 [to YZ] ), the Post-Doctor Research Project of West China Hospital of Sichuan University (Grant No. 2021HXBH025 [to YZ] ), and a Canadian Institutes of Health Research postdoctoral fellowship (Grant No. MFE-171291 [to LD]).
      The authors report no biomedical financial interests or potential conflicts of interest.

      Supplementary Material

      References

        • Korgaonkar M.S.
        • Goldstein-Piekarski A.N.
        • Fornito A.
        • Williams L.M.
        Intrinsic connectomes are a predictive biomarker of remission in major depressive disorder.
        Mol Psychiatry. 2020; 25: 1537-1549
        • Cuthbert B.N.
        The RDoC framework: Facilitating transition from ICD/DSM to dimensional approaches that integrate neuroscience and psychopathology.
        World Psychiatry. 2014; 13: 28-35
        • Siddiqi S.H.
        • Taylor S.F.
        • Cooke D.
        • Pascual-Leone A.
        • George M.S.
        • Fox M.D.
        Distinct symptom-specific treatment targets for circuit-based neuromodulation.
        Am J Psychiatry. 2020; 177: 435-446
        • Arbabshirani M.R.
        • Plis S.
        • Sui J.
        • Calhoun V.D.
        Single subject prediction of brain disorders in neuroimaging: Promises and pitfalls.
        Neuroimage. 2017; 145: 137-165
        • Lebois L.A.M.
        • Li M.
        • Baker J.T.
        • Wolff J.D.
        • Wang D.
        • Lambros A.M.
        • et al.
        Large-scale functional brain network architecture changes associated with trauma-related dissociation.
        Am J Psychiatry. 2021; 178: 165-173
        • Wang D.
        • Tian Y.
        • Li M.
        • Dahmani L.
        • Wei Q.
        • Bai T.
        • et al.
        Functional connectivity underpinnings of electroconvulsive therapy-induced memory impairments in patients with depression.
        Neuropsychopharmacology. 2020; 45: 1579-1587
        • Brennan B.P.
        • Wang D.
        • Li M.
        • Perriello C.
        • Ren J.
        • Elias J.A.
        • et al.
        Use of an individual-level approach to identify cortical connectivity biomarkers in obsessive-compulsive disorder.
        Biol Psychiatry Cogn Neurosci Neuroimaging. 2019; 4: 27-38
        • Liu H.
        • Liu W.J.
        • Wang D.
        • Dahmani L.
        Individual-specific analysis for psychoradiology.
        Neuroimaging Clin N Am. 2020; 30: 45-51
        • Wang D.
        • Li M.
        • Wang M.
        • Schoeppe F.
        • Ren J.
        • Chen H.
        • et al.
        Individual-specific functional connectivity markers track dimensional and categorical features of psychotic illness [published correction appears in Mol Psychiatry 2020; 25:2200].
        Mol Psychiatry. 2020; 25: 2119-2129
        • Woo C.W.
        • Chang L.J.
        • Lindquist M.A.
        • Wager T.D.
        Building better biomarkers: Brain models in translational neuroimaging.
        Nat Neurosci. 2017; 20: 365-377
        • Scarpazza C.
        • Ha M.
        • Baecker L.
        • Garcia-Dias R.
        • Pinaya W.H.L.
        • Vieira S.
        • Mechelli A.
        Translating research findings into clinical practice: A systematic and critical review of neuroimaging-based clinical tools for brain disorders.
        Transl Psychiatry. 2020; 10: 107
        • Gordon E.M.
        • Laumann T.O.
        • Gilmore A.W.
        • Newbold D.J.
        • Greene D.J.
        • Berg J.J.
        • et al.
        Precision functional mapping of individual human brains.
        Neuron. 2017; 95: 791-807.e7
        • Kong R.
        • Li J.
        • Orban C.
        • Sabuncu M.R.
        • Liu H.
        • Schaefer A.
        • et al.
        Spatial topography of individual-specific cortical networks predicts human cognition, personality, and emotion [published correction appears in Cereb Cortex 2021; 31:3974].
        Cereb Cortex. 2019; 29: 2533-2551
        • Braga R.M.
        • Buckner R.L.
        Parallel interdigitated distributed networks within the individual estimated by intrinsic functional connectivity.
        Neuron. 2017; 95: 457-471.e5
        • Gratton C.
        • Laumann T.O.
        • Nielsen A.N.
        • Greene D.J.
        • Gordon E.M.
        • Gilmore A.W.
        • et al.
        Functional brain networks are dominated by stable group and individual factors, not cognitive or daily variation.
        Neuron. 2018; 98: 439-452.e5
        • Mueller S.
        • Wang D.
        • Fox M.D.
        • Yeo B.T.T.
        • Sepulcre J.
        • Sabuncu M.R.
        • et al.
        Individual variability in functional connectivity architecture of the human brain.
        Neuron. 2013; 77: 586-595
        • Wang D.
        • Buckner R.L.
        • Fox M.D.
        • Holt D.J.
        • Holmes A.J.
        • Stoecklein S.
        • et al.
        Parcellating cortical functional networks in individuals.
        Nat Neurosci. 2015; 18: 1853-1860
        • Sylvester C.M.
        • Yu Q.
        • Srivastava A.B.
        • Marek S.
        • Zheng A.
        • Alexopoulos D.
        • et al.
        Individual-specific functional connectivity of the amygdala: A substrate for precision psychiatry.
        Proc Natl Acad Sci U S A. 2020; 117: 3808-3818
        • Buckner R.L.
        • DiNicola L.M.
        The brain’s default network: Updated anatomy, physiology and evolving insights.
        Nat Rev Neurosci. 2019; 20: 593-608
        • Li M.
        • Wang D.
        • Ren J.
        • Langs G.
        • Stoecklein S.
        • Brennan B.P.
        • et al.
        Performing group-level functional image analyses based on homologous functional regions mapped in individuals.
        PLoS Biol. 2019; 17e2007032
        • First M.
        • Spitzer R.
        • Gibbon M.
        • Williams J.
        Structured Clinical Interview for DSM-IV Axis I Disorders (SCID).
        American Psychiatric Press, Washington, DC1997
        • Szegedi A.
        • Jansen W.T.
        • van Willigenburg A.P.P.
        • van der Meulen E.
        • Stassen H.H.
        • Thase M.E.
        Early improvement in the first 2 weeks as a predictor of treatment outcome in patients with major depressive disorder: A meta-analysis including 6562 patients.
        J Clin Psychiatry. 2009; 70: 344-353
        • Widge A.S.
        • Bilge M.T.
        • Montana R.
        • Chang W.
        • Rodriguez C.I.
        • Deckersbach T.
        • et al.
        Electroencephalographic biomarkers for treatment response prediction in major depressive illness: A meta-analysis.
        Am J Psychiatry. 2019; 176: 44-56
        • Spies M.
        • Kraus C.
        • Geissberger N.
        • Auer B.
        • Klöbl M.
        • Tik M.
        • et al.
        Default mode network deactivation during emotion processing predicts early antidepressant response.
        Transl Psychiatry. 2017; 7e1008
        • Pei C.
        • Sun Y.
        • Zhu J.
        • Wang X.
        • Zhang Y.
        • Zhang S.
        • et al.
        Ensemble learning for early-response prediction of antidepressant treatment in major depressive disorder.
        J Magn Reson Imaging. 2020; 52: 161-171
        • Buckner R.L.
        • Krienen F.M.
        • Castellanos A.
        • Diaz J.C.
        • Yeo B.T.T.
        The organization of the human cerebellum estimated by intrinsic functional connectivity.
        J Neurophysiol. 2011; 106: 2322-2345
        • Yeo B.T.T.
        • Krienen F.M.
        • Sepulcre J.
        • Sabuncu M.R.
        • Lashkari D.
        • Hollinshead M.
        • et al.
        The organization of the human cerebral cortex estimated by intrinsic functional connectivity.
        J Neurophysiol. 2011; 106: 1125-1165
        • Holmes A.J.
        • Hollinshead M.O.
        • O’Keefe T.M.
        • Petrov V.I.
        • Fariello G.R.
        • Wald L.L.
        • et al.
        Brain Genomics Superstruct Project initial data release with structural, functional, and behavioral measures.
        Sci Data. 2015; 2: 150031
        • Murtagh F.
        • Legendre P.
        Ward’s hierarchical agglomerative clustering method: Which algorithms implement Ward’s criterion?.
        J Classif. 2014; 31: 274-295
        • Kennedy S.H.
        Core symptoms of major depressive disorder: Relevance to diagnosis and treatment.
        Dialogues Clin Neurosci. 2008; 10: 271-277
        • Heilbronner S.R.
        • Hayden B.Y.
        Dorsal anterior cingulate cortex: A bottom-up view.
        Annu Rev Neurosci. 2016; 39: 149-170
        • Wikenheiser A.M.
        • Schoenbaum G.
        Over the river, through the woods: Cognitive maps in the hippocampus and orbitofrontal cortex.
        Nat Rev Neurosci. 2016; 17: 513-523
        • Rudebeck P.H.
        • Saunders R.C.
        • Prescott A.T.
        • Chau L.S.
        • Murray E.A.
        Prefrontal mechanisms of behavioral flexibility, emotion regulation and value updating.
        Nat Neurosci. 2013; 16: 1140-1145
        • Rushworth M.F.S.
        • Behrens T.E.J.
        Choice, uncertainty and value in prefrontal and cingulate cortex.
        Nat Neurosci. 2008; 11: 389-397
        • Treadway M.T.
        • Zald D.H.
        Reconsidering anhedonia in depression: Lessons from translational neuroscience.
        Neurosci Biobehav Rev. 2011; 35: 537-555
        • Keedwell P.A.
        • Andrew C.
        • Williams S.C.R.
        • Brammer M.J.
        • Phillips M.L.
        The neural correlates of anhedonia in major depressive disorder.
        Biol Psychiatry. 2005; 58: 843-853
        • Borsini A.
        • Wallis A.S.
        • Zunszain P.
        • Pariante C.M.
        • Kempton M.J.
        Characterizing anhedonia: A systematic review of neuroimaging across the subtypes of reward processing deficits in depression.
        Cogn Affect Behav Neurosci. 2020; 20: 816-841
        • Williams L.M.
        Precision psychiatry: A neural circuit taxonomy for depression and anxiety.
        Lancet Psychiatry. 2016; 3: 472-480
        • Zhang H.
        • Chen Z.
        • Jia Z.
        • Gong Q.
        Dysfunction of neural circuitry in depressive patients with suicidal behaviors: A review of structural and functional neuroimaging studies.
        Prog Neuropsychopharmacol Biol Psychiatry. 2014; 53: 61-66
        • Monkul E.S.
        • Hatch J.P.
        • Nicoletti M.A.
        • Spence S.
        • Brambilla P.
        • Lacerda A.L.T.
        • et al.
        Fronto-limbic brain structures in suicidal and non-suicidal female patients with major depressive disorder.
        Mol Psychiatry. 2007; 12: 360-366
        • Jollant F.
        • Lawrence N.S.
        • Giampietro V.
        • Brammer M.J.
        • Fullana M.A.
        • Drapier D.
        • et al.
        Orbitofrontal cortex response to angry faces in men with histories of suicide attempts.
        Am J Psychiatry. 2008; 165: 740-748
        • Zhou H.X.
        • Chen X.
        • Shen Y.Q.
        • Li L.
        • Chen N.X.
        • Zhu Z.C.
        • et al.
        Rumination and the default mode network: Meta-analysis of brain imaging studies and implications for depression.
        Neuroimage. 2020; 206: 116287
        • Stroebe M.
        • Boelen P.A.
        • van den Hout M.
        • Stroebe W.
        • Salemink E.
        • van den Bout J.
        Ruminative coping as avoidance: A reinterpretation of its function in adjustment to bereavement.
        Eur Arch Psychiatry Clin Neurosci. 2007; 257: 462-472
        • Henderson S.E.
        • Vallejo A.I.
        • Ely B.A.
        • Kang G.
        • Krain Roy A.
        • Pine D.S.
        • et al.
        The neural correlates of emotional face-processing in adolescent depression: A dimensional approach focusing on anhedonia and illness severity.
        Psychiatry Res. 2014; 224: 234-241
        • Lu S.
        • Xu R.
        • Cao J.
        • Yin Y.
        • Gao W.
        • Wang D.
        • et al.
        The left dorsolateral prefrontal cortex volume is reduced in adults reporting childhood trauma independent of depression diagnosis.
        J Psychiatr Res. 2019; 112: 12-17
        • Rolls E.T.
        • Cheng W.
        • Gong W.
        • Qiu J.
        • Zhou C.
        • Zhang J.
        • et al.
        Functional connectivity of the anterior cingulate cortex in depression and in health.
        Cereb Cortex. 2019; 29: 3617-3630
        • Raichle M.E.
        The brain’s default mode network.
        Annu Rev Neurosci. 2015; 38: 433-447
        • Dixon M.L.
        • De La Vega A.
        • Mills C.
        • Andrews-Hanna J.
        • Spreng R.N.
        • Cole M.W.
        • Christoff K.
        Heterogeneity within the frontoparietal control network and its relationship to the default and dorsal attention networks [published correction appears in Proc Natl Acad Sci U S A 2018; 115:E3068].
        Proc Natl Acad Sci U S A. 2018; 115: E1598-E1607
        • Buckner R.L.
        • Andrews-Hanna J.R.
        • Schacter D.L.
        The brain’s default network: Anatomy, function, and relevance to disease.
        Ann N Y Acad Sci. 2008; 1124: 1-38
        • Yan C.G.
        • Chen X.
        • Li L.
        • Castellanos F.X.
        • Bai T.J.
        • Bo Q.J.
        • et al.
        Reduced default mode network functional connectivity in patients with recurrent major depressive disorder.
        Proc Natl Acad Sci U S A. 2019; 116: 9078-9083
        • Sheline Y.I.
        • Price J.L.
        • Yan Z.
        • Mintun M.A.
        Resting-state functional MRI in depression unmasks increased connectivity between networks via the dorsal nexus.
        Proc Natl Acad Sci U S A. 2010; 107: 11020-11025
        • Zhao Y.
        • Chen L.
        • Zhang W.
        • Xiao Y.
        • Shah C.
        • Zhu H.
        • et al.
        Gray matter abnormalities in non-comorbid medication-naive patients with major depressive disorder or social anxiety disorder.
        EBioMedicine. 2017; 21: 228-235
        • de Voogd L.D.
        • Fernández G.
        • Hermans E.J.
        Awake reactivation of emotional memory traces through hippocampal-neocortical interactions.
        Neuroimage. 2016; 134: 563-572
        • Groenewold N.A.
        • Opmeer E.M.
        • de Jonge P.
        • Aleman A.
        • Costafreda S.G.
        Emotional valence modulates brain functional abnormalities in depression: Evidence from a meta-analysis of fMRI studies.
        Neurosci Biobehav Rev. 2013; 37: 152-163
        • Price J.L.
        • Drevets W.C.
        Neural circuits underlying the pathophysiology of mood disorders.
        Trends Cogn Sci. 2012; 16: 61-71
        • Barber A.D.
        • Sarpal D.K.
        • John M.
        • Fales C.L.
        • Mostofsky S.H.
        • Malhotra A.K.
        • et al.
        Age-normative pathways of striatal connectivity related to clinical symptoms in the general population.
        Biol Psychiatry. 2019; 85: 966-976
        • Kaiser R.H.
        • Andrews-Hanna J.R.
        • Wager T.D.
        • Pizzagalli D.A.
        Large-scale network dysfunction in major depressive disorder: A meta-analysis of resting-state functional connectivity.
        JAMA Psychiatry. 2015; 72: 603-611
        • Janes A.C.
        • Zegel M.
        • Ohashi K.
        • Betts J.
        • Molokotos E.
        • Olson D.
        • et al.
        Nicotine normalizes cortico-striatal connectivity in non-smoking individuals with major depressive disorder.
        Neuropsychopharmacology. 2018; 43: 2445-2451