利用脫附型電噴灑游離法結合胰蛋白消化開發蛋白質質譜影像平台 利用液相層析質譜法結合機器學習進行原發性醛固酮增多症之非標的代謝體分析以進行輔助診斷

Abstract

目前多數的人體病理組織樣本皆採用福馬林固定石蠟包埋方式進行保存，其檢體具備長期穩定性，並廣泛應用於臨床與研究領域。而蛋白質體學在探討生物系統中扮演關鍵角色，因此蛋白質在福馬林固定石蠟包埋組織中的空間分布資訊，更可提供細胞功能與疾病機制的重要見解。質譜影像技術是一種能夠視覺化組織中蛋白質分布的強大工具。傳統上，以基質輔助雷射脫附游離飛行時間質譜為主流，但近年來，由於大氣游離技術如脫附電噴灑游離法具備免基質、樣本前處理需求低等優勢，亦逐漸受到重視。 然而，目前多數利用脫附電噴灑游離法進行蛋白質的質譜影像研究仍著重於新鮮冷凍組織中直接偵測完整蛋白，該方法不僅蛋白偵測數量有限，亦需低溫保存樣本，不利於長期應用。因此，本研究目標為建構一套整合脫附電噴灑游離法與組織內胰蛋白酶消化之質譜影像平台，以應用於福馬林固定石蠟包埋組織中蛋白質空間分布分析，提升可偵測蛋白的數量。 本研究首先以小鼠腦組織進行方法優化，接著應用於犬黑色素細胞瘤組織進行方法驗證。結果共鑑定出79條胜肽、對應61種蛋白，顯著高於先前報導之脫附電噴灑游離質譜影像的研究成果。犬腫瘤組織中觀察到的蛋白分布亦與既有文獻一致。綜合結果顯示，本研究成功建立一套結合胰蛋白酶消化與脫附電噴灑游離質譜影像進行福馬林固定石蠟包埋組織中蛋白質空間的分析平台，展現其於臨床保存樣本中進行蛋白質體學研究的高度潛力。 原發性高醛固酮症是因腎上腺過度分泌醛固酮所致的次發性高血壓，使腎臟大量再吸收鈉離子及水分，導致血液總量增加而使血壓升高，若能針對腎上腺問題進行治療，即能根治高血壓，因此診斷原發性高醛固酮症便十分重要。現今臨床上以周邊血液中的醛固酮與腎素濃度比值作為初步診斷指標。根據台灣醫療標準，當高血壓患者以藥物控制不佳且醛固酮與腎素濃度比值的數值大於35，即可能罹患原發性高醛固酮症，必須進一步接受確認性檢查，以決定後續的治療方式。 然而，此診斷方法存在諸多限制，例如醛固酮與腎素濃度比值的篩檢具有極高的偽陽性率，而後續的確認性試驗耗時且對於高血壓病人具有健康風險。因此，開發一套更簡便且準確的診斷方法相當重要。本研究旨在利用液相層析質譜儀針對原發性高醛固酮症患者進行非標靶代謝體分析，並結合統計與機器學習模型，篩選出能用於疾病診斷的關鍵代謝分子。 本研究收集163位原發性高血壓及156位原發性高醛固酮症患者的周邊血液樣本，經萃取流程後，以液相層析質譜儀分析其代謝體特徵，並利用統計方法及機器學習模型進行分析。結果顯示，兩組患者的代謝體特徵可有效區分，並透過統計方法找出多個具顯著差異的代謝物。此外，透過機器學習篩選出19個具診斷潛力的代謝分子，並在測試集中達到91.8%的正確率，證明本方法具有良好的診斷效能。本研究結果顯示，結合非標靶代謝體分析與機器學習可作為原發性高醛固酮症診斷之新穎方法，未來具高度臨床應用潛力，可望提高診斷的準確性與效率。

The majority of human pathological specimens are preserved using the formalin-fixed, paraffin-embedded (FFPE) method due to its long-term stability and wide applicability in clinical and research settings. Proteomics plays a crucial role in understanding biological systems, and mapping the spatial distribution of proteins in FFPE tissues can provide valuable insights into cellular function and disease mechanisms. Mass spectrometry imaging (MSI) is a powerful technique for visualizing protein localization within tissue sections. While matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) has been widely used for protein MSI, ambient ionization techniques such as desorption electrospray ionization (DESI) are gaining more application due to their minimal sample preparation and matrix-free operation. However, most DESI-MSI studies have focused on intact proteins in fresh-frozen tissues, which limits protein coverage and requires stringent storage conditions. To overcome these challenges, we established a DESI-MSI platform for spatial proteomic analysis of FFPE tissues by incorporating on-tissue trypsin digestion to enhance protein detectability. We optimized the DESI parameters using FFPE mouse brain tissue and then validated the method on FFPE dog melanocytoma tissue. As a result, we identified 79 peptides corresponding to 61 proteins in mouse brain tissue—exceeding the number reported in previous DESI-MSI studies. The protein distribution patterns observed in melanocytoma tissue were consistent with previously reported biomarkers. This work establishes a robust platform for protein imaging in FFPE tissues using DESI-MSI combined with trypsin digestion, expanding the potential for spatial proteomic analysis in archived clinical specimens. Primary aldosteronism (PA) is a form of secondary hypertension characterized by excessive aldosterone secretion from the adrenal glands, resulting in increased sodium and water reabsorption in the kidneys, thereby elevating blood volume and blood pressure. Clinically, the aldosterone-to-renin ratio (ARR) measured from peripheral blood samples is commonly used as a preliminary diagnostic indicator. According to Taiwanese clinical guidelines, patients whose blood pressure remains uncontrolled despite medication and have an ARR greater than 35 are suspected of having PA and require confirmatory diagnostic tests to determine further therapeutic interventions. However, the current diagnostic approach has significant limitations, including a high false-positive rate associated with ARR screening, along with time-consuming and potentially risky confirmatory procedures. Therefore, this study aims to utilize untargeted metabolomic analysis through liquid chromatography-mass spectrometry (LC-MS) combined with statistical analyses and machine learning algorithms to identify key metabolites for diagnosing PA. We collected plasma samples from 163 patients with essential hypertension and 156 patients diagnosed with primary aldosteronism. After extraction, metabolomic profiles were obtained using LC-MS, and data were analyzed using statistical and machine learning techniques. The results demonstrated clear differentiation between the two groups based on their metabolomic profiles, with several significantly different metabolites identified. Furthermore, machine learning analysis selected 19 metabolites with significant diagnostic potential, achieving an accuracy 91.8% in the testing set. These findings suggest that untargeted metabolomic analysis combined with machine learning represents a promising approach for improving the diagnosis of primary aldosteronism, potentially enhancing clinical efficiency and accuracy in practice.