Structure based drug design

Drug discovery pipeline

• Usually would take 12 years between target protein and drugs that could be sold on the market, but recently, due to structure based drug design, this timeline collapsed down considerably 

• Drug design pipeline involves multiple phases:  

• Prediscovery 

  • Trying to understand a protein in a system that may have sometherepeutic benefit 

  • Understanding the basic mechanism behind disease  

  • Designing assays to assess the system 

• Hit-to lead discovery  

  • Once we’ve identified a target, potentially even a compound 

  • Optimisation steps —> micromolar to nanomolar affinity  

• Preclinical studies 

  • Toxicity tests in animals  

• Move on to clinical trials 

• Goal is to identify candidates early on and reduce the attrition process down  

 

Streamlining this process

• For optimising target identification and validation we can use software for modelling receptor structure 

• Accurate homology models which can be used to test the activity of various compounds 

• Screening in silico of high numbers of compounds with high degree of accuracy  

Methods used in drug design

• Biochemical assays  

  • Thermal stability assays  

  • Binding assays  

  • Quantitative structure activity relationship  

• Computational  

  • Docking  

  • Molecular dynamics  

• Structure based  

  • NMR  

  • X ray crystallography 

Experimental screening methods limitations

• Need to be able to isolate the target molecule in large scale 

• Assays have to be optimised to screen a library of compounds  

  • Analytical centrifugation for example can test up to 6 compunds and takes 2-3 hours  

  • Microscale thermophoresis for example, can be done with 96 well or 384 well plates  

• However, the positive is that there are a diverse range of methods we can use —> some are ore or less amenable to high throughput analysis 

• Needs to be using ow levels of proteins otherwise it takes more money and effort for us to design more 

Characterising Binding

• These methods need to be able to tell us the Kd, the enthalpic and entropic contributions of the binding  

• We can then start to do structure activity relationship analysis which can help us to understand which parts of the protein are important in binding 

Computational methods for design

• We don’t need to express and purify the receptors in silico so advantageous 

• Compounds may be challenging to synthesise but can be made relatively quickly in silico and understand their properties 

• Can take the molecules of interest and use computational methods to assess binding sites  

Determining affinity

• To work out the Kd we must work out the delta G first which has an enthalpic and an entropic contribution  

• Trying to compute terms like entropy (more than entropy) and enthalpy can be challenging  

• Requires a structure or a good homology model and this can be difficult to determine the side chain conformation or the rotomer conformation 

• However, more recently, improvements in the quality of modelling has made this less of an issue 

Molecular dynamics

• Assumed that the chemistry is occurring under a classical regime (not 100% true) which allows you to assume Newton’s law and work out the forces that are acting on a particle 

• We know the mass of each particle so can work out the acceleration of the particles in response to the force  

• In a computer, we can then integrate this as a function of time  

• The main difficulty is working out the forces that are acting on the atoms —> we can do this by looking at the force field of each atom 

• separated into bonded and non-bonded interactions  

  • Bonded interactions – the way that the covalent bonds, hydrogen bonds, steric clashing etc. react to being pushed part/pulled together 

  • Non bonded interactions: electrostatic interactions – more challenging to deal with because you have to integrate across vast quantities of space but has been done 

Why is this good for drug design

• Molecules are free to behave as they want to behave —> can get them to adopt a different conformation etc. 

• We explicitly include the solvent (water molecules) and can understand about the entropic contribution  

• We have an atomistic description of the whole system and can understand how a particular drug interacts with a particular binding site  

Challenges of molecular dynamics

• Simulations nanoseconds in length but proteins bind in microseconds so can be hard to drag these trajectories out  

• Can take a long time for molecules to move from one site to another and means it takes a long time to sample different sites → so poor for determining where binding sites are if you don’t know but good if we insert the molecule into the binding site  

• Classical, not quantum technique so doesn’t deal with how the electrons move → if you have inhibitors which interact with the transporters/receptor this is not taken into account  

• If drugs react with pi-pi and pi-cation interactions can be difficult to capture the polarisation of the bonds in the force field that you’re using 

• Difficult to calculate the delta G of binding because we have water included in this system  

  • However, this can be bypasses by calculating delta G, delta G in water and with the protein bound and because it’s a state function we can then calculate the delta G of binding 

Docking

• Makes the assumptions to simplify the calculation and find a mechanism to reduce the  computational intensity so it can be applied to a large number of compounds:  

• Works off the basis that if you have a drug and a binding site, and the drug enters in the correct orientation, it can fit into the binding site → there’s a good chance if it fits in the binding site, it will be a good drug target 

How do we identofy a binding site

• Calculate all the possible conformations of the durg and calculate the energy of the resulting complex 

Local 

• If you do a chemical labelling study using a native agonist and know that the binding site is in a particular location, we can screen for binding sites in a particular regons.  

• Then take all the conformations of the drug, rotate it in all the possible directions and see what the optimal interaction energy is 

Systematic 

• If you don't have the information about the binding site can do a systematic search: create a box and move your protein around all different areas of the box at all different orientations  

• Calculate an interaction energy for each conformation and rotation and see what the optimal one is 

Random:  

• Random conformations of the drug are placed randomly about the protein and the interaction is scored 

Stimulated:  

• Molecular dynamic driven drug conformations, followed by stimulated annealing based docking 

• However, this doesn’t account for flexibility of the binding site or the protein and so can be refined using molecular dynamics to help define the structure 

Scoring functions

• Calculating interaction energy can be quite time-consuming, so we can use the same forcefield used for MD to calculate interaction energies 

• The downside is we don’t have any quantum effects (cation pi interactions, polarisation of bonds, charge effects)  

Empirical or knowledge based approach

• Faster, less computationally demanding approach which is more suited to screening large libraries, is an empirical or knowledge-based approach  

• We know that certain types of interactions are important and how the directionality impacts the binding affinity 

• Develop an empirical forcefield using observations from previous studies which we can use to screen drugs – gives us a scoring function: the likelihood of a molecule of interacting with that protein, rather than a Kd  

• Computationally more efficient, can screen larger libraries 

• Statistical analysis of pairwise distributions  

Types of docking

Rigid body 

• Receptor and ligands which you assume have a fixed conformation and you assume binding between the two  

• Computationally very efficient only 6 degrees of freedom 

Flexible 

• Ligands and proteins have a range of different states —> usually assume flexibility of the ligand and can extend this to certain regions of the protein  

• But this slows this down and we have to test the energetics of hundreds and thousands of different conformations but is more accurate 

Example: nAchR

• Knew the conformation of the drug so can make the binding site flexible 

• Use docking techniques to see how the receptor responds to the presence of the agonist in the binding site 

Combining molecular docking and molecular dynamics

• Benefits of combining these techniques – not usually independent from one another  

• Use molecular dynamics to generate ensembles of structures of receptor protein  

• Let a molecular dynamics trajectory run and see that structure changes as a function of time  

• Cluster analysis: structures with several key featyres associated, take these and  use the docking type studies to screen against a whole range of different compounds and understand which compounds will bind best and which conformational state its likely to bind to 

When do computational methods work well:

• When the binding pockets are well defined  

• When there’s limited conformational flexibility in the drug and the target protein  

• Struggle with cryptic binding sites (e.g: sites of protein-protein interactions)  

• Easier to find the binding sites for rigid antagonists which do not induce a conformational change 

• Even if you don’t get an exact answer, can be used to: 

  • Opens up new conformational spaces which you can study which you hadn’t previously considered 

  • Can use it to design experimental screens if you know certain motifs favour binding  

NMR based methods

Ligand based 

• Rely on observation of the ligand, not the protein  

• No requirement for labelling of the target which can be good if your protein expressed very poorly  

• Typically, data acquisition is fast  

• No size restriction of the molecule, so can quickly screen large numbers of compounds 

• Examples:  

  • Saturation transfer difference NMR (STD-NMR)  

  • Transferred NOE (trNOE)  

Target based 

• Rely on looking on what the target of the ligand is 

• Need to label the protein with carbon-labelled isotopes 

• Expensive  

• Can then titrate in drugs and see how the structure is changing  

• Can do entire structure elucidation  

• Limited to smaller target molecules 

• Does provide very site-specific information about how the drug is interacting with the protein  

Saturation transfer difference (STD) NMR (ligand based)

• The signal that you obtain for an experiment is based of the difference between the excited and the ground state 

• We can saturate - equally populate the ground and excited state to produce no signal  

• But we can selectively saturate resonances that come from the protein so the signal from the protein will disappear  

• If a small molecule binds to the target protein also becomes saturated to that the signal from the compound will become attenuated as it binds to the receptor 

• Attenuation of the signal can be used to identify which molecule is binding and so is amenable to screening a large number of proteins which can all be screened as a mixture 

 

Example: HSA Binding

• Tryptophan with a  methyl group in 6 and 7 position  

• When you incubate these compounds, you find that the 7H3 binds and the signal is attenuated —> this protein is binding where the other one doesn’t 

Limitations of STD NMR

Doesn’t give us a lot of insight into how the compounds are binding, so we go to protein-based screening 

Protein based screening

• HSQC style plots: every amide in the protein backbone gives rise to one point plotted à the position of the resonances in the spectrum is indicative of the protein conformation  

• If the drugs bind to the protein it changes conformations and the peaks move  and look at how fast the peaks are moving 

• Enables us to measure a Kd and also allows us to identify which peaks have moved and identify a binding site of the protein  

 

Example:

• Titrate in protease from Zikka virus with a  potential drug  

• When you titrate in inhibitor peaks move, can plot the chemical shift perturbation (movement of peaks as a function of concentration )  

• Fit these to models to work out the Kd 

 

Crystallography

• Use X ray crystallography to refine the structures and identify compounds  

• If we have a well-defined protein we can use high-throughput screening and fragment-based screening  

  • High throughput screening: crystallise the protein in a range of different ligands —> see what binds in the binding site 

  • A good set of diffraction data for compound with the target protein can be fast to resolve high resolution structural data which can be used to refine target compounds so you can improve the binding affinity 

• Feeds into a drug development cycle: can be used to design the next range of compounds with higher affinity until you get a clinical candidate 

Drug target/structure – How do we get our drugs into crystal structures?

Cocrystallisation  

• Take a concentrated protein solution and add our drug of interest 

• If you leave this for a few days you will form a crystal of the drug bound to the target protein  

• Don’t need high solubility of drug  

• Protein ligand can be co-expressed  

• Requires screening of crystallisation conditions à conditions which gave you original structure of protein not going to be the ones which gives you a structure of protein and ligand  

• Relatively high throughput and are some techniques of automating it 

Soaking  

• Take protein which you’ve already crystallised 

• 50% of crystal is water/buffer àTake the crystal and put it in the buffer containing ligand of interest  

• Ligand will diffuse into the channels within the crystal and will bind to the binding site 

• Requires high concentrations of drugs so places limitations on the solubility of drugs  

• Assumes that the protein will not change upon ligand binding shape otherwise, the protein will fall apart 

Fragment based drug design

• Can generate chemical structures de novo  

• Not guided by a drug library  

• Take out target protein and incubate it with a fragment library – no longer looking for whole molecules binding woth high affinity 

• Small motifs which will bind t the drug and cause a shift in the thermal shift assay, NMR, or show some evidence of binding 

• Looking for any form of binding rather than high affinity  

• Can identify which fragments will bind and start to assemble them in a mechanism which will enhance the affinity of the target 

• If we link the two fragments:  

  • The delta H doubled  

  • The dela S is largely the same  

  • Can begin to link the candidate drug —> slight decrease in conformational flexibility 

Strategies

Fragment optimisation  

• Bind multiple fragments and see which ones bind best 

Linking  

• Linking and optimising of fragments 

Growing  

• If we know the fragment binds well, we can look at binding it to other fragments to see if this improves 

Merging  

• If 2 fragments which bind with high affinity, can look at merging these compounds in a way that they hold the same conformational space 

• Kd typically in the millimolar range which would not be effective, but by combining these fragments, we can reduce the Kd so that theyre in the nanomolar or micromolar range 

 

Example: targeting SARS-Cov-2

• Can create therapies which target the viral proteases or polymerase associated with the virus  

• Isolated the proteases and tried to identify compounds it was binding 

• X ray crystallography tried and screened a range of compounds 

• Found that the protease had a lot of water filled channels so tried soaking type experiments 

• Studies using fragment screens, mass spec: gave us 23 hits and 3 subsites 

• Ifentified binding to cys145 and therefore build a picture of what lind of molecules will inhibit the protease 

• Identify molecules which binds and target the catalytic triad which as a result targets the protein à Reactive N-chloro-acetyl group inhibits the protein  

• Then used fragments for selective inhibition  

• Example of a very rapid drug development which condensed the process down from 12-2 years