Final Report Summary - TARGETBINDER (Target binders)
Executive Summary:
Most of the objectives and milestones within the project have been achieved, with the only exceptions being deliverable D46 and milestone MS33. Especially noteworthy, we achieved our main objective: we found peptide binders by CIS display for the protein Grb2 without employing any previous knowledge, and we validated these binders in the array format and also by measuring the binding affinities. Indeed, we could verify that every SH3 domain from Grb2 binds to one clearly defined “signature” that could be used to screen the databases for candidate human protein binders. Experiments of this kind might one day elucidate the information processing that takes place in a eukaryotic cell. Other highlights were technical achievements: BIA designed, manufactured, and tested a device that is capable of measuring the affinities for thousands of peptide binders in parallel. KIT and PPP developed a new semi-automated machine that synthesizes peptide arrays in unprecedented density and quality. Both technical achievements are or will be commercialized within our spin-off companies PEPperPRINT (www.PEPperPRINT.com) and Biametrics (www.Biametrics.com). There were difficulties though. The expression of some target proteins proved to be difficult due to problems with proteolytic activity or similar, but the TARGETBINDER consortium provided other interesting target proteins instead that were used to screen for peptide binders. Another problem, however, seems to be of more fundamental nature. Although we employed CIS display for initial screens with its huge peptide libraries (> 10(+13) different peptides), we did not find high affinity binders for the majority of target proteins, only specialized protein domains, e.g. SH2, SH3, and some other protein domains seem to have been designed by evolution to bind to linear short peptides. One reason for this failure might lie in the entropic penalty that linear peptides come along with their many different 3D structures that they could fold into: Only a few of these will be complementary to a binding partners, which directly translates into low affinity binding. Therefore, the consortium tried to find binders against circular peptides. Indeed, ISO’s CIS display screens probably worked with S=S-bond circularized peptides, but we might have failed to synthesize the very same circular peptides also in the array format. Due to much higher concentration of the peptides that are tethered to the array’s surface these might have preferentially formed intermolecular S=S-bonds.
Project Context and Objectives:
Objectives for the whole project:
The ground breaking development of very high-density oligonucleotide arrays (Fodor et al., 1991, Science 251, 767) boosted the field of genomics both in scientific and in economic terms. The main drivers of this development were the US-based companies Affymetrix and Illumina with a combined annual turnover of nearly 1 billion $ in 2009. Recently, the members of this consortium developed several tools and technologies to easily generate (Beyer et al., 2007, Science 318; Stadler et al., 2008, Angew. Chem. Int. Ed. 47, 7132; Odegrip et al., 2004, PNAS 101, 2806) and analyse (Proll et al., 2007, J. Chrom. A 1161, 2-8; Pröll et al., 2005, Anal. Bioanal. Chem. 382, 1889) very high-complexity peptide libraries. Within the TARGETBINDER consortium, we wanted to combine these technologies, and we expected that our procedures might boost the field of proteomics in a similar way as the lithographic technologies did with the field of genomics. Our goal was to develop a new peptide discovery platform based in the EU that should enable high-throughput identification of peptides that bind to interesting target proteins. Furthermore, the platform should allow us to characterize and to optimize these potential therapeutics at a significant time and cost advantage when compared to other biologics. We expected that such a platform might have a huge commercial and scientific impact.
Objectives broken down to work packages:
We wanted to further develop our tools and technologies in order to achieve the following final goals:
(I.) Within WP1 participants KIT, CBL, MS, and TUV wanted to develop a variant of our solid-material-based-synthesis-method (Breitling et al., 2009, Molecular BioSystems 5, 224; Stadler et al., 2008, Angew. Chem. Int. Ed. 47, 7132) that should allow us to synthesize up to 1 million different peptides on a glass slide for chemical costs of ~50€. In order to achieve this goal, we had to find a method that reliably covers >99% of the central area of every synthesis spot with a matrix material that comprises activated amino acid derivatives for peptide synthesis, which should result into improved quality over a laser printout that usually covers only >95% of the central area. When the glass slide with its patterned amino acid derivatives is heated up in an oven, this melting step frees hitherto immobilized amino acids to diffuse to the surface of the synthesis slide where they couple to free amino groups. This combinatorial synthesis procedure is identical to the 40-years-old Merrifield synthesis, with the only difference being an intermittent “freezing” of the coupling reaction at room temperature. Moreover, we wanted to advance this technology up to a level of an automated machine, which should allow us to commercialize this technology within PPP.
(II.) Within WP2 mainly participant BIA wanted to advance BIA’s labelling-free detection method (Proll et al., 2007, J. Chrom. A 1161, 2-8; Pröll et al., 2005, Anal. Bioanal. Chem. 382, 1889) in order to use it for parallel readout of binding affinities for up to 10,000 peptide spots per cm(+2). First experiments were meant to find out how sensitive the method could be, and to identify suitable glass substrates and reliable fluidic cells. Furthermore, BIA wanted to design, construct, and test a device that could be used to perform these affinity measurements in the array format, and, in addition, to write an automated program that should “translate” measured binding curves into affinity constants. We planned to commercialize this device within SMA Biametrics (www.biametrics.com).
(III.) Within WP3 ISO, DKFZ, and PPP wanted to combine acellular transcriptional peptide/protein display technique (“CIS display”; Odegrip et al., 2004, PNAS 101, 2806), high-density peptide arrays (manufactured by PPP, DKFZ, KIT; Angew. Chem. Int. Ed. 47, 7132), and parallel labelling-free detection of binders (iRIfS method developed by BIA; Proll et al., 2007, J. Chrom. A 1161, 2-8; Pröll et al., 2005, Anal. Bioanal. Chem. 382, 1889) in order to develop a standardized, fast, and reliable high-throughput procedure to find peptide binders against, hopefully, any pharmaceutically interesting target protein. The rationale behind this approach was to first use the huge number of 10(+13) to 10(+14) different peptides that are displayed on a ribosome when using the CIS display technique, then use next generation sequencing to generate a comprehensive list of all peptides that possibly bound to a target protein (expressed and purified within WP4), and, finally, validate candidate binders by synthesizing several thousand of them in the array format with PPP’s particle based synthesis method, and staining these arrays with the same target proteins that were used to select displayed peptides.
(IV.) Within WP3 APO, OXF, DKFZ, and PPP wanted to express, purify, and characterize interesting target proteins, and, in addition, to validate our novel high-throughput procedure by characterizing peptide binders. If successful, such an approach might supply big pharmaceutical companies with peptide-based lead structures that bind to target proteins.
Project Results:
Overview: Within the TARGETBINDER project, we advanced several technologies:
WP1: KIT together with CBL, TUV, and SME MS advanced several variants of our solid-material-based synthesis method up to the level of an automated machine that allows us now to synthesize very high-density peptide arrays at very moderate cost and in high quality. We started the process to commercialize the automated cLIFT array synthesizer within SME PEPperPRINT.
WP2: SME BIA designed, manufactured, and tested a device that is capable of measuring the affinities for thousands of peptide binders in parallel, and they wrote automated programs to “translate” their measurements, e.g. into binding affinities. SME Biametrics started to commercialize this device.
WP3: SME ISO implemented a next generation sequencer, and they streamlined their CIS display technology to better handle the massive amount of data that is generated, e.g. by writing automated programs to compare found peptide sequences with each other and with databases. These additions to their CIS display technology certainly help SME Isogenica in their commercialization efforts.
WP4: OXF and APO expressed, purified, and characterized several proteins that were previously pinpointed by them and by others to be excellent targets, which means that a binding lead structure, e.g. a peptide might lead to a therapeutic molecule.
Combining technologies to find peptide binders without any previous knowledge (WP3 and WP4):
The partners combined their technologies, and, indeed, especially noteworthy, we found several Grb2 binding peptides that proved to bind to Grb2 (supplied by OXF) in higher affinity when compared to previously known Grb2 binders. We did that by first screening >10(+13) CIS displayed peptides, then using next generation sequencing to obtain a comrehensive list of candidate binders (done by ISO), next we validated these binders by synthesizing them in peptide array format (done by PPP) and staining them with target proteins (done by DKFZ). Next, we systematically varied the peptide’s sequence to find out the “binding signature”, and, finally, we determined affinities using three different methods (done by BIA, DKFZ, and OXF). Thereby, we achieved our main objective, namely finding peptide binders without using any previously obtained knowledge (milestones MS31 and MS43). We think that some specialized protein domains, e.g. SH2- and SH3-domains have been designed by evolution to bind to linear short peptides, thereby doing information processing. Peptide arrays might be especially well suited to find out which parts of the proteins that are known to form complexes inside a eukaryotic cell code for the switchable peptide binders, e.g. in SH2 domains binding is switched by adding or removing a phosphoryl group, thereby leading to a re-grouping of complexes when a kinase is activated through an external signal.
However, we failed to find high affinity binders for most other target proteins. One reason for this failure might lie in the entropic penalty that linear peptides come along with their many different 3D structures that they could fold into: Only a few of these will be complementary to a binding partner, which directly translates into low affinity binding. Therefore, the consortium tried to find binders against circular peptides, but we might have failed to synthesize the very same circular peptides also in the array format. We will re-address the question of finding peptide binders against any target protein, this time, however, employing circular peptides right from the beginning.
WP1, main S & T results:
All the deliverables and milestones due in WP1 within the TARGETBINDER project have been achieved, with only minor deviations in the delivery date, spent resources and person months.
Laser-induced-melting method:
One highlight clearly is the publication of our laser-induced melting method in a high ranking journal (see D12 appendix and MS11; Maerkle et al., 2014, Advanced Materials 26, 3730-3734). In addition to the data shown in this publication, we added more data to a novel patent application (PCT/EP2013/001141) that describes the laser-induced melting method (see Fig. 1) and additional variants of our particle based synthesis method (see Figs. 2, 3).
Array synthesis based on microstructured glass slides:
When scrutinizing these additional variants of our solid-material-based synthesis method, we realized that these variants might be better suited to bring us close to the overall goal stated for WP1, namely synthesize 1 million peptides per glass slide at very moderate cost and in high quality. One of these additional variants exploits KIT’s unique capacity to microstructure glass slides (Fig. 4; by reactive ion etching). Next, 1 million cavities per cm(+2) from a donor slide are all filled with only one sort of amino acid particles (e.g. Fmoc-Ala-OPfP ester embedded in the same matrix material that is used for laser printed amino acid particles; Stadler et al., 2008, Angew Chem Int. Ed. 120:7241–7244). This is done by simply gently wiping them into the cavities with a smooth towel, which nearly always results into 100% filling of the cavities. Next, particles from selected cavities of the donor slide are transported to the cavities of an aligned synthesis glass slide that also features 1 million cavities per cm(+2) (Fig. 5), which is done with short laser pulses, and using a 2D laser scanning system (Fig. 6). This step is repeated with additional particles that comprise other amino acid derivatives, which results into a microstructured synthesis glass slide that is densely patterned with solid particles that comprise the 20 different amino acids, and, eventually, additional ones (post-translational modified amino acid side chains, artificial catalytic centres, D-amino acids, moieties that generate cyclic amino acids …). When these patterning steps are finished, the synthesis glass slide is simply heated up which frees hitherto immobilized amino acids to diffuse and couple to the surface. Repeating these steps 15x generates an array of 1 Million 15meric peptides per cm(+2). Indeed, we could show that we could synthesize 10x more peptides per area (i.e. 1 Million peptides per cm(+2)) than originally planned when employing glass slides with small microcavities (Fig. 5). Moreover, it should be straightforward to miniaturize the cavities even further – there have been numerous publications to reliably fill even nanoscale cavities with different particles. However, a drawback of that particular method is the time consuming alignment step. The expensive machinery that is used to do such an alignment, e.g. in chip technology up to now prohibited us from developing a fully automated machine. There is a simple solution to this alignment problem, however. Currently, we use a donor slide with smaller cavities that is placed on top of a synthesis slide with larger cavities. Then, the only alignment that has to be done links the larger cavities of the microstructured synthesis slide with the 2D laser scanning system (the same that was used for the laser-induced-melting procedure).
Laser-induced transfer of materials (cLIFT method):
Yet another approach was especially easy to automate. We therefore focused our work on building a fully automated cLIFT-array-synthesizer (Fig. 7). This machine uses the very same 2D laser scanning system that was described above, but this time to puncture out small pieces of a material that all comprise a suitable amino acid derivative that was spin coated as a thin layer onto a flexible foil (Fig. 3; combinatorial laser induced forward transfer of materials, cLIFT). By repeating these procedures with additional amino acid derivatives, an array of peptides is generated. Obviously, the cLIFT-array-synthesizer is much more reliable in the patterning steps in terms of amount and morphology of transferred material when compared to laser printed particles, which – quite surprisingly for us – resulted into arrays with superior quality in terms of staining signals per spot (Fig. 8). Moreover, the only exact positioning system that is needed for this machine is integrated within a commercially available and ready-to-go 2D laser scanning arm module that regularly addresses two reference points on the glass slide with the growing peptide array to calibrate its position ±2µm (built in self adjusting and correcting). Therefore, we hardly need any efforts to position the thin layers with our amino acid containing material on top of the support (see Fig. 3 for a schematic explanation) that is used for peptide synthesis – we only have to shield the machine from dust particles to assure that the donor layer comes close enough to the glass surface with its growing peptides when exchanging the slide coated with material #1 for another slide coated with material #2. This is done with an already well advanced slide loader machine module (see D18). Although we started quite late with the cLIFT procedure during work on our TARGETBINDER proposal, we advanced it meanwhile considerably, especially in terms of automation (see also deliverable report D18).
Please note, that all three methods structure a surface with always the same >20 different solid materials, and afterwards melt the solid matrix material to diffuse the amino acid derivatives for the combinatorial solid phase synthesis according to Merrifield. Currently, we achieve with the cLIFT method a pitch of synthesized peptides of 75µm (see D16). Hopefully, our cLIFT-array-synthesizer will soon be commercialized by PPP.
Conclusions, WP1:
We developed a 2D laser scanning system (see D11) that is used to pattern a glass slide with >20 different amino acid containing materials. Thereby, we can generate very-high-density peptide arrays (see D12, D16, MS11) of up to 1 Million peptides per cm(+2), which is 10-fold more than originally planned (see MS12). Our combined effort led to a fully automated machine prototype (see D18, MS13) including a sophisticated machine controlling program (see D14, D17). Instead of spreading particles onto a flat surface (see D13) – as originally planned – two variant approaches either use a foil with spin coated material, or microstructured glass slides as donor of the to-be-patterned solid material. Moreover, this approach greatly facilitates recycling of expensive materials (see D15).
WP2, main S & T results:
All the deliverables and milestones due in WP2 within the TARGETBINDER project have been achieved, with only minor deviations in the delivery date, spent resources and person months.
iRIfS device for massively parallel labelling-free read out of kinetic data:
One clear highlight of WP2 was the development of a device that is able to readout kinetic data in the array format and without labelling the analyte (see D23; Fig. 9). BIA designed, optimised, assembled, and tested this device that uses the iRIfS method (imaging reflectometric interference sensor; Proll et al., 2007, J. Chrom. A 1161, 2-8; Pröll et al., 2005, Anal. Bioanal. Chem. 382, 1889) to measure growing or decreasing interference of light beams that are reflected from two closely spaced material interfaces. Similar to the Biacore system, their iRIfS setup has one decisive advantage over other methods: The light beam that is used for the measurements doesn’t pass through the buffer solution with its potentially many disturbing influences (light scattering, absorption etc.), but rather is reflected from the bottom of the array (where the resin with synthesized peptides is attached to the surface), and, in addition, from the other side of the peptides, where antibodies from the buffered solution diffuse to the peptides, and, eventually, bind to them. These two reflected light beams interfere with each other, which gives a very sensitive signal that depends over time on the increased (antibodies are added from solution) or decreased (antibodies diffuse away to the washing buffer) optical density of the second interphase. BIA showed that their device is ideally suited to measure over time these binding (on-rate) and dissociation (off-rate) events also in the array format.
For this purpose, BIA used, tested, and optimized different optical components in order to improve the signal-to-noise ratio of the setup. Two different cameras were characterised and their performance was tested against each other. A data acquisition and processing step was established in order to determine kinetic and thermodynamic data of biomolecular interaction. In addition, we performed theoretical consideration of a high density array. In order to validate the device for proper functioning, we performed measurements with a standard antibody / antigen system (anti-PSA / PSA). Indeed, we obtained valid kinetic and thermodynamic constants. When we compared the affinity constants that we measured with our iRIfS device to constants found in the literature they were found to be in very good accordance (Fig. 10). Currently, SME Biametrics started to commercializes this device. It should allow a customer to readout kinetic data for thousands of affinity measurements in parallel (see also D37). Moreover, BIA developed an automated program to kind of “translate” measured signals into binding constants, which can’t be done by hand at such a scale (see D24). BIA programmed tools that automatically acquire, extract, process and evaluate the big amount of data obtained by measurements in this array format. Figure 11 shows the automatic fitting of measured signals (done with our iRIfS device, see D23) with theoretical association and dissociation curves. Next, the program uses these curves to automatically calculate association and dissociation constants, and their resulting affinity constant. This programming tool leads to a tremendous saving of time and work. Moreover, the software was validated against a set of standard data and showed no significant difference to the fits processed by hand.
Suitable glass substrate and flow cell:
The ultimate goal was the massive parallel read out of kinetic data for the biomolecular interactions of up to 10,000 spots per cm(+2). This objective demands a very sensitive and at the same time highly reproducible detection of the iRIfS signals generated from spots that are approx. 100µm in diameter. Therefore, BIA developed a flow cell (see D22) that guarantees a reproducible laminar flow of antibodies and washing buffers over the peptide spots of the array, and they scrutinized different materials between the peptide containing layer and the glass substrate that influence the sensitivity of their iRIfS method (see D21). One goal within this deliverable was to understand the influence of the used transparent substrates, the optical layer coatings, surface chemistries as well as the used measurement wavelength(s) on the iRIfS signals that are generated in response e.g. to antibody molecules that bind to a peptide spot. BIA first performed simulations in order to better understand the theory our method is based upon. In addition, we used a standard system to systematically measure the influence of changing parameters on iRIfS signals in order to verify the simulated results, and, thus, pinpoint especially promising glass substrates. Indeed, BIA could find parameters that granted a constant and very prominent increase of the iRIfS signal over time with an optimized set of parameters and measured on a very small area. The sensitivity of the measured signals indeed suffice to enable the massive parallel read out of kinetic data for the biomolecular interactions of up to 10,000 spots per cm(+2). By these experimental results, and, additionally, by simulations, BIA defined an optimized set of parameters that was used to readout kinetic data in the array format with the Biametrics technology. The optimal parameter set is (see D21):
• N-BK7 glass substrate (from Schott)
• Optical coating 2 (20 nm thick SiO2 layer on top of 45 nm thick Ta2O5 layer)
• AMD as biopolymer (30 nm thick biopolymer on top of SiO2 layer)
• 470 nm (blue light) as evaluation wavelength
Conclusions, WP2:
BIA developed an automated device that allows to readout kinetic date in the array format and up to 10,000 spots per cm(+2). This device is commercialized by SME Biametrics.
WP3, main S & T results:
Most of the deliverables and milestones due in WP3 within the TARGETBINDER project have been achieved, with only minor deviations in the delivery date, spent resources and person months. However, we failed to achieve milestone MS33.
Find peptide binders without any previous knowledge:
One highlight clearly is the proof-of-principle that indeed we can find peptide binders against recombinant proteins without any pre-information about the binder’s nature: We simply screened a CIS display library (10(+13) – 10(+14) different peptides) for Grb2 binders, then, after next generation sequencing, re-screened found peptides in a different format, namely high density peptide arrays, and finally, determined the “binding signature”, i.e. those amino acid positions with the found peptide(s) that are mandatory for antibody binding. This was a major achievement because we thereby achieved the main objective of the TARGETBINDER project (milestone MS31), i.e. we can answer with “yes” to the following question:
“Can we indeed combine CIS display technology (ISO) with particle based synthesis of peptide arrays (PPP) to find peptide binders against target proteins without any pre-information?”
However, when we tried to broaden the experimental data to back our claim that we can find peptide binders for – hopefully any – target protein without any pre-information, we had to realize that other target proteins didn’t yield high affinity peptide binders when using them to screen for linear peptide binders. This finding is probably due to the entropic penalty of linear peptides, i.e. linear peptides can fold into too many different 3D structures, whereby only a few of these 3D structures fit to their binding partner. In other words: although we can use CIS display to screen huge libraries of >10(+13) different linear peptides for most proteins we might only find low affinity binders (see WP4). We therefore failed to achieve milestone MS33.
CIS display screens:
ISO installed a next generation sequencing machine in house (see D31), and, thereby, was able to perform CIS display screens with libraries that comprise more than 10(+13) different peptides up to the level of a list with several ten thousands of potential peptide binders (see D32). Moreover, ISO developed an informatics system for processing next generation sequencing (NGS) data in a high throughput fashion. This includes filters for quality control of the NGS output, sequence trimming to the correct DNA length, translation of the DNA into peptide sequence, frequency analysis of the enriched peptide sequences from the CIS display selections, clustering the peptide sequences into families and depositing these into a database or sub-databases for future reference (see D36).
ISO performed altogether sixteen CIS display selections by panning four pools of libraries (pool 1: short linear; pool 2: long linear; pool 3: constrained, pool 4: β-turn constrained, see D34, Appendix 1) against each target under two conditions, either with or without heparin in the blocking buffer. Next, ISO used their next generation sequencer (see D31) to determine the peptide sequences that were enriched during these pannings. Strikingly, some pannings lead to strong enrichment of some peptides, while other selections lead to only weak enrichment, which is indicative of non-specific binding. From this data, we realized that libraries with linear peptides didn’t give target protein specific enrichment in most cases. Therefore, ISO focused on the peptide libraries with constrained peptides with an intra-peptide disulphide bond. In order to analyse enriched peptides further, ISO performed ELISAs with some of the selected peptides. This was done by cloning the PCR amplified templates into a phagemid vector so as to produce phage-displayed peptides for primary screening (“primary screenings”). Finally, some phagemid displayed peptides were tested in an ELISA for cross reactivity with the other target proteins (“secondary screenings”).
Peptide binders for target proteins EGFR and C-Met:
The presence of multiple tryptophans in the peptides selected on the EGFR target was commonly accompanied by other aromatic amino acid residues and may suggest an enrichment for non-specific “sticky” hydrophobic peptides. Such selection outputs usually occur in the absence of selection of specific target-binding peptides, leading to the conclusion that these selections appear not to have been successful. Yet, sequences of the most enriched peptides were passed to PPP to be printed on arrays and probed with EGFR and an irrelevant target in order to reveal any EGFR-specific candidates.
Peptide binders for target proteins DR4 and CD95:
A number of repeat selections against two TNFR super family members, DR4 and CD95, have identified target-binding peptides with varying degrees of specificity towards the selection antigens and another member of this family, DR5. The behaviour of these outputs can be broadly grouped into four population, (i) those which appeared to bind TNFRs but which showed little/no specificity between the three family members tested (Fig. 12), (ii) peptides which bound DR4 and CD95 with some cross-reactivity to DR5, (iii) one peptide which appeared to bind specifically to the DR4 antigen, failing to bind either DR5 or CD95 and, finally (iv) peptides which appeared to bind the Fc domain fused to each of the selections targets.
Peptide binders for human Fc-antibody fragment:
The target proteins delivered by OXF and by APO were all fused to a human Fc-antibody fragment. ISO therefore analysed selected peptides also for their specificity towards human Fc-fragments. Indeed, especially the disulfide-cyclized β-turn mimetic loop library generated binders to the Fc fragment present on these targets (Fig. 13) and, furthermore, a number of these clones displayed sequences with similarity to know Fc-binding motifs.
Validating peptide binders in the array format:
PPP “translated” the sequences delivered by ISO (see D32, D34, D36) into another format by synthesizing high density peptide arrays with its peptide laser printer (see D35). These were then stained with target proteins delivered by OXF and APO (see D41 to D44). Before, target proteins were chemically conjugated with biotin (done by ISO) or with fluorophores (done by DKFZ).
Binders for target proteins DR4, DR5, CD95, c-Met and EGFR:
When using the list of peptides that PPP and DKFZ got from ISO, we did not find specific peptide binders that were stained also in the array format with commercially extremely interesting target proteins DR4, DR5, CD95, c-Met and EGFR (for all of these proteins peptide binders that interfere with their signalling pathways might be interesting lead structures for the development of therapeutics). When we stained the peptides (“hit peptides”) that were obtained from initial CIS display screens (done by ISO), PPP neither found binders for linear peptides nor for cycled peptides (done by disulphide formation).
There are a number of possible explanations. First, the stringency of binding reaction is probably higher in the array-based assay (addition of 0.1% Tween20 to all buffers; number of washing steps, length of washing) compared to CIS display. Then, binding affinity could simply be too low for detection on peptide arrays.
Binders for target protein Grb2:
For Grb2 we could indeed validate several hundred of the original CIS display hits (see D35). We could identify two different classes of binders showing homologies to known Grb2 binding proteins. One class shows the RXXK consensus motif present in Gab2 and other proteins known to interact with the C-terminal SH3 domain of Grb2. The other class contains the PXXP motif, typical for binders interacting with the N-terminal SH3 domain of Grb2. By competition with class-specific peptides we could demonstrate that the binding domains on Grb2 do not overlap. In other words: Each SH3 domain from Grb2 binds to one kind of peptide signature (Fig. 14), and these “binding signatures” are found in the sequences of known Grb2 binders, e.g. the SOS protein.
Conclusions, WP3:
The observation of conserved peptide motifs, most notably amongst the constrained loop library outputs, suggested the presence of target-specific peptides when doing CIS display screens. This postulation is supported by the similarity seen between the observed motif(s) and that published by Vrielink et al. (FEBS Journal, 2010, 277, 1653-1665) the latter demonstrating DR5-specificity and inhibition of TRAIL binding by this receptor. However, CIS display selection and phage ELISAs cannot predict the affinities of the peptide binders towards the target proteins. Data from CIS display screens clearly show that such peptide motifs were usually only found when using libraries with constrained (looped) peptides.
We could validate these binders in the array format only for Grb2 binders (see D35), where we used linear, non-constrained peptides. Validation experiments with Fc-fragment binding peptides are ongoing.
The most probable explanation for this finding is that cyclic peptides (disulphide bonds) form differently in low concentration in solution (as in CIS display) vs. peptides that are tethered in high concentration to a surface (as in the array format). It might be that oxidized peptides form a meshwork of entangled polymers in the array format that doesn’t bind to an antibody.
WP4, main S & T results:
Most of the deliverables and milestones due in WP4 within the TARGETBINDER project have been achieved, with only minor deviations in the delivery date, spent resources and person months. Although the expression of some target proteins proved to be difficult due to problems with proteolytic activity or similar, OXF and APO were still able to provide the planned number of interesting target proteins that were used by ISO, DKFZ, and PPP to screen for peptide binders. However, except for Grb2 and maybe the human Fc-antibody fragment, we failed to find high-affinity peptide binders (discussed in WP3). Therefore, we didn’t achieve deliverable D46.
Biological active, purified, monomeric target proteins:
One highlight within WP4 was certainly the demonstration of biological activity for the expressed target proteins TRAIL-R1-Fc, EGFR, CD95, and Grb2, which is a pre-requisite to eventually demonstrate an influence of found peptide binders on this activity. Another highlight was the expression of these target proteins in high quality (monomeric forms) and in high quantity (>2mg) that is needed to avoid too high background levels when screening for peptide binders. Yet another highlight was the publication of work done in WP4 in meanwhile several papers. APO expressed and purified TRAIL-R1-Fc protein in monomeric form and in high yield, and then performed an assay to validate its biological activity also in the biotinylated version of the protein (assay was interference with apoptosis; Fig. 15). In addition, APO purified a second interesting target protein, i.e. CD95-Fc protein in monomeric form and in high yield. Again, in order to validate its biological activity APO performed a caspase assay (Fig. 16). However, APO had some trouble to produce human TRAIL-Receptors 3 and 4 as biologically active proteins when they expressed the ligand binding-domain only. APO managed to produce the ligand binding-domain only from macaca fascicularis TRAIL-Receptors 3 and 4 which validated our CHO-based expression system. Afterwards, APO shifted to the production of human TRAIL-Receptor 2. Again, we obtained a homodimeric protein preparation of the human TRAIL-Receptor 2-Fc protein that is biological active (Fig. 17).
Initially, OXF had some trouble in expressing c-Met, but we could indeed express and purify sufficient amounts of a truncated version of EGFR (Δ998-1186) in HEK293T cells. We could also show that the EGFR kinase domain can phosphorylate its downstream target protein Gab2. Meanwhile, OXF also managed to express c-Met in high purity and in high yield (Fig. 18). In addition, OXF supplied the consortium with Grb2 and with individual SH3 domains from Grb2. These proteins were then used by DKFZ to stain peptide arrays, whereby we detected and validated several hundreds of peptide binders. Especially interesting, we identified individual “binding signatures” for each of the SH3 domains from Grb2 (Fig. 14) when we systematically varied some of the peptide binders from initial screens (see D35). OXF could demonstrate high-affinity binding for two peptides that were screened by ISO and validated by PPP for binding to Grb2 SH3 domains (see WP3). Notably, binding affinities for these peptides were higher than for previously identified binding partners. Thereby, we achieved our main aim, i.e. validating our screening procedure (MS43).
Work done with the other target proteins was disappointing though. We failed to identify peptide binders that bound to target proteins in both screens, i.e. in CIS display and in the peptide array format (analysis of Fc-fragment binders in the peptide array format is still ongoing). Therefore, APO investigated four peptides that were identified in ISO’s CIS display screens only. They all proved to be promiscuous, non-specific binders that also bind to unrelated protein, thereby, explaining the negative results in the binding assays employing high density peptide arrays. In competition assays it was shown that the peptides lack an effect in reducing the specific binding of the ligand (a TRAIL mimic) to the receptor target (TRAIL-Receptor 1). Unfortunately, we did not see an effect of these peptides in functional apoptosis assays on Colo205 cells.
Conclusions, WP4:
The consortium expressed and purified several interesting target proteins. In addition, OXF and APO established several functional assays in order to verify that these monomeric proteins still have biological activity.
Problems which have occurred and how they were solved or envisaged solutions:
WP1: We realized that the originally planned laser-induced-melting-method was difficult to automate due to the difficulty to spread our amino acid particles uniformly over the surface. We constructed instead a fully automated cLIFT-method-based array synthesizer (see D18) with programming help from TUV and MS (see D14 and D17). This machine will soon be commercialized by consortium partner PEPperPRINT. Moreover, the donor foils the cLIFT array synthesizer uses guarantee extremely frugal material consumption and easy recycling of expensive materials. In parallel, we came across the idea to use microstructured glass slides for the synthesis of peptide arrays which gave in a not-yet-automated-machine 1 Million peptides per cm(+2) – a 10-fold higher density then originally envisioned.
WP2: The only minor problems we encountered within WP2 centred on combining array synthesis methods and labelling free detection of binding signals: (1) PPP used the Goetheglas from BIA to grow peptides in the array format, (2) while KIT supplied BIA with a stamping method to first sever the peptides from their solid support, and then transferring them in the array format directly onto the specialized Goetheglas from BIA. Both methods worked, but we haven’t yet shown it in the density we envisioned originally. We will do that (hopefully tackling a relevant biological question) in order to generate a publication.
WP3: We managed to find exactly defined binding signatures for the two SH3 domains from Grb2, and, thereby, proved our claim that indeed we could use CIS display and peptide arrays to find binders for in interesting target protein without any previous knowledge. But we failed to do that with other interesting target proteins. We think this failure is on one hand due to the entropic penalty that is associated with linear peptide binders (too many 3D structures), and on the other hand due to difficulties to form disulphide based circular peptides both in solution (for CIS display) and also tethered to a surface in the array format. KIT currently develops very high density peptide arrays that should allow us much better control of peptide cyclisation. As soon as these are available, we will screen them with the target proteins that were generated by the consortium.
WP4: OXF and APO encountered one minor difficulty in the work related to WP4. Due to this nuisance deliverables D42 and D43 were delayed by 4-6 months. The corrective action taken was to try alternative methods to express the proteins. However, it proved to be impossible to express some of these proteins. Therefore, the consortium supplied alternative – and interesting – target proteins that were screened for peptide binders.
Potential Impact:
Future impact of the TARGETBINDER project can be categorized in two fields:
(1) Technological developments in peptide array synthesis, in CIS display screens, and in highly parallel detection of binding affinities will be commercialized within participating SMEs PEPperPRINT, Isogenica, and Biametrics.
(2) Research results from the TARGETBINDER project might have an impact in two different fields. (i) defining binding signatures for proteins of the Grb2 type – that glue together other proteins in order to form bigger complexes – might help to understand information processing within eukaryotic cells. (ii) Finding peptide binders for any target protein – which was our main goal – might have a big impact on screening for novel lead structures.
Procedure to synthesize high-density peptide arrays:
The consortium managed to develop two novel variants of our solid-material-based synthesis method up to the level of an automated cLIFT-array-synthesizer (Fig. 7) that will be commercialized by SME PEPperPRINT. Moreover, these arrays are of much better quality (Fig. 8) when compared to laser printed peptide arrays.
Future work will focus on (1) testing the “cLIFT-array-synthesizer” in order to commercialize it within PPP. KIT and PPP currently negotiate the terms of this commercialization effort including support by the KIT’ venture fund. (2) KIT and PPP will strive to further increase the number of peptides per area that we can synthesize with the cLIFT method by optimizing the laser system and the donor layer of transferred material. We think, that the cLIFT method should allow for a pitch of at least 30 µm, and, thereby, we should achieve approx. 1 million peptide spots per glass slide. These experiments are on-going. (3) We will publish the cLIFT method (see Fig. 3) and also the variant method that is based on microstructured glass slides (Fig. 2). Especially the latter one should allow us to easily synthesize peptide arrays with record densities of peptide spots (Fig. 5). (4) We will try to win additional funding for the variant methods shown in Figs. 2 and 3).
Procedure to determine binding affinities in the array format:
BIA demonstrated a method to perform kinetic evaluations in a density higher than 5,000 spots per cm(+2), which will be commercialized by SME Biametrics in the near future. Due to the excellent optical lateral resolution of the iRIfS device (<10um), and due to its sensitivity such measurements should be feasible in the near future: (1) KIT has developed their cLIFT-based array synthesizer (see D18); (2) KIT meanwhile managed to transfer 10,000 peptide spots per cm(+2) in the array format to an adjacent receptor slide (with BIA’s specialized “Goetheglas” surface; published with lower density in the PhD thesis of Dr. Jakob Striffler, 2014); (3) PPP used BIA’s “Goetheglas” to synthesize arrays of 800 peptides per cm(+2). Future work will comprise the measurement of high-density arrays (with up to 10,000 spots per cm(+2)) that will be produced and provided by PPP and KIT using the Biametrics setup and software to determine the kinetic and thermodynamic constants. We think that such an endeavour should yield a high quality publication.
Information processing within cells, Grb2 binders:
We realized during the TARGETBINDER project that evolution obviously designed several distinct protein domains (e.g. SH3, SH2) to bind to small linear peptide stretches, whereby binding is switched on or off by post-translational modification (e.g. acetylation or methylation of lysine or arginine, which might be the “off-switch” for SH3 domains). In other words: It is a justified assumption that cells do a great deal of their information processing by regrouping protein complexes if a signal induces a new kinase or acetylase or similar. Experiments of the type shown in deliverable D35 (and in Fig. 14) might enable us to use high density peptide arrays (especially those with built in post translational modifications; see D17) to systematically identify a network of these peptide switches, which in the long run should allow us to forecast which protein complexes exist in a cell when knowing what proteins and which enzymes that do the post translational modifications (and, thereby, switching the peptide binding) are present in the first place (from widely available mRNA expression data in the databases).
Therefore, one route of our future work will use existing lists of proteins that were found to from complexes in order to translate the protein’s sequences into arrays of overlapping peptides and then stain them systematically with the different members of these protein complexes. Afterwards, we will systematically vary found peptide binders to determine the binding signature (Fig. 14). Furthermore, our novel cLIFT based method (see D17) is especially well suited to also incorporate all kinds of post translational modifications into the peptides on the array, whereby we should be able to identify the type of on and off switch simply by staining arrays with many different variants of found peptide binders. Next, we will use the binding signature to find out what other proteins might bind via a short peptide stretch to that particular protein. Experiments of that type are ongoing with cooperation partners inside and outside the TARGETBINDER consortium (e.g. Peter Nollau from Univ. Hamburg who has expressed 69 human SH2 domains). We will apply for funding for a DFG Forschergruppe to work on projects of this kind.
Binders to target proteins:
Although up to now our search for peptide binders for other interesting target proteins was disappointing, the argument put forward when applying for the TARGETBINDER project is still valid: Pharmaceutical industry invested heavily to confirm that cancer patients profit from targeting a limited set of interesting target proteins (which is usually done with antibodies). A linear peptide can fold into (too) many 3D structures, resulting in low affinity binding if only a small fraction of these 3D structures fits to the binding partner. One way out of this dilemma is the synthesis of circular peptides that should fold into much fewer 3D structures. Indeed, nature gives us many examples of circular peptides that bind in high affinity to proteins (e.g. alpha-amanitin). The screens from ISO – using circular peptides – clearly showed an enrichment of peptide binders – at least for some of the target proteins – but PPP and DKFZ couldn’t confirm their binding also in the array format. This might have been due to the difficulties in generating circular peptides based on disulphide bonds also in the array format. This reaction is difficult to control in the peptide array format because the density of synthesized peptides is much higher when compared to soluble peptides in CIS display, and it varies from spot to spot, with too much diluted peptides giving low binding signals, and too dense peptides generating a meshwork of polymerized peptides. PPP and KIT currently validate a whole panel of on-array-cyclization methods and using antibodies that recognize conformational epitopes. As soon it is available, PPP and KIT will use the best on-array-cyclisation method to synthesize the panel of cyclic peptides that was identified by ISO and stain them again with our target proteins.
Fc-binders:
ISO identified a panel of peptides that bound to human Fc-fragments (Fig. 13). Although not planned originally, PPP, DKFZ, and KIT will synthesize corresponding peptide arrays to eventually find Fc-binding peptides. These could be used e.g. for antibody purification, or for labelling antibody sera by simply mixing sera with a fluorescently labelled Fc-binding peptide.
List of Websites:
List of participants and contact details:
PEPDIODE web site: www.TARGETBINDER.eu
Coordinator: Dr. Frank Breitling (Karlsruhe Institute of Technology (KIT), Institute for Microstructure Technology (IMT), Germany; www.KIT.edu; email Frank.Breitling@KIT.edu);
Administration at KIT: Mr. Berndt Kronimus (Berndt.Kronimus@KIT.edu); Ms. Daniela Ott (Daniela.Ott@KIT.edu);
CIS display screens:
Dr. Chris Ullman, and Dr. Neil Cooley (Isogenica Ltd. (ISO), United Kingdom; www.Isogenica.com; email chris.ullman@isogenica.com and Neil.Cooley@isogenica.com;
Peptide synthesis:
Dr. Volker Stadler (SME PEPperPRINT (PPP), Germany; www.PEPperPRINT.com; email Volker.Stadler@PEPperPRINT.com); and
Dr. Ralf Bischoff (German Cancer Research Centre (DKFZ), Germany; www.dkfz.de; email R.Bischoff@dkfz.de); and
Dr. Alois Sonnleitner (centre for Advanced Bioanalysis GmbH (CBL), Austria; www.cbl.at; email alois.sonnleitner@cbl.at);
Labelling free detection in the array format:
Dr. Günther Proll (Biametrics GmbH (BIA), Germany www.biametrics.com; email guenther.proll@biametrics.com);
Target proteins:
Dr. Stephan Feller and Dr. Marc Lewitzky (Oxford University (OXF), United Kingdom; www.ox.ac.uk; email stephan.feller@imm.ox.ac.uk and marc.lewitzky@oncology.ox.ac.uk);and Dr. Jürgen Gamer (SME Apogenix GmbH (APO), Germany; www.apogenix.com; email juergen.gamer@apogenix.com);
Prof. Feller meanwhile moved to now University of Halle, Germany; email stephan.feller@uk-halle.de and marc.lewitzky@uk-halle.de
Programming / peptide synthesis reactor:
Prof. Jordan Kolev (Technical University of Varna, (TUV), Bulgaria; http://tu-varna.bg/tu-varnatr/index.php?lang=en(si apre in una nuova finestra); email jkolev@ieee.bg); and
Dr. Nikolay Nikolov (SME Microsystems Ltd. (MS), Bulgaria www.mikrosistemi.com; email n.nikolov@mikrosistemi.com);
Most of the objectives and milestones within the project have been achieved, with the only exceptions being deliverable D46 and milestone MS33. Especially noteworthy, we achieved our main objective: we found peptide binders by CIS display for the protein Grb2 without employing any previous knowledge, and we validated these binders in the array format and also by measuring the binding affinities. Indeed, we could verify that every SH3 domain from Grb2 binds to one clearly defined “signature” that could be used to screen the databases for candidate human protein binders. Experiments of this kind might one day elucidate the information processing that takes place in a eukaryotic cell. Other highlights were technical achievements: BIA designed, manufactured, and tested a device that is capable of measuring the affinities for thousands of peptide binders in parallel. KIT and PPP developed a new semi-automated machine that synthesizes peptide arrays in unprecedented density and quality. Both technical achievements are or will be commercialized within our spin-off companies PEPperPRINT (www.PEPperPRINT.com) and Biametrics (www.Biametrics.com). There were difficulties though. The expression of some target proteins proved to be difficult due to problems with proteolytic activity or similar, but the TARGETBINDER consortium provided other interesting target proteins instead that were used to screen for peptide binders. Another problem, however, seems to be of more fundamental nature. Although we employed CIS display for initial screens with its huge peptide libraries (> 10(+13) different peptides), we did not find high affinity binders for the majority of target proteins, only specialized protein domains, e.g. SH2, SH3, and some other protein domains seem to have been designed by evolution to bind to linear short peptides. One reason for this failure might lie in the entropic penalty that linear peptides come along with their many different 3D structures that they could fold into: Only a few of these will be complementary to a binding partners, which directly translates into low affinity binding. Therefore, the consortium tried to find binders against circular peptides. Indeed, ISO’s CIS display screens probably worked with S=S-bond circularized peptides, but we might have failed to synthesize the very same circular peptides also in the array format. Due to much higher concentration of the peptides that are tethered to the array’s surface these might have preferentially formed intermolecular S=S-bonds.
Project Context and Objectives:
Objectives for the whole project:
The ground breaking development of very high-density oligonucleotide arrays (Fodor et al., 1991, Science 251, 767) boosted the field of genomics both in scientific and in economic terms. The main drivers of this development were the US-based companies Affymetrix and Illumina with a combined annual turnover of nearly 1 billion $ in 2009. Recently, the members of this consortium developed several tools and technologies to easily generate (Beyer et al., 2007, Science 318; Stadler et al., 2008, Angew. Chem. Int. Ed. 47, 7132; Odegrip et al., 2004, PNAS 101, 2806) and analyse (Proll et al., 2007, J. Chrom. A 1161, 2-8; Pröll et al., 2005, Anal. Bioanal. Chem. 382, 1889) very high-complexity peptide libraries. Within the TARGETBINDER consortium, we wanted to combine these technologies, and we expected that our procedures might boost the field of proteomics in a similar way as the lithographic technologies did with the field of genomics. Our goal was to develop a new peptide discovery platform based in the EU that should enable high-throughput identification of peptides that bind to interesting target proteins. Furthermore, the platform should allow us to characterize and to optimize these potential therapeutics at a significant time and cost advantage when compared to other biologics. We expected that such a platform might have a huge commercial and scientific impact.
Objectives broken down to work packages:
We wanted to further develop our tools and technologies in order to achieve the following final goals:
(I.) Within WP1 participants KIT, CBL, MS, and TUV wanted to develop a variant of our solid-material-based-synthesis-method (Breitling et al., 2009, Molecular BioSystems 5, 224; Stadler et al., 2008, Angew. Chem. Int. Ed. 47, 7132) that should allow us to synthesize up to 1 million different peptides on a glass slide for chemical costs of ~50€. In order to achieve this goal, we had to find a method that reliably covers >99% of the central area of every synthesis spot with a matrix material that comprises activated amino acid derivatives for peptide synthesis, which should result into improved quality over a laser printout that usually covers only >95% of the central area. When the glass slide with its patterned amino acid derivatives is heated up in an oven, this melting step frees hitherto immobilized amino acids to diffuse to the surface of the synthesis slide where they couple to free amino groups. This combinatorial synthesis procedure is identical to the 40-years-old Merrifield synthesis, with the only difference being an intermittent “freezing” of the coupling reaction at room temperature. Moreover, we wanted to advance this technology up to a level of an automated machine, which should allow us to commercialize this technology within PPP.
(II.) Within WP2 mainly participant BIA wanted to advance BIA’s labelling-free detection method (Proll et al., 2007, J. Chrom. A 1161, 2-8; Pröll et al., 2005, Anal. Bioanal. Chem. 382, 1889) in order to use it for parallel readout of binding affinities for up to 10,000 peptide spots per cm(+2). First experiments were meant to find out how sensitive the method could be, and to identify suitable glass substrates and reliable fluidic cells. Furthermore, BIA wanted to design, construct, and test a device that could be used to perform these affinity measurements in the array format, and, in addition, to write an automated program that should “translate” measured binding curves into affinity constants. We planned to commercialize this device within SMA Biametrics (www.biametrics.com).
(III.) Within WP3 ISO, DKFZ, and PPP wanted to combine acellular transcriptional peptide/protein display technique (“CIS display”; Odegrip et al., 2004, PNAS 101, 2806), high-density peptide arrays (manufactured by PPP, DKFZ, KIT; Angew. Chem. Int. Ed. 47, 7132), and parallel labelling-free detection of binders (iRIfS method developed by BIA; Proll et al., 2007, J. Chrom. A 1161, 2-8; Pröll et al., 2005, Anal. Bioanal. Chem. 382, 1889) in order to develop a standardized, fast, and reliable high-throughput procedure to find peptide binders against, hopefully, any pharmaceutically interesting target protein. The rationale behind this approach was to first use the huge number of 10(+13) to 10(+14) different peptides that are displayed on a ribosome when using the CIS display technique, then use next generation sequencing to generate a comprehensive list of all peptides that possibly bound to a target protein (expressed and purified within WP4), and, finally, validate candidate binders by synthesizing several thousand of them in the array format with PPP’s particle based synthesis method, and staining these arrays with the same target proteins that were used to select displayed peptides.
(IV.) Within WP3 APO, OXF, DKFZ, and PPP wanted to express, purify, and characterize interesting target proteins, and, in addition, to validate our novel high-throughput procedure by characterizing peptide binders. If successful, such an approach might supply big pharmaceutical companies with peptide-based lead structures that bind to target proteins.
Project Results:
Overview: Within the TARGETBINDER project, we advanced several technologies:
WP1: KIT together with CBL, TUV, and SME MS advanced several variants of our solid-material-based synthesis method up to the level of an automated machine that allows us now to synthesize very high-density peptide arrays at very moderate cost and in high quality. We started the process to commercialize the automated cLIFT array synthesizer within SME PEPperPRINT.
WP2: SME BIA designed, manufactured, and tested a device that is capable of measuring the affinities for thousands of peptide binders in parallel, and they wrote automated programs to “translate” their measurements, e.g. into binding affinities. SME Biametrics started to commercialize this device.
WP3: SME ISO implemented a next generation sequencer, and they streamlined their CIS display technology to better handle the massive amount of data that is generated, e.g. by writing automated programs to compare found peptide sequences with each other and with databases. These additions to their CIS display technology certainly help SME Isogenica in their commercialization efforts.
WP4: OXF and APO expressed, purified, and characterized several proteins that were previously pinpointed by them and by others to be excellent targets, which means that a binding lead structure, e.g. a peptide might lead to a therapeutic molecule.
Combining technologies to find peptide binders without any previous knowledge (WP3 and WP4):
The partners combined their technologies, and, indeed, especially noteworthy, we found several Grb2 binding peptides that proved to bind to Grb2 (supplied by OXF) in higher affinity when compared to previously known Grb2 binders. We did that by first screening >10(+13) CIS displayed peptides, then using next generation sequencing to obtain a comrehensive list of candidate binders (done by ISO), next we validated these binders by synthesizing them in peptide array format (done by PPP) and staining them with target proteins (done by DKFZ). Next, we systematically varied the peptide’s sequence to find out the “binding signature”, and, finally, we determined affinities using three different methods (done by BIA, DKFZ, and OXF). Thereby, we achieved our main objective, namely finding peptide binders without using any previously obtained knowledge (milestones MS31 and MS43). We think that some specialized protein domains, e.g. SH2- and SH3-domains have been designed by evolution to bind to linear short peptides, thereby doing information processing. Peptide arrays might be especially well suited to find out which parts of the proteins that are known to form complexes inside a eukaryotic cell code for the switchable peptide binders, e.g. in SH2 domains binding is switched by adding or removing a phosphoryl group, thereby leading to a re-grouping of complexes when a kinase is activated through an external signal.
However, we failed to find high affinity binders for most other target proteins. One reason for this failure might lie in the entropic penalty that linear peptides come along with their many different 3D structures that they could fold into: Only a few of these will be complementary to a binding partner, which directly translates into low affinity binding. Therefore, the consortium tried to find binders against circular peptides, but we might have failed to synthesize the very same circular peptides also in the array format. We will re-address the question of finding peptide binders against any target protein, this time, however, employing circular peptides right from the beginning.
WP1, main S & T results:
All the deliverables and milestones due in WP1 within the TARGETBINDER project have been achieved, with only minor deviations in the delivery date, spent resources and person months.
Laser-induced-melting method:
One highlight clearly is the publication of our laser-induced melting method in a high ranking journal (see D12 appendix and MS11; Maerkle et al., 2014, Advanced Materials 26, 3730-3734). In addition to the data shown in this publication, we added more data to a novel patent application (PCT/EP2013/001141) that describes the laser-induced melting method (see Fig. 1) and additional variants of our particle based synthesis method (see Figs. 2, 3).
Array synthesis based on microstructured glass slides:
When scrutinizing these additional variants of our solid-material-based synthesis method, we realized that these variants might be better suited to bring us close to the overall goal stated for WP1, namely synthesize 1 million peptides per glass slide at very moderate cost and in high quality. One of these additional variants exploits KIT’s unique capacity to microstructure glass slides (Fig. 4; by reactive ion etching). Next, 1 million cavities per cm(+2) from a donor slide are all filled with only one sort of amino acid particles (e.g. Fmoc-Ala-OPfP ester embedded in the same matrix material that is used for laser printed amino acid particles; Stadler et al., 2008, Angew Chem Int. Ed. 120:7241–7244). This is done by simply gently wiping them into the cavities with a smooth towel, which nearly always results into 100% filling of the cavities. Next, particles from selected cavities of the donor slide are transported to the cavities of an aligned synthesis glass slide that also features 1 million cavities per cm(+2) (Fig. 5), which is done with short laser pulses, and using a 2D laser scanning system (Fig. 6). This step is repeated with additional particles that comprise other amino acid derivatives, which results into a microstructured synthesis glass slide that is densely patterned with solid particles that comprise the 20 different amino acids, and, eventually, additional ones (post-translational modified amino acid side chains, artificial catalytic centres, D-amino acids, moieties that generate cyclic amino acids …). When these patterning steps are finished, the synthesis glass slide is simply heated up which frees hitherto immobilized amino acids to diffuse and couple to the surface. Repeating these steps 15x generates an array of 1 Million 15meric peptides per cm(+2). Indeed, we could show that we could synthesize 10x more peptides per area (i.e. 1 Million peptides per cm(+2)) than originally planned when employing glass slides with small microcavities (Fig. 5). Moreover, it should be straightforward to miniaturize the cavities even further – there have been numerous publications to reliably fill even nanoscale cavities with different particles. However, a drawback of that particular method is the time consuming alignment step. The expensive machinery that is used to do such an alignment, e.g. in chip technology up to now prohibited us from developing a fully automated machine. There is a simple solution to this alignment problem, however. Currently, we use a donor slide with smaller cavities that is placed on top of a synthesis slide with larger cavities. Then, the only alignment that has to be done links the larger cavities of the microstructured synthesis slide with the 2D laser scanning system (the same that was used for the laser-induced-melting procedure).
Laser-induced transfer of materials (cLIFT method):
Yet another approach was especially easy to automate. We therefore focused our work on building a fully automated cLIFT-array-synthesizer (Fig. 7). This machine uses the very same 2D laser scanning system that was described above, but this time to puncture out small pieces of a material that all comprise a suitable amino acid derivative that was spin coated as a thin layer onto a flexible foil (Fig. 3; combinatorial laser induced forward transfer of materials, cLIFT). By repeating these procedures with additional amino acid derivatives, an array of peptides is generated. Obviously, the cLIFT-array-synthesizer is much more reliable in the patterning steps in terms of amount and morphology of transferred material when compared to laser printed particles, which – quite surprisingly for us – resulted into arrays with superior quality in terms of staining signals per spot (Fig. 8). Moreover, the only exact positioning system that is needed for this machine is integrated within a commercially available and ready-to-go 2D laser scanning arm module that regularly addresses two reference points on the glass slide with the growing peptide array to calibrate its position ±2µm (built in self adjusting and correcting). Therefore, we hardly need any efforts to position the thin layers with our amino acid containing material on top of the support (see Fig. 3 for a schematic explanation) that is used for peptide synthesis – we only have to shield the machine from dust particles to assure that the donor layer comes close enough to the glass surface with its growing peptides when exchanging the slide coated with material #1 for another slide coated with material #2. This is done with an already well advanced slide loader machine module (see D18). Although we started quite late with the cLIFT procedure during work on our TARGETBINDER proposal, we advanced it meanwhile considerably, especially in terms of automation (see also deliverable report D18).
Please note, that all three methods structure a surface with always the same >20 different solid materials, and afterwards melt the solid matrix material to diffuse the amino acid derivatives for the combinatorial solid phase synthesis according to Merrifield. Currently, we achieve with the cLIFT method a pitch of synthesized peptides of 75µm (see D16). Hopefully, our cLIFT-array-synthesizer will soon be commercialized by PPP.
Conclusions, WP1:
We developed a 2D laser scanning system (see D11) that is used to pattern a glass slide with >20 different amino acid containing materials. Thereby, we can generate very-high-density peptide arrays (see D12, D16, MS11) of up to 1 Million peptides per cm(+2), which is 10-fold more than originally planned (see MS12). Our combined effort led to a fully automated machine prototype (see D18, MS13) including a sophisticated machine controlling program (see D14, D17). Instead of spreading particles onto a flat surface (see D13) – as originally planned – two variant approaches either use a foil with spin coated material, or microstructured glass slides as donor of the to-be-patterned solid material. Moreover, this approach greatly facilitates recycling of expensive materials (see D15).
WP2, main S & T results:
All the deliverables and milestones due in WP2 within the TARGETBINDER project have been achieved, with only minor deviations in the delivery date, spent resources and person months.
iRIfS device for massively parallel labelling-free read out of kinetic data:
One clear highlight of WP2 was the development of a device that is able to readout kinetic data in the array format and without labelling the analyte (see D23; Fig. 9). BIA designed, optimised, assembled, and tested this device that uses the iRIfS method (imaging reflectometric interference sensor; Proll et al., 2007, J. Chrom. A 1161, 2-8; Pröll et al., 2005, Anal. Bioanal. Chem. 382, 1889) to measure growing or decreasing interference of light beams that are reflected from two closely spaced material interfaces. Similar to the Biacore system, their iRIfS setup has one decisive advantage over other methods: The light beam that is used for the measurements doesn’t pass through the buffer solution with its potentially many disturbing influences (light scattering, absorption etc.), but rather is reflected from the bottom of the array (where the resin with synthesized peptides is attached to the surface), and, in addition, from the other side of the peptides, where antibodies from the buffered solution diffuse to the peptides, and, eventually, bind to them. These two reflected light beams interfere with each other, which gives a very sensitive signal that depends over time on the increased (antibodies are added from solution) or decreased (antibodies diffuse away to the washing buffer) optical density of the second interphase. BIA showed that their device is ideally suited to measure over time these binding (on-rate) and dissociation (off-rate) events also in the array format.
For this purpose, BIA used, tested, and optimized different optical components in order to improve the signal-to-noise ratio of the setup. Two different cameras were characterised and their performance was tested against each other. A data acquisition and processing step was established in order to determine kinetic and thermodynamic data of biomolecular interaction. In addition, we performed theoretical consideration of a high density array. In order to validate the device for proper functioning, we performed measurements with a standard antibody / antigen system (anti-PSA / PSA). Indeed, we obtained valid kinetic and thermodynamic constants. When we compared the affinity constants that we measured with our iRIfS device to constants found in the literature they were found to be in very good accordance (Fig. 10). Currently, SME Biametrics started to commercializes this device. It should allow a customer to readout kinetic data for thousands of affinity measurements in parallel (see also D37). Moreover, BIA developed an automated program to kind of “translate” measured signals into binding constants, which can’t be done by hand at such a scale (see D24). BIA programmed tools that automatically acquire, extract, process and evaluate the big amount of data obtained by measurements in this array format. Figure 11 shows the automatic fitting of measured signals (done with our iRIfS device, see D23) with theoretical association and dissociation curves. Next, the program uses these curves to automatically calculate association and dissociation constants, and their resulting affinity constant. This programming tool leads to a tremendous saving of time and work. Moreover, the software was validated against a set of standard data and showed no significant difference to the fits processed by hand.
Suitable glass substrate and flow cell:
The ultimate goal was the massive parallel read out of kinetic data for the biomolecular interactions of up to 10,000 spots per cm(+2). This objective demands a very sensitive and at the same time highly reproducible detection of the iRIfS signals generated from spots that are approx. 100µm in diameter. Therefore, BIA developed a flow cell (see D22) that guarantees a reproducible laminar flow of antibodies and washing buffers over the peptide spots of the array, and they scrutinized different materials between the peptide containing layer and the glass substrate that influence the sensitivity of their iRIfS method (see D21). One goal within this deliverable was to understand the influence of the used transparent substrates, the optical layer coatings, surface chemistries as well as the used measurement wavelength(s) on the iRIfS signals that are generated in response e.g. to antibody molecules that bind to a peptide spot. BIA first performed simulations in order to better understand the theory our method is based upon. In addition, we used a standard system to systematically measure the influence of changing parameters on iRIfS signals in order to verify the simulated results, and, thus, pinpoint especially promising glass substrates. Indeed, BIA could find parameters that granted a constant and very prominent increase of the iRIfS signal over time with an optimized set of parameters and measured on a very small area. The sensitivity of the measured signals indeed suffice to enable the massive parallel read out of kinetic data for the biomolecular interactions of up to 10,000 spots per cm(+2). By these experimental results, and, additionally, by simulations, BIA defined an optimized set of parameters that was used to readout kinetic data in the array format with the Biametrics technology. The optimal parameter set is (see D21):
• N-BK7 glass substrate (from Schott)
• Optical coating 2 (20 nm thick SiO2 layer on top of 45 nm thick Ta2O5 layer)
• AMD as biopolymer (30 nm thick biopolymer on top of SiO2 layer)
• 470 nm (blue light) as evaluation wavelength
Conclusions, WP2:
BIA developed an automated device that allows to readout kinetic date in the array format and up to 10,000 spots per cm(+2). This device is commercialized by SME Biametrics.
WP3, main S & T results:
Most of the deliverables and milestones due in WP3 within the TARGETBINDER project have been achieved, with only minor deviations in the delivery date, spent resources and person months. However, we failed to achieve milestone MS33.
Find peptide binders without any previous knowledge:
One highlight clearly is the proof-of-principle that indeed we can find peptide binders against recombinant proteins without any pre-information about the binder’s nature: We simply screened a CIS display library (10(+13) – 10(+14) different peptides) for Grb2 binders, then, after next generation sequencing, re-screened found peptides in a different format, namely high density peptide arrays, and finally, determined the “binding signature”, i.e. those amino acid positions with the found peptide(s) that are mandatory for antibody binding. This was a major achievement because we thereby achieved the main objective of the TARGETBINDER project (milestone MS31), i.e. we can answer with “yes” to the following question:
“Can we indeed combine CIS display technology (ISO) with particle based synthesis of peptide arrays (PPP) to find peptide binders against target proteins without any pre-information?”
However, when we tried to broaden the experimental data to back our claim that we can find peptide binders for – hopefully any – target protein without any pre-information, we had to realize that other target proteins didn’t yield high affinity peptide binders when using them to screen for linear peptide binders. This finding is probably due to the entropic penalty of linear peptides, i.e. linear peptides can fold into too many different 3D structures, whereby only a few of these 3D structures fit to their binding partner. In other words: although we can use CIS display to screen huge libraries of >10(+13) different linear peptides for most proteins we might only find low affinity binders (see WP4). We therefore failed to achieve milestone MS33.
CIS display screens:
ISO installed a next generation sequencing machine in house (see D31), and, thereby, was able to perform CIS display screens with libraries that comprise more than 10(+13) different peptides up to the level of a list with several ten thousands of potential peptide binders (see D32). Moreover, ISO developed an informatics system for processing next generation sequencing (NGS) data in a high throughput fashion. This includes filters for quality control of the NGS output, sequence trimming to the correct DNA length, translation of the DNA into peptide sequence, frequency analysis of the enriched peptide sequences from the CIS display selections, clustering the peptide sequences into families and depositing these into a database or sub-databases for future reference (see D36).
ISO performed altogether sixteen CIS display selections by panning four pools of libraries (pool 1: short linear; pool 2: long linear; pool 3: constrained, pool 4: β-turn constrained, see D34, Appendix 1) against each target under two conditions, either with or without heparin in the blocking buffer. Next, ISO used their next generation sequencer (see D31) to determine the peptide sequences that were enriched during these pannings. Strikingly, some pannings lead to strong enrichment of some peptides, while other selections lead to only weak enrichment, which is indicative of non-specific binding. From this data, we realized that libraries with linear peptides didn’t give target protein specific enrichment in most cases. Therefore, ISO focused on the peptide libraries with constrained peptides with an intra-peptide disulphide bond. In order to analyse enriched peptides further, ISO performed ELISAs with some of the selected peptides. This was done by cloning the PCR amplified templates into a phagemid vector so as to produce phage-displayed peptides for primary screening (“primary screenings”). Finally, some phagemid displayed peptides were tested in an ELISA for cross reactivity with the other target proteins (“secondary screenings”).
Peptide binders for target proteins EGFR and C-Met:
The presence of multiple tryptophans in the peptides selected on the EGFR target was commonly accompanied by other aromatic amino acid residues and may suggest an enrichment for non-specific “sticky” hydrophobic peptides. Such selection outputs usually occur in the absence of selection of specific target-binding peptides, leading to the conclusion that these selections appear not to have been successful. Yet, sequences of the most enriched peptides were passed to PPP to be printed on arrays and probed with EGFR and an irrelevant target in order to reveal any EGFR-specific candidates.
Peptide binders for target proteins DR4 and CD95:
A number of repeat selections against two TNFR super family members, DR4 and CD95, have identified target-binding peptides with varying degrees of specificity towards the selection antigens and another member of this family, DR5. The behaviour of these outputs can be broadly grouped into four population, (i) those which appeared to bind TNFRs but which showed little/no specificity between the three family members tested (Fig. 12), (ii) peptides which bound DR4 and CD95 with some cross-reactivity to DR5, (iii) one peptide which appeared to bind specifically to the DR4 antigen, failing to bind either DR5 or CD95 and, finally (iv) peptides which appeared to bind the Fc domain fused to each of the selections targets.
Peptide binders for human Fc-antibody fragment:
The target proteins delivered by OXF and by APO were all fused to a human Fc-antibody fragment. ISO therefore analysed selected peptides also for their specificity towards human Fc-fragments. Indeed, especially the disulfide-cyclized β-turn mimetic loop library generated binders to the Fc fragment present on these targets (Fig. 13) and, furthermore, a number of these clones displayed sequences with similarity to know Fc-binding motifs.
Validating peptide binders in the array format:
PPP “translated” the sequences delivered by ISO (see D32, D34, D36) into another format by synthesizing high density peptide arrays with its peptide laser printer (see D35). These were then stained with target proteins delivered by OXF and APO (see D41 to D44). Before, target proteins were chemically conjugated with biotin (done by ISO) or with fluorophores (done by DKFZ).
Binders for target proteins DR4, DR5, CD95, c-Met and EGFR:
When using the list of peptides that PPP and DKFZ got from ISO, we did not find specific peptide binders that were stained also in the array format with commercially extremely interesting target proteins DR4, DR5, CD95, c-Met and EGFR (for all of these proteins peptide binders that interfere with their signalling pathways might be interesting lead structures for the development of therapeutics). When we stained the peptides (“hit peptides”) that were obtained from initial CIS display screens (done by ISO), PPP neither found binders for linear peptides nor for cycled peptides (done by disulphide formation).
There are a number of possible explanations. First, the stringency of binding reaction is probably higher in the array-based assay (addition of 0.1% Tween20 to all buffers; number of washing steps, length of washing) compared to CIS display. Then, binding affinity could simply be too low for detection on peptide arrays.
Binders for target protein Grb2:
For Grb2 we could indeed validate several hundred of the original CIS display hits (see D35). We could identify two different classes of binders showing homologies to known Grb2 binding proteins. One class shows the RXXK consensus motif present in Gab2 and other proteins known to interact with the C-terminal SH3 domain of Grb2. The other class contains the PXXP motif, typical for binders interacting with the N-terminal SH3 domain of Grb2. By competition with class-specific peptides we could demonstrate that the binding domains on Grb2 do not overlap. In other words: Each SH3 domain from Grb2 binds to one kind of peptide signature (Fig. 14), and these “binding signatures” are found in the sequences of known Grb2 binders, e.g. the SOS protein.
Conclusions, WP3:
The observation of conserved peptide motifs, most notably amongst the constrained loop library outputs, suggested the presence of target-specific peptides when doing CIS display screens. This postulation is supported by the similarity seen between the observed motif(s) and that published by Vrielink et al. (FEBS Journal, 2010, 277, 1653-1665) the latter demonstrating DR5-specificity and inhibition of TRAIL binding by this receptor. However, CIS display selection and phage ELISAs cannot predict the affinities of the peptide binders towards the target proteins. Data from CIS display screens clearly show that such peptide motifs were usually only found when using libraries with constrained (looped) peptides.
We could validate these binders in the array format only for Grb2 binders (see D35), where we used linear, non-constrained peptides. Validation experiments with Fc-fragment binding peptides are ongoing.
The most probable explanation for this finding is that cyclic peptides (disulphide bonds) form differently in low concentration in solution (as in CIS display) vs. peptides that are tethered in high concentration to a surface (as in the array format). It might be that oxidized peptides form a meshwork of entangled polymers in the array format that doesn’t bind to an antibody.
WP4, main S & T results:
Most of the deliverables and milestones due in WP4 within the TARGETBINDER project have been achieved, with only minor deviations in the delivery date, spent resources and person months. Although the expression of some target proteins proved to be difficult due to problems with proteolytic activity or similar, OXF and APO were still able to provide the planned number of interesting target proteins that were used by ISO, DKFZ, and PPP to screen for peptide binders. However, except for Grb2 and maybe the human Fc-antibody fragment, we failed to find high-affinity peptide binders (discussed in WP3). Therefore, we didn’t achieve deliverable D46.
Biological active, purified, monomeric target proteins:
One highlight within WP4 was certainly the demonstration of biological activity for the expressed target proteins TRAIL-R1-Fc, EGFR, CD95, and Grb2, which is a pre-requisite to eventually demonstrate an influence of found peptide binders on this activity. Another highlight was the expression of these target proteins in high quality (monomeric forms) and in high quantity (>2mg) that is needed to avoid too high background levels when screening for peptide binders. Yet another highlight was the publication of work done in WP4 in meanwhile several papers. APO expressed and purified TRAIL-R1-Fc protein in monomeric form and in high yield, and then performed an assay to validate its biological activity also in the biotinylated version of the protein (assay was interference with apoptosis; Fig. 15). In addition, APO purified a second interesting target protein, i.e. CD95-Fc protein in monomeric form and in high yield. Again, in order to validate its biological activity APO performed a caspase assay (Fig. 16). However, APO had some trouble to produce human TRAIL-Receptors 3 and 4 as biologically active proteins when they expressed the ligand binding-domain only. APO managed to produce the ligand binding-domain only from macaca fascicularis TRAIL-Receptors 3 and 4 which validated our CHO-based expression system. Afterwards, APO shifted to the production of human TRAIL-Receptor 2. Again, we obtained a homodimeric protein preparation of the human TRAIL-Receptor 2-Fc protein that is biological active (Fig. 17).
Initially, OXF had some trouble in expressing c-Met, but we could indeed express and purify sufficient amounts of a truncated version of EGFR (Δ998-1186) in HEK293T cells. We could also show that the EGFR kinase domain can phosphorylate its downstream target protein Gab2. Meanwhile, OXF also managed to express c-Met in high purity and in high yield (Fig. 18). In addition, OXF supplied the consortium with Grb2 and with individual SH3 domains from Grb2. These proteins were then used by DKFZ to stain peptide arrays, whereby we detected and validated several hundreds of peptide binders. Especially interesting, we identified individual “binding signatures” for each of the SH3 domains from Grb2 (Fig. 14) when we systematically varied some of the peptide binders from initial screens (see D35). OXF could demonstrate high-affinity binding for two peptides that were screened by ISO and validated by PPP for binding to Grb2 SH3 domains (see WP3). Notably, binding affinities for these peptides were higher than for previously identified binding partners. Thereby, we achieved our main aim, i.e. validating our screening procedure (MS43).
Work done with the other target proteins was disappointing though. We failed to identify peptide binders that bound to target proteins in both screens, i.e. in CIS display and in the peptide array format (analysis of Fc-fragment binders in the peptide array format is still ongoing). Therefore, APO investigated four peptides that were identified in ISO’s CIS display screens only. They all proved to be promiscuous, non-specific binders that also bind to unrelated protein, thereby, explaining the negative results in the binding assays employing high density peptide arrays. In competition assays it was shown that the peptides lack an effect in reducing the specific binding of the ligand (a TRAIL mimic) to the receptor target (TRAIL-Receptor 1). Unfortunately, we did not see an effect of these peptides in functional apoptosis assays on Colo205 cells.
Conclusions, WP4:
The consortium expressed and purified several interesting target proteins. In addition, OXF and APO established several functional assays in order to verify that these monomeric proteins still have biological activity.
Problems which have occurred and how they were solved or envisaged solutions:
WP1: We realized that the originally planned laser-induced-melting-method was difficult to automate due to the difficulty to spread our amino acid particles uniformly over the surface. We constructed instead a fully automated cLIFT-method-based array synthesizer (see D18) with programming help from TUV and MS (see D14 and D17). This machine will soon be commercialized by consortium partner PEPperPRINT. Moreover, the donor foils the cLIFT array synthesizer uses guarantee extremely frugal material consumption and easy recycling of expensive materials. In parallel, we came across the idea to use microstructured glass slides for the synthesis of peptide arrays which gave in a not-yet-automated-machine 1 Million peptides per cm(+2) – a 10-fold higher density then originally envisioned.
WP2: The only minor problems we encountered within WP2 centred on combining array synthesis methods and labelling free detection of binding signals: (1) PPP used the Goetheglas from BIA to grow peptides in the array format, (2) while KIT supplied BIA with a stamping method to first sever the peptides from their solid support, and then transferring them in the array format directly onto the specialized Goetheglas from BIA. Both methods worked, but we haven’t yet shown it in the density we envisioned originally. We will do that (hopefully tackling a relevant biological question) in order to generate a publication.
WP3: We managed to find exactly defined binding signatures for the two SH3 domains from Grb2, and, thereby, proved our claim that indeed we could use CIS display and peptide arrays to find binders for in interesting target protein without any previous knowledge. But we failed to do that with other interesting target proteins. We think this failure is on one hand due to the entropic penalty that is associated with linear peptide binders (too many 3D structures), and on the other hand due to difficulties to form disulphide based circular peptides both in solution (for CIS display) and also tethered to a surface in the array format. KIT currently develops very high density peptide arrays that should allow us much better control of peptide cyclisation. As soon as these are available, we will screen them with the target proteins that were generated by the consortium.
WP4: OXF and APO encountered one minor difficulty in the work related to WP4. Due to this nuisance deliverables D42 and D43 were delayed by 4-6 months. The corrective action taken was to try alternative methods to express the proteins. However, it proved to be impossible to express some of these proteins. Therefore, the consortium supplied alternative – and interesting – target proteins that were screened for peptide binders.
Potential Impact:
Future impact of the TARGETBINDER project can be categorized in two fields:
(1) Technological developments in peptide array synthesis, in CIS display screens, and in highly parallel detection of binding affinities will be commercialized within participating SMEs PEPperPRINT, Isogenica, and Biametrics.
(2) Research results from the TARGETBINDER project might have an impact in two different fields. (i) defining binding signatures for proteins of the Grb2 type – that glue together other proteins in order to form bigger complexes – might help to understand information processing within eukaryotic cells. (ii) Finding peptide binders for any target protein – which was our main goal – might have a big impact on screening for novel lead structures.
Procedure to synthesize high-density peptide arrays:
The consortium managed to develop two novel variants of our solid-material-based synthesis method up to the level of an automated cLIFT-array-synthesizer (Fig. 7) that will be commercialized by SME PEPperPRINT. Moreover, these arrays are of much better quality (Fig. 8) when compared to laser printed peptide arrays.
Future work will focus on (1) testing the “cLIFT-array-synthesizer” in order to commercialize it within PPP. KIT and PPP currently negotiate the terms of this commercialization effort including support by the KIT’ venture fund. (2) KIT and PPP will strive to further increase the number of peptides per area that we can synthesize with the cLIFT method by optimizing the laser system and the donor layer of transferred material. We think, that the cLIFT method should allow for a pitch of at least 30 µm, and, thereby, we should achieve approx. 1 million peptide spots per glass slide. These experiments are on-going. (3) We will publish the cLIFT method (see Fig. 3) and also the variant method that is based on microstructured glass slides (Fig. 2). Especially the latter one should allow us to easily synthesize peptide arrays with record densities of peptide spots (Fig. 5). (4) We will try to win additional funding for the variant methods shown in Figs. 2 and 3).
Procedure to determine binding affinities in the array format:
BIA demonstrated a method to perform kinetic evaluations in a density higher than 5,000 spots per cm(+2), which will be commercialized by SME Biametrics in the near future. Due to the excellent optical lateral resolution of the iRIfS device (<10um), and due to its sensitivity such measurements should be feasible in the near future: (1) KIT has developed their cLIFT-based array synthesizer (see D18); (2) KIT meanwhile managed to transfer 10,000 peptide spots per cm(+2) in the array format to an adjacent receptor slide (with BIA’s specialized “Goetheglas” surface; published with lower density in the PhD thesis of Dr. Jakob Striffler, 2014); (3) PPP used BIA’s “Goetheglas” to synthesize arrays of 800 peptides per cm(+2). Future work will comprise the measurement of high-density arrays (with up to 10,000 spots per cm(+2)) that will be produced and provided by PPP and KIT using the Biametrics setup and software to determine the kinetic and thermodynamic constants. We think that such an endeavour should yield a high quality publication.
Information processing within cells, Grb2 binders:
We realized during the TARGETBINDER project that evolution obviously designed several distinct protein domains (e.g. SH3, SH2) to bind to small linear peptide stretches, whereby binding is switched on or off by post-translational modification (e.g. acetylation or methylation of lysine or arginine, which might be the “off-switch” for SH3 domains). In other words: It is a justified assumption that cells do a great deal of their information processing by regrouping protein complexes if a signal induces a new kinase or acetylase or similar. Experiments of the type shown in deliverable D35 (and in Fig. 14) might enable us to use high density peptide arrays (especially those with built in post translational modifications; see D17) to systematically identify a network of these peptide switches, which in the long run should allow us to forecast which protein complexes exist in a cell when knowing what proteins and which enzymes that do the post translational modifications (and, thereby, switching the peptide binding) are present in the first place (from widely available mRNA expression data in the databases).
Therefore, one route of our future work will use existing lists of proteins that were found to from complexes in order to translate the protein’s sequences into arrays of overlapping peptides and then stain them systematically with the different members of these protein complexes. Afterwards, we will systematically vary found peptide binders to determine the binding signature (Fig. 14). Furthermore, our novel cLIFT based method (see D17) is especially well suited to also incorporate all kinds of post translational modifications into the peptides on the array, whereby we should be able to identify the type of on and off switch simply by staining arrays with many different variants of found peptide binders. Next, we will use the binding signature to find out what other proteins might bind via a short peptide stretch to that particular protein. Experiments of that type are ongoing with cooperation partners inside and outside the TARGETBINDER consortium (e.g. Peter Nollau from Univ. Hamburg who has expressed 69 human SH2 domains). We will apply for funding for a DFG Forschergruppe to work on projects of this kind.
Binders to target proteins:
Although up to now our search for peptide binders for other interesting target proteins was disappointing, the argument put forward when applying for the TARGETBINDER project is still valid: Pharmaceutical industry invested heavily to confirm that cancer patients profit from targeting a limited set of interesting target proteins (which is usually done with antibodies). A linear peptide can fold into (too) many 3D structures, resulting in low affinity binding if only a small fraction of these 3D structures fits to the binding partner. One way out of this dilemma is the synthesis of circular peptides that should fold into much fewer 3D structures. Indeed, nature gives us many examples of circular peptides that bind in high affinity to proteins (e.g. alpha-amanitin). The screens from ISO – using circular peptides – clearly showed an enrichment of peptide binders – at least for some of the target proteins – but PPP and DKFZ couldn’t confirm their binding also in the array format. This might have been due to the difficulties in generating circular peptides based on disulphide bonds also in the array format. This reaction is difficult to control in the peptide array format because the density of synthesized peptides is much higher when compared to soluble peptides in CIS display, and it varies from spot to spot, with too much diluted peptides giving low binding signals, and too dense peptides generating a meshwork of polymerized peptides. PPP and KIT currently validate a whole panel of on-array-cyclization methods and using antibodies that recognize conformational epitopes. As soon it is available, PPP and KIT will use the best on-array-cyclisation method to synthesize the panel of cyclic peptides that was identified by ISO and stain them again with our target proteins.
Fc-binders:
ISO identified a panel of peptides that bound to human Fc-fragments (Fig. 13). Although not planned originally, PPP, DKFZ, and KIT will synthesize corresponding peptide arrays to eventually find Fc-binding peptides. These could be used e.g. for antibody purification, or for labelling antibody sera by simply mixing sera with a fluorescently labelled Fc-binding peptide.
List of Websites:
List of participants and contact details:
PEPDIODE web site: www.TARGETBINDER.eu
Coordinator: Dr. Frank Breitling (Karlsruhe Institute of Technology (KIT), Institute for Microstructure Technology (IMT), Germany; www.KIT.edu; email Frank.Breitling@KIT.edu);
Administration at KIT: Mr. Berndt Kronimus (Berndt.Kronimus@KIT.edu); Ms. Daniela Ott (Daniela.Ott@KIT.edu);
CIS display screens:
Dr. Chris Ullman, and Dr. Neil Cooley (Isogenica Ltd. (ISO), United Kingdom; www.Isogenica.com; email chris.ullman@isogenica.com and Neil.Cooley@isogenica.com;
Peptide synthesis:
Dr. Volker Stadler (SME PEPperPRINT (PPP), Germany; www.PEPperPRINT.com; email Volker.Stadler@PEPperPRINT.com); and
Dr. Ralf Bischoff (German Cancer Research Centre (DKFZ), Germany; www.dkfz.de; email R.Bischoff@dkfz.de); and
Dr. Alois Sonnleitner (centre for Advanced Bioanalysis GmbH (CBL), Austria; www.cbl.at; email alois.sonnleitner@cbl.at);
Labelling free detection in the array format:
Dr. Günther Proll (Biametrics GmbH (BIA), Germany www.biametrics.com; email guenther.proll@biametrics.com);
Target proteins:
Dr. Stephan Feller and Dr. Marc Lewitzky (Oxford University (OXF), United Kingdom; www.ox.ac.uk; email stephan.feller@imm.ox.ac.uk and marc.lewitzky@oncology.ox.ac.uk);and Dr. Jürgen Gamer (SME Apogenix GmbH (APO), Germany; www.apogenix.com; email juergen.gamer@apogenix.com);
Prof. Feller meanwhile moved to now University of Halle, Germany; email stephan.feller@uk-halle.de and marc.lewitzky@uk-halle.de
Programming / peptide synthesis reactor:
Prof. Jordan Kolev (Technical University of Varna, (TUV), Bulgaria; http://tu-varna.bg/tu-varnatr/index.php?lang=en(si apre in una nuova finestra); email jkolev@ieee.bg); and
Dr. Nikolay Nikolov (SME Microsystems Ltd. (MS), Bulgaria www.mikrosistemi.com; email n.nikolov@mikrosistemi.com);
