The word "acquired" is misleading. Most resistance that appears during treatment was already present in the tumor before the first dose; what treatment acquires is not the resistance but the dominance of the resistant cells. A tumor of a billion cells carries roughly thousands of point mutations and numerous copy-number changes, because the same genomic instability that made it a cancer keeps generating variation. A fraction of those variants happen to disable a drug's target, upregulate a pump, or break the apoptosis pathway. At diagnosis those subclones are a small minority, invisible to imaging and irrelevant to the tumor's overall behavior.
Cytotoxic chemotherapy applies a strong selective pressure. A drug that kills ninety-nine percent of the tumor cells removes almost all of the sensitive majority and leaves the pre-existing resistant minority intact. Freed from competition for space, oxygen, and nutrients, that minority expands. The tumor that regrows after cycle four is therefore not the tumor that was treated in cycle one — it is a descendant population enriched for the very mechanisms that survived. This is Darwinian selection operating on a timescale of weeks, and it explains a pattern clinicians see routinely: a tumor that responded well initially and then stopped responding, with no new external cause.
Two features of the process matter for understanding selectivity. First, the resistant subclone is still a cancer cell and still carries the same normal-tissue-independent vulnerabilities — but the drug no longer reaches them, because the mechanism that protects it also protects it from the drug's lethal action. Second, resistance is usually polyclonal and multifocal. Different subclones can acquire different mechanisms in different parts of the tumor, so a single biopsy may not represent the resistance profile of the whole mass, and a drug regimen that overcomes one mechanism may leave another untouched.
There is a second route that is genuinely acquired: a sensitive cell can acquire a new heritable change during treatment, for example a mutation in the target gene selected for by the drug itself, or amplification of a gene whose product is inhibited by the drug. This is the same process that drives resistance to targeted kinase inhibitors, where a secondary mutation in the kinase domain is the classic event. But even here the logic is selective rather than instructive — the drug does not teach the cell to mutate; it kills the cells that did not happen to mutate.
The consequence for the therapeutic index is direct. At diagnosis, the tumor's dose–response curve sits to the left of the normal-tissue curve, and the gap between them is the selectivity window. As resistant subclones expand, the tumor curve shifts right — the same dose now kills fewer cells — while the normal-tissue curve is unchanged, because normal tissues did not undergo selection. The window narrows from the tumor side. If the shift is large enough, the tumor curve crosses the normal-tissue curve, and no dose can kill the tumor without unacceptable toxicity. That crossing point is the biological definition of clinical resistance, and it is why resistance is not a separate topic from selectivity but the same problem viewed from the other end.