Site-directed enzymatic PEGylation of the human granulocyte colony-stimulating factor Carlo Maullu 1, *, Domenico Raimondo 2,3, *, Francesca Caboi 1 , Alejandro Giorgetti 2,4 , Mauro Sergi 1 , Maria Valentini 2 , Giancarlo Tonon 1 and Anna Tramontano 2,3,5 1 Bio-Ker S.r.l., c/o Sardegna Ricerche Scientific Park, Pula, Cagliari, Italy 2 CRS4-Bioinformatics Laboratory, c ⁄ o Sardegna Ricerche Scientific Park, Pula, Cagliari, Italy 3 Department of Biochemical Sciences ‘A. Rossi Fanelli’, University of Rome ‘La Sapienza’, Italy 4 Department of Biotechnology, University of Verona, Italy 5 Pasteur Institute–Cenci Bolognetti Foundation, University of Rome ‘La Sapienza’, Italy Introduction The conjugation of poly(ethylene glycol) (PEG) chains, termed PEGylation, is a useful methodology for drug development that is widely used for the modification of proteins, peptides, and oligonucleotides [1,2]. PEG is a noncharged, highly hydrophilic polymer that has been demonstrated to be nontoxic when its molecular mass is lower than 1000 Da, and its use for conjugation has been approved by the US Food and Drug Administration [3]. The PEGylation of pharma- ceuticals, such as liposomes and therapeutic proteins, has been shown to be an effective strategy for improvement of the biopharmaceutical properties of drugs. PEG–drug conjugates have several advantages: increased stability and water solubility, increased resis- Keywords molecular dynamics; PEGylation; protein–protein docking; site-directed mutagenesis; transglutamination Correspondence A. Tramontano, Department of Biochemical Sciences ‘A. Rossi Fanelli’, University of Rome ‘La Sapienza’, P.le Aldo Moro, 5, 00185 Rome, Italy Fax: +39 06 4440062 Tel: +39 06 49910556 E-mail: anna.tramontano@uniroma1.it Website: http://www.biocomputing.it/ *These authors contributed equally to this work (Received 12 July 2009, revised 14 September 2009, accepted 16 September 2009) doi:10.1111/j.1742-4658.2009.07387.x Poly(ethylene glycol) (PEG) is a widely used polymer employed to increase the circulating half-life of proteins in blood and to decrease their immuno- genicity and antigenicity. PEG attaches to free amines, typically at lysine residues or at the N-terminal amino acid. This lack of selectivity can pres- ent problems when a PEGylated protein therapeutic is being developed, because predictability of activity and manufacturing reproducibility are needed for regulatory approval. Enzymatic modification of proteins is one route to overcome this limitation. Bacterial transglutaminases are enzyme candidates for site-specific modification, but they also have rather broad specificity. The need arises to be able to predict a priori potential PEGyla- tion sites on the protein of interest and, especially, to be able to design mutants where unique PEGylation sites can be introduced when needed. We investigated the feasibility of a computational approach to the prob- lem, using human granulocyte colony-stimulating factor as a test case. The selected protein is therapeutically relevant and represents a challenging problem, as it contains 17 potential PEGylation sites. Our results show that a combination of computational methods allows the identification of the specific glutamines that are substrates for enzymatic PEGylation by a microbial transglutaminase, and that it is possible to rationally modify the protein and introduce PEG moieties at desired sites, thus allowing the selection of regions that are unlikely to interfere with the biological activity of a therapeutic protein. Abbreviations G-CSF, granulocyte colony-stimulating factor; MD, molecular dynamics; mPEG, monomethoxy-poly(ethylene glycol); MTGase, microbial transglutaminase; PEG, poly(ethylene glycol); RMSF, root mean squared fluctuation. FEBS Journal 276 (2009) 6741–6750 ª 2009 The Authors Journal compilation ª 2009 FEBS 6741 tance to proteolytic inactivation, low toxicity, improved pharmacokinetic profiles, and reduced renal clearance and immunogenicity [4,5]. Thanks to these favorable properties, PEGylation plays an important role in drug delivery, enhancing the potential of peptides and proteins as therapeutic agents. PEGylation was first described in the 1970s by Davies and Abuchowsky, and was reported in two key papers on albumin and catalase modification [6,7]. Since then, the procedure of PEGylation has been expanded, and a wide range of chemical and enzymatic methods for conjugation have been developed. The most widely used modification method for pro- tein PEGylation involves the covalent conjugation of activated monomethoxy-PEG (mPEG) at the level of the e-amino group of lysine residues by using acylating mPEG derivatives. This strategy has limitations, because of the potential multiple sites of conjugation and the consequent heterogeneity of the PEGylated proteins. The purification of these mixtures is usually difficult, and this reduces the predictability of their activity and manufacturing reproducibility needed for regulatory approval. The requirements for the approval of new conju- gates are very stringent, and obtaining a single isomer, whenever possible, or at least a well-character- ized mixture of mono-PEGylated isomers is compul- sory. Examples are the two a-interferon conjugates, Pegasys [8] and PEG-Intron [9], for which almost all the binding sites in the primary sequence were charac- terized. In order to obtain site-specific PEGylation, other chemical approaches were developed, such as the selective PEGylation at the level of the thiol group of cysteines or at the N-terminal amino group of a poly- peptide chain [10,11]. More recently, a very promising enzymatic method has been proposed that makes use of the transglutaminase enzyme for the covalent link- age of PEG moieties at the c-carboxamide groups of glutamines of proteins [12,13]. For this purpose, an mPEG derivative bearing a primary amino group is used (mPEG-NH 2 ); this becomes covalently linked to the protein at glutamines through a transglutamination reaction catalyzed by the enzyme according to the following scheme: protein-CONH 2 þ H 2 N-R ! protein-CONH-R þ NH 3 where CONH 2 is a carboxamide group of glutamine side chains, and R is an mPEG molecule. In this work, we investigated the molecular basis of enzymatic conjugation of PEG molecules to glutamines by a microbial transglutaminase enzyme (MTGase) deri- ved from a variant of Streptoverticillium mobaraense. The granulocyte colony-stimulating factor (G-CSF) was used as substrate. It is a challenging case, because it con- tains 17 potential PEGylation sites and, at the same time, an important target, as it acts in hematopoiesis by controlling the production, differentiation and function of granulocytes. It is pharmaceutically available under the names Neupogen or Granulokine (produced by Escherichia coli cells; Amgen, Thousand Oaks, CA, USA ⁄ Roche, Nutley, NJ, USA) and Granocyte (pro- duced in mammalian cells; Rhone-Poulenc, Rorer, Cologne, France), and is used to treat neutropenia, a disorder characterized by an extremely low number of neutrophils in blood. Although widely used, G-CSF is rapidly removed from the body by a combination of renal and active neutrophil clearance processes. As a result, for most practical purposes, repeated injections or continuous infusion of G-CSF are necessary to gener- ate sufficiently elevated neutrophil and mobilized pro- genitor ⁄ stem cell levels in the peripheral blood [14]. For this reason, the PEGylation of G-CSF, and ⁄ or design of new variants with longer circulation times, together with a thorough characterization of the mechanism underly- ing the process of PEGylation, are essential steps for the design of new and more effective therapeutic proteins. We report here a computational approach aimed at identifying the glutamines modified by the enzyme. We used three-dimensional structural analysis, molecular dynamics (MD) simulations, and protein–protein dock- ing calculations. All of these approaches allowed us to identify a single potential PEGylation site in the G-CSF molecule, a prediction that was subsequently validated by site-directed mutagenesis experiments, PEGylation experiments, and analytical analysis of PEGylated G-CSF by peptide mapping and N-terminal sequence analysis. All of the data obtained from these experi- ments confirmed our computational results on the iden- tification of a single G-CSF residue that is the target of PEGylation modification by MTGase. Moreover, the characterization of the dynamic properties of the G-CSF region involved in the transglutamination pro- cess was also demonstrated to be useful for the design of mutants with different PEGylation properties. Results G-CSF sequence and structure analysis The G-CSF primary structure (UniProtKB ⁄ Swiss-Prot accession code: P09919) includes 17 glutamines that, in PEGylation of the G-CSF molecule by MTGase enzyme C. Maullu et al. 6742 FEBS Journal 276 (2009) 6741–6750 ª 2009 The Authors Journal compilation ª 2009 FEBS principle, are candidates for transglutamination by MTGase (Fig. 1). Our aim was to identify the G-CSF reactive gluta- mine(s) involved in the transglutamination process, under the assumptions that they are exposed to the solvent, highly flexible, and in a region that can undergo favorable interactions with the enzyme active site. As a first step, we evaluated the accessible surface area of each of these 17 glutamines. Glutamines were considered to be buried when < 25% of their total area was exposed to solvent: there are eight glutamines satisfying this condition (Table 1). The substrate specificity of MTGase is rather broad [15,16]. In general, broad specificity requires flexibility of the substrate, which is expected to be able to adapt to the enzyme conformation. It follows that the site of PEGylation should not be part of regular secondary structure elements [13,17], and this latter requirement reduced our candidate list to five glutamines: Gln11, Gln67, Gln70, Gln131, and Gln134. Incidentally, Gln131 and Gln134 are very close to Thr133, which is the glycosylation site of natural G-CSF, confirming that they are accessible and potentially more reactive. Note that the nonglycosylated recombinant protein expressed in E. coli is active, and therefore, even if transglutamination impaired glycosylation at the neighboring site, the protein function should not be affected. Both the glycosylated Thr133 and the five candidate glutamines, Gln11, Gln67, Gln70, Gln131, and Gln134, are very well conserved among different species (Fig. 1B). In humans, there are four splicing variants of G-CSF annotated in ensembl (G-CSF ensembl acces- sion code: ENSG00000108342), although only two of them are annotated in the UniprotKB database. They differ in the N-terminal region of the protein, which is far away both in sequence and structure from the glycosylation and putative transglutamination sites, which are conserved in all of them. MD simulations Carefully performed MD simulations can highlight flexible regions of proteins. We performed two differ- ent 10 ns MD simulations on the wild-type G-CSF monomeric subunit and on a G-CSF structure in which we made two single amino acid substitutions (P132Q and Q134N). We selected P132Q and Q134N mutations in order to build a molecule with different transglutamination properties. Removal of Pro132 could lead to increased local flexibility of Gln131, making it an appropriate substrate for transglutamina- tion, whereas the Q134N mutation would remove the putative transglutamination site of the wild-type mole- cule. The MD simulation for the double mutant P132Q ⁄ Q134N (defined as Mut4) was run under the same conditions used for the wild-type protein. A B Fig. 1. (A) G-CSF protein sequence as reported in the Protein Data Bank entry 2D9Q SEQRES records. Secondary structure elements are marked above the sequence. The positions of the 17 gluta- mines present in the wild-type G-CSF are in blue boxes. Glutamines showing high structural flexibility in the MD experiments are indi- cated by stars. (B) A sequence logo representation [35] of the mul- tiple sequence alignment of human G-CSF and its orthologous proteins. It consists of stacks of symbols, one for each position in the protein sequence; the overall height of the stack indicates the sequence conservation at that position, and the height of symbols within the stack indicates the relative frequency of each amino acid at that position. The blue arrows indicate the potential glycosylation and transglutamination sites. V8 protease preferential cleavage sites are marked with red arrows. C. Maullu et al. PEGylation of the G-CSF molecule by MTGase enzyme FEBS Journal 276 (2009) 6741–6750 ª 2009 The Authors Journal compilation ª 2009 FEBS 6743 The analysis of the trajectories of the equilibrated MD simulation showed that the root mean squared fluctuations (RMSFs) of the protein have their highest peaks around the positions corresponding to Gln134 and Gln131, belonging to a highly mobile region of the protein, in good agreement with the b-factor values reported in the Protein Data Bank entry (Fig. 2). We also analyzed the w ⁄ u angle variation during the MD simulation. The Ramachandran plots reported in Fig. 3 show that Gln134 is able to explore a very broad combination of dihedral angles (i.e. all of the allowed conformations of the classic Ramachandran plot), which is not the case for Gln131. The differences in local flexibility observed for Gln131 and Gln134 could be explained by the proxim- ity of Pro132 to Gln131. The rigidity of the proline might reduce the potential flexibility of the neighboring side chain. In conclusion, our analysis suggested that Gln134 is the most likely substrate for PEGylation. In the mutant, both Gln131 and Gln132 were able to explore a broader range of the Ramachandran regions, almost as broad as that of Gln134 in the wild- type protein (Fig. 3). Overall, sequence, structure and dynamic analysis of G-CSF molecule indicate that Gln134 is the most likely transglutamination site, and that the P132Q ⁄ Q134N double mutant should behave differ- Table 1. Solvent-accessible area and secondary structure of the 17 glutamines present in the wild-type G-CSF. The first column reports the position of the glutamines in the wild-type protein sequence. The second column indicates the percentage of residue exposure (we consider a glutamine residue to be exposed when the reported value is grater than 25%). The third column reports the secondary structure context of each of the glutamines. The five candidate glu- tamines that are exposed and outside regular secondary structure elements are in bold type. Gln Solvent-accessible area, G-CSF Secondary structure 11 44.18 At the N-terminus of a1 20 15.66 In a1 25 22.51 In a1 32 22.34 In a1 67 25.53 In the loop closed by the Cis64–Cys74 disulfide bridge 70 80.74 In the loop closed by the Cys64–Cys74 disulfide bridge 77 19.32 In a3 86 12.81 In a3 90 48.65 At the C-terminus of a3 107 14.34 In a4 119 59.5 In a4 120 14.62 In a4 131 69.65 In the loop between helices a4 and a5 134 32.57 In the loop between helices a4 and a5 145 28.71 In a5 158 19.5 In a5 173 72.04 C-terminus 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 Residue 0 0.1 0.2 0.3 0.4 0 .5 RMSF (nm) Gln131 Gln134 Fig. 2. RMSFs of the G-CSF Ca atoms during the entire MD simu- lation. The points corresponding to Gln131 and Gln134 RMSFs are indicated by a red and a green circle, respectively. –180 –120 –60 0 60 120 Phi –180 –120 –60 0 60 120 180 Psi Gln131 –180 –120 –60 0 60 120 180 Phi –180 –120 –60 0 60 120 180 Psi Gln132 –180 –120 –60 0 60 120 Phi –180 –120 –60 0 60 120 180 Psi Gln131AB CD –180 –120 –60 0 60 120 180 Phi –180 –120 –60 0 60 120 180 Psi Gln134 Fig. 3. Ramachandran plots showing the u ⁄ w angle variation along the MD simulations of selected glutamines. (A–D) Plots corre- sponding to Gln131 and Gln134 in the MD simulation of the wild type and of Gln131 and Gln132 in the simulation of Mut4, respec- tively. PEGylation of the G-CSF molecule by MTGase enzyme C. Maullu et al. 6744 FEBS Journal 276 (2009) 6741–6750 ª 2009 The Authors Journal compilation ª 2009 FEBS ently and be transglutaminated on Gln131 and ⁄ or Gln132. Molecular docking analysis of MTGase ⁄ G-CSF One puzzling observation derived from the structural analysis of MTGase (Protein Data Bank accession code: 1IU4) is that the active site of the enzyme is located in a shallow crevice surrounded by two loops, and this is difficult to reconcile with the broad specific- ity of the enzyme. This is confirmed by protein–protein docking calculation. A local version of the rosettadock program [18] was used to predict protein–protein interaction between G-CSF and MTGase (G-CSF–closed-MTGase and G-CSF–open-MTGase). None of the docking solutions that we obtained involved interactions of G-CSF with the enzyme active site, not even when we included distance restrains of 7 A ˚ between the amino acids hypothesized to be involved in the interaction, Gln131 and Gln134 from G-CSF, and Cys64 from MTGase. The active site is surrounded by two loops that are likely to be flexible, and therefore we hypothesized that they can also assume a conformation different from that observed in the X-ray structure. The G-CSF struc- ture was modified by exciting the low-energy modes of the system. In particular, by deforming the structure along the lowest-energy mode, it was possible to gener- ate an ‘open’ conformation of the enzyme (Fig. 4). Next, two different systems were tested by the rosettadock protein–protein docking program, and 100 000 decoys were produced for each of them. The analyzed systems were Gln134-restrained docking of both closed-MTGase–G-CSF and open-MTGase– G-CSF; Gln134-restrained docking means that we per- formed the docking protocol with the inclusion of distance restraints of 7 A ˚ between the G-CSF Gln134 and the active site residue Cys64 of MTGase. Using the open conformation of the enzyme, we were able to retrieve eight configurations fulfilling the distance constraint. Seven of the poses differ by < 1.6 A ˚ rmsd from each other (Fig. 5). Experimental validation To validate our computational predictions about PEGylation site, we analyzed the properties of the wild-type G-CSF and of the following mutants: Q131N, Q134N, Q173N and P132Q ⁄ Q134N (Mut1– Fig. 4. Optimal three-dimensional superposition of the ‘open’ and ‘closed’ MTGase configurations, represented in pale green and blue, respectively. The rmsd values between these two conforma- tions are 1.45 A ˚ and 1.42 A ˚ for all atoms and Ca atoms, respec- tively. The G-CSF interaction site is expected to be near the active site residue Cys64, indicated in ball-and-stick representation. Fig. 5. Model of the interaction between G-CSF (orange) and the ‘open’ conformation of MTGase (blue). The MTGase Asp3, Cys64 (active site residue) and G-CSF Gln134 and Thr133 are shown in ball-and-stick representation. The hydrogen bond between the Thr133 side chain and the Asp3 main chain is shown as a green line. C. Maullu et al. PEGylation of the G-CSF molecule by MTGase enzyme FEBS Journal 276 (2009) 6741–6750 ª 2009 The Authors Journal compilation ª 2009 FEBS 6745 Mut4 in Table 2). Gln173 was chosen because it is located in the very flexible C-terminal region of the protein, very close to an a-helix. The PEGylation reaction results obtained for wild- type G-CSF and for the four mutants are summarized in Table 2. They showed that the Q134N mutant was not PEGylated, whereas PEGylation was only slightly reduced (85%) in the Q131N and Q173N mutants, confirming that Gln134 is the only glutamine, among the 17 present in the molecule, available for the trans- glutamination reaction. These data convincingly validate our computational predictions. Incidentally, it is very relevant that enzymatic PEGylation of G-CSF gives rise to a site-specific monoconjugate derivative, which is interesting mole- cule for therapeutic approaches. The double mutant Mut4 retains the ability to be PEGylated to a similar extent as the wild type (Table 2). As this mutant lacks the Gln134 PEGylation site, it is likely that the P132Q mutation changes the properties of Gln131 and ⁄ or Gln132, increasing its flexibility and making it a better substrate for the enzyme. However, Mut4 contains other glutamines, and the possibility cannot be excluded that one of the others becomes the PEGylation site. To verify which of the glutamines of Mut4 are transglutaminated, the PEGylation sites of native and mutated G-CSF were analyzed by enzymatic digestion with Staphylococ- cus aureus V8 protease, which is specific for cleavage at the C-terminus of glutamic acid and aspartic acid (Fig. 1B). The RP-HPLC profiles of the two enzymatic diges- tion mixtures differed mainly by a few peaks that, in the chromatogram of the PEGylated digestion mixture, were eluted with retention times correspond- ing to more hydrophilic molecules, indicating that these peptides are bound to the PEG chain (data not shown). The peptides obtained by enzymatic digestion were separated by SDS ⁄ PAGE. Figure 6 shows the two SDS ⁄ PAGE gels stained with barium iodine (lane A), which highlights the PEG moiety, and with Coomassie Blue (lane B), which reveals protein and peptides. The spots corresponding to PEG-bearing peptides were then electroblotted onto a poly(vinylidene difluoride) membrane, and the fragments were subjected to N-ter- minal sequencing. All fragments started with the sequence LGMAP- ALQPTQGAMPA and lacked the signal correspond- ing to Gln134, which is diagnostic of its derivatization. This result confirmed that Gln134 is the single PEGylation site of G-CSF, in agreement with the results obtained by the computational calculations. PEGylated Mut4, subjected to the same analytical characterization, did not lack any residue in the N-ter- minal sequencing of its mono-PEGylated fragments. This result can be explained by the presence of two different mono-PEGylated isomers, corresponding to Gln131 and Gln132, in agreement with the calculations performed on the mutant. We are led to conclude that PEGylation of one of the two glutamines impairs the PEGylation of the neighboring one. The computa- tional prediction of the relative abundance of the two PEGylated species would require knowledge of the structure of the mutant, as it is well known that even the most advanced docking technologies cannot cope with cases where the backbone of one of the molecules changes upon binding [19]. Table 2. PEGylation reaction results for G-CSF and its mutants. Name Mutant PEGylation yield (%) G-CSF Wild type 100 Mut1 Q173N 85 Mut2 Q131N 85 Mut3 Q134N 5–6 Mut4 P132Q ⁄ Q134N 80 Fig. 6. SDS PAGE analyses of PEGylated G-CSF and its V8 protease digested mixture, stained with barium iodide (A) and Blue Coomassie (B). PEGylation of the G-CSF molecule by MTGase enzyme C. Maullu et al. 6746 FEBS Journal 276 (2009) 6741–6750 ª 2009 The Authors Journal compilation ª 2009 FEBS Discussion The computational and experimental analysis of the PEGylation properties of the G-CSF residues allows us to confidently conclude the following with regard to PEGylation by MTGase: (a) the substrate reactive site should be exposed to solvent and present in a ‘locally’ flexible region; (b) neighboring residues are unlikely to be PEGylated on the same molecule, possibly because of steric hindrance; and (c) the presence of a proline close to the putative site of PEGylation is a limiting factor that hampers the reaction. In our view, it is relevant that computational predic- tions, based on publicly available methods, are nowa- days sufficiently reliable to allow the identification of targets of enzymatic modifications and the redesign of proteins with the desired properties, as substantiated by the results of our mutant design experiments, where we could redirect the enzyme specificity to different sites. Our study was performed on one protein, selected because it represents a challenging case, with 17 puta- tive transglutamination sites, and because of its high therapeutic interest. We believe that our results are likely to be general, because they are based on reason- able assumptions (flexibility, exposure to solvent, and ability to interact with the enzyme). Further experi- ments on different systems are in progress to substanti- ate this hypothesis. Finally, the mono-PEGylated G-CSF molecule described here is of therapeutic interest, as it is fully characterized, homogeneously modified, easy to pro- duce, and expected to have a longer circulating half- life than the wild-type protein. Pharmacokinetic and pharmacodynamic studies of the recombinant G-CSF– Q134-PEG following subcutaneous administration in normal and neutropenic rats are in progress. Prelimin- ary results show that our molecule has the same phar- macological effect as the nonpegylated G-CSF and better pharmacokinetic parameters. Experimental procedures Materials MTGase from S. mobaraense was purchased from Ajino- moto (Activa WM, Europe Sales GmbH, Hamburg, Germany). Recombinant G-CSF and its mutants were pro- duced by Bio-Ker (c ⁄ o Sardegna Ricerche, Pula, Italy) by a fusion protein technology [20] (US7,410,775 B2, 12 August 12, 2008, Method for making recombinant peptides or proteins using soluble endoptroteases). Endoproteinase Glu-C from St. aureus (V8 protease) was purchased from Sigma Aldrich (St Louis, MO, USA). Methoxy-PEG-NH 2 (M r 20 000) was purchased from SunBio (San Francisco, CA, USA). Restriction and DNA- modifying enzymes were purchased from New England Biolabs (Beverly, MA, USA) and used according to the manufacturer’s instructions. PfuTurbo Hot Start polymerase was purchased from Stratagene (La Jolla, CA, USA). Sequence conservation analysis The alignment shown in Fig. 1 includes all the species where a protein orthologous to G-CSF was found, using the ensembl search for orthology [21]: Bos taurus, Canis familiaris, Cavia porcellus, Dasypus novemcinctus, Dipodomys ordii, Echinops telfairi, Equus caballus, Felis catus, Gorilla gorilla, Loxodonta africana, Macaca mulatta, Macropus eugenii, Microcebus murinus, Monodelphis domestica, Mus musculus, Myotis lucifugus, Ochotona princeps, Ornithorhynchus anatinus, Oryctolagus cuniculus, Otolemur garnettii, Pan troglodytes, Pipistrellus pygmaeus, Procavia capensis, Pteropus vampyrus, Rattus norvegicus, Spermophilus tridecemlineatus, Taeniopygia guttata, Tupaia belangeri, Tursiops truncatus, and Xenopus tropicalis. Neither a psi-blast nor a psi-search run against the NR and UniprotKB databases could identify ortholog-contain- ing species other than the ones listed above (data not shown). Solvent accessibilities and MD simulations The G-CSF coordinates were retrieved from the Protein Data Bank (accession code: 2D9Q, chain A). Modeling of the double mutant of G-CSF was performed with the pro- gram scrwl [22]. Amino acid solvent-accessible surface area was calculated using the molmol program [23] and the Scit web server [24]. MD simulations were performed using the gromacs package of programs (version 3.2) [25] and the gromos 96 force field. All of the structures were placed in a cubic peri- odic box (92 · 92 · 92 A ˚ ) of 24 876 SPC ⁄ E water mole- cules [26]. Four sodium ions were added to ensure electroneutrality of the systems. All of the systems studied were energy relaxed with 1000 steps of steepest descent energy minimization to remove possible unfavorable contacts from the initial structures. The protein–solvent systems were then subjected to 0.5 ns of position-restrained dynamics to allow water mole- cules to soak the protein, followed by 1 ns of equilibration at constant temperature (300 K) and pressure (1 atm), using the Nose–Hoover thermostat and barostat (coupling con- stants were 0.5 ps) [27]. The lincs algorithm [28] was used to constrain all hydrogen bonds. A cut-off of 1.4 nm for Lennard–Jones interactions was used, and the particle mesh Ewald method [29] was employed to calculate longer-range electrostatic contributions on a grid with 0.12 nm spacing C. Maullu et al. PEGylation of the G-CSF molecule by MTGase enzyme FEBS Journal 276 (2009) 6741–6750 ª 2009 The Authors Journal compilation ª 2009 FEBS 6747 and a cut-off of 0.9 nm. The time step used was 2 fs. Root mean square displacement fluctuations were calculated with the program g_rmsf included in the gromacs analysis tools, using the equilibrated trajectories. Normal mode analysis – generation of an ‘open’ MTGase conformation The b-Gaussian network model [30], a coarse-grained model, provides a reliable and not very computationally time-consuming description (with respect to full atom MD simulations) of concerted large-scale rearrangements in pro- teins. In this approach, the concerted motions are calculated within the quasiharmonic approximation of the free energy F around a protein native state (assumed to coincide with the crystallographic structure or with a minimized model structure). Thus, a displacement from the native state dR ={dr 1 , dr 2 , , dr n }(r i being the displacement of Ca atom i) is associated with a free energy change DF =(1⁄ 2)dR  FdR, where F is an interaction matrix derived from the knowledge of contacting Ca and Cb atoms in the native state, and the  superscript indicates the trans- pose matrix. The large-scale motions of the system corre- spond to the eigenvectors of F with the smallest nonzero eigenvalues. The maxsprout algorithm [31] and scrwl software [22] were used to reconstruct the backbone coordinates from the Ca atom positions and the side chains, respectively, after normal mode analysis. G-CSF–MTGase interaction A local version of the rosettadock program [18] running on a 48 node Opteron cluster was used to perform the protein–protein docking experiments. The rosettadock program, also proven to be useful for protein models, uses real-space Monte Carlo minimization on both rigid body and side chain degrees of freedom to identify the lowest- free-energy docked arrangement of two interacting proteins. The ranking of the solutions is based on a free energy func- tion dominated by a Lennard–Jones potential, an orienta- tion-dependent hydrogen bond potential, [32] and an implicit solvation model [33]. Site-directed mutagenesis Four mutants of G-CSF were constructed with the Quik- Change site-directed mutagenesis kit (Stratagene). Mut1– Mut4 correspond to mutants Q173N, Q131N Q134N and P132Q ⁄ Q134N, respectively. Briefly, PCR amplification was performed by PfuTurbo Hot Start polymerase (Stratagene) under standard condi- tions, using approximately 10 ng of a plasmid containing the wild-type G-CSF as a template and, in the case of the Q134N mutant, a pair of complementary primers (forward, 5¢-GCCGGCATGGCACCGTTGGTGGGCTGCAGGG-3¢; and reverse, 5¢-CCCTGCAGCCCACCAACGGTGCCA TGCCGGC-3¢). The PCR product was then digested with 10 U of DpnI, and this was followed by transformation into electrocompetent JM109 E. coli cells. The presence of the desired Q134N mutation was confirmed by direct DNA sequence analysis. The Q173N, Q131N and P132Q ⁄ Q134N mutants were obtained with the same strategy, using suit- able primers. All DNA manipulations, including restriction digestion, ligation, and agarose gel electrophoresis, were performed as described by Sambrook et al. [34]. The PCR amplifications were performed using a PCR thermal cycler (Gene Amp PCR System 2700; Applied Biosystems, Foster City, CA, USA), a high-fidelity PCR system [600320-51, PfuTurbo Hot Start (Stratagene) and 600400-51 Easy A Hi Fi (Strata- gene)], and oligonucleotides synthesized by M-Medical (Milan, Italy). Plasmid extractions, gel extractions and PCR purifications were performed using Qiagen kits. E. coli competent cells {JM109 strain (F¢[traD36, proA + B + , lacI q , D(lacZ)M15], D(lac, proAB)}, glnV44, e14 – , gyrA96, rec A1, rel A1, end A1, thi, hsdR17) from New England Biolabs were transformed using the Bio-Rad E. coli pulser transformation apparatus. The recombinant JM109 cells were cultured using a fed-batch fermentation process with a 10 L bioreactor (Biostat C, B. Braun), and the G-CSF mutant fusion proteins, expressed in the form of insoluble inclusion bodies, were recovered from the cells by high-pressure homogenization, solubilized using a chaotropic agent, and renatured by dilution in urea buffer. Biologically active forms of G-CSF mutants, more than 98% pure, were obtained by enzymatic cleavage of the fusion protein followed by a two-step column chromatogra- phy purification process and a final gel filtration step. PEGylation of G-CSF and its mutants via MTGase Nonglycosylated G-CSF or one of its mutants was dis- solved in a 10 mm (pH 7.4) potassium dihydrogen phos- phate buffer at a concentration of 1 mg proteinÆmL )1 , corresponding to a concentration of about 53 lm. Mono- methoxy-PEG-NH 2 (20 kDa) (Sunbio) was then added to the protein solution to achieve a 10 : 1 PEG ⁄ G-CSF molar ratio. MTGase was then added to the reaction mixture to 0.024 UÆmL )1 of final solution. The reaction took place overnight under mild stirring at room temperature. At the end of the reaction, aliquots of the reaction mixture were analyzed on an RP-HPLC column to determine the yield of the reaction. PEGylated G-CSF and mutant analysis The characterization of the PEGylation sites of the wild- type and mutant G-CSF was performed by combining PEGylation of the G-CSF molecule by MTGase enzyme C. Maullu et al. 6748 FEBS Journal 276 (2009) 6741–6750 ª 2009 The Authors Journal compilation ª 2009 FEBS different analytical methods. The PEGylated proteins were first subjected to enzymatic digestion by V8 protease, and the PEGylated fragments, generated by specific and nonspe- cific enzymatic cuts, were separated from the peptide mixture by SDS ⁄ PAGE. The spots corresponding to PEG-bearing peptides were blotted onto a poly(vinylidene difluoride), membrane and their N-terminal sequences were determined. Acknowledgements This publication was based on work partially sup- ported by the MIUR grant ITALBIONET and by FIRB project PROTEOMICA RBRN07BMCT. We thank F. Ferre ` for insightful discussions. References 1 Jain A & Jain SK (2008) PEGylation: an approach for drug delivery. A review. Crit Rev Ther Drug Carrier Syst 25, 403–447. 2 Veronese FM & Mero A (2008) The impact of PEGyla- tion on biological therapies. BioDrugs 22, 315–329. 3 Harris JM & Chess RB (2003) Effect of pegylation on pharmaceuticals. Nat Rev Drug Discov 2, 214–221. 4 Roberts MJ, Bentley MD & Harris JM (2002) Chemis- try for peptide and protein PEGylation. Adv Drug Deliv Rev 54, 459–476. 5 Wattendorf U & Merkle HP (2008) PEGylation as a tool for the biomedical engineering of surface modified microparticles. J Pharm Sci 97, 4655–4669. 6 Abuchowski A, van EsT, Palczuk NC & Davis FF (1977) Alteration of immunological properties of bovine serum albumin by covalent attachment of polyethylene glycol. J Biol Chem 252, 3578–3581. 7 Abuchowski A, McCoy JR, Palczuk NC, van Es T & Davis FF (1977) Effect of covalent attachment of polyethylene glycol on immunogenicity and circulating life of bovine liver catalase. J Biol Chem 252, 3582– 3586. 8 Bailon P, Palleroni A, Schaffer CA, Spence CL, Fung WJ, Porter JE, Ehrlich GK, Pan W, Xu ZX, Modi MW et al. (2001) Rational design of a potent, long-lasting form of interferon: a 40 kDa branched polyethylene gly- col-conjugated interferon alpha-2a for the treatment of hepatitis C. Bioconjug Chem 12, 195–202. 9 Wang YS, Youngster S, Grace M, Bausch J, Bordens R & Wyss DF (2002) Structural and biological character- ization of pegylated recombinant interferon alpha-2b and its therapeutic implications. Adv Drug Deliv Rev 54, 547–570. 10 Zalipsky S (1995) Functionalized poly(ethylene glycol) for preparation of biologically relevant conjugates. Bioconjug Chem 6, 150–165. 11 Kinstler O, Molineux G, Treuheit M, Ladd D & Gegg C (2002) Mono-N-terminal poly(ethylene glycol)–protein conjugates. Adv Drug Deliv Rev 54, 477–485. 12 Sato H (2002) Enzymatic procedure for site-specific pegylation of proteins. Adv Drug Deliv Rev 54, 487–504. 13 Fontana A, Spolaore B, Mero A & Veronese FM (2008) Site-specific modification and PEGylation of pharmaceutical proteins mediated by transglutaminase. Adv Drug Deliv Rev 60, 13–28. 14 Lord BI, Woolford LB & Molineux G (2001) Kinetics of neutrophil production in normal and neutropenic animals during the response to filgrastim (r-metHu G-CSF) or filgrastim SD ⁄ 01 (PEG-r-metHu G-CSF). Clin Cancer Res 7, 2085–2090. 15 Taguchi S, Nishihama KI, Igi K, Ito K, Taira H, Motoki M & Momose H (2000) Substrate specificity analysis of microbial transglutaminase using proteina- ceous protease inhibitors as natural model substrates. J Biochem 128, 415–425. 16 Ohtsuka T, Sawa A, Kawabata R, Nio N & Motoki M (2000) Substrate specificities of microbial transglutamin- ase for primary amines. J Agric Food Chem 48, 6230– 6233. 17 Coussons PJ, Price NC, Kelly SM, Smith B & Sawyer L (1992) Factors that govern the specificity of transglu- taminase-catalysed modification of proteins and pep- tides. Biochem J 282 (Pt 3), 929–930. 18 Gray JJ, Moughon S, Wang C, Schueler-Furman O, Kuhlman B, Rohl CA & Baker D (2003) Protein– protein docking with simultaneous optimization of rigid-body displacement and side-chain conformations. J Mol Biol 331, 281–299. 19 Wang C, Schueler-Furman O, Andre I, London N, Fleishman SJ, Bradley P, Qian B & Baker D (2007) RosettaDock in CAPRI rounds 6–12. Proteins 69, 758–763. 20 Pozzuolo S, Breme U, Salis B, Taylor G, Tonon G & Orsini G (2008) Efficient bacterial expression of fusion proteins and their selective processing by a recombinant Kex-1 protease. Protein Expr Purif 59, 334–341. 21 Vilella AJ, Severin J, Ureta-Vidal A, Heng L, Durbin R & Birney E (2009) EnsemblCompara GeneTrees: com- plete, duplication-aware phylogenetic trees in verte- brates. Genome Res 19, 327–335. 22 Canutescu AA, Shelenkov AA & Dunbrack RL Jr (2003) A graph-theory algorithm for rapid protein side- chain prediction. Protein Sci 12, 2001–2014. 23 Koradi R, Billeter M & Wuthrich K (1996) MOLMOL: a program for display and analysis of macromolecular structures. J Mol Graph 14, 51–55. 24 Gautier R, Camproux AC & Tuffery P (2004) SCit: web tools for protein side chain conformation analysis. Nucleic Acids Res 32, W508–W511. C. Maullu et al. PEGylation of the G-CSF molecule by MTGase enzyme FEBS Journal 276 (2009) 6741–6750 ª 2009 The Authors Journal compilation ª 2009 FEBS 6749 25 Van Der Spoel D, Lindahl E, Hess B, Groenhof G, Mark AE & Berendsen HJ (2005) GROMACS: fast, flexible, and free. J Comput Chem 26, 1701–1718. 26 Berendsen HJC, Grigera JR & Straatsma TP (1987) The missing term in effective pair potentials. J Phys Chem 91, 6269–6271. 27 Evans DJ & Holian BL (1985) The Nose–Hoover ther- mostat. J Chem Phys 83, 4069–4074. 28 Hess B, Bekker H, Berendsen HJC & Fraaije JGEM (1997) LINCS: a linear constraint solver for molecular simulations. J Comput Chem 18, 1463–1472. 29 Essmann U, Perera L, Berkowitz M, Darden T, Lee H & Pedersen L (1995) A smooth particle mesh Ewald method. J Chem Phys 103, 8577–8593. 30 Micheletti C, Carloni P & Maritan A (2004) Accurate and efficient description of protein vibrational dynam- ics: comparing molecular dynamics and Gaussian models. Proteins 55, 635–645. 31 Holm L & Sander C (1991) Database algorithm for generating protein backbone and side-chain co-ordi- nates from a C alpha trace application to model build- ing and detection of co-ordinate errors. J Mol Biol 218, 183–194. 32 Kortemme T, Morozov AV & Baker D (2003) An orientation-dependent hydrogen bonding potential improves prediction of specificity and structure for proteins and protein–protein complexes. J Mol Biol 326, 1239–1259. 33 Lazaridis T & Karplus M (1999) Effective energy func- tion for proteins in solution. Proteins 35, 133–152. 34 Sambrook J, Fritsch EF & Maniatis T (2000) Molecular Cloning: a Laboratory Manual, 3rd edn. Cold Spring Harbor Laboratory Press, Cold Spring Harbor, NY. 35 Crooks GE, Hon G, Chandonia JM & Brenner SE (2004) WebLogo: a sequence logo generator. Genome Res 14, 1188–1190. PEGylation of the G-CSF molecule by MTGase enzyme C. Maullu et al. 6750 FEBS Journal 276 (2009) 6741–6750 ª 2009 The Authors Journal compilation ª 2009 FEBS . that PEGylation of one of the two glutamines impairs the PEGylation of the neighboring one. The computa- tional prediction of the relative abundance of the. limitations, because of the potential multiple sites of conjugation and the consequent heterogeneity of the PEGylated proteins. The purification of these mixtures