Compulsivity is linked to reduced adolescent development of goal-directed control and frontostriatal functional connectivity

Significance Goal-directed behavior is impaired in disorders of compulsivity. Here, we characterize the developmental trajectory of model-based control and show a progressive strengthening from adolescence to early adulthood. We found that the presence of compulsivity traits impacts on this trajectory as well as on the degree of remodeling in functional connectivity within frontostriatal circuits. These findings have implications for understanding the interplay between compulsivity, the developmental trajectory of model-based planning, and functional connectivity in frontostriatal circuits.

were recruited in an accelerated longitudinal study from schools, colleges, National Health capitalized on this reasoning by using logistic regression to identify whether selection of the fractals in the first stage from one trial to the next was informed only by the outcome of the number of trials was increased to 201 to match previous studies (2). Participants were 100 instructed to win as much reward (play pounds) as possible, and were told they would receive 101 a payment bonus based on task performance. Before completing the two-step reinforcement 102 learning task, participants were always administered instructions and a comprehension test 103 associated with it.

105
Exclusion criteria for the reinforcement learning task. Subjects were excluded if they 106 missed more than 10% of trials (N = 3) either at T1 or T2, if they responded on the same key 107 on more than 95% of trials on which they registered a response (N = 2) or had implausibly fast 108 reaction times, i.e., below 150 ms on more than 20% of the trials (N = 13

111
However, results presented in this paper are not substantially changed by the inclusion or 112 exclusion of these subjects in the analyses. For the two-stage task data, the first trial in each 113 block as well as trials with implausibly fast response times (below 150ms) were omitted from the analysis (less than 1% of the overall trials). 53 of these participants also completed the 115 same reinforcement task at an intermediate time point (T1R), approximately 6 months after 116 the first lab visit at T1. Additionally, following (4), we identified those subjects (T1, N = 20; T2, 117 N = 13) who repeated previously rewarded second-stage responses at a rate lower than 50%.

118
Inclusion or exclusion of these subjects did not impact the main results.

120
Assessment of compulsivity. We investigated LOI concurrent validity by testing its 121 relationship with a separate instrument, the Padua Inventory Washington State University the battery but it was available for a significant lower number of participants (only N=287 124 subjects for whom model-based scores were available at T1 and T2, also had PI-WSUR 125 available for both assessments). At both time points (Figure S1 B, C), there was a high 126 correlation between compulsivity measured by these two questionnaires (T1, N = 277,

127
Pearson's r = 0.58; T2, N = 519, Pearson's r = 0.70). We also tested the concurrent validity of 128 the LOI by using a composite score of compulsivity. We performed principal component analysis on all available items of the available questionnaires that probed individual (6)). This score was constructed independently of the LOI yet was highly correlated with it 132 (mean values across assessments, Pearson's r = 0.76), concurrently validating the LOI as an previously described for this sample (7) and it is summarised here. Following processing of 138 individual structural scans using FreeSurfer v5.3.0 (including skull-stripping, segmentation of 139 cortical grey and white matter and reconstruction of the cortical surface and grey-white matter 140 boundary) (8), all scans were stringently quality controlled by re-running the reconstruction 141 algorithm after the addition of control points and white matter edits (as described previously 142 (9, 10)). The pre-processing of functional data with ME-ICA analysis was performed using 143 AFNI (11). Volumes acquired during steady-state equilibration (15 s) were omitted. The data 144 of the middle TE were used to compute parameters of motion correction and anatomical-145 functional co-registration.The first volume after equilibration was used as the base EPI image. functional (12), using the EPI base image as the LPC weight mask. Matrices for deobliquing, 149 motion correction, and anatomical-functional co-registration were combined into a single 150 alignment matrix using the concatenation approach from the AFNI tool align_epi_anat.py. The

155
We used ME-ICA (13, 14) for pre-processing of functional scans to identify the sources of 156 variance in the fMRI time series that scaled linearly with TE and could therefore be confidently were not BOLD-related and therefore did not scale with TE, were identified by ME-ICA and 159 discarded. The retained independent components, representing BOLD contrast, were 160 optimally recomposed to generate a broadband denoised fMRI time series at each voxel. This  During pre-processing, realignment of scans was used to estimate 6 motion parameters for 164 each participant (3 translation parameters and 3 rotation parameters). Subsequently, these 165 were used to calculate an overall estimate of motion -the framewise displacement (FD; defined as the sum of the absolute derivatives of the six motion parameters, following the surface of a sphere with radius 50 mm as in (15, 16)). Mean FD was used as a measure of head movement in each scan session. Finally, as previously described (7), confounding and FD (7). Residual estimates of overall striatal connectivity (mean-FD corrected) were not 174 correlated with head motion; and there was no significant relationship between FD and any of 175 the measures of interest (i.e., age, model-based control, and compulsivity) (Supplementary Figure S5). Thus we used this movement correction pipeline of ME-ICA followed by FD

199
Firstly, we conducted a basic logistic regression separately for T1 and T2 to test if participants'

228
Thus, here we split the age covariate into its between and within-subject components and     Figure 3 assessing cross-domain coupling between model-based control and compulsivity. T1, baseline; T2, follow-up; d, estimated latent change score 0.039 This model assesses cross-domain coupling between model-based control and compulsivity and includes age, gender and IQ which were regressed both on the observed variables and on the latent change variables of both model-based and compulsivity. The model provided good fit to the data (N = 520; c2 = 0.288, df = 1, P = 0.591; RMSEA = 0.000 [0.000, 0.094], SRMR = 0.004, CFI = 1.000, Yuan-Bentler scaling correction factor = 1.001). Results from this model were not numerically nor inferentially different from those reported in Figure S3 and Table S7. T1, baseline; T2, follow-up; d, estimated latent change score. Model corresponding to Figure 4 assessing cross-domain coupling between model-based control, compulsivity and overall striatal connectivity strength. Site was regressed on functional connectivity measures. T1, baseline; T2, follow-up; d, estimated latent change score