Sim-Bdelloid

Lateral gene transfer as a form of ultrastability

Photo by Damián H. Zanetter, released in the public domain by the author. From Wikipedia.








The Question

Bdelloid rottifers are small animals with an interesting life cycle. They reproduce exclusively asexually, by parthenogenesis; but differently from other parthonegetic species, they have enjoyed a respectable amount of evolutionary success for 35+ millions of years, adapting to different environments and diversifying into more than 450 species.

How do they do that? One theory is that it is because their life cycle permits a considerable amount of lateral gene transfer. In situations of environmental stress, a bdelloid will rapidly dehydrate itself and wait, for up to nine years, for better conditions. Then it will rehydrate and become active again; but in doing so, it will repair its own DNA through a process that can incorporate foreign genes from their external environment (of either bdelloid or non-bdelloid origin). In fact, a significant fraction of a bdelloid genes have fungal or bacterial origin, and have been incorporated in the bdelloid genome in this way!

How does this mechanism for genetic variation compare with standard sexual reproduction? In this experiment, we will attempt to evolve simple (ten neurons big) continuous-time neural network through sexual reproduction and lateral gene transfer. Spoiler alert: lateral gene transfer seems actually much better!

Note, however: of course this does not mean that lateral gene transfer is a better mechanism than sexual reproduction in all circumstances. Details and circumstances do matter! However, it strongly suggests that there may be some real-life scenarios in which this is the case.




Environment

Our simulated critters move around a 1000x1000 bidimensional arena containing, at each time, two sources, a red one and a blue one. These two sources emit different amounts of energy, that our animals absorb; and, to survive, our simulated animals must maintain their internal energy within a certain range. Too little, and they die of starvation; too much, and they die of overeating. Our animals have no access to their internal energy level, but can distinguish between red and blue light. After a certain amount of time, the sources of light disappear and reappear somewhere else in the fields, so they better be ready to chase them around!


In the above image, for example, the red light emits double the energy of the blue one; so the animals have learned to stay farther away from it. Now suppose that, after a while and after the animals evolved to survive in this environment, the amounts of energy emitted by the two lights are changed. Will they be able to adapt to the new circumstances?




The experiments

In one set of experiments, our animals evolved through lateral gene transfer: whenever an animal is close to death, some of its genes are randomly overwritten by those of some completely random other one (and, in a break from biological plausibility, its neural network follows suit immediately). This is the only mechanism of genetic transmission; otherwise, they reproduce asexually.

In another set of experiments, instead, there was sexual reproduction — two of the more successful animals are selected and used to generate a successor — and in yet another set of experiments, neither sexual reproduction nor lateral gene transfer were permitted, but only random genetic variation.




The results

Under lateral gene transfer conditions, the animals were able to evolve to adapt to the environmental change very well. Without sexual reproduction or lateral gene transfer, instead, the animals were able to evolve to adapt to the environment, but had had difficulty adapting to the environmental change. Finally, and somewhat surprisingly, the case of sexual reproduction was the worst of them all, and developed a purely stationary (and less successful) strategy.

Interestingly, if we switch off the possibility of death by overeating the animals can learn, in the sexual reproduction setting, to chase after the lights; however, in this setting, they were not able to evolve the ability to stay st a safe distance from the sources. It would seem that this delicate homeostatic behaviour is not well-preserved by our crossover operation: even if the two parents are good at staying at a safe distance from the light sources, their offspring (whose neural weights come from either parent) will not necessarily be likely to be so.

On the other hand, under lateral gene transfer, successful individuals reproduce essentially asexually while unsuccessful ones absorb genetic information from random (and hopefully more successful) ones. Thus, successful homeostatic adaptations are not lost!

See the following videos for a few examples of our simulation's behaviour:




Videos



Asexual reproduction, before and after env. change:


Horiz. gene transfer, before and after env. change:


Sexual reproduction, before and after env. change:

You can see the videos of all other experiments here (for a description of what they are, see the pdf file).




Details and code

For more information about the simulation, you can look at the pdf file.

You can also download the code from github and play around with it.