Networks

Unlike many traditional phylogenetic trees, a graphical representation known as a network can be used to depict multifurcating, recently evolved lineages in a way that accommodates the co-existence of ancestors with descendants, and the reticulated evolution that accompanies hybridization and recombination (Table 5.2). There are several different ways to construct networks, most of which are distance methods that aim to minimize the distances (number of mutations) among haplotypes (reviewed in Posada and Crandall, 2001). Here we will limit our discussion to what has become one of the most commonly used methods in recent years, known as a statistical parsimony network.

A statistical parsimony network (Templeton, Crandall and Sing, 1992) links haplotypes to one another through a series of evolutionary steps. It is based on an algorithm that first estimates, with 95 per cent statistical confidence, the maximum number of base pair differences between haplotypes that can be attributed to a

Table 5.2 Some characteristics of bifurcating trees versus network analysis, and the relevance of these characteristics to phylogeography

Characteristic

Bifurcating trees

Network analysis

Relevance to phylogeography

Branching

Assumes all

Allows for

Population

pattern

lineages are

multifurcating

genealogies are

bifurcating

lineages

often multifurcated

Divergence

Often requires

Can reconstruct

Within species,

numerous, variable

genealogies from

sequences often show

characters

relatively little

high overall similarity

variation

Ancestral

Assumes

Allows for the

Ancestral and

haplotype

ancestral

co-existence of

descendant haplotypes

haplotypes

ancestral and

often coexist within

no longer exist

descendant

populations

haplotypes

Reticulated

Many algorithms

Networks can

At the conspecific

evolution

assume no

reveal hybridization

level, recombination

recombination

and some methods

and hybridization are

or hybridization

can allow for

often widespread

recombination

series of single mutations at each site. This number is referred to as the parsimony limit. Haplotypes differing by a number of base pairs that exceeds the parsimony limit will not be connected to the network because homoplasy is likely to obscure their evolutionary relationships. Once the parsimony limit is calculated, the algorithm then connects haplotypes that differ by a single mutation, followed by haplotypes that differ by two mutations, three mutations and so on. As long as the parsimony connection limit is not reached, the final product is a single network showing the interrelationships of all haplotypes in a way that requires the smallest number of mutations.

The interpretation of parsimony networks draws on coalescent theory because the connections between haplotypes throughout the network represent coalescent events. By following some of the principles of coalescent theory, there are a number of predictions that we can make about parsimony networks, including:

1. High frequency haplotypes are most likely to be old alleles.

2. Within the network, old alleles are interior, whereas new alleles are more likely to be peripheral.

3. Haplotypes with multiple connections are most likely to be old alleles.

4. Old alleles are expected to show a broad geographical distribution because their carriers have had a relatively long time in which to disperse.

5. Haplotypes with only one connection (singletons) are likely to be connected to haplotypes from the same population because they have evolved relatively recently and their carriers may not have had time to disperse.

Figure 5.6A shows a statistical parsimony network of mitochondrial haplotypes from the migratory dragonfly Anax junius that was sampled from locations across North America spanning a maximum distance of approximately 8600 km between Hawaii and Nova Scotia (after Freeland et al., 2003). Figure 5.6B shows the geographical locations of the different haplotypes. By comparing the network and the map, we can get some idea of whether the previously outlined predictions have been realized in this case. Haplotypes 1 and 25 are of the highest frequency, are central to the network, have more than one connection and show a broad geographical distribution. We cannot state unequivocally that these are the oldest alleles, but they meet the expectations of old alleles according to predictions 1--4. Although it is also true that, contrary to prediction 3, some of the haplotypes with more than one connection appear to be new alleles based on their low frequency and peripheral location in the network, haplotype 1 has considerably more connections (12) than any of the low-frequency haplotypes (maximum of 5).

Figure 5.6 (A) Statistical parsimony network of mitochondrial haplotypes that were identified from partial cytochrome oxidase I sequences for the common green darner dragonfly Anax junius in North America. Small dark circles represent missing or unsampled haplotypes, and each step along a lineage (marked by either a dark or an open circle) represents a single mutation. The sizes of the circles are proportional to the haplotype frequencies. (B) Map of North America showing the approximate sampling locations of the different haplotypes. Redrawn from Freeland et al. (2003)

Figure 5.6 (A) Statistical parsimony network of mitochondrial haplotypes that were identified from partial cytochrome oxidase I sequences for the common green darner dragonfly Anax junius in North America. Small dark circles represent missing or unsampled haplotypes, and each step along a lineage (marked by either a dark or an open circle) represents a single mutation. The sizes of the circles are proportional to the haplotype frequencies. (B) Map of North America showing the approximate sampling locations of the different haplotypes. Redrawn from Freeland et al. (2003)

Prediction 5, however, has not been met because there are many examples of singletons being connected to haplotypes that were found in distant locations, e.g. H3 and H4. Disjunctions such as these reflect the extremely high levels of gene flow in A. junius, which mean that mutations often spread before giving rise to new haplotypes. In fact, gene flow is so high in this migratory species that it shows essentially no phylogeographic structuring across a broad geographical range, despite high levels of genetic diversity (Freeland et al., 2003).

While intuitively appealing and not without merit, it is important to note that network methods are not infallible. In one study, researchers investigating the phylogeography of dusky dolphins (Lagenorhynchus obscurus) compared the results that were obtained using four different methods of network construction (Cassens et al., 2003). Although all four methods yielded networks that showed clear genetic differentiation between Pacific and Atlantic haplotypes, the evolutionary relationships within these two groups varied somewhat, depending on which network method was used. The authors of this study concluded that not all methods for constructing networks have been assessed rigorously under all evolutionary scenarios, and in some cases it may be appropriate to use multiple analytical methods so that any conflicting results can be identified and subsequently interpreted with caution.

Was this article helpful?

0 0

Post a comment