Skip to contents

Tree shape and extinction in metacommunity simulations with comdynamite

Source: vignettes/articles/comdynamite.Rmd
comdynamite.Rmd

treeshapesim and comdynamite divide the work of a simulation study. treeshapesim produces the phylogeny of the regional species pool, with the imbalance and the branching times chosen by the user. comdynamite takes that tree as a fixed input, places traits on it and runs the dynamics of the metacommunity. This article connects the two through a question in which the tree is not just a backdrop: what happens to communities when the pool has lost species, and does it matter which species were lost?

Neither package depends on the other, so comdynamite has to be installed separately.

remotes::install_github("diogoprov/comdynamite")

Why the pattern of extinction matters

If all lineages speciate and go extinct at the same constant rates, the shape of the reconstructed tree is the same as under pure birth (Gernhard 2008). Such extinction changes the branching times, but not the topology. Extinction moves the tree away from that expectation when it differs among lineages.

The same contrast organizes the work on the loss of evolutionary history. For a randomly evolving clade, Nee and May (1997) showed that about 80% of the tree can survive the loss of about 95% of the species, and that choosing the survivors to preserve the most history is not much better than choosing them at random. Current extinction risk, however, is not phylogenetically random, and its clumped distribution puts more higher taxa and more evolutionary history at risk (Purvis et al. 2000). In simulations with phylogenetically patterned speciation rates and extinction risks, Heard and Mooers (2000) found that risks uncorrelated with speciation rates removed only slightly more history than in random trees, whereas risks correlated with speciation rates could increase or decrease the loss strongly. Clustered risk has empirical support, but does not always cause more loss of phylogenetic diversity than random extinction: the shape of the tree also matters (Veron et al. 2017). That dependence is what sim_tree_shape() can vary directly, and comdynamite adds the next question, which is what the losses do to communities.

prune_tips() applies a single round of extinction to a complete tree, as in Nee and May (1997), but the losses need not be random. It removes a fixed fraction of the tips, so that scenarios can be compared at the same richness, in one of three ways.

set.seed(1)
tree <- sim_tree_shape(60, beta = -1.5)
traits <- species_traits(tree, lambda = 1)
risk <- setNames(exp(3 * as.numeric(scale(traits$optimum))), rownames(traits))

random <- prune_tips(tree, 0.4, signal = 0)     # the control
clustered <- prune_tips(tree, 0.4, signal = 1)  # close relatives lost together
linked <- prune_tips(tree, 0.4, risk = risk)    # species with high optima at risk
head(attr(linked, "lost"))
#> [1] "t54" "t1"  "t59" "t58" "t10" "t12"

The tip labels are kept, so the traits of the survivors are found by name. Traits are drawn on the complete tree, before the event, because extinction acts on species that already have them.

One simulation

The remaining tree is the regional pool of the metacommunity.

landscape <- landscape_implicit(20)
sim <- simulate_metacommunity(
  landscape = landscape,
  traits    = traits[linked$tip.label, ],
  alpha     = competition_matrix(linked, type = "phylo"),
  dispersal = 0.01,
  n_steps   = 300
)
dim(sim$abundance)    # sites x species
#> [1] 20 36

A factorial design

The function below runs the whole sequence once and returns the shape of the pool and a few properties of the communities. Environmental optima and the environment of the sites share the range from 0 to 100, so the species at risk under risk are those of the sites with high values.

one_run <- function(beta, scenario, n = 60, fraction = 0.4) {
  tree <- sim_tree_shape(n, beta = beta)
  traits <- species_traits(tree, lambda = 1)
  risk <- setNames(exp(3 * as.numeric(scale(traits$optimum))), rownames(traits))
  pool <- switch(scenario,
    none      = tree,
    random    = prune_tips(tree, fraction, signal = 0),
    clustered = prune_tips(tree, fraction, signal = 1),
    linked    = prune_tips(tree, fraction, risk = risk))
  landscape <- landscape_implicit(20)
  sim <- simulate_metacommunity(landscape, traits[pool$tip.label, ],
                                competition_matrix(pool, type = "phylo"),
                                dispersal = 0.01, n_steps = 300)
  present <- sim$abundance > 0
  richness <- rowSums(present)
  high <- landscape$env > 50
  c(tree_shape(pool),
    pd = sum(pool$edge.length) / sum(tree$edge.length),
    richness_low = mean(richness[!high]),
    richness_high = mean(richness[high]))
}

set.seed(2)
cells <- expand.grid(scenario = c("none", "random", "clustered", "linked"),
                     beta = c(0, -1.5), stringsAsFactors = FALSE)
means <- t(mapply(function(beta, scenario) {
  rowMeans(replicate(100, one_run(beta, scenario)))
}, cells$beta, cells$scenario))
cbind(cells, round(means, 2))
#>    scenario beta gamma colless   pd richness_low richness_high
#> 1      none  0.0 -0.12   -0.03 1.00        16.03         16.36
#> 2    random  0.0 -0.93    0.10 0.76        11.63         11.27
#> 3 clustered  0.0  0.21    0.14 0.68        11.29         10.95
#> 4    linked  0.0 -0.30    0.22 0.71        13.35          9.27
#> 5      none -1.5 -0.01    6.98 1.00        16.28         16.48
#> 6    random -1.5 -0.61    3.79 0.71        11.32         11.33
#> 7 clustered -1.5 -0.14    3.69 0.66        11.24         10.54
#> 8    linked -1.5 -0.38    4.08 0.67        13.06          9.58

The comparison that matters is among the three scenarios of extinction, which have the same richness, and not between them and the complete pool, because the Colless index and gamma also depend on the number of tips. In this example:

  • Phylogenetic diversity. At the same richness, clustered and trait-linked extinction kept less of the total branch length than random extinction (about 0.70 against 0.77 with beta = 0, and 0.66 against 0.71 with beta = -1.5), and each scenario kept less in the imbalanced pools.
  • Gamma. Random losses lowered gamma, as incomplete sampling does, because they remove recent splits throughout the tree. Clustered losses left it near zero: they remove whole clades, and the recent splits of the other clades remain.
  • Local richness. Random and clustered extinction reduced richness evenly along the environmental gradient. Only trait-linked extinction created a difference (about 13 species at sites with low values against 9 at sites with high values), because the species adapted to one end of the gradient were lost.

So the same loss of species can leave the same number of species per site and different phylogenetic structure, or a different distribution of richness along the environment, depending on which species were lost.

Limits of this design

  • The event happens once, at the present. It is not extinction that differs among lineages throughout the history of a clade, for which speciation and extinction have to be simulated together (Maddison et al. 2007; Rabosky 2009; Verboom et al. 2020), and which is outside the scope of treeshapesim.
  • Gamma and the Colless index vary independently only before the event. After it, they are outcomes of the design, and calibrate_shape() does not apply.
  • signal and risk describe the pattern of the losses, not their cause.
  • With a trait that carries phylogenetic signal, as here, extinction linked to the trait is also clustered on the tree. To separate the two, use a lower lambda in species_traits().
  • The numbers above come from 100 replicates of a small example, and only illustrate the workflow. A study needs its own choice of pool size, fraction and dynamics.

References

Gernhard, T. (2008). The conditioned reconstructed process. Journal of Theoretical Biology, 253, 769–778. https://doi.org/10.1016/j.jtbi.2008.04.005

Heard, S. B., & Mooers, A. Ø. (2000). Phylogenetically patterned speciation rates and extinction risks change the loss of evolutionary history during extinctions. Proceedings of the Royal Society B, 267, 613–620. https://doi.org/10.1098/rspb.2000.1046

Maddison, W. P., Midford, P. E., & Otto, S. P. (2007). Estimating a binary character’s effect on speciation and extinction. Systematic Biology, 56, 701–710. https://doi.org/10.1080/10635150701607033

Nee, S., & May, R. M. (1997). Extinction and the loss of evolutionary history. Science, 278, 692–694. https://doi.org/10.1126/science.278.5338.692

Purvis, A., Agapow, P.-M., Gittleman, J. L., & Mace, G. M. (2000). Nonrandom extinction and the loss of evolutionary history. Science, 288, 328–330. https://doi.org/10.1126/science.288.5464.328

Rabosky, D. L. (2009). Heritability of extinction rates links diversification patterns in molecular phylogenies and fossils. Systematic Biology, 58, 629–640. https://doi.org/10.1093/sysbio/syp069

Verboom, G. A., Boucher, F. C., Ackerly, D. D., Wootton, L. M., & Freyman, W. A. (2020). Species selection regime and phylogenetic tree shape. Systematic Biology, 69, 774–794. https://doi.org/10.1093/sysbio/syz076

Veron, S., Davies, T. J., Cadotte, M. W., Clergeau, P., & Pavoine, S. (2017). Predicting loss of evolutionary history: Where are we? Biological Reviews, 92, 271–291. https://doi.org/10.1111/brv.12228