Identifying the general principles governing when non-genetic heterogeneity is biologically relevant has been challenging. By studying correlations of cell fate across single-cell lineages and using barcoding techniques, a widely applicable principle of drug tolerance in cancer has emerged: persister cells probably originate from non-genetic cellular states that exist prior to drug treatment. An important property of these states is that their fluctuations occur on timescales slower than the cell cycle, thereby allowing the state to be inherited across multiple cellular generations, creating a ‘memory’.

However, unbiased discovery of the molecular architecture of these inheritable memory states is challenging. The two main reasons are the rarity of these states, and the destructive nature of genome-wide measurement techniques, such as RNA sequencing (RNA-seq), that make it difficult to link pre-drug cell states to their ultimate cell fates. Emert et al. overcome both these challenges with an approach termed Rewind, which combines transcribed barcodes, in situ hybridization and fluorescence-activated cell sorting to identify gene expression memory states that lead to drug tolerance in melanoma cells.