Improved affinity of engineered streptavidin for the Strep-tag II peptide is due to a fixed open conformation of the lid-like loop at the binding site

Abstract

The Strep-tag II is a nine-amino acid peptide that was developed as an affinity tool for the purification of corresponding fusion proteins on streptavidin columns. The peptide recognizes the same pocket of streptavidin where the natural ligand is normally bound so that biotin or its chemical derivatives can be used for competitive elution. We report here the crystal structures of the streptavidin mutants `1' and `2,' which had been engineered for 10-fold higher affinity towards the Strep-tag II. Both streptavidin mutants carry mutations at positions 44, 45, and 47, that is, in a flexible loop region close to the binding site. The crystal structures of the two apo-proteins and their complexes with the Strep-tag II peptide were refined at resolutions below 2 Å. Both in the presence and absence of the peptide, the lid-like loop next to the ligand pocket—comprising residues 45 through 52—adopts an `open' conformation in all four subunits within the asymmetric unit. The same loop was previously described to be disordered in the wild-type apo-streptavidin and to close over the pocket upon complexation of the natural ligand biotin. Our findings suggest that stabilization of the `open' loop conformation in the absence of a ligand abolishes the need for conformational rearrangement prior to the docking of the voluminous peptide. Because no direct contacts between the flexible part of the loop and the peptide ligand were detected, it seems likely that the higher affinity of the two streptavidin mutants for the Strep-tag II is caused by a predominantly entropic mechanism.

The development of protein purification protocols that utilize affinity tags has been an important factor contributing to the recent increase in the number of protein structures determined by X-ray crystallography or NMR techniques. The oligo-His-tag is widely used in conjunction with metal chelate resins (Hochuli 1990). The Strep-tag was developed as an alternative tool (Schmidt and Skerra 1993, 1994) and has been successfully applied in our own and numerous other laboratories (for a review, see Skerra and Schmidt 2000).

The Strep-tag is a nine-amino acid peptide (sequence: AWRHPQFGG) with high specificity and affinity towards the protein reagent streptavidin. Its sequence was derived by selection from a genetic peptide library (Schmidt and Skerra 1993) and it was found to occupy the same pocket of streptavidin where biotin otherwise becomes complexed (Schmidt et al. 1996). Hence, a corresponding fusion protein can be eluted under gentle buffer conditions—which may be optimized for a given polypeptide and can even contain additives like thiol compounds or detergents—from an affinity column with immobilized streptavidin, just by applying a low concentration of biotin or related compounds as competitor (Skerra and Schmidt 2000). This method usually permits the efficient purification of a recombinant protein in a single step. Another advantage of the Strep-tag is its independence from metal ions in the purification, an aspect often of interest when metallo-proteins are to be studied or when paramagnetic impurities must be avoided for NMR. In addition, the His-tag has a tendency to cause precipitation of proteins after elution from the metal chelate column due to aggregation induced by heavy metal ions.

The original Strep-tag had to be fused to the C-terminus of the recombinant protein in order to maintain a free peptide carboxylate group, giving rise to a salt bridge with an Arg side chain of streptavidin (Schmidt et al. 1996). To also allow for application at the N-terminus of the protein to be purified or even when inserted amid the polypeptide chain, the modified Strep-tag II (sequence: NWSHPQFEK) was subsequently engineered (Schmidt et al. 1996). One initial drawback of the Strep-tag II was its lower affinity towards streptavidin (with a K_D value of approximately 13 μM; Voss and Skerra 1997). Consequently, streptavidin itself was engineered for improved binding of the peptide by preparing a genetic library with the randomized amino acid positions 44 through 47. These residues were found in the vicinity of the C-terminus of the Strep-tag II in the crystallized complex with streptavidin (Schmidt et al. 1996). With the use of a colony screening assay, the streptavidin mutants `1' and `2' (referred to herein as SAm1 and SAm2, respectively) were selected and shown to possess a significantly higher affinity towards the Strep-tag II, with K_D values close to 1 μM (Voss and Skerra 1997).

To investigate the structural reasons underlying the enhanced performance of this generally useful biochemical tool (Skerra and Schmidt 2000), we pursued crystallization of the engineered streptavidin mutants. Here we report the results of the X-ray crystallographic analysis of both apo-proteins and their complexes with the Strep-tag II peptide. The structures will be compared to those of the wild-type streptavidin with the bound Strep-tag II as well as to the natural streptavidin/biotin complex.

Results

Crystal structure determination and quality of the final model

Crystals were obtained at physiological pH, belonging to the space group P2₁ for the two apo-proteins and their peptide complexes (Table 1). Complete X-ray diffraction data sets were collected from single crystals and directly refined (see Materials and Methods) using atom coordinates from the isomorphous structure of the crystallized streptavidin mutant Tyr⁴³→Phe described by Freitag et al. (1999; PDB accession code 1swu) as a starting model. The mutated loop region was manually remodeled to fit the new electron density, and the refinement was extended to resolutions below 1.9 Å in all cases, with good stereochemistry and reaching final R-factors below 20% (Table 2).

Table 1.

Data collection statistics for crystals of streptavidin mutants `1' and `2'

Mutant	apo-SAm1	SAm1•StrepII	apo-SAm2	SAm2•StrepII
Protein	streptavidin mutant `1'	streptavidin mutant `1' in complex with Strep-tag II peptide	streptavidin mutant `2'	streptavidin mutant `2' in complex with Strep-tag II peptide
Crystallization conditions	1.2 M (NH4)2SO4, 100 mM Na2HPO4, pH 7.3	1.3 M (NH4)2SO4, 100 mM Na2HPO4, pH 6.9	50 mH Na2HPO4, 50 mM Na-Citrate, 30% PEG 600, pH 7.3	1.2 M (NH4)2SO4, 100 mM Na2HPO4, pH 7.5
Space group	P21	P21	P21	P21
a, b, c	58.42 Å, 86.92 Å, 47.20 Å	58.70 Å, 86.34 Å, 47.45 Å	58.45 Å, 86.81 Å, 47.08 Å	58.39 Å, 86.51 Å, 47.24 Å
β	98.90°	98.92°	98.82°	99.10°
Resolution	1.9 Å	1.7 Å	1.7 Å	1.8 Å
No. of total reflections	52768	126949	178009	120275
No. of unique reflections	33067	49002	48419	41071
Completeness	89.2% (99.00 − 1.90 Å)	94.8% (99.00 − 1.70 Å)	93.8% (99.00 − 1.70 Å)	94.5% (99.00 − 1.80 Å)
	85.5% (1.97 − 1.90 Å)	90.3% (1.76 − 1.70 Å)	89.7% (1.76 − 1.70 Å)	93.2% (1.86 − 1.80 Å)
Rmerge	0.05	0.04	0.07	0.04

Table 2.

Refinement statistics for streptavidin mutants `1' and `2'

Mutant	apo-SAm1	SAm1•StrepII	apo-SAm2	SAm2•StrepII
R-factor/RFree	17.0/22.9	17.5/21.7	19.1/22.8	17.8/22.1
Protein residues	478	480	478	480
Protein residues visible	A16–A135	A16–A136	A16–A135	A16–A136
	B15–B134	B15–B134	B15–B134	B15–B134
	C16–C134	C15–C134	C16–C134	C15–C134
	D16–D134	D16–D134	D16–D134	D16–D134
Ligand residues	—	24	—	22
Ligand residues visible		E3–E8		E3–E7
		F4–F9		F4–F8
		G4–G9		G4–G9
		H4–H9		H4–H9
No. of solvent molecules	212	240	195	167
R.m.s. deviations from ideality of bond lengths	0.016 Å	0.018 Å	0.018 Å	0.017 Å
R.m.s. deviations from ideality of bond angles	2.868°	2.773°	2.548°	2.679°
Average B-values:
whole molecule	27.5 Å2	27.3 Å2	25.1 Å2	24.4 Å2
residues 46–51	48.6 Å2	43.2 Å2	47.4 Å2	41.8 Å2
peptide ligand	—	33.1 Å2	—	35.1 Å2
PDB accession code	1kff	1kl3	1kl4	1kl5

The resulting structural models of the streptavidin mutants `1' and `2' are made up of one tetramer in the asymmetric unit with 119 to 120 amino acid residues per polypeptide chain, depending on the subunit (full length of the recombinant core streptavidin: 127 residues; Schmidt and Skerra 1994). The binding-site loop itself (residues 45 through 52; for mutations see Table 3) appears well ordered in all four subunits of the two mutants, both in the presence and absence of the ligand (Fig. 1 ▶). In contrast, the same loop was found to be disordered in a large number of streptavidin structures deposited in the PDB (for a compilation see Table 4). In the electron densities from the cocrystals with the Strep-tag II, at least 5 amino acids of the peptide ligand (residues 4 through 8) are clearly visible in all four subunits (Table 2, Fig. 1B ▶) and were thus included in the structural model. Lys⁹ could only be placed in some of the subunits (Table 2) and usually showed weak electron density for its side chain. Spurious density for the N-terminal residues of the peptide ligand could not be fitted unambiguously.

Table 3.

Amino acid sequences of the binding-site loop of wild-type streptavidin and its mutants

Residue	44a	45	46	47
wtSA	Glu	Ser	Ala	Val
SAm1b	Val	Thr	Ala	Arg
SAm2b	Ile	Gly	Ala	Arg

^a Numbering according to the mature full length protein (cf. Schmidt and Skerra 1994).

^b Voss and Skerra (1997).

Fig. 1.

(A) 2F_o-F_c difference Fourier electron density of the binding-site loop in apo-SAm1 (subunit A) contoured at 1 σ. The binding-site loop is well ordered in the absence of the peptide ligand. (B) 2F_o-F_c difference Fourier electron density of the bound peptide in SAm1•StrepII (subunit A) contoured at 1 σ. Apart from the flanking residues, that is, Asn¹ and Trp² as well as Lys⁹, the Strep-tag II peptide is ordered and placed unambiguously in the electron density.

Table 4.

Conformations of the binding-site loop in streptavidin structures from this study and as deposited at the PDB

	Subunit
Structurea	A	B	C	D	Bound ligand	pH	Mutations
apo-SAm1	Ob	O	O	O	—	7.3	E44V/S45T/V47R
SAm1•StrepII	O	O	O	O	Strep-tag II peptide	6.9	E44V/S45T/V47R
apo-SAm2	O	O	O	O	—	7.3	E44I/S45G/V47R
SAm2•StrepII	O	O	O	O	Strep-tag II peptide	7.5	E44I/S45G/V47R
1df8	C				biotin	4.5	S45A
1pts	C	D			FSHPQNT	na
1rst		O			Strep-tag peptide	7.0
1rsu		O			Strep-tag II peptide	7.0
1sld		O			Cyclo-Ac-CHPQFC-NH2	7.5
1sle		O		O	Ac-CHPQGPPC-NH2	5.0
1slf		C		C	2 • (NH4)2SO4	5.6
1slg		O		O	Ac-FCHPQNT	5.6
1sre	C	C			HABA	na
1srf	C	C			3-t-Butyl-HABA	na
1srg	C	C			3-Methyl-HABA	na
1srh	C	C			3`5`-Dimethoxy-HABA	na
1sri	C	C			3`5`-Dimethyl-HABA	na
1srj	C	C			Naphtyl-HABA	na
1stp	C				biotin	7.8
1str	O	O			Ac-CHPQNT-NH2	7.0
1sts	O	O			FCHPQNT-NH2	7.0
1swa	C	D(O)	D(O)	D(O)	—	4.5
1swb	C	D(O)	D(O)	D(O)	—	7.5
1swc	D(O)	O	D(O)	O	—	4.5
1swd	C	D(O)	D(O)	C	biotin	4.5
1swe	C	C	C	C	biotin	4.5
1swh	D	D	D(O)	D(O)	—	4.5	W79F
1swj	D(O)	C	O	D(O)	—	4.5	W79F
1swk	C	C	C	C	biotin	4.5	W79F
1swl	C	D(O)	D	D	—	7.0	W108F
1swn	C	D(O)	C	C	biotin	7.0	W108F
1swo	C	D(O)	D(O)	D(O)	—	7.5	W120F
1swp	C	C	C	C	biotin	7.5	W120F
1swq	D	D(O)	D	D(O)	—	7.5	W120A
1swr	C	C	C	C	biotin	7.5	W120A
1sws	D(O)	O	D(O)	D(O)	—	4.5	D128A
1swt	O	O			biotin	4.5	D128A
1swu	C	D(O)	D(O)	D(O)	—	4.5	Y43F
1vwa		O		O	FSHPQNT	4.5
1vwb		O			Cyclo-Ac-CHPQFC-NH2	11.8
1vwc		O			Cyclo-Ac-CHPQFC-NH2	2.0
1vwd		O			Cyclo-Ac-CHPQFC-NH2	3.0
1vwe		O			Cyclo-Ac-CHPQFC-NH2	3.6
1vwf		O			Cyclo-Ac-CHPQGPPC-NH2	3.7
1vwg		O			(Cyclo-Ac-CHPQGPPC-NH2)2	2.5
1vwh		O			(Cyclo-Ac-CHPQGPPC-NH2)2	3.5
1vwi		O		O	Cyclo-[5S-Valeramido-HPQGPPC]K-NH2	1.5
1vwj		O		O	Cyclo-[5S-Valeramido-HPQGPPC]K-NH2	2.5
1vwk		O		O	Cyclo-[5S-Valeramido-HPQGPPC]K-NH2	3.0
1vw1		O		O	Cyclo-[5S-Valeramido-HPQGPPC]K-NH2	3.5
1vwm		O			Cyclo-Ac-CHPQFC-NH2	4.2
1vwn		O			Cyclo-Ac-CHPQFC-NH2	4.8
1vwo		O			Cyclo-Ac-CHPQGPPC-NH2	2.9
1vwp		O			Cyclo-Ac-CHPQGPPC-NH2	2.5
1vwq		O			Cyclo-[5S-Valeramido-HPQGPPC]K-NH2	2.5
1vwr		O			Cyclo-[5S-Valeramido-HPQGPPC]K-NH2	3.5
2iza	C				—	2.0
2izb	C				—	3.1
2izc	C				—	2.0
2izd	C				—	3.0
2ize	C				—	3.1
2izf		C		C	biotin	4.0
2izg		C		C	biotin	2.0
2izh		C		C	biotin	10.4
2izi		C		C	biotin	2.5
2izj		C		C	biotin	3.5
2izk		C		C	glycoluril	2.6
2izl		C		C	2-iminobiotin	7.3
2rta		C		C	—	3.0
2rtb		C		C	—	3.3
2rtc		C		C	—	3.6
2rtd		C		C	biotin	1.4
2rte		C		C	biotin	1.9
2rtf		C		C	biotin	2.0
2rtg		C		C	biotin	2.4
2rth		C		C	glycoluril	2.5
2rti		C		C	glycoluril	2.5
2rtj		C		C	glycoluril	2.5
2rtk		C		C	glycoluril	2.6
2rtl		C		C	2-iminobiotin-sulfate	2.5
2rtm		C		C	2-iminobiotin-sulfate	3.5
2rtn		C		C	2-iminobiotin	2.0
2rto		C		C	2-iminobiotin	2.6
2rtp		C		C	2-iminobiotin	3.3
2rtq		C		C	2-iminobiotin	3.3
2rtr		C		C	2-iminobiotin	4.0

^a For authors and references of the crystal structures, cf. the PDB deposition.

^b O, `open' loop conformation; C, `closed' loop conformation; D, disordered loop conformation; D(O), disordered loop conformation, but positioning of neighboring residues indicates preference for the `open' loop conformation.

Table 2 shows refinement statistics for the final models of the proteins in their apo-form and in complex with the Strep-tag II peptide. In a Ramachandran plot (Ramachandran and Sasisekharan 1968), no main chain dihedral angles of nonglycine residues lie outside the allowed regions (not shown). After a pairwise least squares superposition of C_α-atoms (LSQKAB; ^{CCP4 1994}), mutual r.m.s. deviations of the subunits within the tetramer lie in a range between 0.23 and 0.30 Å. Average B-values for the whole streptavidin subunits, their binding-site loops, and the peptide ligands are given in Table 2.

Overall topology

The C_α-atoms of the corresponding A subunits of SAm1•StrepII and of the conventional streptavidin•biotin complex (1swu; Freitag et al. 1999) can be superimposed with an overall r.m.s.d. of 0.52 Å (Fig. 2 ▶). The largest deviations in C_α-positions occur within the binding-site loop (up to more than 13 Å), which is `closed' in wtSA•biotin but `open' in our mutant structures. Further deviations larger then 0.5 Å can be seen at residue 26, which is part of the ligand-binding pocket, at residues 66 and 100, which are involved in crystal and tetramer contacts, respectively, and at the C-terminus of the polypeptide chain. The subunit structures of the streptavidin mutant `1' in the presence and absence of the peptide ligand can be superimposed with a significantly smaller C_α r.m.s.d. of 0.32 Å (Fig. 3A ▶).

Fig. 2.

Mutual distances of C_α-positions after pairwise superposition of the A chains from the tetrameric streptavidin crystal structures: SAm1•StrepII versus wtSA•biotin (1swu; Freitag et al. 1999), SAm1•StrepII versus wtSA•StrepII (1rsu; Schmidt et al. 1996), SAm1•StrepII versus apo-SAm1, apo-SAm2 versus apo-SAm1, and apo-SAm2 versus SAm2•StrepII.

Fig. 3.

(A) Superposition of the C_α-backbones from the A chains of apo-SAm1 (blue) with SAm1•StrepII (red) and with the wtSA•StrepII complex (green, 1rsu). The residues that were mutated for the generation of SAm1 (44 to 47) are marked with small dots (green). Residues flanking this segment and other positions with large mutual deviations (cf. Fig. 2 ▶) are labeled. The deviations in positions 67, 100, and 120 are due to tetramer or crystal contacts. The binding-site loop assumes an `open' conformation in all structures, but has undergone partial rearrangement at the mutated positions with respect to wild-type streptavidin. (B) Superimposed binding sites of SAm1•StrepII (red) and wtSA•StrepII (green). The salt bridge between residues Arg⁸⁴ of streptavidin and Glu⁸ of the peptide ligand has been proposed to be crucial for the complexation of the Strep-tag II by streptavidin (Schmidt et al. 1996). This salt bridge has shifted as a whole by 3 Å in SAm1•StrepII but remains intact. Residue Lys⁹ of the Strep-tag II peptide, as deposited in 1rsu (Schmidt et al. 1996), was omitted in both graphs because it exhibits only weak electron density and is probably disordered.

Interaction with the peptide ligand

When compared with the structure of wild-type streptavidin in complex with the Strep-tag II (PDB accession code 1rsu; Schmidt et al. 1996), the conformations of the peptide ligands are very similar (Fig. 3 ▶). C_α-atoms of the Strep-tag II complexed with SAm1 and of the same peptide ligand complexed with wtSA show an overall r.m.s.d. of 0.06 Å when directly superimposed.

In the crystal structure of wtSA•StrepII, the side chain of Lys⁹ from the peptide protrudes into the hydrophobic cleft formed by the side chain Pro⁴ of the ligand together with the side chains Tyr⁴³, Tyr⁵⁴, and Trp⁷⁹ of streptavidin. However, reinspection of the electron density, calculated from the structure factors deposited along with 1rsu, revealed rarely significant density for this side chain (data not shown). Lys⁹ is at least partially disordered both in SAm1•StrepII and SAm2•StrepII, but 2F_o-F_c difference Fourier maps indicate that here its side chain should point into the solvent without specific contacts to protein residues. Lys⁹ was previously shown to significantly contribute to the binding of the Strep-tag II to streptavidin when covalently immobilized via its C-terminus during peptide screening experiments (Schmidt et al. 1996). However, the weak electron densities found for this residue do not permit the identification of a relevant interaction. The situation during complex formation with a C-terminally fused Strep-tag II may therefore be different from the one observed with the free peptide in our crystal structure.

Similarly, the electron density for the side chain of Trp², which was found to be an important residue in the initial selection of the Strep-tag from a random peptide library (Schmidt and Skerra 1993), was rather weak. Nevertheless, both residues may exert an effect on the conformation of the peptide in solution and thus indirectly contribute to the apparent strength of binding (see Discussion). Finally, side chains of Glu⁸ from the ligand and Arg⁸⁴ from streptavidin have been displaced by about 3 Å in the SAm1•StrepII structure with respect to wtSA•StrepII (Fig. 3B ▶). The salt bridge between these residues has been proposed to be crucial for the binding of the Strep-tag II to streptavidin, particularly when used as an N-terminal fusion peptide (Schmidt et al. 1996), and is still intact.

Structure of the binding site

The backbone structure of the binding-site loop comprising residues 45 through 52 is almost identical between the mutants `1' and `2', both in the presence and absence of the ligand (Fig. 2 ▶). Only marginal side chain conformational changes can be observed between SAm1 and SAm2 or upon binding of the Strep-tag II (Fig. 4 ▶). In the crystallographic unit cell, the binding-site loop is located close to packing contacts. However, other crystal structures of streptavidin have been determined in the same space group (e.g., 1swa and 1swb, Freitag et al. 1997; Table 4), where the binding-site loop is still disordered or even in the `closed' conformation. Furthermore, since the tetramer subunits are not related by crystallographic symmetry, all four loops form different contacts with their neighborhood. In subunit C, the closest distance between any of the residues from Val<sup>C44</sup> to Arg<sup>C53</sup> and a symmetry-related molecule is 5.0 Å (i.e., between Arg<sup>C53</sup> and Lys<sup>B134*</sup>). Since noncrystallographic symmetry restraints were also not used during the refinement process, crystal contacts alone cannot be made responsible for the consistently ordered conformation of the loop in the `open' state.

Fig. 4.

(A) Superposition of the ligand-binding sites of SAm1•StrepII (red) and wtSA•StrepII (green, 1rsu). The bound peptide ligands are depicted in corresponding light colors. The labeling of amino acid residues refers to SAm1•StrepII. Partial movement of the binding-site loop towards the hydrophobic cleft lined by residues Tyr⁴³, Tyr⁵⁴, and Trp⁷⁹ as well as Pro⁵ from the peptide ligand results in a shift of Thr⁴⁵ by 1.9 Å. (B) Superposition of the ligand-binding sites of SAm1•StrepII (red) and apo-SAm1 (blue). The bound peptide ligand of SAm1•StrepII is shown in light red. The conformation of the binding-site loop (residues 45 through 52) of SAm1 is nearly the same in the presence and absence of the peptide ligand. The two binding-site loops show an r.m.s.d. of 0.3 Å following C_α superposition of the protomers. (C) Superposition of the ligand-binding sites of apo-SAm1 (blue) and apo-SAm2 (cyan). The conformation of the binding-site loop is almost identical between the two mutants, both in the absence and presence of the peptide ligand (cf. Fig. 2 ▶). (D) Superposition of binding-site loops in wtSA•biotin (green, 1swd) and SAm1•StrepII (red). The biotin ligand of wild-type streptavidin is shown as ball and sticks (light green), whereas the peptide ligand in SAm1•StrepII is depicted as C_α trace (light red). Unlike Val⁴⁷ in wild-type streptavidin, Arg⁴⁷ in SAm1 cannot participate in a hydrophobic interaction with Leu²⁵, and hence the loop is sterically forced into an `open' conformation. The side chain of Arg⁴⁷ itself is partially disordered (cf. Fig. 1A ▶).

When compared with wtSA•StrepII, most of the loop exhibits precisely the same conformation in the SAm1•StrepII structure (Figs. 3A and 4A ▶ ▶). However, the mutated residues 45 to 47 at its N-terminal side show a local deviation. In particular, Thr⁴⁵ has moved by 1.9 Å into the hydrophobic cleft lined by residues Tyr⁴³, Tyr⁵⁴, and Trp⁷⁹, together with Pro⁵ from the peptide ligand. Residues Ala⁴⁶ and Arg⁴⁷ from the binding-site loop are displaced into the same direction. The side chain of Arg⁴⁷ is partially disordered (Fig. 1A ▶). When comparing the structure of apo-SAm1 with that of SAm1•StrepII or even with that of apo-SAm2, the conformation of the binding-site loop is virtually indistinguishable (Fig. 4B,C ▶). Thus, it appears to be fixed in the two streptavidin mutants.

Discussion

Even though the amino acids at the positions 44, 45, and 47 of streptavidin (see Table 3) resulted from a selection experiment for improved binding of the Strep-tag II, these residues do not make structural contacts to the bound peptide. Thr⁴⁵, which replaces a Ser residue in the streptavidin mutant `1', may be involved in a weak hydrophobic interaction with the side chain of Pro<sup>5</sup> from the Strep-tag II (with a 4.2 Å distance between C<sub>β</sub> Pro<sup>5</sup> and C<sub>γ</sub> Thr<sup>45</sup>). However, the same position is occupied by a Gly residue in the streptavidin mutant `2', which possesses essentially the same ligand affinity and loop conformation. Consequently, direct contributions of the mutated side chains to the enhanced binding activity for the peptide are probably not involved.

When the crystal structure of the complex between streptavidin and biotin was first described (Hendrickson et al. 1989; Weber et al. 1989), the binding-site loop was found to be closed almost like a lid over the pocket harboring the small ligand. In contrast, the same loop was seen disordered in the crystal structure of apo-streptavidin (Weber et al. 1989). Therefore, this structural mechanism was postulated to play a role for the remarkably high affinity observed in this system (Kempner 1993). Alternative conformations of loops that are involved in substrate or ligand binding have frequently been observed elsewhere, for example, in the cases of rhamnose isomerase (Korndörfer et al. 2000) and triose-phosphate isomerase (Alber et al. 1987). In these structures, the loop is usually disordered in the apo-protein. Upon binding of the ligand the loop becomes ordered, closes the active site, and shields the ligand from solvent access (for a general review of flexible loops, see Kempner 1993).

Indeed, the binding-site loop (residues 45 through 52) is the conformationally most variable motif in the 83 different structures of streptavidin or its mutants deposited in the PDB (see Table 4). In an attempt to classify the various conformations of this loop, so-called `open' and `closed' states have been distinguished (Weber et al. 1989). As in the protein structures mentioned above, in the absence of a ligand the binding-site loop of streptavidin has been proposed to generally adopt an `open' conformation (Freitag et al. 1997). Still, in the structures deposited it can be seen that the loop may adopt `open' and `closed' conformations even within the same crystal structure (e.g., 1swa and 1swj; Freitag et al. 1997), although in the absence of a ligand it is mostly disordered (Table 4).

In the unique apo-streptavidin structure where the loop is ordered and in the `closed' conformation in all subunits (1slf; Katz 1995), two sulfate ions are actually localized in the binding pocket. Another series of apo-structures where the loop is `closed' has been crystallized at acidic pH (e.g., 2iza and 2izb; Katz 1997), which renders their physiological significance unclear. In the presence of the natural ligand biotin, the binding-site loop always assumes the `closed' conformation (e.g., 1stp; Weber et al. 1989 or 1swd; Freitag et al. 1997), whereas in the presence of a peptide ligand the loop is always `open' (e.g., 1rst; Schmidt et al. 1996). Therefore, the mutants `1' and `2,' which were obtained from a combinatorial mutagenesis experiment (Voss and Skerra 1997), would be the first engineered versions of streptavidin where the binding-site loop is well ordered and exhibits an `open' conformation in all of its four subunits, notably at physiological pH and in the absence of any ligand.

Taken together, the streptavidin mutants `1' and `2' differ from the wild-type core streptavidin in three amino acid positions (Table 3). These substitutions effect a decrease in the K_D for the binding of the Strep-tag II from 13.0 ± 1.3μM to 1.37 ± 0.08 μM and 1.02 ± 0.04μM, respectively (Voss and Skerra 1997). Only the substitution Val⁴⁷→Arg is strictly conserved in both streptavidin mutants. When the loop is closed, the Val⁴⁷ side chain of wild-type streptavidin is positioned such that it faces towards the hydrophobic side chain of Leu²⁵ (Fig. 4D ▶). The side chain of an Arg residue cannot be accommodated in the same orientation due to its larger size and to the highly polar character. Thus, a loop conformation where Arg⁴⁷ is pointing into solution is probably favorable. Arg⁴⁷ is not involved in specific ionic interactions or hydrogen bonds or in crystal packing contacts with symmetry-related molecules; rather, its side chain is disordered.

The `open' loop conformation may furthermore be stabilized by the substitution Glu<sup>44</sup>→Val. This hydrophobic residue might be partially shielded from solvent by the main chain atoms of Gly<sup>48</sup> and the side chain of Arg<sup>53</sup>. In SAm1•StrepII, Val<sup>44</sup> has an average accessible surface area of 9 Å<sup>2</sup> (SURFACE; <sup>CCP4 1994</sup>). A simple substitution of Glu<sup>44</sup> by Val in the crystal structure of 1swb (Freitag et al. 1997), where the loop is in the `closed' conformation, would lead to a much higher hypothetical accessible surface area of 28 Ų for this residue. This effect should be even more pronounced in the mutant `2,' where an Ile residue occupies the corresponding position.

To understand the influence of the loop conformation on the ligand affinity, a thermodynamic consideration may be applied. Even though streptavidin constitutes a stable tetramer in solution, its four ligand-binding sites are independent and structurally remote from each other, including the corresponding flexible loops, so that it may be treated here to behave generally like a monomeric protein. Let us assume a simple association/dissociation equilibrium between such a protein, occurring in an active state A, and its ligand L, characterized by the dissociation constant:

Let us further assume that the active state A corresponds to the wild-type streptavidin monomer with the `open' loop conformation, which participates in a dynamic equilibrium with a population of monomers X that have loop conformations in any other conformation (e.g., `closed') being sterically incompatible with the complexation of the ligand (i.e., the Strep-tag II):

If we cannot distinguish between these two conformational states of the protein by means of physicochemical or spectroscopic methods, as is in fact the case for the system treated here, a ligand-binding experiment may just reveal an `apparent' dissociation constant according to the following Law of Mass equation:

Here P constitutes the sum of the protein in both states: [P] = [A] + [X].

Since [P•L] must equal [A•L] from above, the following equation can be deduced:

This means that the measurable apparent dissociation constant is raised by a factor (1 + K_Conf) compared with the intrinsic K_D of the complexation-competent conformation A (here the `open' loop form of streptavidin) as a result of its equilibrium with the complexation-incompetent conformation X (here the closed-loop or related conformation). If this equilibrium is far on the side of the A state, K_Conf is small and this factor approaches unity; that is, the true K_D will be measured. However, if the equilibrium is on the side of the X state, K_Conf becomes significantly larger than 1 and the apparent K_D is increased to an appreciable extent.

For example, K_Conf = 9 (i.e., just 10% of the free protein is in the active state A) would result in K_D^app = 10 • K_D. This number should roughly reflect the situation for the complex formation between the wild-type streptavidin with its flexible loop and the Strep-tag II peptide, where the apparent dissociation constant was measured to be approximately ten-fold higher than for the two streptavidin mutants with the loop fixed in the `open' conformation. By the way, a similar theoretical consideration would in turn apply to distinct conformational states of the peptide ligand in solution and a corresponding indirect effect on the binding strength. Hence, it appears that the stabilization of a flexible loop in a conformation that is `competent' for the binding of the cognate ligand or substrate would be a general means to increase the affinity for this target via an entropic mechanism.

The same interpretation would be valid for the `induced fit' phenomenon, which is well known in the structural biology of antibody–antigen reactions (Stanfield and Wilson 1994). In this case, some of the six hypervariable loops, predominantly CDR-H3, are found in a different conformation when complexed with the antigen than in the crystal structure of the antibody fragment alone. Again, an equilibrium may be postulated between the two conformations, whereby the latter is preferentially populated in the absence of the antigen. Although it has been speculated that induced fit may be a general requirement for tight antigen binding, the affinity should actually be diminished by this phenomenon according to the argument given above. Of course, it could be the case that there is no alternative solution provided by the immune system that would lead to an equally good structural fit in the absence of an induced conformation. Nevertheless, if some somatic mutation could be acquired during maturation of the immune response that would effect fixation of the loop conformation in a productive manner, this in itself should result in an improved antigen affinity.

In conclusion, this study demonstrates that the introduction of three amino acid substitutions into a flexible loop that is part of the streptavidin binding site lowers its conformational entropy by freezing it in an `open' state. Thus, without making direct contacts to the artificial Strep-tag II ligand, its binding becomes considerably favored. This observation illustrates that it is not sufficient to consider the structural situation in the preformed biomolecular complex alone when attempting to improve molecular recognition.

Materials and methods

Protein crystallization

Recombinant core streptavidin mutants `1' and `2' (for amino acid replacements, see Table 3), comprising residues 13 to 139 of the mature protein, were overexpressed as inclusion bodies in E. coli using the expression vector pSA1 and refolded as described (Schmidt and Skerra 1994; Voss and Skerra 1997). The purified synthetic Strep-tag II peptide (Ac-Ser-Ala-Trp-Ser-His-Pro-Gln-Phe-Glu-Lys-NH₂) was a gift from IBA (Göttingen, Germany). Sparse matrix crystallization screens (Jancarik and Kim 1991) were performed using the hanging-drop vapor diffusion technique (McPherson 1999). Three μL of protein solution (22 mg/mL in 5 mM Tris/HCl, pH 8.0) were mixed on a siliconized glass cover slip with 3 μL of the precipitant and equilibrated against 0.5 mL of the precipitant solution at 20°C. Typically, crystals of the streptavidin mutants were obtained after 4 to 5 days. Following optimization of the crystallization conditions, crystals of streptavidin mutants without the Strep-tag II peptide were obtained using (NH₄)₂SO₄ or PEG 600 as precipitant. For cocrystallization experiments with the Strep-tag II peptide, 2 μL of the protein solution (11.5 mg/mL in 2.5 mM Tris, pH 8.0) were mixed with 1 μL of the peptide (8 mg/mL in H₂O) and 2 μL of the reservoir solution. Precise crystallization conditions for the crystals described in this work are given in Table 1.

Data collection and processing

Crystals were mounted directly from the drop in thin wall glass capillary tubes (Müller, Schönwalde, Germany). Essentially complete X-ray diffraction data sets were collected from single crystals at room temperature on a mar345 imaging plate detector (X-RAY Research, Norderstedt, Germany) with CuK_α radiation from an RU-300 rotating anode generator (Rigaku, Tokyo, Japan) monochromatized and focused through MaxFlux Confocal Multilayer Mirror Optics (Osmic, Troy, MI). Crystal-detector distances were 110 mm or 130 mm. The data were processed and scaled using the programs DENZO and SCALEPACK (Otwinowski and Minor 1997). Reduction to structure factor amplitudes was performed with TRUNCATE (French and Wilson 1978; CCP4 1994). Data collection statistics are given in Table 1.

Model building and refinement

The unit cell dimensions of the streptavidin crystals investigated here (Table 1) matched those of the crystallized streptavidin mutant Y43F described by Freitag et al. (1999) (PDB accession no. 1swu). Consequently, all solvent and ligand molecules were stripped from this model and, starting from these coordinates, rigid body refinement was directly carried out with data to 2.5 Å using the program CNS (Brunger et al. 1998). After 20 steps of rigid-body minimization of the whole streptavidin tetramer and 20 steps of refinement of the four individual chains in the asymmetric unit, the R-factor dropped from 35.2% (for apo-SAm1) and 38.1% (for SAm1•StrepII) to 31.3% and 32.8%, respectively. The loop from residues 45 to 52 was manually rebuilt using QUANTA (Oldfield 2001) in order to fit the 2F_o-F_c and F_o-F_c electron densities, which also confirmed the identity of the mutated side chains. Numbering of residues was adjusted to match that used in previous publications from our lab, thus referring to the mature full-length streptavidin instead of the truncated core protein (Schmidt and Skerra 1994). For apo-SAm2 and SAm2•StrepII, the refined structures of apo-SAm1 and SAm1•StrepII, respectively, were used as starting models. Refinement was continued with REFMAC5 (CCP4 1994; Murshudov et al. 1997). Several sessions of manual rebuilding alternated with further rounds of refinement against all data. Water molecules were positioned where stereochemically plausible and 2F_o-F_c and F_o-F_c difference Fourier maps showed densities of 1.0 and 3.0 σ, respectively. The Free-R factor (Brunger 1992) was used to monitor the refinement and solvent building process. However, it must be kept in mind that its significance is limited here due to the presence of noncrystallographic symmetry in the crystals and to the unavailability of the `free' data set from the 1swu model structure (Freitag et al. 1999) to allow for a corresponding selection. Noncrystallographic symmetry restraints were not used in the refinement process. Table 2 summarizes refinement statistics for the models presented in this work. Illustrations of molecular models and electron densities were prepared with MOLSCRIPT (Kraulis 1991) and MINIMAGE (Arnez 1994). The coordinates and structure factors of all four crystal structures have been deposited at the Protein Data Bank (Berman et al. 2000).

Abbreviations

• KD, equilibrium constant of dissociation

• NMR, nuclear magnetic resonance

• PDB, protein data bank

• r.m.s.d., root mean square deviation

• SAm1, streptavidin mutant `1'

• SAm2, streptavidin mutant `2'

• StrepII, Strep-tag II peptide

• wtSA, recombinant wild-type core streptavidin