Flowchart of the Cross-scan Functional Connectivity Analysis
Fig. 1 displays the flowchart of the cross-scan functional network connectivity (FNC) analysis. We first applied a Neurmark framework to extract robust intrinsic connectivity networks (ICNs) that are comparable across subjects, scans, and sessions. FNC was estimated using the time-courses (TCs) of ICNs from each scan. After obtaining the FNC matrix of each scan, cross-scan FNC similarity was measured by the correlation between FNC from different scans. Then individual identification was performed based on the cross-scan FNC similarity. Finally, we examined the associations between intra-subject FNC similarity and individuals' behaviors via a linear mixed-effect model (LMM).
Functional Networks
53 ICNs were extracted by the Neuromark framework, with activation peaks falling on the cortical and subcortical gray matter areas across the whole brain. The ICNs were assigned to seven different functional domains according to their anatomical locations and functional information 14, including subcortical (SC), auditory (AUD), visual (VS), sensorimotor (SM), cognitive-control (CC), default-mode (DM), and cerebellar (CB) domains. Details of the spatial maps and coordinates of ICNs are provided in the supplementary materials.
Intra-subject FNC Shows High Similarity across Scans
Fig. 2a displays the FNC of subjects with maximum and minimum intra-subject FNC similarities between scans. Children show different levels of cross-scan FNC similarity than each other. For example subject 1, the FNC of scan 1 and the FNC of scan 2 share the highest intra-subject similarity (r = 0.9448). In contrast, for example subject 2, the FNC show less stability between scan 1 and scan 2, where the intra-subject FNC similarity is only r = 0.1914. Fig. 2b displays the percentage of children with intra-subject FNC similarity larger than a given percentage of inter-subject FNC similarity, from 60% to 99%. Our results show that intra-subject FNC similarity is larger than a majority of inter-subject FNC similarity, though intra-subject FNC variability exists. The FNC shows the highest intra-subject similarity between scan 1 and scan 2. More than 90% of subjects have intra-subject FNC similarity larger than 60% of inter-subject FNC similarity and more than 65% of subjects have intra-subject FNC similarity larger than 99% of inter-subject FNC similarity. The intra-subject FNC shows the lowest similarity between scan 1 and scan 4. Still, more than 80% of subjects have intra-subject FNC similarity larger than 60% of inter-subject FNC similarity, and about 40% of subjects have intra-subject FNC similarity larger than 99% of inter-subject FNC similarity.
The results are replicated by examining the scans from the second-year session. Similarly, subjects can have different levels of cross-scan FNC similarity. FNC shows the highest intra-subject similarity between scan 1 and scan 2 and the lowest intra-subject similarity between scan 1 and scan 4. FNC also shows intra-subject similarities between longitudinal scans. Although a two-year time interval between scans incurred a significant decrease in intra-subject similarity, the intra-subject similarity is still larger than the majority of inter-subject FNC similarity, especially when the FNC was averaged within the session.
Individual Identification using Whole-brain FNC
Fig. 3 shows the identification results of each pair of identification. At the baseline session, the identification accuracy was 93.99%, 84.78%, 81.87%, and 93.10% based on the database-target of scan 1-scan 2, the target-database of scan 1-scan 3, the target-database of scan 1-scan 4, and the target-database of scan 1-scan mean respectively. The identification was replicated by using the FNC of the second-year scans. Similar to the results from the baseline, the highest identification accuracy 95.16% was achieved based on the database-target of scan 1-scan 2, while the lowest identification accuracy 82.80% was achieved based on the database-target of scan 1-scan 4.
The individual identification was further performed using the FNC from longitudinal scans. Scans from the baseline session were the database and scans from the second-year session were the target. Although more intra-subject FNC variations were introduced, the FNC of a child from the baseline session can still be used to identify his/her FNC from the second-year follow-up session. The highest accuracy was 91.43%, which was achieved by averaging the FNC across all four scans within each session before identification.
The nonparametric permutation testing shows that the average identification accuracy was 50% if the identity was shuffled for each scan. The real identification accuracy was significantly higher than the accuracy obtained by the permutation tests.
FNC Stability Correlates with Cognitive Performance
We noted that besides the intrinsic patterns, FNC also shows significant variability across scans and the mechanisms underlying the cross-scan FNC stability are still unknown. In this study, we focused on children’s cognitive performance, mental problems, sleep conditions, and screen usage. These scores have been linked to a wide range of brain functions and structures in previous studies 33–36. The investigation of the associations between these scores and FNC stability might advance our understanding of the neural underpinnings of children’s behaviors. The cognitive measures were positively correlated with the intra-subject FNC stability (False discovery rate [FDR] corrected, q < 0.05). Specifically, 10 out of 10 of the cognitive summary scores were positively correlated with FNC stability, with correlation r values ranging from 0.0376 to 0.1070. The Total Composite Score was the score most significantly positively correlated with the FNC stability (r = 0.1070, Cohen’s d = 0.2152, p = 4.82×10-24). For the neurocognitive battery in the subdomain, TPVT was the score most significantly positively correlated with the FNC stability (r = 0.0841, Cohen’s d = 0.1688, p = 1.54×10-15) while TFT was the score least significantly positively correlated with intra-subject FNC stability (r = 0.0376, Cohen’s d = 0.0753, p = 3.68×10-4). To better visualize the associations, we divided the children into four groups from low cognitive performance to high cognitive performance according to each cognitive score (group 1: 0%~25%, group 2: 25%~50%; group 3: 50%~75%, and group 4: 75%~100%) and the averaged cross-scan FNC stability within each group is displayed using bar plots in Fig. 4a. Clear increasing trends can be observed along group 1 to group 4. Along with the scatter plots in Fig. 4a, our results indicate that children with good cognitive performance tended to have higher FNC stability.
FNC Stability Correlates with Psychiatric Problems
The psychopathological measures of children were negatively correlated with the intra-subject FNC stability. 12 out of 20 psychiatric problem scores show significantly negative correlations with FNC stability, with r values ranging from -0.0257 to -0.0496 (FDR corrected, q < 0.05). The social problem score was the score most significantly negatively correlated with the FNC stability (r = -0.0496, Cohen’s d = -0.0992, p = 2.38×10-6). Similarly, we divided the children into four groups according to each psychopathological measure. The mean and the standard error of the mean for the cross-scan FNC stability of each group were displayed in Fig. 4b. The FNC stability show decreasing trends along group 1 to group 4, indicating that children with high psychiatric problem scores tended to have lower FNC stability.
FNC Stability Correlates with Sleep Conditions and Screen Usage
We further found significant associations between FNC stability and the sleep conditions of children. The cross-scan FNC stability was negatively correlated with the sleep duration score (r = -0.0752, Cohen’s d = -0.1508, p = 7.74×10-13). In the ABCD measurement system, high sleep duration score indicates short sleep duration (1 = 9-11 hours; 2 = 8-9 hours; 3 = 7-8 hours; 4 = 5-7 hours; 5 = Less than 5 hours). The FNC stability was also negatively correlated with the score that evaluates how long an adolescent falls asleep (sleepdisturb2_p). A higher score in sleepdisturb2_p indicates a longer time to fall asleep. The FNC stability was negatively correlated with other sleep behaviors of adolescents, such as sleepdisturb24_p (evaluates a child feels unable to move when waking up in the morning) and sleepdisturb26_p (evaluates a child falls asleep suddenly in inappropriate situations). Higher scores in these measurements indicate more frequently that the event happens (1 = Never; 2 = Occasionally (once or twice per month or less); 3 = Sometimes (once or twice per week); 4 = Often (3 or 5 times per week); 5 = Always). The overall results indicate that children with worse sleep conditions (e.g., shorter sleep duration or longer time to fall asleep) tended to have lower FNC stability (Fig. 4d).
Children’s screen usage is also negatively correlated with cross-scan FNC stability. 14 out of 14 youth screen time utilization scores, including the use of television, internet, cell phone, and video games, show negative correlations with individuals’ FNC stability (FDR corrected, q<0.05). Children with more screen usage tended to have lower FNC stability. Details of the correlations statistics can be found in the supplementary materials (Fig. 4c).
FNC Stability Correlates with Parent Psychopathology and Prenatal Exposure
Moreover, parental dimensional psychopathology showed significant correlations with their children’s FNC stability (Fig. 5a and b). Specifically, the positive questions in the parents’ psychopathology assessment, including asr_q15_p (I am pretty honest), asr_q73_p (I meet my responsibilities to my family), asr_q88_p (I enjoy being with people), asr_q98_p (I like to help others), asr_q106_p (I try to be fair to others), asr_q123_p (I am a happy person), were positively correlated with the FNC stability of children with r values ranging from 0.0315 to 0.0583 (FDR corrected, q < 0.05). In contrast, the negative questions in the parents’ psychopathology assessment were negatively correlated with the FNC stability of children with r values ranging from -0.0287 to -0.0482 (FDR corrected, q < 0.05). These results indicate that parents with positive behaviors will result in higher FNC stability in children while parents with negative behaviors will result in lower FNC stability in children.
Our analysis also showed that prenatal exposure before and during pregnancy was associated with FNC stability in children. Parents with prenatal exposure to tobacco and marijuana will result in lower FNC stability in children (Fig. 5d). Also, a planned pregnancy will result in higher FNC stability in children. The age of the parents during the pregnancy showed significant correlations with FNC stability as well. While older mothers will result in higher FNC stability in children, fathers’ ages between 30~40 years old (when the child was born) result in the highest FNC stability in adolescents (Fig. 5c).