Frontoparietal functional connectivity moderates the link between time spent on social media and subsequent negative affect in daily life

Evidence on the harms and benefits of social media use is mixed, in part because the effects of social media on well-being depend on a variety of individual difference moderators. Here, we explored potential neural moderators of the link between time spent on social media and subsequent negative affect. We specifically focused on the strength of correlation among brain regions within the frontoparietal system, previously associated with the top-down cognitive control of attention and emotion. Participants (N = 54) underwent a resting state functional magnetic resonance imaging scan. Participants then completed 28 days of ecological momentary assessment and answered questions about social media use and negative affect, twice a day. Participants who spent more than their typical amount of time on social media since the previous time point reported feeling more negative at the present moment. This within-person temporal association between social media use and negative affect was mainly driven by individuals with lower resting state functional connectivity within the frontoparietal system. By contrast, time spent on social media did not predict subsequent affect for individuals with higher frontoparietal functional connectivity. Our results highlight the moderating role of individual functional neural connectivity in the relationship between social media and affect.

head motion and susceptibility distortions.Next, the BOLD time series were resampled again into standard space, generating a preprocessed BOLD run in MNI152NLin2009cAsym space.We performed all resamplings with a single interpolation step by composing pertinent transformations (i.e., head motion transform matrices, susceptibility distortion correction when available, and co-registrations to anatomical and output spaces).Gridded (volumetric) resamplings were performed using antsApplyTransforms (ANTs), configured with Lanczos interpolation to minimize the smoothing effects of other kernels 13 .Non-gridded (surface) resamplings were performed using mri_vol2surf (FreeSurfer).

SI2. Head motion correction
We estimated head motion parameters with respect to the BOLD reference (transformation matrices, and six corresponding rotation and translation parameters) before any spatiotemporal filtering using mcflirt (FSL 5.0.9) 6 .We slice-time corrected BOLD runs using 3dTshift from AFNI 20160207 12 .We resampled the BOLD time series onto their original native space by applying a single, composite transform to correct for head motion and susceptibility distortions.Next, the BOLD time series were resampled again into standard space, generating a preprocessed BOLD run in MNI152NLin2009cAsym space.We performed all resamplings with a single interpolation step by composing pertinent transformations (i.e., head motion transform matrices, susceptibility distortion correction when available, and co-registrations to anatomical and output spaces).Gridded (volumetric) resamplings were performed using antsApplyTransforms (ANTs), configured with Lanczos interpolation to minimize the smoothing effects of other kernels 13 .Nongridded (surface) resamplings were performed using mri_vol2surf (FreeSurfer).We calculated various confounds (e.g., framewise displacement [FD], DVARS, global signal) for each TR and logged in a confounds file (for additional details, see https://fmriprep.org/en/20.0.6/outputs.html#confounds).Across the 300 total volumes of the participants included in the current study, the average FD was 0.13mm (SD= 0.05), the average standardized DVARS was 1.24 (SD=0.09),and on average, 0.2% (SD=0.25) of the scans showed spikes across 300 volumes (Fig. SI2a).We manually quality-checked the fMRIPrep outputs, by looking for gross distortions, to ensure adequate preprocessing.None were excluded based on manual checks.
We further denoised resting-state data using the XCP Engine pipeline (Version 1.0) 14 .Specifically, we used XCP Engine to remove motion-related confounds from BOLD sequences: (1) de-meaning and detrending (removal of linear and quadratic trends from time series), (2) despiking using AFNI's 3dDespike utility, (3) temporal bandpass filtering using a first-order Butterworth filter to retain signal in the range 0.01-0.08Hz, (4) 36-parameter confound regression including 6 realignment parameters, mean signal in white-matter, cerebrospinal fluid, and mean global signal, as well as the first power and quadratic expansions of their temporal derivatives.Instead of removing volumes identified as motion outliers, we denoised BOLD timeseries voxel-wise by truncating large spikes.None of the participants were entirely excluded for excessive head motion.Participants' resting state frontoparietal functional connectivity levels were not associated with head motion descriptors including FD (r=-0.153,p=0.266),DVAR (r=-0.094,p=0.501), or percentage of scans that showed spikes (r=0.072,p=0.603).Please see Fig. SI2b for temporal signal to noise ratio (tSNR) maps which were computed by taking the mean signal over time divided by the standard deviation over time.

Fig.SI2a Distributions of motion metrics
Fig. SI2b Temporal Signal to Noise Ratio (tSNR) maps tSNR maps were constructed using the (A) raw resting-state data and (B) denoised resting-state data.For each map, we averaged across participants and projected the distribution of tSNR across the cortex onto an inflated fsaverage brain surface.Results indicate overall improvement in signal quality, particularly in the occipital and parietal cortices, as well as areas prone to signal dropout such as the temporal lobes and the orbitofrontal cortices.

SI3. MNI coordinates and corresponding regions of frontoparietal nodes.
Nodes making up the frontoparietal system were based on the original regions of interest from Power et al., 2011 15 .Regions were determined using the Talairach Daemon 16 .L = left hemisphere, R = right hemisphere.

SI4. Subregions within the frontoparietal system
Subsystems of the frontoparietal system are implicated in different types of executive control.Specifically, frontoparietal network A (FPA) connects to the default mode network and supports introspective processes, whereas FPB connects to the dorsal attention network and helps regulate perceptual attention 17 .To further understand which subsystems are more responsible for moderating the link between social media use and affect, we extracted average resting state functional connectivity within these subsystems separately using Schaefer 17 systems atlas with 400 parcels 18,19 .
We ran two separate multilevel models that included the amount of time spent on social media, FPA / FPB subsystems within the frontoparietal system (in separate models), and their interaction term as predictors of subsequent negative affect.We found that the main interaction we observed in the main text (frontoparietal functional connectivity x time spent on social media predicting negative affect) was unique to overall average frontoparietal connectivity, but not to the specific subsystems: We did not find any significant interaction between time spent on social media and FPA (b=-1.028,95% CI [-11.96,9.90], t=-0.184,p=0.854) or FAB (b=-6.335,95% CI [-18.46,5.79], t=-1.025,p=0.310).These results suggest that the overall executive control, rather than its subdomains, might be more important for regulating complex social stimuli such as the ones presented in social media.

SI5. Functional connectivity within visual and auditory systems as control measures
As one test of specificity, we repeated all main analyses using functional connectivity within the visual and auditory systems 15 .We chose these regions because, compared to the frontoparietal system, we had less theoretical reason to believe they are involved in cognitive control, emotion regulation, or negative affect.Among the additional regions examined, we found that the results reported in the main text were unique to the frontoparietal system.Specifically, individual differences in functional connectivity within the visual and auditory systems were not significantly associated with the levels of depression (visual: b=3.

SI7. Current positive affect
In addition to the negative EMA questions reported in the main text, participants also reported their current positive affect by responding to two positive affect items, twice a day (How positive / happy do you feel right now?) on a scale of 1 (not at all) to 100 (extremely) with higher scores indicating higher positive affect.We combined these two ratings to produce mean positive affect scores (Rc =0.830).Using these positive affect composite scores, we tested the relationships between time spent on social media, frontoparietal functional connectivity, and positive affect.We did not find any association between time spent on social media reported since the previous time point and current positive affect (b=-1.813,95% CI [-6.16, 2.54], t=-0.817,p=0.418).In the same model, we also did not find any significant interaction between time spent on social media and frontoparietal functional connectivity in predicting positive affect (b=8.586,95% CI [-11.83,29.01], t=0.824, p=.414).

SI8. Difficulties in Emotion Regulation (DERS) subscales
The short form 18-item Difficulties in Emotion Regulation Scale (DERS) 20,21 measures the degree of difficulty experienced when regulating emotions, rated on a 1 (almost never) to a 5 (almost always) scale.DERS includes six subscales that measure different types of difficulties in emotion regulation, including lack of emotional awareness, nonacceptance of emotions, impulse control difficulties, restricted access to emotion regulation strategies, reduced emotional clarity, and difficulties participating in goal-directed behavior.In the main text, we report that the greater emotion regulation dysfunction, indicated by higher DERS composite scores, was associated with weaker frontoparietal functional connectivity.Here, we examined this relationship across different subscales within DERS.Correlation analysis showed that, of the six DERS subscales, two subscales likely drove the inverse relationship between the frontoparietal functional connectivity (FC, in the figure below) and emotion dysregulation, including goaldirected behavior (r=-.39495% CI [-0,60, -0.14], p=0.003) and restricted access to emotion regulation strategies (r=-.422,95% CI [-0,62, -0.17], p=0.002).These results are consistent with previous evidence supporting the role of the frontoparietal system in engagement of executive control that support emotion regulation, such as goal-directed cognition and strategic planning 22- 25 .

Fig.SI8 Bivariate correlations between frontoparietal functional connectivity and Difficulties in Emotion Regulation (DERS) subscales
Notes: FC=Functional Connectivity within the frontoparietal system, mean=average DERS score

SI9. Temporal relationships between time spent on social media and affect
To further unpack the temporal link between social media use and subsequent affect, we also explored the possibility that the negative or positive affect at a previous time point would predict later social media use, and whether these links might be moderated by frontoparietal functional connectivity.Two multilevel models included the negative or positive affect scores at a previous time point (in two separate models), frontoparietal connectivity, and their interaction term as predictors of subsequent reported time spent on social media.The slope of the positive affect scores was allowed to vary randomly across participants; the random effect variance of the negative affect was close to zero and was removed from the model.We found no evidence linking previous negative (b=0.

SI11. Correction for multiple comparisons
All main results were robust to False Discovery Rate (FDR) correction (FDR-adjusted p values for the association between frontoparietal functional connectivity and depression (0.018), anxiety (0.036), and emotion dysregulation (0.036); for the link between social media and negative affect at lower (0.003), mean (0.026), and higher (0.892) levels of frontoparietal functional connectivity).

SI12. Coefficients and statistics of all
969, p=0.552; auditory: b=-8.516,p=0.242), anxiety (visual: b=7.845, p=0.577; auditory: b=-3.771,p=0.807), or difficulty regulating emotions (visual: b=0.089, p=0.914; auditory: b=-1.618,p=0.065).In addition, visual/auditory functional connectivity did not moderate the relationship between minutes spent on social media and negative affect (visual: b=3.921, p=0.227; auditory: b=-5.440,p=0.138).In the main text, we report results that converted these raw scores into minutes by taking the midpoint value of the answer range.All findings remained consistent when we used the raw scores: More minutes spent on social media since the previous time point predicted greater increases in negative mood (b=4.581, 95% CI[1.52, 7.64], t=2.932, p=0.005).In the same model, we also observed that this link was moderated by frontoparietal functional connectivity, such that greater functional connectivity was associated with a weaker relationship between social media use and negative affect (b=-18.953,95% CI [-33.32,-4.59], t=-2.586,p=0.013).

main analyses models.
Standardized (β)and unstandardized (b) regression coefficients, standard error for unstandardized regression coefficients (se), and 95% confidence intervals (CI) are displayed.FC=Functional Connectivity within the frontoparietal system, Depression=Epidemiologic Studies Depression Scale (CES-D), Anxiety=State-Trait Anxiety Inventory (STAI), Emotion dysregulation=Difficulties in Emotion Regulation Scale, social media use=Minutes spent on social media.Time-varying variables (social media use, negative affect) were within-person standardized (N=54; 2424 observations).All analyses controlled for potential covariates, including demographic variables (age, gender, and race/ethnicity), the condition assignment as part of a parent study, and perceived social status (in analyses linking FC and depression, anxiety, and emotion dysregulation).Please see https://github.com/cnlab/social_media_brain/for the complete model output statistics.