Machine-Directed Evolution of an Imine Reductase for Activity and Stereoselectivity

Biocatalysis is an effective tool
to access chiral molecules that
are otherwise hard to synthesize or purify. Time-efficient processes
are needed to develop enzymes that adequately perform the desired
chemistry. We evaluated machine-directed evolution as an enzyme engineering
strategy using a moderately stereoselective imine reductase as the
model system. We compared machine-directed evolution approaches to
deep mutational scanning (DMS) and error-prone PCR. Within one cycle,
it was found that machine-directed evolution yielded a library of
high-activity mutants with a dramatically shifted activity distribution
compared to that of traditional directed evolution. Structure-guided
analysis revealed that linear additivity might provide a simple explanation
for the effectiveness of machine-directed evolution. The most active
and selective enzyme mutant, which was identified through DMS and
error-prone PCR, was used for the gram-scale synthesis of the H4 receptor
antagonist ZPL389 with full conversion, > 99% ee (R), and a 72% yield.

生物催化（Biocatalysis）是获取手性分子（chiral molecules）的高效工具，此类分子若不借助该手段通常难以合成或纯化。开发可充分实现目标化学反应的酶，亟需具备时间效率优势的研发流程。本研究以中等立体选择性亚胺还原酶（imine reductase）为模型体系，评估了机器学习指导的定向进化（machine-directed evolution）作为酶工程策略的应用效果。我们将该方法与深度突变扫描（deep mutational scanning, DMS）、易错PCR（error-prone PCR）进行了对比。实验结果显示，仅经过一轮迭代，机器学习指导的定向进化即可获得高活性突变体文库，其活性分布相较于传统定向进化方法发生了显著偏移。结构导向分析表明，线性加和性或可为机器学习指导定向进化的有效性提供简洁解释。通过深度突变扫描与易错PCR筛选得到的活性与选择性最优的酶突变体，被用于克级合成H4受体拮抗剂（H4 receptor antagonist）ZPL389，实现了完全转化，(R)型对映体过量值>99%，产率达72%。