This page outlines my understanding (so far) of the mechanisms and concepts involved in symbiosis and its role in evolutionary adaptation. Although these ideas provide the motivation for my thesis work, my published work to date is necessarily more conservative. The final section below discusses how my papers relate to the ideas I discuss.

 The facts from nature
 Units of selection
 Fuzzy units-of-selection
 Research directions
 References

The facts from nature

In its strongest form, symbiosis can lead to symbiogenesis: the genesis of new species via the genetic integration of symbionts [Merezhkovsky 1909]:

"...different bacteria form consortia that, under ecological pressures, associate and undergo metabolic and genetic change such that their tightly integrated communities result in individuality at a more complex level of organization." [Margulis, 1995].

"entities that were capable of independent replication before the transition can replicate only as part of a larger whole after it." [Maynard-Smith & Szathmary, 1995].

Units of selection

Although this point of view may seem natural and straight forward it opens up a whole can of worms. The whole "group selection" debate has been a messy one - with very strong opinions on both sides. As I understand it, it used to be fashionable to talk about selection at whatever level makes a good story: sometimes it might be the individual level, sometimes it might be the family level, often it might be the species level, or anything upto the kingdom - plants vs animals. Quite rightly, this was criticised harshly. It transpired that nearly all of the adaptations observed in nature that were previously explained using higher levels of selection - 'for the good of the species' - could be explained using individual selection if you cared to look a little more closely.

Subsequently, 'group selection' has become a tabu subject, and the field has been left, for the most part, with the position that 'group selection does not exist; selection at the individual level is the only mechanism in evolution and is sufficient to explain all adaptations'. Some go further, Richard Dawkins, for example, will tell you that the correct unit-of-selection is the gene and that selection at this level is sufficient to explain everything. But thats another story.

In light of this background from the group selection debate it is very hard to even discuss selection at any higher level these days. Any mention of adaptive mutualism is often interpreted as ignorant sentimentalism. The prefered explanation for observed mutualism is either that it is merely an abberation (on the otherwise relentless path of mutually exclusive competition), or if treated seriously at all, it is 'explained away' by reducing the interactions involved to the individual level and showing that each individual is merely acting in its own self-interest. Well duh! What did you expect? Perhaps you expected bacteria, for example, that were acting for the good of the group out of the kindness of their hearts?

Are they doing it out of the kindness of their hearts? - No. Are they doing it? - Yes.

The prevailing position is that it is selfish individualism (hence, The Selfish Gene) that drives everything. But this language is just as emotive as the group selection descriptions that have been rejected - 'for the good of the species'. When we take a stance of either 'nature is nice and everything wants to work together', or 'nature is a cut-throat nasty place (and you might not like that, but thats the way it is)', we are resorting to sentamentalism. The intentions and morals of the entities involved, is neither here nor there. What matters is the adaptive mechanisms. Its nothing to do with selfishness vs altruism - even competition and cooperation might not be very useful terms. "Mutualism" - the living together of entities that gain mutual benefit from the association - has nothing to say about selfishness or altruism. The entities involved are just doing what they do. If only we could get past the emotive language we might be able to talk about higher-level selection without flipping-out.

So where are the tools with which we can begin to talk about mutualism and its role in evolutionary adaptation?

Fuzzy units-of-selection

• Firstly, we need to talk about systems rather than genes, individuals, species etc. A system might be one of these things, but it might also be a sub-group of different species, an ecosystem, a 'network' of species that percolates through an ecosystem, or whatever. I will consider all of these as candidates.
• Secondly, we need to recognise that a system may be coupled to other systems to a greater or lesser degree. For example, individual organisms are generally only loosely coupled to other organisms, whereas genes are relatively tightly coupled to other genes in the same organism.  To the extent that a system is tightly coupled internally and only loosely coupled to other systems, we are likely to think of it as an entity in its own right. But, in general, systems have fuzzy boundaries. We will use the term "coherence" to describe systems that are strongly coupled internally and weakly coupled externally.
• Thirdly, we need to understand that the mechanisms of heritable variation and differential reproduction may apply at many levels, systems and sub-systems, simultaneously. The extent to which it is useful to explain the adaptation of a system under these mechanisms is dependent on both the extent to which the characteristics of the system are heritable, and the extent to which the system can be differentiated from other systems. That is, selection acts on an entity to the extent that that entity is coherent.

However, note that neither of these levels alone provides a full explanation. The complete set of characteristics that describes an individual is not perfectly heritable under sexual reproduction because subsets of genes are crossed-over. And, conversely, one gene is not completely differentiable from another - they are often tightly coupled within an individual. Put simply, genes are more likely to reproduce together if they come from the same individual than if they come from different individuals. Niether the gene level nor the individual level tells the whole story. On the one hand, as Dawkins will agree, you cant tell a complete story about the adaptation of individuals without mentioning genes. But on the other hand, you cant tell a complete story about the adaptation of genes without mentioning individuals. Neither level of description is 'the right level'.

So, there are many levels being acted on by selection, and all these levels are (potentially) valuable for describing the system. The usefulness of a level (the extent to which a level influences adaptation) is dependent on the coherence of components at that level. The interesting thing about the major transitions in natural evolution is that, one way or another, the coherence of the system at a higher level became dramatically increased. For example, when the DNA of one bacteria gets 'under the skin' of another bacteria (as in the formation of the eukaryotes) the coherence of the pair is dramatically increased. In other cases, the increase in coherence may be more subtle. For example, it may involve the behavioural maintenance of the group - that is, adaptation at the individual level produces behaviours that maintain the coherence of the group. Having been formed, the group is now a unit of selection and adaptation occurs at the group level.

Note that the individual level is still present, to some degree - adaptation at the individual level may still occur. But, to the extent that the coherence of the group is maintained, adaptation at the individual level is subsumed by adaptation at the group level. If we assume that variations are generaly reversible then, in some cases, an individual may arise that does not maintain the coherence of the group, and perhaps it may gain subsequent advantage. However, it is also possible to imagine that a group, by chance, may get itself into a part of variation space such that all individual variations that break the coherence of the group are more detrimental than advantageous. If this is the case then the coherence of the group will be further supported and 'solidified'. Steven Frank talks about "thresholds" in "mutual policing" - and I think this is where we are headed.

The fact of the matter is that mutualism is, at least sometimes, adaptive. Moreover, it sometimes has significant impact on the course of evolution, as cited above. My argument has been that this is because selection may have an effect on many levels (not just the individual or the gene). The strength of its influence on a group is related to the coherence of that group. If the coherence of the group becomes self-perpetuating then selection of at the group level will subsume selection at the individual level. In fact, as with the eukaryotes, it is no longer useful to talk about the original individuals - as Margulis says,  we have "individuality at a more complex level". Such a transition may, on rare occasions, bring a major transition in evolutionary history. But just as interesting, is the notion that more subtle groups may be having more subtle influence not just on rare occasions but in a widespread, even pandemic, sense.  Thus mutualism, far from being an insignificant curio, may be an important key mechanism in evolutionary adaptation.

Research directions

• Under what circumstances will symbiosis be influential in shaping evolution?, and
• How are higher-level units-of-selection formed and maintained in natural evolution?

My hunch is that the formation of mutually benefitial groups will be adaptively useful when the environment presents a problem that is 'partially decomposable'. That is, a problem with identifiable sub-parts, but where these sub-problems are not completely independent. In this case, the coupled structure of a problem will be reflected in the structure of coupling in the (eco)system that adapts to it. I have spent more than half of my research effort on trying to formalise what kind of coupling I might be talking about. The result of this is an abstract building-block problem which I call HIFF (hierarchical if-and-only-if) (Modeling Building-Block Interdependency), (Hierarchically-Consistent GA Test Problems), (C code for the HIFF test function). This problem has a hierarchical structure of coupled building-blocks. It makes a good test problem for regular GAs but, more importantly for me, there are variants of this problem that cannot be solved by the regular GA, but may, I suggest, be solved by a form of symbiotic algorithm.

Now I have begun to make modifications to the GA to address its inadequacies with respect to this class of problems. (Incremental Commitment in Genetic Algorithms) The important characterisation of this algorithm is that mating between two parents produces the combination of building-blocks they contain, as distinct from a 'bit of this one, bit of that one' that you get in normal recombination. This paper shows how individuals can be used to represent sub-solutions explicitly such that their combination provides whole-solutions.

In the meantime, I've done some (almost unrelated) excursions into how the formation of mutualist groups can have a shaping influence on evolution. My first paper illustrated how the formation of mutualist groups can guide genetic variation so as to enable the evolution of ultimately independent organisms that would otherwise be unobtainable.(How Symbiosis Can Guide Evolution), My second paper, with Torsten Reil, shows that this effect applies not just in genetically related organisms but may also occur from symbiosis between distinct species. (Mutualism, Parasitism and Evolutionary Adaptation) Also available in html.

The next, and critical, step in the research is to bring the group evaluation used in the symbiosis work of the last two papers together with the partial specification (incremental commitment) demonstrated in the GA work. I just figured out how to do this.  This will demonstrate a problem solving method utilising the evolution of groups and collective problem solving rather than individual problem solving. In addition, it will demonstrate the encapsulation of these groups into new individuals so that the process can recurse, making groups at increasing levels of complexity.

However, even when this works, it will be quite some way from a model of symbiosis in natural evolution. It wont describe any behavioural methods for group formation or self-sustaining groups, for example, since the required mechanisms are primitive operations in this algorithm. This reflects my computer science background, and, in fact, I'm not sure how to go about transforming the algorithm into a model that biologists might appreciate. However, in the meantime, it will provide an illustration of how hierarchical selection can be an effective adaptive mechanism - and this is likely to have something to say about the conditions under which symbiosis will be influential in natural evolution.

References

Margulis L, 1992, Symbiosis in Cell Evolution: Microbial Communities in the Archean and Ptroterozoic Eons. 2d ed., W.H. Freeman, New York.

Merezhkovsky KS, 1909 "The Theory of Two Plasms as the Basis of Symbiogenesis, a New Study or the Origins of Organisms," Proceedings of the Studies of the Imperial Kazan University, Publishing Office of the Imperial University. (In Russian, see Khakhina 1992).