Patterns of Neural Network Functional Connectivity Associated With Mania/Hypomania and Depression Risk in 3 Independent Young Adult Samples

Key Points Question Are there objective, reproducible neural markers that can distinguish mania/hypomania from depression risk? Findings In 3 independent samples comprising 299 young adults, neural response patterns differentially associated with mania/hypomania and depression risk were identified and replicated. Greater bilateral amygdala–left amygdala functional connectivity was associated with greater mania/hypomania and depression risk, and greater bilateral ventrolateral prefrontal cortex–right dorsolateral prefrontal cortex functional connectivity and greater right caudate deactivation were associated with greater mania/hypomania and depression risk, respectively. Meaning Neural markers reliably associated with mania/hypomania and depression risk may help identify young adults at risk of bipolar disorder and provide treatment targets for early interventions.

eMethods.Exclusion criteria, power calculation, clinical measures, neuroimaging data acquisition, neuroimaging data preprocessing, functional imaging task, first-level analyses and extraction of neural measures, cross validation procedure eResults.Additional risk models, post-hoc sensitivity analyses, all facial emotions exploratory analyses eFigure 1. Facial emotion-processing task eFigure 2. Comparison of the bilateral amygdala-left amygdala cluster target and bilateral anatomical amygdala seed mask in the three samples eFigure 3. Scatterplots of Pearson correlations between MOODS-SR risk scores and current anxiety, depression and mania/hypomania severity in the discovery sample eFigure 4. Scatterplots of Pearson correlations between MOODS-SR risk scores and current anxiety, depression and mania/hypomania severity in test sample 1 eFigure 5. Scatterplots of Pearson correlations between MOODS-SR risk scores and current anxiety, depression and mania/hypomania severity in test sample 2 eFigure 6. Figure 2C, third panel, with N=2 outliers removed eTable 1. MOODS-SR manic and depressive mood domain descriptive statistics eTable 2. Significant activity and functional connectivity to approach-related facial emotions in the three samples eTable 3. Additional mania/hypomania risk models eTable 4. Additional depression risk models eTable 5. Post-hoc sensitivity analyses with mania/hypomania risk in unmedicated participants eMethods.Exclusion criteria, power calculation, clinical measures, neuroimaging data acquisition, neuroimaging data preprocessing, functional imaging task, first-level modeling and extraction of neural measures, cross validation procedure

Exclusion criteria
Discovery sample: n=133 participants were scanned on a Siemens Trio scanner; n=19 were excluded due to: excessive motion (>4 mm/mean framewise displacement>0.05)(n=13), incomplete neuroimaging assessments or missing data (n=4), having a diagnosis of Bipolar Disorder (n=2).See eTable 15 for comparisons of demographic and clinical measures in included participants versus excluded participants from the discovery sample.
Test sample 1: n=136 participants were scanned on a Siemens Prisma scanner; n=33 were excluded due to: excessive motion (n=17), incomplete neuroimaging assessments or missing data (n=13), having a diagnosis of Bipolar Disorder (n=3).See eTable 16 for comparisons of demographic and clinical measures in included participants versus excluded participants from test sample 1.
Test sample 2: n=106 participants were scanned on a Siemens Prisma scanner; n=24 were excluded due to: excessive motion (n=9), incomplete neuroimaging assessments or missing data (n=15).None of the participants in test sample 2 had a diagnosis of Bipolar Disorder.See eTable 17 for comparisons of demographic and clinical measures in included participants versus excluded participants from test sample 2.
Additional exclusion criteria included: visual disturbance (<20/40 Snellen corrected visual acuity); presence/questionable history of metallic foreign objects in body; positive pregnancy test/self-reporting of pregnancy for female individuals; claustrophobia; Mini-Mental State Examination score<24; a premorbid IQ estimate<85 (determined by the National Adult Reading Test); substance use disorder (SUD) or illicit substance use (except cannabis, given its common usage in young adults) in the prior three months; any history of serious medical/physical conditions: head injury, neurological disorder, pervasive developmental disorder (e.g., autism), history of brain tumor/brain surgery, progressive endocrine disorder, heart disorder, or other major systemic medical conditions (kidney disease, multiple sclerosis, cerebral palsy, blindness, serious physical disability) or chronic/acute condition including any managed by medication (chronic back problem, recent surgery); current psychosis and/or lifetime history of primary psychotic disorder; taking medication for an excluded medical condition; current treatment with psychotropic medication for >2 weeks; previous psychotropic medication treatment in the past three months; inability to comprehend English.All participants were right-handed.The Structured Clinical Interview for DSM-5 Research Version (SCID-5-RV) was administered to all participants to assess for current/past month and lifetime DSM-5 psychiatric disorders 1 .Diagnosis and subsequent exclusion of Bipolar Disorder were defined with DSM-5 criteria, based on administration of the SCID-5-RV.For information on family psychiatric history of the three samples, see eTable 18.

Power calculation
With our sample sizes (discovery sample n=114, test sample 1 n=103, test sample 2 n=82), for continuous outcome measures, we had 90%, 86%, and 75% power, respectively, to detect a significant(P=0.05)effect(f 2 =0.15) for the independent variables, controlling for up to five independent variables.

Clinical measures
Mania/hypomania and depression lifetime risk were measured by the Mood Spectrum Self-Report (MOODS-SR) [2][3][4][5][6] , a 160-item count measure assessing lifetime occurrence of threshold/subthreshold manic/hypomanic and depressive symptomology, behavioral traits, lifestyles, and features related to mood/affective dysregulation characterizing subsyndromal-prodromal states of affective disturbance.The present study specifically focused on the manic and depressive 'mood' domains, which assess mood lability and associated changes in interest towards interpersonal relationships, work, hobbies, and sports.These scales are included within the MOODS-SR, which also includes five other domain scales: manic energy, depressive energy, manic cognition, depressive cognition, and rhythmicity.
In addition to the MOODS-SR mania/hypomania and depression risk assessments described in the main text and as above, additional clinical scales measured present symptom severity as part of a larger study (R01MH100041/R37MH100041).These scales included the Young Mania Rating Scale 7 to measure current mania/hypomania severity, the Hamilton Rating Scale for Depression 8 to measure current depression severity, and Hamilton Anxiety Rating Scale 9 to measure state anxiety.

Neuroimaging data acquisition
All neuroimaging data were collected at the University of Pittsburgh.

Neuroimaging data preprocessing
Imaging data for both the discovery sample and test sample 1 were preprocessed using a combination of software packages (SPM, FSL 10,11 , and AFNI 12 ) implemented in Nipype 13 .Standard preprocessing steps were applied including realignment to the first volume of the time series, co-registration of realigned images to each participant's anatomic scan, normalization of anatomic images to the MNI152 template, despiking using AFNI 3dDespike, and spatial smoothing using a Full-width at half maximum (FWHM) of 6 mm.
Imaging data for test sample 2 was preprocessed using fMRIPrep 14 , FSL (5.0.9), and AFNI 20160207 (RRID:SCR_005927).The first volume of the time series and its skull-stripped version were generated from the shortest echo of the BOLD run using fMRIPrep, followed by co-registration to the T1-weighted reference image using FreeSurfer bbregister and estimation of head-motion parameters using FSL mcflirt.Slice-time correction was performed using AFNI 3dTshift.BOLD time-series were then resampled into native space.Multi-echo (ME) independent components analysis (ME-ICA) 15,16 was then applied to preprocessed BOLD images using Tedana 17 for ME integration.Despiking using AFNI 3dDespike and spatial smoothing using a FWHM of 6 mm were then applied.

Functional imaging task
Participants completed a well-validated facial emotion-processing task (eFigure 1) [18][19][20][21] .During the 12.5-minute block design task, participants viewed facial expressions from the NimStim dataset 22 that morphed in 5% increments from neutral (0% emotion) to angry, fear, sad, or happy (100% emotion) over the course of a one-second movie.In control trials, movies comprised a simple dark oval superimposed on a light-grey oval that subsequently morphs into a larger shape.There were three blocks for each of the four face emotions (happy, fear, angry, and sad) with 12 stimuli per block, and six control blocks with six stimuli per block.Blocks were presented in a pseudorandomized order.Participants used one of three fingers to press a button indicating the color of a semi-transparent foreground color flash (orange, blue, or yellow) that appeared during the mid 200-650 ms of the one-second face movie.

First-level modeling and extraction of neural measures
SPM8 (discovery sample and test sample 1) and SPM12 (test sample 2) were used to generate a first-level fixedeffects general linear model for each participant in each sample.In all samples, the model included six motion parameters and a regressor controlling for physiologic noise derived from the mean signal of white matter and CSF as covariates of no interest.A high-pass frequency filter (256 seconds) and autoregressive modeling (AR(1) for discovery sample and test sample 1, FAST for test sample 2) were also applied.In all samples, from the first-level activation maps, statistical parametric maps were generated for our emotion contrast of interest, the main effect of approach-related emotions (angry and happy faces combined) versus implicit baseline, to identify neural activity and functional connectivity related to processing approach-related emotions across the entire sample using a one sample t-test in SPM12.In all samples, parameter estimates for clusters in which there was significant activity were then extracted using a family-wise error (FWE) correction threshold of α=0.05, within an anatomical region-of-interest (ROI) mask that was created in SPM using the WFU Pickatlas, Wake Forest University, Winston-Salem 23 (further described in the main text Methods 'Activity and Functional Connectivity' section).

Cross-validation procedure
Standard cross-validation evaluated the replication of the risk models' predictive ability 24 in all three samples via comparing the average in-sample and out-of-sample difference between predicted MOODS-SR scores and actual MOODS-SR scores.Raw MOODS-SR scores were natural log-transformed and scaled up by 1 to account for scores of 0. Formula: mean(abs(ln(raw MOODS-SR score+1)-predicted MOODS-SR score)).
In test sample 1, a Poisson log-link regression model revealed that right caudate activity (FDR q<0.001) was negatively, and amygdala-left amygdala functional connectivity (FDR q<0.001) and vlPFC-right dlPFC functional connectivity (FDR q=0.004) were positively, associated with mania/hypomania risk (eTable 3B).Age, dACC-left mPFC functional connectivity, dACC-left dlPFC functional connectivity, dlPFC-right dlPFC functional connectivity, and putamen-left dACC functional connectivity were not significantly associated with mania/hypomania risk.Left dlPFC activity, right vlPFC activity, and putamen-right mPFC functional connectivity were not observed in test sample 1 and were thus not included in the model.
In test sample 2, a Poisson log-link regression model revealed that right caudate activity (FDR q=0.002) was negatively, and amygdala-left amygdala functional connectivity (FDR q<0.001) and dlPFC-right dlPFC functional connectivity (FDR q=0.021) were positively, associated with mania/hypomania risk (eTable 3C).Age, vlPFC-right dlPFC functional connectivity, dACC-left mPFC/dlPFC functional connectivity, putamen-left dACC functional connectivity, and putamen-right mPFC functional connectivity were not significantly associated with mania/hypomania risk.Left dlPFC activity and right vlPFC activity were not observed in test sample 2 and were thus not included in the model.

Additional depression risk models
In the discovery sample, a Poisson log-link regression model revealed that left dlPFC activity, right caudate activity, and dACC-left dlPFC functional connectivity (FDR q<0.001) were negatively, and amygdala-left amygdala functional connectivity (FDR q<0.001), dACC-left mPFC functional connectivity (FDR q=0.002), and dlPFC-right dlPFC functional connectivity (FDR q<0.001) were positively, associated with depression risk (eTable 4A).Age, right vlPFC activity, vlPFC-right dlPFC functional connectivity, putamen-right mPFC functional connectivity, and putamen-left dACC functional connectivity were not significantly associated with depression risk.
In test sample 1, a Poisson log-link regression model revealed that right caudate activity (FDR q<0.001) was negatively, and amygdala-left amygdala functional connectivity (FDR q<0.001), vlPFC-right dlPFC functional connectivity (FDR q=0.002), dlPFC-right dlPFC functional connectivity (FDR q<0.001), and putamen-left dACC functional connectivity (FDR q<0.001) were positively, associated with depression risk (eTable 4B).Age, dACCleft mPFC functional connectivity, and dACC-left dlPFC functional connectivity were not significantly associated with depression risk.Left dlPFC activity, right vlPFC activity, and putamen-right mPFC functional connectivity were not observed in test sample 1 and were thus not included in the model.
In test sample 2, a Poisson log-link regression model revealed that age, amygdala-left amygdala functional connectivity, and dACC-left mPFC/dlPFC functional connectivity (FDR q<0.001) were positively, and right caudate activity, putamen-right mPFC functional connectivity (FDR q<0.001), and putamen-left dACC functional connectivity (FDR q=0.002) were negatively, associated with depression risk (eTable 4C).vlPFC-right dlPFC functional connectivity and dlPFC-right dlPFC functional connectivity were not significantly associated with depression risk.Left dlPFC activity and right vlPFC activity were not observed in test sample 2 and were thus not included in the model.

Post-hoc sensitivity analysis: excluding participants taking medication
In the discovery sample, after excluding participants taking medication, a Poisson log-link regression model revealed that amygdala-left amygdala functional connectivity, vlPFC-right dlPFC functional connectivity, and dACC-left mPFC functional connectivity were positively associated with mania/hypomania risk (FDR q<0.001); left dlPFC activity (FDR q=0.003) and right vlPFC activity (FDR q=0.008) were negatively associated with mania/hypomania risk (eTable 5A).Age was not significantly associated with mania/hypomania risk.
In test sample 1, after excluding participants taking medication, a Poisson log-link regression model revealed that amygdala-left amygdala functional connectivity and vlPFC-right dlPFC functional connectivity were positively associated with mania/hypomania risk (FDR q<0.001) (eTable 5B).Age and dACC-left mPFC functional connectivity were not significantly associated with mania/hypomania risk.Left dlPFC activity and right vlPFC activity were not observed in test sample 1 and were thus not included in the model.
In the discovery sample, after excluding participants taking medication, a Poisson log-link regression model revealed that right vlPFC activity (FDR q=0.02), right caudate activity (FDR q<0.001), and dACC-left dlPFC functional connectivity (FDR q<0.001) were negatively associateed with depression risk; dlPFC-right dlPFC functional connectivity and amygdala-left amygdala functional connectivity were positively associated with depression risk (FDR q<0.001) (eTable 6A).Age, putamen-left dACC functional connectivity, and putamen-right mPFC functional connectivity were not significantly associated with depression risk.
In test sample 1, after excluding participants taking medication, a Poisson log-link regression model revealed that right caudate activity was negatively, and amygdala-left amygdala functional connectivity, dlPFC-right dlPFC functional connectivity, and putamen-left dACC functional connectivity were positively, associated with depression risk (FDR q<0.001) (eTable 6B).Age and dACC-left dlPFC functional connectivity were not significantly associated with depression risk.Right vlPFC activity and putamen-right mPFC functional connectivity were not observed in test sample 1 and were thus not included in the model.
None of the participants in test sample 2 were taking psychotropic medication.

Post-hoc sensitivity analysis: excluding participants with Major Depressive Disorder (MDD)
In the discovery sample, after excluding participants with MDD, a Poisson log-link regression model revealed that age (FDR q=0.036), amygdala-left amygdala functional connectivity, vlPFC-right dlPFC functional connectivity, and dACC-left mPFC functional connectivity were positively associated with mania/hypomania risk (FDR q<0.001); left dlPFC activity (FDR q=0.038) was negatively associated with mania/hypomania risk (eTable 7A).Right vlPFC activity was not significantly associated with mania/hypomania risk.
In test sample 1, after excluding participants with MDD, a Poisson log-link regression model revealed that amygdala-left amygdala functional connectivity and vlPFC-right dlPFC functional connectivity (FDR q<0.001) were positively associated with mania/hypomania risk; age was marginally positively associated with mania/hypomania risk (FDR q=0.05) (eTable 7B).dACC-left mPFC functional connectivity was not significantly associated with mania/hypomania risk.Left dlPFC activity and right vlPFC activity were not observed in test sample 1 and were thus not included in the model.
In test sample 2, after excluding participants with MDD, a Poisson log-link regression model revealed that amygdala-left amygdala functional connectivity and vlPFC-right dlPFC functional connectivity were positively associated with mania/hypomania risk (FDR q<0.001); age and dACC-left dlPFC functional connectivity were not significantly associated with mania/hypomania risk (eTable 7C).Left dlPFC activity and right vlPFC activity were not observed in test sample 2 and were thus not included in the model.
In the discovery sample, after excluding participants with MDD, a Poisson log-link regression model revealed that age (FDR q=0.028), right caudate activity, dACC-left dlPFC functional connectivity, and putamen-right mPFC functional connectivity (FDR q<0.001) were negatively associated with depression risk; amygdala-left amygdala functional connectivity, dlPFC-right dlPFC functional connectivity (FDR q<0.001), and putamen-left dACC functional connectivity (FDR q=0.008) were positively associated with depression risk (eTable 8A).Right vlPFC activity was not significantly associated with depression risk.
In test sample 1, after excluding participants with MDD, a Poisson log-link regression model revealed that right caudate activity (FDR q<0.001) and amygdala-left amygdala functional connectivity (FDR q=0.013) were negatively associated with depression risk; dlPFC-right dlPFC functional connectivity and putamen-left dACC functional connectivity (FDR q<0.001) were positively associated with depression risk (eTable 8B).Age and dACCleft dlPFC functional connectivity were not significantly associated with depression risk.Right vlPFC activity and putamen-right mPFC functional connectivity were not observed in test sample 1 and were thus not included in the model.
In test sample 2, after excluding participants with MDD, a Poisson log-link regression model revealed that right caudate activity was negatively associated with depression risk (FDR q<0.001) and age, amygdala-left amygdala functional connectivity (FDR q<0.001), and dACC-left dlPFC functional connectivity (FDR q=0.002) were positively associated with depression risk (eTable 8C).dlPFC-right dlPFC functional connectivity, putamen-left dACC functional connectivity, and putamen-right mPFC functional connectivity were not significantly associated with depression risk.Right vlPFC activity was not observed in test sample 2 and was thus not included in the model.

Post-hoc sensitivity analysis: excluding participants with Attention Deficit Hyperactivity Disorder (ADHD)
In the discovery sample, after excluding participants with ADHD, a Poisson log-link regression model revealed that left dlPFC activity (FDR q=0.024) and right vlPFC activity (FDR q=0.009) were negatively associated with mania/hypomania risk; amygdala-left amygdala functional connectivity, vlPFC-right dlPFC functional connectivity, and dACC-left mPFC functional connectivity (FDR q<0.001) were positively associated with mania/hypomania risk (eTable 9A).Age was not significantly associated with mania/hypomania risk.
In test sample 1, after excluding participants with ADHD, a Poisson log-link regression model revealed that amygdala-left amygdala functional connectivity and vlPFC-right dlPFC functional connectivity were positively associated with mania/hypomania risk (FDR q<0.001) (eTable 9B).dACC-left mPFC functional connectivity was not significantly associated with mania/hypomania risk, and age was no longer significantly positively associated with mania/hypomania risk after FDR correction (P=0.048;FDR q=0.068).Left dlPFC activity and right vlPFC activity were not observed in test sample 1 and were thus not included in the model.
In test sample 2, after excluding participants with ADHD, a Poisson log-link regression model revealed that amygdala-left amygdala functional connectivity (FDR q<0.001), vlPFC-right dlPFC functional connectivity (FDR q=0.005), and dACC-left mPFC functional connectivity (FDR q=0.017) were positively associated with mania/hypomania risk (eTable 9C).Age was not significantly associated with mania/hypomania risk.Left dlPFC activity and right vlPFC activity were not observed in test sample 2 and were thus not included in the model.
In the discovery sample, after excluding participants with ADHD, a Poisson log-link regression model revealed that right vlPFC activity (FDR q=0.007), right caudate activity and dACC-left dlPFC functional connectivity (FDR q<0.001) were negatively associated with depression risk; amygdala-left amygdala functional connectivity and dlPFC-right dlPFC functional connectivity (FDR q<0.001) were positively associated with depression risk (eTable 10A).Age, putamen-left dACC functional connectivity, and putamen-right mPFC functional connectivity were not significantly associated with depression risk.
In test sample 1, after excluding participants with ADHD, a Poisson log-link regression model revealed that right caudate was negatively associated with depression risk (FDR q<0.001) and amygdala-left amygdala functional connectivity, dlPFC-right dlPFC functional connectivity, and putamen-left dACC functional connectivity (FDR q<0.001) were positively associated with depression risk (eTable 10B).Age and dACC-left dlPFC functional connectivity were not significantly associated with depression risk.Right vlPFC activity and putamen-right mPFC functional connectivity were not observed in test sample 1 and were thus not included in the model.
In test sample 2, after excluding participants with ADHD, a Poisson log-link regression model revealed that age, amygdala-left amygdala functional connectivity, dACC-left dlPFC functional connectivity, and putamen-right mPFC functional connectivity were positively associated with depression risk (FDR q<0.001); right caudate activity (FDR q<0.001) and putamen-left dACC functional connectivity (FDR q=0.002) were negatively associated with depression risk (eTable 10C).dlPFC-right dlPFC functional connectivity was not significantly associated with depression risk.Right vlPFC activity was not observed in test sample 2 and was thus not included in the model.

All facial emotions exploratory analysis
The patterns of neural activity and functional connectivity to the all facial emotions contrast in the two test samples were mostly consistent with that shown in the discovery sample (eTable 11), which allowed the two MOODS-SR risk models generated in the discovery sample to be tested for replication in the two independent test samples using the same procedures described in the main text for the main contrast i.e., approach-related emotions.
First, in the discovery sample, elastic-net variable selection after cross-validation using the Poisson family identified age, right vlPFC activity, right dlPFC activity, caudate-left caudate functional connectivity, dACC-right vlPFC functional connectivity, dlPFC-right insula functional connectivity, and vlPFC-right dlPFC functional connectivity to the all facial emotions contrast as measures associated with mania/hypomania risk (eTable 12A).A Poisson loglink regression model then revealed that age, dACC-right vlPFC functional connectivity, and dlPFC-right insula functional connectivity (FDR q<0.001) were positively associated with mania/hypomania risk; right vlPFC activity (FDR q<0.001) and right dlPFC activity (FDR q=0.001) were negatively associated with mania/hypomania risk (eTable 12A).Caudate-left caudate functional connectivity and vlPFC-right dlPFC functional connectivity were not significantly associated with mania/hypomania risk.
In test sample 1, a Poisson log-link regression model revealed that age(FDR q<0.001), dACC-right vlPFC functional connectivity (FDR q=0.013), and vlPFC-right dlPFC functional connectivity (FDR q=0.004) were positively associated with mania/hypomania risk(eTable 12B).Right dlPFC activity, caudate-left caudate functional connectivity, and dlPFC-right insula functional connectivity were not significantly associated with mania/hypomania risk.Right vlPFC activity was not observed in test sample 1 and was thus not included in the model.
In test sample 2, a Poisson log-link regression model revealed that age and caudate-left caudate functional connectivity (FDR q<0.001) were positively associated with mania/hypomania risk (eTable 12C).dACC-right vlPFC functional connectivity and vlPFC-right dlPFC functional connectivity were not significantly associated with mania/hypomania risk.Right vlPFC activity, right dlPFC activity, and dlPFC-right insula functional connectivity were not observed in test sample 2 and were thus not included in the model.
Next, in the discovery sample, elastic-net variable selection after cross-validation using the Poisson family identified gender, age, right vlPFC activity, right caudate activity, amygdala-left amygdala functional connectivity, caudateleft caudate functional connectivity, dACC-left dlPFC functional connectivity, dlPFC-right dlPFC functional connectivity, putamen-right OFC functional connectivity, and putamen-left dACC functional connectivity to the all facial emotions contrast as measures associated with depression risk (eTable 13A).A Poisson log-link regression model then revealed that gender, right vlPFC activity, right caudate activity (FDR q<0.001) and dACC-left dlPFC functional connectivity (FDR q=0.024) were negatively associated with depression risk; age, amygdala-left amygdala functional connectivity, and dlPFC-right dlPFC functional connectivity (FDR q<0.001) were positively associated with depression risk (eTable 13A).Caudate-left caudate functional connectivity, putamen-right OFC functional connectivity, and putamen-left dACC functional connectivity were not significantly associated with depression risk.
In test sample 1, a Poisson log-link regression model revealed that gender (FDR q=0.006) and right caudate activity (FDR q=0.033) were negatively associated with depression risk; age (FDR q<0.001) and caudate-left caudate functional connectivity (FDR q=0.026) were positively associated with depression risk (eTable 13B).Amygdala-left amygdala functional connectivity, dACC-left dlPFC functional connectivity, and dlPFC-right dlPFC functional connectivity were not significantly associated with depression risk.Right vlPFC activity, putamen-right OFC functional connectivity, and putamen-left dACC functional connectivity were not observed in test sample 1 and were thus not included in the model.
In test sample 2, a Poisson log-link regression model revealed that age (FDR q<0.001), dACC-left dlPFC functional connectivity (FDR q=0.011), and putamen-right OFC functional connectivity (FDR q<0.001) were positively associated with depression risk; dlPFC-right dlPFC functional connectivity and putamen-left dACC functional connectivity (FDR q<0.001) were negatively associated with depression risk (eTable 13C).Gender, right caudate activity, amygdala-left amygdala functional connectivity, and caudate-left caudate functional connectivity were not significantly associated with depression risk.Right vlPFC activity was not observed in test sample 2 and was thus not included in the model.eFigure 1. Facial emotion-processing task.The top panel represents a happy trial of our facial emotion-processing task.Over a one-second duration, the face changed in 5% increments from neutral (0% emotion) to a happy, sad, angry, or fearful face (100% emotion).The bottom panel represents the changing shape condition.In both cases, participants were asked to identify the color flash presented in the middle of the dynamic change.The correlations among MOODS-SR mania and HRSD and HAMA and among MOODS-SR depression and YMRS and HAMA likely reflect the general aspects of affective psychopathology and arousal, respectively, that are captured by the MOODS-SR mania and depression mood domains.eTable 16.Demographic and clinical measures of participants with missing or incomplete data versus participants with complete data from Test sample 1. Age, gender, education, and lifetime psychiatric diagnoses were assessed at time of fMRI scan.Participants may be diagnosed with more than one disorder.Abbreviations: ADHD, Attention Deficit Hyperactivity Disorder; MDD, Major Depressive Disorder; MOODS-SR, Mood Spectrum Self-Report; N/A, not applicable.

eFigure 2 . 3 .
Comparison of the bilateral amygdala-left amygdala cluster target and bilateral anatomical amygdala seed mask in the three samples.Black arrows highlight the location of the bilateral amygdala seed mask used in the functional connectivity analysis (shown on the left in red) which is overlaid on each sample's left amygdala cluster target from the bilateral amygdala-left amygdala functional connectivity finding that was associated with both mania/hypomania and depression risk in all three samples.Neural activity and functional connectivity displayed in the sagittal slices reflect the position in the y/z axis.Scatterplots of Pearson correlations between MOODS-SR risk scores and current anxiety, depression and mania/hypomania severity in the discovery sample.Correlations with the MOODS-SR manic mood domain are on the left; correlations with the MOODS-SR depressive mood domain are on the right.Abbreviations: HAMA=Hamilton Anxiety Rating Scale; HRSD=Hamilton Rating Scale for Depression; MOODS-SR=Mood Spectrum Self-Report; YMRS=Young Mania Rating Scale.eFigure 4. Scatterplots of Pearson correlations between MOODS-SR risk scores and current anxiety, depression and mania/hypomania severity in test sample 1.Correlations with the MOODS-SR manic mood domain are on the left; correlations with the MOODS-SR depressive mood domain are on the right.Abbreviations: HAMA=Hamilton Anxiety Rating Scale; HRSD=Hamilton Rating Scale for Depression; MOODS-SR=Mood Spectrum Self-Report; YMRS=Young Mania Rating Scale.eFigure 5. Scatterplots of Pearson correlations between MOODS-SR risk scores and current anxiety, depression and mania/hypomania severity in test sample 2.Correlations with the MOODS-SR manic mood domain are on the left; correlations with the MOODS-SR depressive mood domain are on the right.Abbreviations: HAMA=Hamilton Anxiety Rating Scale;

Significant activity and functional connectivity to approach-related facial emotions in the three samples.
Test sample 2. Abbreviations: MOODS-SR, Mood Spectrum Self-Report; SD, standard deviation; SE, standard error.Abbreviations: dACC, dorsal anterior cingulate cortex; dlPFC, dorsolateral prefrontal cortex; FWE, family-wise error corrected; k, cluster size; L, left; MNI, Montreal Neuroimaging Institute; mPFC, medial prefrontal cortex; OFC, orbitofrontal cortex; R, right; vACC, ventral anterior cingulate cortex; vlPFC, ventrolateral prefrontal cortex.Discovery sample activity and functional connectivity related to MOODS-SR manic mood domain using all non-zero coefficients from both initial mania and depression risk models.B. Test sample 1 activity and functional connectivity related to MOODS-SR manic mood domain using all non-zero coefficients from both initial mania and depression risk models.C. Discovery sample activity and functional connectivity related to MOODS-SR depressive mood domain using all non-zero coefficients from both initial mania/hypomania and depression risk models.B. Test sample 1 activity and functional connectivity related to MOODS-SR depressive mood domain using all non-zero coefficients from both initial mania/hypomania and depression risk models.C.
HRSD=Hamilton Rating Scale for Depression; MOODS-SR=Mood Spectrum Self-Report; YMRS=Young Mania Rating Scale.eFigure 6. Figure 2C, third panel, with N=2 outliers removed.Abbreviations: R=Right.Test Sample 2 © 2023 Schumer MC et al.JAMA Psychiatry.eTable 1. MOODS-SR manic and depressive mood domain descriptive statistics.A. Discovery sample; B. Test sample 1; C.a Cluster identified in elastic-net variable selection models in the discovery sample.bClusterreplicated in multivariable regression models in one or both of the independent test samples.cNearestpeakandassociated t-statistic are reported.eTable3.Additional mania/hypomania risk models.A.a Raw P values were significant at FDR-corrected thresholds; FDR q values are reported in the eResults.bFindingreplicatedin all 3 samples.cAgewasz-scored for the multivariable regression analyses.eTable4.Additional depression risk models.A.a Raw P values were significant at FDR-corrected thresholds; FDR q values are reported in the eResults.bFinding replicated in all 3 samples.c Age was z-scored for the multivariable regression analyses.eTable 5.

Post-hoc sensitivity analyses with mania/hypomania risk in unmedicated participants. A.
Discovery sample activity and functional connectivity related to MOODS-SR manic mood domain after variable selection with elasticnet.B. Test sample 1 activity and functional connectivity related to MOODS-SR manic mood domain.Age was z-scored a Raw P values were significant at FDR-corrected thresholds; FDR q values are reported in the eResults.bFindingreplicated in all 3 samples.c Age was z-scored for the multivariable regression analyses.

Post-hoc sensitivity analyses with depression risk in unmedicated participants. A.
Discovery sample activity and functional connectivity related to MOODS-SR depressive mood domain after variable selection with elasticnet.B. Test sample 1 activity and functional connectivity related to MOODS-SR depressive mood domain.
a Raw P values were significant at FDR-corrected thresholds; FDR q values are reported in the eResults.b Finding replicated in all 3 samples.c Age was z-scored for the multivariable regression analyses.

Post-hoc sensitivity analyses with mania/hypomania risk in participants without MDD. A.
Discovery sample activity and functional connectivity related to MOODS-SR manic mood domain after variable selection with elasticnet.B. Test sample 1 activity and functional connectivity related to MOODS-SR manic mood domain.C.
a Raw P values were significant at FDR-corrected thresholds; FDR q values are reported in the eResults.b Finding replicated in all 3 samples.c Age was z-scored for the multivariable regression analyses.

Post-hoc sensitivity analyses with depression risk in participants without MDD. A.
Discovery sample activity and functional connectivity related to MOODS-SR depressive mood domain after variable selection with elasticnet.B. Test sample 1 activity and functional connectivity related to MOODS-SR depressive mood domain.C. Test sample 2 activity and functional connectivity related to MOODS-SR depressive mood domain.Abbreviations: dACC, dorsal anterior cingulate cortex; dlPFC, dorsolateral prefrontal cortex; Exp, exponentiated; Exp(β), incidence rate ratio or a 1-unit change in independent variable is an Exp(β) increase in the dependent variable (depression risk score); L, left; MDD, Major Depressive Disorder; mPFC, medial prefrontal cortex; N/A, not applicable; R, right; vlPFC, ventrolateral prefrontal cortex.
a Raw P values were significant at FDR-corrected thresholds; FDR q values are reported in the eResults.bFindingreplicated in all 3 samples.c Age was z-scored for the multivariable regression analyses.

Post-hoc sensitivity analyses with mania/hypomania risk in participants without ADHD. A.
Discovery sample activity and functional connectivity related to MOODS-SR manic mood domain after variable selection with elasticnet.B. Test sample 1 activity and functional connectivity related to MOODS-SR manic mood domain.C. Test sample 2 activity and functional connectivity related to MOODS-SR manic mood domain.Abbreviations: ADHD, Attention Deficit Hyperactivity Disorder; dACC, dorsal anterior cingulate cortex; dlPFC, dorsolateral prefrontal cortex; Exp, exponentiated; Exp(β), incidence rate ratio or a 1-unit change in independent variable is an Exp(β) increase in the dependent variable (mania risk score); L, left; mPFC, medial prefrontal cortex; N/A, not applicable; R, right; vlPFC, ventrolateral prefrontal cortex.
a Raw P values were significant at FDR-corrected thresholds; FDR q values are reported in the eResults.b Finding replicated in all 3 samples.c Age was z-scored for the multivariable regression analyses.

eTable 10. Post-hoc sensitivity analyses with depression risk in participants without ADHD. A.
Discovery sample activity and functional connectivity related to MOODS-SR depressive mood domain after variable selection with elasticnet.B. Test sample 1 activity and functional connectivity related to MOODS-SR depressive mood domain.C. Test sample 2 activity and functional connectivity related to MOODS-SR depressive mood domain.Abbreviations: ADHD, Attention Deficit Hyperactivity Disorder; dACC, dorsal anterior cingulate cortex; dlPFC, dorsolateral prefrontal cortex; Exp, exponentiated; Exp(β), incidence rate ratio or a 1-unit change in independent variable is an Exp(β) increase in the dependent variable (depression risk score); L, left; mPFC, medial prefrontal cortex; N/A=not applicable; R, right; vlPFC, ventrolateral prefrontal cortex.
a Raw P values were significant at FDR-corrected thresholds; FDR q values are reported in the eResults.bFindingreplicatedin all 3 samples.cAge was z-scored for the multivariable regression analyses.©2023 Schumer MC et al.JAMA Psychiatry.

eTable 11. Significant activity and functional connectivity to all facial emotions in the three samples.
Abbreviations: dACC, dorsal anterior cingulate cortex; dlPFC, dorsolateral prefrontal cortex; FWE, family-wise error corrected; k, cluster size; L, left; MNI, Montreal Neuroimaging Institute; mPFC, medial prefrontal cortex; OFC, orbitofrontal cortex; R, right; vACC, ventral anterior cingulate cortex; vlPFC, ventrolateral prefrontal cortex.Cluster identified in elastic-net variable selection models in the discovery sample.b Cluster replicated in multivariable regression models in one or both of the independent test samples.c Nearest peak and associated t-statistic are reported.

eTable 12. All facial emotions analysis results for mania/hypomania risk. A.
Discovery sample activity and functional connectivity related to MOODS-SR manic mood domain after variable selection with elastic-net.B. Test sample 1 activity and functional connectivity related to MOODS-SR manic mood domain.C. Test sample 2 activity and functional connectivity related to MOODS-SR manic mood domain.Abbreviations: dACC, dorsal anterior cingulate cortex; dlPFC, dorsolateral prefrontal cortex; Exp, exponentiated; Exp(β), incidence rate ratio or a 1-unit change in independent variable is an Exp(β) increase in the dependent variable (mania risk score); L, left; mPFC, medial prefrontal cortex; N/A, not applicable; R, right; vlPFC, ventrolateral prefrontal cortex.Raw P-values were significant at FDR-corrected thresholds; FDR q-values are reported in the eResults.b Age was z-scored for the multivariable regression analyses. a

eTable 13. All facial emotions analysis results for depression risk
. A. Discovery sample activity and functional connectivity related to MOODS-SR depressive mood domain after variable selection with elastic-net.B. Test sample 1 activity and functional connectivity related to MOODS-SR depressive mood domain.C. Test sample 2 activity and functional connectivity related to MOODS-SR depressive mood domain.Abbreviations: dACC, dorsal anterior cingulate cortex; dlPFC, dorsolateral prefrontal cortex; Exp, exponentiated; Exp(β), incidence rate ratio or a 1-unit change in independent variable is an Exp(β) increase in the dependent variable (depression risk score); L, left; mPFC, medial prefrontal cortex; N/A, not applicable; R, right; OFC, orbitofrontal cortex; vlPFC, ventrolateral prefrontal cortex.Raw P-values were significant at FDR-corrected thresholds; FDR q-values are reported in the eResults.b Age was z-scored for the multivariable regression analyses. a

MOODS-SR manic and depressive mood domain correlations with state anxiety, depression severity, and mania/hypomania severity
. State anxiety was measured by the Hamilton Anxiety Rating Scale 9 .Depression severity was measured by the Hamilton Rating Scale for Depression 8 .Mania/hypomania severity was measured by the Young Mania Rating Scale 7 .A. Discovery sample Pearson correlations.B. Test sample 1 Pearson correlations.C. Test sample 2 Pearson correlations.Abbreviations: HAMA, Hamilton Anxiety Rating Scale; HRSD, Hamilton Rating Scale for Depression; MOODS-SR, Mood Spectrum Self-Report; YMRS, Young Mania Rating Scale.HAMA score missing for n=1 Test sample I participant.d HAMA score missing for n=1 Test sample II participant.e YMRS score missing for n=1 Test sample II participant.

eTable 18. Family psychiatric history information for the three samples.
There were too few participants to provide numbers without compromising identifiability.Number of participants with a family history of Bipolar Disorder, Major Depressive Disorder, Attention Deficit Hyperactivity Disorder, or anxiety disorder. a