© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEWS AND VIEWS when the crucial drivers, as well as their major recruits (to ‘spread determinants’), are all tolerized. This is the tack pursued by Robinson et al., where the DNA encoding each of the antigens whose specific T cells are easily recruitable is used in a toleranceinducing regimen to ask whether the number of disease relapses usually found in autoimmune animals can be reduced. The last question remaining is whether DNA coding for the putatively toleranceinducing segments of the protein antigens would actually induce tolerance, rather than exacerbating autoimmunity. Robinson et al. pursued the notion that the breadth of response to the initiating peptide (they tested two), after determinant spreading, would point to candidate antigenic molecules that should be employed in the attempted tolerance trial. Previous studies12,13 had shown clearly that DNA encoding the antigen in question tended to induce deletion/anergy, or to activate regulatory helper T cells (TH2 cells), which provide a suppressive environment for self-reactive T cells. Robinson et al. attempted treatment of previously PLP(139–151)-primed mice, on day 18, with a DNA cocktail encoding several antigens to which responses had spread in their earlier array experiments. They found that this DNA cocktail treatment was successful in reducing subsequent relapses, and it was even more potent if DNA encoding the cytokine interleukin (IL)-4 were added. Treatment with DNA representing PLP(139–151) alone was a relatively poor substitute for the cocktail. Although the authors claim that the cocktail constituents were ‘array determined,’ the four neuroantigens chosen would have been a reasonable choice by anyone with knowledge of the EAE literature. Nevertheless, these results strongly suggest that DNA cocktail administration leads to a diversified tolerogenic presentation of the antigens in the cocktail, probably owing to the production of a TH2 cytokine milieu, which helps to maintain the downregulated state achieved following recovery from the acute paralytic attack. Additional studies need to be undertaken to discover the most efficacious constitution of the tolerizing cocktail for different subjects and each disease condition. By assessing the diversification from an initially narrow antibody response using a protein array, Robinson et al. have gained insight into the complexity of the tolerogenic DNA that would have to be used to induce and maintain a state of tolerance in each mouse. Apparently, using DNA encod-
ing four proteins led to no obvious exacerbation of disease in the mice. Therefore, in the human situation, a sequential study of serum reactivity in patients or of T-cell reactivity in vitro to a large set of neuroantigenic peptides could predict the course of future tolerogenic protection studies. The approach may succeed in providing a therapeutic avenue, rather than a preventive strategy, which would fill the long-sought need for a treatment for TH1 autoimmune diseases such as multiple sclerosis, type I diabetes and arthritis. Further mechanistic studies to elaborate the connection between determinant spreading at the antibody level, the choice of antigens and cytokines to be used in the DNA tolerization vaccine and the true extent of reduction in the selfreactive T- (or B-) cell responses should establish the generality of these exciting
results, and their usefulness in devising therapy for patient populations. 1. Robinson, W.H. et al. Nat. Biotechnol. 21, 1033–1039 (2003). 2. Jenner, E. The Three Original Publications on Vaccination against Smallpox by Edward Jenner, in Harvard Classics, vol. 38 (P.F. Collier & Son, New York, 1910). 3. Billingham, R.E., Brent, L. & Medawar, P.B. Nature 172, 603–606 (1953). 4. Lehmann, P.V., Forsthuber, T., Miller, A. & Sercarz, E.E. Nature 358, 155–157 (1992). 5. Tuohy, V.K., Yu, M., Yin, L., Kawczak, J.A. & Kinkel, R.P. J. Exp. Med. 189, 1033–1042 (1999). 6. Genain, C.P., et al. Nat. Med. 5, 170–175 (1999). 7. Benjamin, D.C. et al. Annu. Rev. Immunol. 2, 67–101 (1984). 8. Simitsek, P.D. et al. J. Exp. Med. 181, 1957–1963 (1995). 9. Sercarz, E.E. J. Autoimmun. 14, 275–277 (2000). 10. Sercarz, E.E. Nat. Immunol. 3, 110–112 (2002). 11. Maverakis, E. et al. Proc. Natl. Acad. Sci. USA 100, 5342–5347 (2003). 12. Bot, A. et al. J. Immunol. 167, 2950–2955 (2001). 13. Garren, H. et al. Immunity 15, 15–22 (2001).
Two hands (or four) are better than one Ben M Dunn & Jörg Bungert A bacterial protease inhibitor provides a scaffold for the in vitro evolution of specific inhibitors of human serine proteases. The ideal of a stable protein scaffold onto which a variable loop could be grafted to provide for specific interaction with protein targets has been a goal since the beginnings of the protein engineering discipline1,2. Nature has provided a number of suitable systems, but achieving the desired selectivity has been a sizable barrier. Two papers, one in a recent issue of the Proceedings of the National Academy of Sciences USA3 and the other published here4, take giant steps toward the discovery of truly selective inhibitors of peptidases. In the first report, Komiyama et al.3 optimize active site interactions and randomize potential ‘adventitious’ contacts in eglin c from the medicinal leech, Hirudo medicinalis. In this issue, Stoop and Craik4 describe a combinatorial method for achieving wide variation in four loops that contact protease targets to provide an even finer control of selectivity. Ben M. Dunn and Jörg Bungert are in the Department of Biochemistry and Molecular Biology, University of Florida College of Medicine, P.O. Box 100245, 1600 SW Archer Road, Gainesville, Florida 32610-0245, USA. e-mail:
[email protected]
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 9 SEPTEMBER 2003
Protein proteinase inhibitors5 typically provide a tightly folded protein core with at least one prominent loop that protrudes, to some degree, from the surface and that can interact with the active site of a proteolytic enzyme (see Figs. 1a,b). In most but not all cases, the loop binds through an active site cleft in a ‘canonical’6 manner, much as a substrate would. Because of subtle geometric differences, a substrate would be cleaved, whereas an inhibitor is not. In some cases, the inhibitor is slowly cleaved, but remains tightly associated with the enzyme7. In either case, the enzyme active site is occupied, preventing substrate processing. Nature has provided many examples of the coevolution of peptidases and peptidase inhibitors. For a select number of these systems, we now have excellent structural details of the atomic contacts involved in the interactions between the two protein molecules. Unfortunately, for many proteases of medical importance, naturally occurring protein inhibitors have not arisen or have not been discovered. Therefore, there has been great interest, both theoretical and practical, in adapting known pro-
1019
tein proteinase inhibitors to do the new job of inhibiting deleterious proteolysis. In addition, as the binding of a proteinase inhibitor to a proteinase can be a welldefined example of a protein-protein interaction, these studies have a wider value in the current era of defining protein ‘interactomics.’ Therefore, studies of the energetic changes involved in these macromolecular interactions8 can provide predictive algorithms. Furthermore, attempts have been made to minimize the macromolecular scaffold to permit the use of peptide synthesis to create mimics that would allow combinatorial methods9. Perona and Craik have previously provided a thorough analysis of interaction between substrates and serine proteinases10. Three separate loops of the enzyme structure are involved in contacting a polypeptide substrate, thus providing a large surface for interaction with a molecule larger that a simple peptide. In addition, the interactions between protein protease inhibitors and their target enzymes involve additional contacts beyond the active site loop that binds to the catalytic cleft. These two factors, when combined, lead to a large surface area of contact between the two proteins, which can permit fine-tuning of the potency and specificity of inhibition. Komiyama et al.3 have taken advantage of the ‘adventitious contacts’ (interactions that occur outside of the canonical loop–active site cleft) between eglin c and proteases to convert the natural inhibitor, which has a specificity toward bacterial serine proteinases, into an inhibitor of the furin/kex-2 class of intracellular processing enzymes. The furin/kex-2 enzymes are members of the superfamily defined by the subtilisins. They used a two-step procedure in which the first step was the optimization of the sequence of the reactive site loop for binding to furin and the second step was the randomization of the sequences outside of the active site contact area. Ten codons of eglin c (codons 33, 35, 37, 39, 40, 47, 49, 50, 65 and 68) were randomized in a series of libraries of different combinations, resulting in ∼500 variants. These were expressed and screened for inhibition of several enzymes. Mutations at four of the codons (33, 39, 40 and 49) provided reproducible enhancement of inhibition of furin. Further randomization of codon 49 produced an inhibitor with a Ki value of 310 pM for furin, which, when added to the culture medium, blocked furin-dependent processing of von Willebrand factor in
1020
a
b Secondary site 100s 60s
IV
I
II III
III II IV
I 50s 80s Primary site
c I
II
III
IV
1. Synthetic DNA shuffling using degenerate oliogonucleotides
I/III
II/III
III/IV
Phage display, selection and combining clones expressing substrateinteracting proteins 2. Fragmentationbased DNA shuffling
©Bob Crimi
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEWS AND VIEWS
Figure 1 Inhibitors get to grips with serine protease. (a) Crystal structure of an ecotin dimer bound to two protease molecules. (b) Schematic diagram showing an ecotin dimer in the center, each with four arms that interact with two separate molecules of trypsin. One arm of each ecotin (the 80s loop) interacts with the active-site cleft of one trypsin molecule and is supported by a second arm (the 50s loop). Two arms from the other octopus (the 60s loop and the 100s loop, different color) make additional contacts with the trypsin molecule to complete the large interaction area. (c) Two-step procedure for introducing mutations into all four surface loops of ecotin. Numbers I to IV correspond to the 50s, 60s, 80s, and 100s loops of ecotin.
African green monkey kidney (COS-1) cells. Stoop and Craik4 have been working on a different system—the bacterial protein ecotin, which has been found to be an excellent inhibitor of serine proteases of the trypsin/chymotrypsin family. Their work takes advantage of a unique aspect of ecotin inhibition; it forms a dimer and inhibits two molecules of, for example, trypsin. The 80s loop of one ecotin interacts with the active site cleft of one trypsin molecule, with the 50s loop of the same ecotin supporting the 80s loop and providing additional contacts. In addition, the 60s and 100s loops of the second ecotin molecule interact with the same trypsin molecule to
complete the 2,850 Å2 contact surface. The reciprocal contacts are made with the second trypsin molecule; that is, the 80s loop of the second ecotin serves as the reactive site loop with the assistance of the 50s loop, and the 60s and 100s loops come from the first ecotin molecule. The ability to rapidly introduce diversity into a protein by molecular evolution in vitro is greatly enhanced by a procedure called DNA shuffling, first introduced by Stemmer12,13. In this method, a DNA clone encoding the protein of interest is fragmented by DNase I digestion, leading to a pool of random DNA fragments. Fragments of a specific size range are isolated, annealed and subjected to PCR for mutage-
VOLUME 21 NUMBER 9 SEPTEMBER 2003 NATURE BIOTECHNOLOGY
© 2003 Nature Publishing Group http://www.nature.com/naturebiotechnology
NEWS AND VIEWS nesis (by error-prone PCR) and reassembly of the gene. The pool of mutated expression clones is then subjected to selection procedures, for example, by assaying for proteinprotein interactions or enzymatic activity. The DNA shuffling and selection steps can be repeated several times leading to the rapid evolution of a protein with a desired activity, such as improvement of interactions with other proteins (or substrates), altered substrate specificity or enhanced enzyme activity. Protein engineering by DNA shuffling represents a milestone in the rapid evolution of proteins. The fragmentation-based DNA shuffling procedure has distinct advantages over random mutagenesis approaches, including the possibility of eliminating deleterious mutations. In addition, Crameri et al.14 used fragmentationbased DNA shuffling to recombine variants present in parent clones of highly related genes into a DNA library of expression clones, a process referred to as ‘family shuffling.’ A limitation of the original DNA shuffling technique is the fact that closely linked mutations recombine with poor efficiency, because of inefficient annealing of DNA fragments harboring many mismatches in close proximity. This is the basis for the observation that individual clones of expression libraries generated by fragmentation-based DNA shuffling do not deviate as much from the parental clones as desired. This problem was recently solved by the development of a new technique called synthetic DNA shuffling15. Here, small overlapping synthetic oligonucleotides covering the protein region of interest are generated and assembled into expression clones by PCR. The use of synthetic DNA shuffling has the advantage that desired mutations can be introduced into individual oligonucleotides that then randomly recombine during the assembly step. This allows amino acids to recombine independently, thus destroying the linkages between amino acids found in parental genes. Indeed, Ness et al.15 compared libraries generated by fragmentation-based or synthetic DNA shuffling. They found that synthetic DNA shuffling leads to a greater diversity and deviation from parental clones than fragmentation-based DNA shuffling. Furthermore, synthetic DNA shuffling allows investigators to introduce mutations based on optimal species-specific codon usage. Finally, by taking advantage of the degenerate code, degenerate oligonucleotides can be synthesized to increase the
diversity at single amino acid positions, known as tetranomial diversity. Another limitation commonly encountered during the development of proteins using molecular evolution is the lack of an appropriate selection procedure for clones expressing the protein with the desired activity. Experiments leading to the evolution of proteins with altered substrate specificity or increased affinities in proteinprotein interactions are facilitated by the use of phage display. Stoop et al.11 combined DNA shuffling with phage display, termed ‘shuffled proteins on phages’ or SPOP, to characterize the role of amino acids in the interaction between plasminogen activation inhibitor 1 (PAI-1) with the serine proteinase tissue plasminogen activator (tPA). PAI-1 expression clones generated by DNA shuffling were used to generate a phage display library. Variant PAI-1 proteins, displayed on the phage surface, were incubated with tPA and the complexes were captured using a tPA antibody. The work by Stoop and Craik published in this issue represents a major step toward the molecular evolution of proteins. In a tour de force, the authors have combined synthetic DNA shuffling, fragmentationbased DNA shuffling and phage display to engineer and select for macromolecular protease inhibitors with increased specificity for distinct serine proteases (see Fig. 1c). It would not be possible to create expression libraries of variant ecotin with mutations in all four surface loops by synthetic DNA shuffling only. The authors elegantly circumvented this problem by first generating three phage display libraries with mutations in two of the four loops by synthetic DNA shuffling and restricted tetranomial diversity. They then selected clones that express proteins capable of interacting with the substrate using phage display. All of the clones expressing interacting proteins and derived from the three libraries were then combined into a single phage library and subjected to fragmentation-based DNA shuffling (‘family shuffling’) to generate expression clones that contain mutations in all four surface loops of ecotin. The authors succeed in generating a population of clones that bear mutations in all of the loops and demonstrate quite convincingly that this method is very useful in generating and selecting ecotin variants with high specificity for plasma kallikrein, membrane-type serine protease 1 and factor XIIa. The mutagenesis performed by Stoop and Craik was restricted to 20 amino
NATURE BIOTECHNOLOGY VOLUME 21 NUMBER 9 SEPTEMBER 2003
acids on the four surface loops, and they limited the number of substituted residues in the 20 positions to four amino acids by restricted tetranomial diversity using degenerate oligonucleotides. By designing the synthetic oligonucleotides, they carefully took into account natural species-specific variations that are known to be tolerated during evolution and preferentially introduced amino acid residues enriched in protein-protein interaction surfaces16. The feature of four contact loops in the ecotin dimer provides additional flexibility in inhibitor design, especially considering that the amino acid residues in the secondary contact regions of trypsin are not likely to be as conserved as the contact regions near the active site cleft. Therefore, achieving additional levels of selectivity is possible in the ecotin system because of the extra contacts, which are missing in the eglin c and other Kunitz-type domain inhibitors. Thus, even using a limited diversity in the 20 positions that were modified, Stoop and Craik were able to achieve a tight-binding (11 pM) inhibitor of plasma kallikrein that is selective when compared with binding to factor Xa, factor XIa, urokinase plasminogen activator, thrombin and membranetype serine protease 1 by four to seven orders of magnitude. Up to now, selectivity of 10,000-fold has been achieved only with mutations in the reactive site region. This line of research clearly represents a major tool in the molecular evolution of proteins that exhibit complex protein-protein interaction interfaces. 1. Ulmer, K.M. Science 219, 666–671 (1983). 2. Perutz, M. New Scientist 106, 12–15 (1985). 3. Komiyama, T., et al. Proc. Natl. Acad. Sci. USA 100, 8205–8210 (2003). 4. Stoop, A.A. & Craik C.S., Nat. Biotechnol. 21, 1063–1068 (2003). 5. Fritz, H., et al. Hoppe-Seylers Zeitschrift fur Physiologische Chemie 353, 1950–1952 (1972). 6. Bode, W. & Huber R. Eur. J. Biochem. 204, 433–451 (1992). 7. Laskowski, M. & Kato I. Annu. Rev. Biochem. 49, 593–625 (1980). 8. Laskowski, M., Qasim M.A., & Yi Z.P. Curr. Opin. Struct. Biol. 13, 130–139 (2003). 9. McBride, J.D. & Leatherbarrow R.J. Curr. Med. Chem. 8, 909–917 (2001). 10. Perona, J.J. & Craik C.S. Protein Sci. 4, 337–360 (1995). 11. Stoop, A.A., et al. J. Mol. Biol. 301, 1135–1147 (2000). 12. Stemmer, W.P.C. Proc. Natl. Acad. Sci. USA 91, 10747–10751 (1994). 13. Stemmer, W.P.C. Nature 370, 389–291 (1994). 14. Crameri, A., Raillard, B. S.-A. & Stemmer W.P.C. Nature 391, 288–291 (1998). 15. Ness, J.E., et al. Nat. Biotechnol. 20, 1251–1255 (2002). 16. Bogan, A.A. & Thorn K.S. J. Mol. Biol. 280, 1–9 (1998).
1021
