Question 11 of 20 Intermediate Interpretation
Why do IC50 values depend on how fast the cells divide, and what fixes it?
Short answer
A conventional IC50 compares treated cells with an untreated control at one endpoint, and the control grows more in a fast-dividing line than a slow one. Faster lines therefore look more sensitive to the same drug. Growth rate inhibition metrics, such as GR50, normalize to the number of divisions and remove the artifact.
Why
Hafner and colleagues showed that IC50 and Emax depend strongly on how many divisions happen during the assay. In their simulations a line dividing every 1.8 days and one dividing every 3.9 days gave different IC50 values for the same cytostatic drug, and in the slow line the total cell number barely doubled in a three day assay, so no IC50 could be defined. This creates false correlations between genotype and sensitivity, because growth rate itself correlates with genotype, and it obscures real biology in biomarker discovery.
The GR approach compares the growth rate with and without drug rather than the endpoint count. GR values are largely independent of division rate and assay length. It needs a cell count or an equivalent measurement at the start of treatment as well as at the end, which is a modest change to the protocol.
GR metrics do not repair everything. They assume exponential growth, which breaks when cells reach confluence or in three-dimensional culture, and they still depend on the readout chosen. Reporting both the metric and the assay duration lets another lab interpret the number.
A strong answer covers
- Endpoint ratios depend on control growth during the assay
- Fast lines appear more sensitive artifactually
- False genotype to sensitivity correlations
- GR metrics normalize to growth rate
- A time-zero measurement is required
Common mistakes
- Comparing IC50 values across cell lines with very different doubling times
- Omitting the time-zero count so GR metrics cannot be computed
- Applying exponential-growth metrics to cultures that reached confluence
The interviewer might ask next
What would you measure at time zero to compute GR metrics in a 384-well screen?