Introduction

Neurobiological and neuroimaging studies have demonstrated that the dorsal lateral prefrontal cortex (DLPFC), a major component of the executive control network, is crucial in modulating top-down control and regulating decision-making ^1,2. Transcranial magnetic stimulation (TMS) targeting the DLPFC has been shown to effectively manipulate inter-temporal behaviors in healthy individuals ³ and drug-using behaviors in cocaine users ⁴. However, the neural mechanism through which the DLPFC is involved in the alteration of these behaviors remains unclear. In this study, we utilized resting-state fMRI (rs-fMRI) to investigate the relationship between the DLPFC functional connectivity (FC) and relapse in cocaine addiction.

Methods

Forty-five cocaine use disorder (CUD) patients and 22 healthy control (HC) subjects participated in this study. The CUD participants underwent one of three residential cocaine-dependence treatment programs that utilized the Minnesota Model psychosocial treatment approach. Urine drug screens were conducted throughout the residential treatment to verify abstinence. The CUD participants were followed for 168 days post-treatment or until relapse to stimulant use. The study was approved by the Institutional Review Boards and written informed consent of each subject was obtained prior to the study ⁵. Resting-state fMRI data were obtained on a Philips 3T scanner using a GE-EPI sequence (TE/TR=25/1700 ms, 212 volumes). The rs-fMRI datasets were analyzed with AFNI, SPM8 and Matlab. Slice timing correction, head motion correction, white matter, CSF signal removal were performed. The necessity of global signal removal was determined individually by evaluating global negative index of each subject ⁶. Band-pass filter was used to keep low-frequency fluctuation between 0.012-0.1Hz. The datasets were then normalized to standard MNI space and resampled to isotropical 2mm. The seed ROI located on the left DLPFC was selected in accordance with literature ⁴. The cross-correlation coefficient map of each subject was generated by correlating time course of each voxel with that of the seed. After applying Fisher’s Z-transformation, the datasets of HC subjects were used to generate a pattern with significant FC to the seed which served as a mask for following analyses. Then we utilized a customized Matlab program to conduct relapse prediction based on the Cox regression model, in a voxel-wise fashion, to generate the relative hazard ratio (HR) map. Age, gender, years of education and mean head movement during the scan were served as covariates in the Cox regression analysis. To further evaluate the potential ability of the Cox model in predicting relapse at an individual level, we also performed the leave-one-out (LOO) cross validation. All analyses were thresholded at the rigorous p=0.001 and Monte-Carlo simulation implemented with 3dclustsim program of AFNI was used to correct for multiple comparisons. To evaluate the prediction power of the above analyses, we conducted receiver operating characteristic (ROC) analyses and calculated the area under curve (AUC) of the ROC of each voxel to generate the AUC maps.

Results

The voxel-wise Cox regression analysis revealed that two regions in the left DLPFC and left inferior frontal gyrus (IFG) survived the statistical threshold and both had HR values < 1 (Fig.1). Individuals with higher connectivity between these regions and the seed had lower relative relapse HR (or longer abstinence following the treatment), suggesting protective effects of the connectivity of these brain circuits against relapse. Fig. 2 shows the result of the ROC analyses for both without and with the LOO analyses of these two regions. AUCs of the ROC curves are ranging from 0.82~0.96 for the left DLPFC (Fig.2B left), and from 0.85~0.99 for the left IFG (Fig.2B right) without LOO. With the LOO cross-validation, the AUC ranges dropped to 0.72~0.89 for the left DLPFC (Fig.2C left) and 0.76~0.91 for the left IFG (Fig.2C right).

Discussion and Conclusion

Emerging concepts based on neuroimaging studies suggest that cognitive behaviors were supported by neural networks connected among discrete brain regions ⁷. The observed regions in the current study (the DLPFC and IFG) that serve as a protective factor against relapse are in accordance with principle of TMS with long-term potentiation effects that stimulating the target could lead to a momentary elevation of the BOLD signal in the directly stimulated area, as well as regions monosynaptically connected to the stimulated site ^8,9. These findings may serve as the neural mechanism of the effective manipulation of the behavioral outcome by TMS targeting the DLPFC, and extend our understanding of the neural underpinnings of the high relapse in addiction. The approach used in the study may have the potential to guide informed treatment such as evaluating the effectiveness of the stimulus site in TMS treatment.

Acknowledgements

The research was supported by the Intramural Research Program of the NIH, NIDA. We thank Dr. Bryon Adinoff for providing data on cocaine users.