Published in the Proceedings of the National Academy of Sciences (PNAS), a groundbreaking study led by researchers from the University of Vermont and the University of Cambridge has used computer simulations to replay evolution hundreds of times. This research sheds light on how fluctuating environmental conditions can shape or derail the long-term success of populations, challenging conventional assumptions about evolutionary predictability. The findings could have far-reaching implications, from climate adaptation strategies to advancements in artificial intelligence.
Simulating Evolution Over Thousands of Generations
Utilizing digital organisms exposed to 105 distinct environmental scenarios, the research team conducted large-scale simulations that tracked evolutionary outcomes across thousands of generations. These scenarios mirrored real-world environmental fluctuations, such as temperature swings and alternating drought-rainfall cycles. The key revelation was the remarkable diversity in how populations responded to these changes.
“We found remarkable variation in how populations evolved in variable environments,” the researchers reported. “In some cases, changing the environment helped populations find higher fitness peaks; in others, it hindered them.”
This result underscores that the nature of environmental change, not just its frequency, can profoundly impact evolutionary outcomes. Certain types of variability acted as a training ground, nudging populations toward higher fitness levels. However, in many instances, unstable conditions created evolutionary dead ends, forcing repeated restarts that impaired progress.
A Wider Lens on Evolutionary Research
Traditionally, evolutionary studies have focused on a single population in a fixed or simplified environment, limiting the scope of conclusions. This study, however, adopted a panoramic view.
“Researchers often watch the long-term trajectory of one population in a specific environment,” says Lapo Frati, a computer scientist at UVM. “We picked an array of environments and see how the specifics of each one influence the trajectory of many populations.”
This shift in approach was central to uncovering what the authors call evolutionary divergence by condition. The results imply that no single population can represent an entire species, a critical insight as scientists attempt to forecast how organisms will cope with climate change, disease pressures, or habitat disruption.
The researchers emphasize that real-world species are exposed to different environmental pressures, often within the same species. For instance, two populations of fruit flies might both belong to Drosophila melanogaster, but while one group in the U.S. contends with winter-summer swings, another in Kenya cycles through drought and heavy rain.
When Variability Helps: And When It Hurts
One of the most eye-opening discoveries in the PNAS paper was how certain types of environmental fluctuation promote adaptation, while others impede it.
“Temperature fluctuations might promote better adaptation to both cold and warm seasons,” Csenge Petak explains. “But repeated cycling between dry and wet seasons might actually impede adaptation to drought, forcing the population to ‘restart’ evolution after they experience a long period of rainfall — leading to worse traits than in populations exposed only to drought.”
This insight reveals that evolution doesn’t always reward flexibility. The sequence and timing of stressors can set back evolutionary progress, even in populations capable of adapting in principle. Environments that flip-flop in certain ways may trap organisms in low-fitness cycles, never allowing enough time to consolidate advantageous traits before the next shift hits.
Lessons Beyond Biology: AI and Meta-Learning
The implications of this work extend well beyond biological systems. Co-author Nick Cheney, also a UVM computer scientist, sees parallels in the world of artificial intelligence, particularly in continual learning systems that must evolve new skills without losing old ones.
“My research is about meta-learning — the capability of systems to learn to learn,” adds Frati.
Much like biological populations in a volatile world, AI systems often struggle to retain previous learning when faced with new challenges. The researchers suggest that understanding how evolution copes with changing environments could help design more resilient AI architectures.
Just as no single test environment suffices to evaluate evolutionary strength, a one-task AI cannot prove its intelligence. Resilience, whether in evolution or machine learning, depends on the ability to adapt, and to do so without forgetting past lessons.
Rethinking Evolution as a Process of History
One of the study’s senior authors, Melissa Pespeni, emphasizes that the ability to replay evolution hundreds of times offered an unprecedented level of insight.
“What’s exciting about this study is that we replayed evolution hundreds of times. This gave us a bird’s-eye view of how evolution played out across many different environments, something that would be impossible to test in the lab,” she said.
These repetitions revealed that starting conditions matter enormously. A population’s history, what challenges it faced first, how fast conditions changed, and what traits were useful early on, sets the course for its long-term potential. In this view, evolution is not just shaped by selection, but by historical momentum.
The study challenges the traditional view of evolution as a linear, predictable process, highlighting instead its sensitivity to historical context and environmental variability. As scientists continue to explore the complexities of evolution, this research provides a critical framework for understanding the intricate dance between organisms and their ever-changing environments.
