Relationships between depressive symptoms and brain responses during emotional movie viewing emerge in adolescence

Journal Pre-proof Relationships between depressive symptoms and brain responses during emotional movie viewing emerge in adolescence David C. Gruskin, Monica D. Rosenberg, Avram J. Holmes PII:

S1053-8119(19)30808-0

DOI:

https://doi.org/10.1016/j.neuroimage.2019.116217

Reference:

YNIMG 116217

To appear in:

NeuroImage

Received Date: 14 March 2019 Revised Date:

14 August 2019

Accepted Date: 19 September 2019

Please cite this article as: Gruskin, D.C., Rosenberg, M.D., Holmes, A.J., Relationships between depressive symptoms and brain responses during emotional movie viewing emerge in adolescence, NeuroImage (2019), doi: https://doi.org/10.1016/j.neuroimage.2019.116217. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2019 Published by Elsevier Inc.

Title: Relationships between depressive symptoms and brain responses during emotional movie viewing emerge in adolescence Authors: David C. Gruskina, Monica D. Rosenberga,b, Avram J. Holmesa,c* Affiliations: aDepartment of Psychology, Yale University, New Haven, Connecticut 06520, USA b

Department of Psychology, University of Chicago, Chicago, IL 60637, USA

c

Department of Psychiatry, Yale University, New Haven, Connecticut 06520, USA

*Corresponding author. Author email addresses: [email protected]; [email protected]; [email protected]

Correspondence: Avram J. Holmes, Yale University, Department of Psychology, 402 Sheffield Sterling Strathcona Hall, 1 Prospect Street, New Haven, CT 06511, Phone: 203436-9240, Email: [email protected].

Keywords: Depression, Brain development; Emotion; Naturalistic fMRI; Adolescence; Inter-subject correlation

*Color should be used for all figures in print*

1

ABSTRACT Affective disorders such as major depression are common but serious illnesses characterized by altered processing of emotional information. Although the frequency and severity of depressive symptoms increase dramatically over the course of childhood and adolescence, much of our understanding of their neurobiological bases comes from work characterizing adults’ responses to static emotional information. As a consequence, relationships between depressive brain phenotypes and naturalistic emotional processing, as well as the manner in which these associations emerge over the lifespan, remain poorly understood. Here, we apply static and dynamic inter-subject correlation analyses to examine how brain function is associated with clinical and non-clinical depressive symptom severity in 112 children and adolescents (7-21 years old) who viewed an emotionally evocative clip from the film Despicable Me during functional MRI. Our results reveal that adolescents with greater depressive symptom severity exhibit atypical fMRI responses during movie viewing, and that this effect is stronger during less emotional moments of the movie. Furthermore, adolescents with more similar item-level depressive symptom profiles showed more similar brain responses during movie viewing. In contrast, children’s depressive symptom severity and profiles were unrelated to their brain response typicality or similarity. Together, these results indicate a developmental change in the relationships between brain function and depressive symptoms from childhood through adolescence. Our findings suggest that depressive symptoms may shape how the brain responds to complex emotional information in a dynamic manner sensitive to both developmental stage and affective context.

2

INTRODUCTION When it comes to interpreting emotional events, truth is subjective. Across individuals, variability in state and trait psychological characteristics can influence the ways in which we attend to, interpret, and make decisions about affective information. In their pathological form, biases in emotional information processing, including increased focus on negative stimuli and overly pessimistic interpretations of emotional events, reflect core features of affective illnesses such as major depressive disorder (1–3). Depression symptom prevalence, however, is not static across the lifespan. Rather, risk for illness onset increases dramatically during adolescence (4, 5), and depression is currently the largest single contributor to adolescent disability (6). Although substantial progress has been made in identifying psychological characteristics and cognitive biases that increase the likelihood of future mental health problems in children and adolescents (7), relationships between individual responses to complex emotional stimuli, symptoms of depression, and neurodevelopmental processes remain poorly understood. Accordingly, the aim of the present study is to investigate relationships between brain responses to emotional information and depression symptom profiles in children and adolescents. Underpinning the emergence of affective symptoms and associated illness risk, human brain development is influenced by a complex series of dynamic processes ranging from shifts in profiles of gene expression to changes in environmental pressures (8, 9). The passage from childhood to adulthood, in particular, is marked by significant social and biological transitions (10, 11), including hierarchical changes in brain structure and function that underlie the gradual development of adaptive emotion reactivity and regulation (12). As one example, the staggered development of the amygdala and the medial prefrontal cortex (mPFC) contributes to prominent increases in emotional difficulties that characterize adolescence (12). This scheduled maturation of subcortical and cortical systems is reflected in a developmental switch in amygdala-mPFC functional connectivity and alterations in amygdala responses to emotionally evocative stimuli (13–15). The characterization of age-dependent changes in brain function would provide a tremendous opportunity to understand how neurodevelopment shapes both individual differences in emotion processing and the emergence of depression. Developmental changes with consequences for individual differences in affect are pronounced in frontolimbic circuitry but are also evident in widely distributed patterns of brain function. In adults, distributed patterns of brain functioning reflect individual symptom profiles of patients with depression (16, 17) and other psychiatric disorders (18, 19). Much

3

like a fingerprint, spatiotemporal patterns of brain function allow for the identification of individuals within a broader population (20) and the prediction of behavioral phenotypes across cognition, personality, and emotion (21, 22). These individual-specific functional profiles emerge over the course of development, settling into a more stable, idiosyncratic pattern during adolescence (23). Delays in this process of differentiation are associated with the expression of psychiatric symptoms, including those related to depression (23). Accordingly, data-driven approaches that consider the broad neurobiological changes associated with adolescence can complement work focused on specific regions or circuits of interest to illuminate the sources of individual variation that contribute to symptom burden and risk for depression across development. Embedded within the gradual process of neurodevelopment, brain responses continually change over shorter time-scales as individuals react to, and interact with, their environments. The diverse repertoire of these moment-to-moment fluctuations in brain activity and functional connectivity are instrumental for higher-order cognition (24, 25). Furthermore, impairments in functional brain dynamics have been implicated in psychiatric illness (18, 26). Although task-based and task-free fMRI (functional magnetic resonance imaging) have dominated modern research on the biological basis of depression, much of the work in these domains has relied on static analytic approaches that assume stable brain responses across time (27). An emerging dynamic paradigm that combines the targeted nature of task-based imaging with the minimal demands of resting-state research is movie-watching fMRI (28, 29). In addition to providing high-quality data by reducing head motion and increasing compliance (28, 30), movie-watching fMRI facilitates the study of naturalistic affective processing. Movies, with their dynamic, feature-rich content, can elicit emotional responses similar to those experienced in real-world situations (29, 31). A powerful approach used to study these film-induced brain responses is inter-subject correlation (ISC) (32). This method describes the similarity between multiple individuals’ brain responses to a time-locked stimulus. ISC has proven useful for identifying the possible brain bases of emotion processing in healthy individuals (33–35) and group differences in adult depression (36, 37). However, relationships between brain responses to naturalistic affective stimuli and depressive symptoms remain poorly understood, especially in developing populations. To characterize relationships between functional brain responses to emotional information and depression symptoms, we analyzed neuroimaging and questionnaire data acquired from 112 children and adolescents by the Healthy Brain Network project (38).

4

Informed by recent work showing a distributed decrease in ISC in adults with depression (36), we tested the hypotheses that participants with greater depressive symptom severity would exhibit lower ISC, and that pairs of participants who were more similar in their depressive symptom profiles would share more similar fMRI responses to the movie. First, we show that synchronization of brain activity, indexed with ISC, scales with depressive symptom severity in adolescents but not children. Next, we demonstrate that the synchrony of brain responses, as well as the relationship between this synchrony and depressive symptoms, is influenced by the emotional content of the movie. Finally, we show that patterns of movie-evoked brain activity are common between adolescents, but not children, with similar item-level depressive symptom profiles. These results suggest that depressive symptoms influence how the brain responds to emotional information in a dynamic manner that is sensitive to both developmental stage and moment-to-moment fluctuations in affective content.

METHODS To understand relationships between naturalistic emotional processing and depressive symptoms across development, we characterized children’s and adolescents’ fMRI signal responses to an emotionally evocative clip from the film Despicable Me (Fig. 1). Specifically, we asked whether the “typicality” of an individual’s blood-oxygenation-leveldependent (BOLD) signal time-courses—that is, their similarity to the rest of the group— was related to their depressive symptom severity measured with self-report responses to the Moods and Feelings Questionnaire (MFQ-SR) (39). We next asked whether relationships between spatiotemporal activity patterns and depressive symptoms were influenced by the emotional content of the film, and whether individuals with more similar patterns of brain activity reported more similar symptom patterns. Of note, our participant sample includes children and adolescents with and without diagnoses of depression and other comorbid disorders (Supplementary Table 1). Here we focus on symptom severity rather than clinical diagnosis to characterize dimensional, transdiagnostic relationships between brain function and behavior.

5

Figure 1. Typicality, dynamic typicality, and pairwise similarity analysis pipelines. (A) Typicality analysis: Functional typicality for each participant and each brain region (parcel) was assessed with the Pearson correlation between each individual’s BOLD response time-course and the group average time-course from the remaining participants. The resulting functional typicality vector was Spearman-correlated with depressive symptom scores. Permutation testing was used to assess significance. (B) Dynamic typicality analysis: ISC was estimated at each functional time point using a sliding window/weighted average approach (see Methods). The typicality correlation described in (A) was applied at each time point. The association between fluctuations in emotional content and the functional typicality/symptom severity relationship was evaluated using feasible Generalized Least Squares (fGLS) regression. (C) Pairwise similarity analysis: BOLD timecourse similarity between every pair of participants was calculated via Pearson correlation. The resulting values were related to similarity in depressive symptom profiles through Spearman correlations and significance was determined through permutation testing.

6

Participants Neuroimaging and questionnaire data from 563 participants were downloaded from the data portal for the Child Mind Institute’s Healthy Brain Network (HBN) project (38), a large, ongoing initiative that has been collecting data from a high-risk community sample of children and adolescents with perceived clinical concern. All participants provided written consent or assent, and consent was obtained from the parents or legal guardians of participants younger than 18 years old. The HBN project was approved by the Chesapeake Institutional Review Board. Of all the participants whose anatomical scans passed visual inspection (n=324) and had complete Despicable Me data (n=313), those with acceptable levels of head motion (defined a priori as maximum head displacement <3 mm and mean framewise head displacement <.15 mm on the functional scan; n=153) were included in our analyses. Finally, participants without self-report scores on the Moods and Feelings Questionnaire (n=41) were excluded, leaving 112 participants (65 F; mean age=12.7±3.3 years, range=7-21 years) for the final analyses.

MRI data collection All MRI data used in this study were collected at the HBN Rutgers University Brain Imaging Center site on a Siemens 3T Tim Trio magnet. Key parameters for the functional scan are as follows: TR=800ms, TE=30ms, # slices=60, flip angle=31˚, # volumes=750, voxel size=2.4mm. Complete information regarding the scan parameters used for the Healthy Brain Network project can be found at: http://fcon_1000.projects.nitrc.org/indi/cmi_healthy_brain_network/mri_protocol.html#mriscan-parameters.

Depressive symptom inventory Self-report responses to the Moods and Feelings Questionnaire (MFQ-SR) were used to quantify depressive symptom severity and profiles. The MFQ-SR is a 33-item questionnaire used to inventory core depressive symptomatology in clinical and sub-clinical pediatric populations (39). All HBN participants over eight years of age as well as a subset of seven-year-olds were administered the MFQ-SR. Missing item-level responses (3/3696) were imputed based on the mean response across the full sample. Mean MFQ-SR scores did not differ between the child and adolescent age-groups (child: M=14.7±9.4; adolescent: M=13.4±11.4; t110=0.65, P=0.52), and a Kolmogorov-Smirnov test revealed no significant difference in score distributions between the groups (D=.20, P=.18). Independent samples

7

t-tests revealed that mean item-level responses were indistinguishable between groups on

all but one question (individual question |t-statistics|<1.98, Ps>.05; MFQ item 13 [“I was talking more slowly than usual”]: t110 =2.16, P=.03), suggesting that average depressive symptom profiles were consistent across the two groups. Although the MFQ-SR authors do not recommend a specific diagnostic cut-off, previous work has determined that a score of 29 can distinguish individuals experiencing a current major depressive event (40). All participants were also administered the KSADS-COMP (41), a semi-structured DSM-5-based psychiatric interview, by a licensed clinician. Consensus DSM-5 diagnoses were then generated for each participant following the completion of the clinical interview and other study procedures. In the adolescent group, 12/50 of the participants received at least one depressive disorder diagnosis, as opposed to 6/62 of the participants in the child group. Because both groups shared similar depression score distributions but differed in their proportions of diagnoses, we calculated Cronbach’s alpha for MFQ-SR responses in both groups as a way of ensuring that our participants were accurate self-reporters of their symptoms. Internal reliability was high for both groups (child: alpha =.86, adolescent: alpha =.94). Data reflecting symptom onset were unavailable at the time of writing, but disease duration data were available for participants who entered the HBN with a depression diagnosis and history of treatment (6/112). Of these six participants, five were in the adolescent group (mean disease duration=2.46 years) and one was in the child group (disease duration=4.5 years). Only 16/112 (10 children, 6 adolescents) participants did not receive any consensus DSM-5 diagnosis. Reflecting the significant heterogeneity of our sample, common (non-depressive) primary diagnoses included attention deficit hyperactivity disorder (39/112), anxiety disorder (20/112), and specific learning disorder (9/112). Movie clip and emotional content ratings Functional MRI data were collected while participants viewed a ten-minute clip from the movie Despicable Me (01:02:09–01:12:09; presentation details available at http://fcon_1000.projects.nitrc.org/indi/cmi_healthy_brain_network/mri_protocol.html#mrimovies) (28). Twenty adult raters provided continuous ratings of the emotional valence of the clip on a scale from 1 to 9 (most negative to most positive) three times each using a custom PsychoPy (v3.1.2) script (42). Across all 60 trials, the average valence rating was 4.97 (5=neutral) with a standard deviation of 1.44. Inter-rater correlation (the average of every participant’s average rating time series Pearson-correlated with the corresponding

8

time series from every other participant) was 0.74, and average intra-rater rater correlation (the average correlation of every pair of rating runs for each participant) was 0.79. The 60 rating trials were individually z-scored (to account for variability in raters’ use of the numeric scale) and averaged across raters and within TRs to yield an emotional valence vector. An emotional intensity vector was then generated using the absolute value of the emotional valence vector. Line plots showing the group and individual emotional valence and intensity ratings are shown in Supplementary Figure 1. To control for the effects of non-emotional aspects of the film on ISC, one rater (author D.G.) recorded the number of faces present on the screen at each TR using a PsychoPy script (averaged across three viewings). Visual and auditory intensity (brightness and loudness) of the movie were also calculated at each TR using the SaliencyToolbox (43) and a custom Matlab script (which operationalized volume as dynamic peak signal amplitude of the clip’s audio), respectively.

fMRI data preprocessing Functional data were preprocessed in AFNI (44). Three volumes were first removed from the start of each ten-minute run. Functional images were then despiked, corrected for head motion, aligned to the corresponding skull-stripped anatomical image with a linear transformation and then to the MNI atlas via nonlinear warping, and spatially smoothed with a 4 mm full-width at half-maximum filter. Covariates of no interest, including a 24-parameter head motion model (6 motion parameters, 6 temporal derivatives, and their squares) and mean signal from subject-specific eroded white matter, ventricle, and whole-brain masks were regressed from the data. Data were band-pass filtered from .01 to .1 Hz. Temporal filtering was performed because our research question primarily concerned dynamic emotional information, which is understood to fluctuate at frequencies below .10 Hz (45). ISC is also most pronounced at low (<.10 Hz) frequencies, although ISC in frequency bands below .01 Hz could reflect spurious effects (46), thus motivating our lower boundary of .01Hz. Voxel-wise BOLD signal time-courses were averaged within regions of interest using a 268-node whole-brain parcellation (47) to reduce both the number of analyses performed and the impact of age-dependent differences in functional topography on our results. MNI coordinates and network affiliations for all of the parcels mentioned in this study can be found at https://bioimagesuiteweb.github.io/webapp/connviewer.html.

Static inter-subject correlation (ISC) analyses

9

Two static ISC analyses, evaluating participants’ functional typicality and pairwise symptom profile similarity, were performed using the parcellated fMRI data to characterize relationships between depressive symptoms and BOLD responses to the emotional movie.

Functional typicality: For every parcel, each participant’s BOLD signal time-course was z-

standardized and Pearson correlated with the average BOLD signal time-course from the rest of the group (Fig. 1A). This procedure resulted in a 112 subject x 268 parcel typicality matrix. The columns of this typicality matrix were related to MFQ-SR scores through Spearman partial correlations (controlling for age, sex, and mean framewise head motion) to reveal parcels in which BOLD signal time-course typicality was associated with depressive symptom severity. Because each participant’s BOLD time-course is used to calculate every other participant’s degree of ISC, parametric statistical tests that assume independence across measurements should not be used with ISC values (46). Statistical significance of the relationships between BOLD signal typicality and depressive symptom severity was therefore evaluated using nonparametric significance testing. First, the order of the questionnaire variables was randomly shuffled such that one participant’s typicality score was paired with a random participant’s MFQ-SR score, age, mean framewise head motion, and sex. Spearman partial correlations were then performed, and the process was repeated 10,000 times to generate a null distribution. Given that we hypothesized that adolescents would show the negative association between functional typicality and depression that had previously been reported in the adult literature (Guo et al., 2015), we conducted one-tailed significance testing. We also expected the pairwise similarity analysis to yield a similar effect due to its conceptual and mathematical overlap with the typicality analysis. The reported P values were accordingly calculated using the following one-tailed formula:

P = (1+ # of null Fisher’s z-values <= observed Fisher’s z-value)/10001

In order to test the extent to which the relationship between functional typicality and depression score was consistent across the whole brain, we averaged the Fisher’s z-values obtained for each parcel in the real data to get the observed mean rho value. We then computed the same average z-value for each permutation of the null data and performed the following one-tailed test:

10

P = (1+ # of null mean Fisher’s z-values <= observed mean Fisher’s z-value)/10001

Pairwise symptom profile similarity: For every possible pair of 112 participants, a

phenotypic matrix was constructed by finding the L1 (Manhattan) distance, a measure appropriate for ordinal and trinary data (e.g. Chekroud et al., 2017 (48); high L1 distance=less similar), between each pair’s item-level responses to the MFQ (Fig. 1C). Next, parcel-level BOLD time-courses from the first participant were z-standardized and Pearson-correlated with corresponding time-courses from the second to create a 112 x 112 BOLD similarity matrix for every parcel. Correlation coefficients were Fisher z-transformed, averaged across all subject pairs, and transformed back to r values to generate the average ISC maps shown in Fig. 2A. A representational similarity analysis (49) using Spearman partial correlations (controlling for differences in age, sex, and mean framewise head motion) was then performed to reveal parcels in which more similar BOLD timecourses are associated with more similar depressive symptom profiles. Significance for these results was tested using a permutation design analogous to the one used in the typicality analyses. Specifically, each participant pair’s item-level MFQ responses, as well as their similarity in age and head motion (absolute value of the difference between the two participant’s ages and mean framewise displacement values) and sex (binary classification for same vs. different), were associated with a random pair’s functional similarity correlation coefficient, and the analysis was repeated 10,000 times. Differences in the static ISC/depressive symptom maps (typicality and similarity; Supplemental Figure 2) for the two age groups were assessed by first Fisher transforming the observed correlation coefficients and the null coefficients generated during permutation testing. For each parcel, the adolescent group’s Fisher’s z-value was subtracted from that of the child group, and the significance of this difference was evaluated using the following one-tailed formula, as we expected the adolescents, being closer in age to the adults, to more strongly exhibit the negative relationship previously observed in adults (Guo et al., 2015): P = (1+ # of null differences in child − adolescent Fisher’s z-values >= observed difference

in child − adolescent Fisher’s z-value)/10001

Fisher’s z-values were then transformed back to rho values for visualization.

11

Dynamic inter-subject correlation analyses Dynamic ISC: In order to evaluate how the content of the emotional movie drives changes

in neural synchronization across participants, we performed an exploratory dynamic ISC analysis. Parcel-level dynamic ISC (i.e., functional typicality) time-courses were calculated for every participant using a tapered cosine sliding window approach (width=30 TRs, taper=15 TRs; Fig. 1B). A tapered window was selected to reduce the influence of timepoints at the extremes of the window by assigning greater weight to time-points that are closer to the window’s center. This yields an observed statistic (here, a mean of correlation values) that is not dramatically affected by the inclusion of new points as the window shifts. The static typicality analysis detailed above was initially performed on the first 24 seconds of the time-courses and then repeated after sliding the window by 1 TR until the end of the clip was reached. The sliding window parameters chosen for this analysis were motivated by prior research and represent a balance between capturing enough time points to calculate stable correlation coefficients and keeping the window short enough to isolate transient ISC dynamics (34, 50, 51). The correlation coefficients in the resulting ISC timecourses were then Fisher z-transformed to generate a 268 parcel x 718 TR matrix for each subject reflecting their parcel-level fluctuations in synchrony with the rest of the group over the course of the movie clip. To relate the windowed ISC time-courses to the movie vectors (emotional intensity and emotional valence, controlling for the number of faces on the screen, brightness, and loudness), estimates of ISC at each TR were obtained by first identifying every window in which a given TR was included and then taking the weighted average of those windows’ ISC z-values (with the weights being derived from the tapered window) (50). This approach yielded a 268 parcel x 747 TR matrix for each subject containing dynamic ISC time-courses for every parcel. The parcel x TR matrices were then z-standardized, averaged across participants, and zero-padded with three columns to account for the TRs discarded at the beginning of the run. The first and final 15 TRs were excluded from the following analyses as the ISC estimates for these TRs were based on a small number of windows, and 6 TRs were removed to account for hemodynamic delay, yielding 714 TRs for regression analysis. Given the high degree of autocorrelation present in the movie feature and ISC timecourses, feasible Generalized Least Squares regression (fGLS) was used to identify how ISC scales with the emotional content of the movie. FGLS accounts for autocorrelation by minimizing the squared Mahalanobis distance of the ordinary least squares (OLS) error

12

terms based on an estimation of their variance-covariance matrix, and GLS has previously been leveraged to analyze fMRI/movie feature time-courses (52). A fourth-order autoregressive model was used to estimate the covariance structure of the OLS error terms based on a visual inspection of the partial sample autocorrelation functions of the OLS residual time series according to the Box-Jenkins methodology (53). Of note, the residuals for each fGLS regression have a different autocorrelation structure. However, because much of the autocorrelation in the fMRI data was imposed by our analyses in a uniform fashion across all parcels, these differences were not extreme. In addition, although visual inspection revealed that some regressions may have been better characterized with higherorder AR models, lower p-values were not associated with higher degrees of autocorrelation for any of our analyses. Thus, using a fourth-order model did not bias our results towards significance. Importantly, because relationships between clinical symptoms and high-level features of the stimulus such as emotional content (here, emotional intensity and valence) can be confounded by lower-level features such as visual intensity (54), time-courses of low-level stimulus features (brightness and loudness) as well as baseline social content (faces) were also included in the fGLS regression. A t distribution (709 degrees of freedom) was used for two-tailed significance testing of the standardized regression coefficients. Two tailed-significance testing was performed for the dynamic analyses because we did not have strong a priori reasons to expect that the fGLS results would be restricted to a particular direction (positive or negative).

Dynamic typicality: The typicality/symptom severity analysis detailed above was repeated

at each TR using the (non-normalized) window-derived ISC estimates (Fig. 1B). The resulting time-courses (714 TRs) were then entered into two fGLS regression models to characterize how fluctuations in the strength of the typicality/symptom severity relationship are related to the moment-to-moment emotional intensity and valence of the movie (controlling for brightness, loudness, and baseline social content). Age-related differences in the effects of emotional movie content on ISC and the ISC/symptom severity relationship were evaluated by dividing the child – adolescent difference in standardized regression coefficients by the square root of the sum of the coefficients’ squared standard errors. Two-tailed significance testing of the resulting z scores was conducted against a normal distribution.

13

All analyses were conducted in Matlab. Data visualizations were generated using Connectome Workbench following projection of the volumetric Shen atlas parcels to the cortical surface (v1.3.1) (55).

RESULTS Greater depressive symptom severity is associated with reduced functional typicality during emotional movie viewing in adolescence The free viewing of an emotionally evocative Despicable Me clip elicited stereotyped BOLD responses across participants, as measured by pairwise inter-subject correlation (Fig. 2A). Consistent with recent work from Guo and colleagues in adults (36), greater self-reported depressive symptom severity, measured with the MFQ-SR, was associated with participant-specific brain responses that were less similar to the group average (Fig. 3A). In other words, individuals with more severe depressive symptoms showed less typical fMRI response time-courses, a measure we call “functional typicality” (Fig. 1A). This inverse relationship was preferentially expressed in 6 out of 268 parcels derived from a whole-brain parcellation (47), encompassing aspects of left dorsolateral prefrontal cortex (dlPFC; ρ=−.27, P<.01, uncorrected; Fig. 3A; Table S2) and medial temporal lobe (ρ=−.28, P<.01, uncorrected).

14

Figure 2. Parcel-level inter-subject correlations and depression score distributions across childhood and adolescence. (A) Inter-subject correlation analyses reveal consistent participant responses to movie viewing with a non-uniform spatial distribution across cortex. (B) Depression (MFQ-SR) score means and distributions were similar across the child (14.7±9.4 [SD]) and adolescent (13.4±11.4) groups (t110=0.65, P=0.52; KolmogorovSmirnov test; D=.20, P=.18).

15

The transition from childhood to adolescence is marked by changes in brain and behavioral responses to emotionally evocative and naturalistic stimuli (14, 50, 56), including age-dependent shifts in functional synchronization. Accordingly, analyses that group children and adolescents together could fail to identify age-dependent associations. To examine how the relationship between depressive symptoms and naturalistic emotion processing changes with development, we divided the full sample into two age-groups (children: 7-12 years old, n=62; adolescents: 13-21 years old, n=50) using a split point informed by prior work (57) and repeated the functional typicality analysis separately in each group (Fig. 3B). Findings in the adolescent group mirrored those observed across the full sample, such that depressive symptom severity was negatively associated with the typicality of movie-evoked functional time-courses in 8/268 parcels (e.g. left orbitofrontal cortex, ρ=−.36, P<.01, uncorrected, left hippocampus, ρ=−.37, P<.01, uncorrected; Table S14). This relationship was consistent across the brain such that the average correlation coefficient across all parcels was significantly stronger than that of a corresponding null distribution (whole-brain average ρ=−.15, P<.05; Fig. 3C). We report uncorrected p-values in Figures 3 and 5 as the presence of a whole-brain effect provided additional evidence for our results not captured by frequentist statistics. However, we note that in our parcel-level follow-up analyses, ~2 false positive relationships can be expected per sub-figure (each visualizing 197 cortical parcels) given our P<.01 threshold. Data tables detailing our results for all cortical and subcortical/cerebellar parcels (which are not visualized in Figures 3−5) that passed this threshold are available in the Supplementary Materials and contain FDR adjusted P values. In the child group, on the other hand, no parcel-level correlations between symptom severity and functional typicality were significant at P<.01, and the mean correlation coefficient across all parcels was nominally positive (parcel-level |ρschildren|<.33, Ps>.01, uncorrected; whole-brain average ρ=.005, P=.53). The whole-brain relationship between symptom severity and functional typicality nominally differed between the child and adolescent groups (whole-brain average ρchildren−ρadolescents=.15, P=.08; Supplemental Figure 2). Importantly, MFQ score means and distributions as well as responses to 32 of the 33 individual scale items did not significantly differ between the child (14.7±9.4[SD]) and adolescent (13.4±11.4) groups (mean MFQ scores: t110=0.65, P=.52; KolmogorovSmirnov D=.20, P=.18; individual question |t-statistics|<1.98, Ps>.05; MFQ item 13: “I was talking more slowly than usual”, t110=2.16, P=.03), suggesting that differences in the

16

functional typicality/symptom severity relationship cannot be explained solely by disparities in symptom severity or profiles between the two groups (Fig. 2B).

17

Figure 3. Functional typicality is related to depressive symptom severity in adolescence. (A) Spearman partial correlations controlling for age and sex reveal parcels in which less typical BOLD time-courses are associated with greater (blue) depressive symptom severity in the whole sample (bordered parcels significant at P<.01 uncorrected). (B) The same analysis shown in (A), repeated for the child and adolescent groups. (C) The histograms reflect permutation tests demonstrating that the observed mean functional typicality/symptom severity correlation coefficient across all parcels was stronger than that from the corresponding null distribution in the adolescent, but not child, group (whole-brain averages: ρchildren=.005, P=.53; ρadolescents=−.15, P<.05).

18

Emotional movie content increases functional synchrony Our analyses revealed a relatively consistent pattern of time-varying brain activity across participants during movie viewing, replicating prior work in this domain (32). However, the specific stimulus features that evoke this inter-subject consistency have yet to be characterized in developmental populations. To test whether dynamic emotional movie content drives functional synchronization, we first estimated fluctuations in ISC using a Tukey sliding window approach (Fig. 1B; 24-second window, 30 time points; see Methods for details). This resulted in functional typicality scores for every participant and every fMRI volume, which we averaged across individuals to generate a dynamic ISC time-course for each parcel in our whole-brain atlas. Next, we related these dynamic ISC time-courses to independent ratings of the emotional content of the movie using feasible Generalized Least Squares (fGLS) regression, a technique that addresses the autocorrelation present in the movie and imaging data (52). We report results that pass a Benjamini-Hochberg False Discovery Rate (FDR) adjusted threshold of P<.05 for all fGLS analyses. Although the aim of fGLS is to improve on OLS regression’s tendency to yield spurious relationships when given serially correlated inputs, this approach has not been widely applied in functional MRI research and thus should be considered exploratory. Consistent with prior work in adults (34), emotional intensity (the absolute value of the emotional valence ratings) was positively associated with ISC in the full participant sample (age 7–21) such that brain synchrony across individuals was greater during more emotionally evocative moments of the movie (Fig. 4). This relationship was strongest in parcels encompassing aspects of frontolimbic circuitry implicated in emotion processing, including right anterior insula (β=0.09, P<.05, FDR adjusted, Table S4) and anterior cingulate cortex (β=0.11, P<.05, FDR adjusted). Conversely, emotional valence was negatively associated with ISC in several parcels, including bilateral dlPFC (right hemisphere β=−0.15, P<.01, FDR adjusted; left hemisphere β=−0.12, P<.01, FDR adjusted; Table S5) and anterior insula (right hemisphere β=−0.15, P<.01, FDR adjusted; left hemisphere β=−0.09, P<.05, FDR adjusted). When independently considering children and adolescents, the dynamic ISC findings were similar across both groups, with positive inter-group spatial correlations for both ISC/emotional intensity (ρ=.23, P<.001) and ISC/emotional valence relationships (ρ =.14, P<.05; see Supplementary Figure 3 for age-group-specific maps). While the topographic distributions of effects were preserved across the two age-groups, differences

19

in effect size were apparent (Supplementary Figure 4) such that the relationship between increased synchronization and more emotional content was stronger in children than in adolescents. This age-related difference was notable in several regions including bilateral mPFC (right hemisphere z=−3.63, P<.01, FDR adjusted; left hemisphere z=−4.42, P<.001, FDR adjusted) and dlPFC (right hemisphere z=−3.67, P<.01, FDR adjusted; left hemisphere z=−4.46, P<.001, FDR adjusted). These results suggest that emotional film content synchronized brain responses in both an age-invariant (topography) and an agespecific (effect size) manner.

20

Figure 4: Fluctuations in emotional movie content influence both inter-subject correlations and the relationship between functional typicality and depressive symptom severity (full sample). Across the full sample, more emotionally charged moments of the movie were, on average, associated with stronger inter-subject correlations (top left) and weaker functional typicality/depressive symptom severity relationships (bottom left), as determined by fGLS regression (parcel threshold P<.05, FDR adjusted). BOLD synchronization was higher in several parcels when the clip was more negatively valenced (top middle). Positively valenced moments of the movie maximized the inverse relationship between functional typicality and symptom severity in more parcels than did negatively valenced moments (bottom middle). Line plots showing the emotional valence/ISC and emotional valence/functional typicality fGLS regressors in two exemplar parcels are included for visualization purposes (right).

21

Emotional movie content weakens the relationship between ISC and depressive symptom severity in adolescence Developmental studies have reported information processing biases in depression, including increased engagement with negative information and excessively negative interpretations of emotional events (58, 59). Our analyses suggest that inter-subject synchronization is related to the emotional content of the stimulus. Given that individuals with depression can exhibit emotion-driven attentional and cognitive biases, the atypical BOLD time-courses we observed in more depressive individuals may reflect dynamic biases in attention to, appraisal of, or difficulty disengaging from emotional information. Consequently, we hypothesized that the emotional movie content would dynamically influence the relationship between functional typicality and depressive symptom severity. Using the dynamic ISC time-courses described above (Fig. 1B), we first calculated the relationship between functional typicality and symptom severity at each time point. We next related the resulting time-courses of correlation coefficients (reflecting the fluctuating strength of the functional typicality/symptom severity relationship) to the emotional intensity and emotional valence time-courses using fGLS. The functional typicality/symptom severity relationship was stronger (more negative) during less emotional moments of the movie in parcels encompassing aspects of bilateral anterior insula (right hemisphere β=0.11, P<.05, FDR adjusted, left hemisphere β=0.10, P<.05, FDR adjusted; Fig. 4; Table S6), left mPFC (β=0.12, P<.05, FDR adjusted), and right PCC (β=0.13, P<.01, FDR adjusted). Similar to the dynamic ISC results, emotional valence was negatively associated with the strength of the functional typicality/symptom severity relationship in 8/268 parcels, including left ACC ( β=−0.13, P<.05, FDR adjusted; Table S7) and dlPFC (β=−0.18, P<.001, FDR adjusted).

Adolescents with similar depressive symptom profiles share more similar brain responses during emotional movie viewing Our results demonstrate that functional typicality during movie watching is related to depressive symptom severity in adolescence. However, depression is a heterogeneous syndrome and the assessment of gross depressive symptom severity can mask the presence of diverse symptom profiles. This is especially the case during development, when substantial variability in emotional, social, and cognitive maturity is thought to give rise to diverse clinical presentations (60). Recent work suggests that different depressive symptom profiles track with distinct patterns of brain function (16, 17) and cognitive

22

impairments (61). Accordingly, two individuals who share similar levels of depressive symptom severity might express qualitatively different symptom profiles and associated brain phenotypes. To determine whether BOLD time-courses during emotional movie watching index depressive symptom profiles, we used the L1 (Manhattan) distance to investigate whether pairs of individuals who were more alike in their item-level responses to the MFQ-SR also exhibited more similar functional time-courses (Fig. 1C). In the full sample (age 7–21), this pairwise similarity analysis revealed that participant pairs with more similar symptom profiles were also more similar in terms of their movie-induced BOLD activity in a subset of parcels (controlling for similarity in age, sex, and mean framewise head motion; Fig. 5A). The hypothesized global effect was present in the adolescent group (whole-brain average ρ=−.05, P<.05; Fig. 5C), where it was preferentially expressed in right orbitofrontal cortex ( ρ=−.11, P<.01, uncorrected; Fig. 5B; Table S15), right precuneus (ρ=−.18, P<.01, uncorrected), and left angular gyrus (ρ=−.13, P<.01, uncorrected). Analogous to our typicality findings, there were no global or parcel-level relationships between symptom profile and functional similarity in children (whole-brain average ρchildren=.004, P=.54; parcel-level |ρschildren|<.18, Ps>.01, uncorrected), resulting in a nominal difference in wholebrain effects between the two groups (whole-brain average ρchildren−ρadolescents=.06, P=.08; Supplemental Figure 1). Of note, the relationships between symptom profile similarity and functional time-course correspondence in adolescents were largely accounted for by similarity in overall symptom severity. Controlling for the absolute value of the difference between every two participants’ MFQ scores considerably diminished the whole-brain average of symptom profile/time-course similarity relationships in the adolescent group (ρ =−.03, P=.14). However, the nominal negative direction of this relationship when controlling for similarity in symptom severity suggests that while depressive symptom severity and profile similarity are largely overlapping, symptom profiles may reflect additional unique aspects of brain function.

23

Figure 5: BOLD time-course similarity scales with depressive symptom similarity in adolescents. Spearman partial correlations controlling for similarity in age and sex reveal parcels in which pairwise BOLD time-course similarity is associated with more similar (blue) depressive symptom profiles in the whole sample (bordered parcels significant at P<.01, uncorrected). (B) The same analysis shown in (A), repeated for the child and adolescent groups. (C) The histograms reflect permutation tests demonstrating that the observed mean functional similarity/symptom profile similarity correlation coefficient across all parcels was stronger than that from a corresponding null distribution in the adolescent, but not child, group (whole-brain averages: ρchildren=.004, P=.54; ρadolescents=−.05, P<.05).

24

DISCUSSION Children and adolescents with severe depressive symptoms often describe viewing the world around them through a filter, focusing on negative information and forming overly pessimistic interpretations of the thoughts and actions of others (59). A fundamental question facing human neuroscience is how these emotional information processing biases and the associated expression of depression symptoms emerge during development. Here we extend research on the brain mechanisms underlying this phenomenon. First, we demonstrated that greater depressive symptomatology was associated with less typical brain responses to an emotional clip from the film Despicable Me in adolescents but not children. Second, we established that the strength of this brain/behavior relationship scaled with the moment-to-moment affective content of the movie, such that the depressive symptom severity and functional typicality association was more prominent when the movie was less emotional. Finally, we found that adolescents with more similar depressive symptom profiles shared more similar brain responses to the movie. Collectively, these results highlight time-varying features of brain function and emotional information processing that reflect the presence and severity of depression symptoms across development. The transition from childhood to adulthood is characterized by changes in brain function and structure that constrain moment-to-moment responses to affective information (10, 12, 62). In both children and adolescents, depression is associated with altered brain responses to static emotional images (1, 14, 63). In adults, patients with melancholic depression exhibit reduced group-level inter-subject functional synchronization during film viewing (36). Building upon this literature, our ISC typicality and similarity analyses revealed that adolescents with more severe depressive symptoms exhibited less typical brain-wide responses during emotional movie viewing, and that these responses were more similar in pairs of individuals who were more alike in their symptom profiles. These effects were evident in several regions commonly implicated in affective integration and regulation (e.g., orbitofrontal cortex, precuneus, dlPFC) (64). Prior work suggests that specific changes in amygdala–mPFC functional connectivity during emotional face viewing underlie shifts in psychiatric symptoms during the transition to adulthood (13). The distributed relationships between patterns of brain function and depression observed here provides novel insight into the neurobiological basis of affective processing in adolescents and suggests an intriguing model for the more general development of emotion-relevant regulatory systems throughout the brain.

25

Although much of the current literature treats emotion processing as time-invariant by focusing on average responses to static stimuli, the brain possesses a dynamic organizational structure that adjusts in response to explicit task demands. Here, we establish the importance of dynamic approaches to the study of emotion processing with an exploratory analysis demonstrating that the observed relationship between functional typicality and depressive symptom severity was sensitive to the emotional content of the movie. More specifically, the inverse functional typicality/symptom severity relationship was weaker when the movie was more emotionally charged. This effect was evident in several higher-order brain regions, including left mPFC and right PCC. Recent studies have found that functional similarity in vmPFC and PCC/posteromedial cortex increases when individuals share similar interpretations of an event (65, 66). Based on this literature, participants in our study with greater depressive symptoms may have formed more similar interpretations of the movie when it was more emotional, leading to relative population-level increases in ISC, thereby weakening the inverse functional typicality/symptom severity relationship. Similarly, these participants may have engaged in more mood congruent stimulus-independent cognition and/or idiosyncratic interpretation during less emotional (and perhaps less engaging) moments, which could account for the inverse association between functional typicality and symptom severity observed across the clip. Although speculative, this candidate explanation aligns with interpretation bias, a phenomenon in which depressed individuals adopt rigid, overly pessimistic responses to emotional (and especially negatively valenced) information (3, 59, 67, 68). This set of findings deserves close consideration in future studies given the preliminary nature of our fGLS analyses as well as the fact that interpretation bias represents just one possible explanation for these observed effects. For example, it may be the case that brain function and symptom severity were less related during emotionally negative compared to positive moments of the clip because more depressed individuals may preferentially attend to negative stimuli, regardless of how those stimuli are interpreted. Psychiatric research has increasingly focused on a dimensional perspective of illness that incorporates transdiagnostic conceptions of neurobiology and behavior (69). This approach is particularly relevant to the study of major depressive disorder, a heterogeneous syndrome with hundreds of possible clinical presentations (70). Although the relationships between symptom profile and BOLD time course similarity were attenuated by controlling for overall symptom severity, the fact that the corresponding whole-brain relationship still trended in the expected direction suggests that an individual’s

26

brain responses to naturalistic stimuli may be sensitive to their specific pattern of symptoms. Furthermore, the influence of symptom profile similarity on brain synchrony might be more pronounced in samples enriched for depression. Future work focusing on clinically depressed participants may be necessary to evaluate whether fMRI responses to naturalistic stimuli could contribute to symptom profile biotypes, which will be instrumental in developing new taxonomies and treatments of depression (71, 72). Understanding the neurobiological signatures of depressive symptom profiles in children and adolescents is especially crucial, as both the prevalence and consequences of depression are magnified during the transition to adulthood (59). Here, the presence of relationships between time-course typicality/similarity and depressive symptoms in adolescents and not children may reflect effects of symptom duration and/or developmental stage. More specifically, it is possible that inter-subject synchrony scales with symptomatology once individuals have spent a considerable amount of time living with depressive symptoms, irrespective of age. It is impossible to confirm or counter this explanation without being able to control for clinical features such as age of onset and time since last episode. However, the recent discovery of similar adolescent-emergent relationships between neurogenetic/functional profiles and psychiatric symptomatology (23, 73) suggests that our present findings likely relate to developmental stage. One such relationship was identified by Kaufmann and colleagues, who reported that while adolescents with increased overall psychiatric symptom severity exhibited decreased resting-state connectome distinctiveness compared to healthy individuals, no significant relationship between these variables was found in children (23). Although the association between functional brain typicality and symptom severity was negative in our analyses, we note that functional typicality has been shown to scale positively with some clinically relevant features (trait paranoia (74)) and negatively with others (depression (36) and autism (75) symptom severity). Therefore, while functional typicality appears to serve as an index of psychopathology that emerges during adolescence, future work should clarify how these relationships may differ across psychiatric diagnoses and analytic approaches. Several limitations should be considered when evaluating the current findings. First, individual differences in ISC vary significantly with the specific characteristics of the stimulus and population being studied (29, 30). This is particularly relevant to the study of emotional processing, as films with different affective structures might reveal distinct relationships between depressive symptoms and brain function. Ongoing deep-phenotyping efforts involving a diverse array of stimuli and behavioral assays may be helpful in

27

illuminating general relationships between affective brain response and behavior (76). Second, the emotional content of the movie was rated by twenty independent adults (mean age=25 years old, s.d.=3.74 years) whose affective experiences while viewing the movie might have qualitatively differed from those of the participants. Although this would likely only hinder our ability to relate emotional film content to brain function, it is difficult to predict exactly how this might influence our analyses. Finally, the design of this study prevented us from identifying specific cognitive and affective processes associated with the ISC/depressive symptom relationships and how these relationships might change within individuals across development. These findings motivate longitudinal research to examine how age-related changes in individuals’ brain responses to naturalistic emotional stimuli relate to the emergence and maintenance of depressive symptoms. Such efforts hold considerable promise in advancing both our ability to predict clinical outcomes and our understanding of neurobiological mechanisms underlying psychiatric illness (22, 77).

CONCLUSION How the presence and severity of psychiatric symptoms colors emotional experiences across development is a central and challenging question in affective neuroscience. Our data suggest that atypical brain responses to an emotional movie may constitute functional markers of depression that emerge in adolescence and serve as signatures of item-level depressive symptom profiles. The sensitivity of the observed functional typicality/symptom severity relationship to the dynamic affective content of the movie is consistent with the proposed core role of affective information-processing biases in depression (3). Furthermore, the preferential expression of these effects in frontolimbic areas aligns with an extensive literature implicating these regions in the processing of static emotional images (1, 64, 78). However, the presence of whole-brain relationships suggests that naturalistic stimuli may afford the opportunity to characterize qualitatively different patterns of activity in response to emotional content. Due to the method’s ecological validity (29), ability to provide high quality data (28, 30), and support of sophisticated analytic techniques (28), we join a growing number of researchers in recommending that naturalistic paradigms be used to study the brain bases of psychiatric disorders. To conclude, our discovery of a developmental change in the relationships linking brain function and depressive symptoms encourages further development of naturalistic and biologically informed methods for the early detection and prevention of psychiatric illness across development.

28

ACKNOWLEDGMENTS Analyses were made possible by the high-performance computing facilities provided through the Yale Center for Research Computing and data provided by the Child Mind Institute as part of the Healthy Brain Network Biobank. We thank BJ Casey and Danilo Bzdok for their invaluable feedback on earlier versions of this manuscript, Kevin Anderson for assisting with visualizations, and members of the Yale Department of Psychology for their help rating the emotional content of Despicable Me. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. The authors have no competing interests to declare.

29

REFERENCES 1. Leppänen JM (2006) Emotional information processing in mood disorders: a review of behavioral and neuroimaging findings. Curr Opin Psychiatry 19:34–39. 2. Beck AT, Clark DA (1988) Anxiety and depression: an information processing perspective. Anxiety Stress Copin 1:23–36. 3. Gotlib IH, Joormann J (2010) Cognition and depression: current status and future directions. Annu Rev Clin Psychol 6:285–312. 4. Hankin BL, et al. (1998) Development of depression from preadolescence to young adulthood: emerging gender differences in a 10-year longitudinal study. J Abnorm Psychol 107:128–140. 5. Lee FS, et al. (2014) Adolescent mental health—opportunity and obligation. Science 346:547–549. 6. Gore FM, et al. (2011) Global burden of disease in young people aged 10–24 years: a systematic analysis. Lancet 377:2093–2102. 7. Kovacs M, Gatsonis C, Paulauskas SL, Richards C (1989) Depressive disorders in childhood. IV. A longitudinal study of comorbidity with and risk for anxiety disorders. Arch Gen Psychiatry 46:776–782. 8. Stiles J, Jernigan TL (2010) The basics of brain development. Neuropsychol Rev 20:327–348. 9. Kang HJ, et al. (2011) Spatio-temporal transcriptome of the human brain. Nature 478:483–489. 10. Casey BJ, Jones RM, Hare TA (2008) The adolescent brain. Ann N Y Acad Sci 1124:111–126. 11. Paus T, Keshavan M, Giedd JN (2008) Why do many psychiatric disorders emerge during adolescence? Nat Rev Neurosci 9:947–957. 12. Casey BJ, Heller AS, Gee DG, Cohen AO (2019) Development of the emotional brain. Neurosci Lett 693:29–34.

30

13. Gee DG, et al. (2013) A developmental shift from positive to negative connectivity in human amygdala-prefrontal circuitry. J Neurosci 33:4584–4593. 14. Somerville LH, Fani N, McClure-Tone EB (2011) Behavioral and neural representation of emotional facial expressions across the lifespan. Dev Neuropsychol 36:408–428. 15. Hare TA, et al. (2008) Biological substrates of emotional reactivity and regulation in adolescence during an emotional go-nogo task. Biol Psychiatry 63:927–934. 16. Drysdale AT, et al. (2017) Resting-state connectivity biomarkers define neurophysiological subtypes of depression. Nat Med 23:28–38. 17. Maglanoc LA, et al. (2019) Data-driven clustering reveals a link between symptoms and functional brain connectivity in depression. Biol Psychiatry Cogn Neurosci Neuroimaging 4:16–26. 18. Reinen JM, et al. (2018) The human cortex possesses a reconfigurable dynamic network architecture that is disrupted in psychosis. Nat Commun 9:1157. 19. Xia CH, et al. (2018) Linked dimensions of psychopathology and connectivity in functional brain networks. Nat Commun 9:3003. 20. Finn ES, et al. (2015) Functional connectome fingerprinting: identifying individuals based on patterns of brain connectivity. Nat Neurosci 18:1664–1671. 21. Kong R, et al. (2018) Spatial topography of individual-specific cortical networks predicts human cognition, personality, and emotion. Cereb Cortex. https://doi.org/10.1093/cercor/bhy123 22. Rosenberg MD, Casey BJ, Holmes AJ (2018) Prediction complements explanation in understanding the developing brain. Nat Commun 9:589. 23. Kaufmann T, et al. (2017) Delayed stabilization and individualization in connectome development are related to psychiatric disorders. Nat Neurosci 20:513–515. 24. Cole MW, Bassett DS, Power JD, Braver TS, Petersen SE (2014) Intrinsic and taskevoked network architectures of the human brain. Neuron 83:238–251.

31

25. Breakspear M (2017) Dynamic models of large-scale brain activity. Nat Neurosci 20:340–352. 26. Calhoun VD, Miller R, Pearlson G, Adalı T (2014) The chronnectome: time-varying connectivity networks as the next frontier in fMRI data discovery. Neuron 84:262–274. 27. Bullmore E, Sporns O (2009) Complex brain networks: graph theoretical analysis of structural and functional systems. Nat Rev Neurosci 10:186–198. 28. Vanderwal T, Eilbott J, Castellanos FX (2018) Movies in the magnet: naturalistic paradigms in developmental functional neuroimaging. Dev Cogn Neurosci https://doi.org/10.1016/j.dcn.2018.10.004. 29. Hasson U, et al. (2008) Neurocinematics: the neuroscience of film. Projections 2:1–26. 30. Vanderwal T, Kelly C, Eilbott J, Mayes LC, Castellanos FX (2015) Inscapes: a movie paradigm to improve compliance in functional magnetic resonance imaging. Neuroimage 122:222–232. 31. Gross JJ, Levenson RW (1995) Emotion elicitation using films. Cogn Emot 9:87–108. 32. Hasson U, Nir Y, Levy I, Fuhrmann G, Malach R (2004) Intersubject synchronization of cortical activity during natural vision. Science 303:1634–1640. 33. Nanni M, Martínez-Soto J, Gonzalez-Santos L, Barrios FA (2018) Neural correlates of the natural observation of an emotionally loaded video. PLoS One 13:e0198731. 34. Nummenmaa L, et al. (2012) Emotions promote social interaction by synchronizing brain activity across individuals. Proc Natl Acad Sci U S A 109:9599–9604. 35. Nummenmaa L, et al. (2014) Emotional speech synchronizes brains across listeners and engages large-scale dynamic brain networks. Neuroimage 102:498–509. 36. Guo CC, Nguyen VT, Hyett MP, Parker GB, Breakspear MJ (2015) Out-of-sync: disrupted neural activity in emotional circuitry during film viewing in melancholic depression. Sci Rep 5:11605.

32

37. Hyett MP, et al. (2015) Scene unseen: disrupted neuronal adaptation in melancholia during emotional film viewing. Neuroimage Clin 9:660–667. 38. Alexander LM, et al. (2017) An open resource for transdiagnostic research in pediatric mental health and learning disorders. Sci Data 4:170181. 39. Angold A, Costello EJ, Messer SC, Pickles A (1995) Development of a short questionnaire for use in epidemiological studies of depression in children and adolescents. Int J Methods Psychiatr Res 5:237–249. 40. Daviss WB, et al. (2006) Criterion validity of the Mood and Feelings Questionnaire for depressive episodes in clinic and non-clinic subjects. J Child Psychol Psychiatry 47:927–934. 41. Kaufman J, et al. (1997) Schedule for Affective Disorders and Schizophrenia for School-Age Children-Present and Lifetime Version (K-SADS-PL): initial reliability and validity data. J Am Acad Child Adolesc Psychiatry 36:980–988. 42. Peirce JW (2007) PsychoPy—psychophysics software in Python. J Neurosci Methods 162:8–13. 43. Walther D, Koch C (2006) Modeling attention to salient proto-objects. Neural Netw 19:1395–1407. 44. Cox RW (1996) AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res 29:162–173. 45. Nastase SA, Gazzola V, Hasson U, Keysers C (2019) Measuring shared responses across subjects using intersubject correlation. Soc Cogn Affect Neurosci 14:667–685. 46. Kauppi J-P, Jääskeläinen IP, Sams M, Tohka J (2010) Inter-subject correlation of brain hemodynamic responses during watching a movie: localization in space and frequency. Front Neuroinform 4:5. 47. Shen X, Tokoglu F, Papademetris X, Constable RT (2013) Groupwise whole-brain parcellation from resting-state fMRI data for network node identification. NeuroImage 82:403–415.

33

48. Chekroud AM, et al. (2017) Reevaluating the efficacy and predictability of antidepressant treatments. JAMA Psychiatry 74:370–378. 49. Kriegeskorte N, Mur M, Bandettini P (2008) Representational Similarity Analysis – connecting the branches of systems neuroscience. Front Syst Neurosci 2:4. 50. Geerligs L, Cam-CAN, Campbell KL (2018) Age-related differences in information processing during movie watching. Neurobiol Aging 72:106–120. 51. Trost W, Frühholz S, Cochrane T, Cojan Y, Vuilleumier P (2015) Temporal dynamics of musical emotions examined through intersubject synchrony of brain activity. Soc Cogn Affect Neurosci 10:1705–1721. 52. Dayan E, et al. (2018) Motion cues modulate responses to emotion in movies. Sci Rep 8:10881. 53. Box GEP, Jenkins GM (1970) Time series analysis: forecasting and control (HoldenDay). 54. Raila H, Scholl BJ, Gruber J (2015) Seeing the world through rose-colored glasses: people who are happy and satisfied with life preferentially attend to positive stimuli. Emotion 15:449–462. 55. Marcus DS, et al. (2011) Informatics and data mining tools and strategies for the human connectome project. Front Neuroinform 5:4.52. 56. Petroni A, et al. (2018) The variability of neural responses to naturalistic videos change with age and sex. eNeuro 5:ENEURO.0244-17.2017. 57. Cohen AO, et al. (2016) When is an adolescent an adult? Assessing cognitive control in emotional and nonemotional contexts. Psychol Sci 27:549–562. 58. Klein AM, de Voogd L, Wiers RW, Salemink E (2018) Biases in attention and interpretation in adolescents with varying levels of anxiety and depression. Cogn Emot 32:1478–1486.

34

59. Platt B, Waters AM, Schulte-Koerne G, Engelmann L, Salemink E (2017) A review of cognitive biases in youth depression: attention, interpretation and memory. Cogn Emot 31:462–483. 60. Ginicola MM (2007) Children’s unique experience of depression: using a developmental approach to predict variation in symptomatology. Child Adolesc Psychiatry Ment Health 1:9. 61. Castaneda AE, Tuulio-Henriksson A, Marttunen M, Suvisaari J, Lönnqvist J (2008) A review on cognitive impairments in depressive and anxiety disorders with a focus on young adults. J Affect Disord 106:1–27. 62. Hutchison RM, Morton JB (2015) Tracking the brain’s functional coupling dynamics over development. J Neurosci 35:6849–6859. 63. Kerestes R, Davey CG, Stephanou K, Whittle S, Harrison BJ (2014) Functional brain imaging studies of youth depression: a systematic review. Neuroimage Clin 4:209– 231. 64. Lindquist KA, Wager TD, Kober H, Bliss-Moreau E, Barrett LF (2012) The brain basis of emotion: a meta-analytic review. Behav Brain Sci 35:121–143. 65. Yeshurun Y, et al. (2017) Same story, different story: the neural representation of interpretive frameworks. Psychol Sci 28:307–319. 66. Nguyen M, Vanderwal T, Hasson U (2019) Shared understanding of narratives is correlated with shared neural responses. Neuroimage 184:161–170. 67. Everaert J, Bronstein MV, Cannon TD, Joormann J (2018) Looking through tinted glasses: depression and social anxiety are related to both interpretation biases and inflexible negative interpretations. Clin Psychol Sci 6:517–528. 68. Everaert J, Podina IR, Koster EHW (2017) A comprehensive meta-analysis of interpretation biases in depression. Clin Psychol Rev 58:33–48. 69. Insel T, et al. (2010) Research domain criteria (RDoC): toward a new classification framework for research on mental disorders. Am J Psychiatry 167:748–751.

35

70. Fried EI (2017) Moving forward: how depression heterogeneity hinders progress in treatment and research. Expert Rev Neurother 17:423–425. 71. Williams LM (2016) Precision psychiatry: a neural circuit taxonomy for depression and anxiety. Lancet Psychiatry 3:472–480. 72. Williams LM (2017) Defining biotypes for depression and anxiety based on large-scale circuit dysfunction: a theoretical review of the evidence and future directions for clinical translation. Depress Anxiety 34:9–24. 73. Gee DG, et al. (2016) Individual differences in frontolimbic circuitry and anxiety emerge with adolescent changes in endocannabinoid signaling across species. Proc Natl Acad Sci U S A 113:4500–4505. 74. Finn ES, Corlett PR, Chen G, Bandettini PA, Constable RT (2018) Trait paranoia shapes inter-subject synchrony in brain activity during an ambiguous social narrative. Nat Commun 9:2043. 75. Salmi J, et al. (2013) The brains of high functioning autistic individuals do not synchronize with those of others. Neuroimage Clin 3:489–497. 76. Holmes AJ, Patrick LM (2018) The myth of optimality in clinical neuroscience. Trends Cogn Sci 22:241–257. 77. Casey BJ, et al. (2018) The Adolescent Brain Cognitive Development (ABCD) study: imaging acquisition across 21 sites. Dev Cogn Neurosci 32:43–54. 78. Fitzgerald PB, Laird AR, Maller J, Daskalakis ZJ (2008) A meta-analytic study of changes in brain activation in depression. Hum Brain Mapp 29:683–695.

36

SUPPLEMENTARY MATERIAL: Relationships between depressive symptoms and brain responses during emotional movie viewing emerge in adolescence Authors: David C. Gruskina, Monica D. Rosenberga,b, Avram J. Holmesa,c* Affiliations: aDepartment of Psychology, Yale University, New Haven, Connecticut 06520, USA b

Department of Psychology, University of Chicago, Chicago, IL 60637, USA

c

Department of Psychiatry, Yale University, New Haven, Connecticut 06520, USA

*Corresponding author. Correspondence: Avram J. Holmes, Yale University, Department of Psychology, 402 Sheffield Sterling Strathcona Hall, 1 Prospect Street, New Haven, CT 06511, Phone: 203436-9240, Email: [email protected].

37

Figure S1. Group and individual ratings of the emotional content of the Despicable Me clip. (A) The mean ± 1 standard deviation of the normalized emotional valence ratings of the stimulus produced by 20 independent raters. (B) Normalized emotional valence ratings for each rater, averaged across three trials per rater (each color corresponds to a separate rater). (C and D) The same plots shown in A and B for emotional intensity (calculated as the absolute value of the emotional valence ratings).

38

Figure S2. Age-group differences (children – adolescents) in the ISC/depressive symptom relationships. (A) Subtracting the ISC/depressive symptom correlation coefficients (typicality and similarity) obtained in the adolescent group from those of the child group reveals parcel-level age-group differences in the relationships visualized in Figures 3 and 5 (bordered parcels significant at P<0.01, uncorrected). (B) The histograms reflect permutation tests showing the whole-brain age-group differences in the relationships between functional typicality/symptom severity and functional similarity/symptom profile similarity (whole-brain averages: ρchildren−ρadolescents=0.15, P=0.08 (functional typicality/symptom severity); ρchildren−ρadolescents =0.06, P=0.08 (functional similarity/symptom profile similarity).

39

Figure S3. Fluctuations in emotional movie content influence both inter-subject correlations and the functional typicality/depressive symptom severity relationship. (A) In children, inter-subject correlations were, on average, greatest during more emotionally-charged moments of the movie when the (negative) typicality-depression score relationship was weakest (as determined by fGLS regression; bordered regions significant at P<0.05, FDR adjusted). Emotional valence was largely negatively related to increased BOLD synchronization. (B) The same analyses in (A) for the adolescent group show consistent effects. In adolescents, relationships between emotional valence and ISC were mostly negative, while emotional intensity and the strength of the functional typicality/symptom were positively associated in two cortical parcels.

40

Figure S4. Age-related differences (children – adolescents) in how emotional movie content influences inter-subject synchronization and the functional typicality/depressive symptom severity relationship. These maps indicate the difference in fGLS-derived beta coefficients between the two age groups for each of the analyses shown in Figure S2 (P<0.05, FDR adjusted).

41

Table S1. Participant demographics and head motion Children Adolescents Mean±SD n Mean±SD n Age MFQ-SR total score Mean framewise displacement (mm) Max framewise displacement (mm) Sex (M:F) Clinician: no diagnosis Clinician: depression diagnosis Past depression diagnosis Current psych medication

10.1±1.4 14.7±9.4 0.11±0.03

62 62 62

15.8±1.9 13.4±11.4 0.09±0.03

50 50 50

t-test / chi-square Test P statistic N/A N/A t110 = 0.65 0.52 t110 = 1.21 0.23

1.69±0.64

62

1.53±0.63

50

t110 = 1.35

0.18

—

25:37

—

22:28

0.70

—

10

—

6

—

6

—

12

—

1

—

5

—

10

—

12

χ21,112 = 0.15 χ21,112 = 0.39 χ21,112 = 4.2 χ21,112= 3.84 χ21,112= 1.09

0.53 <0.05

Table S1. Participant demographics and head motion. Independent samples t-tests show that mean total scores on the MFQ-SR in the child and adolescent groups did not significantly differ, and mean/max framewise head motion during the movie-viewing fMRI scan was similar across the two age groups. Clinical characteristics of the two age groups were compared using chi-squared tests, which revealed that a significantly higher proportion of participants in the adolescent group received a depression diagnosis from the lead clinician as part of their participation in the study. Similarly, five participants in the adolescent group entered the study with a past depression diagnosis, compared to only one participant in the child group. Both groups were similar in their proportions of participants currently taking psychiatric medication and participants who did not receive a clinical diagnosis.

42

0.05 0.30

Tables S2-7: Full sample results (P<.01) Table S2. Functional typicality/depressive symptom severity volume mm3 MNI x MNI y MNI z ρ P

parcel #

region

193 149 154 230 254 61

L Temporal L Prefrontal L Prefrontal L Limbic L Cerebellum R Temporal

parcel # 230 182

parcel #

14

6128 6700 6106 2470 3363 8377

-60 -39 -43 -32

-27 17 42 -40 -53 -3

-21 59

-18 47 11 -4 -24 3

-0.283 -0.271 -0.253 -0.242 -0.237 -0.241

.002 .003 .006 .008 .009 .009

P (FDR adjusted) .380 .380 .380 .380 .380 .380

Table S3. Functional similarity/depressive symptom profile similarity region volume mm3 MNI x MNI y MNI z ρ P P (FDR adjusted) -40 .532 L Limbic 2470 -32 -4 -0.120 .008 .532 L Parietal 7464 -42 -66 42 -0.084 .008

region

Table S4. Inter-subject synchronization/emotional intensity volume mm3 MNI x MNI y MNI z β P

R Prefrontal 5968 34 R Insula 239 L Cerebellum 56 R Temporal 185 L Temporal 264 L Subcortical 27 R Motor strip 9 R Prefrontal 20 R Prefrontal 24 R Motor strip 28 R Motor strip 266 L Brainstem 15 R Prefrontal 119 R Cerebellum 263 L Subcortical 196 L Temporal 87 R Limbic 230 L Limbic 268 L Brainstem 52 R Temporal 216 L Occipital 260 L Subcortical 205 L Occipital 243 L Cerebellum

41 3555 2492 3812 2681 3084 4733 8091 3744 4343 3970 2664 5070 3861 4523 2411 2176 2470 2439 4094 3709 2379 4912 2544

14 42 -9 54 -38 -12 49 29 37 6 6 -5 7 30 -5 -52 28 -32 -6 40 -22 -15 -17 -19

43

48 5 -55 -8 6 -26 -4 51 21 -22 14 -38 21 -36 -10 -18 -54 -40 -19 19 -67 -4 -51 -46

0.144 -8 -52 -32 -38 15 48 19 6 66 49 -53 31 -31 6 -29 7 -4 -37 -34 7 21 1 -53

<.001 0.140 -0.129 0.115 0.116 0.105 -0.121 0.107 0.122 0.107 0.104 0.079 0.106 -0.082 0.063 0.099 0.083 0.071 0.074 0.090 0.080 0.085 0.067 0.079

P (FDR adjusted)

<.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 .001 .002 .002 .002 .002 .002 .003

<.001 <.001 .002 .002 .002 .004 .004 .008 .008 .011 .011 .011 .013 .015 .015 .015 .021 .021 .021 .023 .026 .027 .031

258 75 261 161 36 58 186 30 255 237 171 187 257

parcel #

region

236 34 59 24 256 123 122 15 47 11 146 22 185 220 20 201 243 94 10 115 55 19 169

L Subcortical R Occipital L Subcortical L Motor strip R Insula R Temporal L Temporal R Motor strip L Cerebellum L Cerebellum L Parietal L Temporal L Subcortical

3263 7048 3935 4319 4090 3389 3885 4790 2438 2705 6869 2833 1446

-13 19 -25 -6 37 40 -35 25 -7 -9 -51 -49 -11

12 -82 6 -4 21 -11 19 12 -55 -51 -24 11 24

9 42 0 48 -10 -36 -32 49 -26 -40 41 -31 10

0.045 0.074 0.081 0.072 0.093 0.078 0.080 -0.071 0.080 0.083 0.066 0.081 0.062

Table S5. Inter-subject synchronization/emotional valence volume mm3 MNI x MNI y MNI z β P

L Cerebellum R Insula R Temporal R Motor strip L Cerebellum R Subcortical R Subcortical R Prefrontal R Parietal R Prefrontal L Prefrontal R Prefrontal L Temporal L Limbic R Prefrontal L Temporal L Cerebellum R Limbic R Prefrontal R Cerebellum R Temporal R Prefrontal L Insula

3147 3555 3171 4343 1749 3729 2597 5070 9524 6389 7833 4650 2681 1376 3744 4180 2544 2275 7549 3194 5545 6344 3888

-7 42 43 6 -24 13 14 7 54 38 -27 40 -38 -4 37 -47 -19 36 8 8 61 48 -39

44

-66 5 -26 -22 -38 20 -4 21 -45 35 34 18 6 -5 21 -40 -46 -15 53 -57 -23 36 8

-38 -8 -25 66 -44 -1 21 31 37 31 36 29 -38 33 6 -24 -53 -18 24 -51 -22 15 -5

-0.254 -0.198 -0.194 -0.181 -0.174 0.165 0.162 -0.152 -0.133 -0.150 -0.123 -0.148 -0.136 -0.151 -0.150 -0.130 -0.111 -0.133 0.122 0.089 -0.059 -0.065 -0.087

.003 .003 .003 .003 .003 .004 .004 .004 .004 .005 .006 .006 .007

.031 .031 .031 .031 .031 .031 .033 .036 .036 .041 .042 .042 .049

P (FDR adjusted)

<.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 .001 .002 .003

<.001 <.001 <.001 <.001 <.001 <.001 <.001 .001 .001 .001 .001 .001 .003 .004 .004 .008 .010 .013 .013 .013 .017 .022 .030

143 70 48 218 248 265 58

parcel #

237

parcel #

183

L Prefrontal R Temporal R Parietal L Limbic L Cerebellum L Brainstem R Temporal

4507 7893 7536 4170 4563 3632 3389

-43 61 48 -8 -8 -5 40

47 -43 -62 -22 -68 -22 -11

-7 -18 35 46 -20 -16 -36

0.103 -0.087 -0.078 -0.103 0.094 0.098 -0.097

.003 .003 .003 .003 .004 .004 .005

.032 .032 .032 .032 .038 .039 .048

Table S6. Functional typicality & symptom severity relationship/emotional intensity region volume mm3 MNI x MNI y MNI z β P P (FDR adjusted)

L Cerebellum 2705 249 L Cerebellum 86 R Limbic 92 R Limbic 129 R Brainstem 190 L Temporal 169 L Insula 127 R Subcortical 214 L Occipital 141 L Prefrontal 24 R Motor strip 205 L Occipital 268 L Brainstem 153 L Prefrontal 36 R Insula

-9 3934 4903 3590 2325 4258 3888 3154 5211 5640 4343 4912 2439 3883 4090

-51 -35 12 31 5 -58 -39 12 -22 -12 6 -17 -6 -32 37

-40 -50 -57 4 -37 -6 8 -28 -97 65 -22 -51 -19 20 21

-0.181 -54 18 -22 -53 -23 -5 14 -10 4 66 1 -37 -16 -10

<.001 0.140 0.129 -0.109 -0.120 0.162 -0.132 0.119 -0.117 0.117 0.127 0.117 -0.117 0.100 0.112

<.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 .001 .001 .001 .002 .002 .003

.002 .007 .007 .007 .007 .008 .010 .025 .033 .033 .033 .048 .048 .048

Table S7. Functional typicality & symptom severity relationship/emotional valence region volume mm3 MNI x MNI y MNI z β P P (FDR adjusted)

L Parietal 7604 209 L Occipital 147 L Prefrontal 80 R Occipital 237 L Cerebellum 179 L Parietal 199 L Temporal 76 R Occipital 166 L Motor strip 150 L Prefrontal 244 L Cerebellum

-51 6172 7059 5914 2705 6156 8544 6226 3471 4360 5232

-56 -48 -46 8 -9 -36 -60 18 -28 -5 -7

45

20 -67 28 -89 -51 -39 -50 -83 -9 18 -50

0.190 1 27 12 -40 48 -14 -11 56 46 -11

<.001 -0.218 -0.182 0.159 0.181 -0.178 -0.161 0.149 -0.174 -0.142 -0.148

<.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 .001

<.001 <.001 .001 .001 .002 .005 .008 .016 .016 .026

219

L Limbic

5821

-6

34

26

-0.134

.001

.027

Tables S8-13: Child group results (P<.01)

parcel #

region

Table S8. Functional typicality/depressive symptom severity volume mm3 MNI x MNI y MNI z ρ P

P (FDR adjusted)

(no results significant at P<.01)

parcel #

Table S9. Functional similarity/depressive symptom profile similarity region volume mm3 MNI x MNI y MNI z ρ P P (FDR adjusted) (no results significant at P<.01)

parcel #

34 20 140 239 185 56 259 243 266 119 267 108 263 187 84 9 169 8 75 64 10

region

Table S10. Inter-subject synchronization/emotional intensity volume mm3 MNI x MNI y MNI z β P

R Insula 3555 R Prefrontal 3744 L Prefrontal 5575 L Cerebellum 2492 L Temporal 2681 R Temporal 3812 L Subcortical 6997 L Cerebellum 2544 L Brainstem 2664 R Cerebellum 3861 L Brainstem 2798 R Cerebellum 3263 L Subcortical 4523 L Temporal 2833 R Limbic 3038 R Prefrontal 8091 L Insula 3888 R Prefrontal 5142 R Occipital 7048 R Temporal 6013 R Prefrontal 7549 205 L Occipital 36 R Insula 115 R Cerebellum 153 L Prefrontal

42 37 -6 -9 -38 54 -10 -19 -5 30 -7 16 -5 -49 5 29 -39 45 19 56 8 4912 4090 3194 3883

5 21 48 -55 6 -8 11 -46 -38 -36 -33 -47 -10 11 -1 51 8 46 -82 -9 53 -17 37 8 -32

46

-8 6 12 -52 -38 -32 -8 -53 -53 -31 -39 -52 6 -31 36 19 -5 -5 42 -14 24 -51 21 -57 20

0.158 0.137 -0.131 -0.115 0.103 0.111 0.099 0.103 0.093 -0.099 0.103 -0.102 0.077 0.096 0.100 0.092 0.077 0.074 0.072 -0.089 -0.087 1 -10 -51 -16

<.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 .001 .001 .001 .002 .002 .002 0.076 0.091 -0.069 -0.083

P (FDR adjusted)

<.001 .002 .002 .002 .005 .005 .006 .006 .006 .006 .013 .013 .015 .016 .016 .017 .020 .022 .030 .031 .031 .003 .003 .003 .004

.031 .031 .032 .038

parcel #

region

236 34 256 15 146 268 115 10 58 140 220 153 59 94 11 123 131 51 122 60 22 24 174 55 235 145 8 265 67 237 143 117 194 201 73 144 53 19

Table S11. Inter-subject synchronization/emotional valence volume mm3 MNI x MNI y MNI z β P

L Cerebellum R Insula L Cerebellum R Prefrontal L Prefrontal L Brainstem R Cerebellum R Prefrontal R Temporal L Prefrontal L Limbic L Prefrontal R Temporal R Limbic R Prefrontal R Subcortical R Brainstem R Temporal R Subcortical R Temporal R Prefrontal R Motor strip L Parietal R Temporal L Limbic L Prefrontal R Prefrontal L Brainstem R Temporal L Cerebellum L Prefrontal R Cerebellum L Temporal L Temporal R Occipital L Prefrontal R Temporal R Prefrontal

3147 3555 1749 5070 7833 2439 3194 7549 3389 5575 1376 3883 3171 2275 6389 3729 1608 3019 2597 3128 4650 4343 6596 5545 3281 5600 5142 3632 4550 2705 4507 2783 3027 4180 5017 6283 6898 6344

-7 42 -24 7 -27 -6 8 8 40 -6 -4 -32 43 36 38 13 6 27 14 31 40 6 -7 61 -21 -10 45 -5 36 -9 -43 7 -49 -47 30 -29 53 48

47

-66 5 -38 21 34 -19 -57 53 -11 48 -5 20 -26 -15 35 20 -22 12 -4 1 18 -22 -34 -23 -4 56 46 -22 -69 -51 47 -54 -5 -40 -83 50 11 36

-38 -8 -44 31 36 -37 -51 24 -36 12 33 -16 -25 -18 31 -1 -42 -39 21 -44 29 66 67 -22 -29 30 -5 -16 -17 -40 -7 -34 -37 -24 21 22 -22 15

-0.268 -0.190 -0.202 -0.193 -0.164 0.176 0.106 0.171 -0.157 0.167 -0.138 0.146 -0.125 -0.128 -0.144 0.114 0.118 -0.133 0.128 -0.107 -0.128 -0.088 -0.125 -0.077 0.070 0.115 -0.094 0.086 -0.083 -0.109 0.109 0.133 0.094 -0.109 -0.103 -0.080 0.106 -0.067

P (FDR adjusted)

<.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 .001 .002 .002 .002 .002 .002 .002 .003 .004 .004 .004 .004

<.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 .001 .002 .002 .003 .003 .003 .003 .003 .004 .004 .007 .007 .009 .010 .010 .015 .015 .015 .015 .016 .019 .020 .023 .028 .028 .028 .029

162 128 5 47 32 31 243 121 2

parcel #

49 135 88 239 182

7259 4408 6436 9524 3273 3460 2544 3796 3367

-9 5 8 54 32 40 -19 13 10

0 -10 46 -45 -5 3 -46 13 18

67 5 -2 37 52 34 -53 11 -19

-0.086 0.082 -0.102 -0.073 0.090 -0.089 -0.091 0.056 -0.091

.004 .004 .005 .006 .007 .007 .007 .007 .008

.029 .029 .032 .037 .041 .042 .043 .043 .045

Table S12. Functional typicality & symptom severity relationship/emotional intensity region volume mm3 MNI x MNI y MNI z β P P (FDR adjusted)

112 46 207 161 86 98 135 42 79 205 227

parcel #

L Motor strip R Subcortical R Prefrontal R Parietal R Motor strip R Motor strip L Cerebellum R Subcortical R Prefrontal

R Cerebellum R Parietal L Occipital L Motor strip R Limbic R Limbic L Prefrontal R Parietal R Occipital L Occipital L Limbic

4877 7635 6059 4319 4903 5112 2855 4344 3576 4912 1691

20 58 -26 -6 12 15 -18 15 7 -17 -7

-74 -29 -63 -4 -57 -46 19 -68 -76 -51 -42

-50 20 -12 48 18 3 -21 35 -3 1 13

0.166 -0.151 0.175 0.156 0.167 0.145 0.139 0.138 0.129 0.120 -0.122

<.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 .001 .001 .002

.001 .004 .009 .010 .010 .015 .020 .026 .031 .031 .043

Table S13. Functional typicality & symptom severity relationship/emotional valence region volume mm3 MNI x MNI y MNI z β P P (FDR adjusted)

R Parietal 6082 12 R Prefrontal L Prefrontal 2855 117 R Cerebellum R Limbic 2103 L Cerebellum 2492 L Parietal 7464

41 8570 -18 2783 7 -9 -42

-75 14 19 7 -19 -55 -66

28 37 -21 -54 30 -52 42

-0.260 49 -0.199 -34 -0.160 0.158 -0.174

<.001 0.200 <.001 0.186 <.001 <.001 <.001

<.001 <.001

.001 .004

<.001

.007 .010 .015 .018

Tables S14-19: Adolescent group results (P<.01) Table S14. Functional typicality/depressive symptom severity volume mm3 MNI x MNI y MNI z ρ P

parcel #

region

159 125 104

L Motor strip R Subcortical R Cerebellum

5095 5085 1543

-58 14 24

-6 8 -36

48

27 -9 -43

-0.455 -0.387 -0.380

.002 .007 .008

P (FDR adjusted)

0.268 0.268 0.268

157 97 153 95 229

parcel #

1 104 90 182 95

parcel # 268

97 131 14 110 146 138 21 261 172 15 27 124 218 3 19 247 258 203 205

parcel # 47

14 97 100

L Prefrontal R Limbic L Prefrontal R Limbic L Limbic

6830 3004 3883 4512 2338

-46 25 -32 28 -21

8 -3 20 -28 -37

29 -31 -16 -14 6

-0.373 -0.380 -0.363 -0.380 -0.369

.008 .008 .008 .008 .009

0.268 0.268 0.268 0.268 0.268

Table S15. Functional similarity/depressive symptom profile similarity region volume mm3 MNI x MNI y MNI z ρ P P (FDR adjusted)

R Prefrontal R Cerebellum R Limbic L Parietal R Limbic

region

14 24 6 -42 28

57 -36 -57 -66 -28

-17 -43 38 42 -14

-0.108 -0.154 -0.177 -0.130 -0.180

.002 .003 .005 .005 .007

Table S16. Inter-subject synchronization/emotional intensity volume mm3 MNI x MNI y MNI z β P

L Brainstem R Limbic R Brainstem R Prefrontal R Cerebellum L Prefrontal L Prefrontal R Prefrontal L Subcortical L Parietal R Prefrontal R Motor strip R Subcortical L Limbic R Prefrontal R Prefrontal L Cerebellum L Subcortical L Occipital L Occipital

region

4965 1543 5764 7464 4512

2439 3004 1608 5968 4146 7833 4749 7170 3935 3404 5070 4733 4105 4170 4248 6344 5762 3263 7160 4912

-6 25 6 41 21 -27 -7 55 -25 -24 7 49 27 -8 5 48 -10 -13 -41 -17

-19 -3 -22 14 -55 34 48 10 6 -32 21 -4 6 -22 35 36 -82 12 -75 -51

-37 -31 -42 48 -24 36 -6 22 0 64 31 48 0 46 -17 15 -32 9 23 1

0.211 0.205 0.201 0.156 -0.159 0.130 0.112 0.110 0.096 0.078 0.102 -0.076 0.094 0.095 0.071 0.082 0.087 0.054 -0.067 0.065

<.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 .001 .002 .002 .002 .002 .003 .003 .003

Table S17. Inter-subject synchronization/emotional valence volume mm3 MNI x MNI y MNI z β P

R Parietal R Prefrontal R Limbic R Cerebellum

9524 5968 3004 5424

54 41 25 32

-45 14 -3 -78

49

37 48 -31 -40

-0.267 -0.228 -0.229 -0.235

<.001 <.001 <.001 <.001

0.240 0.240 0.240 0.240 0.240

P (FDR adjusted)

<.001 <.001 <.001 <.001 <.001 <.001 <.001 0.001 0.008 0.008 0.009 0.010 0.023 0.030 0.030 0.030 0.038 0.041 0.044 0.044

P (FDR adjusted)

<.001 <.001 <.001 <.001

146 30 19 89 124 22 131 164 169 241 117 74 50 91 221 123 161 139 2 114 34 190 53 44 25 226 5 126 60 145 51 238 122 1 138 264 265 227 178 41 4 63 246

L Prefrontal R Motor strip R Prefrontal R Limbic R Subcortical R Prefrontal R Brainstem L Motor strip L Insula L Cerebellum R Cerebellum R Occipital R Parietal R Limbic L Limbic R Subcortical L Motor strip L Prefrontal R Prefrontal R Cerebellum R Insula L Temporal R Temporal R Parietal R Motor strip L Limbic R Prefrontal R Subcortical R Temporal L Prefrontal R Temporal L Cerebellum R Subcortical R Prefrontal L Prefrontal L Subcortical L Brainstem L Limbic L Parietal R Parietal R Prefrontal R Temporal L Cerebellum

7833 4790 6344 3531 4105 4650 1608 7552 3888 8228 2783 6589 6030 4394 3729 3729 4319 5427 3367 4651 3555 4258 6898 8121 3690 4799 6436 3726 3128 5600 3019 4468 2597 4965 4749 3084 3632 1691 8536 6539 2385 6510 6772

-27 25 48 8 27 40 6 -23 -39 -40 7 45 49 8 -5 13 -6 -18 10 23 42 -58 53 7 7 -9 8 10 31 -10 27 -37 14 14 -7 -12 -5 -7 -10 25 16 62 -43

34 12 36 -23 6 18 -22 11 8 -74 -54 -74 -58 -40 13 20 -4 57 18 -72 5 -6 11 -57 -8 -43 46 -27 1 56 12 -53 -4 57 48 -26 -22 -42 -66 -52 34 -24 -64

50

36 49 15 45 0 29 -42 54 -5 -29 -34 3 14 48 29 -1 48 -14 -19 -29 -8 -23 -22 62 53 50 -2 -2 -44 30 -39 -31 21 -17 -6 15 -16 13 55 68 -23 -3 -46

-0.202 -0.215 -0.189 -0.192 -0.180 -0.193 -0.186 0.162 -0.128 -0.158 -0.169 0.130 0.123 -0.160 -0.124 0.103 -0.127 0.094 -0.127 -0.094 -0.122 -0.126 -0.126 -0.124 -0.101 -0.111 -0.087 0.094 0.094 -0.099 0.118 -0.097 0.125 -0.118 -0.125 -0.095 0.111 -0.069 -0.100 0.081 -0.077 -0.085 0.079

<.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 .001 .001 .001 .001 .002 .002 .002 .003 .004 .004 .006 .006 .006 .006

<.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 0.001 0.001 0.001 0.003 0.003 0.003 0.003 0.003 0.005 0.005 0.005 0.005 0.005 0.006 0.006 0.006 0.006 0.006 0.008 0.008 0.010 0.011 0.014 0.015 0.016 0.017 0.026 0.028 0.034 0.035 0.036 0.036

248 59 258 16 189 192 191

L Cerebellum R Temporal L Subcortical R Prefrontal L Temporal L Temporal L Temporal

4563 3171 3263 7575 2641 7466 7044

-8 43 -13 54 -23 -58 -59

-68 -26 12 25 9 -47 -30

-20 -25 9 1 -39 5 3

0.080 -0.108 -0.077 -0.106 -0.071 -0.088 -0.073

.007 .007 .007 .007 .008 .008 .010

0.036 0.038 0.038 0.039 0.039 0.040 0.049

Table S18. Functional typicality & symptom severity relationship/emotional intensity region volume mm3 MNI x MNI y MNI z β P P (FDR adjusted)

parcel # 128 R Subcortical

226 194

L Limbic L Temporal

4408 4799 3027

5 -9 -49

-10 -43 -5

5 50 -37

-0.158 0.157 0.180

<.001 <.001 <.001

<.001 0.005 0.014

Table S19. Functional typicality & symptom severity relationship/emotional valence parcel # region volume mm3 MNI x MNI y MNI z β P P (FDR adjusted) 101 R Cerebellum 5245 6 -51 -12 -0.332 <.001 <.001

227 92 157 147 179 79 71 240 137 169 209 158

L Limbic R Limbic L Prefrontal L Prefrontal L Parietal R Occipital R Temporal L Cerebellum L Prefrontal L Insula L Occipital L Motor strip

1691 3590 6830 7059 6156 3576 4787 3949 4548 3888 6172 4931

-7 31 -46 -46 -36 7 42 -21 -8 -39 -48 -42

-42 4 8 28 -39 -76 -46 -70 40 8 -67 -15

51

13 -22 29 27 48 -3 -23 -49 -21 -5 1 45

-0.244 0.208 -0.194 -0.190 -0.193 0.153 0.174 0.200 -0.165 0.180 -0.144 0.167

<.001 <.001 <.001 <.001 <.001 <.001 <.001 <.001 .001 .001 .002 .002

<.001 0.001 0.002 0.006 0.008 0.010 0.011 0.019 0.028 0.028 0.046 0.046

Title: Relationships between depressive symptoms and brain responses during emotional movie viewing emerge in adolescence Highlights: • Adolescents with worse depressive symptoms show less typical movie-driven brain activity • Relationships between depressive symptoms and brain responses are stronger during less emotional moments of the movie • Pairs of adolescents with more similar depressive symptom profiles share more similar brain responses to the movie • Brain response typicality and pairwise similarity are unrelated to depressive symptoms in children