Lipophilicity: logP and logD

Lipophilicity is the single most consequential number on a medicinal chemistry project. It correlates with potency, with promiscuity, with metabolic clearance, with plasma protein binding and with toxicity — and with all of them in the same direction, which is what makes it a trap.

Shake a compound with octanol and water, let it settle, and measure the concentration in each layer. The ratio, logged, is the partition coefficient.

logP is the ratio for the neutral species alone. It is a property of the molecule and does not depend on pH.

logD is the ratio for whatever is actually present at a stated pH — neutral and ionised forms together. It does depend on pH, and it is written with the pH attached: logD₇.₄.

For a compound with no ionisable group, the two are identical at every pH. For everything else they differ, sometimes by orders of magnitude, and logD at pH 7.4 is the number that matters physiologically, because that is the pH of plasma.

The relationship is straightforward. Only the neutral form partitions appreciably into octanol, so logD falls below logP by the log of the neutral fraction — and the neutral fraction is exactly what Henderson-Hasselbalch gives you from the pKa. Which is why this module and the next are really one topic.

Three routes, in decreasing order of directness.

• Shake-flask. The definition, done literally. Accurate, slow, and it needs a lot of compound. It struggles above about logP 4, where too little stays in the water layer to measure.
• Chromatographic. Retention on a reversed-phase column, calibrated against standards. Fast and needs almost no compound, which is why it is what most projects actually use. It measures something correlated with partition, not partition itself.
• Calculated. Free, instant, and available before the compound exists — which is its real advantage. You can compute logP for a molecule you are only thinking about.

You will meet several calculated values — cLogP, ALogP, XLogP, MLogP and others — and they disagree with each other and with experiment. Fragment-based methods add up contributions from pieces of the molecule; atom-based methods add up contributions from atoms in environments; property-based methods regress against calculated descriptors. Each was fitted to a training set, and each is least reliable on molecules unlike that set.

Disagreements of one log unit between methods are routine. Two log units is not unusual for an unusual molecule. Zwitterions, very large molecules and heavily halogenated compounds are where they diverge most.

Raising logP tends to improve the assay numbers that are easy to measure early, and to ruin the properties that only show up later. Both halves of that sentence matter.

A more lipophilic molecule makes more and better van der Waals contacts with a hydrophobic pocket, and it pays less desolvation penalty on the way in. So potency usually rises with logP. It also crosses membranes more easily up to a point, so permeability improves.

This is why raising logP is such a reliable way to make an assay number look better, and why it happens by accident on almost every project: a chemist adds a lipophilic group, potency improves, the compound is progressed.

• Promiscuity. Lipophilic compounds bind more things. Off-target activity, including hERG, rises with logP.
• Metabolic clearance. Cytochrome P450 enzymes recognise lipophilic substrates. Clearance rises with logP, so half-life falls.
• Plasma protein binding. More drug bound to albumin means less free drug available to act.
• Solubility. Falls, roughly one log unit of solubility for each log unit of logP. Module 13 takes this apart.
• Toxicity. Analyses of large datasets consistently find higher attrition for compounds that are both large and lipophilic.

Here is the trap, stated plainly. Potency is measured in week one. Clearance, toxicity and solubility are measured in month six. So a chemist optimising on the data in front of them will drift upwards in logP, compound by compound, each step justified by a real improvement in the only number available.

The project then discovers, much later, that the series is too greasy to develop, and has to walk back changes that each looked correct at the time.

The defence is to watch a metric that combines potency with lipophilicity rather than potency alone. That metric is lipophilic ligand efficiency, LLE, and it is simply potency minus logP. Module 22 treats it properly; the point to take now is that it exists because logP-driven drift is the commonest failure mode in lead optimisation.

Rules of thumb, to be held loosely. For an oral drug acting outside the central nervous system, logP roughly 1 to 3 and logD₇.₄ roughly 0 to 3 is comfortable. For a CNS drug, slightly higher logP but low polar surface area matters more. Below logP 0, permeability usually becomes the problem; above logP 5, everything else does.

Lipinski's limit of 5 in module 20 is not a wall. Plenty of marketed drugs sit above it. It is a marker for where the risks start to compound.