Kumawat, Lalejini, Acosta, and Zaman (PNAS, 2025) evolved self-replicating digital organisms in changing environments and found that the populations developed two distinct strategies for coping with change — simultaneously, not sequentially.
The first strategy is positional. Populations relocate in genotype space to regions where mutations are more likely to produce beneficial phenotypes. They sit on “phenotypic boundaries” — points in the genetic landscape where small mutations step into useful territory rather than neutral or harmful territory. This strategy works for environments the population has encountered before. The landscape around the organism has been shaped by prior selection so that the most probable mutations point in useful directions. The system doesn't need to search widely because it's already positioned at a good starting point.
The second strategy is exploratory. Populations evolve higher mutation rates, which increase the volume of genetic space sampled per generation. This strategy works for environments the population has never encountered. No amount of good positioning helps when the required adaptation lies in a region of genotype space that has never been explored. Higher mutation rates increase the probability of reaching that region, at the cost of producing more deleterious mutations along the way.
The structural observation: these are different answers to different questions. The first strategy answers “how do I readapt to something I've seen before?” — by remembering the right neighborhood. The second answers “how do I adapt to something I've never seen?” — by searching more broadly. The two strategies are orthogonal. A population can be well-positioned (low exploration cost for recurring environments) while also having a high mutation rate (wide search for novel environments). Evolution produces both simultaneously because the problems they solve are independent.
Evolvability is usually discussed as a single property — how quickly a population can adapt. This study shows it decomposes into at least two independent components: memory (landscape positioning that encodes prior solutions) and reach (mutation rate that determines how far the next generation can land). High memory, low reach: fast readaptation, no novelty. High reach, low memory: wide search, no direction. The evolved populations have both. Neither component alone explains their adaptive speed.