Directed in Vitro Evolution and Crystallographic Analysis of a Peptide-binding Single Chain Antibody Fragment (scFv) with Low Picomolar Affinity
2004; Elsevier BV; Volume: 279; Issue: 18 Linguagem: Inglês
10.1074/jbc.m309169200
ISSN1083-351X
AutoresChristian Zahnd, Silvia Spinelli, Béatrice Luginbühl, Patrick Amstutz, Christian Cambillau, Andreas Plückthun,
Tópico(s)Glycosylation and Glycoproteins Research
ResumoWe generated a single chain Fv fragment of an antibody (scFv) with a binding affinity of about 5 pm to a short peptide by applying rigorous directed evolution. Starting from a high affinity peptide binder, originally obtained by ribosome display from a murine library, we generated libraries of mutants with error-prone PCR and DNA shuffling and applied off-rate selection by using ribosome display. Crystallographic analysis of the scFv in its antigen-bound and free state showed that only few mutations, which do not make direct contact to the antigen, lead to a 500-fold affinity improvement over its potential germ line precursor. These results suggest that the affinity optimization of very high affinity binders is achieved by modulating existing interactions via subtle changes in the framework rather than by introducing new contacts. Off-rate selection in combination with ribosome display can evolve binders to the low picomolar affinity range even for peptide targets. We generated a single chain Fv fragment of an antibody (scFv) with a binding affinity of about 5 pm to a short peptide by applying rigorous directed evolution. Starting from a high affinity peptide binder, originally obtained by ribosome display from a murine library, we generated libraries of mutants with error-prone PCR and DNA shuffling and applied off-rate selection by using ribosome display. Crystallographic analysis of the scFv in its antigen-bound and free state showed that only few mutations, which do not make direct contact to the antigen, lead to a 500-fold affinity improvement over its potential germ line precursor. These results suggest that the affinity optimization of very high affinity binders is achieved by modulating existing interactions via subtle changes in the framework rather than by introducing new contacts. Off-rate selection in combination with ribosome display can evolve binders to the low picomolar affinity range even for peptide targets. Directed evolution of proteins in vitro has become a widely applied strategy to generate proteins with a desired property (1Amstutz P. Forrer P. Zahnd C. Plückthun A. Curr. Opin. Biotechnol. 2001; 12: 400-405Google Scholar). Especially in the generation of high affinity binders, the iterative succession of randomization and selection was shown to efficiently mimic natural affinity maturation. The success of this approach is dependent on the size and the quality of the library and the power of the selection method used. Technologies that work entirely in vitro such as ribosome display (2Hanes J. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 1997; 94: 4937-4942Google Scholar) and mRNA display (3Roberts R.W. Szostak J.W. Proc. Natl. Acad. Sci. U. S. A. 1997; 94: 12297-12302Google Scholar) are more favorable than methods that work partially in vivo such as phage display (4Smith G.P. Virology. 1988; 167: 156-165Google Scholar) or fully in vivo such as the yeast two-hybrid system (5Fields S. Song O. Nature. 1989; 340: 245-246Google Scholar) or the protein complementation assay (6Mössner E. Koch H. Plückthun A. J. Mol. Biol. 2001; 308: 115-122Google Scholar) as the in vitro technologies do not need transformation of cells after each new round of randomization. Therefore, they allow the handling of much larger libraries and more cycles of randomization, and the experimental work is accelerated greatly. A prerequisite of any evolution is randomization between different selection rounds. In ribosome display this is facilitated by the use of linear DNA. The randomization occurs at a low rate by the intrinsic error rate of the polymerase used but can be enhanced by error-prone PCR (7Zaccolo M. Gherardi E. J. Mol. Biol. 1999; 285: 775-783Google Scholar), by DNA shuffling (8Stemmer W.P. Nature. 1994; 370: 389-391Google Scholar), or both, thereby generating highly diverse pools. While the generation of binders having binding constants in the subnanomolar range can be achieved, e.g. with synthetic antibody libraries and established techniques (9Knappik A. Ge L. Honegger A. Pack P. Fischer M. Wellnhofer G. Hoess A. Wölle J. Plückthun A. Virnekäs B. J. Mol. Biol. 2000; 296: 57-86Google Scholar), the generation of very high affinity binders with binding constants in the lowest picomolar affinity range is difficult for several reasons. First, a very stringent selection pressure must be applied to separate improved binders from the already very high affinity precursors. Second, selected binders must be eluted efficiently, which may become very difficult for binders with very slow off-rates. Here ribosome display offers a significant advantage since bound binders must not be eluted, but the ribosomally bound mRNA can be recovered by the addition of chelating agents, which destabilize the ribosomal complex (1Amstutz P. Forrer P. Zahnd C. Plückthun A. Curr. Opin. Biotechnol. 2001; 12: 400-405Google Scholar). In particular the generation of very high affinity peptide binders is made difficult by the relatively high flexibility of the peptide in the unbound state and the corresponding loss of entropy upon binding. This is less of a problem for comparatively rigid antigens such as hydrophobic small molecular weight compounds for which subpicomolar binders are known (10Boder E.T. Midelfort K.S. Wittrup K.D. Proc. Natl. Acad. Sci. U. S. A. 2000; 97: 10701-10705Google Scholar). We applied a competitive selection for increased off-rates to affinity mature a high affinity peptide binder previously selected with ribosome display from a murine library (11Hanes J. Jermutus L. Weber-Bornhauser S. Bosshard H.R. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 1998; 95: 14130-14135Google Scholar). The peptide was derived from the yeast transcription factor GCN4. We constructed different mutants of a high affinity binder and generated from them second generation libraries using DNA shuffling and error-prone PCR. From these libraries we successfully isolated binders in the low picomolar affinity range. By determining the crystal structure both in the free and antigen-bound state we could show that the gain in affinity of 500-fold, compared with its likely germ line precursor, was almost exclusively a result of second sphere mutations not being in direct contact with the antigen. These findings may have great impact on future library design and affinity maturation strategies. Expression and Purification of Single Chain Fv Fragments—The scFv 1The abbreviations used are: scFv, single chain Fv fragment; VL, variable domain of the light chain; VH, variable domain of the heavy chain; 8-oxo-dGTP, 8-oxo-2′-deoxyguanosine 5′-triphosphate; dPTP, 6-(deoxy-β-d-erythro-pentofuranosyl)-3,4-dihydro-8H-pyrimido-[4,-5c][1,2]oxazine-7-one-5′-triphosphate; RIA, radioimmunoassay; CDR, complementarity-determining region. genes were cloned into the periplasmic expression vector pAK400 (12Krebber A. Bornhauser S. Burmester J. Honegger A. Willuda J. Bosshard H.R. Plückthun A. J. Immunol. Methods. 1997; 201: 35-55Google Scholar) introducing a His6 tag, expressed in Escherichia coli SB536 (13Bass S. Gu Q. Christen A. J. Bacteriol. 1996; 178: 1154-1161Google Scholar), and purified by immobilized metal ion affinity chromatography and subsequent antigen affinity chromatography as described previously (11Hanes J. Jermutus L. Weber-Bornhauser S. Bosshard H.R. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 1998; 95: 14130-14135Google Scholar). For structure determination, a seleno-Met-containing variant of clone H6 was grown in 1 liter of M9 minimal medium to an OD600 of 0.6. An amino acid mixture containing 100 mg/liter each of Lys, Thr, and Phe and 50 mg/liter each of Leu, Ile, Val, and seleno-Met was added. After 1 h, isopropyl-1-thio-β-d-galactopyranoside was added to a final concentration of 1 mm for expression overnight. The protein was purified as described above. The Library Construction—The scFv fragments C11L34, L24, L107, L135, L107–135, H6, and H67 in the vector pAK400 were PCR-amplified using primers SDAla+ (5′-AGACCACAACGGTTTCCCTCTAGAAATAATTTTGTTTAACTTTAAGAAGGAGATATATCCATGGCGGACTACAAAGAT) and Sfi_rescue (5′-GCCCTCGGCCCCCGAGGC). A total of 1 μg of PCR product of an equimolar mixture of all clones was used for DNase I shuffling (14Adey N.B. Stemmer W.P.C. Kay B.K. Kay B.K. Winter J. McCafferty J. Phage Display of Peptides and Proteins. Academic Press, Cambridge1996: 280-292Google Scholar) as described previously (15Jermutus L. Honegger A. Schwesinger F. Hanes J. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 2001; 98: 75-80Google Scholar). Some of the reassembled PCR products were further randomized by error-prone PCR using primers SDAla+ and Sfi_rescue. Error-prone PCR was performed using the dNTP analogues 8-oxo-dGTP and dPTP according to Ref. 7Zaccolo M. Gherardi E. J. Mol. Biol. 1999; 285: 775-783Google Scholar with small modifications. Twenty-six cycles of error-prone PCR were performed in the presence of 85 μm dPTP, 85 μm 8-oxo-dGTP, and 50 μm dGTP, dATP, dTTP, and dCTP each. The final mutation rate after DNase I shuffling and error-prone PCR and the distribution of the mutations were determined by sequencing about 2000 bp. A gene III linker was fused to the library as described earlier (11Hanes J. Jermutus L. Weber-Bornhauser S. Bosshard H.R. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 1998; 95: 14130-14135Google Scholar). Selection for Improved Affinities—The library was transcribed and translated in vitro as described in Ref. 11Hanes J. Jermutus L. Weber-Bornhauser S. Bosshard H.R. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 1998; 95: 14130-14135Google Scholar. The ternary complexes of ribosome, mRNA, and displayed proteins were equilibrated with 1 nm biotinylated peptide GCN4(7P14P) (16Berger C. Weber-Bornhauser S. Eggenberger J. Hanes J. Plückthun A. Bosshard H.R. FEBS Lett. 1999; 450: 149-153Google Scholar) at 4 °C overnight. Every sample was split into two aliquots, and only to one, 1 μm non-biotinylated GCN4(7P14P) was added. The aliquots were incubated for a defined time span, which was increased from round to round, starting with 2 h in the first round and going up to 10 days after the fourth round in a rollover shaker at 4 °C. The complexes were recovered by binding to streptavidin-coated magnetic beads (Roche Applied Science) for 30 min. The beads were washed five times, and the RNA was eluted and purified as described in Ref. 11Hanes J. Jermutus L. Weber-Bornhauser S. Bosshard H.R. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 1998; 95: 14130-14135Google Scholar. Affinity Comparison of Pools and Single Clones by Radioimmunoassay (RIA) and Inhibition BIACORE—For the analysis of single clones, the selected pools were cloned into pTFT74 (17Ge L. Knappik A. Pack P. Freund C. Plückthun A. Borrebaeck C.A.K. Antibody Engineering. Oxford University Press, New York1995: 229-266Google Scholar). RNA of single clones was transcribed directly from plasmid pTFT74, whereas the pools were transcribed from a PCR product. RIAs were performed as described previously (18Auf der Maur A. Zahnd C. Fischer F. Spinelli S. Honegger A. Cambillau C. Escher D. Plückthun A. Barberis A. J. Biol. Chem. 2002; 277: 45075-45085Google Scholar). Of some clones, absolute affinities were measured using the inhibition method on a BIACORE 3000 (19Nieba L. Krebber A. Plückthun A. Anal. Biochem. 1996; 234: 155-165Google Scholar). The purified protein was diluted to 1 nm and incubated with different concentrations of antigen overnight at 4 °C. The samples were injected over a chip, which was coated to maximal density with the antigen used for selection. The slope of the association curves in the linear phase was plotted against the concentration of soluble antigen. KD was determined from at least three independent curves as described previously (11Hanes J. Jermutus L. Weber-Bornhauser S. Bosshard H.R. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 1998; 95: 14130-14135Google Scholar). Crystallization—Crystals of clone H6 in complex with the antigen and clone C11L34 in the absence of the peptide were obtained using the hanging drop method. Drops of 1 μl (protein concentration, 8 mg/ml) of the unliganded Fv GCN4 were mixed with 1 μl of the well solution (1.1 m ammonium sulfate, 150 mm sodium citrate, pH 4.8). Crystals appeared after 2 days at 20 °C belonging to space group P212121 (a = 35.08 Å, b = 60.53 Å, and c = 123.05 Å) and contained one molecule per asymmetric unit (Vm = 2.29 Å3/Da, 46% solvent (20Matthews B.W. J. Mol. Biol. 1968; 33: 491-497Google Scholar)). Crystals of the complex of H6 and the peptide YHLENEVARLKK were obtained by mixing the scFv and the peptide in a 1:2 molar ratio. Drops of 1 μlofthe complex (protein concentration, 7.4 mg/ml) were mixed with 1 μl of a solution containing 32–28% (w/v) polyethylene glycol 4000, 0.1 m Tris/HCl, pH 7.5. The space group was P21 (a = 37.24 Å, b = 36.29 Å, c = 84.46 Å, β = 90.5°) with one complex per asymmetric unit and a specific protein volume of 2.03 Å3/Da. Data Collection and Processing—Data of the free scFv fragment were collected at 100 K on a MAR-Research Imaging Plate (MAR, Hamburg, Germany) placed on a Rigaku RU200 rotating anode using the CuKα radiation. Crystals were frozen using 25% glycerol, and they diffracted to 2.6 Å. Data integration and reduction were performed using DENZO (21Otwinowski Z. Minor W. Carter Jr., C.W. Sweet R.M. Methods in Enzymology. Vol. 276. Academic Press, New York1997: 307-326Google Scholar) and SCALA (22Collaborative Computational Project No. 4 (CCP4)Acta Crystallogr. Sect. D Biol. Crystallogr. 1994; 50: 760-763Google Scholar). A crystal of the complex was frozen with 16% glycerol. Data were collected at 100 K in the beam-line ID14-EH4 at the European Synchrotron Radiation Facility (Grenoble, France), λ = 0.988 Å. The data reduction was performed with SCALA (22Collaborative Computational Project No. 4 (CCP4)Acta Crystallogr. Sect. D Biol. Crystallogr. 1994; 50: 760-763Google Scholar). The data of the complex were merged with TRUNCATE (22Collaborative Computational Project No. 4 (CCP4)Acta Crystallogr. Sect. D Biol. Crystallogr. 1994; 50: 760-763Google Scholar) in CCP4i using the anisotropic correction option. The RSym for the free scFv and the complex were 5.9 and 6.2%, respectively. The statistics of the data sets are shown in Table I.Table ICrystal parameters and refinement statisticsFree scFvComplexData collection Space groupMP212121P21 Total/unique number of reflections51,966/6,870129,374/10,027 Percentage of data >1σ (overall/last shell)aLast shell, 2.9-2.8 Å (free scFv); 2.43-2.35 Å (complex)96.0/96.097.2/97.6 Overall I/σ (I) (last shell)11.9/2.68.5/4.1 Resolution limits24.6/2.820.0/2.3 Rmerge (%) (overall/last shell)aLast shell, 2.9-2.8 Å (free scFv); 2.43-2.35 Å (complex)5.9/28.16.2/18.4Refinement Number of protein/solvent atoms1,65817,648/66 Number of reflections6,7969,940 Resolution limits (Å)12.0-2.815.0-2.35 R/Rfree value (%)20.05/26.418.0/21.5 r.m.s.d.br.m.s.d., root mean square deviation on bonds (Å)/angels (°)0.012/2.00.025/2.1 r.m.s.d. on impropers/dihedrals (°)1.14/27.80.86/25.2 Mean B-factor (Å2) (main/side chain)35.423.8/27.0 Peptide (main/side chain)25.0/28.0 Water33.0a Last shell, 2.9-2.8 Å (free scFv); 2.43-2.35 Å (complex)b r.m.s.d., root mean square deviation Open table in a new tab Structure Determination and Refinement—Both structures were solved by molecular replacement (23Rossman M.G. Blow D.M. Acta Crystallogr. 1962; 15: 24-31Google Scholar) using the program AMoRe (24Navaza J. Dodson E.J. Gover S. Wolf W. Molecular Replacement: Proceedings of the CCP4 Study Weekend. Science and Engineering Research Council (SERC) Daresbury Laboratory, Warington, UK1992: 87-90Google Scholar). Rotation and translation searches were performed using the murine Fv SE155-4 (Protein Data Bank entry 1MFA) for the unliganded Fv. The structure of the unliganded Fv was used to solve the structure of the complex. The molecular replacement solution for the unliganded Fv gave an initial correlation coefficient of 45% and R-factor of 47%, while the values for the complex were 46 and 48%, respectively, between 8.0 and 3.0 Å. Refinement was performed first with CNS (25Brunger A.T. Adams P.D. Clore G.M. DeLano W.L. Gros P. Grosse-Kunstleve R.W. Jiang J.S. Kuszewski J. Nilges M. Pannu N.S. Read R.J. Rice L.M. Simonson T. Warren G.L. Acta Crystallogr. Sect. D Biol. Crystallogr. 1998; 54: 905-921Google Scholar) using standard protocols followed by REFMAC (26Murshudov G.N. Alexei A.V. Dodson E.J. Acta Crystallogr. Sect. D Biol. Crystallogr. 1997; 53: 240-255Google Scholar) using maximum likelihood, incorporating bulk solvent corrections and translation-libration-screw (TLS) anisotropic refinement. After each refinement cycle, a new map was calculated, and the model was fitted with Turbo-Frodo (27Roussel A. Cambillau C. Silicon Graphics Geometry Partners Directory. Silicon Graphics Corp., Mountain View, CA1991: 81Google Scholar). Final refinement data are summarized in Table I. The coordinates and structure factors of the free and peptide-bound Fv have been deposited in the Protein Data Bank at Research Collaboratory for Structural Bioinformatics as entries 1P4I and 1P4B, respectively. Library Construction—In a previous study, we selected a group of scFv fragments from a murine library by using ribosome display, a selection method that works entirely in vitro and therefore allows the selection from very large libraries (Fig. 1). The selected scFv fragments bound with very high affinity to the peptide GCN4(7P14P) derived from the yeast transcription factor GCN4 (11Hanes J. Jermutus L. Weber-Bornhauser S. Bosshard H.R. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 1998; 95: 14130-14135Google Scholar, 16Berger C. Weber-Bornhauser S. Eggenberger J. Hanes J. Plückthun A. Bosshard H.R. FEBS Lett. 1999; 450: 149-153Google Scholar). The highest affinity clone, named "C11L34," had an affinity of 40 pm and acquired a crucial amino acid substitution during ribosome display rounds that led to a 65-fold affinity improvement compared with its likely progenitor from the murine B-cells (11Hanes J. Jermutus L. Weber-Bornhauser S. Bosshard H.R. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 1998; 95: 14130-14135Google Scholar). We constructed a second generation library in two steps. First, we generated five different point mutants of the high affinity clone and a double mutant, based on mutations that had been enriched in the first selection experiment, and measured their affinities (Table II and see Fig. 4). These mutants were taken as a template for the construction of three second generation libraries. The first pool, named "S," was generated by using DNA shuffling of all seven clones including C11L34, thereby recombining all mutations (8Stemmer W.P. Nature. 1994; 370: 389-391Google Scholar). Two other pools were generated by the use of error-prone PCR. Pool "R" was generated by error-prone amplification of C11L34, and pool "SR" was generated by the combination of the two approaches: the mutants were amplified by error-prone PCR and then subjected to DNA shuffling.Table IISummary of the mutations and affinities of different GCN4 bindersCloneaThe first eight clones listed were found in a previous selection (11), and their mutations compared with C11L34 were introduced by site-directed mutagenesis. Clones H6 and L135H6 were constructed because it was shown recently that the residue on position H6 is often critical for stability and affinity (32, 34). All these clones were used for the generation of different libraries by DNA shuffling and error-prone PCR with the exception of clone L112H6, which was constructed at a later point in the project. The last five clones listed were found during the directed evolution and showed promising signals in a competition RIAMutationsbThe mutations are given in the AHo nomenclature (35)KDcThe affinities were determined by inhibition BIACORE. The affinity improvement during the directed evolution was monitored mainly by competition RIA (see text)pmC11L34dClone C11L34 was found to have a single point mutation compared with its likely progenitor (11). All other clones are derived from C11L34 and contain its mutation, L42(Asn → Ser). For historical reasons, its name is indicated in the Kabat nomenclature (36) and was retainedL42(Asn → Ser)40 ± 4L24L24(Arg → His)32 ± 13L107L107(Ala → Val)48 ± 6L135L135(Asn → Asp)23 ± 4H6H6(Glu → Gln)20 ± 2H67H67(Ile → Val)47 ± 7L107L135L107(Ala → Val)L135(Asn → Asp)47 ± 5L135H6L135(Asn → Asp)H6(Glu → Gln)16 ± 1352SR4L13(Thr → Ser)L107(Ala → Val)5.2 ± 2.3L135(Asn → Asp)H30(Ser → Leu)63S3L145(Leu → Pro)NDeND, not determined70SR4L13(Thr → Ala)L135(Asn → Asp)NDH96(Leu → Pro)82SR4H70(Tyr → His)H94(Asn → Ser)ND84SR4H32(Thr → Ala)H70(Tyr → His)NDH94(Asn → Ser)a The first eight clones listed were found in a previous selection (11Hanes J. Jermutus L. Weber-Bornhauser S. Bosshard H.R. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 1998; 95: 14130-14135Google Scholar), and their mutations compared with C11L34 were introduced by site-directed mutagenesis. Clones H6 and L135H6 were constructed because it was shown recently that the residue on position H6 is often critical for stability and affinity (32Honegger A. Plückthun A. J. Mol. Biol. 2001; 309: 687-699Google Scholar, 34Jung S. Spinelli S. Schimmele B. Honegger A. Pugliese L. Cambillau C. Plückthun A. J. Mol. Biol. 2001; 309: 701-716Google Scholar). All these clones were used for the generation of different libraries by DNA shuffling and error-prone PCR with the exception of clone L112H6, which was constructed at a later point in the project. The last five clones listed were found during the directed evolution and showed promising signals in a competition RIAb The mutations are given in the AHo nomenclature (35Honegger A. Plückthun A. J. Mol. Biol. 2001; 309: 657-670Google Scholar)c The affinities were determined by inhibition BIACORE. The affinity improvement during the directed evolution was monitored mainly by competition RIA (see text)d Clone C11L34 was found to have a single point mutation compared with its likely progenitor (11Hanes J. Jermutus L. Weber-Bornhauser S. Bosshard H.R. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 1998; 95: 14130-14135Google Scholar). All other clones are derived from C11L34 and contain its mutation, L42(Asn → Ser). For historical reasons, its name is indicated in the Kabat nomenclature (36Kabat E.A. Wu T.T. Perry H.M. Gottesmann K.S. Foeller C. Sequences of Proteins of Immunological Interest.5th ed. National Institutes of Health Publication No. 91-3242, United States Department of Health and Human Services, Bethesda, MD1991Google Scholar) and was retainede ND, not determined Open table in a new tab To enhance the mutation rate of the polymerases, we used high concentrations of the nucleotide analogues 8-oxo-dGTP and dPTP leading to mismatch incorporation (7Zaccolo M. Gherardi E. J. Mol. Biol. 1999; 285: 775-783Google Scholar). The final mutation rates of the pools were determined as 8.9 kbp-1 for pool S, 61 kbp-1 for pool R, and 78 kbp-1 for pool SR. Taking into account that many mutations are silent or have no effect on binding, e.g. because they are located in the linker or the lower framework, we estimated that the randomized pools will have two to three relevant amino acid substitutions per gene. As a consequence of the PCR-based randomization we found that mutations that were introduced during an early PCR cycle were more prominent in the final library. This clustering of mutations could be circumvented in future experiments by the use of a high template concentration and very high error rates, thereby reaching the final mutation rate within fewer amplification cycles. Ribosome Display and Off-rate Selection—For ribosome display, all three libraries had to be fused to a protein spacer derived from gene III to allow the displayed protein to fold properly (15Jermutus L. Honegger A. Schwesinger F. Hanes J. Plückthun A. Proc. Natl. Acad. Sci. U. S. A. 2001; 98: 75-80Google Scholar). Since panning followed by extensive washing would hardly be able to discriminate between binders with different affinities all lying in the picomolar range, we used a competitive selection for decreased off-rates. The ribosomal complexes formed after in vitro translation were equilibrated overnight with a 1 nm solution of biotinylated antigen. A 1000-fold excess of competitor antigen carrying no biotin label was then added, and the pools were incubated at 4 °C. All complexes dissociating from the biotinylated antigen, to which they were initially bound, will be captured by the competitor carrying no biotin label and thus cannot be bound to streptavidin-coated magnetic beads. Hence the duration of incubation with competitor is defining the stringency of the selection. The incubation was prolonged from round to round (2, 10, and 240 h), thereby increasing the selection pressure. After every round, selected mRNA was isolated and reverse transcribed, and the enriched pools were subjected to DNA shuffling. The pool SR was further randomized after the second round using the same conditions as in round 1. Interestingly the mRNA of pool R, which had been generated by error-prone amplification of C11L34 only but which had not been subjected to DNA shuffling, could not be restored after the first round of selection. This indicates the importance of recombination in conjunction with high mutation rates to preserve a fraction of the pool in an active form. Furthermore the long off-rate selection times underscore the stability of the non-covalent ribosomal complex, which can survive more than 20 days at 4 °C. Analysis of the Pools after Off-rate Selection—After every round of ribosome display, the pools were checked for improved binding by RIA (18Auf der Maur A. Zahnd C. Fischer F. Spinelli S. Honegger A. Cambillau C. Escher D. Plückthun A. Barberis A. J. Biol. Chem. 2002; 277: 45075-45085Google Scholar). The pools were expressed in vitro in the presence of [35S]Met, equilibrated with different amounts of free antigen, and allowed to bind to surface-immobilized antigen. The amount of competitor antigen needed to inhibit the binding of the scFv fragments to surface-immobilized antigen correlates with the mean affinity of binders found in the pool under investigation and decreased from round to round (Fig. 2). In the initial error-prone randomized pool even high concentrations of competitor did not affect binding of the pool to the plate. After 240 h of off-rate selection, however, 0.1 nm antigen was sufficient to prevent 50% of the pool from binding to the surface compared with the uninhibited signal, giving evidence that the mean affinity of the binders in the pool had improved affinities compared with clone C11L34. The total signal intensity of the pools decreased from round to round, indicating that the percentage of rescued binders decreased from round to round due to the very stringent selection pressure. Screening for Binders—After the third round, pools SR and S were cloned, plasmids of single colonies were isolated, and in vitro expressed protein was analyzed by RIA. Only 7 of 54 clones (13%) showed a significant binding signal to the surface-immobilized antigen in the absence of competitor. All of these clones could be completely inhibited with 10 nm antigen. Four of them were inhibited at even lower concentrations than C11L34 and were therefore ranked as affinity improved. In pool S, which was generated by DNA shuffling, 4 of 15 (25%) clones analyzed showed binding to the antigen after the third round of directed evolution, but only one had an improved inhibition signal (Table II). Only a few molecules had reached improved affinities that let them survive the applied selection pressure, whereas the background signal, consisting of unspecific complexes and RNA sticking to the streptavidin-coated magnetic beads, remained constant. To improve the ratio of binders over background, a nonstringent enrichment round was performed. After this enrichment, 14 of 16 (87%) randomly picked clones showed binding to the antigen, of which eight showed improved inhibition patterns compared with C11L34. Furthermore the signal intensity of the pool in the RIA increased by a factor of 120, indicating a strong enrichment of the binders in the pool over the nonspecific RNA. Thus, a non-selective enrichment round after extensive off-rate selection may be useful in general to amplify the selected clones. Sequences of the Clones with Improved RIA Signal—The clones showing the most promising RIA signal were sequenced. They carried an average mutational load of one to four amino acid substitutions, whereas zero to two mutations derived from shuffled input DNA (Table II). The mutations were distributed over both domains, and some mutations showed up several times. It is likely that they were found independently since they had different codon usage. Interestingly the only mutation lying in CDRs, L135(Asn → Asp), originated from the clones used for DNA shuffling. All other mutations were located in framework positions. BIACORE Measurements—The affinity of all clones generated by site-directed mutagenesis and used for the library construction was measured. In addition, the binding constant of the evolved clone showing the best RIA signal after off-rate selection was determined. All clones were expressed in the periplasm of E. coli and purified by immobilized metal ion affinity chromatography and antigen affinity chromatography (18Auf der Maur A. Zahnd C. Fischer F. Spinelli S. Honegger A. Cambillau C. Escher D. Plückthun A. Barberis A. J. Biol. Chem. 2002; 277: 45075-45085Google Scholar). The dissociation constant KD of the purified proteins was determined in solution by competitio
Referência(s)