Constrained binding peptides (peptide aptamers) may serve as tools to explore protein conformations and disrupt protein−protein interactions. The quality of the protein scaffold, by which the binding peptide is constrained and presented, is of crucial importance. SQT (Stefin A Quadruple Mutant—Tracy) is our most recent development in the Stefin A-derived scaffold series. Stefin A naturally uses three surfaces to interact with its targets. SQT tolerates peptide insertions at all three positions. Peptide aptamers in the SQT scaffold can be expressed in bacterial, yeast and human cells, and displayed as a fusion to truncated pIII on phage. Peptides that bind to CDK2 can show improved binding in protein microarrays when presented by the SQT scaffold. Yeast two-hybrid libraries have been screened for binders to the POZ domain of BCL-6 and to a peptide derived from PBP2’, specific to methicillin-resistant Staphylococcus aureus. Presentation of the Noxa BH3 helix by SQT allows specific interaction with Mcl-1 in human cells. Together, our results show that Stefin A-derived scaffolds, including SQT, can be used for a variety of applications in cellular and molecular biology. We will henceforth refer to Stefin A-derived engineered proteins as Scannins.
Molecular tools that bind to proteins with high specificity and affinity are increasingly important in basic, translational and applied research. This is relevant in diagnostics (to detect disease proteins or for guided imaging) and therapeutics (to disrupt protein−protein interactions or for targeted therapies), as well as in systems-wide proteome analysis, including the emerging discipline of interactomics (dissection of cellular protein networks).
Antibodies represent the best-studied group of binding molecules to date. Although they have been applied successfully to a variety of biological and medical problems, their use can be limited by their structural properties (Ladner et al., 2004; Skerra, 2007; Gebauer and Skerra, 2009). Antibody molecules are comparatively large, multimeric proteins that require disulphide bonds and glycosylation for stability. Consequently, antibodies are functionally confined to the extracellular environment and are difficult to express in high throughput or commercially attractive bacterial systems. Additionally, commercial antibodies are commonly raised against short, linear peptides, making it difficult to explore different protein conformations that are relevant in biological and clinical research (Pontén et al., 2008).
Advances in the construction and screening of combinatorial protein libraries have led to the development of almost 50 different non-immunoglobulin protein scaffolds designed to constrain and present variable peptide inserts for protein recognition (reviewed in Binz et al., 2005; Groenwall and Stahl, 2009). The rationale behind constraining the binding peptide within a scaffold, rather than using free peptides, is 2-fold. Firstly, due to their flexible nature, unconstrained peptides can interact non-specifically and with low affinity with a range of different surfaces. When inside a scaffold, peptides should be restricted to a limited range of conformations and binding will occur with increased specificity and at high affinity, as the entropic cost of binding is decreased (Ladner, 1995). Secondly, scaffold-constrained peptides are less susceptible to degradation in cells, resulting in a longer half-life compared with their unconstrained counterparts (Saveanu et al., 2002; Reits et al., 2003). Engineered scaffolds are variable in size and fold, and range from small protein domains such as the protein A domain (Affibody, Nord et al., 1995), PDZ domains (Schneider et al., 1999) and ankyrin repeat proteins (Darpins, Binz et al., 2003, 2004), through small full-length proteins, such as the commonly used thioredoxin scaffold (La Vallie et al., 1993; Colas et al., 1996), to higher-molecular-weight β-barrels and Ig-like structures such as lipocalins (Anticalins, Beste et al., 1999), green fluorescent protein (GFP) (Abedi et al., 1998) and the T-cell receptor (Holler et al., 2000). Examples of very recent applications include cystine-knot peptides that bind to αvβ3 integrin and inhibit cell binding to vitronectin (Silverman et al., 2009), or a HER2-specific Affibody that was used for tumour imaging in vivo (Orlova et al., 2007), for nano-particle targeting in tumour therapy (Alexis et al., 2008) or targeting of adenovirus to tumours in gene therapy (Myhre et al., 2009). Affibody-affinity columns have been used to deplete cerebrospinal fluid of four high-abundance proteins prior to proteomic analysis (Ramström et al., 2009). Colas and co-workers coined the term ‘peptide aptamer’ to describe binding peptides constrained within a constant, full-length scaffold (thioredoxin) identified using an intracellular screening platform (yeast two hybrid), starting with the identification of high-affinity CDK2 binders (Colas et al., 1996). Thioredoxin is one of the most commonly used scaffolds to date (e.g. LaVallie et al., 1993; Colas et al., 1996; Hoppe-Seyler et al., 2004; Borghouts et al., 2008; Wickramasinghe et al., 2010).
Despite its unquestionable success, thioredoxin contains a disulphide bridge that may affect the folding of inserted peptides and possesses biological activity when expressed in human and mouse cells (Tao et al., 2004). In seeking alternatives, we focused on the protease inhibitor Stefin A (cystatin A) (Woodman et al., 2005). Low-molecular-weight protease inhibitor proteins are highly stable and use exposed peptide loop(s) to engage their target protease, making them excellent candidate scaffolds and indeed Kunitz domains have been engineered successfully in the past (Dennis and Lazarus, 1994). Stefin A is a 98 amino acid (aa), single-chain protein, which inhibits members of the cathepsin family of proteases, including cathepsins B, C, H, L and S (Brzin et al., 1984; Green et al., 1984; Abrahamson et al., 1986). It interacts with its target via three distinct features, namely the amino-terminus and two hairpin structures referred to as loop 1 and loop 2 (Jenko et al., 2003). Guided by the available NMR and crystal structures (Bode et al., 1988; Stubbs et al., 1990; Martin et al., 1994, 1995; Tate et al., 1995; Pavlova and Björk, 2002; Jenko et al., 2003), we replaced specific residues to abolish the binding of Stefin A to cathepsins. Thus, by replacing glycine at position 4 with tryptophan and valine 48 with aspartate, the N-terminal and loop 1 interaction sites were successfully abolished. Furthermore, the introduction of an RsrII restriction endonuclease site in the open reading frame corresponding to loop 2 allows the introduction of randomised oligonucleotide, as well as removing the final interaction with cathepsins. The combination of these changes into a single scaffold was designated Stefin A Triple Mutant (STM) (Woodman et al., 2005). Work since has shown that STM is capable of presenting peptide loops for specific interaction with CDK2, allowing the design of a range of biophysical assays (Evans et al, 2008; Johnson et al., 2008; Shu et al., 2008, Davis et al., 2009; Estrela et al., 2008; 2010).
More recently, we set out to further improve Stefin A-based scaffolds. Studies from antibodies have shown that bivalency (i.e. the use of two interaction sites in one molecule to bind a single target) can lead to increased binding affinities (avidity effect, Hollinger and Hudson, 2005) and we wished to apply this idea to engineered Stefin A variants. We further reasoned that the N-terminus, loop 1 and loop 2 may each constrain peptides differently, increasing the versatility of the scaffold. This led to the design of the SQM scaffold (Stefin A Quadruple Mutant, Hoffmann et al., 2010). SQM can accept peptide insertions into the N-terminus and loop 1, respectively, whereas loop 2 can be replaced by alternative peptides of at least 17aa in length. However, insertions into the N-terminus and loop 1 individually or in combination with loop 2 had varying effects on scaffold stability, raising concerns about the usefulness of SQM as a scaffold for large combinatorial libraries.
In the present study we show that the stability issues with SQM are due to specific mutations introduced in the β-sheet adjacent to loop 2. A new variant scaffold, SQT (Stefin A Quadruple Mutant—Tracy), is able to present a range of peptides and shows thermostability similar to that of the parental Stefin A protein. Peptide aptamers, which we now call Scannins, presenting ‘epitope tag’ peptides in the N-terminus, loop 1 and/or loop 2 of SQT are well expressed in Escherichia coli, yeast and human cells, as well as at the surface of phage. Amino-terminal insertions are uniformly well tolerated in SQT, opening up the possibility of utilising this site for the generation of combinatorial libraries. SQT is able to present peptides in loop 2 or the N-terminus for interaction with CDK2 (in protein microarrays) and Mcl-1 (in human cells), respectively. We have constructed yeast two-hybrid libraries using loop1, loop2 or both simultaneously and used them to identify specific interactors for the POZ domain of BCL6 and for a peptide derived from the penicillin binding protein PBP2’ of methicillin-resistant Staphylococcus aureus. Together with our previous work, our data suggest that Scannins are likely to be broadly useful in presenting a wide range of peptides for interaction using three independent sites, with protein binding affinities and stabilities that can be tuned to the experimental setting.
Materials and methods
Enzymes, reagents and chemicals
Unless otherwise specified, enzymes for molecular biology were from NEB (Ipswich, MA, USA), and reagents and chemicals from BDH (Darmstadt, Germany) or Sigma (Dorset, UK) and were the highest grade available.
Plasmids and DNA manipulation for bacterial expression
Stefin A variants STM and SQM were reported previously (Woodman et al, 2005; Hoffmann et al., 2010). SQT was created by moving the second RsrII site of SQTM, so that amino acid changes occurred in loop 2 rather than in the β-strand. This was achieved by carrying out two consecutive site-directed mutagenesis reactions (SDM) using the following oligonucleotides:
SDM 1 forward: 5'AACGGACCGCC CGGACAAA ATGCGGACCGGGTACACTCCGG ATACCAGGTT3'
SDM 1 reverse:
SDM 2 forward:
SDM 2 reverse:
5′CCTTGTTTTTGTCAACCTGGTATCCGGTGAGTACCCGGTCCGCATTTTGTCCG3′. Site-directed mutagenesis was performed in a Tetrad thermal cycler (BioRad Corp, Hercules, CA, USA) with 12 cycles of polymerase chain reaction catalysed by Taq, followed by DpnI treatment to degrade the template DNA. The remaining PCR product was transformed into XL10 gold competent cells (Stratagene, Santa Clara, CA, USA).
Peptide insertion into the Stefin A-based scaffolds
Double-stranded (ds) DNA cassettes encoding each peptide (Supplementary data, Table A) and flanked by the relevant restriction enzyme sites were made by annealing oligonucleotides (Supplementary data, Table B). Restriction enzyme digested dsDNA cassettes were ligated into the appropriate restriction sites of the scaffold-encoding open reading frame in pET30a(+) (Novagen, Gibbstown, NJ, USA).
Peptide aptamer cloning and expression in mammalian cells
SQM, SQT and pepNoxa were amplified by PCR (forward SQM and SQT 5′-CCGCGGCCGCAGATCATGATACCTAGGGGCTTATC-3′; forward pepNoxa 5′-CCGCGGCCGCAGATCATGATACCTAGGCCTGCTGA-3′; and reverse 5′-GAGAGGGGCGCCATGCTAAAAGCCCGTCAGCTCG-3′) and cloned into the Retro-X ProteoTuner™ retroviral expression vector (Clontech, Mountain View, CA, USA) using the In-Fusion™ PCR cloning system (Clontech). Clones were sequenced using the ABI PRISM Big dye terminator kit V1.1 (Applied Biosystems, Carlsbad, CA, USA) on an ABI 3100 Genetic Analyser. Constructs were transfected into the PhoenixA packaging cell line (ATCC, Teddington, UK), which produce the gag-pol, and envelope proteins for amphitropic viruses. After 48 h, medium was harvested, 0.4 µm filtered and mixed in equal amounts with fresh growth medium containing 8 µg/ml of polybrene (Sigma). Polybrene is a cationic polymer that increases retrovirus gene transfer efficiency. MCF-7 or A375 cells were grown to 50% confluence and then incubated with retroviral supernatants for 8h. Forty-eight hours after infection, cells were transferred into selection medium containing puromycin. Cells were cultured with Shield (5 µM; Clontech) to induce expression of SQM, SQT or pepNoxa and either lysed in 1% Triton buffer (1% Triton-X 100, 2 mM EDTA and protease inhibitor cocktail in phosphate-buffered saline (PBS)) followed by centrifugation (10,000 rpm, 4°C, 10 min in an Eppendorf microfuge) for western blotting or RNA was harvested in RLT buffer (Qiagen) for reverse-transcriptase polymerase chain reaction (RT–PCR).
Yeast two-hybrid assay
The yeast two-hybrid screening system used was the ProQuest™ two-hybrid system (Invitrogen) using the yeast strains MaV103 (MATa) and MaV203 (MATα) (the latter a generous gift of Professor Marc Vidal, Dana-Farber Cancer Institute). Both yeast expression vectors, pEXP22 (prey, encoding a fusion protein of the peptide aptamers and the GAL4 transcriptional activation domain) and pEXP32 (bait, encoding a fusion protein of the target protein and the GAL4 DNA binding domain), were obtained from Invitrogen.
Bait sequences were cloned into the ProQuest™ (Invitrogen) vector pDEST32 (POZ domain of BCL6: POZ-BCL6_F: 5'-ATG GGT TCC CCC GCC-3', POZ-BCL6_R: 5'-AGG TTC ATA GCT GGC CTG G-3'; a peptide fragment of the penicillin binding protein PBP2' (designated MecA_3): MecA_3_F:, MecA_3_R:) thus generating the yeast expression vectors pEXP32-POZ and pEXP32-MecA_3, which was transformed into yeast cells (MaV203, MATα).
A loop 1 only (L1), a loop 2 only (L2) and a loop 1/loop 2 combined (L12) library of random Scannins were generated. Oligonucleotides were designed with homology to the linking region between the loop 1 and loop 2 coding sequences flanked by a series of randomised codons. For the libraries, the forward primers included 6 (L12) or 10 (L1 only) and 12 (L2, L12) randomised codons encoding random peptide inserts of 6, 10 or 12aa in length. Hybridisation of forward and reverse primers followed by a Klenow fill-in reaction and digestion with NheI and RsrII restriction enzymes created a mixed pool of random inserts that were integrated in the pEXP22-SQT vector by standard ligation. Yeast two-hybrid screens were performed by mating MaV103 cells expressing one of the libraries and MaV203 cells expressing the protein baits, followed by selection with three different reporter systems (growth on 3AT, growth on medium lacking uracil and LacZ). To confirm interactions observed in the screen, the haploid yeast strains MaV103 and MaV203 carrying either the prey vector or the bait vector, respectively, were mated by replica-plating in a mating grid.
Production of recombinant proteins in E.coli
All scaffold proteins were expressed as described previously (Hoffmann et al., 2010). Briefly, cells were grown in Luria Bertani broth at 37°C to A600= 0.3 and treated with 0.4 mM isopropyl-β-d-thio-galactoside. Cultures were centrifuged at 3220×g after 3 h and cell pellets resuspended in lysis buffer (50 mM sodium phosphate buffer (pH 7.4), 300 mM NaCl and protease inhibitor lacking EDTA (Roche)), followed by lysis in a cell disrupter (Constant Systems, Daventry, UK). Recombinant protein was captured on Ni-NTA columns (Qiagen), washed (lysis buffer plus 30 mM imidazole) and eluted (lysis buffer plus 250 mM imidazole). The purified protein was buffer exchanged into 50 mM sodium phosphate (pH 7.4) using Amicon Ultra 10 kDa columns (Millipore, Billerica, MA, USA).
Far-UV circular dichroism spectroscopy
Circular dichroism (CD) spectra were recorded on a Jasco J715 spectropolarimeter at 10°C. Folding spectra were collected from 190 to 260 nm. Thermal denaturation curves were performed at a temperature range of 30–98°C. Melting temperatures were determined by following the ellipticity at 208 nm. The experimental data were fitted to a sigmoidal or a double-sigmoidal function by non-linear least-squares regression using R software. For SQT, the best fit was obtained using a double-sigmoidal function. The raw output is given in ellipticity (θ (mdeg)). The data were normalised by calculating the mean residue ellipticity (Equation (1)):http://plot.micw.ec/).
Immunoprecipitation and co-immunoprecipitation
For immunoprecipitation 10 μl anti-Myc tag antibody-coated Agarose resin (Abcam, Cambridge, UK) was blocked in 250 μl blocking buffer (50 mM sodium phosphate (pH 7.4), 4% bovine serum albumin (BSA) and 0.1% Nonidet P-40 (NP-40, Calbiochem, Gibbstown, NJ, USA)) for 1 h at 4°C, followed by three washes in wash buffer (50 mM sodium phosphate with 0.05% BSA and 0.1% NP-40). Subsequently, the resin was incubated with 1 μg of purified peptide aptamer in 200 μl of wash buffer for 2 h at 4°C, followed by seven washes. Protein sample buffer was added (Laemmli, 1970), the sample boiled for 5 min and analysed by sodium dodecyl sulphate (SDS) polyacrylamide gel electrophoresis followed by western blotting. Co-immunoprecipitation was performed with the Crosslinking Immunoprecipitation kit (Pierce, Rockford, IL, USA) according to the manufacturer's instructions using an anti-Stefin A antibody (R&D Systems, Minneapolis, Minnesota, USA).
Western blotting was preceded by SDS polyacrylamide gel electrophoresis and transfer onto nitrocellulose membrane. Antibodies were anti-S-tag monoclonal antibody (1:5000, Novagen, USA), anti-Stefin A (R&D Systems; 1:1000), anti-tubulin (1:10,000; Serotec, Kidlington, UK) and anti- Mcl-1, Bcl-2, Bcl-xl (all 1:1000; all NEB). Western blots were quantified using Quantity One Software (Biorad) where relevant.
Reverse-transcriptase polymerase chain reaction
Total cellular RNA was extracted using Qiagen RNeasy Mini Kit (Qiagen, Crawley, UK) and 1 µg was reverse transcribed in the presence or absence of reverse transcriptase (Advantage RT-for-PCR kit, Clontech) according to the manufacturer's instructions. SQM and SQT sequences were amplified using the primers described above and hypoxanthine-guanine phosphoribosyltransferase primers were: forward: 5′- CCTGCTGGATTACATTAAAGCACTG-3′, reverse: 5'–CTGAAGTACTCATTATAGTCAAGGG-3′.
Samples were printed using a BioOdessey and 100 µm capillary pins (Biorad Corp) onto amine slides (Genetix, New Milton, UK). SQT samples were printed at concentrations of 10 or 40 μM in PBS-T (PBS with 0.05% Tween-20) with 10% glycerol. Sample grids were printed as quadruplicate spots for each peptide aptamer or control sample and the entire grid was repeated three times across the array. Sixty minutes after printing, slides were blocked in 1% BSA PBS-T for 60 min. For epitope tag detection, antibodies (anti-Myc, -HA, -AU1; all Abcam) were labelled 1:1 with Atto 550-NHS ester or Atto 647-NHS ester. Slides were probed with antibody concentrations of 7–28 nM in 1% BSA in PBS-T for 40 min using 50 μl volume Lifterslips (Thermo Scientific, USA). Slides were then washed in PBS-T twice for 10 min then twice in deionised water for 2 min. The slides were dried under a stream of nitrogen and scanned at a resolution of 5 μm with 543 and 633 nm lasers under the Cy3/Cy5 detection protocol (Scan Array Express, Perkin Elmer).
For CDK2 binders, STM/SQM/SQT carrying pep2, 6, 9 or 10m in loop 2 were spotted on Ni2+ glass slides (Xenopore Corp, Hawthorne, NJ, USA) using a QArray mini (Genetix) in quadruplicate at each of four different concentrations (10, 5, 2.5, 1.25 μM). Slides were incubated with 100 ng/ml recombinant active GST-CDK2/Cyclin A (NEB) in PBS-T, followed by three washes in PBS-T and incubation with Cy3-labelled anti-CDK2 antibody (NEB). Slides were scanned as above.
Surface plasmon resonance
Surface plasmon resonance (SPR) experiments (Autolab Esprit, Amsterdam, Netherlands) were performed on gold surfaces cleaned in hot piranha solution and functionalised overnight with a mixed 1:1 monolayer of OEG-OH thiol (1-mercapto-11-undecyl)tri(ethylene glycol) (Asemblon) and OEG-COOH thiol (HS(CH2)10(OCH2CH2)3OCH2COOH (Prochimia Surfaces) at 625 μM in ethanol. SPR chips were then equilibrated to a stable time-dependent signal under 10 mM pH 7.4 phosphate buffer conditions. Terminal acid groups were activated with 100 μl of a 1:1 EDC, 0.2 M and NHS, 0.05 M solution for 15 min. Each peptide aptamer was injected at a concentration of 10 μM and incubated on the surface until no more adsorption was observed. Any unreacted NHS groups were quenched by incubation in 50 mM ethanolamine for 5 min. Antibodies were injected at concentrations of 7 nM, and absorption/desorption profiles measured. Bound antibodies were eluted with 0.1 M glycine solution pH 2.4 for 2 min.
Plasmids pCD87SA-GFP and pDsbAss-GFP were kindly provided by Matthias Paschke (Charite, Berlin). The vectors pOmpA, pPhoA, pPelB and pTorA are derivatives of pDsbAss and share the same vector backbone. The SQT-AU1-Myc ORF was amplified from pET30a+ by PCR using the primer pair SQT-SfiI_F 5′-TGCGACTGCG GCCCAGCCGGCCATGATACCTAGGGGCTTATC-3′ and SQT-SfiI-R 5′-GTTCTGCGGCCACCGAGGCCGAAAA GCCCGTCAGCTCGTCATC-3′, followed by digestion with BglI and ligation with BglI digested vector backbones of pCD87SA, pOmpA, pPhoA, pPelB, pDsbAss and pTorA.
Production of phage particles
Wells containing 200 µl 2xTY and chloramphenicol (2xTYC) were inoculated with 12 single E.coli colonies each from ER2738 cells transformed with pCD87SA-GFP, pCD87SA-SQT, pTorA-SQT, pOmpA-SQT, pPhoA-SQT, pPelB-SQT and pDsbAss-SQT and incubated overnight at 37°C and with shaking. A volume of 190 µL 2xTYC were inoculated with 10 µl of the overnight culture and grown at 37°C to OD600 = 0.5. Bacteria were infected with 2 × 109 helper phage and incubated with shaking for 30 min at 25°C. Following addition of 50 µg/ml kanamycin the plate was incubated for another 15 min. Cells were harvested by centrifugation and the pellet resuspended in 200 µl 2xTYC, 50 µg/ml kanamycin and 1 µg/ml tetracyclin and incubated at 25°C with shaking overnight. Cells were removed from phage containing supernatant by centrifugation at 4°C.
ELISA was performed in 96 microtitre well plates (MaxisorbTM, Nunc) coated overnight at 4°C with 1 µg/ml goat anti-human cystatin A antibody in PBS. Wells were blocked with 2× blocking buffer (Sigma)/PBS overnight at 37°C, followed by washing with PBS-T. Some 20 µl of phage-containing supernatant were diluted 1:1 in 2× blocking/PBS and incubated with the antigen for 2 h, followed by washing with PBS-T. Bound phage was detected using rabbit anti-fd phage (1:1000) (Sigma) and anti-rabbit-horseradish peroxidase (1:6000). The visualisation was carried out after 20 washing steps with 50 µl TMB (Seramun, Heidesee, Germany) as substrate and the staining reaction was stopped with 1 N sulphuric acid. Absorbance was measured at 450 nm.
Thermal stability lost in SQM can be restored in SQT
We have designed a new variant scaffold protein, based on the human protease inhibitor Stefin A, called SQT (Fig. 1). Site-directed mutagenesis of the scaffold open reading frame allows us to use engineered restriction sites for the insertion of peptide-encoding oligonucleotides. These peptides become part of the primary sequence of the expressed protein, with the goal that the three-dimensional folding of the scaffold enables the peptides to adopt specific shapes to mediate between high-affinity and specific protein recognition. We first used CD to determine whether the amino acid changes in SQT lead to loss of secondary structure (Fig. 1A). Stefin A is a very stable protein and has previously been shown to be highly resistant to heat-induced denaturation (Zerovnik et al, 1992). The temperature at which half the protein is unfolded (T50%) has been reported to be either 90.8 or 94.5°C at different pH (pH 5.0/pH 6.5 and pH 8.1; Zerovnik et al., 1992). Here we employed CD spectroscopy to measure the T50% for STM, SQM and SQT at pH 7.4. For STM we measured a T50% of 82.9°C (Fig. 1B), which is in agreement with our previously reported T50% = 81°C (Woodman et al., 2005). SQM, which bears L82R and T83S substitutions with respect to Stefin A and STM due to the second RsrII site in the open reading frame (Fig. 1C), showed a T50% of 52.5°C (Fig. 1B). SQT, where the second RsrII site results in an E78A and L80R substitution with respect to Stefin A (Fig. 1C) has a measured T50% = 79.7°C (Fig. 1B), similar to STM. Interestingly, the absorbance spectrum for SQT exhibits biphasic behaviour, with the ellipticity peaking at about 69°C, before sharply dropping off (see the Discussion).
SQT tolerates short peptides in all three insertion sites
All three Stefin A-derived scaffolds differ with respect to their loop 2 insertion sites. In STM, oligonucleotides encoding peptide sequences are inserted into a single RsrII site corresponding to the amino terminus of loop 2, thus leading to an extension of that loop on insertion. Both SQM and SQT contain two RsrII sites at positions corresponding to the beginning and end of loop 2, and consequently inserting a peptide sequence results in a replacement of the original Stefin A loop 2 (Fig. 1C). In SQM the excision removes eight amino acids, including some from the adjacent β-sheet, whereas in SQT only four amino acids are removed. We expected that due to these differences, all three scaffolds would behave differently on peptide insertion. We tested this using three short peptides corresponding to the epitope tags AU1 (6aa), HA (9aa) and Myc (10aa). In SQM we found that insertions of the AU1 tag into loop 1 and either the HA or Myc tag into loop 2 (individually or in combination) were generally well tolerated, judging both from the expression levels in E.coli as well as CD spectroscopy (Hoffmann et al., 2010). Inserting HA or Myc into the N-terminal site, however, had a disruptive effect on SQM, especially in combination with insertions in loop 1 and/or loop 2 (Hoffmann et al., 2010). In a similar manner, we cloned the epitope tag sequences into the different positions of SQT and assessed the protein yield following bacterial expression and purification using Ni-NTA columns (Fig. 2A). With all combinations investigated, SQT expresses uniformly at high levels. The lowest yield was obtained with a Myc tag in both loop 1 + 2 (25 mg/l culture) and the highest with AU1 in loop 1 and Myc in loop 2 (69 mg/l culture). This compares favourably with epitope-tagged SQM, where the protein yield is consistently very low (lowest yield: 2 mg/l culture), except for two extremely highly expressed versions (AU1 in loop 1: 103 mg/l culture; AU1, loop 1, HA loop 2: 206 mg/l culture; Fig. 2A). When we measured absorption in the far-UV range to assess protein folding (Fig. 2B and C), the resulting curves show a broad range of behaviour for SQM (Fig. 2B) but not for SQT (Fig. 2C), indicating that insertion of epitope tags into the primary sequence of SQT does not significantly disrupt the protein's secondary structure. As with both STM and Stefin A, the inflection is always at 218 nm and all sample curves follow the shape of the empty SQT reference spectrum. This is in stark contrast to SQM where alterations in the protein fold are readily observed (shift of inflection from 218 to 209 nm), especially when the N-terminus is tagged (Fig. 2B).
In a further test of the folding and stability of these model Scannins, we asked whether they could be exported to the periplasmic space for assembly at the surface of phage via either the E.coli Sec- or Tat-mediated pathways (Fig. 3). Sec-mediated transport usually requires protein unfolding prior to translocation across the membrane (Nilsson et al, 1991; Paschke and Höhne, 2005), whereas Tat-mediated transport requires correctly folded protein structures, allowing, for example, the expression of functional GFP (Paschke et al., 2007). We constructed several alternative phagemids to express SQT-AU1-MYC in bacteria using leader sequences specific for each of these pathways (the OmpA, PelB and PhoA leader sequences direct proteins to the Sec pathway, while TorA directs proteins to the Tat pathway; Paschke and Höhne, 2005; Paschke et al, 2007). In order to ask whether the Sec or Tat pathways could export SQT to the periplasmic space for assembly into phage particles, we used an anti-scaffold antibody to recruit the expressed Scannins to the surface of microtitre ELISA plates and an anti-phage antibody to detect immobilised phage that express the Scannins (Fig. 3). The data show that, in at least the case of the model Scannin used here, export via the Sec and Tat pathways and assembly of Scannin fusions to the truncated pIII protein into phage particles is possible (Fig. 3A), although quantification of the results reveals differences between Sec pathway leader sequences (Fig. 3B; see the Discussion). These experiments indicate that it will be possible to create phage display Scannin libraries using one or both pathways.
In combination, these results establish SQT as a scaffold where insertions of peptides of at least 10aa in length can be tolerated at all three positions, and the resulting Scannin proteins expressed in a range of systems relevant to molecular biology.
SQT presents peptides for interaction with cognate binding proteins
We next set out to test the ability of SQT to present inserted peptides for interaction. It is conceivable that for entropic reasons it is more favourable for a given peptide insert to be incorporated within the secondary structure of the protein, rather than being exposed to solvent. We used model peptide aptamers with a Myc tag inserted in either the N-terminus, loop 1 or loop 2 and performed immuno-precipitation (IP) experiments using an anti-Myc antibody (Fig. 4A). For model peptide aptamers with the Myc peptide in the amino terminal site or in loop 2, the HA and AU1 tags were inserted into the two remaining positions to test whether multiple inserts might interfere with target recognition. The IP was followed by a western blot to detect the scaffold protein. In all cases the scaffold is efficiently immuno-precipitated, leading to the conclusion that the Myc peptide is equally well recognised at all three positions and that peptides inserted simultaneously at several sites do not interfere with presentation of this epitope.
In order to further investigate the scaffold's ability to present peptides for interaction, we have devised a microarray assay which enables us qualitatively to test several epitope-presenting Scannins simultaneously using cognate antibodies. Model SQT Scannins containing various combinations of the HA, AU1 and Myc epitope tags were spotted on an amine surface followed by blocking and incubation with each labelled antibody. After washing, the slides were scanned using a standard microarray scanner (Fig. 4B). Print buffer control spots and tag-less Scannins all gave the expected lack of signal. The Myc and HA tags gave strong signals when binding was expected, whereas samples containing the AU1 tag typically gave a very weak signal when it was present in loop 1. All Scannins were spotted from 10 μM solutions and spotting consistency was confirmed using an anti-Stefin A antibody, which also allowed normalisation of all measurements to protein loading on the slide (Fig. 4C). Interestingly, our results reveal that, for the anti-AU1 antibody, binding strength depends on the position of the short AU1 tag in the SQT scaffold, with the tag in loop 2 giving the strongest signal. In order to confirm and quantify the microarray data, we employed SPR to measure relative binding affinities of the anti-AU1 antibody to SQT-AU1 in loop 1, SQT-AU1 in loop 2, and to SQT with the tag inserted in both positions simultaneously (Fig. 4D). Again, the strongest binding is observed with the AU1 epitope inserted in loop 2 and negligible binding is seen with the tag in loop 1. Interestingly, when two AU1 tags are present, no avidity effect was observed, as might be expected with a bivalent binding molecule, but instead the affinity is inferior in comparison with SQT with AU1 in loop 2 alone. We attribute these variations to the short length of the AU1 peptide (6 residues, compared with 9 or 10 for the HA and MYC peptides), with steric hindrance and perhaps a bending of the peptide imposed by the scaffold decreasing the interaction surface area available to the antibody.
In order to extend this result to a biologically relevant system, we sought to test whether SQT (and SQM) bearing the CDK2 binding peptides pep2, 6, 9 and 10m in loop 2 could interact with recombinant CDK2 (Fig. 5). Pep2, 6, 9 and 10m were first identified in the context of the thioredoxin scaffold (Colas et al., 1996) and were later shown to maintain their CDK2 binding ability when transferred to loop 2 of STM (Woodman et al., 2005; Evans et al, 2008). In order to test all three scaffolds (STM, SQM, SQT) with all four peptides in loop 2, we made use of the protein microarray format. The Scannins were spotted onto nickel-coated glass slides, followed by incubation with recombinant CDK2 and subsequent detection with fluorescently labelled anti-CDK2 antibody (Fig. 5A). As a negative control, we used the MYC epitope which showed no affinity for CDK2 when presented by either STM, SQM or SQT (Fig. 5A). Importantly, strength of binding varied not just with peptide insert − as has been shown in the context of thioredoxin − but also depending on the scaffold. The best binder in this experimental set-up was pep10m in the context of SQT, which gave the strongest signal, followed by pep6 in SQM (Fig. 5B). These data indicate that interactions between the peptide and the scaffold that presents it may contribute to target binding affinity.
SQT can be efficiently expressed in yeast cells
There is considerable precedent for the identification of useful peptide aptamers by yeast two-hybrid screening. In order to ask whether the SQT scaffold could be used to present libraries of random Scannins in this format, we constructed and screened three libraries. One of these comprised randomised peptides 10aa in length presented and constrained by loop 1. The second comprised randomised peptides 12aa in length presented and constrained by loop 2. In the third library, we randomised loop 1 and loop 2 simultaneously; we call this library loop 12. Because these two loops are relatively close to each other in the folded protein, for the combined loop 12 library we decided to use a shorter (6 aa long) randomised peptide insert in loop 1 so as to avoid potential steric hindrance with loop 2, where we used a 12aa -long randomised peptide.
Each library comprised ∼107 unique peptide aptamer sequences. We used two very different baits for this experiment: the 130aa-long POZ domain from the transcriptional repressor BCL6, a bait we have previously characterised (Chattopadhyay et al., 2006), and a newly identified 18aa-long peptide (MecA_3) from the penicillin binding protein 2' from methicillin-resistant S. aureus (Flanagan, Cockell et al., manuscript in preparation). Interestingly, although Scannin binders to both of these baits were identified successfully, they were not identified from all libraries. For example, binders to BCL6 were identified in the loop 2 and the loop 12 library, whereas MecA binders were predominantly derived from the loop 1 library. Figure 6 shows a yeast two-hybrid interaction grid demonstrating that binders we identified to each protein are specific for that protein. The sequences of these Scannins may be obtained on request.
SQT can be efficiently expressed in mammalian cells
Although peptide aptamers and other artificial binders are most commonly selected in vitro (e.g. by phage display, ribosome display) or in screens involving simple organisms (e.g. yeast), their applications will very often involve expression in cells derived from multi-cellular organisms. In order to assess the ability of SQT to be expressed in cultured human cells, MCF-7 breast cancer cells were transduced with viral vectors carrying open reading frames for SQT or SQM. After selecting for stably transduced cells, scaffold protein expression was induced for 24 h and measured by western blotting with an anti-Stefin A antibody (Fig. 7A). Strikingly, SQT levels were ∼7-fold higher than SQM protein levels. RT–PCR showed that mRNA levels were equal for both scaffolds, indicating that the higher protein levels of SQT are due to increased translation efficiency and/or protein stability compared with SQM (Fig. 7A).
We next asked whether SQT can present peptides for biological interaction within human cells. To this end, we inserted 26 amino acids which form the BH3 alpha-helix in the weakly pro-apoptotic protein Noxa (Chen et al., 2005) into the N-terminus of SQT, to create pepNoxa. CD spectroscopy indicated that the peptide sequence was indeed constrained as an alpha helix, as the SQT reference spectrum (which is indicative of predominantly beta-sheet) switches to a characteristically alpha-helical spectrum on peptide insertion (Fig. 7B). In cells, the Noxa BH3 alpha-helix is thought to interact with the anti-apoptotic protein Mcl-1 (Liu et al., 2003). We immuno-precipitated pepNoxa from A375 cells and asked whether anti-apoptotic proteins including Mcl-1 were present (Fig. 7C). Whereas Mcl-1 was bound by pepNoxa as expected, the related anti-apoptotic factors Bcl-2 and Bcl-xl were not. These data show that peptide aptamers in SQT can efficiently compete against natural interactors in vivo, a key requirement for their use as tools to dissect the role of specific protein interactions in cell biology. They also show that peptides presented by the SQT scaffold are capable of highly specific discrimination between proteins related in both sequence and function. These findings further support the hypothesis that the SQT scaffold is capable of presenting peptides for specific interaction with a target protein.
We have developed a Stefin A-derived scaffold with three insertion sites instead of one. In addition to increasing the surface area of interaction and introducing a combinatorial nature to the scaffold, we find that a given peptide may be constrained differently at each position in each version of the scaffold. This work thus extends the scope and versatility of our peptide aptamer scaffold. As with all engineered scaffold proteins, scaffold stability must not be compromised by peptide insertion.
Re-engineering loop 2
When the idea to move from the monovalent STM to the trivalent SQM was originally conceived, the decision was made to replace loop 2 of Stefin A with inserted peptides by introducing a second RsrII restriction site in the open reading frame corresponding to the C-terminal end of the loop (Hoffmann et al., 2010). This resulted in a leucine-to-arginine substitution as well as a threonine-to-serine substitution in the β-strand adjacent to loop 2. The latter replacement (T83S) is conservative and unlikely to cause major disruption to the stability of the fold. It was conceivable, however, that the introduction of the charged arginine could be detrimental, especially, since it replaces a branched, non-polar residue (leucine), commonly found in β-strands (Solovyev and Salamov, 1994). SQT differs from SQM only in that the second RsrII site has been moved so that the corresponding amino acid substitutions are moved to the base of loop 2 leading to substitutions E78A and L80R. We hypothesised that this would result in a more stable scaffold than SQM, as the β-strand should no longer be affected. Using CD spectroscopy to monitor thermal stability, we find that SQT has a very similar heat resistance to STM, indicating that the mutations in the β-strand caused the instability of SQM. The temperature scan of SQT revealed a two-phase transition that might indicate a heat-induced dimer-formation through three-dimensional domain swapping as seen in Stefin A by Jerala and Zerovnik (1999). A possible explanation could be that the V48L mutation in SQT is not as effective as the V48D mutation in STM in abolishing dimer formation due to the stronger ß-breaking propensity of asparagine compared with lysine (Japelj et al., 2004).
SQT is a robust and versatile peptide aptamer scaffold
In SQM peptide insertions into loop 1 and 2 individually were generally well tolerated, but insertions in combination or into the N-terminus caused structural problems (Hoffmann et al., 2010). For SQT we report that epitope tags inserted into individual sites or in combination were tolerated without affecting secondary structure or protein yield for any of the combinations tried. Even the N-terminal insertions, which were particularly harmful in SQM, did not have a negative effect on the folding of SQT. This is somewhat surprising, but could be explained if the mutations in the C-terminal β-strand of the SQM scaffold predispose it to destabilisation by changes at the N-terminus. Effects of changes in protruding loops on distant structural elements of a protein have been reported. For example, Dun et al. (1999) and Clarke et al. (2006) showed that a mutation in a loop of the hERG potassium channel exerts an allosteric effect on the gating and permeation properties of the channel. Jenko et al., (2003) have shown the N- and C-termini of Stefin A to be in close spatial proximity. An interaction between the two regions might then contribute to the stability of the scaffold providing a mechanism to explain the effects of the ß-sheet mutation in triggering detrimental effects of N-terminal insertions into SQM, although other explanations that involve interactions between the N- and C-termini and other elements of the ß-sheet are also possible.
It was interesting to note that, although a model Scannin could be exported to the surface of phage via both the Sec and the Tat pathways (Fig. 3), export via the Sec pathway may be more efficient. Because our data indicate that SQT-based Scannins are expressed as soluble proteins to high levels in bacteria (e.g. Fig. 2), we expected that tat-mediated transport would be favoured over the Sec pathway that requires protein unfolding for translocation, and refolding in the periplasm. If SQT-based Scannins exported via the Sec pathway are found to be functionally folded, this would indicate that chaperones are not required for periplasmic re-folding of the protein. We also noted that although both the PelB and PhoA leader sequences were compatible with the expression of functional pIII fusions, the use of the OmpA leader sequence was not: induction of OmpA leader-SQT-pIII fusion protein expression led to rapid growth arrest of bacterial cultures, indicating that the fusion was toxic, perhaps by leading to a specific block in the OmpA export pathway. This is, to our knowledge, the first example of a protein that shows differential effects when targeted to the Sec pathway via distinct leader sequences, and indicates that the ‘Sec pathway’ comprises more than one path. Further work will be required to investigate this.
SQT can present binding peptides for target interaction
We show that the Myc epitope tag is available for interaction with an anti-Myc antibody from the N-terminus, loop 1 and loop 2, respectively. When the peptide is in the N-terminus or loop 2, insertions into the remaining two positions do not inhibit interaction with the antibody. This is an important finding, as the ability to use several insertion sites in one scaffold was the premise for developing SQT as well as SQM. These conclusions can be extended beyond the Myc epitope tag on the basis of our microarray results, where presentation of the AU1 and HA tag as well as the Myc tag are explored. Although intended to serve only as a qualitative assay for monitoring multiple protein interactions in parallel (Fig. 4B), this platform can be used semi-quantitatively, by normalising the interaction signal to the amount of spotted Scannin at each feature on the array, as shown in Fig. 4C. Where subtle or unexpected differences arise, as was the case for the AU1 epitope tag, we found it useful to turn to a more quantitative SPR-based assay (Fig. 4D). Interestingly, as is made clear with the example of the AU1 tag, binding affinities to the target protein vary strongly depending on the insertion site. These findings highlight the value in developing a multivalent peptide aptamer scaffold as the number of protein conformations that can be explored is increased. The simultaneous insertion of two AU1 tags into the loop 1 and loop 2 positions did not result in an increase in affinity beyond that seen with the tag in loop 2 alone. This so-called avidity effect is usually achieved when a bivalent IgG molecule binds to two identical epitopes on a target protein with each ‘arm’ of the antibody engaging one epitope each (Oda and Azuma, 2000). The lack of an avidity effect in our assay is most probably due to dissimilar spacing between loop 1 and loop 2 of SQT and the two arms of the anti-AU1 IgG molecule.
It has previously been shown that peptides pep2, 6, 9 and 10m retain CDK2-binding activity when transferred from thioredoxin to STM (Woodman et al., 2005). Here, using a microarray platform, we show that in loop 2 of SQT (and SQM) this activity is maintained. Differences in the way the inserts are constrained lead to differential binding affinities, with pep10m in SQT giving a considerably elevated fluorescence signal compared with STM. These data indicate that highly related alternative versions of a single engineered scaffold are capable of imposing very different conformational constraints on a given peptide. In addition, this use of a Scannin microarray may prove useful in future experiments, for example, when seeking to improve the binding affinity or specificity of a particular Scannin.
SQT as a functional scaffold in yeast and human cells
The yeast two-hybrid screening method is ideally suited for the identification of specific protein interactors in a competitive intracellular environment, and has been used successfully with other peptide aptamer libraries (Colas et al., 1996; Schneider et al., 1999). Here we show that SQT is capable of presenting a large library of peptides in yeast, thus further corroborating the idea that it is a viable peptide aptamer scaffold. Interestingly, when we used a long protein as a bait, Scannins could be identified from either loop 2 or loop 12 libraries, whereas the short peptide bait we used yielded hits almost entirely from the loop 1 library. Given that we have used only two baits so far, the importance of this observation is impossible to determine, but it does indicate that different libraries will be better suited to screening for Scannins that bind to specific baits.
In human cells Stefin A can be over-expressed to high levels (Takahashi et al., 2001). This is a highly desirable feature for a peptide aptamer scaffold, as it facilitates biological studies in cell culture. Using an inducible expression system we have shown that SQT can achieve higher protein levels in cells than SQM in at least one human cell line. We speculate that SQM's relative lack of stability leads to accelerated degradation of the molecule, as partially unfolded proteins are susceptible to degradation (Prakash et al., 2004; Marques et al., 2006). In contrast to SQM, where peptides in the N-terminal insertion site proved disruptive to the scaffold, we show here that in SQT these insertions are not only well tolerated structurally, but also interactions with a target are possible. By inserting the Noxa BH3 helix into this site we were able to mimic Noxa exactly, which has been shown to bind Mcl-1, but not the related Bcl-2 and Bcl-xl (Chen et al., 2005).
The results obtained in the course of this study indicate that SQT is a significant improvement over the previously described SQM scaffold in several respects. Firstly, whereas for SQM we had to reject the N-terminus as a potential site for the construction of a combinatorial peptide aptamer library, in SQT this becomes a possibility, as peptide insertions into this site do not disrupt the stability of the scaffold. Secondly, SQT is significantly more heat stable than SQM and protein yield following bacterial expression is superior. Finally, in mammalian cells SQT can be expressed at several fold higher levels than SQM, making the former a useful tool for biological studies in cell culture. Importantly, we find that the three peptide insertion sites display peptides in different ways, whether between SQM and SQT, or even within SQT itself. In combination, these findings establish SQT as a legitimate peptide aptamer scaffold of high structural integrity and functional diversity, and indicate that each version of the scaffold may be useful when considering a large number of target proteins.
This work was supported by BBSRC Grant BB/F011296/1 to J.J.D. and P.K.F. and grant number 7508 from the Leukaemia Research Fund (to S.D.W. and P.K.F.). L.K.J.S. acknowledges support from AstraZeneca (to P.K.F.). D.C.T. and M.A.K. were supported by Yorkshire Cancer Research grant number L346. Work on PBP2' was performed as part of the EPSRC-funded AptaMEMS-ID project (http://aptamems.ncl.ac.uk/; EP/G061394/1 to Calum McNeil, Institute of Cellular Medicine, University of Newcastle).
Conflict of interest: P.K.F. is a shareholder and Director of, and E.G. and C.T. are employees of, Aptuscan Ltd. The sequences of the Scannins described here represent potentially commercially sensitive information that will be disclosed under CDA upon request to P.K.F.
We would like to thank Marc Vidal (Dana-Farber Cancer Institute) for generously sharing yeast strains and plasmids. Circular dichroism spectroscopy was performed at the Astbury Center for Structural Molecular Biology (Leeds University). We would also like to thank Keith Ainley (from Prof. Sheena Radford's group) for introducing us to the spectropolarimeter and giving helpful advice in using the equipment. Finally, we thank Joan Segura Mora for help with the R software package. Q.S. and P.K.F. gratefully acknowledge generous support provided by Professor Peter Selby (LIMM).