Learning mitigates genetic drift Scientific Reports

Genetic drift describes random fluctuations in allelic diversity in a population. In this article, we have reasoned that a systematic increase of organisms’ chances of surviving in response to a random source of mortality, eg, learning from experience, mitigates genetic drift (decrease in the number of alleles) even when overall mortality rates remain the Same. To test this hypothesis, we developed an agent-based simulation model. The results of our simulations show that even though the same number of prey die per time step in both the learning and non-learning scenarios, their allele numbers decrease at quite different rates. This effect is explained by the tendency of learning to create a pool of older experienced individuals who have a very small chance of dying from predation and thus create a pool of alleles that are unlikely to be lost. In addition to confirming our hypothesis in general, our results also demonstrate that this effect applies regardless of whether the affected organism is haploid and reproduces asexually or diploid and reproduces sexually.

Our results suggest that learning from experience may be an overlooked factor affecting effective population size. Contrary to most other factors such as an unequal number of males and females4,9,10the ability to learn from the experience possibly increases the effective population size.

The purpose of this article is to point out the existence of a complex relationship between various mechanisms improving organisms’ ability to survive and the strength of genetic drift. The results presented here cannot, obviously, be expected to tell us anything about how strong this effect is in nature. With regard to learning from experience, the crucial factor for determining this would be assessing how effectively animals can learn to avoid or escape a deadly danger. For simulations in this article, we operated with an efficiency of 90% or more. Arguably, this probability may be too high in real-world settings. However, under certain circumstances, it is plausible. For example, a study on barn owl (Tyto alba) has shown that these owls have a 90% chance of catching a stationary food item and only a 21% chance of catching moving items11. Furthermore, the owls were even less likely to catch an item moving towards them (18%), and most interestingly, if the food items were moving sideways the barn owls were not able to catch them at all. Therefore, it seems feasible that prey may learn simple strategies that would allow them to escape at least some types of predators with surprisingly high probability. Nevertheless, even precise information about how effectively animals can learn to avoid or escape different sources of mortality may not be enough to predict the effect of learning on the genetic drift in a real-world population because the ability to learn is by itself an adaptive trait and including selection into the model would further complicate interpreting its results. One way or another, our message remains the same; predicting the strength of genetic drift is more complex than conventionally assumed.

At first glance, it may seem that the effect of learning presented in this article is caused merely by the fact that learning increases generational interval (the average age of parents when their offspring are born), which is known to result in increased effective population size and thus smaller genetic drift10. However, as we have shown, while learning increases the generation interval in most simulation settings, there are situations where it decreases the loss in the number of alleles without prolonging the generation interval. Therefore, the lengthening of generation intervals likely explains part but not the entire effect of learning. However, this is only because the generation interval is an average metric that cannot reflect changes resulting from changes in age distribution that will not change average values. This can occur in some settings with learning scenarios as learning not only creates a pool of older hard-to-kill individuals but also decreases the number of individuals surviving up to some age (specific simulation); see Fig. 1c, d.

We exemplified our idea of ‚Äč‚Äčlearning mitigating genetic drift by considering the effect of learning in the context of predation. However, for the same reasons, learning exerts the same influence on genetic drift caused by any source of mortality, which individuals of a given species may learn to avoid. For example, if individuals of a given species were able to learn to avoid traffic from their experience, the effect on genetic drift would be identical to the learned ability to avoid or escape predators. To go even further, any learned improvement in the ability to find a partner for mating, to find food and thus decrease the risk of starvation, or to care for offspring and thus increase their chance of reaching adulthood, will, to some extent, help preserve genetic diversity despite the raw effects of drift. Accordingly, there may be other types of reactions, such as immunological memory, which could help preserve genetic diversity by mitigating genetic drift through the same principle.

It is important to note that our model assumes overlapping generations, and thus all our results apply only to such a situation. With nonoverlapping generations, the benefits of nonhereditary learning are likely to be nonexistent or much smaller as there is much less time for them to have any effect (provided that learning is immediate). In our scenarios with overlapping generations, individuals that got lucky and learned to evade predators could effectively reproduce several times before their death. With nonoverlapping generations, they would be able to reproduce only once regardless of how well they can escape predators which would undoubtedly reduce the benefits of learning.

The here-described relationship between learning and genetic drift also leads to some testable predictions. First, species that can learn more effectively should, on average, better conserve genetic diversity than species of less capable learners living in populations of similar size. The second and connected prediction is that species that are “better learners” should have a lower risk of extinction. This prediction is in good agreement with a recent study showing that bird species with greater behavioral plasticity indeed have a lower risk of extinction12, even though there are also likely other reasons responsible for this connection, such as the likely ability of behavioral plasticity to reduce mortality rates. The third prediction. In regard to our findings, there are two types of excess mortality, one which individuals of a given species may learn to avoid, and second that members of that species cannot learn to avoid or escape. Our findings suggest that the first type of excess mortality should reduce genetic diversity to a smaller degree than the second type, even if they cause the same increase in mortality rates. The analogous predictions can also be made with regard to immunological memory or any process reducing mortality by a systematic reactive mechanism.

Overall, our results show that the conventional view considering genetic drift as independent of underlying species behavior is incomplete and that genetic drift may be affected by common processes such as learning or immunological memory even if overall population-level mortality rates remain the same. Furthermore, we have shown that the level of protection against genetic drift varies in different situations, suggesting that loss of genetic diversity by genetic drift is a more complex issue than previously thought. We hope that these findings will make existing models of evolution more precise and could prove useful in a variety of topics, including the development of effective species conservation strategies, studies of the evolutionary past as well as the evolutionary future.