Stanford scientists watched African killifish for their entire lives; despite similar genes and conditions, early swimming and sleep patterns predicted lifespan

Two African turquoise killifish of the same age, raised under a carefully controlled environment, could take different developmental trajectories into their maturity. One may grow up to be active, swim vigorously and concentrate most of its sleep at night. Meanwhile, the other might gradually become less active and spend more time resting during daylight hours. Stanford University researchers found that these behavioural differences could emerge very early in adulthood and help predict whether an individual fish would have a shorter or longer lifespan. The scientists also discovered, while continuously monitoring the animals from adolescence until death, that ageing involved distinct periods of behavioural stability interrupted by relatively abrupt transitions.A camera system followed fish from adolescence to deathThe researchers picked the African turquoise killifish, a small vertebrate that usually lives for just four to eight months in laboratory conditions. The naturally compressed lifespan of the species makes it possible to study changes associated with ageing over a shorter time period than what would be required for longer-lived animals. Despite having a short lifespan, the fish has a complex brain and various biological characteristics that make it useful for testing vertebrate ageing.The findings were published in Science in 2026, and the experiment was led by Claire Bedbrook and Ravi Nath. The Stanford researchers developed an automated system that housed individual fish in separate tanks and documented their behaviour constantly. After reaching maturity, the animals were placed in the system at approximately three to four weeks of age. The overhead cameras set up for the experiment recorded their movements at 20 frames per second, thus allowing the team to follow each fish throughout its adult life. A cohort of 81 fish were included for the principal whole-lifespan behavioural analysis and were recorded for an average of 220 days. Their observed lifespans ranged from 23 to 371 days.In addition, computer vision and machine-learning techniques to analyse the recordings were used by the researchers. They identified 100 recurring behavioural patterns, which they called “behavioural syllables.” Therefore, by organising the documented information into daily behavioural patterns, the crew could compare how individual fish evolved and identify differences between animals that lived shorter or longer lives.Swimming and sleep patterns showed early hintsOne of the most significant findings included the behavioural trajectories of longer-lived and shorter-lived fish. The behaviour started diverging while the animals were still relatively young. The team noted that prediction models performed poorly at ages below 70 days but improved after that point. By around 70-110 days, behavioural differences were sufficiently visible for machine-learning models to distinguish between fish that would eventually fall into shorter-or longer-lifespan groups.Among other differences, sleep and rest patterns were the clearest of them. Fish that ultimately lived shorter lives increasingly spent more time in sleep-like or inactive states during daylight hours. Meanwhile, those that tended to concentrate these states at night showed a clearly defined daily rhythm. Additionally, the researchers noted differences in movement. Fish with longer lifespans usually displayed more vigorous daytime activity and reached higher levels of swimming speed when moving in the tanks.These daily behavioural measurements were used by the team to place fish into shorter- or longer-lived categories. The shorter-lived group consisted of fish that died before 200 days but lived at least 120 days. This excluded those that died prematurely. For fish that survived past the early baseline, the models achieved useful predictive performance using behavioural data collected at 70, 90 and 110 days. Instead of calculating the exact death date for each fish, the researchers estimated which lifespan category an individual was more likely to belong to.Ageing unfolded through distinct behavioural stagesThe continuous recordings also contested the idea that ageing necessarily appears as a smooth, gradual decline. When the researchers analysed each fish’s behaviour across its lifespan, they found periods when daily activity patterns remained relatively stable, followed by abrupt transitions into different behavioural states. Many of the animals experienced between two and six such transitions. These stages were identified through statistical modelling of behavioural patterns rather than being established as universal biological stages that every fish must experience in exactly same way.The team also tested gene expression across eight organs to see whether the behavioural differences were reflected in the body’s molecular activity. The liver in particular showed pronounced differences between fish following shorter- and longer-lifespan trajectories. They identified changes in groups of genes linked with cellular processes, including production of protein and maintenance. While these findings drew a connection between observable behaviour and organ-specific molecular differences, they did not demonstrate that the liver change directly caused the different lifespan.