Medicinal chemistry and drug design

Most drugs act by binding to a protein target through the same intermolecular forces you meet elsewhere: hydrogen bonds, ionic interactions, permanent dipoles and London forces. The arrangement of groups a molecule needs for binding is its pharmacophore.

Shape matters, including chirality. Proteins are chiral, so the two enantiomers of a drug can behave differently. Only (S)-ibuprofen inhibits the target enzyme, although the body slowly converts the R form into the S form. With thalidomide, one enantiomer is linked to birth defects, and because the enantiomers interconvert in the body, giving the 'safe' one alone does not solve the problem.

Some drugs bind permanently. Aspirin transfers its ethanoyl group to a serine residue in the cyclooxygenase enzyme, so the enzyme is switched off for good.

A typical discovery path:

1. choose a target linked to the disease; 2. screen thousands of compounds to find hits that bind weakly; 3. pick a lead and make analogues to improve it, studying how each change affects activity (structure–activity relationships); 4. test the best candidate in clinical trials.

A drug taken by mouth must cross cell membranes, which are lipid bilayers. Chemists measure this with the partition coefficient between octan-1-ol and water:

Too low and the drug cannot cross membranes; too high and it dissolves poorly and sticks to fatty tissue. Lipinski's rule of five sums up what most oral drugs look like: no more than 5 hydrogen-bond donors, no more than 10 acceptors, relative molecular mass under 500 and log P no higher than 5.

Measuring binding

Potency is reported as IC_{50} (the concentration that halves activity in an assay) or as a dissociation constant K_d, usually quoted as pK_d = -\log_{10} K_d. Because \Delta G = RT\ln K_d, each tenfold gain in affinity is worth about 5.7 kJ mol^{-1} at 298 K.

Bigger molecules usually bind more tightly simply because they make more contacts, so chemists normalise. Ligand efficiency is the binding energy per heavy (non-hydrogen) atom:

where the 1.37 converts pK_d to kcal mol^{-1} at 298 K. A compound with K_d = 10 nM (pK_d = 8) and 30 heavy atoms has LE = 1.37 × 8/30 ≈ 0.37 kcal mol^{-1} per atom, which is respectable; values below about 0.3 suggest the molecule is carrying dead weight. Lipophilic efficiency, LipE = pIC_{50} − log P, checks whether potency is coming from specific interactions or just from greasiness.

SAR and bioisosteres

Structure–activity relationships are built by changing one feature at a time: remove an OH, methylate an NH, swap a ring. A bioisostere replaces a group with another that keeps the important interactions but fixes a problem. A tetrazole mimics a carboxylic acid (similar pK_a, more lipophilic and metabolically sturdier), as in losartan; fluorine replaces hydrogen at a metabolic hot spot; an oxetane can stand in for a gem-dimethyl or carbonyl group, adding polarity without extra hydrogen-bond donors.

ADME

• Absorption: depends on solubility and permeability; at gut pH ionisation matters, so log D_{7.4} (which accounts for ionisation) is more useful than log P.
• Distribution: binding to plasma proteins and partitioning into tissues determine how much free drug reaches the target.
• Metabolism: cytochrome P450 enzymes in the liver oxidise drugs (phase I), and conjugation with glucuronic acid or sulfate makes them more water-soluble (phase II). A drug absorbed from the gut passes through the liver first, so first-pass metabolism can destroy much of an oral dose.
• Excretion: mainly via the kidneys or bile.

Veber's guidelines add that good oral bioavailability is likely with no more than 10 rotatable bonds and a polar surface area of at most 140 Å^2.

Hit finding has diversified well beyond high-throughput screening. Fragment-based discovery screens small molecules (under about 300 Da, following the rule of three) by X-ray crystallography, NMR or surface plasmon resonance, then grows or links them; vemurafenib, approved in 2011, was the first marketed drug from this approach. DNA-encoded libraries let a single tube hold billions of compounds, each tagged with a DNA barcode read out by sequencing after affinity selection.

Covalent drugs have returned to favour. Acrylamide warheads on kinase inhibitors such as ibrutinib target a non-conserved cysteine, and sotorasib (approved 2021) traps the G12C mutant of KRAS, a protein long thought undruggable. The design problem is tuning electrophile reactivity so the drug labels its target but not glutathione or off-target proteins.

Targeted protein degradation changes the goal from inhibiting a protein to destroying it. PROTACs link a target ligand to an E3 ligase ligand; molecular glues such as lenalidomide reshape the ligase surface. Both often break the rule of five, which has prompted work on 'beyond rule of five' space: macrocycles and some PROTACs reach oral bioavailability by being chameleonic, hiding polar groups through intramolecular hydrogen bonds in membranes.

Structure-based design now draws on cryo-EM structures and predicted structures from AlphaFold, while free-energy perturbation calculations rank close analogues before synthesis. Machine-learning models for ADME and toxicity are used routinely, but they predict poorly outside their training chemistry. Open problems include predicting human pharmacokinetics and idiosyncratic toxicity from early data, drugging flat protein–protein interfaces and transcription factors, and shortening the design–make–test cycle with automated synthesis.