Construction of automated high-throughput screening method for finding efficient 3-ketosteroid 1,2-dehydrogenating strains

Abstract

Dehydrogenation reaction at C1(2) positions is typical and representative of industrial production of steroid drugs. Anti-inflammatory activity can be doubled when the nucleus of the anti-inflammatory steroid hormone drug introduces double bonds at the C1(2) positions. Arthrobacter simplex is currently the most widely studied and used strain for C1(2) dehydrogenation. Therefore, breeding Arthrobacter simplex with high-efficiency dehydrogenation ability is of great significance. In order to obtain high-efficiency strains, the research proposed a new screening strategy based on image process technique: firstly, a color reaction between 2,4-dinitrophenylhydrazine (DNPH) and 9α-hydroxyandrost-4-ene-3,17-dione (9α-OH-AD) was established to characterize the dehydrogenation ability of the strain; secondly, the color data of strains mutated by atmospheric and room temperature plasma (ARTP) in the "color reaction" were automated and analyzed for dehydrogenation ability prediction using optimized support vector machine model. Result showed that the prediction accuracy reached as high as 96% in verification experiments. After a series of mutagenesis, including breaking the bottleneck of a single mutation in ARTP, the dominant strain ARLU-146 was finally obtained from 5168 strains. Its initial conversion rate was 0.8059 g/L/h, with a conversion of 94.41% at 24 h, compared to the original strain ASP which increased the transformation rate by more than 10%. By further process optimization, a high conversion (94.34% within 20 h) with high substrate (85 g/L cortisone acetate) was achieved. According to literature research, it is the highest conversion at this substrate concentration. KEY POINTS: • A high-throughput screening method was developed by using image processing and machine learning technique. • "Mutation bottleneck" of single ARTP mutagenesis was surpassed by complex mutagenesis. • A high substrate (85 g/L CA) and high transformation rate craft (94.34% within 20 h) were built.

Keywords: ARTP compound mutagenesis; Arthrobacter simplex; Automated high-throughput image processing; C1(2) dehydrogenation; Machine learning.