A Study on Clinical variables and Blood Proteins related to Diagnosis, reflect Severity, and Progress of High-Risk State of Psychiatric Disorders

Introduction: Early diagnosis and treatment of major mental disorders are very important in improving the prognosis and quality of life of young age. The existing diagnostic system has limitations in diagnosing high-risk patients with non-specific symptoms, and it is necessary to find biological markers that can reflect the disease status of high-risk groups. Therefore, in this study, blood proteins were quantitatively analyzed for high-risk groups of major mental illness, and protein markers reflecting the state of high-risk groups and disease progress were explored in various ways. Method: 90 clinical high-risk subjects recruited from 7 medical institutions in Seoul Clinical evaluation and blood collection were performed on the same day, and blood proteins were quantified using 158 target peptides with multiple reaction monitoring-mass spectrometry. The analyzed protein expression in the high-risk group was compared with that in the patient group and the normal control group. Protein markers reflecting the high-risk group status were explored(Study1), In addition, through latent profile analysis, heterogeneous characteristics of high-risk were categorized, and correlation analysis was performed to search for proteins reflecting symptom severity(Study2). 10-crossfold LASSO regression was repeated 100 times for each model, features predict transition to major psychiatric disorders, and recovery of symptoms were selected. Predictive performance of the models were measured by area under the receiver operating characteristics (Study3). Results: Through study 1, Twenty-seven protein markers reflecting high-risk conditions were identified. Most of proteins were reported in previous studies related to psychiatric disorders. In study 2, clinical high-risk groups were categorized into three subgroups, and it was different from categorized group in study 1. Twenty-nine protein markers with significant correlation with two or more of the 19 clinical indicators could be selected. In study 3, clinical variables (9 markers in clinical models, 5 markers in mixed models) and protein (15 markers in proteomic models, 2 markers in mixed models) were selected to predict the transition to psychiatric disorders. The performance of the mixed model (Model 1c) was showed excellent performance in the test data (AUC 0.90-0.97). On the other hand, clinical variables (8 markers in clinical models, 9 markers mixed models) and protein variables (8 markers in clinical models, 5 markers in mixed models) were selected to predict symptom recovery of symptom. However, the performance of the prediction model was fair at AUC 0.70-0.75 in the test data. Protein markers discovered through study 1-3 were integrated and most of the significant proteins were related to the blood coagulation cascade, immune response, lipid metabolism, and antioxidant response pathways reported in previous study. Conclusion: Through this study, it was confirmed that meaningful biological alterations reported in the patient group were also observed in the patients with high-risk state. In addition, it was possible to identify common protein markers that could reflect the severity of symptoms, and predict the progression of disease. However, in order to validate our results and confirm the stability of the discovered protein markers, it will be necessary to evaluate the longitudinal variation of the protein markers in the future. Nevertheless, the protein and clinical indicators discovered in this study are expected to be used as useful markers for early diagnosis of the conditions of high-risk states in the future and predicting the course of the disease.서론: 주요정신질환의 조기진단과 치료는 젊은 연령의 고위험군 환자들의 경과 호전 및 삶의 질 향상에 매우 중요하다. 기존진단분류체계로는 증상이 비특이적인 고위험군 환자를 진단하는 것에 한계가 있으며, 고위험군의 질병 상태를 반영할 수 있는 생물학적 표지자의 발굴이 필요하다. 이에 본 연구에서는 주요정신질환의 고위험군을 대상으로 혈액 단백체를 정량 분석하여, 고위험군 상태와 질병 경과를 반영하는 단백체 마커를 다양한 방법으로 탐색하고자 하였다. 방법: 서울지역 7개의 의료기관으로부터 의뢰된 90명의 고위험군을 모집하였고, 임상평가와 채혈을 같은 날에 시행하였다. 혈중 단백질은 158개의 표적 펩타이드를 사용하여 다중 검지법으로 정량화하였다. 분석된 고위험군의 혈중 단백질 발현량을 환자군과 정상 대조군의 단백질 발현량과 비교하였고 이를 통해 고위험군 상태를 반영하는 단백체 마커를 탐색하였다(연구 1), 또한, 잠재프로파일 분석을 통해 고위험군의 이질적인 임상특성을 유형화하고, 증상 중증도 반영하는 단백체를 탐색하기 위해 상관분석을 수행하였다(연구 2). 마지막으로, 초기 평가 시 수집된 혈액 단백질 중 주요정신질환의 경과를 예측하는 임상, 단백체 마커를 탐색하기 위하여 LASSO 회귀분석을 통해 통계적으로 의미 있는 변수를 선별하였다. 결과 변수로 주요정신질환으로의 발병(Model 1), 임상 경과의 호전(Model 2)으로 두 개의 모델을 구성하였으며, 각 모델에서 3개의 하위 모델(Model a –임상 변인 모델, Model b-단백체 변인 모델, Model c-혼합 모델)로 나누어, 선별된 변수들이 포함된 모델의 성능을 AUROC로 평가하였다 (연구3). 결과: 연구 1을 통해 고위험 상태를 반영하는 27종류의 단백체 마커를 확인하였고, 이 중 대부분의 마커들이 선행 환자군 연구에서 보고된 마커임 확인하였다. 연구 2에서는 잠재프로파일 분석을 통하여, 3개의 하위군으로 고위험군을 분류하였으며, 증상으로 유형화된 분류는 연구 1의 범주와 다름을 확인하였다. 또한, 19개의 임상지표 중 2개 이상의 지표와 유의한 상관관계를 보인 단백체 마커 29개를 선별할 수 있었다. 연구3에서는 고위험군의 발병 여부를 예측하는 임상 변인(Model 1a 9개, Model 1c 5개) 및 단백체(Model 1b 15개, Model 1c 2개) 변인을 선별하였고, 선별된 단백체 변인으로 분석한 모델의 성능을 평가하였다. 혼합 모형인 Model 1c의 성능이 테스트 데이터에서 안정적으로 우수 하였다(AUC 0.90-0.97). 반면 고위험군의 증상 호전을 예측하는 임상 변인(Model 2a 8개, Model 2c 9개), 단백체 변인(Model 2b 8개, Model 2c 5개)을 선별하고, 예측모델의 성능은 테스트 데이터에서 Model 2c AUC 0.70-0.75로 가장 높았다. 연구 1-3을 통해 발굴된 단백체 마커의 대부분은 단백질들이 선행 환자군 대상연구에서 보고된 것이었으며, 관련된 생물학적 경로로는 혈액 응고과정, 면역반응, 지질대사, 항산화 반응 경로였다. 결론: 본 연구를 통해 주요정신질환의 발병 전 고위험군 상태에서도 환자군에서 보고되었던 의미 있는 생물학적 변이가 관찰됨을 확인할 수 있었다. 뿐만 아니라, 고위험군의 증상 중증도를 반영하고, 고위험군의 발병 여부 및 증상 호전경과를 예측할 수 있는 공통된 단백체 마커를 확인할 수 있었다. 그러나 발굴된 단백체 마커의 정확도 안정성의 확보와, 예측 모델의 정확도를 향상시키기 위해서 향후 단백체 마커의 종단적 변화량을 평가하는 것이 필요 할 것이다. 그럼에도 본 연구에서 발굴된 단백질 및 임상지표는 향후 주요정신질환 고위험군의 상태를 생물학적으로 조기진단하고, 질병의 경과를 예측하는 유용한 마커로 활용될 수 있을 것으로 전망한다.초록 i 표 목차 vi 그림 목차 ix 약어 목록 xi 제 1 장 서 론 1 제 2 장 연구 1 7 제 1 절 연구대상 및 방법 8 제 2 절 결과 18 제 3 절 고찰 24 제 3 장 연구 2 28 제 1 절 연구대상 및 방법 29 제 2 절 결과 39 제 3 절 고찰 43 제 4 장 연구 3 47 제 1 절 연구대상 및 방법 48 제 2 절 결과 54 제 3 절 고찰 63 제 5 장 종합요약 66 표 목차 [표 1-1]Inclusion and exclusion criteria of clinical high risk group 68 [표 1-2]Demographic and clinical characteristics of the study subjects related to HC, CHR-BD, and BD 70 [표 1-3]Classification of subgroup of CHR-BD according to BARS criteria 71 [표 1-4]Differentially expressed peptides(ANOVA) identified by LC-MRM analysis between HC, CHR-BP, and BD. 72 [표 1-5]Differentially expressed peptides(ANCOVA) identified by LC-MRM analysis between HC, CHR-BD, and BD 74 [표 1-6]Demographic and clinical characteristics of the HC, CHR-PRS and FEP 76 [표 1-7]Differentially expressed peptides(ANOVA) identified by LC-MRM analysis between HC, CHR-PRS, and FEP. 77 [표 1-8]Differentially expressed peptides(ANCOVA) identified by LC-MRM analysis between HC, CHR-PRS, and FEP 78 [표 1-9]Comparison of baseline demographics and clinical characteristics between CHR-PRS vs. CHR-BD 79 [표 1-10]Differentially expressed peptides identified by LC-MRM analysis between CHR-PRS and CHR-BD 80 [표 2-1]Fit indices of LPA 81 [표 2-2]Comparison of baseline demographics between LS, DPD, and MD group 82 [표 2-3]Comparison of clinical characteristics between Low LS, DPD, and MD group 83 [표 2-4]Differentially expressed peptides(ANOVA) identified by LC-MRM analysis between LS, DPD, and MD 85 [표 2-5]Pearson's correlation coefficients(r) between clinical variables and 29 peptides 88 [표 3-1]List of baseline 110 clinical features included in models 89 [표 3-2]List of baseline 158 proteomic features included in models 90 [표 3-3]Summary of prediction model. 91 [표 3-4]Comparison of baseline demographical characteristics between CHR-T vs.CHR-NT 92 [표 3-5]Comparison of baseline symptom characteristics between CHR-T vs. CHR-NT 93 [표 3-6]Comparison of baseline proteomic data between CHR-T vs. CHR-NT 95 [표 3-7A]Unique models of Model 1a based on the combinations of features selected 96 [표 3-7B]Unique models of Model 1b based on the combinations of features selected 97 [표 3-7C]Unique models of Model 1c based on the combinations of features selected 98 [표 3-8]Features included in prediction models(Model 1) 99 [표 3-9]Comparison of baseline demographical characteristics between Recovered vs. Not-recovered group 100 [표 3-10]Comparison of baseline symptom characteristics between Recovered vs. Not-recovered group 101 [표 3-11]Comparison of baseline proteomic data between Recovered vs. Not-recovered group 104 [표 3-12A]Unique models of Model 2a based on the combinations of features selected 105 [표 3-12B]Unique models of Model 2b based on the combinations of features selected 106 [표 3-12C]Unique models of Model 2c based on the combinations of features selected 108 [표 3-13]Features included in prediction models(Model 2) 110 그림 목차 [그림 1-1]Determination of quantifiable targets for CHR of psychiatric disorders 111 [그림 1-2]Summary of DEPs in CHR-BD and CHR-PRS 112 [그림 1-3]Correlation plot showing Pearson’s correlation analysis between clinical variable(PANSS) and DEPs 113 [그림 1-4]Correlation plot showing Pearson’s correlation analysis between clinical variable(HAM-D, HAM-A) and DEPs 114 [그림 1-5]Correlation plot showing Pearson’s correlation analysis between clinical variable(A-YMRS, B-GAF, CGI) and DEPs 115 [그림 1-6]Bioinformatics analysis of the proteins that differentially expressed in CHR 116 [그림 2-1]Mean and SD of clinical variables between Class 1 (Low severity group), Class 2 (Depressive-psychosis dominant group), and Class 3 (Manic-dominant group) 117 [그림 2-2]Correlation plot showing Pearson’s correlation analysis between 19 clinical variable and peptides 118 [그림 2-3]Bioinformatics analysis of the proteins significantly reflect severity of symptoms 119 [그림 3-1]Volcano plot of p-values for DEPs between CHR-T and CHR-NT 120 [그림 3-2]Feature selection among 100 models analyzed from repeated application of LASSO regression with 10-fold CV on the training data(Dependent variable : Transition) 121 [그림 3-3]ROC curves showing model performance in predicting the probability of Transition(CHR-T). 122 [그림 3-4]Volcano plot of p-values for DEPs between Recovered and Not-recovered group 123 [그림 3-5]Feature selection among 100 models analyzed from repeated application of LASSO regression with 10-fold CV on the training data(Dependent variable: Symptom recover) 124 [그림 3-6]ROC curves showing model performance in predicting the probability of disease recover 125 [그림 4] Summary of the results of study 1-3 126 참고문헌 127 Abstract 144박