模拟与实验整合改进酶的计算设计

As an important subject of protein design, computational enzyme design has in recent years witnessed substantial progresses, proving that de novo enzymes can in principle be designed by combining existing theoretical and computational tools. However, in its current state, computational enzyme design can achieve only minimum success rate and catalytic activity. One important cause of this is that in current approaches, backbone dynamics has been mostly neglected. In addition, the one-by-one analysis of design results with conventional experimental techniques is time and labor consuming, and it cannot produce systematic experimental feedbacks needed for method improvement. On the theoretical and computational side, we propose new design approaches, in which backbone dynamics will be considered to improve the matching between the designed active sites and the transition state catalytic core, and to reduce potential sensitivity of design results to the accuracy of energy functions. Such approaches will be applied to design artificial enzymes to catalyze the threonine synthase reaction. By choosing an essential bacterium biosynthesis reaction as the design target, the activity of designed enzymes, native enzymes, and their various sequence variants can be characterized systematically and in high throughput with auxotroph bacterium strains, so that the various theoretical hypotheses on which the designs are based can be tested, and more experimental feedbacks can be obtained to support method development. The results can also add insights into the catalytic selectivity and design principle of pyridoxal-5’-phosphate-dependent enzymes, helping future design of enzymes based on this widely-distributed co-factor to execute novel catalytic functions.

作为蛋白质设计的重要课题，酶计算设计近年来取得实质性进展，表明基于现有理论计算工具从头设计人工酶在原理上是可能的。然而，目前酶设计还只能达到最低的成功率和催化活性，其重要原因之一是设计过程中几乎未考虑和利用主链动力学。此外，采用常规实验方法逐一分析设计结果耗时耗力，不能为计算方法改进提供系统的反馈数据。从理论计算角度，我们提出新设计策略，考虑和利用主链构象动力学，以改进活性中心与过渡态催化核心的匹配、降低结果对能量模型精度的敏感度等。这些策略将被用于设计能催化苏氨酸合成酶反应的人工酶。选择此反应，我们可以用营养缺陷型细菌生长特性为指标，高效、系统地表征人工酶、天然酶及各种变体序列的催化活性，检验设计所依据的理论假设，获得更充分的实验反馈信息以改进设计方法。这些结果还将有助于深入理解PLP依赖酶的催化选择性和设计原理，为基于PLP辅基创建新型催化功能的人工酶打下基础。

本项目中，我们在酶计算设计方法方面，建立了以下能完成不同子任务的方法。（1）DEPACT，用于在给定底物结构的情况下，设计与底物相互作用的氨基酸残基类型和空间位置的cluster模型，cluster中一个或多个蛋白质基团与底物的结合可以是由水分子或金属离子介导的。cluster模型中各个局部的分子间空间构型直接来自于实验观察，避免了用力场模型优化可能带来的精度问题。DEPACT统计评分函数能对计算机自动生成的大量cluster模型进行评价，以选出最合理的模型；（2）SCUBA，用于构建具有高可设计性的主链骨架。SCUBA实质是从蛋白质结构数据学习得到的统计能量函数。我们采用一种新的近邻计数-神经网络（NC-NN）方法来学习SCUBA中不同类型的统计能量，忠实再现可设计主链构象在高维几何空间中的复杂分布，能够用于在序列待定前提下广泛搜索可设计性高的主链构象空间。（3）PACMatch用于在候选蛋白质骨架上设计活性中心，能够在保持cluster模型中各个蛋白质基团与底物相对空间关系的前提下，把cluster模型中各氨基酸残基定位到蛋白质主链上的特定位置，同时维持主链结构的合理性。在与实验结合方面，我们演示了用SCUBA和ABACUS(我们前期发展的氨基酸序列设计方法)可以设计具有自然界不存在的新颖结构的蛋白；我们从头设计了含有特定形状空腔的全新骨架蛋白，并解析了其晶体结构；建立了基于营养缺陷型对苏氨酸脱氨酶活性进行高效筛选的实验流程，并用其定向进化获得了不受代谢物别构抑制调节的苏氨酸脱氨酶变体，解析了变体的晶体结构。在Nature，Bioinformatics，Biotechnology and Bioengineering等杂志发表论文7篇；登记软件著作权1项；申请发明专利1项。