The Pill That Could · The Generalization Problem

The Generalization Problem, part 3

What the adjuvant Herceptin story shows, in the larger frame of this chapter, is that targeted therapy's biggest contribution in some cancers has come not in the headline-grabbing metastatic setting but in the quieter, longer-tail setting of preventing recurrence. The patient who never has a metastatic recurrence does not show up in dramatic before-and-after photographs. She shows up in the survival curves, decades after her surgery, alive and well, having taken a year of an antibody she would not have received in 1998. The adjuvant Herceptin trials saved more lives, in absolute terms, than any individual development in CML did. They are also, partly because there is no single dramatic moment of cure, less famous than they should be.

There are several other cancers where adjuvant or maintenance targeted therapy has produced effects of a similar shape — not the dramatic shrinkage of metastatic disease that makes for a good newspaper article, but a steady reduction in the rate at which a cancer comes back, integrated over years and across populations. PARP inhibitors as maintenance therapy in BRCA-mutant ovarian cancer have done something analogous. Adjuvant osimertinib in EGFR-mutant lung cancer, established by the ADAURA trial in 2020, is now standard. The pattern is consistent: where targeted therapy can be given to patients before the cancer has had a chance to advance, the long-tail benefits compound.

* * *

Now the cancers where the paradigm has not worked as well.

Most solid tumors are not driven by a single, dominant, druggable kinase or by any single, simple, druggable event. Most solid tumors, when they are sequenced, contain dozens of mutations across dozens of genes, with substantial diversity within any one tumor between cells in different regions, and substantial diversity between different patients with the same diagnosis. This is not a small problem for the targeted approach. It is the central problem.

Pancreatic ductal adenocarcinoma, the most common form of pancreatic cancer, is the locus classicus of this difficulty. It is one of the deadliest cancers — five-year survival, as of 2026, remains in the low double digits, and most diagnoses are still effectively terminal. Roughly ninety percent of pancreatic cancers are driven, in part, by a mutation in a gene called KRAS — a small GTPase, a kind of molecular switch involved in the same general signaling pathway as many of the kinases the rest of this chapter has discussed. KRAS would seem, on the face of it, to be the perfect target. It is mutated in nearly every patient. It drives the cancer. It has been the subject of intense drug-discovery effort for forty years.

It also turned out, for most of those forty years, to be undruggable. KRAS is small. Its surface is smooth in a structural sense — there are no obvious deep pockets where a small molecule could bind tightly. The active site, where it cycles between active (GTP-bound) and inactive (GDP-bound) states, holds onto its native substrate so avidly that no competitor molecule could realistically displace it. Multiple generations of medicinal chemists, working in multiple laboratories, attempted KRAS inhibitors and found that the protein simply did not provide them with the kind of binding surface a drug needs. KRAS sat, for forty years, on the list of validated cancer targets that nobody could drug.

This changed, partially, in the late 2010s. A group at the University of California, San Francisco, led by Kevan Shokat, identified a particular subtype of KRAS mutation — KRAS-G12C, in which a glycine at position 12 is replaced by cysteine — that introduces, into the otherwise smooth protein surface, a reactive cysteine residue that a covalently-binding drug could grab onto. The covalent KRAS-G12C inhibitors that emerged from this work — sotorasib, approved 2021, and adagrasib, approved 2022 — were the first KRAS-targeted drugs ever approved. They worked, in patients carrying the specific G12C mutation, in cancers including non-small cell lung cancer.

But here is the catch. KRAS-G12C accounts for only about thirteen percent of KRAS mutations. The other eighty-seven percent — G12D, G12V, G12R, G13D, and others — do not have the same reactive cysteine and cannot be targeted the same way. In pancreatic cancer specifically, G12D and G12V are the most common KRAS mutations; G12C is rare. The drugs that exist, as of 2026, for KRAS-mutant cancer are useful for a slim fraction of patients, and the broader population of KRAS-mutant cancers — including the great majority of pancreatic cancers — remains, in a meaningful sense, undruggable. Newer compounds aimed at G12D and the broader KRAS family are in clinical development, with some early results in late 2024 and 2025 looking promising; whether they will match what G12C inhibitors deliver, or whether they will hit a wall the field has not yet fully understood, remains an open question in 2026.

Pancreatic cancer, in short, has the molecular driver and does not yet have the drug. The targeted-therapy paradigm has not yet rescued it. Whether it ever will is uncertain.

Glioblastoma, the most aggressive primary brain tumor, is another cancer that has resisted the targeted approach. Its driver landscape is heterogeneous within any single tumor — different regions of the same tumor have different mutational profiles — and the blood-brain barrier, the cellular barrier that protects the brain from circulating molecules, makes drug delivery to glioblastoma cells unusually difficult. Targeted drugs that work in other cancers often cannot reach effective concentrations in the brain. Glioblastoma's median survival, despite forty years of incremental improvement in surgery and radiation and chemotherapy, has remained in the range of fifteen to eighteen months. The targeted-therapy revolution has, so far, largely passed it by.

All 13 chapters of The Pill That Could · Library · All courses