Probabilistic modeling of the evolution of gene synteny within reconciled phylogenies

Abstract

Background: Most models of genome evolution concern either genetic sequences, gene content or gene order. They sometimes integrate two of the three levels, but rarely the three of them. Probabilistic models of gene order evolution usually have to assume constant gene content or adopt a presence/absence coding of gene neighborhoods which is blind to complex events modifying gene content.

Results: We propose a probabilistic evolutionary model for gene neighborhoods, allowing genes to be inserted, duplicated or lost. It uses reconciled phylogenies, which integrate sequence and gene content evolution. We are then able to optimize parameters such as phylogeny branch lengths, or probabilistic laws depicting the diversity of susceptibility of syntenic regions to rearrangements. We reconstruct a structure for ancestral genomes by optimizing a likelihood, keeping track of all evolutionary events at the level of gene content and gene synteny. Ancestral syntenies are associated with a probability of presence.

Figure 1

Principle of the method. Given two gene trees (dark blue tree and light blue tree) reconciled within a species tree (black tree), and sharing adjacencies in some extant species (species A and C), we reconstruct hypothetical ancestral adjacencies (in species D and E) using a model of evolution and maximum likelihood algorithm. Our method allows for losses (cross in light blue tree between species B and species D), and duplications (empty square in dark blue tree) of genes.

Figure 2

The tree of possible adjacencies. A tree of possible adjacencies is constructed from two reconciled gene trees (top trees). The nodes of the tree are annotated as speciation nodes (grey nodes), duplication nodes (white squares), or losses (crossed grey nodes). A binary state is attributed to each leaf according to the presence/absence pattern of the adjacency in extant species (light blue square). The true evolutionary history of the adjacency is represented on the blue tree. An adjacency exists between genes a₁ and a₂, c₁ and c₂, d₁ and d₂. It is absent between genes a₁ and $a_2^{\prime }$ and genes $c_1^{\prime }$ and c₂. If one extremity of the adjacency is lost (species B), the adjacency node is given an undefined state "?".

Figure 3

Model of evolution with duplications. A duplication in the same species in each of the two gene trees leads to a duplication node with four children (white square) in the tree of possible adjacencies between the two gene trees. Immediately after a duplication event, the adjacency is broken for the duplicated branch (branch leading to the rightmost red node). The second duplication leads to the simultaneous apparition of two other branches (leading to left and middle red nodes). The adjacency is also broken at the beginning of these two branches. The probabilities of transition between the duplication node and its four children are then given by the (., 0, 0, 0) → (., ., ., .) components of P(t₁) ⊗ N¹¹(t₁). In the likelihood computation, all positions for the blue and red nodes are considered.

Figure 4

Drosophila phylogeny. The 12 Drosophila species tree with branch lengths optimized according to the model and the synteny data.

Figure 5

Ancestral adjacencies reconstruction with our method and with DeCo. The posterior probabilities of presence of ancestral adjacencies reconstructed with our model (in grey). In red the part reconstructed by a parsimony method.