Participants
A total of 1,154 healthy right-handed young adults (666 males, mean age 20.79 years, SD = 1.89 years and 488 females, mean age 20.60 years, SD = 1.61 years) participated in this study as part of our ongoing project to explore the associations among brain imaging characteristics, cognitive functions, aging, genetics, and daily habits. Indeed, from our database, we used the data from 1,154 subjects that had questionnaire data about internet dependence, fMRI imaging data, and behavioral data of the N-back task without apparent artifacts. Therefore, these 1,154 subjects are those who were included in the study after these exclusions were considered.
All subjects were undergraduate or postgraduate students from Tohoku University in Japan. All had normal vision and none had a history of neurological or psychiatric illness. None reported recent use of any psychoactive drugs or other drugs that could negatively impact cognitive abilities. History of psychiatric illnesses and recent drug use were assessed by our laboratory’s routine questionnaire. We used the Edinburgh Handedness Inventory to evaluate handedness in subjects 72. The Ethics Committee of Tohoku University approved all procedures, which were performed in accordance with relevant guidelines and regulations. Written informed consent was obtained from each subject for the projects in which they participated. Descriptions in this subsection are adapted from a previous study using similar methods 73.
Internet Addiction Tendency assessment
We used the Japanese version of Young’s IAT scale to assess condition severity 74. This IAT instrument consists of 20 items answered on a 1–5 scale from 1 = “rarely” to 5 = “always”. The scale is self-administered and requires 5 to 10 minutes. The IAT scale minimum and maximum scores are 20 and 100, with higher scores reflecting a greater tendency toward internet addiction. The Japanese version of this scale has demonstrated high reliability and validity 75.
fMRI task
Functional MRI was used to map brain activity during working memory. The n-back task is a widely used task consisting of 0-back (simple cognitive processing) and 2-back (working memory) conditions. In the 2-back task, subjects viewed a series of stimuli presented sequentially (one of four Japanese vowels) and were instructed to judge if a target stimulus appearing “2” presentations earlier was the same as the current stimulus by pushing a button. In the 0-back task, subjects were instructed to determine whether a presented letter was the same as the target stimulus by pushing a button (Figure 1). We used a simple block design. Descriptions in this subsection were mostly adapted from our previous studies using similar methods 73, 76.
In this study, our focus was TID in the DMN. TID in the DMN occurs in mostly similar areas regardless of whether the task is 2-back or 0-back, although there are differing magnitudes. Furthermore, differences in brain activity between patients with schizophrenia and control subjects were similar regardless of whether the task was a 0-back task or 2-back. These included areas of DMN (i.e., subtracting the activity during the 0-back task from the brain activity during the 2-back task substantially eliminates group differences) 77, 78. Therefore, we did not analyze the contrast of 2-back – 0-back, as was done in another study that focused on TID in the DMN 78.
Image acquisition
The MRI acquisition methods were described in our previous study 79. Briefly, all MRI data were acquired using a 3T Philips Achieva scanner. Diffusion-weighted data were acquired using a spin-echo EPI sequence (TR = 10293 ms, TE = 55 ms, FOV = 22.4 cm, 2×2×2 mm3 voxels, 60 slices, SENSE reduction factor = 2, number of acquisitions = 1). The diffusion weighting was isotropically distributed along 32 directions (b value = 1,000 s/mm2). In addition, three images with no diffusion weighting (b value = 0 s/mm2 or b = 0 images) were acquired using a spin-echo EPI sequence (TR = 10293 ms, TE = 55 ms, FOV = 22.4 cm, 2 ´ 2 ´ 2 mm3 voxels, 60 slices). For the n-back session, 174 functional volumes were obtained 73. For more details, see our previous study 53.
Preprocessing of structural data
Preprocessing of the structural and functional data was performed using Statistical Parametric Mapping software (SPM12; Wellcome Department of Cognitive Neurology, London, UK) implemented in MATLAB (Mathworks, Inc., Natick, MA). For analyses, T1-weighted structural images of each individual were segmented using the new segmentation algorithm implemented in SPM12 and normalized to Montreal Neurological Institute (MNI) space to yield images with 1.5 × 1.5 × 1.5 mm3 voxels using the diffeomorphic anatomical registration through exponentiated lie algebra registration process implemented in SPM12. In addition, we performed a volume change correction (modulation) 80. Subsequently, generated rGMV and rWMV images were smoothed by convolution using an isotropic Gaussian kernel of 8 mm full width at half maximum. These descriptions were mostly adapted from our previous study using similar methods. For full descriptions of these procedures, see our previous work 73.
Pre-processing and data analysis for functional activation data
Pre-processing and data analysis of functional activation data were performed using SPM. The following procedures for functional activation data analysis were reproduced from our previous study, as described previously 81. Prior to analysis, BOLD images were re-aligned and resliced to the mean BOLD image of the run. They were then corrected for slice timing, co-registered, and resliced to the voxel space of images of diffusion tensor imaging. All images were subsequently normalized using a previously validated 82 two-step segmentation algorithm of the diffusion images and the previously validated diffeomorphic anatomical registration through exponentiated lie algebra-based registration process. The voxel size of normalized BOLD images was 3 ´ 3 ´ 3 mm3 and taken to the second-level analyses of functional activities. The full description of these procedures are provided in our previous study 76.
A design matrix was fitted to each participant with one regressor for each task condition (0-, 2-back in the n-back task) using the standard hemodynamic response function. The design matrix weighted each raw image according to its overall variability, to reduce the impact of movement artifacts 83. The design matrix was fit to the data for each participant individually. After estimation, beta images were smoothed (8 mm full-width half-maximum) and taken to the second level or subjected to a random effect analysis. We removed low-frequency fluctuations using a high-pass filter and a cutoff value of 128 s. The individual-level statistical analyses were performed using a general linear model.
In the individual analyses, we focused on activation related to the condition (0-back or 2-back versus rest). The resulting maps representing brain activity during the working memory condition (2-back) and simple cognitive processing condition (0-back) for each participant were then forwarded for group analysis.
The fMRI images with artifacts based on the visual inspection had been removed from the images. Thorough instruction to prevent motion during the scan was given to educated participants. Other exclusions based on motion parameters were not performed in this study.
In a previous study, we validated normalization procedures of fMRI using diffusion tensor images using SPM 8 84. Our internal preliminary survey also showed these procedures work better using SPM8. Conversely, VBM procedures work better with SPM12. In other words, the segmentation of the diffusion images obtained, which were part of our preprocessing procedures of fMRI, were not adequate for SPM12. Misclassifications that were apparent by visual inspection were systematically found when SPM 12 was used. In the second-level analysis, the use of SPM8 or SPM12 does not affect the results of TFCE based on permutation.
Generally, thorough instructions and thorough fixation by the pad were provided to prevent head motion during the scan as much as possible, and we utilized the software to reduce the impact of movements 83, as described in the subsection below.
Thus, we did not exclude any subject from the fMRI analyses based on excessive motion that did not cause evident artifacts during the scan. The subjects were young adults and the scan duration was very brief. Only the maximum movement of several subjects detracted from the original point, and in one of the directions exceeded 3 mm. Removing these subjects from analyses did not substantially alter the significant results of the present study.
Similarly, the subjects enrolled in the study were educated young adults, and thorough instruction and sufficient practice was provided. Subjects whose responses were properly recorded showed acceptable accuracies and only seven subjects showed accuracies lower than 80% in the 0-back or 2-back task (but accuracies were at least 50% or greater). Removing these subjects also did not substantially alter the significant results of the present study.
Effects of interaction between sex and the score of Young’s IAT scale on imaging measures
We also performed a supplementary investigation of the potential regions displaying significant effects of interaction between the subject’s sex and score on the Young’s IAT scale (that is, we investigated whether some regions showed sex-related differences in the correlations patterns based on the Young’s IAT scale score). For this purpose, we performed whole-brain analyses of covariance (ANCOVAs). The dependent variables in these analyses were same as those in the whole-brain multiple regression analyses that were conducted to investigate the correlation with score of Young’s IAT scale in each voxel across sexes. In these whole-brain ANCOVAs, sex was a group factor (using the full factorial option in SPM8), whereas and all other covariates are same as those of the abovementioned whole-brain multiple regression analyses. In addition, all covariates were modeled to enable unique relationships with imaging measures (dependent variables) (using the interactions option in SPM8) for each sex. The interaction between sex and the score of Young’s IAT scale (contrasts of [the score of the Young IAT scale for males, the effect of the score of the Young IAT scale for females] were [−1 1] or [1 –1]) were assessed using t-contrasts. Correction for multiple comparisons was performed using the same method used in the whole-brain multiple regression analyses.
Supplemental methods
Supplemental analyses of the comparison between subjects using Young’s IAT scale.
In accordance with the considerable literature available that classified subjects based on the Young’s IAT scale score, we also divided subjects into two groups (IAT score ≥50 and IAT score <50). This classification was used to compare dependent variables between those who used the internet excessively and those who used it less frequently. We hypothesized that excessive use of the internet would be associated with additional changes in brain structure and functional characteristics. For this reason, we also conducted the supplemental analyses of comparisons between subjects who scored ≥50 using Young’s IAT scale and those who scored <50 using Young’s IAT scale, on the basis of the criteria described previously 74.
For this comparison, we conducted multiple regression analyses in which all dependent and independent variables of the main analyses remained the same, except that Young’s IAT score was replaced by the dichotomized value (Young’s IAT scale ≥50 = 1, Young’s IAT scale < 0).
Supplemental region of interest (ROI) analyses of the associations between activity in key nodes of the DMN and IAT scores.
We conducted a supplemental partial correlation analyses of the associations between mean beta estimates of functional ROIs of important nodes of the DMN and IAT after controlling for covariates. In these analyses, ROI masks were defined by the areas that are mostly significantly deactivated during the 2-back task using an appropriate threshold that successfully segregated each area in the representative DMN nodes for the 63 subjects from which the template of normalization was created (when there were multiple clusters in one area, those that showed the strongest statistical values at the peak were selected). The mean beta estimates of the 2-back task as well as the 0-back task within each ROI were extracted. For these analyses, control variables were same as those of the covariates in the whole-brain multiple regression analyses in the main text.
ROIs were mPFC (peak coordinate: x = −6, y = 57, z = −6, T score threshold = 15, 425 voxels), PCC/precuneus (peak coordinate: x = −6, y = −57, z = 12, T score threshold = 15, 305 voxels), left hippocampus (peak coordinate: x = −27, y = −21, z = −24, T score threshold = 9, 27 voxels), right hippocampus (peak coordinate: x = 24, y = −15, z = −27, T score threshold = 9, 66 voxels), left temporoparietal junction (peak coordinate: x = −45, y = −72, z = 21, T score threshold = 7, 322 voxels), and right temporoparietal junction (peak coordinate: x = 54, y = −69, z = 27, T score threshold = 7, 32 voxels).
Results with a threshold having p < 0.05, and corrected for the false discovery rate (FDR) using the two-stage sharpened method 85, were considered statistically significant.
Statistical analysis
Statistical analyses of imaging data were performed with SPM8. Structural whole-brain multiple regression analyses were performed to investigate associations of IAT scores with rGMV and rWMV. Age, sex, and total intracranial volume calculated using voxel-based morphometry (for details of calculation see 86) were added as covariates.
For the functional images, we used multiple regression analysis to investigate the relationship between IAT score and brain activity levels during the 0-back, 2-back, and 2-back-0-back tasks. Age, sex, n-back task accuracy, and n-back task reaction time were entered into the multiple regression model as covariates.
A multiple comparison correction was performed using threshold-free cluster enhancement (TFCE) 87 with randomized (5,000 permutations) nonparametric testing using the TFCE toolbox (http://dbm.neuro.uni-jena.de/tfce/). We applied a threshold of family-wise error corrected at P < .05. SPM8 was used for analyses because of better compatibility with TFCE software and our in-house scripts 54.