Computational protocol: Nymphalid eyespots are co-opted to novel wing locations following a similar pattern in independent lineages

Similar protocols

Protocol publication

[…] For each clade, we restricted our analyses to those pairs of eyespots that were inferred to have a shared history in the clade being investigated. A shared history was inferred when both eyespots had a higher likelihood of being present than absent in at least one shared ancestral node. Ancestral state likelihoods were calculated in Mesquite [], using an asymmetrical (2-parameter) likelihood model. For each pair of eyespots identified as having a shared history, we categorized one eyespot as “ancestral” (the eyespot that evolved first) and the other eyespot as “derived”. To objectively categorize the eyespots, we again used the likelihoods of presence/absence for each eyespot using an asymmetrical likelihood model in Mesquite []. If one eyespot had a greater likelihood of being present at a node ancestral to the “shared node” identified above, while the other eyespot had a greater likelihood of being absent at that ancestral node, the first eyespot was categorized as the ancestral eyespot. When the deepest node where each eyespot was more likely present than absent was the same for each eyespot, the eyespot with the greater likelihood of being present was categorized as the ancestral eyespot.After categorizing eyespots as ancestral or derived, we employed an iterative model comparison approach, based on estimated rates of eyespot gains. “Origination rates” for each eyespot are defined as the rate of transition from “absent” to “present” for an individual eyespot character. All analyses were calculated using the Discrete module of BayesTraits [], in which we estimated the maximum likelihood ancestral states under a variety of evolutionary models (see below for model details). In all cases, simpler nested models were rejected when more complex models had likelihood scores 2 log likelihood units higher than the simpler models. Multiple instantaneous transitions were never allowed. We first compared the likelihoods of two models: (1) an independent model of evolution, where gains and losses of one eyespot were independent of the state of the other eyespot, but in which both eyespots were constrained to the same origination rate (Figure A; the simpler model); and (2) an independent model of evolution where the origination rates of one eyespot were not constrained to equal the origination rates of the other eyespot (Figure B; the more complex model). For those pairs of eyespots in which the more complex independent unconstrained model provided a better fit, we then performed another model comparison: between it (Figure B; now the simpler model) and a dependent constrained model of evolution, where the transition rates between present and absent for the derived eyespot were conditional on the state of the ancestral eyespot and the transition rate from absent to present in the derived eyespot when the ancestral eyespot was absent was set to zero (Figure C; the more complex model). This latter model restricts the origin of derived eyespots to those lineages in which the ancestral eyespot is present, as expected if the derived eyespot is a re-deployment of the ancestral eyespot. For those pairs of eyespots in which the more complex dependent constrained model provided a better fit, we performed one additional model comparison. We compared the dependent constrained model as described above (Figure C; now the simpler model) to a dependent unconstrained model, where gains of the derived eyespot were possible in the absence of the ancestral eyespot (Figure D; the most complex model). This last comparison effectively tests if the gain of the derived eyespot in the absence of the ancestral eyespot is likely, a condition that is not congruent with a model of re-deployment, but instead supports de novo evolution of positional information for the eyespot gene network. Only those pairs of eyespots for which we failed to reject the dependent constrained model (i.e. the rate of gain of the derived eyespot in the absence of the ancestral eyespot is not significantly different from zero) were identified as pairs that fit a model of network re-deployment via modification of pre-existing positional information within the network.Figure 3 […]

Pipeline specifications

Software tools Mesquite, BayesTraits
Application Phylogenetics