Enzymatic Synthesis of Amoxicillin Avoiding Limitations of the Mechanistic Approach for Reaction Kinetics Luciana R.B. Gonc ¸ alves, 1 Ruy Sousa, Jr., 2 Roberto Fernandez-Lafuente, 3 Jose M. Guisan, 3 Raquel L.C. Giordano, 2 Roberto C. Giordano 2 1 Departamento de Engenharia Quı ´mica, Universidade Federal do Ceara ´, Brazil 2 Departamento de Engenharia Quı ´mica, Universidade Federal de Sa ˜o Carlos, Brazil; telephone: +55 16 260 8264; fax: +55 16 260 8266; e-mail: roberto@deq.ufscar.br 3 Instituto de Cata ´ lisis y Petroleoquı ´mica, CSIC, Madrid, Spain Received 8 February 2002; accepted 14 May 2002 DOI: 10.1002/bit.10417 Abstract: A recurrent doubt that occurs to the enzyme- kinetics modeler is, When should I stop adding param- eters to my mechanistic model in order to fit a non- conventional behavior? This problem becomes more and more involving when the complexity of the reaction net- work increases. This work intends to show how the use of artificial neural networks may circumvent the need of including an overwhelming number of parameters in the rate equations obtained through the classical, mechanis- tic approach. We focus on the synthesis of amoxicillin by the reaction of p-OH-phenylglycine methyl ester and 6-aminopenicillanic acid, catalyzed by penicillin G acyl- ase immobilized on glyoxyl-agarose, at 25°C and pH 6.5. The reaction was carried on a batch reactor. Three kinetic models of this system were compared: a mechanistic, a semi-empiric, and a hybrid–neural network (NN). A semi- empiric, simplified model with a reasonable number of parameters was initially built-up. It was able to portray many typical process conditions. However, it either un- derestimated or overestimated the rate of synthesis of amoxicillin when substrates’ concentrations were low. A more complex, full-scale mechanistic model that could span all operational conditions was intractable for all practical purposes. Finally, a hybrid model, that coupled artificial neural networks (NN) to mass-balance equa- tions was established, that succeeded in representing all situations of interest. Particularly, the NN could predict with accuracy reaction rates for conditions where the semi-empiric model failed, namely, at low substrate con- centrations, a situation that would occur, for instance, at the end of a fed-batch industrial process. © 2002 Wiley Periodicals, Inc. Biotechnol Bioeng 80: 622–631, 2002. Keywords: -lactamic antibiotic; penicillin G acylase; ar- tificial neural network; hybrid model; immobilized en- zyme INTRODUCTION Nowadays, amoxicillin is produced in industry through a chemical route, which involves the protection of reactive groups, low temperatures (approximately -30°C), and or- ganic media. In spite of the high yields that this process has achieved, it has been criticized for using toxic and not easily biodegradable solvents. Accordingly, enzymatic synthesis of -lactamic antibiotics has become more interesting, due to the increasingly tight environmental regulations. Mild reaction conditions and aqueous solutions may be used in the enzymatic reactor (Ospina et al., 1996). Figure 1 shows one of the possible routes for the kinetically controlled syn- thesis of amoxicillin, using Penicillin G acylase (PGA), EC 3.5.1.11, from Escherichia coli, as catalyst. PGA from E. coli is a heterodimer (two subunits with 209 and 566 amino acids, respectively). The smaller subunit, , has a molecular mass of 20,500 Da and the larger subunit, , of 69,000 Da (Bo ¨ck et al., 1983). The role of the  subunit is to recognize the side chain of the substrate. The serine residue essential for catalytic activity is in the  subunit. Therefore, both subunits take part in the constitu- tion of the active site of PGA (Daumy et al., 1985). Bacte- rial PGAs catalyze the synthesis/hydrolysis of acyl deriva- tives of phenylacetic acid by the formation of a covalent intermediate (the acyl–enzyme complex) and accept a broad range of nucleophiles (Margolin et al., 1980; Plaskie et al., 1978). The use of derivatives of p-hydroxyphenylglycine (either esters or amides) is necessary because the direct, thermo- dynamically controlled synthesis of amoxicillin is not fa- vored. Many authors have been studying the catalytic mechanism of PGA, and its hydrolytic pathway has already been elucidated (Duggleby et al., 1995; McVey et al., 1997; Morillas et al., 1999). However, a fully mechanistic kinetic model is still not available for either the hydrolytic reaction or the synthetic reaction. Conformational changes of the 3-D structure of this enzyme in the presence of different binding molecules (Done et al., 1998) and during its hydro- lytic action (McVey et al., 2001) have been reported. How- Correspondence to: Roberto C. Giordano, DEQ/UFSCar, Via Washing- ton Luiz, km 235, C.P. 676, 13564-210, Sa ˜o Carlos, SP, Brazil. © 2002 Wiley Periodicals, Inc.