Two groups of heroin-dependent individuals were enrolled, one in MT and the other undergoing a medication-free TC program. The participants in MT were recruited from the Far Eastern Memorial Hospital (New Taipei City, Taiwan) and En Chu Kong Hospital (New Taipei City, Taiwan); both hospitals provided low-threshold MT programs. The participants in TC programs were recruited from Operation Dawn, a Christian-based organization providing an 18-month residential program for individuals with drug use problems. An advanced 3-year voluntary missionary training program was provided for those who completed the basic programs. The TC programs were abstinence-oriented, and the recovering drug users and professionals provided supportive care and spiritual guidance for drug users to reconstruct their orderly lives and reshape their values. Psychoactive substances, including coffee and tobacco, were prohibited. Psychopharmacological medications were also not allowed; therefore, individuals with disturbing mental symptoms requiring pharmacological treatment were excluded from TC programs.

The enrolled MT and TC participants had to meet the following criteria: 1) aged 20 or above, 2) a lifetime diagnosis of opioid dependence, on the basis of the SCID-I (Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision (DSM-IV-TR), Axis-I disorder), and 3) remaining in treatment (MT or TC) for at least 3 months. Given that the participants in TC programs were separated by gender, this study only enrolled male participants based on the accessibility. The study was approved by the institutional review board of the National Health Research Institutes (Miaoli, Taiwan). A total of 11 MT and 11 TC individuals were enrolled between August 2010 and July 2011. Written informed consent was obtained from all participants.

All participants were interviewed by research nurses. Sociodemographics (age, education, religion) and clinical information (onset of heroin use, current treatment duration, current daily dose of methadone for MT) were collected. The SCID-I was used to assess the current comorbid mood, psychotic, and anxiety disorders (18). The severity of drug use, indexed by the days of drug use in the past 4 weeks, was assessed using the Treatment Outcome Profiles (19). Handedness was assessed using the Briggs–Nebes modified Annett’s 12-item Hand Preference Questionnaire with scores ranging from −100 for extreme left hand preference to +100 for extreme right hand preference (20). Urine drug testing for morphine and amphetamine was performed before MR experiments.

The decision-making capability was assessed using a computerized CGT, a subset of the Cambridge Neuropsychological Test Automated Battery (Cambridge Cognition Ltd). The task was composed of eight blocks of nine trials. At the beginning of each block, a total of 100 cumulative points were offered to participants. On each trial, the participants were presented with an array of 10 blue and red boxes. The ratio of box colors varied pseudo-randomly over the course of the task. Participants were informed that a token had been hidden inside one of these boxes, and they had to decide whether the token had been hidden inside a red or blue box. After choosing the box color (blue or red), participants were required to bet on their decision. Five available bet options, which represented 5%, 25%, 50%, 75%, and 95% of the cumulative points, were offered in an ascending or descending sequence. In the ascend condition (half the blocks), the available bet was 5% upwards, and vice versa for the descend condition (half the blocks). In betting, participants had to decide and wait for their desirable choice (21). Several key features of CGT should be highlighted. First, the varied ratio of box colors provides the explicit outcome probability of gain and loss, which enables the assessment of the decision-making performance under risk. Second, the variable proportion of cumulative points of the bets reflects the willingness to risk on the already acquired wage for further unknown reward. Finally, the inclusion of both ascend and descend conditions allows for distinguishing the participants’ impulsive and risk-taking betting strategies. Risk adjustment (RA) is defined by the degree to which participants calibrate their bets in response to the ratio of box colors. RA was the primary measure for decision-making performance in this study considering heroin users were reported to have deficit in RA (11). A high RA score indicated a tendency to bet a larger proportion of available points on the high ratio rather than on the low ratio trials, whereas, a low RA score suggested the failure to use the available information in decision-making.

The experiments were performed on a 3T MRI scanner (MR750, GE Healthcare, Milwaukee, WI, USA) with an 8-channel head radio frequency (RF) receive array coil. For anatomical image acquisition, a T2-weighted fast spin-echo sequence was employed with repetition time of 5,806 ms, echo time of 102 ms, number of averages of 1, 20 slices, slice thickness of 5 mm, and in-plane resolution of 0.45 mm × 0.45 mm. For rs-fMRI, functional images were acquired using a gradient-echo echo-planar imaging sequence with repetion time/echo time (TR/TE) of 2,000/30 ms, 40 slices, slice thickness of 3 mm, field of view of 230 mm, matrix size of 64 × 64, in-plane resolution of 3.6 mm × 3.6 mm, and 320 repetitions. For the purpose of feasibility, the MR experiments were arranged with months of lag after neurocognitive functioning assessments.

The overall analysis workflow of this study is illustrated in Figure 1 . To identify DMN subregions from the rs-fMRI data, we applied the group independent component analysis (G-ICA) approach on the rs-fMRI data of all participants from both groups (22 individuals). For each participant, the rs-fMRI data were first preprocessed using the Data Processing Assistant for Resting-State fMRI (DPARSF) toolbox (22) and Statistical Parametric Mapping (SPM) software program (https://www.fil.ion.ucl.ac.uk/spm/). The preprocessing procedures included realignment, slice timing adjustment using windowed Fourier interpolation, coregistration between fMRI and T2-weighted images, spatial normalization of T2-weighted images to Montreal neurological institute (MNI) space [International consortium for brain mapping (ICBM) 152 template] using Diffeomorphic anatomical registration through exponentiated lie algebra (DARTEL) (23), and spatial smoothing using Gaussian kernel with full width at half maximum of 5 mm. The preprocessing procedures for DMN identification differed from those for deriving functional connectivity and rs-fMRI measures, which are introduced in a subsequent section. G-ICA was performed on the preprocessed rs-fMRI data using the group ICA of fMRI toolbox (http://mialab.mrn.org/software/gift/). For each participant, the independent components (ICs) were extracted using the infomax ICA algorithm (24). To increase the reliability, we employed the ICASSO (25) algorithm, which could output the most reliable set of ICs from 100 repeated ICA runs. To determine the optimum number of ICs, we performed the G-ICA using four IC numbers (i.e., 10-, 15-, 20-, and 30-component ICAs were performed). For each IC, we calculated the similarity between the IC and a DMN template obtained from a previous study (26). The similarity of each IC was defined as the correlation coefficient (r) between its Z-score map and the Z-score map of the DMN template. For G-ICA results derived from each IC numbers (i.e., 10, 15, 20, and 30), the IC with highest similarity to the DMN template was selected as the matched component to represent the DMN. We then compared the four ICA settings and discovered that the matched IC obtained from 15-component ICA had the highest similarity (r = 0.559) to the DMN template. Therefore, this matched IC was identified and used to extract the DMN subregions ( Figure 2A ). The DMN subregions extraction was mainly based on the local maxima of the Z-score map of the matched IC. The matched ICs were divided into spatially connected clusters, and clusters containing less than 20 voxels were regarded as spurious clusters and were removed to avoid potential confounds. For each of the remaining cluster, the local maximum was identified as the center of the region-of-interest (ROI). The ROI was then extracted as a spherical region with a radius of 6 mm centering at the local maxima. Through this approach, in total, four ROIs within the DMN located at the mPFC, left inferior parietal lobe (IPL_L), right inferior parietal lobe (IPL_R), and PCC were identified and used for further analysis ( Figure 2B ).

To derive both the interregional functional connectivity and regional rs-fMRI measure, a dissimilar preprocessing approach from the one we used in DMN identification was employed on the rs-fMRI data of each participant. The preprocessing procedures included realignment, slice timing adjustment using windowed Fourier interpolation, coregistration between fMRI and T2-weighted images, spatial normalization of T2-weighted images to MNI space (ICBM512 template) using DARTEL (23), 24-parameter head motion correction (27), and nuisance covariate regression (white matter signals and cerebrospinal fluid signals). The rs-fMRI time series of each voxel were further detrended and band-pass filtered (0.01–0.1 Hz). We evaluated the maximum head motion for each participant and confirmed that it was within 3 mm (voxel size) among all participants. In addition, we performed a statistical comparison and confirmed that the two participant groups did not differ significantly in the amount of head motion (p = 0.931). On completing rs-fMRI data preprocessing, the mean time series of each ROI were obtained by averaging the voxel-wise preprocessed rs-fMRI time series within the selected ROI. To calculate the interregional functional connectivity, we first computed a Pearson’s correlation coefficient (PC) for connectivity measures in the time courses. Because the sample size is relatively small, we then adapted a Fisher’s Z-transformation for the PCs to stabilize the variance of correlations. In this study, we decided to use ALFFs to investigate the regional rs-fMRI activity within each ROI. ALFFs were derived from the voxel-wise preprocessed rs-fMRI time series (28) and accounted for the ALFFs of the rs-fMRI signals, in which the resting-state activity is considered to be more prominent.

The statistical analysis was composed of three parts as follows: functional connectivity analysis in the DMN, between-group comparison for the quantitative rs-fMRI measure (ALFF), and multiple linear model analysis for the association of the decision-making index (RA) and rs-fMRI measure (ALFF). To study the functional connectivity, the Fisher’s Z-transformation for the PCs was used to measure the significance of the difference between the two groups. In the study of functional connectivity, we had six sets of comparisons from four ROIs in the null hypothesis. To address the multiple comparison issue, a Benjamini–Hochberg (BH) procedure (29) was used to control a false discovery rate (FDR) at level 0.05. Specifically, for the i-th participant in the j-th group, i = 1,…,11, j = 1,2, the sample PC ρ ^ j,k, k ′ ,i between the k-th and k’-th ROIs was computed from a set of pairs [(Y _j,k,t,i ,Y _j,k′,t,i ): t = 1,…,320], where Y _i,k,t,i and Y _i,k’,t,i represent the mean rs-fMRI signals at time t across the k-th and k’-th ROIs, respectively, for the i-th participant in the j-th group. The sample correlation ρ ^ j,k, k ′ ,i was then transformed to Z j,k, k ′ ,i =0.5ln( 1+ ρ ^ j,k, k ′ ,i 1− ρ ^ j,k, k ′ ,i ) by a Fisher’s transformation. When the number of temporal points T is large, Z ¯ j,k, k ′ ,i approximately follows a normal distribution with mean 0.5ln( 1+ρ 1−ρ ) and variance 1 T−3 (30). In this study, the variance of Z _j,k,k′,i is approximately 0.0032. Z ¯ j,k, k ′ = ∑ i=1 11 Z j,k, k ′ ,i /11 denotes an average of the transformed correlations between the k-th and k’-th ROIs in the j-th group. The transformed PCs have been standardized by corresponding variances in the Fisher’s Z-transformation. Therefore, with the Fisher’s transformation, the standard deviation for Z ¯ j,k, k ′ was 0.017 in all cases. D _k,k’ = Z _1,k,k – Z _2,k,k′ denotes a difference in averages for the transformed PCs between the two groups. Considering the variance was estimated from the data, a two-sided t-test with degrees of freedom equal to 20 was used to evaluate the test statistic ζ _k,k’ = D _k,k’ /SD(D _k,k’ ), where SD(D _k,k’ ) denotes a standard deviation of D _k,k’ . This approach was then applied to test the significance of the functional connectivity difference between the MT and TC groups.

For the regional rs-fMRI activity, a two-sample t-test was used to evaluate the between-group difference of ALFFs. The BH procedure was again applied to control the FDR at level 0.05 for four sets of comparisons (corresponding to four ROIs).

The associations of the decision-making measure (RA) with fMRI signals were examined. W _i denotes the values of RA for the i-th participant. Moreover, X _i,k denotes the averaged ALFFs over the time period and across the k-th ROI for the i-th participant. δ _i denotes an indicator variable for the group effect as δ_i = 1 if the i-th participant belongs to the MT group. We then fit a model by W _i = b ₀ + b ₁ X _i,k + b ₂δ _i , k = 1,…,4. A two-sided t-test for each estimated coefficient was used to examine whether the coefficient was significantly away from zero. Given that we have to compare the ALFF from four ROIs, the BH procedure was again applied to control the FDR at level 0.05 for four sets of comparisons.

Table 1 summarizes the clinical characteristics and CGT decision-making measures (RA) of MT and TC participants. The age of the participants in the MT group was 39.1 ± 5.9 years, which was significantly older than the age of the participants in the TC group (34.3 ± 4.1 years) (p < 0.01). At the rs-fMRI session, 54.5% (6/11) and 9.1% (1/11) of the participants in the MT group were tested positive for morphine and amphetamine, respectively. No use of these illicit drugs was detected in the participants in the TC group by urine testing. The CGT measure (i.e., RA) was not significantly different between the MT and TC groups.

As illustrated in Figure 2A , the subregions of DMN for each group were identified using the ICA approach. The most closely corresponding components exhibiting DMN characteristics were selected for the following connectivity and statistical analysis ( Figure 2B ).

The results of functional connectivity analysis are presented in Table 2 and Figure 3 . Herein, the subregions 1 to 4 represented the mPFC, IPL_L, IPL_R, and PCC subregions in DMN, respectively. For the MT and TC groups, the averages of the transformed PCs between two subregions, k and k’, were denoted by Z ¯ MT,k, k ′ and Z ¯ TC,k, k ′ , respectively. The values of Z ¯ MT,1,2 , Z ¯ MT,1,3 , Z ¯ MT,1,4 , Z ¯ MT,2,3 , Z ¯ MT,2,4 , and Z ¯ MT,3,4 were 0.63, 0.46, 0.81, 0.70, 0.89, and 0.53, respectively, and the values of Z ¯ TC,1,2 , Z ¯ TC,1,3 , Z ¯ TC,1,4 , Z ¯ TC,2,3 , Z ¯ TC,2,4 , and Z ¯ TC,3,4 were 0.51, 0.49, 0.80, 0.77, 0.84, and 0.60, respectively. The values of the difference between groups D_1,2, D_1,3, D_1,4, D_2,3, D_2,4, and D_3,4 were 0.12 (p < 0.001), −0.03 (p = 0.23), 0.01 (p = 0.68), −0.07 (p = 0.0085), 0.05 (p = 0.050), and −0.07 (p = 0.0085), respectively, with the standard deviation at a value of 0.024 for all cases.

Using the BH procedure to control the FDR at level 0.05 results in thresholds for p-values in a descending order as 0.008, 0.017, 0.025, 0.033, 0.042, and 0.05. Therefore, after conducting the BH procedure to screen out p-values, we concluded that D_1,2 (t-ratio = 5.0), D_2,3 (t-ratio = 2.91), and D_3,4 (t-ratio = 2.91) were significantly away from zero.

The two-sample t-test conducted with the BH procedure to control the FDR revealed there was no significant difference among all cases.

The estimation results for the association between ROI-based rs-fMRI measure (ALFF) and CGT measure (RA) are presented in Table 3 and Figure 4 . Our results revealed that, after conducting the BH procedure to control the FDR, RA exhibited a significant association with ALFFs in IPL_L (₁ = 0.51, t = 3.005, p = 0.007).