【作者】 李锐；

【导师】 陈蕾；

【作者基本信息】 四川大学 ， 神经病学（专业学位）， 2023， 硕士

【摘要】 目的:通过对健康对照、癫痫患者癫痫发作前不同时间内的尿蛋白进行提取和分析,挖掘预警癫痫发作特异性的尿蛋白指标,构建基于尿蛋白组学的癫痫发作预警模型,从特异性尿蛋白的新角度探索癫痫发病机制的方向。材料和方法:本研究纳入了2020年1月至2022年12月在四川大学华西医院神经内科就诊的癫痫患者,按照年龄和性别匹配了健康对照。收集受试者的人口学、癫痫相关特征、既往病史、饮食习惯和生活习惯。受试者连续采集每日晨尿,癫痫患者还需收集癫痫发作后2小时内的尿液。根据癫痫发作时间,癫痫组尿液可分为癫痫发作前3天、前2天、前1天和发作后。使用Thermo Fisher Orbitrap Exploris 480蛋白质谱仪检测尿蛋白,使用串联质谱标签(Tandem Mass Tags pro16 plex)制备和标记肽样品。对癫痫组和健康组的蛋白进行差异蛋白分析和蛋白筛选,进行受试者工作特征曲线分析,通过受试者工作曲线下面积(area under curve,AUC)、灵敏度和特异度来评估模型的性能。使用自身交叉验证和训练集开发+验证集验证两种方法,分别建立了癫痫发作前3天、发作前2天和发作前1天的癫痫预警模型。结果:共纳入16例癫痫患者的61次癫痫发作及22例健康对照。与癫痫组相比,健康对照组学历较高(P=0.004),BMI更标准(P=0.004),在吸烟、饮酒、熬夜、早产、出生缺氧、伴随疾病、伴随用药、癫痫家族史及高热惊厥史等方面的差异无统计学意义(P>0.05)。纳入的患者癫痫均为局灶起源,癫痫起病年龄为16.50±6.84岁,癫痫病程为16.25±9.77年,癫痫的严重程度评分为9.56±5.61分。我们通过TMT pro 16 plex标记,在肽和蛋白水平上采用1%的错误发现率的标准,共筛选出3803个蛋白。原始样本变异系数的中位数为6.7%,技术可重复性的中位数为0.88,生物可重复性的中位数为0.94。蛋白的平均缺失率为38.2%,通过剔除缺失率大于80%的蛋白,填充缺失值,最小值替换、批次和组别效应移除之后,癫痫组和健康对照组共定量得到2937个蛋白。在差异倍数>1.2及P<0.05时,健康对照与癫痫患者癫痫发作前3天、前2天、前1天和当天的尿蛋白比较,共鉴定出41、355、23和365个差异蛋白。在癫痫发作前1天,我们选取了溶酶体α-甘露糖苷酶、POTE锚蛋白家族成员J和溶酶体α-葡萄糖苷酶等15个蛋白,将上述蛋白联合预测癫痫发作,采用100次10折交叉验证,模型的AUC、灵敏度和特异度分别为0.931、77.1%和94.1%。另外,我们选取了免疫球蛋白λ常数7、肌球蛋白重链7和丝氨酸蛋白酶抑制剂B13等8个尿蛋白进行了模型的构建并在验证集中进行验证。训练模型的AUC、灵敏度和特异度分别为1、100%和100%,验证模型的AUC、灵敏度和特异度分别为0.763、57.1%和95.5%。在癫痫发作前2天,我们选取了溶酶体α-甘露糖苷酶、神经节苷脂GM2激活剂和泛素结合酶E2变异体1等8个蛋白,将上述蛋白联合,采用100次10折交叉验证,预警模型的AUC、敏感度及特异度分别为0.884、88.4%和85.9%。另外,我们选取溶酶体α-甘露糖苷酶、内因子-维生素B12受体和Tudor结构域蛋白5等24个蛋白进行了模型的构建并在验证集中进行验证。训练模型的AUC、灵敏度、特异度和准确性分别为1、100%和100%。验证模型的AUC、灵敏度和特异度分别为0.550、23.1%和87.0%。在癫痫发作前3天,选取了β-半乳糖苷α-2,6唾液酸转移酶1、C4b结合蛋白β链、和丝氨酸蛋白酶抑制剂B13等9个蛋白,将上述蛋白联合,采用100次10折交叉验证,预警模型的AUC、灵敏度和特异度分别为0.838、87.1%和70.6%。另外,我们选取了血纤维蛋白肽B、纤维蛋白原γ链和丝氨酸蛋白酶抑制剂B13等6个尿蛋白进行了模型的构建并在验证集中进行验证。训练模型的AUC、灵敏度和特异度分别为0.974、89.5%和100%,验证模型的AUC、灵敏度和特异度分别为0.790、66.7%和91.3%。结论:癫痫患者尿蛋白与健康对照尿蛋白有明显差异,通过尿蛋白差异性分析,可筛选出能预警癫痫发作的生物标志物。基于尿蛋白组学的分析,可提前3天预警癫痫发作,模型的AUC多大于0.8,同时拥有较高的灵敏度和特异度。未来应针对筛选出的尿蛋白进一步研究与癫痫的关系,改进模型。更多还原

【Abstract】 Objective:By extracting and analyzing the urine protein of healthy controls and patients with epilepsy at different times before seizures,excavate the specific urinary protein indicators of early warning seizures,construct seizure early warning models based on urinary proteomics,and explore the direction of the pathogenesis of epilepsy from the new perspective of specific urine protein.Materials and Methods:From January 2020 to December 2022,patients with epilepsy admitted to the Department of Neurology,West China Hospital,Sichuan University were included in this study,and matched healthy controls by age and sex.The demographic characteristics,epileptic-related characteristics,past medical history,dietary habits and lifestyle habits of the subjects were collected.Daily morning urine was continuously collected from subjects,and urine was also collected from patients with epilepsy within 2 hours after seizure onset.According to the time of seizure,the urine of the epilepsy group could be divided into 3 days before seizure,2 days before seizure,1 day before seizure,and after seizure.Urine protein was detected using the Thermo Fisher Orbitrap Exploris 480 protein spectrometer,and peptide samples were prepared and labeled using the Tandem Mass Tags pro 16 plex.Differential protein analysis and protein screening were performed on proteins in the epilepsy group and healthy group,and the performance of the model was evaluated by the area under the receiver working curve(AUC),sensitivity and specificity.Using the two methods of self-crossvalidation and training set development combined with validation set verification,the epilepsy early warning model of 3 days before seizure,2 days before seizure and 1 day before seizure was established,respectively.Results:A total of 61 seizures from 16 patients with epilepsy and 22 healthy controls were enrolled.Compared with the epilepsy group,the healthy control group had a higher education level(P=0.004)and a more standard BMI(P=0.004).There was no significant difference in smoking,drinking,staying up late,preterm birth,hypoxia at birth,concomitant diseases,concomitant medication,family history of epilepsy,and history of febrile convulsion between the two groups(P > 0.05).All the patients with epilepsy were of focal origin.The mean age of seizure onset was 16.50±6.84 years,the mean duration of epilepsy was 16.25±9.77 years,and the epilepsy severity score was 9.56±5.61.We screened a total of 3803 proteins using the criteria of 1% false discovery rate at the peptide and protein levels by TMT pro 16 plex labeling.The median coefficient of variation of the original sample was 6.7%,the median technical reproducibility was0.88,and the median bioreproducibility was 0.94.The average missing rate of proteins was 38.2%.By eliminating proteins with missing rate greater than 80%,filling missing values,minimum replacement,batch and group effect removal,a total of 2937 proteins were quantified in the epilepsy group and the healthy control group.In the fold change > 1.2 and P < 0.05,41,355,23 and 365 differentially expressed proteins were identified between healthy controls and patients with epilepsy 3,2,1and the same day before seizures.One day before seizure onset,we selected 15 proteins,including lysosomal α-mannosidase,POTE anserine family member J and lysosomal α-glucosidase,and combined the above proteins to predict seizures.Using 100 times of 10-fold crossvalidation,the AUC,sensitivity and specificity of the model were 0.931,77.1% and94.1%,respectively.In addition,we selected 8 urinary proteins,including immunoglobulin λ constant 7,myosin heavy chain 7 and serine protease inhibitor B13,to construct the model and verify it in the validation set.The AUC,sensitivity and specificity of the training model were 1,100% and 100%,respectively,and those of the validation model were 0.763,57.1% and 95.5%,respectively.Two days before seizure onset,we selected 8 proteins,including lysosomal α-mannosidase,ganglioside GM2 activator and ubiquitin conjugated enzyme E2 variant1,and combined the above proteins.Using 100 times of 10-fold cross-validation,the AUC,sensitivity and specificity of the early warning model were 0.884,88.4% and85.9%,respectively.In addition,we selected 24 proteins,such as lysosomal α-mannosidase,intrinsic factor-vitamin B12 receptor and Tudor domain protein 5,to construct the model and verify it in the validation set.The AUC,sensitivity,specificity and accuracy of the training model were 1,100% and 100%,respectively.The AUC,sensitivity and specificity of the validation model were 0.550,23.1% and 87.0%,respectively.Three days before seizure onset,we selected 9 proteins,including β-galactosidaseα-2,6-sialyltransferase 1,C4 b binding protein β chain,and serine protease inhibitor B13.After 100 times of 10-fold cross-validation,the AUC,sensitivity and specificity of the early warning model were 0.838,87.1% and 70.6%,respectively.In addition,we selected 6 urinary proteins,including blood fibrinopeptide B,fibrinogen gamma chain and serine protease inhibitor B13,to construct the model and verify it in the validation set.The AUC,sensitivity and specificity of the training model were 0.974,89.5% and 100%,respectively,and those of the validation model were 0.790,66.7%and 91.3%,respectively.Conclusion:There was significant difference between the urine protein of patients with epilepsy and healthy controls.By analyzing the difference of urine protein,biomarkers that could be used to predict seizure could be screened.Based on the analysis of urinary proteomics,the seizure could be predicted 3 days in advance,and the AUC of the model was more than 0.8,with high sensitivity and specificity.In the future,the relationship between urinary protein and epilepsy should be further studied to improve the model.更多还原