Water molecules in crystal structures?

[danaklug]

I went back and had a look at the NGL viewer of Fragment 373 for MurD (#1) and noticed some water molecules that could potentially be displaced by substituents off of the aromatic ring:

Computational chemists/crystallographers - any insights as to whether these displacing these waters would improve binding? Are they always there, or possibly just artifacts of crystal soaking? Any advice/discussion greatly appreciated! @drc007 @tkrojer @LizbeK

[drc007]

@danaklug As a rule of thumb unless a water is buried in a hydrophobic pocket displacement does not afford significant advantages.

[danaklug]

@drc007 Makes sense - could we still potentially pick up the H-bond interaction that the water was making and improve binding that way?

[drc007]

Ok, this is gross simplification. Displacement of "unhappy" waters in a hydrophobic environment, reasonable chance of benefit. Replacing a water in a hydrogen bonding interaction, may gain in affinity but on the other hand you lose because you have to desolvate your ligand, gain on the swings lose on the roundabouts. Result might be useful for specificity but difficult to predict increase in affinity. Replacing full solvated water molecules no benefit, and may be detrimental. Challenge is to identify "unhappy" waters. This is computationally challenging, you can read more about it here https://siremol.org/pages/apps/waterswap.html. You should consider 24-72 h CPU time (depending on the number of WaterSwap iterations you carry out) on a 16-36 core machine for a single WaterSwap calculation.

[mattodd]

Nice thread. @BenedictIrwin are you able to pitch in a little here? We spoke about this a little at the AI3SD meeting.

[BenedictIrwin]

Hi all, I'm new to this project.

I worked on a method to predict the free energy change of extracting a hydration site (time averaged water molecule), using molecular dynamics (MD) simulations.

We used it in the prediction of a binding of two proteins : https://onlinelibrary.wiley.com/doi/full/10.1002/prot.25589?af=R

But it was also used to find hydrophobic patches on proteins: https://iopscience.iop.org/article/10.1088/0953-8984/28/34/344007/meta

If we have a .pdb of a structure that can be used for MD in a water box, I could try to run these algorithms if that helps? It's been a while so I might need some time to get back up to speed. This would identify the free energy associated with desolvating each hydration site around the protein.

We used to run the simulations on the HPC, and the analysis could run overnight.

Ben

[drc007]

@BenedictIrwin This is very interesting. Would it be worth asking the Diamond folks about the X-ray resolution before embarking?

[BenedictIrwin]

@drc007 We would need to be able to see the individual sidechains to build an appropriate model of the protein. If there are frozen waters visible in the structure (as in the snapshot), I would guess the resolution is good enough for MD.

I will need to get up to speed about which files are available. Do we already have a .pdb? If it is just 3UAG, then the resolution of 1.77A should be fine.

[drc007]

@BenedictIrwin Here is the PDB file https://github.com/opensourceantibiotics/murligase/blob/master/docs/MurD_pdbs_forNGL/MurD/373.pdb I used MOE to correct a few minor errors.

[hannahbrucemacdonald]

Sorry for chiming in - I was sent this thread on twitter!

I agree with what is said above - it's hard to tell if there's gain in displacing a water molecule or not - as @drc007 says, it depends on how 'happy' the water is, and if you could replace it with a 'happier' R-group on the ligand. I'd also add that there is generally an entropic gain of displacing a localized water molecule into bulk.

Waterswap is able to perform absolute free energy calculations, where the ligand you are turning off is replaced with a volume of water. This will tell you about the affinity of the ligand, but won't tell you about the binding free energy of the waters themselves (which are circled in red). There are other methods that are able to afford the binding free energy (or the happiness) of the water molecule itself. It is weakly bound (and therefore more easily displaceable), then it might be worth looking at substituents on the ring that could displace them.

[hannahbrucemacdonald]

https://pubs.acs.org/doi/pdf/10.1021/ja066980q

Is a good reference! tightly bound water molecules are generally located in highly polar cavities and they can make three or four hydrogen bonds with the protein and the ligand; loosely bound water molecules are generally located in partially apolar cavities and they can make less than three hydrogen bonds with the protein and the ligand

given the binding free energy of a water molecule, it is possible to calculate the probability of the water molecule being conserved or displaced by a ligand. This knowledge may then be used to focus the synthesis of ligands ad hoc, to maximize the interactions with conserved water molecules and target those that may be displaced

[mattodd]

Interesting @hannahbrucemacdonald and @BenedictIrwin (and don't feel the need to apologise for chipping in - that's the whole point!)

[mattodd]

Paul Brennan said by email "It depends on what kind of water it is. If buried and making h-bonds between frag and protein we tend to leave it. If on the edge of the protein we try to displace it. If you want to do it right, you use water map to see if it is a happy or sad water and try to displace the sad ones" which I think captures this very well and matched the thoughts above.

Also, Peter Kenny: H2O accepting HB from target likely to be easier to displace than H2O donating HB to target (H2O is better HB donor than acceptor). This blog post discusses HB donor/acceptor asymmetry (post)"

[danaklug]

Really useful information, thanks so much to everyone who has contributed so far. I'm currently working on the synthesis of follow-up compounds with substituents at ortho, meta, and para positions relative to the amide bond containing some H-bond donors and acceptors. These should be fairly simple to access synthetically.

I think where some computational work would be most useful would be pointing to analogs that are maybe less synthetically accessible, but have the potential to displace an unhappy water. If that's possible with the information we have now, great. If not, we can potentially circle back once data comes back from the follow-up analogs.

[drc007]

@BenedictIrwin I've been reading about the Swarmdock algorithm, has anyone compared the results with other tools such as Watermap, waterswap, AcquaAlta, 3DRISM?

[BenedictIrwin]

@drc007 Sorry, I should have been clearer. Swarmdock is not the method for water calculations. The method used is Solvaware, by David Huggins etc.

[drc007]

@BenedictIrwin That makes more sense! Thanks.