利用質譜開發感染性疾病的臨床診斷工具

Abstract

感染相關疾病仍是全球健康的重大威脅，涵蓋從常見細菌感染到如敗血性休克等危及生命的併發症，不僅導致高罹病與死亡率，亦造成沉重的醫療與經濟負擔。這類疾病常需即時處置，尤其是在急重症情境中，若無法于黃金時機啓動適當治療，可能導致臨床惡化甚至死亡。然而，現行診斷流程多仰賴耗時的培養程序或複雜的分子檢測技術，導致診斷遲滯與治療延誤，影響臨床决策與病患預後。隨著抗藥性病原體不斷出現，臨床迫切需要能够快速、準確、易于應用的診斷工具，以因應感染相關臨床挑戰。 本論文致力于開發快速、有效且具臨床轉譯潜力的感染性疾病診斷策略，核心目標在于縮短診斷時程，提升急重症病患照護效率。本研究以質譜（mass spectrometry, MS）技術爲平臺，建構兩種可應用于臨床樣本的診斷工具，分別針對感染後果與感染源進行偵測。質譜在此不僅爲分析工具，更被設計爲解决特定臨床瓶頸的診斷手段。 爲同時回應宿主與病原層面的診斷延遲問題，我們開發了兩種以質譜爲基礎的工作流程，分別對應感染性疾病管理中的不同面向。第一項研究針對急診中敗血性休克的早期預測問題，采用高解析液相層析質譜技術分析入院時的血清代謝體，建立能够預測休克風險的代謝特徵。結合機器學習後所建構的模型可顯著早于傳統臨床評分系統辨識高風險病患，爭取治療先機。值得一提的是，以低解析度質譜建構的模型表現亦具可比性，顯示未來臨床應用的可行性。 第二項研究聚焦于加速細菌鑒定流程。我們建立了一套結合流動注射與質譜的快速策略，可直接分析早期液體培養樣本中的代謝指紋，無需菌落分離即可完成鑒定。本方法于培養後6至9小時內即可達成菌種層級的辨識，幷在標準菌株與臨床分離株（含抗藥株）間展現高度準確性。相較于目前臨床標準的MALDI-TOF MS，本方法顯著縮短診斷時間，亦可整合至現有自動化培養系統，具高度實用潜力。 綜合而言，這兩項研究展示了以質譜爲基礎的代謝組學可作爲解决感染性疾病診斷瓶頸的統一架構。無論是預測宿主惡化風險，或是鑒定致病菌來源，本研究所提出的快速代謝指紋技術皆展現其支持即時、準確臨床决策的潜力。除診斷性能外，對于具區辨力代謝物的生物學詮釋亦揭示了宿主與病原間的互動機制，包含細菌壓力反應與跨菌種的代謝適應行爲。這些發現凸顯了代謝物爲基礎的診斷工具在推動臨床應用與深化感染疾病生物學理解上的廣泛轉譯潜力。

Infection-related diseases remain a major threat to global health, encompassing conditions ranging from common bacterial infections to life-threatening complications. These diseases contribute to high morbidity and mortality worldwide, placing significant burden on healthcare systems and society. Timely intervention is critical, especially in acute settings, where delays in diagnosis can lead to rapid deterioration. However, conventional diagnostic workflows rely heavily on time-consuming or complex techniques, which often fail to meet the clinical need for speed and accuracy. The growing prevalence of antimicrobial resistance further underscores the urgent demand for rapid, precise, and accessible diagnostic tools. This dissertation focuses on the development of rapid, efficient, and clinically translatable diagnostic strategies for infectious diseases, with an emphasis on minimizing diagnostic delays and enhancing decision-making in acute care settings. The central aim is to leverage mass spectrometry (MS), a highly sensitive and versatile analytical technique, to build practical platforms that can detect infection consequences and identify causative pathogens directly from patient samples or culture media. Rather than focusing on MS as a general analytical method, this work frames it as a targeted clinical tool designed to solve specific diagnostic bottlenecks. To address the diagnostic delays at both host and pathogen levels, we developed two MS-based workflows tailored to different aspects of infectious disease management. The first project addresses the clinical challenge of early septic shock recognition in the emergency department. Using liquid chromatography-high-resolution mass spectrometry, we profiled serum metabolites collected at admission and identified metabolic signatures that precede the clinical onset of shock. Machine learning models trained on metabolic features successfully predicted septic shock risk several hours earlier than conventional clinical scores, offering a window for early intervention. Importantly, models built on low-resolution MS showed comparable performance, indicating feasibility for broader clinical translation. The second project focuses on accelerating bacterial identification. We developed a direct identification strategy using flow injection analysis–mass spectrometry, enabling rapid metabolite fingerprinting of bacteria from early-stage liquid culture without the need for colony isolation. This approach achieved species-level identification within 6 to 9 hours of incubation and demonstrated high accuracy across both type strains and clinical isolates, including drug-resistant strains. Compared to gold-standard MALDI-TOF MS, which requires prolonged culture and biomass accumulation, our method significantly reduces turnaround time and streamlines sample processing, while remaining compatible with automated incubation systems used in clinical microbiology laboratories. Together, these studies demonstrate how MS-based metabolomics can provide a unified framework for tackling diagnostic bottlenecks in infectious diseases. Whether predicting host deterioration or identifying causative pathogens, both approaches highlight the utility of rapid metabolic profiling in supporting timely, informed clinical decision-making. Beyond diagnostic performance, the biological interpretation of discriminative metabolites offered mechanistic insights into host-pathogen interactions, unveiling stress responses and metabolic adaptations across species. These findings illustrate the broader translational potential of metabolite-based diagnostics in advancing both patient care and our understanding of infectious disease biology.