Altered Gray Matter Structure And Networks In Patients With Paroxysmal Kinesigenic Dyskinesia Based On Structural Magnetic Resonance Imaging

Objective:Paroxysmal kinesigenic dyskinesia(PKD)is a rare movement disorder characterized by brief attacks of dystonia,choreoathetosis,ballism or any combination of these movements which are mainly triggered by sudden voluntary movement.The disease often occurs in children and adolescence,so the normal neurodevelopmental process of patients may be affected by the disease.Family genetic tendency is commonly noted in most PKD patients,and the inheritance mode is mainly autosomal dominant.The proline-rich transmembrane protein 2(PRRT2)has been identified as the causative gene for PKD and accounts for 27-65% of PKD.However,the underlying pathophysiological mechanisms of PKD are not fully understood.Its diagnosis mainly depends on clinical symptoms and clinical experience of neurologists.Because of the rarity,complex clinical manifestations and lack of specific diagnostic markers,PKD is usually misdiagnosed.And previous studies on PKD are relatively insufficient,especially neuroimaging studies which are less than 10,limiting our understanding of this disease.Therefore,in this study we explored the alterations of gray matter structure and gray matter network in PKD patients based on 3DT1 structural MRI,hoping to clarify the pathophysiological mechanism of PKD and assist in its clinical diagnosis.Meanwhile,although PRRT2 gene has been identified as the causative gene for PKD,the specific pathogenesis of PRRT2 gene needs to further investigate.For this reason,we combined the genetic information of PKD patients with image data to explore how the PRRT2 mutation affects the gray matter structural network of PKD patients in order to find the pathogenic mechanism of PRRT2 mutation in PKD patients.On the basis of the above research,we tried to use machine learning method to classify PKD patients and normal people based on gray matter structural measures,hoping to build a robust classifier which can assist the clinical diagnosis of PKD.Materials and Methods:Part 1: Forty-five drug-naive patients with PKD and 50 healthy controls were recruited in this study.All participants were scanned with MRI to obtain 3DT1 structural images.First of all,we used Freesurfer(version 6.0.0)software to preprocess the structural images.Then,we extracted the cortical thickness,cortical surface area and gray matter volume of all subjects for group-wise statistical analysis based on the Desikan-killiany template;Next,the correlation between the structural changes of gray matter and clinical features was analyzed.Part 2: We totally recruited 87 patients with PKD and 115 healthy controls in the present study.3DT1 structural images were also obtained for all participants.We used SPM12 software and DARTEL methods to preprocess the structural images to obtain the smoothed and modulated gray matter images,which comprise morphological volume information for each voxel.Then we used the automated anatomical labeling algorithm to parcellate the brain gray matter into 90 anatomical regions.The kernel density estimation method was used to estimate the probability density function(PDF)of the gray matter intensity value in each anatomical region.We used the KullbackLeibler divergence-based methods to describe the similarities of PDFs of different anatomical regions to generate a 90 × 90 similarity matrix for each participant.We used the MATLAB-based GRETNA toolbox to calculate the topological properties of the brain gray matter network,including small-world parameters,network efficiency parameters,and node topological centralities.Between-group differences in brain network parameters were analyzed by non-parametric permutation tests.After that,the correlation analyses between altered brain network parameters and clinical symptoms were performed using SPSS.To further investigate whether individual gray matter measures can be used to detect PKD at the individual level,we applied support vector machine(SVM)to the preprocessed gray matter images and gray matter network matrices(90 × 90 Pearson correlation matrices)to classify PKD out of healthy controls.Statistical significance was estimated using the permutation method(1000permutations).Part 3: Ninety-seven PKD patients(47 PRRT2 mutation carriers,15 PRRT2 nonmutation carriers)and 60 matched healthy controls were recruited in this study.We also used SPM12 and DARTEL to preprocess the structural images to obtain the smoothed and modulated gray matter images.Then we used the statistical similarity of regional gray matter volume which contains three-dimensional structure of cerebral cortex to construct gray matter structural network.A 90 × 90 similarity matrix was also generated for each subject.Then,we used the GRETNA software to calculate the topological properties of gray structural network,including global and nodal network properties.Then,the network parameters were also calculated and compared among PRRT2 mutation carriers,PRRT2 non-mutation carriers and healthy controls.The correlation analyses between altered brain network parameters and clinical features were also performed using SPSS.Results:The main results are as follows:Part 1: In terms of the gray matter morphological features,we found that drugnaive PKD patients exhibited an increase in volume of bilateral thalamus and left postcentral gyrus compared with healthy controls.And the increased volume in left thalamus was negatively correlated with the age of onset and positively correlated with the duration of diseasePart 2: Compared with health controls,we found that:(1)at the global level,the PKD patients showed a significant increase in Lp and decreases in Eloc,Cp,γ and σ;(2)at the nodal level,the PKD patients showed nodal centrality changes in multiple brain regions,mainly in the basal ganglia-thalamus circuitry,default-mode network(DMN)and central executive network(CEN);(3)the Cp of the gray matter structural network in patients with PKD was negatively correlated with the duration of disease and positively correlated with the age of onset;(4)in the SVM analysis,when using whole-brain normalized gray matter images data,the mean balanced accuracy of classification was close to the chance level,in particular,the balanced accuracy is60.06%,sensitivity is 38.19%,and specificity is 81.89%.In contrast,using gray matter network matrices,the mean balanced accuracy of classification was 87.8%,with sensitivity 87.6% and specificity 88.0%.Having identified gray matter structural network matrices based SVM classifier as a powerful measure for detecting PKD,we examined the regions contributing to their superior performance.The top 20 brain regions with the highest classification values mainly located in basal ganglia-thalamocortical circuit.Part 3: Compared with health controls,the global properties of PRRT2 nonmutation carriers had no significant alteration,while the global properties of PRRT2 mutation carriers,including Eglob and Eloc,were significantly decreased,and the Lp was significantly increased.At the nodal level,compared with health controls,the nodal centrality of bilateral thalamus,left middle cingulate gyrus and paracingulate gyrus were decreased in both PRRT2 mutation carriers and non-carriers;Compared with health controls and PRRT2 non-mutation carriers,PRRT2 mutation carriers showed decreased nodal centralities in left supplementary motor area,bilateral angular gyrus,right caudate nucleus,bilateral globus pallidus and left superior temporal gyrus,and increased nodal centralities in right cingulate gyrus.Conclusions:Our findings of volume abnormalities in the bilateral thalamus and left posterior central gyrus in drug-naive PKD patients provided neuroanatomical evidence for the pathogenesis of PKD.Meanwhile,we found the global integration and local separation ability of gray matter structural networks in PKD patients were decreased,showing a "weaker small-worldization" pattern.And further analysis found that such "weaker small-worldization" changes in PKD patients may be caused by the mutation of PRRT2,which is helpful to elucidate the pathogenesis of PRRT2.Moreover,the regions with nodal centrality alterations of PKD patients were mainly located in the basal ganglia-thalamic loop,suggesting that basal ganglia-thalamic loop plays a key role in the neuropathology mechanism of PKD.Meanwhile,using machine learning methods,we found that the individual gray matter network information could differentiate individual PKD patients from healthy individuals with high accuracy.These findings may provide new insights into the pathophysiology of PKD and aid development of new biomarkers for clinical diagnosis.The innovation of this study is to combine gray matter structure,gray matter structural network,pathogenic gene and machine learning to explore the alterations of gray matter anatomical structure and gray matter structural network in patients with PKD,and whether these alterations are affected by PRRT2 genes.Our study further clarifies the pathological mechanism of PKD and provides important clues for clinical individual diagnosis of PKD.