WO2001066567A2 - Design, synthesis and use of affinity ligands - Google Patents

Abstract

An affinity ligand is constructed from at least 3 domains: (1) a recognition domain, e.g. a 3-6 amino acid chain specific for a peptide region of a target; this may be discovered and/or optimised by fitting candidate structures or parts thereof to a model of the target in a computer memory; (2) an affinity domain, e.g. a domain capable of binding to a particular type of region of the target such as a glycosylation region; and (3) a flexible linker having a functional group enabling the ligand to be immobilised to a sensor surface to create a specific biosensor.

Description

DESIGN, SYNTHESIS AND USE OF AFFINITY LIGANDS

The present invention relates to affinity ligands, particularly for proteins, and especially for proteins having non-peptide portions such as glycosylated proteins. In different aspects it relates to the ligands themselves; design and synthesis of the ligands; support surfaces having such ligands immobilised thereon; and uses thereof particularly as sensors for detecting the presence of, and/or quantifying, substances for which the ligands have affinity.

The exponential increase in the number of structures available in the Brookhaven Protein Databank (PDB) has led to a growing interest in the direct approach to drug design. Our interest has been to exploit the tools developed and used by pharmaceutical companies to aid the development of synthetic receptors for diagnostic purposes. Using this technology we can characterize the target protein molecule efficiently, estimate the interactions between individual combinatorial building blocks and the proposed binding site, rationally design new peptide sequences and also investigate the interactions between combinatorial lead ligands and the binding site. The present invention provides a novel concept of using a "building blocks" approach for designing affinity ligands by molecular modelling. Ligands were designed by combining different domains each targeting a different part of a binding pocket of the protein or other target species. In the case of designing specific ligands for glycosylated haemoglobin (HbAι_c) , the ligands were constructed using three domains. A specific domain for interacting with residues in the Aχ_c binding pocket comprises a small molecule or a short peptide of three to six units. They were obtained through virtual screening of chemical databases or probing the binding site with amino acids. An affinity domain, such as boronic acid and metal chelating compounds, recognises the glucose on the glycosylated ends of β-chains and a flexible domain allows the ligands to be functionalised so that labels or anchors can be implemented to facilitate the assay. The integration between the specific domain and affinity domain using a suitable linker will improve the affinity and specificity exponentially in comparison with the domains presented separately. Virtual libraries can then be constructed using combination of these domains. The validation of this approach was carried out by chemical synthesis of these libraries and the real-time binding of the ligands was examined using Biacore 3000. A lead compound was identified. It was shown that the lead compound has a superior binding capacity over the individual domain. This approach can be generally applied for drug design and for designing affinity ligands for any protein with known structural information and binding mechanism.

Thus in a first aspect the invention provides a receptor molecule comprising a first portion adapted to bind specifically to a first region of a target molecule, a second portion adapted to bind specifically to a second region of the same target molecule, and a flexible linker portion including a functional group whereby the receptor molecule is immobilised or immobilisable to a sensor surface. One region is generally a recognition domain, which is a specific binding member that binds to a specific structural fingerprint. Another region may be an affinity domain that has affinity for a particular type of region of the target, e.g. a glycosylation region. It may rely on charge interaction. In most cases the target molecule comprises a peptide chain, and at least one of the receptor portions is adapted to bind specifically to a specific peptide region. Different portions may be adapted to bind in different binding pockets of the target.

The linker generally includes a chain of 4 or more atoms to provide flexibility, e.g. comprising glycol or polyglycol residues. The receptor may be bound to a biosensor surface, e.g. a surface plasmon resonance sensor such as a BIAcore sensor .

Some embodiments of the invention will now be described in more detail with reference to the accompanying drawings in which:

Fig. 1 (i) , (ii) and (iii) show BIAcore sensor responses for carboxyphenyl boronic acid (PBA) immobilised to the sensor surface via a spacer: (i) during immobilisation; (ii) and (iii) subsequent responses to HBA₀ and HBAι_c respectively;

Fig. 2 (i) , (ii) and (iii) are a set of sensor responses like those of Fig. 1 but relating to a ligand having a PBA region, a peptide region (DAG) , and a spacer. The following examples relate to ligands (receptors) for glycosylated haemoglobin. This is a convenient testing ground for the invention because of the ready availability of data and materials. It should be understood that the invention is not limited thereto. Since haemoglobin is one of the most studied proteins, a large variety of structural datasets exist for haemoglobin. The objective selection of a dataset is assisted using informatics, by examination of the proposed dihedral angles of the protein chain and by comparing the proposed bond lengths, angles and side chain planarity against a group of structures known to possess high structural precision. This is routinely achieved using the HAT_CHECK and PROCHECK packages. Examination of the dihedral angles is usually performed using Ramachandran analysis. Within protein structures, rotations about the individual bonds of both the sidechain and backbone are generally close to one of the conformations favoured in the individual isolated amino acids. Plotting the dihedral angles φ against ψ of the protein in a good protein structure produces data points in discreet regions allocated to each conformation. A high quality haemoglobin structure will have accurate bond lengths and angles. Using the infomatics approach it is readily possible to compare the extent of measurement faults between sets of structural data.

Used with physical data from the literature the molecular model is an ideal technique for the characterisation of the binding site, allowing the rapid identification and location of hydrophobic regions, electrostatic residues, hydrogen bond donors and acceptors .

More and more often in computer-aided molecular discovery work, a useful model of the receptor site has been generated, either directly, as an experimental or homology-built model of the receptor cavity. The use of de novo tools enables the automatic generation of ligand molecules. The program LEAPFROG is a second generation de novo drug discovery program, LeapFrog combines a number of pioneering ideas from authors including Kuntz, Boehm, Karplus, Goodford and others. LeapFrog performs electrostatic screening, by repeatedly making some structural change and then either keeping or discarding the results depending on the quality of the interaction. This technique can be used to directly propose peptide ligand sequences, but is best used as an aid to combinatorial chemistry. Information can be gained by the use of binding site "mapping" (goodford, mess, leapfrog) . This technique allows the binding site to be "probed" either with small moieties such as hydroxyl groups or large fragments and amino acids (MCSS and LEAPFROG) . This binding site "mapping" can therefore be used as an aid to accelerate design of combinatorial libraries by allowing the trial of rationally selected building blocks which may comprise one or more amino acids .

The invention will now be explained further by means of specific examples which relate to the provision of ligands for glycosylated haemoglobin ("HbAι_c").

Example One : Building and refining a molecular model for glycosylated haemoglobin

a) . Objective Selection of a Structural Dataset to Construct a Model of HbA₀

To produce a model that was representative as possible required the use of the best starting x-ray crystallography co-ordinates available. The haemoglobin datasets available are listed in Table 1. The criteria for selection included deviation from expected bond length and angle, chirality, omega angle and dihedral angles, which can be illustrated using Ramachandran plots and presented as RMS values Table 2.

Table 1 Deoxyhaemoglobin datasets available from the Brookhaven Protein Database Date Filen Resoluti Source ame on (A)

18-7-84 2HHB 1.74 Human Red Blood Cells (HRBC) 18-7-84 4HHB 1.74 HRBC

22-1-98 1A3N 1. HRBC

4-1-96 1GHB HRBC & Recombinant α-Chain.

It has been shown that the entry 1A3N is the best-suited dataset. The refinement of the haemoglobin model in the absence of explicit waters using a multi-step approach with a distance related dielectric function and a dielectric constant of 20 provided a reasonable model for the construction of HbAlc model.

Table 2 Comparison of Structural Quality

Example Two The Generation and Refinement of the Molecular Model of HBAχ_c

The 1A3N structure was imported from the PDB file into the SYBYL environment (Tripos) . All the waters of crystallisation were removed and all hydrogen atoms for the protein were added using the automated procedure in BIOPOLYMER. The capping groups of each peptide chain were then added and were treated as charged (NH3⁺, COO^") . Remaining hydrogen atoms were added manually so that the residues Vallα, Lys99α, Hisl03α, Vallβ, His2β, Lys82β, Argl04β and Hisl43β were all correctly protonated. The whole molecule was charged using the Kollman "ALL ATOM" approach.

A step-wise approach to refinement by energy minimisation was used. All hydrogen atoms were first energy minimised, first with the POWELL algorithm and then with BFGS algorithm.

To reduce the overall time taken to produce a reproducible model of the binding site, to overcome the memory restrictions and to reduce the impact of removing the iron from the centre of the haem groups, only the proposed binding site (considered 2θA from the amino β- termini) was considered when doing the final steps of the energy minimisation. The side-chain, followed by the complete binding site, were refined using energy minimisation using the POWELL algorithm only.

Since we are interested in studying HBAι_c we need to manually glycate the HbA₀ by adding glycopyranose adduct to the β valine termini. A model of the glycated valine adduct was first hand "sketched" before being converted into a 3-D structure using SYBYL "Sketch". The nitrogen of the former terminal amine was changed from type ^λN.3' to ^ΛN.am' and the model was charged using the Gasteiger method. The conformational space of the rotating bonds was explored using random impulse search techniques.

Suitably low energy conformations were then applied to the model. The glycosylated model was further refined for rational ligand design and functional binding site mapping.

Example Three Binding Site Mapping

An aid to establishing the optimum amino acids to be included in a combinatorial library is the use of a site mapping methodology. By modifying the parameters conventionally used with the ligand design program LEAPFROG it has been possible to propose interactions between amino acids and the binding site of HBAι_c. We have modified several parameters associated with normal use of the software, so as to allow us to probe the binding site with all of the 21 amino acids and establish which of those could be used to produce an optimised combinatorial library. The program LEAPFROG was used with the default values apart from the changes. The binding site along the β dyad including the residues Vallβ, His2β, Lys82β, Argl04β and Hisl43β was selected as the active site for use with LEAPFROG. The program identifies the locations of hydrogen bond donors and acceptors, charged residues and hydrophobic residues. Entries from a virtual peptide library are then placed at random within the binding site and the interaction energies are estimated. If a suitable position is found for a particular amino acid, attempts are made to refine its position. After a large number of interactive steps, it is possible to identify key interactions. Results have been obtained both with the supplied 21 natural amino library and with a user defined unnatural amino acid and capping group library. Additional entries in this user defined virtual library include:

Unnatural Amino Acids

4-Chlorophenylalanine 3, 5-Diiodotyrosine 2-Fluorophenylalanine 3-Fluorophenylalanine 4-Fluorophenylalanine Citrulline HomoCitrulline HomoSerine 4-Nitrophenylalanine Phosphotyrosine Phosphoserine Phosphothreonine

Benzafibrate Derivatives

4-Chlorobenzene 2, 4-Dichlorobenzene 2, 3, 4-Trichlorobenzen

Results of Natural Amino Acids Mapping The LEAPFROG package completed 100,000 iterations in fewer than 24 hours, storing the best results of the "mapping" in a spreadsheet. This spreadsheet contained the relative co-ordinates of the best interaction with the binding site, together with an empirical binding score in Kcal/mol^-1. The key results are shown in Table 3.

The top two amino acids, glutamic and aspartic acid are anionic hydrophilic residues. This result is not unexpected because of the cationic nature of the binding site. Four of the six highest ranked amino acids feature in the AE1 band-3 peptide sequence.

Table 3 Results from the LEAPFROG Algorithm for HBA_lc with 21 natural amino acids

Although the binding site formed between the β- chains nearest the glycosylated termini consists of a large number of charged polar residues there are still important hydrophobic regions. The alanine, isoleucine and leucine residues bind to a cluster of residues that bring the hydrophobic side chains close to the Leu 78 β₂, Leu 81 β₂, Val 1 β₂ and part of the Lys 82 β₂. This region is the same one that the phenyl moiety of the BCCEP, a crosslinking reagent, is known to bind.

Results of Unnatural Amino Acid Mapping A virtual combinatorial library of unnatural amino acids and chlorobenzenes were tested with the LEAPFROG package. Similarly to the natural amino acids, it was possible to complete 100,000 iterations in less than 1_. day, storing the best results of the "mapping" in a spreadsheet. The key results are shown in Table 4.

Table 4 Results with unnatural amino acids

These results give a clear indication that the affinity can be significantly improved by modification of peptide libraries with unnatural amino acids. For example, a tyrosine residue has been estimated to interact with the protonated Lys 82β residue with an affinity of -32.80 kCal Mol^"1. The unnatural amino acid phosphotyrosine has been estimated to bind to the same residue with an affinity of -200 kCal Mol^-1, a six-fold increase in affinity. It can be observed from the results that affinity increases with the number of chlorine substitutions, the single substituted derivative has a very poor affinity and so did not feature in the LEAPFROG "hit list".

Example Four Design of de novo Ligands for HBAlc. The rational design of ligands was facilitated using the LEAPFROG package. This software can be used in two modes, either to optimise ligands that have already being discovered or to propose new lead ligand compounds directly. The LEAPFROG package was used with the HBAχ_c model produced previously to discover completely new lead sequences i.e. virtual combinatorial chemistry. The binding site along the β dyad including the residues Vallβ, His2β, Lys82β, Argl04β and Hisl43β were selected as the active site. The LEAPFROG package was configured as in Table 4.14.

Results

After 100,000 iterations of the LEAPFROG algorithm several interesting novel peptide structures were proposed. During these experiments the software failed several times, usually causing a system crash. The problem was isolated to the "Bridge" move part of the LEAPFROG algorithm. In situations where this was happening regularly, then this "Bridge" move was switched off and the number of iterations doubled. Many of these structures did not directly interact with the glycosylated termini and these have been rejected from this discussion. Three examples of peptide sequences obtained from the "de novo" modelling, are shown in Table 5. Table 5 Examples of some Novel Lead Ligands discovered using ^Λde novo ' design techniques.

The similarity of the GGAVL de novo ligand to the

GAIIVL combinatorial ligand gives an insight into the type of interaction that the combinatorial ligand provides. In this ligand the hydrophobic residues interact with the hydrophobic residues for example Val β₂, His β₂, whilst the more polar backbone gives hydrogen bond interactions with the pyranose, charged amine residues and backbone of the residues in the binding site. These interactions between the backbone of the ligand and the binding site simply do not provide enough discrimination between two forms. This result has several implications for the design of new optimised combinatorial libraries for the discovery of ligands for HBAι_c. Firstly, that specificity to the glycosylated valine termini will be improved considerably with the inclusion of charged polar groups such as DEQ etc. Examining some of the other lead sequences obtained from the LEAPFROG de novo experiment reinforces this hypothesis. The sequence ADCGE contains two such charged polar groups. This compound is estimated to bind better in the region of the glycosylated valine, and furthermore the charged groups are observed to form more bonds with the pyranose ring than the GGAVL sequence. Table 6 The Parameters used to apply the program LEAPFROG for Ligand Discovery Purposes .

Energy Startup

Include Hydrogen Bonding On

Grid Separation 0.1 A Runtime Binding Energy Tailor

Include Ligand Desolvation On

Include Cavity Desolvation On

Calculate Charges after Any Change

Charge Dictionary to Use Gasteiger Tactics/Evaluations

Strain Minimisation Always Move Types Default Move Frequencies

"Join" Move 3

"Fuse" Move 0

"New" Move 5

"Fly" Move 1

"Twist" Move 1

"Refine" Move 4

"Bridge" Move 1

"Complement" Move 1 " Save " Move 2

"Crossover" Move 0

"Prune" Move 0

Synthetic Difficulties Off

The functional mapping, "de novo" design and virtual high throughput screening techniques described above were able to provide many important prompts for rational design. Used together with information gathered from literature it was possible to design and model different types of ligand compound for the detection of HBAι_c. We have found that the pyranose ring is situated close to the 2,3-DPG binding site. Our "de novo" design experiments found that glutamic acid and aspartic acids were observed to interact with this binding site.

Glutamic acid interacts with Lys 132β2. Aspartic acid interacts with Lys 82β2. Furthermore our results indicate that the phosphorylated amino could be capable of inducing protonation in the terminal NH₂ group of the lysine (pi 9.74) .

The phosphinic acid moiety of phosphothreonine interactsstrongly with the terminal amine of residue β Lys 82. Example 5: Ligand portions binding to sugar residues

It has been found from extensive modelling and BIAcore studies that natural amino acids alone will be of restricted use for the detection of the presence of the pyranose adduct. Boronic acids have been described many times in the literature for the detection of carbohydrates, forming reversible covalent bonds with cis-diols present in cyclic glucose. Using the limited parameterisation, it was not possible to model the energy of interaction occurring during the condensation reaction. Approximate binding scores were calculated by breaking the O-B bonds and using the scoring algorithm of the DOCK program. A preferred technique for docking this ligand into the binding site would be the use of FLEXIDOCK, a genetic minimisation algorithm that allows the flexing of both the ligand and selected residues within the binding site. However, the use of boron prevents the use of this program since parameterisation of boron with FLEXIDOCK is not available. These technical problems resulted in a more manual route to ligand design, using a knowledge based rational design approach with the use of molecular mechanics for docking simulations. Our rational design approach has proposed ligands that interact directly with the cis- diols from the pyranose ring, and also with the 2,3-DPG binding site. In the limited time available to develop new chemistries, new ligands were proposed that included a phenylboronic acid moeity. Initially, three different carboxyphenyl boronic acid (PBA) based ligands were considered: PBA-Glutamic acid, PBA-Aspartic acid, PBA- phosphorylated serine. Models were developed and manually docked into the binding site, and bonds were created between the boronic acids and the cis-diols to simulate the reversible condensation reaction. After a final unconstrained minimisation to a convergence of O.OlKcal the O-B bonds were removed and the molecule was scored using the DOCK program. The results of each evaluation are shown in Table 7.

Ligand Binding Free Energy

Kcal Mol^"1

PBA-Glutamic acid (PBA-E) -75.28 PBA-Aspartic acid (PBA-D) -72.62 PB A-phosphorylated Serine, -200.11

Table 7 Docking of the rationally designed ligands

Although these empirical binding scores are approximate it can be observed that the PBA- phosphorylated serine ligand has a much higher quality interaction compared with the glutamic and aspartic acid ligands, a similar result to that found in the functional mapping experiments. The results for the ligands are higher than those for the single amino acids, since_. the phenyl ring forms additional hydrophobic interactions with the valine and leucine residues surrounding the pyranose ring.

After docking the PBA-aspartic acid ligand, it was found to form a hydrogen bond (2.07A) with the amine group on the sidechain of Lys β82. The approximate binding scores for the aspartic and glutamic ligands are similar; it is difficult to rank the better ligand from this result.

Further additions to the ligand would improve specificity to the binding site. Around the lysine β82 residue are a group of hydrophobic residues, forming a suitable cleft for binding. Adding a small complementary hydrophobic peptide chain would enhance specificity. This approach was explored by the addition of one of three hydrophobic amino acids to the chain A, I,V.

A model was constructed of PBA-DA to investigate this approach. After manual docking and minimisation using the DOCK program, the binding free energy score was reduced to -60.35 Kcal Mol^-1. The desired interactions between both the boronic acid moiety and the cis-diols, and between lysine and the aspartic acid can be observed in models .

The empirical nature of the binding free energy score calculations resulted in some uncertainty in the identity of the optimal ligand. This led to the proposal of a small diversity of compounds to be synthesised and evaluated using physical methods (as shown in table 8) .

PBA-DV PBA-EV PBA-PsV

PBA-DA PBA-EA PBA-PsA

PBA-DI PBA-EI PBA-PsI

Table 8 : Lead Compounds

Example 6 : Synthesis of new rationally designed 1igands/1ibrary

(A) An introduction to the synthesis of PBA-DAG, DIG,

DVG, EAG, EIG and EVG (where PBA is phenyl boronic acid)

In order to synthesise the six rationally designed ligands shown below, we have to look into the type of building blocks that may be required in order to complete the synthesis successfully.

Firstly, O-Bis- (aminoethyl) ethylene glycol trityl resin was chosen as a solid support because it has been proved successful in previous studies as a hydrophilic spacer that helps in the solubility of the final ligand as opposed to the e spacer which has been used in the past. The resin itself is very labile to acid, and therefore allows mild cleavage of the final ligands from resins .

Secondly, Fmoc amino acids used had to be properly protected to enable the synthesis and subsequent cleavage. The structure of the side-chain protected amino acids is illustrated below:

The six tripeptides with spacer HNCCOCCOCCNH₂ were prepared using the PE Biosystems peptide synthesiser. Standard Fmoc chemistry was employed for the synthesis. Aspartic acid and Glutamic acid were both O^fcButyl protected on their respective side chains.

Thirdly, once the tripeptides were prepared (after final deprotection, i.e., Fmoc removal) they were tested with the Kieser test (ninhydrin reagent) to confirm the presence of the terminal amine. A blue colour after. the test indicated that peptides were then ready to be coupled to the corresponding ester of carboxyphenyl boronic acid.

(B) Selection of Boron Esters Initially some feasibility studies were carried out using a preloaded amino ethylene glycol resin and carboxyphenyl boronic acid as a model system to see if we could directly couple the carboxylic acid group in carboxyphenyl boronic acid to the amine on the ethylene glycol resin. We used manual and automated synthesis and carried out a number of reactions using a variety of standard coupling techniques to see if this synthesis was possible.

However, there seemed to be no reaction at all under a variety of reaction conditions. It was clear that there was insignificant activation of the carboxylic acid for reaction with the amine on the resin due to the presence of the boronic acid group:

In order to overcome this problem we prepared a number of esters of the boronic acid which have been used extensively in the literature in solution phase synthetic organic chemistry:

Diethanolamine Neopentyl glycol Pinanediol Ester Ester Ester

The three esters can be removed using acid to liberate the free boronic acid. The preparation of the three esters can be found in the experimental section. Generally the boronic acid captures the diol rapidly forming a stable boronate in very high yield with little or no need for purification. These esters were then used to couple to the amino group on the resin manually using PyBOP as the condensation reagent in DMF.

The diethanolamine ester is known in the literature as a good stable protecting group (cyclic boronate) . Its preparation is shown below:

Therefore, it may perform a dual function of helping overcome the boronic acid synthesis problem and being able to exchange with the cis-diol on the pyranose ring of HbAi_c and therefore help in its efficiency to bind to it. However, in the model system the diethanolamine boron ester did not couple well to the resin under a variety of other conditions. Kiesers test showed that a significant amount of uncoupled resin was present. This is partly due to lack of solubility of the ester, which results in large volumes of solvent required for synthesis and hence high dilution of reagents slowing down the reaction

The coupling of diethanolamine ester with glycol resin is shown below:

The neopentyl ester and especially the pinanediol diol ester have been used widely in boron chemistry as a protecting group for boron acids.

The neopentyl glycol boron ester did couple well to the resin after 4 h. Kieser test showed that very little unreacted resin was present. The pinanediol boron ester coupled completely to the resin after only 1 h. Kieser 's test showed that no unreacted resin was present and that the carboxyl group had completely coupled to the amino group of the resin. We therefore decided that the pinanediol ester was the best ester and should give us complete reaction with the six tripeptides.

(C) Synthesis of Pinanediol Boron Esters of DAG, DIG, DVG, EAG, EIG and EVG

(Coupling of peptides to Carboxyphenyl Pinanediol Boron Ester)

250 mg of each of these resins were suspended in dry DMF and DIEA and 4-carboxybenzene pinanediol boron ester was added. To this mixture was added PyBOP and the resulting reaction mixture was allowed to stir at room temperature. After 1 h the reactions had almost completed and they were left to fully react for another 3 h.

Kieser tests were carried out every 30-min to monitor the progress of the reactions which were shown to be complete after the 4 h period (i.e. the test was negative) .

After purification the resins were then treated with 95% TFA. The TFA would :- a) Remove the O^fcBu protecting groups from the aspartic and glutamic acids b) Remove the pinanediol ester to form the free boronic acid c) Cleave the peptide from resin.

The peptides were isolated as oils although attempts were made to solidify them. The presence of small quantities of pinanediol would account for the oily nature of the product that was not removed fully for subsequent analysis and evaluation. However, LC-MS and CHN confirmed the purity and presence of the correct structure for each of the peptides.

(D) Experimental Section A. Synthesis of Carboxyphenyl Boronic Esters

Synthesis of diethanolamine ester 4-Carboxyphenyl boronic acid (1.65g, 1 mmol) was suspended in THF and the minimum amount of DMF was added until all the acid had dissolved. To it was added diethanolamine (1.05g, 1 mmol) in THF and the resulting reaction mixture was heated under reflux for 16 h. The white precipitate formed was then filtered and washed repeatedly with copious amounts of hot THF to remove any unwanted starting material. The resulting solid, the diethanolamine ester (2.2g, 96%) was used without further

purification, mp. , 223 °C (lit., 223-225 °C, M⁺ 235)

Synthesis of Neopentyl ester

A suspension of 4-carboxyphenyl boronic acid (1.65 g, 1 mmol) in acetone was stirred vigorously and to it was added an equimolar quantity of neopentyl glycol (1.04 g, 1 mmol) . On addition of the neopentyl glycol the solution went completely clear and this solution was allowed to stir for 16 h at room temperature.

The acetone was then removed under reduced pressure to give a white solid (2.3 g) in quantitative yield. TLC of the solid confirmed that there were no starting materials present. M⁺ 234.

Synthesis of Pinanediol ester A suspension of 4-carboxyphenyl boronic acid (4.95 g, 3 mmol) in acetone was stirred vigorously and to it was added an equimolar quantity of pinanediol (5.1 g, 3 mmol) . On addition of all the pinanediol the solution went completely clear and this solution was allowed to stir for 16 h at room temperature.

The acetone was then removed under reduced pressure to give a white solid (3.0 g) in quantitative yield. TLC of the solid confirmed that there were no starting materials present. Result: M⁺ 303.

B . Reactivity of Carboxyphenyl Boronic Esters

(Coupling of 4-Carboxybenzene Boron Esters O-Bis- (aminoethyl) ethylene glycol trityl resin)

Diethanolamine ester

Under standard coupling conditions, the diethanolamine boron ester did not couple well to the resin. Kiesers test showed that significant amount of uncoupled resin was present.

Neopentyl glycol ester

Under standard coupling conditions, the neopentyl glycol boron ester did couple well to the resin after 4 h. Kieser test showed that very little unreacted resin was present.

Pinanediol ester

Under standard coupling conditions, the pinanediol boron ester coupled completely to the resin after only 1 hr . Kiesers test showed that no unreacted resin was present and that the carboxyl group of boron ester had completely coupled to the amino group of the resin.

C. Synthesis of Rationally Designed Peptides: DAG, DIG, DVG, EAG, EIG and EVG on Ethylene Glycol Resin

The six tripeptides with spacer HNCCOCCOCCNH₂ were prepared as described using standard methods. 0.25 mmol Fmoc Chemistry was used to carry out the synthesis for all six tripeptides.

In each case 0.25g (0.49 mmol/g) of O-Bis- (aminoethyl) ethylene glycol trityl resin was reacted with 1 mmol cartridges of Fmoc protected amino acids to give approx 250 mg of resin.

D. Synthesis of Pinanediol Boron Esters of DAG, DIG, DVG, EAG, EIG and EVG

( Coupling of peptides to Carboxyphenyl Pinanediol Boron Ester)

250 mg of each of these resins (0.12 mmol) were suspended in 2 ml dry DMF and 2.4 mmol of DIEA (410 Dl) and 1.2 mmol of 4-carboxybenzene pinanediol boron ester (0.72 g) was added. To this mixture was added PyBOP (2.4 mmol, 0.62 g) and the resulting reaction mixture allowed to stir at room temperature for 4 h.

Kieser tests were carried out every hour to monitor the progress of the reactions and they were completed after 4 h. Resins were filtered separately and washed several times with DMF, DCM and then dried under vacuum. The resins were used without further purification.

E. Synthesis of PBA-DAG, DIG, DVG, EAG, EIG and EVG (Cleavage of Resin)

The resin was then treated with 2.5 ml 95% TFA in DCM for 2 h at room temperature. After this time the resin was washed with 1 ml 95% TFA in DCM and the 3 x 1 ml DCM. 20 ml of ether was added to the resulting filtrate but no significant amounts of precipitate were formed for all six tripeptides. The six tripeptides were evaporated down to dryness under high vacuum, which results in the loss of all volatiles to liberate six light brown oils. In each case approximately 100-150 mg of product was obtained and sent for analysis of structure. Small traces of pinanediol present in the reaction after cleavage of the ester were not removed.

Mass spec data and elemental analysis of the synthesised peptides were obtained from Micromass UK Limited (Wythenshaw, Manchester) . A LC/Tof- MS fitted with a Z Spray™ Source was used. The spectrometer was operated in positive-ion electrospray (ES) . The samples were injected (20 l) into a mobile phase of acetonitrile and 0.1% formic acid at a flow rate of 1.0 ml/min delivered from a Waters-Alliance LC gradient system.

Results PBA-DAG-NHCCOCCOCCNH₂ M⁺¹, Expected 540.2398, C₂₂H₃₄BιN₅Oιo; Found 540.2484, C₂₂H₃ B]N₅θιo.

PBA-DIG-NHCCOCCOCCNH2

M⁺¹, Expected 582.2868, C₂5H₄oB₁N₅Oιo; Found 582.2955,

C₂₅H₄oBιN₅Oιo.

PBA-DVG-NHCCOCCOCCNH2

M⁺¹, Expected 568.2711, C₂ H₃₈BιN₅Oιo; Found 568.2802,

PBA-EAG-NHCCOCCOCCNH2

M⁺¹, Expected 554.2555, C₂₃H₃₆BιN₅O₁₀; Found 554.2636,

C₂₃H₃₆BιN₅Oιo. PBA-EIG-NHCCOCCOCCNH2

M⁺¹ , Expected 596 . 3024 , C₂₆H₄₂BιN₅Oι₀ ; Found 596 . 3100 ,

PBA-EVG-NHCCOCCOCCNH2

M⁺¹, Expected 582.2868, C₂₅H₄oBιN₅θ₁₀; Found 582.2970,

The average mass measurement error of that acquired relative to the theoretical for the six tripeptides is 1.6 ppm.

Biacore evaluation of new ligands

The carboxyphenyl boronic acid (PBA) ligands devised from rational design were evaluated using the BIAcore 3000 biosensor.

Each lead ligand was in the form of highly water- soluble oil. When a carboxylated dextran surface was used initially, the ligands were found in general to physically adsorb strongly to the sensor chip surface. A flat (no extended matrix) carboxylated sensor surface dextran surface, used to investigate whether this binding was due to the interaction between the ligand and the dextran and found that this was not the case. The ligand still physically adsorbed to the surface of the flat chip; a lOμl injection of a 0.5 mgml^"1 concentration of the ligand produced a 147 RU increase in signal. It was found that desorption of this material was possible using a lOμl injection of sodium lauryl sulphate (SDS) 0.1%. An attempt to prevent this physical adsorption was made by adding 20% DMSO into the ligand solution. This reduced the physical adsorption by approximately 70%. Another method used was the addition of 20% glycerol to the ligand solution. This also was found to reduce the physical adsorption effect to acceptable limits, this was the preferred approach, since the BIAcore flow cells are not very tolerant to DMSO. After immobilisation, the remaining physically adsorbed ligands could be removed using a number of small pulses (lOμl) of 0.1% SDS.

The affinity peptides were functionalised by means of a polyglycol spacer group that allowed direct coupling to a carboxylated dextran modified sensor surface. The free amine group was linked via a peptide condensation reaction using the reagents l-ethyl-3,3- diethylaminocarbodiimide (EDC) and N-hydroxysuccinimide (NHS) : NH — PEPTIDE

By using the sensorgram recorded during immobilisation it was possible to monitor the activation, the capturing of the affinity peptide and the blocking of the activated groups on the sensor surface directly.

An example of a typical affinity peptide immobilisation by amine coupling follows:

BIAcore Configuration : -

Running Buffer : Hepes Buffered Saline pH 7 . 4 ( Physiological )

+ 0. 01% T een 20 + 0. 005% Na Dextran Flow Rate : 5μl per minute . Activation Reagent : 0. 05M NHS/0.2M EDC

Deactivation Reagent : Ethanolamine Hydrochloride IM pH 8 . 5

A solution of the affinity peptide was dissolved in the correct coupling buffer. The carboxyl groups on the sensor surface are activated by a seven-minute pulse of 0.05M NHS/0.2M EDC. When a suitable quantity of peptide was immobilised, ideally 700 to 1000 resonance units, the surface was then deactivated using a seven-minute pulse of ethanolamine. Reference surfaces were prepared by activating and then deactivating the separate sensor surface. In this way each of the lead ligands were immobilised onto a biosensor surface. It was not, however possible to obtain a completely uniform level of immobilisation for each of the ligands, as they coupled with differing efficiencies. The more hydrophobic ligands were found to be difficult to immobilise, an additional 0.1% Tween 20 was added to the ligand mixture and the contact time was increased to 21 minutes. Even after these further steps the ligands were found not to couple particularly well. A summary of the immobilisation of these ligands is shown in Table 9.

Ligand Immobilised Ligand

RU

PBA 2687

PBA-EVG 950 PBA-DVG 372

PBA-EIG 235

PBA-DIG 225

PBA-EAG 3000

PBA-DAG 966

Table 9: Immobilisation of the Lead Ligands.

After immobilisation, each of the ligands was screened for suitability using standards of HBAι_c and

HBAo . Each standard was prepared by exchanging the buffer with HEPES pH 8.4 as this was found (by Lifescan) to promote the stability of haemoglobin) . The addition of 2- 3 molar (ion) excess of sodium dithionate improved the binding interactions. This is because the deoxy form has a more cationic charged β-dyad, and the conformation of

the binding site changes, increasing the size of the β-

dyad cavity. Thus 0.5mg.ini^"1 of oxy-form gave 204RU while 0.5mg.ml^_1 of deoxy-form gave 296RU. For the screening of the ligands the BIAcore was configured as follows:

BIAcore Configuration: -

Running Buffer : Hepes Buffered Saline pH 8.5 + 0.01% Tween 20 Flow Rate 20-30μl per minute.

The BIAcore was found to be a very useful tool for this evaluation since it was possible to monitor the interactions between ligand and the haemoglobin in real time. The results of the screening exercise are shown in Table 10:

Sensor Response (Resonance Units)

Ligand HBAi_c HBAo

PBA 250μg 405.4 250μg 340.0

PBA-ENG 250μg 303.8 250μg 288.3

PBA-DNG 250μg 285.5 250μg 250.0

PBA-EIG 250μg* 12.8 250μg 3.8

PBA-DIG 500μg* 264.2 250μg 78.0

PBA-EAG 250μg 6370.0 250μg -6.5

PBA-DAG 250μg 7810.0 250μg 208.5

Table 10: Summary of the ligand screening results. Note * these ligands are immobilised at a very low surface concentration . After each experiment the sensor surface was regenerated. Previously regeneration has been achieved using pulses of 0.1% SDS and 20mM HC1. However, regeneration of these surfaces with such treatment reduced the activity of the sensor rapidly. The approach used for the analysis of the haemoglobin standards applied the HBA₀ standard first, then if any significant binding occurred only a single pulse of 0.1% SDS was used to regenerate the surface. The surface was allowed to equilibrate for several minutes before the next injection of standard, before injection of HBAι_c. After only a few injections over the surface, the binding decreased significantly. Often it was possible to regain the binding by an extended buffer wash of 30-40 minute. It is clear a better regeneration protocol is needed to allow for an advanced characterisation study of these lead ligands .

The carboxyphenyl boronic acid attached to a polyalcohol spacer moiety was evaluated in the same way as the other ligands. The performance of this material alone was quite poor giving only a slight increase in binding affinity, and appeared to have a very fast K_off rate. Typical results from this experiment are shown in Figure 1. As shown in Fig. 1 (i) the immobilisation of PBA- spacer on to a CM5 sensor and blocking with ethanolamine, resulted in 2687 RU. Fig. 1 (ii) shows that injection of 250 μgml^"1 HBA₀ onto the bound PBA gave a binding response at 360 RU. Fig. 1 (III) shows that injection of 250 μgml^{" 1} HB i_c onto PBA gave a binding response of 405 RU.

Fig. 2 shows the immobilisation of PBA-DAG-spacer and subsequent sensor response to haemoglobin. Fig. 2 (i) relates to the immobilisation process and shows response c/w baseline at (a) EDC-NHS 103.4RU; (b) after ligand 1096; and (c) after Ethanolamine 966. Fig. 2 (ii) shows the effect of injection of 250μgml HBA_ic onto PBA- DAG:- Response c/w baseline at (a) 7810 RU and (b) 612 RU. Fig. 2 (iii) shows the effect of the injection of 250μgml^"1 HBA₀ onto PBA-DAG:- Response c/w baseline at (a) 208 RU (b) 30 RU. Other ligands were tested in the same way.

From this initial screening work we found the ligands PBA-DAG and PBA-EAG to be the most suitable ligands for further investigation and development. They present high level of immobilisation, whilst providing high specificity to HBA_lc. The ligand PBA-DAG was selected as the main lead ligand for further characterisation. To examine whether the batch of standard haemoglobins had an effect on the binding results, to eliminate errors due to poor surface regeneration, or to realise any artefact due to the sensor chip, new standards were extracted from whole blood and the experiments were repeated using different sensor chips. The results of this experiment are shown in Table 11.

Experiment PBA-DAG HBAi_c HBAo loading Cone Response Cone Response

1 1670 RU 250μgmr' 9375 RU 250μgmr' 875 RU

2 996 RU 250μgml^"1 7810 RU 250μgml^"1 108 RU

Table 11 : Additional screening results obtained from different standards of HBAic and HBAO and different levels of ligand loading.

These results indicate that the ligand PBA-DAG has at best a 2% cross-reactivity. This indicates that the level of ligand loading on the sensor surface is another factor in the final device selectivity towards HBAι_c. The cross-reactivities of the ligand towards other haemoglobin variants and serum proteins were briefly examined. A solution containing 900 μgml^"1 of non-HBAlc haemoglobins was injected over a PBA-DAG modified CM5 sensor. In a typical experiment, injection of a 900μg ml^{" 1} solution of Non-Ai_c proteins gave a binding response of

493 RU.

Blood plasma taken from the hemolysation process was processed by exchange of buffer to HBS (pH 8.5 lOmM containing 0. IM NaCl and 0.1% Tween), dilute 10:1 and injected over a PBA-DAG surface. The proteins were found to bind non-specifically to the ligand modified surface, giving a signal of 1106.1 RU after a 40μl injection. The binding response was 495 RU. In further work we sought to rebuild the oxyhaemoglobin model and re-dock some of the ligands into the binding pocket to examine the binding characteristics. A new molecular model of the dimer form of oxygenated haemoglobin was generated using the Brookhaven Protein Databank data. There was only a single entry for this R state variant of haemoglobin. The pdb file was imported into the Sybyl environment. The water molecules present in the crystal were removed as were the haem groups. The hydrogen atoms were added to the model such that the "anionic sink" had a zero net charge.

The methodology and parameters evaluated found to give the best resulting structures in the absence of explicit water were applied to the whole haemoglobin model. A step-wise approach was used to refine the deoxyhaemoglobin model as indicated in Table 12.

Parameter Value

Convergence 0. 001

Dielectric 20

Constant

Charges KOLLMAN-AΣ

Algorithm BFGS (a)

POWELL (b)

Table 12: Parameters used to refine the structure using a step^→wise energy minimisation approach.

The program MAXIMIN2 was used to energy minimise the hydrogen atoms using the parameters shown in Table 12, with the BFGS (second derivative) algorithm. The side chains were energy minimised using the Powell algorithm to accommodate the larger number of calculations. Finally, a subset of atoms to a distance of 10 A around each of the terminal amino acids were energy minimised using the Powell algorithm. The remaining part of the main chain was not optimised. A model of valine- cyclopyranose was produced using the cyclic glucose (GLA) entry from the carbohydrate database and valine from the amino acid database. The model was charged using the "Gasteiger-Huckel" method and energy minimised to a convergence of 0.001 Kcal mol^"1 and using the distance varied dielectric function with a dielectric constant of 1. The rotatable torsion bonds were identified between the valine side chain and the pyranose ring. Using an α random impulse perturbation dynamics search it was possible to calculate the molecular mechanical energy at each allowable conformation. The lowest energy conformation was used to modify the HBAo model.

The refined model of deoxyhaemoglobin was modified by replacing the β-valine residue with the valine- cyclopyranose residue and assigning the lowest energy conformations selected by the search. The suitability was evaluated manually and the energy of the molecule was evaluated to avoid selecting a conformation that would cause atoms on the protein to become too close, and therefore unfavourable, to the pyranose ring. After conjugation of the carbohydrate onto the valine terminal amino acid, energy minimisation was used to locate local minima. An aggregate was formed using the entire structure except for the glucose-valine residue and minimisation was performed using the Powell method with a convergence of 0.001, a distance varied dielectric variable and a dielectric constant of 20. The charge parameter was set to "use current".

This model was then used with LEAPFROG and with FLEXIDOCK to try and visualise and analyse the results from the Jerini "screening" analysis.

a) Mapping of the Binding Site using LEAPFROG.

The LEAPFROG program was configured to only probe the binding site with prospective functional groups, but not to attempt to join together successful hits from the database amino acids. The binding site was defined to include atoms 8A around the N-terminal valine group~ This

was achieved by modifying the default "move frequency" parameters as shown in Table 12. The start-up energy was configured to include hydrogen bonding and the scoring function was adjusted to the "QUALITY" setting to improve the quality of the scoring function. The software was configured to run for 200,000 iterations, and the results were stored in a database for analysis.

Energy Startup

Include Hydrogen Bonding On

Grid Separation 0.1 A Runtime Binding Energy Tailor

Include Ligand Desolvation On Include Cavity Desolvation On

Calculate Charges after Any Change

Charge Dictionary to Use Gasteiger-Huckel Tactics/Evaluations

Strain Minimisation Always Move Types Default Move Frequencies

"Join" Move 0

"Fuse" Move 0

"New" Move 10

"Fly" Move 1

"Twist" Move 1

"Refine" Move 4

"Bridge" Move 0

"Complement" Move 0

"Save" Move 2

"Crossover" Move 0

"Prune" Move 0

Synthetic Difficulties Off

Table 12 The Parameters used to apply the program LEAPFROG for Binding Site Mapping Purposes .

The results of these experiments proved interesting, since they reproduced some of the original combinatorial library results. In the uncharged site the hydrophobic amino acids were found to be far more important than in previous work with the charged deoxygenated haemoglobin binding site. This is illustrated in Table 13

Amino Acid Binding Energy Optimal

Score Binding

(Kcal mol^"1) Interactions

Tyrosine -46.99 Valine 1 β GLA-OH

Serine -44.32 Valine l β GLA-OH

Alanine -38.72 Leucine 78 β Leucine 81 β

Theorine -33.0 Aspartic Acid 79

Isoleucine -32.15 Leucine 78 β Leucine 81 β

Leucine -19.29 Leucine 78 β Leucine 81 β

Methionine -18.5 Leucine 81 β Leucine 78 β

Table 13. Results from the LEAPFROG Algorithm for HBA_lc with 21 natural amino acids.

These results also complement the results found with the Jerini Libary in that the polar amino acids tyrosine and serine have been observed to interact with the glucose residues. From these results it appears that if part of the peptide forms some kind of complementary fit to the binding site then a correctly orientated polar amino acid such as serine or tyrosine will sense the presence of the glycosylated residue.

(b) Docking Simulations

The Jerini library, as screened by Lifescan, produced some very interesting results. Some of the "lead" peptides found in this way include: Acetyl-GSIIVL Acetyl-GYIIVL Acetyl-GAIIVL

Models of each peptide were constructed using a biopolymer. Hydrogen atoms were added to fill all valences. The peptides were charged using the Gasteiger- Huckel method to produce a 0 net overall charge. The program FLEXIDOCK was used to dock the peptide into the binding site. This is a powerful genetic algorithm (GA) that allows the exploration of conformational space of the peptide and of selected residues within the protein binding site, during docking simulations that run for days rather than weeks. The binding site was defined as the cleft surrounding valine βl, extending to a distance of 10 Amstrong radius. The entire backbone of the peptide was considered flexible. The backbone of residues lys 82β, val lβ were configured as being

flexible. All hydrogen bond donors and acceptors on both the peptide and the binding site were considered during docking. The seed generation was random. The number of generations to be considered was 100,000.

In each case the GA found several docking solutions. However, using the potential energy function results it is impossible to rank any of the peptides in terms of binding affinity.

It was observed by modelling that the polar amino acid tyrosine interacts with the fructosamine hydroxyl groups as predicted by the LEAPFROG algorithm discussed previously.

These modelling results may help to explain the results obtained from the Jerini library screening. In previous reports that use deoxyhaemoglobin as a model, amino acids that fit to the template of the band 3 AE1 proteins and the 2,3-DPG allosteric effector are prevalent. However in the oxygenated form that lacks specifically, charged amino acids the shape specificity, hydrophobicity and more subtle interactions such as those between the serine and fructosamine hydroxyl groups are more important.

Claims

CLAIMS :

1. A receptor molecule comprising a first portion adapted to bind specifically to a first region of a target molecule, a second portion adapted to bind specifically to a second region of the same target molecule, and a flexible linker portion including a functional group whereby the receptor molecule is immobilised or immobilisable to a sensor surface.

2. A receptor molecule according to claim 1 wherein one of the first and second portions is adapted to bind specifically to a peptide region of the target molecule .

3. A receptor molecule according to claim 2 wherein said one portion is adapted to bind specifically in a binding pocket of the target molecule. . A receptor molecule according to claim 2 or claim 3 wherein said one portion comprises an oligopeptide preferably of 3-6 amino acids.

5. A receptor molecule according to claim 2, 3 or 4 wherein the other one of said first and second portions is adapted to bind to a sugar region of the target molecule.

6. A receptor molecule according to claim 5 wherein said other portion comprises a boronic acid or a metal species able to chelate with a cis-diol of a sugar residue of said sugar region.

7. A receptor molecule according to any preceding claim wherein the linker region comprises a chain of at least 4 atoms.

8. A biosensor comprising a sensing surface having a receptor molecule according to any preceding claim immobilised to it via the flexible linker portion.

9. Use of a biosensor according to claim 8 to detect the target molecule of the receptor molecule.

10. A process of producing a receptor molecule according to any of claims 1-6 comprising providing a computer with a structural dataset for a target molecule; providing data for a multiplicity of species which are candidates for the first region of the receptor molecule and using the computer to determine which have the best binding properties; and synthesising the receptor molecule by linking a selected candidate, a species for providing said second region, and a flexible linker portion.