Hotspots of diversity of wild Australian soybean relatives and their conservation in situ

Abstract

Mapping diversity hotspots of key species, such as the crop wild relatives, is an essential task for their conservation and for their further exploration. In this paper, we develop and apply methods to locate centres of species richness (SR), endemism, phylogenetic diversity and phylogenetic endemism (PE) for the Australian perennial diploid species of Glycine (Fabaceae). The study taxa are congeneric with the cultivated soybean Glycine max. The DNA sequence data for the phylogenetic analysis are histone H3D gene sequences for these Glycine species. The highest 2.5 % grid cell scores of diversity were defined as the Glycine diversity “hotspots”. The hotspots for the four types of diversity are located in the Kimberley district Western Australia, the Wet Tropics and south-eastern Queensland. The observed frequency distribution of SR values were compared with a theoretical distribution that assumed a species-specific but geographically constant probability for the occurrence of each individual species. The comparison showed broad trends of geographic dispersion overlaying localised high diversity. Simulations of endemism scores supported these themes. No grid cell scored highly for all four diversity metrics, as each index captured specific types of diversity. The inclusion of phylogenetic data pinpointed new areas of biodiversity that were less obvious from other metrics. The Kimberley district emerged as a crucial centre of Glycine diversity with two related lineages of narrowly endemic species. Overall, ~16 % of the endemism centres, and 24 % of the PE centres are conserved in situ in protected areas.

Appendix: Testing the distribution of SR values

The presence versus the absence of a particular Glycine taxon in a grid cell is analogous to the zygosity status (heterozygous vs. homozygous) of a particular genetic locus in an individual. The frequency distribution of the number of cells with a particular value of SR is then analogous to that of the number of individuals with a given number of heterozygous loci. The hypothesis of independent occurrence of species among grid cells (and hence the absence of diversity hot-spots, or indeed the lack of regional divergence) is analogous to the hypothesis of independence of heterozygosity for loci (and hence the lack of multilocus structure). The distribution of multilocus heterozygosity and its sampling behaviour has been formulated for the case of independent heterozygosity and applied to genes in plant populations (Brown and Burdon 1983).

Let H _j denote the observed proportion of grid cells in which the jth species of Glycine is present and 1 − H _j the proportion it is absent. The complement of Glycine species or SR for each grid cell is determined and summarized as the observed distribution of the number (Y_K) of grid cells having a richness of (K = 1, 2, 3, … , 27) species. The total number of grid cells is N = Σ Y_K (=448 here). It is immediately clear that the analogy between species and genes has one problem in that no value for Y₀, the number of cells with K = 0 is observable. To fit a random independent distribution, we must invoke an unknowable variable (Y₀) to be this value. Assuming an estimate of this variable, it is possible to compute s ² _k the variance of this adjusted empirical distribution.

The theoretical distribution of richness assuming independence is obtained by convolution of the generating functions for the binomial distributions for the presence or absence of each single species (j = 1,2, … 27 here). The combined generation function is:

$$ G({\text{X}}) = \prod {\left[ {\left( {1 - H_{j} } \right) + \left( {H_{j} {\text{X}}} \right)} \right]} $$

The function {G(X)} is expanded numerically, multiplying over all 27 species, and the coefficient of the term X ⁱ in this expansion is the probability (P _j ; j = 0, 1, 2, … , 27) that the number of species (richness) in a grid cell is i. In this case, we obtain an expected value for the ‘unobservable’ proportion (=P₀) of grid cells with zero Glycine, and total number of grid cells in the model is adjusted to N + Y₀. The H _j are estimated from the observed frequency of each species, but with the total cells adjusted by the presumed value of Y₀ (= N P₀/[1 − P₀]). However the estimates of {H _j } are no longer appropriate and must be adjusted for the unobservable zero cells, and a new set of estimates of joint probabilities obtained. The iteration is repeated until the value of Y₀ that gives the random model best fitting the observed richness distribution is obtained. This theoretical expected distribution under the null hypothesis of species independence is compared with the observed distribution with goodness-of-fit test and its variance calculated, or it can be calculated directly from the fitted {H _j } values

$$ \sigma_{K} = \sum {H_{j} } - \sum {H_{j}^{2} } $$

The observed distribution of the 448 cells with at least one record of a Glycine species is given in the table below. Estimates of the theoretical distribution of SR assuming independence of species occurrence were then obtained from the observed frequency of cells where a given species occurs by the convolution of generating functions. Expected numbers were adjusted to give the same total of positive cells (488), and in the first iteration the model yielded an estimate of 117 cells without any Glycine (Y₀). However the total Glycine species-grid cell occurrences in model 1 (810) exceeds the observed (641). Model 2 is a second iteration that used the species frequencies corrected for the hypothetical zero cells (117) yields a new estimate of Glycine-zero cells (187). By repeating this iterative procedure, the estimate of Y₀ converged to 313 negative cells as yielding the best fitting random model (model 3 in Table 5) to the observed joint species frequencies.

Table 5 Observed and expected distribution of Glycine SR

The comparison of the observed with this model is instructive. The Chi-square goodness of fit of the four classes (conservatively 1, 2, 3 and >3 species) was 10.41, (P = 0.0055). The observed distribution has excess cells with the lowest (SR = 1) and higher (SR > 3) values, and deficiency in the intermediate (SR = 2, 3). The variance of the observed distribution was 0.836, higher than the theoretical value of 0.763. Increased variance arises from two causes, broad geographic divergence (e.g. dryland species like G. canescens, or temperate species like G. latrobeana occur in particular portions of the total range of the genus). At the other extreme, the occurrence of hot spots of SR also increases this variance.