Drugs Against a Part

In 2001 a capsule aimed at one enzyme, out of about five hundred human enzymes that share its pocket, was approved for a leukaemia. That same year a laboratory read, from the blood of patients who had relapsed, how the leukaemia was getting around it.

Every pathway in this course is a chain of parts: a receptor, a switch, a kinase, a docking site. A targeted drug is a molecule made to act on one of those parts. This lesson sorts such drugs by where on the part they land, because the landing place decides how selective a drug can be and how a disease will escape it.

A decoy in the receptor: fit without the push

Some of the first drugs designed against a signalling part were aimed at something nobody had seen. In 1948 Raymond Ahlquist sorted the actions of adrenaline-like compounds into two families by their order of potency on tissues, and named two receptor types, alpha and beta. No receptor had been purified. James Black joined ICI Pharmaceuticals in 1958 with a plan for angina, the chest pain of a heart whose arteries have narrowed: make the heart respond less to adrenaline, so that it asks for less oxygen.

A known compound, dichloroisoprenaline, showed the difficulty. In some of the group’s tests it blocked the heart’s response to adrenaline, and in others it acted as a stimulant itself, depending on the tissue. The chemist John Stephenson proposed swapping its ring for a naphthyl group. In the group’s tests the new compound, pronethalol, blocked the receptor without stimulating it. A safer successor, propranolol, followed.

The result means that fitting a receptor and switching it on are separable properties. The lesson on binding and dose-response calls them affinity and efficacy, and calls a molecule with the first and without the second an antagonist. Picture a key that goes into a lock and will not turn: while it sits there, the right key cannot get in. Black later did the same at the histamine receptor of the stomach lining, which led to the ulcer drug cimetidine.

That is the first class: a molecule that sits in a receptor’s own site and does not switch it on.

One pocket, five hundred kinases: “Highly Conserved— target for lots of drugs”

“Nearly Every Kinase Has Been Associated with Disease.” That is the case for kinase drugs. The case against was the first half of this section’s heading. Every kinase runs the same reaction, moving the end phosphate of ATP onto a protein, so the pocket that holds ATP looks much alike across about five hundred human kinases. Around 1990 the reasonable expectation was that a molecule shaped to jam that pocket in one kinase would jam it in dozens.

The test case was chronic myeloid leukaemia, a cancer of white blood cells. Its cause is a fusion of two genes whose product, BCR-ABL, is a tyrosine kinase that is always on and that no healthy cell contains. At Ciba-Geigy in Basel a group that included the chemist Juerg Zimmermann was changing single groups on one core structure to shift its preference from kinase to kinase. One added methyl group removed its activity against protein kinase C while keeping it active against ABL. Elisabeth Buchdunger’s pharmacology showed what the resulting compound hit: ABL and the PDGF receptor, and later KIT, with the EGF and insulin receptors barely touched. In Brian Druker’s cultures in Oregon it killed BCR-ABL leukaemia cells and left normal cells alive. The compound became imatinib.

Two kinds of selectivity are in that story. One belongs to the molecule: a short list of kinases out of hundreds, measured on purified enzymes. The other belongs to the disease. Normal cells also use ABL, the PDGF receptor and KIT, and they get through a period without them. A leukaemia cell does not get through a period without BCR-ABL. That second selectivity can only be measured in a living cell or a patient.

A structural account of the molecule’s selectivity came after the fact. As the lesson on the phosphorylation switch shows, imatinib binds ABL in its switched-off shape, and switched-off shapes are thought to differ between kinases more than working shapes do, though how much of the selectivity this explains is still studied. That is how the second half of the heading came true. The second class is a small molecule in the ATP pocket of a kinase: imatinib (approved 2001), gefitinib (2003), erlotinib (2004), and the JAK inhibitors from the JAK-STAT lesson.

When the drug stops working: one residue changed

Some patients in the early imatinib trials, mostly those with advanced disease, got well and then relapsed. In Charles Sawyers’s laboratory at UCLA, Mercedes Gorre sequenced BCR-ABL from their blood. In several patients one DNA letter had changed, ACT to ATT, which replaces threonine 315 with isoleucine.

Threonine 315 is one of the contacts that hold imatinib in its pocket. Isoleucine is bulkier and cannot make that contact. The enzyme does not need this residue for its own reaction, so the kinase keeps working while the drug fails to bind.

A relapse looks like a failure of the idea. Read it as an experiment instead. To survive, the leukaemia changed the one position the drug depended on and left the rest of the enzyme alone. That is strong evidence that the drug had been working through BCR-ABL, and that the leukaemia still needed BCR-ABL. It also wrote the specification for the next drug: grip ABL without needing residue 315. Dasatinib, a second-generation inhibitor, kills cells with other resistance mutations such as G250E, but T315I resists it too. Ponatinib was then built with a slim carbon-carbon triple bond where the isoleucine side chain sticks out, and it inhibits the T315I kinase.

Design problem

A patient’s leukaemia answers to a kinase inhibitor for a year and then returns. List three different ways the leukaemia cells could have escaped the drug while still depending on the same pathway. For each, say what you would measure to show it, and what kind of second drug you would design.

The target nobody could hit: a bond to the mutation itself

Ras, the GTP switch from the lesson on GTPase switches, was the standard example of an undruggable target. Ras holds its nucleotide with picomolar affinity while the cell keeps GTP near the millimolar range, so a competitor for that site must beat a ligand that binds very tightly and is present in vast excess. And the known structures of Ras showed no groove deep enough to hold a small molecule.

A first plan went around the protein. Ras needs a farnesyl lipid to reach the membrane, as the lesson on lipid anchors explains, so chemists made inhibitors of the enzyme that adds it. In clinical trials these drugs did not control cancers in which K-Ras is commonly mutated. One reason is that when that enzyme is blocked, a related enzyme gives K-Ras a different lipid, geranylgeranyl, and K-Ras reaches the membrane anyway.

Kevan Shokat’s laboratory went at the mutation itself. Glycine 12 has no side chain. One cancer mutation replaces it with cysteine, which puts a thiol, a reactive sulphur-hydrogen group, on the surface of the protein. The group held purified K-Ras G12C in its GDP-bound shape, passed 480 sulphur-seeking fragments over it, and weighed the protein to see what had attached. Two fragments attached to most of the protein. Normal K-Ras, which has three cysteines elsewhere, showed no detectable reaction.

The crystal structure showed something new. The compound lay in a pocket beneath the switch II loop that no earlier Ras structure had shown. Once bonded, it tilts Ras toward holding GDP and gets in the way of Raf. Because the bond lasts for the life of that protein molecule, the compound never has to out-compete GTP. In 2019 Canon and colleagues reported sotorasib, a drug for patients that bonds to the same cysteine. This is a class of its own: a molecule that bonds to a residue only the mutant has. Its limit is in its definition. Other mutations at position 12, such as aspartate or valine, leave no sulphur there for the chemistry to find.

Outside, inside, beside: the sorting, and what the network does next

You have three classes so far. Two more complete the sorting.

• Outside. An antibody is a large protein that does not cross the membrane, so it works on the outer face of a receptor. Trastuzumab (1998) and cetuximab (2004), from the lesson on receptor tyrosine kinases, are examples. The checkpoint antibodies in the lesson on antigen receptors land outside too.
• Beside. Some small molecules bind next to the active site and hold the enzyme in a shape that cannot work. The MEK inhibitors in the lesson on Ras and MAPK sit in a pocket beside the ATP site whose shape is particular to MEK.

The part need not be an enzyme. Venetoclax, from the lesson on cell death, fills the groove on BCL-2 where a partner protein would bind.

Wherever a drug lands, it lands in a network, and the lesson on signalling circuits tells you what to expect. ERK normally sends negative feedback to steps above it. Block MEK and that feedback is released, so the upstream steps push harder and the pathway can partly switch itself back on. A bypass is the other route. In some lung cancers treated with an EGF receptor inhibitor, extra copies of the gene for a different receptor, MET, switch the same downstream pathways on again.

For the K-Ras G12C drugs the feedback has a particular bite. The drug can bond only to Ras in its GDP-bound shape, and in cell experiments, signals from receptors that load Ras with GTP leave less of the protein in the shape the drug needs.