全基因組之結構TCR-pMHC模型剖析跨病原體同源胜肽抗原家族

Abstract

後天免疫反應(adaptive immune response)引發途徑之一是當特異性T細胞受體(T-cell receptor, TCR)與peptide-major histocompatibility complex(MHC)結合後才會被活化。儘管已有數種預測方法可識別胜肽與MHC的結合，它們通常缺乏考慮胜肽與TCR的交互作用且無法提供詳細的接合模型。隨著高通量實驗而來之抗原表位(epitope)和TCR-MHC結構數量的增加，我們提出"胜肽抗原家族(peptide antigen family)"概念，同時考慮peptide-MHC和peptide-TCR交互作用介面，可快速地進行全基因組規模且結構的剖析跨病原體胜肽以及其同源胜肽抗原(homologous peptide antigens)。每個抗原胜肽家族中的同源胜肽抗原跨多病原體且共有相似的交互作用環境。本概念著重於胜肽抗原和蛋白質體之間的特異性結合，可提供觀察於揭示抗原辨識性和免疫機制的過程。 為了瞭解TCR-peptide-MHC的交互作用模型，我們首先觀察TCR-pMHC的共結晶結構與蛋白質-蛋白質交互作用介面的相同與相異處。根據結果分析，發現peptide-MHC結合介面近似於蛋白質交互作用介面；反之，peptide-TCR結合介面則更近似於抗體-抗原結合介面。我們以抗體-抗原結合介面上最常出現的胺基酸(如酪氨酸和色氨酸)之結合區域(binding pocket)與結合能量(binding energy)來判讀的此交互作用介面的特性。由於當前TCR-peptide-MHC共結晶結構數目的稀少，我們透過收集結構抗體-抗原複合體來提升我們建立之交互作用模型的合理性。 我們建立了以模板為基礎之peptide-TCR和peptide-MHC能量函式來探討TCR-peptide-MHC接合模型。為了鑑定出查詢目標 (the query) 中潛在的胜肽抗原，我們利用兩種不同的計分矩陣，分別是以抗體-抗原計分矩陣(命名iMatrix)量度peptide-TCR交互作用介面，及以蛋白質-蛋白質交互作用計分矩陣量度peptide-MHC的交互作用介面。我們收集了抗體-抗原資料庫來建立iMatrix。iMatrix憑藉分別評估凡得瓦力或特殊作用力(氫鍵和靜電力)、以及交互作用發生位置(胺基酸鏈側鏈-側鏈或側鏈-主鏈交互作用)來預測結合能量。我們以兩個構面評估iMatrix：1) 由ASEdb收集了70個丙氨酸點突變(alanine mutagenesis)的能量改變，觀察預測能量與實驗記錄自由能改變之相關性來評估能量函式對於親和力的預測；2) 從IEDB收集2,287物種中80,057條實驗紀錄之胜肽，來驗證預測的準確度及同源胜肽抗原的合理性。對於每個模板，iMatrix搜尋病原體全基因組數據庫(來自389個病原體中的864,628條蛋白質序列，總數大於108個候選胜肽)並賦予其同源胜肽抗原。鑑定出的胜肽抗原呈現了iMatrix著重氫鍵和對保留交互作用的重視。此外，TCR-pMHC交互作用模型在peptide-TCR和peptide-MHC交互作用介面皆視覺化了接合機制(的氫鍵和空間用作用力)，更強調胜肽抗原上的關鍵位置。實驗結果顯示，本概念具有相當高的預測準確度並提供了跨病原體中可能潛在的胜肽抗原。這個概念對於未來發展胜肽疫苗及研究MHC的限制性(MHC restriction) 能提供了相當有價值的見解。 總結以上，我們發展"胜肽抗原家族"和"免疫複合體家族"來探索胜肽抗原和同源胜肽抗原。本概念是第一個同時考慮TCP-peptide-MHC兩側交互作用介面的方法，且從完整病原體基因組和實驗胜肽資料庫中找尋同源胜肽抗原。此外，本模型對於胜肽觸發免疫過程以及免疫複合體家族在T細胞活化過程中所扮演的角色提供了見解。我們相信利用此這種新穎的觀點專研胜肽抗原和蛋白質體間獨特辨識度或許能揭示免疫進化上的寶貴洞燭力。最後，本研究對於疫苗設計、移植器官的排斥反應(rejections of transplanted organs)、以及腫瘤免疫療法 (cancer immunotherapy) 具有發展潛力。

One of the most adaptive immune responses is triggered by specific T-cell receptors (TCR) binding to peptide-major histocompatibility complexes (pMHC). Despite the availability of many prediction approaches to identify peptides binding to MHC, they are often lack of peptide-TCR interactions and detailed atomic interacting models. Due to an increasing number of high-throughput binding epitopes and TCR-peptide-MHC (TCR-pMHC) structural complexes are available, we proposed "peptide antigen family" to investigate both peptide-MHC and peptide-TCR interfaces and do fast genome-wide structural inferring peptide antigens and its homologous peptide antigens from whole pathogen genomes. These homologous peptide antigens share a similar binding model across multiple species. Our idea focusing on the unique binding between peptide antigen and proteasome can provide the insights of antigen recognition and the mechanisms of immune process. For understanding the TCR-peptide-MHC binding model, we first observed the consistency and divergence between TCR-pMHC complexes and protein-protein interaction. According to the analysis, the interacting propensities of peptide-MHC interfaces are similar to those of protein-protein interfaces; conversely, the interacting propensities of peptide-TCR interfaces are similar to those of antibody-antigen interfaces. We determined properties of antibody-antigen interfaces from frequent residues (i.e. Tyrosine and Tryptophan) through dimensions of binding pockets and binding energy. Because of the limit number of TCR-pMHC co-crystal structures in the Protein Data Bank presently, we collected antibody-antigen complexes to promote the reliability of our binding model. We derived template-based peptide-TCR and peptide-MHC scoring functions for investigating TCR-peptide-MHC binding models. For identifying the potential peptide antigen of a query, we used these two different antibody-antigen (called iMatrix) and protein-protein interacting scoring matrices for peptide-TCR and peptide-MHC interfaces, respectively. We prepared non-redundant antibody-antigen dataset to generate iMatrix. iMatrix can predict binding energies by separating the van der Waals forces from special forces (hydrogen bonds and electrostatic interactions), and can discriminate sidechain-sidechain or sidechain-backbone interactions. We evaluated iMatrix through two dimensions: 1) we collected 70 alanine mutagenesis from the ASEdb and estimated the relationship between predicted energies from our scoring function and experimental free energies to validate the predicted binding affinity; 2) we prepared 80,057 experimental peptides in 2,287 species from the IEDB to validate the predictive accuracy and the reliability of homologous peptide antigens. For a TCR-peptide-MHC template, iMatrix inferred its homologous peptide antigens from complete pathogen genome databases (≥ 108 peptide candidates from 864,628 protein sequences of 389 pathogens). iMatrix keeps hydrogen-bond energies and consensus interactions from these identified peptide antigens. In addition, our TCR-pMHC models can visualize detailed binding mechanisms (e.g., hydrogen bonds and steric interactions) and highlight the key region of peptide antigen on both peptide-TCR and peptide-MHC interfaces. Experimental results demonstrate that our models can achieve high prediction accuracy and offer potential peptide antigens across pathogens. The peptide antigen family is able to provide valuable insights on the peptide vaccine and MHC restriction. In summary, we have developed “peptide antigen family” and “immune complex family” to identify peptide antigens and homologous peptide antigens. Our concept is the first method to infer homologous peptide antigens by considering two TCP-peptide-MHC interfaces from complete pathogen genome and experimental peptide databases. Additionally, our model provides the insights of the peptide trigger processes for immune and the role of immune complex family during T cell activation. We believe that this novel idea focusing the unique recognition between peptide antigen and proteasome reveals valuable insights of immunity evolution. Finally, this study has potential for vaccine design, rejections of transplanted organs, and cancer immunotherapy.