© 2006 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 531 Biotechnol. J. 2006, 1, 531–536 DOI 10.1002/biot.200600029 www.biotechnology-journal.com Enzymes are increasingly used in industrial processes be- cause of their ability to catalyze reactions with high ve- locity and specificity. It is estimated that the number of in- dustrially established biocatalytic processes will double every decade [1]. The main factor preventing the wide- spread adoption of biocatalysts is their lack of stability un- der processing conditions at elevated temperatures, which are often used for their beneficial effects on reac- tion rates and reactant solubility, and to reduce the risk of microbial contamination. Therefore, enzymes with high- er thermostability are necessary to create economically viable biocatalytic processes. Nongenetic methods of protein stabilization include the use of additives such as sucrose, glycerol, or trehalose to guard against denaturation, chemical modifications, and the immobilization of the enzyme on a solid support [2, 3]. Biologically based methods of obtaining highly ther- mostable enzymes include isolating enzyme variants from extremophilic organisms (by traditional cloning methods [4] or by metagenomic isolation [5]) or obtaining ther- mostable enzyme variants by protein engineering. The first approach, isolation from thermophilic organisms, is very time and cost intensive, and often yields enzyme candidates that do not catalyze the desired reaction with high turnover rates or specificity, two properties with few- er routes to optimization than thermostability. Addition- ally, thermophilic organisms can be rather hard to grow in the laboratory and may require specialized equipment in biotechnological production [6]. The latter approach, pro- tein engineering, has been extensively reviewed in the lit- erature [7], and can be further subdivided into three main methods: rational design, combinatorial design, and data- driven protein design. Rational protein design requires extensive knowledge of the enzyme to be optimized such as a crystal structure, knowledge of the mechanism of de- activation, and a general idea regarding the weak spots of the enzyme, but is far less labor intensive than screening for new enzymes as a limited number of variants are cre- ated. A barrier to adopting rational design to improve Short Communication Structure-guided consensus approach to create a more thermostable penicillin G acylase Karen M. Polizzi 1 *, Javier F. Chaparro-Riggers 1 *, Eduardo Vazquez-Figueroa 1 and Andreas S. Bommarius 1,2 1 School of Chemical & Biomolecular Engineering, Parker H. Petit Institute of Bioengineering and Bioscience, Atlanta, GA, USA 2 School of Chemistry and Biochemistry, Georgia Institute of Technology, Atlanta, GA, USA The thermostabilization of penicillin G acylase (PGA) is a difficult problem due to the large size of the protein and its complex maturation process. We developed a data-driven protein design method that requires fewer homologous sequences than the traditional consensus approach and utilizes structural information to limit the number of variants created. Approximately 50% of our 21 single-point mutants were found experimentally to be more thermostable than the wild-type PGA, two had almost threefold longer half-life at 50°C, with very little effect on activity. An analy- sis of four programs that predict the thermostability conferred by point mutations shows little agreement between the programs and with the experimental data, emphasizing that the chosen stabilizing mutations are very difficult to predict, but that our data-driven design method should prove useful. Keywords: Protein engineering · Thermostability · Consensus approach · Penicillin G acylase Correspondence: Professor Andreas S Bommarius, School of Chemical and Biomolecular Engineering, Georgia Institute of Technology, 315 Ferst Drive, Atlanta, GA 30332-0363, USA E-mail: andreas.bommarius@chbe.gatech.edu, Fax: +1-404-894-2291 Abbreviations: aa, amino acid; acc. code, accession code, PGA, penicillin G acylase *These authors contributed equally to this work. Received 10 March 2006 Revised 26 March 2006 Accepted 27 March 2006