Bacterial display using circularly permuted outer membrane protein OmpX yields high affinity peptide ligands

Authors

  • Jeffrey J. Rice,

    1. Department of Chemical Engineering, University of California, Santa Barbara, California 93106, USA
    2. Institute for Collaborative Biotechnologies, University of California, Santa Barbara, California 93106, USA
    Search for more papers by this author
  • Aaron Schohn,

    1. Department of Chemical Engineering, University of California, Santa Barbara, California 93106, USA
    Search for more papers by this author
  • Paul H. Bessette,

    1. Department of Chemical Engineering, University of California, Santa Barbara, California 93106, USA
    2. Institute for Collaborative Biotechnologies, University of California, Santa Barbara, California 93106, USA
    Search for more papers by this author
  • Kevin T. Boulware,

    1. Department of Chemical Engineering, University of California, Santa Barbara, California 93106, USA
    Search for more papers by this author
  • Patrick S. Daugherty

    Corresponding author
    1. Department of Chemical Engineering, University of California, Santa Barbara, California 93106, USA
    2. Institute for Collaborative Biotechnologies, University of California, Santa Barbara, California 93106, USA
    3. Biomolecular Science and Engineering, University of California, Santa Barbara, California 93106, USA
    • Department of Chemical Engineering, University of California, Santa Barbara, CA 93106, USA; fax: (805) 893-4731.
    Search for more papers by this author

Abstract

A bacterial display methodology was developed for N- and C-terminal display and demonstrated to enable rapid screening of very large peptide libraries with high precision and efficiency. To overcome limitations of insertional fusion display libraries, a new scaffold was developed through circular permutation of the Escherichia coli outer membrane protein OmpX that presents both N and C termini on the external cell surface. Circularly permuted OmpX (CPX) display was directly compared to insertional fusion display by screening comparable peptide libraries in each format using magnetic and fluorescence activated cell sorting. CPX display enabled in situ measurement of dissociation rate constants with improved accuracy and, consequently, improved affinity discrimination during screening and ranking of isolated clones. Using streptavidin as a model target, bacterial display yielded the well-characterized HPQ/M motif obtained previously using several alternative peptide display systems, as well as three additional motifs (LI/V CQNVCY, CGWMYF/YxEC, ERCWYVMHWPCNA). Using CPX display, a very high affinity streptavidin-binding peptide was isolated having a dissociation rate constant koff = 0.002sec−1 even after grafting to the C terminus of an unrelated protein. Comparison of individual clones obtained from insertional fusion and terminal fusion libraries suggests that the N-terminal display yields sequences with greater diversity, affinity, and modularity. CPX bacterial display thus provides a highly effective method for screening peptide libraries to rapidly generate ligands with high affinity and specificity.

Molecular recognition is essential in a broad range of scientific and technological applications ranging from proteomics to therapeutic discovery. Consequently, improved methodologies for producing high quality molecular recognition reagents, including antibodies and peptides, have been pursued vigorously since the development of hybridoma monoclonal antibody technology by Köhler and Milstein (1975). In particular, the advent of filamentous bacteriophage display technology (Smith 1985) enabled generation of tailor-made affinity reagents in weeks rather than months. The fundamental principle behind display technologies is a physical linkage of a functional polypeptide to the gene that encodes it. As such, the linkage between polypeptide function and gene enables rapid identification of polypeptides that recognize and bind to an arbitrary target molecule from a collection of billions of unique molecules in a process referred to as “panning.” Despite the development of several alternative display approaches, phage display remains, by far, the most widely used display methodology.

Among display methods, cell surface display (Boder and Wittrup 1997; Georgiou et al. 1997; Shusta et al. 1999; Wittrup 2001; Lee et al. 2003) is attractive since fluorescence activated cell sorting (FACS) can be applied for sensitive, quantitative library analysis and screening (Daugherty et al. 1998; Georgiou 2000). Furthermore, cell display enables simple clonal and library manipulation and propagation and direct measurement of the relative binding constants of isolated clones (Boder and Wittrup 1997; Daugherty et al. 1998; Feldhaus et al. 2003). Cell surface display in conjunction with FACS has enabled antibody epitope mapping (Tripp et al. 2001; Bessette et al. 2004), direct selection of peptides and antibodies from diverse repertoires (Feldhaus et al. 2003; Bessette et al. 2004), T-cell receptor engineering, and antibody affinity maturation into the femtomolar range (Boder et al. 2000). This unprecedented affinity improvement was directly attributed to the combination of polyvalent display and FACS-based screening for slow dissociation rate constants. Recently, a yeast display human antibody library of 109 members was constructed enabling isolation of antibodies binding to a diverse group of antigens (Feldhaus et al. 2003). Thus, cell surface display has proven to be an exciting, alternative display methodology with unique capabilities.

Bacterial cell surface display approaches have been vigorously pursued since bacteria possess short doubling times, allow simple manipulation, and enable construction of libraries containing as many as 5 × 1010 clones (Bessette et al. 2004). However, the general utility of bacterial display for screening peptide and protein libraries is currently limited by the suitability of typical display scaffolds–including outer membrane proteins (Brown 1992), fimbria (FimH) (Klemm and Schembri 2000), and flagella (FLITRX) (Lu et al. 1995). Several different outer membrane proteins have been used for construction of loop insertion libraries, including LamB (Brown 1992), OmpA (Bessette et al. 2004), FhuA (Taschner et al. 2002), and OmpS (Lang 2000). These scaffolds for insertional display have been most useful for antibody and protein epitope mapping experiments (Tripp et al. 2001; James et al. 2003). However, these approaches have yielded less success in the isolation of soluble affinity reagents (Bessette et al. 2004), which typically requires grafting of the peptides into constrained, soluble scaffold proteins. Given these problems, bacterial display systems have not proven generally useful for generating affinity reagents.

The performance of a given display methodology has been evaluated previously by screening peptide libraries with streptavidin as a model target protein (Giebel et al. 1995; Wilson et al. 2001; Lamla and Erdmann 2003). For example, several different phage display studies have yielded streptavidin binding peptides with moderate affinity, ranging from 13 to 70 μM for linear peptides termed Strep-tag (SAWRHPQFGG) (Schmidt and Skerra 1993) and Strep-tag II (SNWSHPQFEK) (Korndorfer and Skerra 2002). The best reported cyclic peptide isolated using phage display (AECHPQGPPCIEGRK) possessed a KD of 0.23 μM (koff = 0.2 sec−1) (Giebel et al. 1995). Thus far, the majority of streptavidin binding peptides isolated from display libraries are dependent upon the presence of the signature HPQ tripeptide (Giebel et al. 1995; Katz 1995). Crystallography studies of streptavidin–peptide complexes subsequently have revealed that HPQ binds to the biotin binding site, through hydrogen bonds and hydrophobic interactions (Katz 1995). Similarly, screening of even very large (1013) mRNA and ribosome display peptide libraries against streptavidin has yielded peptides with HPQ motifs (Giebel et al. 1995; Wilson et al. 2001; Lamla and Erdmann 2003). One peptide motif isolated using ribosome display (M-DVEAWL) exhibited an affinity of 4–240 nM, when presented at the N terminus of a fatty-acid binding protein (Lamla and Erdmann 2003). This study raised the question of whether the intrinsic properties of a display host, other than the scaffold and the library size, influence the spectrum of binding peptides identified. Recently, we identified streptavidin binding peptides using a combinatorial 15-mer library inserted into an extracellular loop of Escherichia coli outer membrane protein OmpA (Bessette et al. 2004). Isolated peptides possessed an unusual CQNVC consensus and exhibited high affinity on the cell surface (KD = 5 nM). However, transfer of these loop peptides to a soluble scaffold resulted in a nearly 100-fold decrease in binding affinity to levels comparable to reported affinities for phage derived peptides (Giebel et al. 1995). Given these limitations, display methods that identify peptides that retain high affinity in the absence of the host and the scaffold used for selection would be of substantial value.

To overcome problems associated with insertional fusion display, we sought to determine whether peptides could be displayed on the E. coli cell surface as terminal fusions to a circularly permuted outer membrane protein. If peptides could be tethered to the cell surface via a flexible linker fused to a single terminus, we hypothesized that their behavior on the cell surface would more closely approximate solution behavior, and, consequently, one could effectively isolate peptide ligands that retain their affinity in the absence of the display scaffold. Here, we demonstrate for the first time that OmpX can be circularly permuted to enable display of “passengers” at either the N or the C terminus. Furthermore, this permuted OmpX scaffold enabled isolation of peptides from large display libraries with affinities surpassing those obtained using insertional fusion libraries against the same target protein.

Results

Circularly permuted OmpX display vectors

To enable N-terminal display on the surface of E. coli, we hypothesized that appropriate circular permutations of outer membrane proteins might retain the necessary information required for membrane insertion, resulting in the presentation of both N and C termini on the cell surface. One ideal candidate for protein display applications is E. coli outer membrane protein OmpX (Mecsas et al. 1995), a small (16.3 kDa), monomeric, β-barrel protein that is highly expressed (Lai et al. 2004). OmpX possesses four extracellular loops, with loops 2 and 3 forming a semirigid β-sheet protruding from the cell surface (Vogt and Schulz 1999; Fernandez et al. 2004; Fig. 1). As a starting point, we determined whether the second and the third extracellular loops of the native OmpX would tolerate peptide insertions. Separately, peptides recognized by an anti-T7 tag (MASMTGGQQMG) antibody or by streptavidin (RLE ICQNVCYYLGTL) (Bessette et al. 2004) were genetically inserted within loop 2 or loop 3, with flexible amino acid linkers flanking the insertion. The araBAD promoter (Guzman et al. 1995) was employed in combination with a low copy plasmid to minimize negative selection pressure (Daugherty et al. 1999). Both the streptavidin binding peptide and T7 tag epitope were efficiently presented on the cell surface as loop 2 insertions between residues S53 and S54, just 30 min after induction of expression, as determined by flow cytometry with a streptavidin R-phycoerythrin (SA-PE) conjugate or anti-T7 tag monoclonal antibody (Fig. 2A). Similarly, peptide insertions within loop 3, between amino acids Y95 and T97 (with P96 deleted), were also surface localized (data not shown). Finally, whole cells displaying streptavidin binding epitopes within OmpX loop 2 bound to the surface of streptavidin-coated microtiter plates at a level fivefold greater than when displayed within OmpA loop 1 under identical conditions (data not shown). These results demonstrate that epitope insertions within OmpX (loops 2 and 3) are efficiently displayed on the E. coli surface in a manner accessible to soluble or immobilized proteins.

Next, circularly permuted variants of OmpX (CPX) were designed with the aim of presenting passengers as fusions to cell surface-exposed termini (Fig. 1). We reasoned that by fusing the native N and C termini, and opening the backbone between nonconserved residues within extracellular loops, both termini would become exposed on the cell surface. For terminal display at S54 (Fig. 1B), the scaffold protein consists of the native OmpX signal sequence, which is cleaved after translocation; a linker sequence GQSGQ (with an embedded SfiI restriction site) after which peptides may be inserted; a second flexible GGQSGQ linker; amino acids S54–F148 of the mature OmpX; a GGSG linker joining the native N and C termini; and finally, amino acids A1–S53 of the mature OmpX. A flexible, four-residue GGSG linker was chosen since this length appeared sufficient to span the distance between the C and N termini without imposing steric constraints (Vogt and Schulz 1999). Thus to display a given epitope at the N terminus, sequences encoding the desired peptides are inserted downstream of the signal sequence and first linker. In this manner, two vectors were constructed wherein the T7 tag epitope or streptavidin-binding peptide were fused to the resultant N terminus of CPX immediately downstream of the first linker. After incubating bacterial cells expressing the CPX constructs with the corresponding fluorescent conjugates, both clones exhibited fluorescence intensities more than 20-fold greater than background autofluorescence, as determined using flow cytometry. However, the extent of labeling of the CPX-displayed epitopes was reduced relative to that using OmpX (Fig. 2A,B). Even so, after a 2-h induction period, the fluorescence labeling of epitope-displaying cells was sufficient to enable library screening using FACS. A second CPX display vector was designed to display epitopes as C-terminal fusions at loop 3 residue Y95. Display of the T7 tag epitope resulted in a more than 10-fold increase in cellular fluorescence over background autofluorescence by FACS analysis (data not shown). All further experiments utilized the S53/S54 permutant.

To investigate whether both N and C termini were simultaneously surface-exposed, the streptavidin binding peptide and T7 tag epitope were genetically fused upstream of S54 and downstream of S53, respectively, within CPX. Cells expressing this fusion protein were capable of binding both streptavidin and T7 tag IgG, as determined using flow cytometry, albeit at a reduced level when compared to individual N- or C-terminal fusions (data not shown). The binding of streptavidin-conjugated quantum dots to cells displaying a streptavidin binding peptide on CPX, upstream of S54, was visualized using fluorescence microscopy (Fig. 3). To aid visualization, bacterial display clones were made to coexpress a green fluorescent protein (Bessette and Daugherty 2004) simultaneously in the cytoplasm. Cells displaying streptavidin binding peptides, but not those expressing CPX scaffold alone, were extensively labeled with quantum dots. Taken together, these results suggest that CPX possesses the topology of Figure 1 and demonstrate that the CPX scaffold is capable of displaying peptides at either the N or the C terminus, or both simultaneously, on the E. coli cell surface.

Library design and screening

The application of N-terminal bacterial display peptide libraries to identify peptide ligands has not been reported previously. For this reason, we sought to directly compare N-terminal and insertional fusion combinatorial peptide libraries within OmpX. Thus, large, random libraries of the form X2CX7CX2, where X is encoded by NNS codons, were constructed in each display format (Fig. 1). The CPX library was constructed as two separate pools (CX71, CX72) as described in Materials and Methods. An additional constrained library of the form X4CX3CX4 was constructed (OmpX3C), since a previous report using a 15-mer random peptide library identified a unique motif (CxNVC) with high affinity for streptavidin (Bessette et al. 2004). All libraries contained 109–1010 independent transformants, a number substantially smaller than the sequence space accessible to peptides with 11 random residues (>1016). Streptavidin binding peptides were enriched using two cycles of magnetic selection (MACS) with streptavidin-coated magnetic beads. Subsequently, the enriched population was screened using two cycles of FACS (Fig. 4A,B). After the first round of FACS, 85% of CPX library cells were highly fluorescent above background when labeled with 5 nM SA-PE (Fig. 4C). Thus, the reduced display level obtained using CPX did not interfere with enrichment of binding peptides. In the final round of FACS, biotin was used as a competitor to favor the detection and sorting of clones with slow dissociation kinetics.

Characterization of streptavidin binding peptides

The majority of clones identified from the OmpX library included an HPQ/M motif (Table 1), as has been observed in selections of phage displayed peptide libraries. However, the CPX library yielded three distinct motifs, including HPQ/M, a unique motif (CGWMYF/YxEC), and the sequence (ERCWYVMHWPCNA). In contrast to the X4CX7CX4 libraries, the X4CX3CX4 library showed a strong preference for the motif LI/VCQNVCY, in agreement with our previous results using a different display scaffold (OmpA) (Bessette et al. 2004). Interestingly, one peptide isolated from the X4CX7CX4 library (OX7-S7 FVCENVCYWVCDN) shared the same consensus (CxNVC) as clones from the X4CX3CX4 library. This sequence contained a third cysteine between the two fixed cysteines, allowing for an alternative disulfide bond to form a three-residue constrained loop. These results indicate that libraries having fixed cysteines to bias loop length do not preclude identification of shorter loop lengths.

The relative affinities of the identified streptavidin binding peptides were estimated by measuring the apparent dissociation rate constant for peptide displaying cells using flow cytometry (Fig. 5A), a well-established and validated method for ranking clonal affinity (Daugherty et al. 1998; Feldhaus et al. 2003). A single clone, CX71-S1, exhibited an unusually slow dissociation rate constant koff = 0.0005 sec−1. The majority of clones, however, exhibited koff values ranging from 0.01 to 0.06 sec−1. Values larger than this could not be measured accurately by this method. Previously identified streptavidin binding peptides, AECHPQGPPCIEGRK (Giebel et al. 1995), and “Strep-tag II” (WSHPQFEK) (Korndorfer and Skerra 2002) were inserted into both OmpX and CPX display scaffolds for comparison with peptides identified in this study. Both peptides remained functional in the bacterial display format and bound SA-PE, whether displayed within OmpX or CPX. However, addition of biotin resulted in rapid dissociation of SA-PE from the cell surface preventing determination of dissociation rate constants using this method for three of four constructs. Peptide AECHPQGPPCIEGRK displayed on bacteria as a constrained insertion into OmpX exhibited a dissociation rate constant of 0.029 sec−1. These results indicate that for identical peptides, insertional fusions result in higher apparent affinities than terminal fusions, and that the best peptide isolated in the present study possesses a slower dissociation rate constant than reported peptides isolated using phage display.

To determine whether the isolated peptides retained binding affinity independent of the CPX scaffold, a subset of the identified peptides were fused, via recombinant methods, to the terminus of a monomeric yellow fluorescent protein (Nguyen and Daugherty 2005) via a GSGS linker. Fluorescent protein–peptide fusions were expressed in E. coli and purified. In order to measure dissociation rates, streptavidin-functionalized microspheres were incubated with purified fusion proteins, washed, and resuspended in the presence of biotin to prevent rebinding. Then, the time-dependent bead fluorescence was measured by flow cytometry (Fig. 5B; Table 2). Consistent with cell surface measurements, the fusion protein corresponding to clone CX71-S1 exhibited the slowest dissociation rate constant (koff = 0.002 sec−1), an unusually slow dissociation rate constant among streptavidin binding peptides (Giebel et al. 1995; Lamla and Erdmann 2003). Fusion proteins containing both CGMWYF/YxEC and HPM motifs exhibited dissociation rates more than 20-fold greater than the best clone, CX71-S1 (Table 2). Three additional fusion proteins exhibited binding to streptavidin in this format, but their dissociation rate constants were >0.1 sec−1 and could not be measured accurately. To verify that the peptide sequences were binding to the biotin binding pocket, streptavidin-phycoerythrin was pre-incubated with an excess of biotin, prior to labeling cells. None of the streptavidin binding clones bound to the streptavidin–biotin complex.

Overall, dissociation rates for soluble, monomeric fusion proteins ranged from two-to fourfold faster than the corresponding values measured for cell surface displayed peptides (Table 2), consistent with the loss of entropic constraints and any possible avidity effects imposed by surface tethering. Even so, the relative affinity ranking of individual clones was equivalent for each method of analysis. The slow dissociation rate of peptide OX7-S1 (YNCCHPMNNLCKE) was not due to the presence of an unpaired cysteine, since deletion of the first cysteine resulted in a nearly identical dissociation rate (data not shown). Peptide OX7-S2 was not functional as either an N- or a C-terminal fusion to YFP, suggesting that this peptide is scaffold dependent. Some fusion proteins appeared to possess an orientation preference since C-terminal YFP fusions of CX72-S2 and CX72-S3 exhibited increased koff values when compared with N-terminal fusions. Most importantly, some peptides were both scaffold and orientation independent since both N- and C-terminal fusion proteins of CX71-S1 yielded identical koff values (data not shown).

In an effort to more directly compare the sequences identified using bacterial display with previous streptavidin binding peptides obtained using phage (Giebel et al. 1995) or ribosome display (Lamla and Erdmann 2003), two additional N-terminal YFP fusions were constructed with the phage peptide ligand AECHPQGPPCIEGRK and ribosome display-derived peptide DVEAWLDERVPLVET. The ribosome-display-derived peptide fusion was not fluorescent and thus was not considered further. However, the phage-derived peptide fused to YFP retained fluorescence and exhibited binding to streptavidin-conjugated microspheres. In agreement with the reported koff of 0.2 sec−1 measured using surface plasmon resonance (Giebel et al. 1995), the dissociation rate was too fast to measure using cytometry. Collectively, these results demonstrate that at least three peptides identified here possess dissociation rates that are slower than those for reported phage peptides and comparable to that for peptides generated using ribosome or mRNA display (Giebel et al. 1995; Wilson et al. 2001; Lamla and Erdmann 2003).

Finally, the specificity of a subset of isolated clones for streptavidin was investigated by measuring the extent of binding to several unrelated target proteins (Fig. 6). All clones were specific for streptavidin, and did not bind appreciably to unrelated proteins at 20-to 50-fold higher concentrations, although weak cross-reactivity for HSA and IgG was observed with a few clones (CX71-S1, OX7-S1, and OX3-S1). Thus, all three consensus groups were target-specific.

Discussion

Here, we demonstrate for the first time that N- and C-terminal bacterial display provide highly effective alternatives to widely used phage display systems for screening peptide libraries. Peptides with affinities at least equivalent to those obtained using phage display were easily generated in just a few days, using either loop insertions within outer membrane protein OmpX or terminal fusions to a circularly permuted variant of OmpX (CPX). To expand the utility of cell surface display in protein and peptide library screening, a new scaffold, termed CPX, was developed to enable display of peptides as N- or C-terminal fusions. When compared to an insertional fusion peptide library, a CPX N-terminal library yielded improved affinity discrimination and somewhat greater diversity among the identified sequences. Both CPX and OmpX libraries yielded streptavidin binding peptides exhibiting strong consensus sequences, including known HPQ/M motifs identified using phage, ribosome, and mRNA display (Giebel et al. 1995; Wilson et al. 2001; Lamla and Erdmann 2003). However, N-terminal display with CPX also yielded an additional group with the consensus CGWMYF/YxEC, as well as a sequence possessing an unusually slow dissociation rate among streptavidin binding peptides (ERCWYVMHWPCNA). Interestingly, this peptide did not share significant identity with other peptides isolated. It is tempting to speculate that that this sequence represents a rare solution to streptavidin binding, reflecting a rugged fitness landscape. Using CPX, all three clones characterized were not dependent upon the display scaffold for binding, since they could be readily grafted to the terminus of a monomeric yellow fluorescent protein without loss of binding. Finally, we attribute the unusual affinity of the best peptide isolated here primarily to the large library employed and the fine affinity discrimination achievable using FACS for library screening.

Previous bacterial display peptide libraries have relied primarily upon use of insertional fusions into outer membrane proteins (OMPs) (Brown 1992; Etz et al. 2001) and fimbrial and flagellar proteins (Lu et al. 1995; Kjaergaard et al. 2000). Probably the most widely used bacterial display system is the FLITRX random 12-mer peptide library, developed by McCoy and coworkers (Lu et al. 1995). With FLITRX display, peptides are presented as constrained insertions into an exposed loop of thioredoxin, which in turn is fused into the major flagellar protein FliC. The FLITRX library has proven useful for mapping antibody epitopes (James et al. 2003) and for identifying cell binding motifs (Brown 2000). Constrained peptides identified using insertional fusion approaches typically exhibit poor affinities in solution, where constraints upon the peptide's termini are absent (Bessette et al. 2004). To overcome this limitation, one could potentially use bacterial display vectors that enable N- or C-terminal display from scaffolds derived from ice nucleation protein (INP) (Jung et al. 1998; Shimazu et al. 2001), EaeA intimin (Wentzel et al. 2001), and EstA (Yang et al. 2004; Becker et al. 2005). These scaffolds may also prove useful for peptide ligand engineering. However, thus far, their use has been restricted to epitope mapping (Christmann et al. 2001; Lu et al. 2003) and identification of cell binding motifs (Taschner et al. 2002).

OmpX possesses several characteristics rendering it attractive as a peptide display scaffold. First, OmpX, the smallest known member of the outer membrane protein family (Koebnik 1996), can be overexpressed rapidly without detrimental effects upon cell viability or growth rate. In fact, high-level display of peptides within OmpX could be detected just 20 min after induction of expression when using the araBAD promoter on a low-copy plasmid with a p15A origin of replication.

For these reasons, OmpX and CPX mediated display did not result in growth biases that interfere with selection and screening (Christmann et al. 1999; Daugherty et al. 1999). Furthermore, OmpX is nonessential, monomeric, and thought to be well distributed throughout the outer membrane (Lai et al. 2004). Extracellular loops 2 and 3 project from the bacterial outer membrane, thereby attenuating steric clashes between large extracellular targets and membrane lipopolysaccharide. Interestingly, loops 2 and 3 of OmpX form a mini-β-sheet that has been observed to mediate cell adhesion or invasion (Vogt and Schulz 1999; Maisnier-Patin et al. 2003), indicating that OmpX/CPX-based libraries may be useful for cell targeting applications.

To our knowledge, CPX is the first example of an outer membrane protein having the N and C termini simultaneously exposed on the cell surface. The circular permutation approach described here may be useful in the development of other surface display vectors. Since a variety of well-expressed, surface-localized proteins in many hosts (phage, viruses, bacteria, yeast, and mammalian cells) possess flexible loops, circular permutation could provide a general approach to achieve N- or C-terminal display within an arbitrary surface exposed loop. Circular permutation of OmpX enabled epitope display at either or both N and C termini generated by opening the backbone within loop 2 or 3. In agreement with an earlier investigation of membrane insertion of a permuted OmpA (Koebnik and Kramer 1995), surface localization of CPX occurred more slowly after induction, relative to OmpX. This result indicates that membrane insertion of CPX is less efficient than that of OmpX, but supports the idea that the essential determinants for secretion and insertion into the outer membrane are maintained within the permuted sequence. Given that the native N and C termini of OmpX were joined by an arbitrarily chosen flexible linker (GGSG), we speculate that alternative linkers and/or point mutations within CPX that stabilize the interaction between the adjacent β-strands might improve expression and/or insertion. We are currently attempting directed evolution of CPX to shorten the time required to achieve sufficient display for screening. Nevertheless, the present CPX display system is clearly effective for screening peptide libraries to identify peptide ligands.

Library diversity can be skewed during numerous rounds of regrowth after selection causing certain clones to outgrow others, as has been observed in some display selections (Christmann et al. 1999; Daugherty et al. 1999). While we cannot formally exclude the possibility that some sequences were de-enriched during amplification, the diversity of isolated peptide sequences (Table 1) suggests that many different clones were selected and amplified at similar rates. Consistent with this idea, many unique high affinity clones were isolated, rather than a small number of dominant clones. This diversity among binding clones enabled identification of several binding motifs. For example, the CPX scaffold yielded three distinct motifs. The HPQ/M motif identified using both insertional and terminal fusion scaffolds has been observed previously using mRNA, ribosome, and phage display. In fact, one of the peptides isolated using CPX (CX72-S9) exhibits six identities (CHPQGP) to a peptide isolated using phage display having high affinity for streptavidin (KD =200 nM) (Giebel et al. 1995). However, two additional motifs were identified having equivalent or improved dissociation kinetics relative to HPQ/M containing peptides. Ribosome display has also yielded noncanonical streptavidin binding motifs when peptides were displayed as fusions to the N terminus of a fatty acid binding protein (Lamla and Erdmann 2003). Similarly, recent application of T7 phage display to isolate streptavidin binding peptides yielded HPQ/M motifs as well as motifs not found using bacterial or M13 display (T. Mori, pers. comm.). Collectively, these results support the idea that different display hosts and scaffolds bias the recovered sequence set and can yield different consensus information.

An important consideration in peptide library screening is whether the sequences identified retain sufficient function independent of the scaffold used for display. Remarkably, the majority of isolated peptides retained binding function as fusions to the N or the C terminus of YFP. The highest affinity sequence (CX71-S1) also exhibited the slowest dissociation rate as a soluble fusion protein. Peptides containing the CGWMYF/YxEC motif, isolated exclusively using the N-terminal CPX library, were functional as N-terminal, but not C-terminal, fusions to YFP. Moreover, two different loop libraries used in this study yielded peptides possessing the CxNVC motif, previously identified using a combinatorial 15-mer loop-displayed peptide library (Bessette et al. 2004). In contrast, N-terminal display with CPX did not yield clones with the CxNVC motif. The fact that this CxNVC motif mediated high affinity binding as an insertion into OmpX or OmpA (Bessette et al. 2004), but not as a terminal fusion to YFP, highlights the scaffold's influence upon the selected sequences. Thus, the mode of display apparently influences the spectrum of sequences identified. And sequences identified in a particular orientation can be dependent upon this orientation while scaffold independent.

Our results suggest that loop insertion and terminal fusion libraries displayed on bacteria can be complementary. Insertional libraries enable enrichment of ligands over a wide range of affinity values. In other words, ligands with low to moderate affinity can be easily detected since the scaffold constraint increases their apparent affinity. Meanwhile, terminal fusion display coupled with FACS enables improved discrimination of clonal affinity in the moderate to high affinity range. More generally, our results further support the idea that no single polypeptide display technology is best for all applications. Rather, with improved understanding of the strengths and weaknesses of existing display technologies, our tool set will evolve to meet the needs of particular applications. We anticipate that N- and C-terminal fusion bacterial display peptide libraries will be especially effective for automated and large-scale affinity reagent development and for affinity maturation and specificity engineering owing to the unique capabilities of existing and emerging cell sorting instrumentation.

Materials and methods

Bacterial strains, reagents, and plasmids

All experiments were performed with E. coli strain MC1061 (F-araD139 Δ(ara-leu)7696 galE15 galK16 Δ(lac)X74 rpsL (StrR) hsdR2 (rK – mK +) mcrA mcrB1) (Casadaban and Cohen 1980). All plasmid constructs utilize pBAD33 (Cmr) (Guzman et al. 1995), which contains the promoter of the araBAD operon and the p15A origin of replication (low-copy number). Reagents and their suppliers were as follows: primers (Integrated DNA Technologies and Operon), restriction enzymes (New England BioLabs), streptavidin-R-phycoerythrin (SAPE) (Molecular Probes), biotinylated anti-T7 tag monoclonal IgG (Novagen); MyOne streptavidin-coated magnetic microbeads (Dynal), B-PER II bacterial protein extraction reagent (Pierce Biotechnology), anti-biotin mAb R-phycoerythrin conjugate (Miltenyi Biotec), Qdot655 streptavidin conjugate (Quantum Dot Corp), and Ni-NTA agarose (Qiagen).

Vector and library construction

Oligonucleotide primers are listed in Supplemental Table 1. The OmpX encoding gene and RBS were amplified from MC1061 genomic DNA using primers encoding a flanking 5′ KpnI restriction site and 3′ SfiI and HindIII sites (primers PD515 and PD516, respectively). Plasmid pB33OmpX was constructed by ligation of KpnI/HindIII digested ompX into similarly digested pBAD33. The T7 tag epitope was inserted into the second extra-cellular loop of OmpX, between S53 and S54, using overlap PCR with PD517/PD516 and PD518/PD515 with pB33OmpX as the template, generating flanking residues of GQSGQ (encoded by an SfiI containing DNA sequence) upstream of the epitope and GGS downstream of the epitope. The overlap product was digested with KpnI/HindIII and ligated into similarly digested pBAD33 vector yielding pB33OmpXT2. Plasmid pB33OmpX-temp was constructed with template pB33OmpXT2 and primers PD538/PD492, resulting in removal of the SfiI sites and the T7 tag epitope. The resulting product was used for PCR with PD179 and pB33OmpXT2 as template. The product was digested with KpnI/HindIII and ligated into similarly digested pBAD33 vector.

For circularly permutated OmpX (CPX) display, overlap PCR was used to assemble a gene comprised of the following elements: (1) the native signal sequence, (2) amino acids GQSGQ (with an embedded SfiI restriction site), (3) the passenger peptide, (4) amino acids GGQSGQ, (5) amino acids S54–F148 of OmpX, (6) GGSG (joining the native OmpX C and N termini), and, finally, (7) OmpX amino acids A1-S53. Standard recombinant DNA methods and overlap PCR were used to construct N- or C-terminal CPX constructs to display T7 tag or streptavidin binding peptide on the cell surface. Assembled genes were digested with HindIII and KpnI and ligated into similarly digested pBAD33.

Peptide insertion libraries within OmpX of the form X2CX7CX2 and X4CX3CX4 were constructed using PCR with primers PD671/PD516 or PD539/PD516, respectively, using as a template pB33OmpXtemp. The resulting products were digested with SfiI and ligated into SfiI-digested pB33OmpXT2. Ligation products were desalted and electroporated into electro-competent MC1061 yielding 4 × 109 and 1.3 × 109 independent transformants, respectively. For CPX library construction, a PCR template was created (pB33CPX-template) lacking a passenger peptide and incorporating a silent mutation that destroys the SfiI restriction site. Two X2CX7CX2 libraries were generated in two separate CPX vectors. In the first library (CX71), PCR was performed with primers PD707 and PD634 using pB33CPX-template. The resulting PCR product was lengthened in a second PCR to enable efficient digestion, using primers PD180 and PD753. This product was digested with SfiI and ligated into HincII/SfiI digested pB33CPX-template. Transformation of the ligation mixture into MC1061 yielded 4×108 independent transformants. A second X2CX7CX2 library (CX72) was constructed using a CPX scaffold containing the substitutions L(–17)V, L(–14)V, L(–10)V, L26V, L37I, L113V, L123V, where the residue numbering is based on the wild-type OmpX. This leucine deficient CPX scaffold displayed T7 tag and SA binding peptides equally as well as “wild-type” CPX (P.H. Bessette and P.S. Daugherty, unpubl.). A plasmid encoding a CPX without leucine codons (pB33NLCPX) was constructed by overlap PCR. PCR products generated with primers PD515/PD703, PD704/PD632, and PD633/PD634 were used for overlap PCR with outside primers PD515/PD634. The resulting product was cloned into KpnI/HindIII digested pBAD33. The CX72 library was then constructed by PCR amplification of pB33NLCPX with primers PD707/PD180, amplification of the product with PD753/PD180, digestion with SfiI, and ligation into SfiI-digested pB33NLCPX. Transformation of the ligation reaction yielded 1×1010 independent transformants.

Known streptavidin binding peptide sequences were inserted into both OmpX and CPX. The sequence AECHPQGPPCIEGRK (Giebel et al. 1995) was introduced into OmpX using PCR with primers PD1219/PD1220/PD180, and into CPX using primers PD1221/PD1222/PD180. The sequence WSHPQFEK (Korndorfer and Skerra 2002) was introduced into OmpX using PCR with primers PD1215/PD1216/PD180, and into CPX using primers PD1217/PD1218/PD180. The PCR products were digested with SfiI and ligated into a similarly digested vector.

Fluorescent protein fusions

Fluorescent protein–peptide fusions were constructed using PCR with an Aequorea GFP-based yellow fluorescent protein variant (YPet) (Nguyen and Daugherty 2005) gene as template and representative primers PD494 and PD860 in the first reaction and PD494 and PD859 in the second reaction. This resulted in the fusion of the peptide and a GSGS linker at the N terminus of YPet, and a hexahistidine tag to the C terminus. The C-terminal peptide fusions to YPet were constructed with primers PD179/PD915 in the first round and PD179/PD914 in the second round. The product was digested with SfiI and ligated into similarly digested pBAD33 containing two asymmetric SfiI sites. In a similar manner, the terminal fusion of peptide OX7-S1 was made lacking the first cysteine, allowing for only one possible disulfide bond to form, creating OX7-S1b.

Magnetic selection and library analysis and screening by FACS

Typically, for magnetic selection, cells corresponding to 5× the library diversity were inoculated to LB medium containing 34 μg/mL chloramphenicol (Cm) for a final cell concentration of 0.05 OD600 and grown to mid-log phase at 37°C with shaking (250 rpm), at which time the culture was moved to room temperature (22°C) to equilibrate and then induced with L-arabinose to a final concentration of 0.02% (w/v). The OmpX library was induced at room temperature for 30 min, and the CPX library was induced at room temperature for 2 h, both with shaking (250 rpm). A volume of cells corresponding to 5× the library diversity was concentrated by centrifugation (3000g, 4°C, 10 min) and resuspended in cold PBS to 10–30 OD600. Dynal MyOne SA beads were added to a ratio of one bead per 100 cells in the initial round, and one bead per 10 cells in later rounds when the population had enriched. Magnetic separation was used to wash the beads three times with a volume of PBS equivalent to the volume used in the initial labeling, and the beads plus bound cells were finally resuspended in LB with Cm and 0.2% glucose (w/v) for overnight growth.

For flow cytometric analysis and sorting, induced cells (5 × 105 cells/μL) were labeled with 5–25 nM SA-PE in PBS at room temperature for 30 min, pelleted by centrifugation, and the supernatant was removed. Cells were resuspended in ice-cold PBS at ∼107 cells/mL and immediately analyzed on a FACSAria cytometer using 488-nm excitation. For analysis, 104–106 cells were interrogated, and for sorting, at least fivefold oversampling of the expected clonal diversity was used. For kinetic-based sorting, biotin was added to a final concentration of 2–10 μM to inhibit rebinding of streptavidin. After sorting, retained cells were amplified for further rounds of analysis and/ or sorting by growing overnight in medium containing glucose and plated directly on agar for isolation of single clones. Typically 5–15 selected clones were assayed for antigen binding by flow cytometry, and the identity of each peptide insert was determined by DNA sequencing.

Clonal characterization

Cells were labeled with 50 nM SA-PE for 30 min on ice. The cells were then washed once with 1 mL room temperature PBS and the initial fluorescence was measured by flow cytometry. Biotin was then added to 2–10 μM and cells were immediately analyzed by flow cytometry. Fluorescence data were collected continuously for ∼3 min. The apparent dissociation rate constants were then determined as described previously (Daugherty et al. 1998), using the initial linear region.

For analysis of peptide affinity in a soluble scaffold, YPet– peptide fusions were prepared by cytoplasmic expression in E. coli strain MC1061. Soluble protein was isolated using BPER II bacterial protein extraction reagent following the manufacturer's protocol. For hexahistidine tagged proteins, the lysate was then purified using immobilized metal affinity chromatography. YPet fusions were dialyzed overnight against PBS (pH 7.4). Approximately 107 streptavidin coated magnetic beads (Dynal) were added to 10 μL of the YPet–peptide fusion with 15 μL of PBS and equilibrated at room temperature for 20 min. The beads were washed once in PBS (100 μL) and resuspended in 2 mL of PBS. The initial fluorescence was measured by flow cytometry; biotin was added to a final concentration of 2–10 μM, and cells were analyzed as above.

To determine specificity, clones were assayed for binding to various proteins using flow cytometry. C-reactive protein (CRP) was used at a concentration of 500 nM and was fluorescently conjugated to Alexa 488. Human serum albumin (HSA) and anti-T7 tag monoclonal antibody were biotinylated and used at concentrations of 500 nM and 200 nM, respectively. A secondary label of anti-biotin R-phycoerythrin was used at a 1 nM concentration for the biotinylated samples. A fourth probe, a fusion of the Mona/Gads SH3-C domain to the YFP mutant YPet (Nguyen and Daugherty 2005), was used for direct labeling at a final concentration of 200 nM.

Fluorescence microscopy

A gene encoding the fluorescent protein AlajGFP1 (Bessette and Daugherty 2004) was genetically inserted into the plasmid pLAC22 (Warren et al. 2000) using EcoRI and HindIII restriction sites, and the resulting plasmid was transformed into MC1061/pCX71-S1 and MC1061/pB33CPX-template to allow for fluorescent imaging. Cells were grown to log phase and induced with 0.02% arabinose and 1 mM IPTG for 2 h at room temperature. Approximately 5 × 106 cells were then incubated for 30 min at room temperature with 20 μL of a 1:20 dilution of streptavidin-conjugated quantum dots (emission max. 655 nM) and washed once with PBS. The unfixed cells were imaged using a Zeiss Axiovert microscope with a 100× oil emersion objective with standard green and red filter sets from Chroma. Images were acquired using a cooled CCD camera CoolSNAP HQ (Roper Scientific).

Electronic supplemental material

Supplemental material consists of one table entitled “Oligonucleotide Primers Used for Vector and Library Construction.”

Figure Figure 1.

Diagram of bacterial display scaffolds OmpX and CPX. (A) Structure of outer membrane protein OmpX (Vogt and Schulz 1999) with extracellular loops L2 and L3 projecting beyond the lipopolysac charide (LPS) layer, alongside a topological depiction of OmpX and CPX, where the native OmpX N and C termini are fused together by a GGSG linker (not to scale), and the newly formed N and C termini reside on the cell surface. (B) Gene map of OmpX and CPX depicting the araBAD promoter and ribosomal binding site as an arrow, followed by the OmpX signal sequence, and the remainder of the gene and stop codon. The vertical lines depict the portion of the gene encoding A1-S53, and the diagonal lines are the remaining codons. The linking sequences are shown as black lines, the asymmetric SfiI sites are indicated with subscripts, and the wild-type residues are labeled.

Figure Figure 2.

Display of a SA-binding peptide using OmpX or CPX. Flow cytometric analysis of E. coli displaying a SA binding peptide (RLEICQNVCYYLGTL) (Bessette et al. 2004) as an insertion into OmpX (A) or an N-terminal fusion to CPX (B), after varying durations of induction. The cells were induced at room temperature for time increments between 0 and 180 min, then labeled with 100 nM SA-PE.

Figure Figure 3.

Fluorescence microscopy analysis of peptide surface localization using CPX display. Bacterial cells coexpressing GFP and CPX1-S1 (A) or bacterial cells coexpressing GFP and CPX (no peptide) (B) were labeled with streptavidin-conjugated quantum dots. Images were acquired using GFP and PE filter sets and then overlaid.

Figure Figure 4.

Enrichment of streptavidin-binding clones from a CPX library as measured using flow cytometry. Flow cytometric analysis after labeling with 5 nM SA-PE revealed 0.01% SA-PE positive cells prior to selection (A), 1% after two rounds of MACS (B), and 85% after two rounds of MACS and one round of FACS (C). Plots use a four-decade log scale.

Table Table 1. Streptavidin binding peptides identified using OmpX (OX) or CPX (CX) bacterial display libraries
original image
Figure Figure 5.

Dissociation rate constants of selected clones measured on the cell surface or for soluble proteins. (A) Measurement of the dissociation rate of several streptavidin binding bacterial clones after labeling with SA-PE and addition of biotin as a competitor. (Diamonds) CX71-S1, (triangles) OX7-S1, (circles) CX72-S2, (asterisks) OX7-S2, (squares) CX72-SA3. (B) Measurement of SA binding peptide–YFP fusions bound to SA-conjugated polymeric beads after addition of biotin as a competitor. (Diamonds) CX71-S1, (circles) CX72-S2, (triangles) OX7-S1b.

Table Table 2. Comparison of the dissociation rates of cell surface peptides and soluble fusion proteins
original image

Constants determined using flow cytometry as described in the Materials and Methods section.

aBinding was not detected.

bBinding of the protein to streptavidin was detected, but the off rate was too fast to determine using cytometry.

ND, Not determined.

Figure Figure 6.

Specificity of isolated clones for streptavidin. The extent of peptide binding to unrelated proteins, including anti-T7 tag monoclonal antibody at 200 nM, human serum albumin (HSA) at 500 nM, C-reactive protein (CRP) at 500 nM, and the Mona/Gads SH3 domain at 500 nM, as measured using flow cytometry.

Acknowledgements

We thank Brian Kay and Michael Scholle for critically reading the manuscript. We thank Chinmay Pangarkar for help with the microscopy. This project was supported, in part, by National Science Foundation CAREER award (BES-0449399) to P.S.D., and by the Institute for Collaborative Biotechnologies through grant no. DAAD19-03-D-0004 from the U.S. Army Research Office.

Ancillary