Tautomers: which structure did you model?

A tautomer shift moves one proton and rearranges one double bond. The formula does not change, the name does not change, the molecular weight does not change — and a hydrogen bond donor becomes an acceptor. Everything a binding site cares about has changed, and nothing a database index cares about has.

Students conflate these constantly, so it is worth being precise.

Resonance structures are different drawings of one molecule. No atom moves. The real molecule is a single thing, somewhere between the drawings, and the drawings are an artefact of our notation. A carboxylate is one species with the charge spread over two oxygens.

Tautomers are different molecules in equilibrium. An atom — specifically a hydrogen — genuinely moves, and a bond genuinely shifts. They can in principle be separated, they have different energies, and at any moment a given molecule is definitely one or the other.

Three families account for nearly everything you will meet.

The classic. A carbonyl with an adjacent C-H can shift that hydrogen to the oxygen, giving an alcohol and a C=C. For a simple ketone the keto form dominates overwhelmingly — acetone is about one part in a million enol. But put two carbonyls either side of the same carbon and the enol becomes competitive, because the enol can form an internal hydrogen bond and conjugate.

An amide can in principle shift its N-H onto the carbonyl oxygen, giving C(OH)=N. For a simple amide this barely happens — the amide is far more stable. But build the amide into a ring, particularly an aromatic one, and the balance changes completely. The 2-pyridone/2-hydroxypyridine pair is the textbook case, and both forms are populated in water.

Imidazoles, triazoles, tetrazoles and pyrazoles all have a mobile N-H that can sit on more than one nitrogen. The forms are often close in energy, sometimes essentially degenerate, and they interconvert rapidly.

This matters a great deal in practice because these rings are everywhere in drug space, and because the two tautomers present donor and acceptor in different places around the ring. A histidine in a binding site has exactly this problem, and which tautomer it adopts is a real question in structure preparation.

Here is why any of this matters.

Take a group written as C=O with an adjacent N-H. That is a donor next to an acceptor. Shift the proton and you get C-OH next to C=N: an acceptor next to a donor, in the other order. The hydrogen bonding pattern the protein sees has been reversed, and:

• the molecular formula is identical;
• the molecular weight is identical;
• the name is the same;
• most database identifiers are the same or nearly so.

So there is no signal anywhere in your metadata that anything has changed, and a docking run with the wrong tautomer will produce a confident, plausible, wrong pose — and often a better score than the right one, because a wrong donor-acceptor pattern can be complementary to something.

The equilibrium position depends on the environment. Two consequences.

First, ratios measured in water need not apply inside a binding site. A pocket has a different dielectric, different hydrogen bonding partners available and no bulk solvent. A minor tautomer in water can be the bound form, if the protein pays for it — which it will, if that tautomer makes better contacts.

Second, a tautomer that is 5% populated is not negligible. Binding is a coupled equilibrium: if only the minor form binds, the system simply replenishes it. The cost is the energy of forming it, which by the arithmetic of module 8 is small — a 95:5 ratio is under 2 kcal/mol.

A chemical database stores one structure per compound. If the molecule exists as several tautomers, the stored one is whichever the depositor drew, or whichever the registration software normalised it to. Different databases normalise differently.

The practical consequences:

• Substructure searches miss things. A query drawn as the keto form will not match a record stored as the enol.
• Descriptors differ. Donor and acceptor counts change between tautomers, so TPSA and every filter built on it change too.
• Duplicate detection fails. The same compound registered as two tautomers looks like two compounds — which is why canonical tautomer generation exists.

So where in a workflow do you have to think about this explicitly? Three points, and they are worth memorising.

• Before docking or scoring. Enumerate plausible tautomers and dock them. If they score differently, you have learnt something important about how much you can trust the result.
• Before a similarity or substructure search. Standardise to a canonical tautomer, or search several forms.
• When preparing a protein structure. Histidine's tautomer and protonation state are a choice someone has to make, and the default is not always right.