To test this approach, a candidate synbody was constructed using two peptides, AHKVVPQRQIRHAYNRYGSG and FRGWAHIFFGPHVIYRGGSG, referred to as peptides 1 and 2, respectively. The sequences of the first 17 aa of these peptides were generated using a random number generator and three constant positions on the C-terminus that served as a spacer between the linker scaffold and the active portion of each peptide. A divergent synthetic method was used in which peptide 1 was synthesized from C to N terminus from the α-amine of lysine and peptide 2 was synthesized from the ε-amine of lysine (, Materials and Methods S1
, Figure S1
). The synbody was tagged on the C-terminus with biotin and screened against a Life Technologies ProtoArray™. ProtoArrays are nitrocellulose coated slides onto which 8,303 recombinant human proteins have been spotted. After a 2-hour incubation, the array was washed and bound synbody was detected using fluorescently labeled streptavidin. The synbody bound few proteins on the array as seen in the distribution plot of the top 50 background-subtracted spots (). Analysis of the brightest spots revealed that the synbody bound several different proteins including PCCA, CASZ1, GRP58, NOB1, and AKT1 (). It should be noted that AKT1, FBXO21, PDE7B, and FBXO4 have multiple variants (full-length, partial length, or transcript variants) present on the array, hence the appearance of the protein kinase, RAC-alpha serine/threonine protein kinase (AKT1) twice in the top 10 proteins bound by the synbody. It is likely that some of the proteins bound might be false positives as binding levels on protein arrays can be highly variable, thought to be caused by partial to complete denaturation of the immobilized protein that arises from protein printing and array storage 
. We choose two proteins, AKT1 and protein disulfide-isomerase A3 (GRP58), for further analysis as recombinant proteins are commercially available.
Synbody binding to proteins on protein array.
The synbody (100nM concentration) was screened by Surface Plasmon Resonance (SPR) against immobilized AKT1 and GRP58 on a Biacore T-100 SPR. The synbody appeared to have high affinity for AKT1 and low affinity for GRP58 (Figure S2
). To accurately determine the binding kinetics of the synbody-AKT1 complex, several concentrations of synbody were injected over a low-density surface of biotin-labeled AKT1 captured on a Neutratvidin SPR chip using a Biacore A-100. The synbody bound AKT1 and the resulting sensorgram was fit with a 1
1 binding model to reveal that the synbody had a rapid association rate, ka
and a slow dissociation rate, kd
for a dissociation constant, Kd
1.5 nM (). A residual plot was constructed from the difference between the measured sensorgram and the 1
1 fit of the data at each point in time and the residuals were randomly distributed around zero, which indicated that the 1
1 binding model was an appropriate model. The synbody's binding affinity is in the same range as that of a monoclonal antibody binding its cognate antigen 
but was achieved in a single screening step with no further evolution to achieve nanomolar binding affinity.
Characterization of synbody binding to AKT1.
The synbody was tested for its ability to bind native protein in solution. A variant of the synbody was prepared in which the C-terminal Cys was replaced with a Lys-Biotin. The synbody was used in an immunoprecipitation (IP) experiment in which it was bound to streptavidin magnetic beads and incubated overnight with 89 nM to 0.7 nM solutions of AKT1. The beads were washed extensively and the bound protein was eluted. Samples were run on a SDS-PAGE gel and detected by Western Blot with an anti-AKT1 monoclonal antibody (). As can be seen, the synbody bound AKT1 from solution.
In order to validate the Kd
determined by SPR, an additional IP was performed using 35
S-labeled AKT1 produced by in vitro translation (IVT). The translation mixture contained 12 µg of AKT1 in the presence of 400 µg of total protein and a concentration series of 35
S-labeled AKT1 from 41.5 nM to 80 pM was prepared by dilution of the 35-S-labeled AKT1 IVT mixture. The synbody was then used to precipitate the 35
S-labeled AKT1 in the same manner as before. Bound AKT1 was quantified using liquid scintillation counting and plotted as a function of AKT1 concentration. The resulting isotherm was fit to a 1
1 binding model using GraphPad Prism and showed the synbody had a Kd
of 4.9±1.1 nM, in agreement with the SPR results (). It should be noted that this experiment was performed in the presence of a cell lysate in which AKT1 was approximately 3% of the total protein in solution, indicating that the synbody specifically binds AKT1 from a complex background.
As a preliminary test of specificity, 200 ng of AKT1 (33.2 nM) was spiked into increasing amounts of cell lysate prepared from un-stimulated A549 lung epithelial cells and streptavidin beads coated with 25 pmols of synbody (7.6
1 synbody-to-AKT1 ratio) were used to pull down AKT1 using the same IP procedure and Western Blot protocol as before. As can be seen in , the synbody successfully pulled AKT1 out of the cell lysate in the presence of an increasing concentration of A549 lysate with little reduction in AKT1 signal intensity. This result indicates that the synbody has sufficient affinity and specificity for AKT1 to precipitate it in the presence of 2,500 fold excess cell extract. Note that in lane 6 at higher A549 lysate concentration, the synbody precipitates the very low level endogenous AKT1 from the un-stimulated cell lysate. This result was confirmed in lane 8 where the synbody was used to precipitate AKT1 from the un-stimulated A549 lysate when no recombinant AKT1 was present.
However, to more rigorously test the specificity of the synbody, we reduced the molar ratio of synbody to protein to approximately 2 to 1 and performed an IP to examine other proteins that were non-specifically bound by the synbody. In this experiment, 400 ng of AKT1 (66.4 nM) was spiked into 100 and 500 µg of A549 cell lysate pre-cleared on streptavidin beads (Figure S3
) and precipitated using streptavidin beads coated with 15 pmols of synbody. Synbody coated beads and uncoated streptavidin beads were incubated overnight, washed, and bound proteins were eluted. Elution samples were split with half of the sample analyzed by silver stain and half analyzed by Western Blot (). From the silver stained gel, it can be seen AKT1 was precipitated along with two other prominent proteins from the 500 µg (5mg/ml) sample. This result indicates the synbody has useful, though not perfect specificity.
When the synbody was screened on the ProtoArray, it showed little binding to AKT2, or AKT3 on the array, which share 92% and 87% sequence identity with AKT1, respectively. This apparent specificity was investigated further by performing an IP using 250 ng of each isoform of AKT spiked into 500 µg of A-549 cell lysate. As can be seen in , the synbody selectively precipitates AKT1 while precipitating relatively little AKT2 or AKT3, which suggests that the synbody makes unique contacts on AKT1 that are not present on either AKT2 or AKT3. This result is in contrast to an anti-AKT1 monoclonal antibody that recognized all three AKT variants on the protein array (Figure S4
The simplicity in creating this high affinity ligand to AKT1 was based on the hypothesis that each unstructured peptide bound a different site on AKT1 and that the synbody affinity was driven by the product of the two peptide's affinity 
. However, it is possible that the high-affinity binding of just one of the peptides drives the synbody affinity. To test if the binding was driven by a single peptide, we used SPR to screen the individual peptides that made up each arm of the synbody against immobilized AKT1. Each peptide was synthesized on an automated peptide synthesizer and purified to >95% purity by HPLC with confirmation by MALDI-TOF-MS. Each peptide was analyzed by SPR against immobilized AKT1 and the individual peptides bound AKT1 with micromolar affinity (Kd
~2 to 20 µM). Representative sensorgrams are shown in Figure S5
and are indicative of a low affinity interaction between the peptides and AKT1. These data imply that the high affinity exhibited by the synbody is not simply driven by a single high affinity peptide but could be the consequence of a bivalent interaction between each peptide and AKT1.
As an orthogonal assessment of whether the peptides bind different sites on AKT1, each peptide was crosslinked to AKT1 in vitro
using a commercially available set of deuterated / non-deuterated crosslinkers (BS3
)). The crosslinkers chemically conjugate primary amines that are in close proximity to one another (11.4 Å) and the crosslinked complex is then subjected to enzymatic digestion and analysis by mass spectrometry. Crosslinked peptides are identified by the presence of a pair of ions that are separated by 4 mass units indicating crosslinking with both non-deuterated and deuterated crosslinkers. Peptide 1 has two possible crosslinking sites (N-terminus and K at position 3) and peptide 2 has one possible crosslinking site at the N-terminus. A 10 µM solution of each peptide was added to 2.5 µM AKT1 and incubated for 1 hour prior to the addition of a 100 µM solution of a 1
1 mixture of crosslinkers. After crosslinking, the mixture was digested with trypsin and analyzed by MALDI-TOF-MS. It was found that peptide 1 crosslinked to a region of AKT1 that spans the Protein Kinase (PK) and AGC Kinase domain while peptide 2 crosslinked to multiple regions on AKT1 (, , Table S1
), with the majority of crosslinking occurring in the PK domain, in the region of the ATP binding site (residue 179) and the active site (residue 274). The location of each AKT1 fragment was mapped onto a crystal structure of residues 139–480 of AKT1 (PDB #3CQU) (). As amine reactive cross-linkers will only react with the side chains of surface exposed lysines, we analyzed the identified AKT1 fragments for surface exposed lysines and found that each peptide had 1 exposed lysine. Peptide 1 most likely cross-linked to 419
Lys given that the side chain is exposed while peptide 2 likely cross-linked to 182
Lys, and 481
The two component peptides of bind different sites on AKT1.
The relative intensity data for each peptide 2-AKT1 fragment and the close proximity of 182
Lys and 276
Lys, suggests that the primary binding site for peptide 2 is likely on the PK domain and that a secondary binding site lies on the AGC domain. The locations of the cross-linked lysines indicate that it is possible for the N-terminal region of each peptide to bind AKT1 when linked as a bivalent synbody. These data and the micromolar binding affinity of the individual peptides suggest that the high-affinity and high specificity of the synbody is the product of a bivalent interaction. Additionally, as the binding site of peptide 1 is in a region of low homology between AKT1, AKT2, and AKT3 (Figure S6
), the selectivity for AKT1 is likely a product of a bivalent interaction.
To test if this method to produce synbodies backwards is generally applicable and amenable to HTP parallel processing, 8 other synbodies were synthesized and screened on ProtoArrays (Materials and Methods S1
, Figure S7
, Figure S8
, Figure S9
, Table S2
). Additionally, we screened 4 monoclonal antibodies to compare the synbody binding profiles to those of high-affinity antibodies. A heat map was constructed from the median normalized fluorescent data and in each case the synbody produced a unique binding profile suggesting that each synbody behaves as a unique binding entity (). Each synbody bound several proteins. It should be noted that monoclonal antibodies often bind several proteins when screened on protein arrays 
, as was the case for the 4 monoclonal antibodies that we screened.
Discovery and characterization of a synbody to ABL1.
From the heat map, some proteins, such as the proto-oncogene tyrosine kinase ABL1, appeared to bind to multiple synbodies (). We performed an additional screen of 5 synbodies that bound ABL1 on the protein array plus 2 negative control synbodies that were not run on the ProtoArray to test if the backwards method could be used to identify a functional synbody to ABL1. We purchased recombinant full-length ABL1 and used the same IP protocol as before to precipitate 1µg of ABL1 (~700 ng of full-length ABL1 for a concentration of ~54 nM) in a phosphate buffered saline solution. The silver stained gel showed that only synbody 9 pulled down ABL1 (). Subsequent IP assays that used a concentration gradient of ABL1 demonstrated that synbody 9 had high affinity for ABL1, Kd
12 nM (). We tested the specificity of synbody 9 for ABL1 in a pull-down assay in which 800 ng of ABL1 (though by western blot only a small portion of this was full-length protein) was spiked into 100 or 500 µg of pre-cleared A549 lysate (1 mg/mL and 5 mg/mL protein concentration) and precipitated with synbody 9. Fifteen pmols of biotinylated synbody 9 was captured on streptavidin coated beads (~3
1 synbody-to-ABL1 ratio) and the proteins precipitated by the synbody were analyzed by silver stain and Western Blot (Figure S10
). The synbody precipitated full-length ABL1 in the presence of >150 and >850 fold excess protein. Note that the synbody only appears to bind full-length protein. The Western blot indicates the synbody also precipitated one other prominent protein from the extract. These results demonstrate that a high-affinity, moderate specificity synbody can be readily generated to another protein using this procedure.
Generating ligands to the human and other proteomes is a major challenge. The unique features of synbodies in combination with protein arrays may offer a solution. Synbodies are synthesized by standard peptide chemistry, enabling a large number of them to be produced in parallel. The ligand screening step is rapid and can be run in parallel, but the major cost in this system as demonstrated is the protein arrays. This limitation could be overcome by multiplexing the synbodies on each array or using other protein array technologies 
. It should be noted that these synbodies were generated against native protein. While we have found that they do function in Western blots, synbodies optimized for functioning in Western blots could be isolated in screens against denatured proteins.
The two initial synbodies discovered by this method demonstrated binding affinities in the range of antibodies with specificities approaching those of commercial antibodies. These synbodies have low sequence diversity, as synbody 9 and synbody 1 share the same stretch of 20 amino acids; yet have completely different protein targets. This echoes recent work that has shown that low complexity ligand libraries that are highly biased for Tyr and Ser can be used to produce high affinity and high specificity ligands for a variety of protein targets 
. By incorporating these findings into an improved design of the peptides that make up the synbody, it should be possible to improve the binding affinity and binding specificity. Additionally, we are developing an affinity optimization method for the peptide arms of the synbody (M. Greving, N. Woodbury, P.E.B, C.W.D and S.A.J., unpublished results) that could also be used to improve affinity and specificity. With these improvements to the first generation of synbodies, we believe that the backward process of ligand generation could offer the possibility of creating binding agents to the human proteome.