Phylogénie Et Evolution Du Comportement Social Chez Les Blattes ...

An n e x e sDefining and including a character in a phylogenetic matrix implies severalassumptions, themselves depending on a homology hypothesis. In this respect, this methodis conceptually very straightforward by coding the occurrence of transitions between the actsin the sequences. It applies a better and more accurate concept of behavioural homology,each act being considered according not only to its special quality but also to its positionwithin sequences. Three main assumptions are implied and must be discussed with regard tobehavioural characters: heritability, sampling, and independence.All kind of characters used in phylogenetic analyses, either morphological, molecular,or behavioural, should be checked from this point of view. In practice, this is usually madeonly for non-traditional characters such as behavioural or physiological ones, the variability ofwhich is intuitively more questioned by scientists (Wenzel, 1992). Heritability is the first andmost basic concern since non-heritable traits would be nonsensical if used in a phylogeneticcontext of descent with modification (see Grandcolas & D’Haese, 2003 for a review). Exceptwith breeding and genetic studies, the only a priori way to assess the heritability of charactersis to control for epigenetic effects by observing every species in the same conditions and byvarying and repeating the conditions of observation. A posteriori, phylogenetic congruencebetween behaviour and other markers including those molecular ones reputed to be neutralis also a mean to assess heritability. In the present case, both criteria have been employed:repertoires and kinds of interactions have been found stable in repeated studies (Gautier,1974; Grandcolas, 1991; van Baaren et al., 2002, 2003a, 2003b) and different data sets arereasonably congruent.As for the sampling effort, behaviour is not more difficult to sample than morphologyor molecules and it only requires to have living specimens placed in controlled and relevantconditions and to follow classical protocols designed in this respect (e.g., Martin & Bateson,1986; Wenzel, 1992). In our case, accumulation curves were computed showing that thesampling of behavioural transitions has reached a plateau in every species, which allowed asound interspecific comparison. Given that samples are large enough, the successive eventpairingmethod has also the advantage to bring more characters – potentially up to squarepower more – than the acts considered in isolation.Character independence has been mentioned for a long time as a potential problemin phylogenetics but also as one without solution. The most obvious cases of dependencebetween characters must be checked for and discarded but some dependence will necessarilyexist between characters observed in a same organism which cannot be extirpated (Wiley,1981; Simmons and Freudenstein, 2002). For instance, molecular phylogenetic studiesusually do not consider that base pairs in the stem regions of ribosomal RNA are not trulyindependent. This problem does not occur with our case for the successive event-pairingmethod as shown by the χ 2 goodness-of-fit tests.Finally, this method successfully proved to be efficient and informative according tothe present example of sequences of dyadic interactions in cockroach groups taken from a291