A Study On The Neurophysiological Basis Of Response Inhibition In Adolescent Depression With Non-Suicidal Self-Injury

PartⅠ:An ERP study on the response inhibition in adolescent depression with non-suicidal self-injuryObjective:Nonsuicidal self-injury(NSSI)may be a type of behavioral addiction,and cue reactivity is one of the characteristics of addiction disorders.The purpose of this study was to explore whether the behavioral performance and electrophysiological indexs of response inhibition during exposure to self-injury-related cues were abnormal in adolescent depression with NSSI.Methods:Forty-seven patients with major depressive disorder(MDD)with NSSI,39 patients with MDD without NSSI,and 25 healthy controls(HC)between 12-18 years old completed two-choice Oddball paradigm and EEG was acquired.Accuracy cost and reaction time(RT)cost were used as behavioral indexes;the latencies and amplitudes of difference waves N2d and P3d derived from deviation stimulus minus standard stimulus were used as electrophysiological indexes.We used the ERP Reliability Analysis(ERA)toolbox to estimate the reliability for N2 and P3,with a threshold of 0.70.Results:Demographic analysis showed that there were significant differences in age between groups,mainly because the age of the MDD+NSSI group was significantly lower than that of the HC group and the MDD group(p=0.005 and p=0.048,respectively).Age was not significantly different between the MDD group and the HC group.The results of behavioral indexs showed that the main effect and interaction effect of accuracy cost were not significant regardless of whether considering age as a covariate.Regarding the RT cost,the main effect of cue was significant[F(1,103)=24.65,p<0.001,η_p~2=0.19]and the main effect of group was significant[F(2,103)=3.46,p=0.04,η_p~2=0.06].After considering age as a covariate,the main effect of cue was no longer significant;the main effect of group remained significant[F(2,102)=5.25,p=0.007,η_p~2=0.09],mainly due to the RT cost of the MDD+NSSI and MDD group was higher than that of the HC group.The results of electrophysiological index showed that the group×cue interaction effect of P3d amplitude was significant[F(2,108)=4.54,p=0.01,η_p~2=0.08].After considering age as a covariate,the interaction effect of group×cue remained significant[F(2,107)=5.85,p=0.004,η_p~2=0.10].Simple effects analysis showed that the P3d amplitude in the MDD+NSSI group was significantly larger than that in the HC group during exposure to self-injury related cues(p=0.046 after Bonferroni correction).In the MDD and MDD+NSSI groups,the P3d amplitudes with self-injury cue were significantly larger than those with neutral cues(Bonferroni-corrected p=0.038 and 0.009,respectively);however,in the HC group,the P3d amplitudes with self-injury related cues were significantly smaller than those with neutral cues(p=0.038 after Bonferroni correction).The N2 and P3 reliability estimates showed that the MDD+NSSI group,the MDD group and the HC group all have acceptable reliability under various conditions(all greater than 0.7).Conclusions:We provide electrophysiological evidence of altered neural reactivity to self-injury related cues in depressed adolescents with NSSI.Specifically,self-injury related cues induce significantly larger P3amplitudes in adolescents with NSSI.Future longitudinal studies are needed to explore the role of cue reactivity in predicting future self-harm and suicidal behavior.Part Ⅱ: A study on the microstate dynamics and sequence during response inhibition in adolescent depression with non-suicidal self-injuryObjective: EEG microstate refers to maintaining a relatively stable state in a short period(about 80-120 milliseconds),and is an important biomarker for neuropsychiatric diseases.However,there is little study on microstates related to nonsuicidal self-injury(NSSI).We aimed to explore whether the underlying dynamics and sequences of microstates during response inhibition are abnormal in depressed adolescents with NSSI.Methods: The traditional dynamics of microstates as well as microstate sequences were extracted using the Microstate 0.3 plugin in MATLAB.The microstate dynamics of interest include the average duration,contribution,and occurrence.The sample entropy of microstate sequences with different conditions in three groups was calculated separately.We used Phenotype Seeker to extract k-mers(short segments)in microstate sequences.Three-way analysis of variance(ANOVA)was performed on the dynamics of microstates and k-mers of microstate sequences,respectively.The cue condition was regarded as a within-subject factor(2 levels: neutral cue and self-injury related cue),the stimulus was regarded as a within-subject factor(2 levels,standard stimulus and deviation stimulus),and the group was regarded as a between-subject factor (3 levels: HC,MDD and MDD+NSSI).Bonferroni correction was used for multiple comparisons.Results: 1.The ANOVA results considering age as a covariate showed that the cue main effect of the mean duration of microstate A was significant [F(1,107)=7.814,p=0.006,η_p~2 =0.068],mainly due to the average duration of microstate A under self-injury cues was longer than that under neutral cues in all groups.The stimulus × group interaction of microstate B was significant [F(1,107)=4.272,p=0.016,η_p~2 =0.074] and simple effect analysis showed that: in the MDD+NSSI group,the duration of microstate B under deviant stimuli was significantly longer than that under standard stimuli,while no such significant difference was found in the HC group and MDD group.Regarding the occurrence,the cue main effect of microstate C was significant [F(1,107)=4.917,p=0.029,η_p~2=0.044],mainly because the occurrence of microstate C under self-injury cues was less frequently than that under neutral cues.2.Regarding the analyses on sample entropy,the sample entropy of microstate sequence in HC group,MDD group and MDD+NSSI group with various conditions is significantly smaller than that of random sequence.However,no significant difference was found between groups and conditions.3.Regarding the analyses on k-mers of microstate sequences,there are significant non-random short segments of microstate sequences in all three groups,and there are significant group main effects and significant interaction effects of group × cue,group × stimuli and group × stimuli ×cue for some short segments.Conclusions: 1.Depressed adolescents with NSSI have abnormal brain network dynamics on a large scale compared with MDD and HC groups;2.The microstate sequence in adolescent depression with or without NSSI still maintains the regularity similar to that of the healthy control group;3.There are significant non-random microstate sequence short segments in the HC group,MDD group and MDD+NSSI group during the two choice Oddball task.The microstate sequence k-mers may be a potential neural biomarker for NSSI and MDD.Part Ⅲ: A machine learning research based on ERP and microstate sequence k-mersObjective: We aimed to investigate whether ERP or EEG microstate features during response inhibition can correctly classify depressed patients with non-suicidal self-injury(NSSI)and those without NSSI at the individual level,and whether ERP or EEG microstate features during response inhibition can predict the short-term prognosis of NSSI at the individual level.Methods: 1.Based on the amplitudes and latencies of N2 d and P3 d or short segments of microstate sequence,we performed the cross-sectional classification of depressed patients with NSSI and those without NSSI.2.Based on the amplitudes and latencies of N2 d and P3 d or short segments of microstate sequence,we tried to predict the short-term prognosis of NSSI.We used t-test to filter original features,and selected 20% of the features to input into the machine learning model.We used the Support Vector Machine(SVM)linear kernel model for training,and used the grid search method to tune hyperparameters.Then,we select the optimal hyperparameters to train the model.We used the Leave-One-Out Cross Validation(LOOCV)to calculate the average accuracy for evaluating model performance.The receiver operating characteristic curve(ROC)was drawn and area under the curve(AUC)was calculated.Results: 1.The cross-sectional classification results showed that:when only considering the amplitudes and latencies of difference wave N2 d and P3 d as features,the average accuracy of LOOCV is 61.63% and the AUC is 0.40.The results suggested that considering only the amplitudes and latencies of N2 d and P3 d is not effective in classifying depressed patients with or without NSSI,and the model performance is similar to random classifiers.When short segments of microstate sequences are used as features,the average accuracy of LOOCV is 75.58%,the sensitivity is0.83,the specificity is 0.67,and the AUC is 0.79.The top five features according to the weights are: ADACDCDC under neutral cue and deviant stimuli,CDCDCBCA under neutral cue and deviant stimuli,CACDCAC under self-injury cue and standard stimuli,CDCDADCD under self-injury cue and deviant stimuli,and DCDCACDA under neutral cue and deviant stimuli.2.The machine learning results of longitudinal prediction in 3 months show that when considering only the amplitudes and latencies of N2 d and P3 d as features,the average accuracy of LOOCV is 64.29%.The results suggested that considering only the amplitudes and latencies of N2 d and P3 d is not effective to predict short-term prognosis of NSSI at the individual level,and the model performance is similar to the random classifiers.When short segments of microstate sequences are used as features,the average accuracy of LOOCV is 85.71%,the sensitivity is 0.95, the specificity is 0.67,and the AUC is 0.87.The top five features according to the weights are: CACDCC under self-injury cue and deviant stimuli,CDCACAC under neutral cue and deviant stimuli,CDCDCAD and CDCACAC under self-injury cue and standard stimuli,and ADCDCDCA under self-injury cue and deviant stimuli.Conclusion: The amplitudes and latencies of N2 d and P3 d based on the two-choice Oddball paradigm cannot correctly distinguish depressed patients with NSSI from those without NSSI at the individual level,and cannot predict the short-term prognosis of NSSI.However,short segments of microstate sequences during performing the two-choice Oddball paradigm can correctly distinguish depressed patients with NSSI from those without NSSI at the individual level,and can predict the short-term prognosis of NSSI.This suggested that short segments of microstate sequences can serve as potential biomarkers for diagnostic identification and prognosis prediction in adolescents with NSSI.