Using phylogenetic diversity to identify ancient rain forest refugia and diversification zones in a biodiversity hotspot

BIODIVERSITYRESEARCH

Using phylogenetic diversity to identifyancient rain forest refugia anddiversification zones in a biodiversityhotspotCraig M. Costion1,2*, Will Edwards3,4, Andrew J. Ford5, Daniel J.

Metcalfe5, Hugh B. Cross6, Mark G. Harrington1,4, James E. Richardson7,8,

David W. Hilbert5, Andrew J. Lowe2,6,† and Darren M. Crayn1,4,9,†

1Australian Tropical Herbarium, James Cook

University, Cairns Campus, PO Box 6811,

Cairns, Qld, 4870, Australia, 2Australian

Centre for Evolutionary Biology and

Biodiversity, University of Adelaide,

Adelaide, SA, 5005, Australia, 3Centre for

Tropical Environmental Sustainability

Science, School of Tropical Biology, James

Cook University, Cairns Campus, PO Box

6811, Cairns, Qld, 4870, Australia, 4Centre

for Tropical Environmental Sustainability

Science, James Cook University, Cairns

Campus, PO Box 6811, Cairns, Qld, 4870,

Australia, 5CSIRO Ecosystem Sciences,

Tropical Forest Research Centre, Atherton,

Qld, Australia, 6Department of Environment

and Natural Resources, State Herbarium of

South Australia, North Terrace, Adelaide,

SA, 5005, Australia, 7Royal Botanic Garden

Edinburgh, 20A Inverleith Row, Edinburgh,

EH3 5LR, UK, 8Universidad de los Andes,

Apartado A�ereo, 4976, Bogot�a, Colombia,9Centre for Tropical Biodiversity and

Climate Change, James Cook University, PO

Box 4811, Townsville, Qld, 4811, Australia

†These authors contributed equally to the

manuscript.

*Correspondence: Craig M. Costion,

Australian Tropical Herbarium, James Cook

University, Cairns Campus, PO Box 6811,

Cairns, Qld 4870, Australia.

E-mail: [email protected]

ABSTRACT

Aim The plight of the world’s biodiversity hotspots has been paralleled by a

debate over how to best prioritize or maximize gain of biodiversity for conser-

vation. Approaches to date have focused on quantifying species, habitat, phylo-

genetic or other types of diversity. The importance of preserving evolutionary

distinctiveness or phylogenetic diversity (PD) has gained popularity due to its

ability to identify evolutionary patterns in the landscape that traditional taxon

richness measures cannot. Here, we expand upon the application of PD as a

biodiversity index and incorporate data on historical biogeography to under-

stand the processes that shaped the assembly of a tropical flora.

Location Tropical north-east Queensland, Australia.

Methods We generated a genus-level molecular phylogeny for the bioregion to

calculate PD. We then integrated data on historical biogeography into a model

to explain the distribution of PD and the PD residuals and further tested for a

correlation between rain forest stability through time and community assembly.

Results We identified a strong correlation between PD residuals and the bio-

geographic origin of the lineages in the extant flora. Areas with higher PD than

expected based on generic richness (GR) contain a higher proportion of immi-

grant plant lineages dispersed into northern Australia mostly from Southeast

Asia within the past few million years. Areas with lower PD than predicted by

genus richness are rich in ancient Australian relict lineages and are correlated

with previously identified rain forest refugia that have remained stable through-

out the last glacial cycle.

Main conclusions Maximizing PD without historical interpretation may yield

unintended or undesirable conservation outcomes such as deprioritizing ancient

refugia with lower PD values. By understanding the biome assembly of a

region, better-informed decisions can be made to ensure different stages of a

region’s evolutionary history are preserved.

Keywords

Biome assembly, conservation prioritization, Gondwanan, relict species, Sahul,

Sunda.

INTRODUCTION

The conservation of biodiversity to date has largely focused

on maximizing gain of feature diversity of some kind

whether it is species, functional traits, genetic diversity or

phylogenetic diversity. Although much progress has been

made in the development of these types of indices, the sci-

ence of maximizing biodiversity gain has operated separately

from the science of understanding how species and eco-

systems have evolved and dispersed across the globe through

DOI: 10.1111/ddi.12266ª 2014 John Wiley & Sons Ltd http://wileyonlinelibrary.com/journal/ddi 1

Diversity and Distributions, (Diversity Distrib.) (2014) 1–11A

Jou

rnal

of

Cons

erva

tion

Bio

geog

raph

yD

iver

sity

and

Dis

trib

utio

ns

time until relatively recently. Phylogenetic measures in par-

ticular have become increasingly relevant to applied conser-

vation (Mace et al., 2003; Rodrigues et al., 2005; Cadotte &

Davies, 2010). Here, we use a dramatic example to illustrate

how important it is to bring these two disciplines closer

together in applied conservation prioritization contexts. We

use a case study from one of the world’s oldest known con-

tinuous rain forest regions, the Wet Tropics World Heritage

bioregion in north-east Queensland, Australia. We develop a

new application of the phylogenetic diversity (PD) index

(Faith, 1992), which accounts for the biogeographic origin of

the component taxa, to understand patterns of biome assem-

bly and long-term rain forest continuity.

PD is a measure of biodiversity that incorporates informa-

tion on evolutionary relationships (phylogeny) between taxa,

calculated as the total branch length connecting a subset of

taxa on a phylogenetic tree. The phylogenetic trees are usu-

ally inferred using nucleotide sequence data, where branch

lengths are proportional to the number of genetic mutations

estimated to have occurred. Instead of counting the number

of species in a particular location, PD sums the total branch

length represented by those taxa in the phylogeny. In this

sense, two areas with the same number of species may not

have the same PD value. It has been argued that the value of

biodiversity measures that incorporate phylogeny (such as

PD) lies in the expectation that these will better predict fea-

ture diversity of organisms. Thus, maximizing PD should

maximize feature diversity and hence biodiversity option

value (Faith, 1992).

Although several studies have reported a direct correlation

between taxonomic diversity and PD (Polasky et al., 2001;

Rodrigues & Gaston, 2002; Sechrest et al., 2002; Davies et al.,

2008) raising questions as to its usefulness, the residuals of a

regression of taxonomic richness and PD have been shown

to identify important evolutionary patterns in the landscape

that are not otherwise apparent using taxonomic diversity

alone (Forest et al., 2007). Here, we expand upon this use of

PD by incorporating data on the historical biogeography

and community assembly into a model to explain the extant

distribution of genera in north-east Queensland.

The Wet Tropics bioregion of north-east Queensland

(Fig. 1) is globally recognized for its so-called relict Gondwa-

nan flora (UNESCO, 1988). Patches of continuous rain forest

in this location have served as arks for remnant lineages

from Australia’s mesic past. It is generally accepted that rain

forest and mesic habitat were more widespread across

Australia up until c. 40–35 mya (Byrne et al., 2011). Austra-

lia’s separation from Antarctica initiated continent-wide

Figure 1 Phylogenetic diversity (PD)

(left) and residuals of PD (middle)

indicated by colour-coded scale bar at

0.0625° grid cell resolution. Shaded areas

(right) have maintained stable rain forest

habitat since the Last Glacial Maximum.

The Wet Tropics bioregion of north-east

Queensland (inset, upper right). Shaded

areas indicate extant rain forest and vine

thickets in Australia.

2 Diversity and Distributions, 1–11, ª 2014 John Wiley & Sons Ltd

C. M. Costion et al.

aridification with rain forest retreating to a fragmented east

coast distribution. This large-scale contraction of rain forest

habitat was exacerbated by periodic expansions and contrac-

tions of rain forest coinciding with the glacial fluctuations of

the Quaternary period (Hilbert et al., 2007) that would have

acted as an extinction filter. Notwithstanding these condi-

tions, Australia’s persisting rain forest fragments have

remained stable enough to allow the survival of many

ancient lineages of plants and animals. The Wet Tropics in

particular contain the highest concentration of basal angio-

sperm groups in the world (Endress, 1986) with high levels

of endemism including one monotypic family and 62 mono-

typic genera (Metcalfe & Ford, 2008). Further to this excep-

tional evolutionary legacy, accumulating molecular evidence

(Crisp et al., 2010; Byrne et al., 2011; Sniderman & Jordan,

2011) now suggests that the ‘Gondwanan’ elements of the

contracted rain forest flora have had to compete with an

influx of species well adapted to warm, moist conditions

from Southeast Asia, as a result of the Miocene (c. 25 mya)

collision of Sahul (Australian plate) and Sunda (Southeast

Asian plate), facilitating intercontinental dispersal of the

biota (Crayn et al., 2014).

This complex and ancient biotic history would have

undoubtedly left a signature on the extant distribution of

plants. We aimed to determine whether the integration of

the PD index and data on historical biogeography could be

used to explain the extant distribution of genera. Where are

the hotspots of evolutionary history? Do they coincide with

refugial areas thereby highlighting the museum effect (Erwin,

1991), is phylogenetic diversity concentrated in areas where

active speciation is occurring suggesting the importance of

the ‘cradle’ (May, 1990), or are high PD areas associated

with biogeographic convergence zones indicating immigra-

tion and admixture of biotas is a key process to identify for

preserving evolutionary potential?

We explored these questions by first reconstructing the

genus-level phylogeny of the north-east Queensland Wet

Tropics flora using nucleotide sequences of the plastid ribu-

lose-1,5-bisphosphate carboxylase/oxygenase large subunit

(rbcL) for all genera present in a plot network spanning the

region. The high amplification rate of the rbcL gene across

all angiosperm families and its designation as a universal bar-

code for land plants (CBOL Plant Working Group 2009)

make it an optimal choice for measuring feature diversity at

a bioregional scale (Forest et al., 2007). After calculating the

PD at multiple spatial scales from our phylogeny, we mod-

elled PD as a function of GR (genus richness) then mapped

residuals from this model to investigate the spatial variation

in PD that cannot be attributed to the effects of GR. On

examination of the spatial distribution of residuals, we

undertook a further round of model fitting identical to

above but including two further predictor variables, altitude

and an estimate of biogeographic history of the assemblage.

Estimate of biogeographic history was based on per cent

Sahul plate derived (Sahul) versus Sunda plate derived

(Sunda) species per site. In further identifying a correlation

between the spatial patterns of our results and the degree to

which rain forest extent has fluctuated in association with

glacial cycles in north-east Queensland, we demonstrate a

general and broad scale approach that can be used to identify

ancient rain forest refugia and more recent speciation zones

in hyperdiverse regions of the world.

METHODS

Distribution and DNA sequence data

We utilized presence/absence of families, genera, and species,

elevation and GPS location data from a plot network of 238

0.1 hectare plots spanning the Queensland Wet Tropics bio-

region. Leaf tissue was sampled from fresh leaf material or

from herbarium specimens ≤ 10 years of age for at least one

species per genus present in the plot network (Appendix S1

in Supporting Information). Total genomic DNA was

extracted using the Machery Nagel Plant II DNA Extraction

Kit at the Australian Genome Research Facility (AGRF).

Amplification of the rbcLa fragment was performed following

the Consortium for the Barcode of Life (CBOL) Plant

Working Group PCR protocol (CBOL Plant Working

Group 2009) using the primers rbcLa forward (ATGTC-

ACCACAAACAGAGACTAAAGC) and rbcLa reverse (GTAA-

AATCAAGTCCACCRCG). Sanger sequencing was performed

by the Australian Genome Research Facility and the Cana-

dian Centre for DNA Barcoding according to their protocols

and submitted to the Barcode of Life Data Systems (BOLD)

database (Ratnasingham & Hebert, 2007). In total, 585 new

rbcL sequences were generated, representing 585 genera, 43

orders and 129 families of flowering plants. Sequences for 73

genera were obtained from GenBank (http://www.ncbi.nlm.

nih.gov/genbank/).

Phylogenetic tree and PD

Consensus sequences were assembled using CHROMASPRO

v.1.32 and DNA Baser Sequence Assembler v.3, aligned with

MAFFT online v. 6 (http://mafft.cbrc.jp/alignment/server/),

then checked manually. The final alignment of rbcLa (560

base pairs) for 658 genera was analysed using maximum like-

lihood in PhyML (Guindon et al., 2010) using the HKY85

substitution model with estimated gamma shape parameters

and optimized topology and branch lengths. The final tree

was then imported into Biodiverse (Laffan et al., 2010) to

calculate PD, and the richness of species and genera were

plotted against PD per taxonomic unit at four different

spatial scales (grid cells of 0.1 hectares, 0.0625°, 0.125° and

0.25°). To verify the accuracy of PD estimated from our ori-

ginal tree, we then ran a second maximum-likelihood analy-

sis in raxmlGUI (Silvestro & Michalak, 2012) with 1000

bootstrap pseudo replicates using the GTRGAMMA substitu-

tion model. As neither of the maximum-likelihood trees were

completely congruent with the Angiosperm Phylogeny Group

III (Bremer et. al, 2009) topology, we ran a third analysis

Diversity and Distributions, 1–11, ª 2014 John Wiley & Sons Ltd 3

Phylogenetic diversity and rain forest refugia in NE Queensland

that was constrained to the APG III tree (Bremer et al.

(2009)). We constrained the tree by defining monophyletic

groups at the level of order and above following APG III

classification in BEAST (Drummond & Rambaut, 2007). We

then ran a Bayesian analysis using the HKY substitution

model with a lognormal relaxed clock and Yule process tree

prior for 100 million iterations. The maximum clade credi-

bility tree from this analysis and the most likely tree from

the raxmlGUI analysis were then both imported into Biodi-

verse to compute PD as a proportion of total tree length

(PD_P). These results were correlated against the results

obtained for the PhyML tree. The PD_P values from all

three phylogenetic methods were strongly correlated to each

other (PhyML on raxmlGUI, R = 0.9976, P < 0.0001; PhyML

on BEAST, R = 0.9681, P < 0.0001; raxmlGUI on BEAST,

R = 0.9683, P < 0.0001) providing confidence that the results

are independent of the method used to generate the phyloge-

netic structure underlying the calculation of PD. The APG

III constrained tree was selected for display due to its sub-

stantially higher support values below the order level com-

pared with the two unconstrained trees. The maximum clade

credibility tree with posterior support values for each node is

displayed as a cladogram in Fig. S1, and a phylogram of a

randomly selected tree from the results of the same analysis

is displayed in Fig. S2.

Linear regression

To examine the ability of GR to predict PD, we tested the

relationship between them via regression at four spatial scales

(0.1 Ha, 0.0625°, 0.125° and 0.25°). Presence/absence data

from the 0.1 Ha spatial resolution were amalgamated into

grid cells for the three higher spatial scales in Biodiverse. To

account for any components of PD not explained by GR, we

plotted the residual values from previous analyses and exam-

ined the distribution of residuals in geographic space. We

separated the residuals into six classes (see Results), colour

coded to indicate sign (+/�) and size proportional to their

relative values. The spatial analyses were performed, and the

results mapped using Biodiverse and DIVA-GIS v.5.2

(Hijmans et al., 2001).

Spatial autocorrelation

Spatial autocorrelation in PD and residual PD was first

examined via Moran’s I correlogram using spatial eigenvector

mapping (SEVM), generated through Principal Coordinates

Neighbour Matrices (PCNM) (Diniz-Filho & Bini, 2005).

Geographic location information was based on decimal lati-

tude and longitude of the centre of each grid cell, and a

truncation distance (calculated in SAM – Spatial Analysis in

Macroecology; Rangel et al., 2010) of 0.428 decimal units

was used to create spatial filters. Three eigenvector filters

were chosen based on their influence on PD being both sta-

tistically significant (P < 0.05) and having sufficient explana-

tory power (r2 > 0.2) (Huang et al., 2011). Competing

regression models were generated based on all possible com-

binations of GR and the three spatial filters and compared

for their ability to describe PD based on the sample size-cor-

rected Akaike information criterion (AICc). Although the

model with the lowest AICc score was considered to be most

informative (Burnham & Anderson, 1998), all models having

DAICc values ≤ 2 showed substantial support as approximat-

ing models. To visualize the geographic distribution of the

component of PD that could not be described by either GR

or any spatial filter, we then mapped the residuals from the

best approximating model.

Incorporating historical biogeography

A published dataset of floristic origins data (Richardson

et al., 2012) was utilized which assigned three categories of

floristic origin to all genera present in north-east Queens-

land; (1) Sunda – plants which have dispersed southeast

from the Sunda plate, (2) Sahul – plants which originated on

the Sahul plate and persisted in Australia or dispersed north-

west (also popularly known as ‘Gondwanan’) and (3) unre-

solved – lineages of plants for which data are lacking to

assign a floristic origin. Here, we use the term Sahul instead

of ‘Gondwanan’ to distinguish Australasian lineages from lin-

eages that have dispersed to or dispersed back to Australia

from the Sunda plate. In this sense, some Sunda lineages

may have a more ancient Gondwanan origin having rafted to

the Sunda region on the Indian plate. Richness of Sahul and

Sunda taxa was calculated in Biodiverse. Per cent of each flo-

ristic element was then calculated by dividing these numbers

by the total number of species in each plot. We then

included two further predictor variables [altitude (alt)] and

the proportion of species within each local community that

were determined to be from Sahul lineages [p(Sahul)] in

another round of model fitting with spatial filters identical

to that described above. Model selection was based on AICc.

We also checked for the ability of inclusion of spatial filters

to remove possible influence of spatial autocorrelation by

plotting residuals from the best approximating model via

Moran’s I correlogram (Diniz-Filho & Bini, 2005).

We used Spearman’s rank correlation test comparing our

floristic origin data per site, p(Sahul), and the rain forest

stability index modelled for the Wet Tropics bioregion dur-

ing the Quaternary to one hectare resolution (Hilbert et al.,

2007) to test for correlation between the per cent Sahul

species per site and the level of rain forest continuity.

RESULTS

Hotspots of PD and genus richness

PD was found to be concentrated in two primary areas in

north-east Queensland (see Fig. 1, blue). A large southern

hotspot in the southern Cairns–Cardwell lowlands (hotspot

1) and a smaller northern hotspot concentrated in the Dain-

tree region (hotspot 2). Moderately high PD values were also

4 Diversity and Distributions, 1–11, ª 2014 John Wiley & Sons Ltd

C. M. Costion et al.

observed in smaller isolated areas in the Paluma range and

the northern edge of the Black Mountain Corridor. Low PD

is observed between the northern and southern hotspots,

south of hotspot 1 and north of hotspot 2 (Fig. 1, yellow).

Hotspots of genus richness (GR) are nearly identical (Fig.

S3) and when GR is plotted against PD, GR is an accurate

predictor of PD at four spatial scales (Fig. 2).

Significant spatial autocorrelation was associated with PD

among sites (Fig. S4). In our first round of analyses, GR was

identified as a significant explanatory variable in determining

PD (Table 1), although in all cases, the best approximating

model included (at least) one spatial filter (Table 1) and GR.

Coefficients for GR in models were 0.05 at 0.1 Ha scale, but

slightly lower for scales amalgamating sites (0.033, 0.032 and

0.028 for 0.0625°, 0.125° and 0.25° grid cells, respectively)

(Table 1). In all cases, a high proportion of the variation in

PD could be explained by GR and spatial filters alone [R2

values; 87.5% (0.1 Ha); 91.9% (0.065°); 97.3% (0.125°);97.6% (0.25°)]. A total of 12.5% of observed variation in PD

from our point data (0.1 Ha) were not explained by GR and

spatial filters. When this 12.5% variation (PD residuals) was

mapped, a spatial pattern was revealed.

Mapping the residuals of PD

Three spatial patterns were identified from the residuals of

the regression models. The area encompassing hotspot 2 had

high PD overall but lower PD than expected based on genus

richness (negative residuals). The most northern and south-

ern areas and the area between the two hotspots (Black

Mountain Corridor) all had low PD but higher PD than

expected (positive residuals). Hotspot 1 and the Paluma

range areas notably had high PD, which was higher PD than

expected (positive residuals). There also appeared to be a

spatial pattern correlated to elevation. Most of the positive

PD residuals appeared to occur in lowland areas.

Floristic origin data

When the per cent Sunda and per cent Sahul species present

per site were plotted against elevation, the per cent Sunda

species decreased with an increase in elevation (Fig. S5),

although a correlation with per cent Sahul species (Fig. S6)

is not evident. Sites that are rich in Sahul elements are pres-

ent in both the uplands and lowlands; however, sites that are

rich in Sunda elements tend to be restricted to the lowlands.

These results supported the inclusion of floristic origin and

elevation data in a more complex model to explain the varia-

tion of PD. The per cent Sahul species was also significantly

correlated to the rain forest stability index estimated for each

site (Rs = 0.15, Z2 = 2.25, P = 0.024, n = 216).

Modelling PD with historical biogeography data

In the second round of model fitting conducted for 0.1 Ha

resolution, which included the two new terms altitude

(Log10Alt) and proportion of Sahul taxa [p(Sahul)], three

spatial filters were again identified as potentially important.

Of the 61 candidate models, four were considered to have

substantial support as approximations for PD (Table 2). As

in the initial analysis, all four models included both GR and

SF3. All four models also included one of the new predictor

variables p(Sahul). The most informative model contained

these three factors only, while the other three models

Figure 2 The relationship between phylogenetic diversity (PD)

and genus richness (GR) at four different spatial resolutions; 0.1

hectares (black), 0.065° (red), 0.125° (green) and 0.25° (blue).

Table 1 Summary of the best fit multiple regression model at

four spatial scales

Variable b t P

Scale 0.1 ha

Constant 1.943 18.788 < 0.001

GR 0.05 39.673 < 0.001

SF3 1.92 4.087 < 0.001

Scale 0.0625°

Constant 2.675 15.99 < 0.001

GR 0.033 34.789 < 0.001

SF2 2.466 4.735 < 0.001

Scale 0.125°

Constant 2.78 15.299 < 0.001

GR 0.032 42.243 < 0.001

SF1 �0.799 �1.832 0.071

SF2 1.848 4.691 < 0.001

SF3 �2.303 �5.161 < 0.001

Scale 0.250°

Constant 3.6 13.707 < 0.001

GR 0.028 34.225 < 0.001

SF1 �1.528 �3.297 0.002

SF2 2.718 6.398 < 0.001

Standardized regression coefficients (b), t statistics, and associated

P-values of the best fit multiple regression model to explain phylo-

genetic diversity (PD) based on a single predictor (GR = generic

richness), and three spatial filtering variable conducted at four differ-

ent spatial resolutions. The single model presented for each spatial

resolution was that with the lowest AICc value.

Diversity and Distributions, 1–11, ª 2014 John Wiley & Sons Ltd 5

Phylogenetic diversity and rain forest refugia in NE Queensland

included these plus an additional spatial filter or altitude,

either SF2 (model 2), SF1 (model 3) or log10Alt (model 4)

(Table 2). Based on Akaike weights (Wi), we selected model

1 for more detailed analysis. The final model showed signifi-

cant positive association between PD and GR, and PD and

SF3. However, the standardized coefficient for p(Sahul) was

negative which supports the hypothesis that increased pro-

portion of immigrant lineages within local species assem-

blages is associated with higher PD values (Table 3). This

model was a significant improvement over the original

(R2adj = 0.908). Further, the spatial correlogram based on

Moran’s I index suggested that the inclusion of the single

spatial filter was sufficient to overcome the pattern of spatial

autocorrelation in the residuals of the regression (Fig. S4).

While we found that the best model did not include both

Log10Alt and p(Sahul), one model with substantial support

did include both variables (model 4) (Table 2). Spatial corre-

lation between p(Sahul) and Log10Alt was significant and

positive (n = 236, Pearson’s r = 0.452, Fcorrected = 58.607,

d.f.corrected = 227.773, P < 0.001), suggesting that the propor-

tion of species in a local assemblage that are from Sahul

lineages increases with increasing altitude in these forests.

Equivocal lineages

A proportion of taxa could not be resolved to a particular

geographic origin with currently available data. The per cent

unresolved taxa that were calculated as p(Sahul) and p(Sun-

da) (proportion of Sunda taxa) per site were plotted against

p(unresolved species) per site to assess whether unresolved

taxa were more likely to fall within one or the other

categories (Figs S7 and S8). There was no significant spatial

correlation between p(Sunda) and p(unresolved) (n = 236,

Pearson’s r = 0.018, Fcorrected = 0.066, d.f.corrected = 209.31,

P < 0.798). There was, however, a significant spatial

correlation between p(Sahul) and p(unresolved), and the

correlation was negative (n = 236, Pearson’s r = �0.835,

Fcorrected = 415.47, d.f.corrected = 180.89, P < 0.001). Thus, as

the proportion of all taxa identified from Sahul lineages went

up, the proportion of unresolved went down, while the pro-

portion identified as Sunda lineages had little influence on p

(unresolved). This suggests that unresolved species are more

likely to turn out to be of non-Sahul origin. The results thus

present a potentially conservative picture of the influence of

the immigrant element on the Queensland Wet Tropics flora.

It also supports our selection of p(Sahul) as opposed to p

(Sunda) as a more accurate influential factor in the model

test to explain the distribution of PD.

DISCUSSION

The two hotpots of phylogenetic diversity (PD) and genus

richness (GR) identified are consistent with two centres of

endemism previously identified for north-east Queensland

(Crisp et al., 2001). The correlation between hotspots of PD

and hotspots of taxonomic richness (GR) is statistically sig-

nificant and initially suggested that both of these areas may

be significant long-term rain forest refugia. Plotting the lin-

ear regression residuals, however, showed that there were sig-

nificant differences in the evolutionary history of these two

hotspots. Hotspot 1 has a higher concentration of established

Indomalayan or Sunda lineages, and the Daintree region

(hotspot 2) has a much stronger ‘Gondwanan’ character with

a higher proportion of Sahul lineages. In general, we found

that areas that have higher PD than expected based on GR

are positively correlated with areas that have a high percent-

age of Sunda lineages and to a lesser extent also correlated

with low elevation. That the per cent Sahul taxa was a more

accurate predictor of residual PD than elevation explains

exceptions to the elevation trend, particularly in the lowlands

of hotspot 2, which had negative PD residuals.

The fossil record

A wealth of macro and micro fossil data supports the domi-

nance of a widespread cool-adapted rain forest flora for

much of Australia’s history (Dettmann, 1994; Greenwood &

Christophel, 2005). Forests went through warmer (more

tropical) and cooler (more temperate) phases but persisted

until Australia’s final separation from Antarctica at c. 38 mya

brought about the formation of the Antarctic circumpolar

current. This resulted in an increase in the latitudinal

temperature gradient in the Southern Hemisphere, and in

Table 2 Summary of model selection for the top four models

describing the distribution of PD

Model

Adjusted

R2 AICc DAICc K Wi

GR, p(Sahul), SF3 0.908 227.114 0 4 0.304

GR, p(Sahul), SF2, SF3 0.908 228.679 1.565 5 0.139

GR, p(Sahul), SF1, SF3 0.908 228.785 1.67 5 0.132

GR, p(Sahul), logAlt, SF3 0.908 229.065 1.951 5 0.115

DAICc values were compared with the best fitting model. Wi is the

Akaike weight, K is the number of variables, including intercept. Pre-

dictor variables are GR = generic richness, p(Sahul) = proportion of

species derived from Gondwanan lineages, logAlt = log10 altitude of

1Ha study site and SF1–SF3 = spatial filtering variables.

Table 3 Summary statistics of the final multiple regression

model

Variable b t P

Constant 3.084 19.913 < 0.01

GR 0.05 46.135 < 0.01

p(Sahul) �1.873 �9.012 < 0.01

SF3 1.032 2.473 < 0.01

Standardized regression coefficients (b), t statistics, and associated

P-values of the best fit multiple regression model identified in

Table 2. GR = generic richness, p(Sahul) = proportion of species in

community sample known to be derived from Gondwanan lineages,

SF3 = spatial filtering variable.

6 Diversity and Distributions, 1–11, ª 2014 John Wiley & Sons Ltd

C. M. Costion et al.

Australia, cooler temperatures and aridification caused a

massive contraction of rain forest to small refugia along the

eastern coast and ranges (Greenwood & Christophel, 2005).

From the largest of these refugia in north-east Queens-

land’s uplands, a prevalence of Nothofagus and Podocarpa-

ceae in the fossil record up until the early Pliocene suggests

a cooler, more temperate climate than today for much of the

Miocene (Kershaw, 1988). The extant lowland flora of the

region has also been considered the closest analogue to

numerous macrofossil sites of south-eastern Australia (Chris-

tophel, 1981, 1994), thus the extant flora maintains a diverse

assemblage of vegetation types, many of which bear much

similarity to fossil floras of different eras during Australia’s

history (Kershaw et al., 2005). Both upland sites rich in

cooler temperate affinities and lowland sites in the bioregion

have been inferred as long-term refugia (Webb & Tracey,

1981; Hilbert et al., 2007). This is consistent with our find-

ings of the Sahul lineage-rich sites occurring in both the

uplands and lowlands. Hotspot 2, a lowland area in the

Daintree rain forest, has long been regarded as an important

long-term refugium due to its concentration of rare, narrow

range endemic genera. Its distinction from hotspot 1 in the

south, in terms of richness of PD and GR, however, was only

evident after plotting the residuals from our analysis.

Affinities of the extant flora

Although the ‘Gondwanan’ heritage of this region is well

accepted, the significance of Sunda or ‘intrusive’ elements has

remained contentious. Previous authors have noted fossil evi-

dence for long distance dispersals from Southeast Asia to

Australia during its isolation phase in the Oligocene, but

argued this invasion was insignificant (Truswell et al., 1987).

Our data, however, support a strong presence of the Indoma-

layan element in the extant flora. Although few of these lin-

eages have left a fossil record in Australia, nearly all of the

lineages designated as Sahul derived in this study are charac-

teristic of Australia’s fossil floras (Hill, 1994). This is consis-

tent with the premise that lineages that have recently

dispersed and adapted to the Queensland Wet Tropics are less

likely to have left a fossil record.

Elevation and glacial cycle effects

Our results indicate that Sunda plant lineages have more

successfully colonized the lowland areas of Australia’s Wet

Tropics, although with notable exceptions in the northern

Daintree area, whereas the upland areas have retained a

higher percentage of Sahul lineages. Although the proportion

of species of Sunda origin decreases with elevation, the Sahul

element is abundant at higher elevations and also in specific

localities in the lowlands. This is consistent with Australia’s

fossil record, which attests to the resemblance of both extant

upland and lowland sites to fossil floras (see above).

Previous authors have argued that invading Sunda lineages

were unlikely to have been able to compete with an already

established rain forest flora in Australia, one that was

adapted to habitat in gradual decline as aridification swept

across the continent (Adam, 1992). However, periodic glacial

cycles forced rain forest to contract and re-expand. The re-

expansion periods would have provided opportunities for

invading lineages (Richardson et al., 2012). We found direct

support for this hypothesis in the correlation between the

per cent Sahul species per site and the level of rain forest sta-

bility throughout the Quaternary period. Areas that were less

stable through time had a higher proportion of immigrant

lineages.

Figure 1 includes a summary map showing areas of con-

tinuous rain forest during the Quaternary based on a previ-

ously published study that modelled rain forest expansion

and contraction in north-east Queensland (Hilbert et al.,

2007). Our PD hotspot 1 clearly overlaps with large areas of

low stability (white). These areas are currently rain forest but

were not consistently covered by rain forest vegetation over

the Last Interglacial cycle. Hotspot 1 contains high stability

areas (black) but they are notably fragmented and smaller in

comparison with the larger refugia areas directly to the north

and in hotspot 2. Hotspot 2 is notably intact with a high

proportion of high stability areas and lower concentration of

introduced lineages. The Black Mountain Corridor, the

Cooktown area north of hotspot 2, and the region from the

Paluma range south to Townsville are all mostly low stability

and high PD residual areas with a high concentration of

introduced lineages.

These patterns lend support to the view that glacial cycles

influenced the establishment of the Sunda element in north-

east Queensland and can help explain the distribution of the

extant species assemblages. In addition and perhaps most

striking are the patterns visible in the high stability area

(black) directly north of hotspot 1 (See Fig. 1, Atherton-

Bellenden Ker uplands). This area, mostly upland habitat

stretching from the Atherton Tablelands to the Bellenden

Ker and Lamb mountain ranges, has notably low PD but a

strong Sahul or ancient ‘Gondwanan’ character. This case

clearly illustrates the limitations of using diversity measures

(species or phylogenetic) alone. Both of the two largest rain

forest refugia or high stability zones in north-east Queens-

land are mostly comprised of Sahul floristic elements, but

one is a diversity hotspot rich in species, whereas the other

is notably less diverse perhaps due to its more upland char-

acter. Of the two major diversity hotspots, hotspot 2 is

clearly a long standing ancient rain forest refugium whilst

the other, hotspot 1, is diverse due to the intermixing of ref-

ugial Sahul elements and an exogenous flora that likely

became established there during periods when rain forest

re-expanded from refugia.

Conservation implications

Although PD has mostly been promoted for its use in guid-

ing conservation policy by identifying priority areas, recent

studies (Davies & Buckley, 2011; Fritz & Rahbek, 2012;

Diversity and Distributions, 1–11, ª 2014 John Wiley & Sons Ltd 7

Phylogenetic diversity and rain forest refugia in NE Queensland

Kissling et al., 2012) have identified and interpreted links

between PD and historical processes shaping extant species

assemblages. Our study integrates historical biogeography

data into a large-scale PD analysis for an entire flora and

demonstrates how important this is for interpreting biodiver-

sity patterns in a species-rich region. In this case, we identi-

fied distinct zones of both evolutionary relicts and fronts.

But which are more important? ancient refugia of ‘Gondwa-

nan’ heritage or zones of recent intercontinental intermixing

and evolutionary potential? Although there is no clear answer

to this question, it brings up important points for the field

of conservation. Simply choosing a site or a series of sites

that maximize PD in this scenario misses a fundamental

story and the underlying processes that created the PD to

begin with. In this case, maximizing PD without historical

interpretation would deprioritize Australia’s largest remain-

ing tropical rain forest refugium (the Atherton-Bellenden Ker

uplands Fig. 1). This area has low genus diversity but has

remained stable enough through time to serve as an ark for

the survival of relict lineages from Australia’s ancient past.

Being one of the last major fragments of continuous rain

forest in Australia with a history stretching back over 40 mil-

lion years, an approach which deprioritizes this area in

favour of areas with high diversity due to the incursion of

foreign lineages from Southeast Asia in comparatively more

recent times may result in undesirable or unintended conser-

vation outcomes.

Alternatively, if the origin of lineages that dispersed into

Australia comparatively recently was equally ancient or if the

habitat of origin became extinct, these incursion zones (high

PD residuals, Fig. 1) could be viewed as a different type of

refugium. Indeed, with the unabated habitat loss occurring

in Southeast Asia and other tropical regions, this logic may

prove to be a useful argument in other case studies around

the world. Fortunately, the majority of extant rain forest in

north-east Queensland is well protected, thus we have

Figure 3 Summary of our methodology

combining phylogenetic diversity (PD)

and historical biogeography data.

8 Diversity and Distributions, 1–11, ª 2014 John Wiley & Sons Ltd

C. M. Costion et al.

explored these complex questions without the ‘agony of

choice’. Many other species-rich tropical regions of the world

are not in the same position and may benefit from the meth-

ods proposed here to help make informed long-term man-

agement decisions under a range of different conservation

criteria.

In Fig. 3, we outline a generalized summary of our

approach so that it may be tailored and applied in other

regions. Where generating new community phylogenies from

sequence data is unfeasible, bioinformatics tools such as Phy-

lomatic (Webb & Donoghue, 2005) can be used to generate

trees suitable for estimating PD. As data on historical bioge-

ography accumulates, the biogeographical origin of groups

can be determined at a minimal cost with a species list and a

dedicated literature review. This integrated approach may be

used to verify or provide further support for other rain forest

refugia inferred using traditional approaches, such as climate

modelling and population genetics for individual species,

and can rapidly advance knowledge on the evolutionary his-

tory of large species assemblages in other parts of the globe.

ACKNOWLEDGEMENTS

We thank the University of Adelaide, Australian Tropical

Herbarium and Barcode of Life Data Systems molecular

biology labs for helping generate the sequence data for this

project.

REFERENCES

Adam, P. (1992) Australian rainforests. Clarendon Press,

Oxford.

Bremer, B., Bremer, K., Chase, M.W., Fay, M.F., Reveal, J.L.,

Soltis, D.E., Soltis, P.S., Stevens, P.F., Anderberg, A.A.,

Moore, M.J., Olmstead, R.G., Rudall, P.J., Sytsma, K.J.,

Tank, D.C., Wurdack, K., Xiang, J.Q.Y., Zmarzty, S. &

Angiosperm Phylogeny, G. (2009) An update of the Angio-

sperm Phylogeny Group classification for the orders and

families of flowering plants: APG III. Botanical Journal of

the Linnean Society, 161, 105–121.

Burnham, K.P. & Anderson, D.R. (1998) Model selection and

inference: a practical information-theoretic approach.

Springer-Verlag, New York, 353 p.

Byrne, M., Steane, D.A., Joseph, L., Yeates, D.K., Jordan,

G.J., Crayn, D., Aplin, K., Cantrill, D.J., Cook, L.G.,

Crisp, M.D., Keogh, J.S., Melville, J., Moritz, C., Porch,

N., Sniderman, J.M.K., Sunnucks, P. & Weston, P.H.

(2011) Decline of a biome: evolution, contraction,

fragmentation, extinction and invasion of the Austra-

lian mesic zone biota. Journal of Biogeography, 38, 1635–

1656.

Cadotte, M.W. & Davies, T.J. (2010) Rarest of the rare:

advances in combining evolutionary distinctiveness and

scarcity to inform conservation at biogeographical scales.

Diversity and Distributions, 16, 376–385.

CBOL Plant Working Group (2009) A DNA barcode for

land plants. Proceedings of the National Academy of Sciences

USA, 106, 12794–12797.

Christophel, D.C. (1981) Tertiary megafossil floras as indica-

tors of floristic associations and paleoclimate. Ecological

biogeography of Australia (ed. by A. Keast), pp. 379–390.

Junk, The Hague.

Christophel, D.C. (1994) The early tertiary macrofloras of

continental Australia. History of the Australian vegetation,

cretaceous to recent (ed. by R.S. Hill), pp. 262–275. Cam-

bridge University Press, New York.

Crayn, D.M., Costion, C., Harrington, M.G. (2014) The

Sahul-Sunda floristic exchange: dated molecular phyloge-

nies document Cenozoic intercontinental dispersal

dynamics. Journal of Biogeography, Early View published 9

September 2014, DOI 10.1111/jbi.12405.

Crisp, M.D., Laffan, S., Linder, H.P. & Monro, A. (2001)

Endemism in the Australian flora. Journal of Biogeography,

28, 183–198.

Crisp, M.D., Isagi, Y., Kato, Y., Cook, L.G. & Bowman, D.

(2010) Livistona palms in Australia: ancient relics or

opportunistic immigrants? Molecular Phylogenetics and

Evolution, 54, 512–523.

Davies, T.J. & Buckley, L.B. (2011) Phylogenetic diversity as

a window into the evolutionary and biogeographic

histories of present-day richness gradients for mammals.

Philosophical Transactions of the Royal Society B, 366,

2414–2425.

Davies, T.J., Fritz, S.A., Grenyer, R., Orme, C.D.L., Bielby, J.,

Bininda-Emonds, O.R.P., Cardillo, M., Jones, K.E., Gittle-

man, J.L., Mace, G.M. & Purvis, A. (2008) Phylogenetic

trees and the future of mammalian biodiversity. Proceedings

of the National Academy of Sciences USA, 105, 11556–

11563.

Dettmann, M.E. (1994) Cretaceous vegetation, the microfos-

sil record. History of the Australian vegetation. Cretaceous to

recent (ed. by R.S. Hill), pp. 143–170. Cambridge Univer-

sity Press, New York.

Diniz-Filho, J.A.F. & Bini, L.M. (2005) Modelling geographi-

cal patterns in species richness using eigenvector-based spa-

tial filters. Global Ecology and Biogeography, 14, 177–185.

Drummond, A.J. & Rambaut, A. (2007) BEAST: Bayesian

evolutionary analysis by sampling trees. BMC Evolutionary

Biology, 7, 214.

Endress, P.K. (1986) Tropical rainforests of North Queens-

land. Their conservation significance. Special Australian

Heritage Publication Series, 3, 1–195.

Erwin, T.L. (1991) An evolutionary basis for conservation

strategies. Science, 253, 750–752.

Faith, D.P. (1992) Conservation evaluation and phylogenetic

diversity. Biological Conservation, 61, 1–10.

Forest, F., Grenyer, R., Rouget, M., Davies, T.J., Cowling,

R.M., Faith, D.P., Balmford, A., Manning, J.C., Proches, S.,

van der Bank, M., Reeves, G., Hedderson, T.A.J. & Savolai-

nen, V. (2007) Preserving the evolutionary potential of

floras in biodiversity hotspots. Nature, 445, 757–760.

Diversity and Distributions, 1–11, ª 2014 John Wiley & Sons Ltd 9

Phylogenetic diversity and rain forest refugia in NE Queensland

Fritz, S.A. & Rahbek, C. (2012) Global patterns of amphibian

phylogenetic diversity. Journal of Biogeography, 39, 1373–

1382.

Greenwood, D.R. & Christophel, D.C. (2005) The Origins

and Tertiary History of Australian “Tropical” Rainforests.

Tropical Rainforests: past, present, and future (ed. by E.

Bermingham, C.W. Dick and C. Moritz), pp. 336–373.

University of Chicago Press, Chicago, IL.

Guindon, S., Dufayard, J.-F., Lefort, V., Anisimova, M.,

Hordijk, W. & Gascuel, O. (2010) New algorithms and

methods to estimate maximum-likelihood phylogenies:

assessing the performance of PhyML 3.0. Systematic Biol-

ogy, 59, 307–321.

Hijmans, R.J., Guarino, L., Cruz, M. & Rojas, E. (2001)

Computer tools for spatial analysis of plant genetic

resources data: 1. DIVA-GIS. Plant Genetics Resources

Newsletter, 127, 15–19.

Hilbert, D.W., Graham, A. & Hopkins, M.S. (2007) Glacial

and interglacial refugia within a long-term rainforest refu-

gium: the wet tropics bioregion of NE Queensland, Austra-

lia. Palaeogeography Palaeoclimatology Palaeoecology, 251,

104–118.

Hill, R.S. (1994) History of the Australian vegetation. Creta-

ceous to recent. Cambridge University Press, New York.

Huang, Y., Dai, Q., Chen, Y., Wan, H., Li, J. & Wang, Y.

(2011) Lizard species richness patterns in China and its

environmental associations. Biodiversity and Conservation,

20, 1399–1414.

Kershaw, A.P. (1988) Australasia. Vegetation history (ed. by

B. Huntely and T. Webb III), pp. 237–306. Kluwer aca-

demic publishers, Dordrecht.

Kershaw, A.P., Moss, P.T. & Wild, R. (2005). Patterns and

causes of vegetation change in the Australian Wet Tropics

region over the last 10 million years. Tropical rainforests,

past, present, and future (ed. by E. Bermingham, C.W. Dick

and C. Moritz), pp. 374–400. University of Chicago Press,

Chicago, IL.

Kissling, W.D., Eiserhardt, W.L., Baker, W.J., Borchsenius,

F., Couvreur, T.L.P., Balslev, H. & Svenning, J.-C. (2012)

Cenozoic imprints on the phylogenetic structure of palm

species assemblages worldwide. Proceedings of the National

Academy of Sciences USA, 109, 7379–7384.

Laffan, S.W., Lubarsky, E. & Rosauer, D.F. (2010) Biodiverse,

a tool for the spatial analysis of biological and related

diversity. Ecography, 33, 643–647.

Mace, G.M., Gittleman, J.L. & Purvis, A. (2003) Preserving

the tree of life. Science, 300, 1707–1709.

May, R.M. (1990) Taxonomy as destiny. Nature, 347, 129–

130.

Metcalfe, D.J. & Ford, A.J. (2008) Floristics and plant bio-

diversity of the rainforests of the Wet Tropics. Living in

a dynamic tropical forest landscape (ed. by N.E. Stork

and S.M. Turton), pp. 123–132. Blackwell Publishing,

Oxford.

Polasky, S., Csuti, B., Vossler, C.A. & Meyers, S.M. (2001) A

comparison of taxonomic distinctness versus richness as

criteria for setting conservation priorities for North Ameri-

can birds. Biological Conservation, 97, 99–105.

Rangel, T.F., Diniz-Filho, J.A.F. & Bini, L.M. (2010) SAM: a

comprehensive application for Spatial Analysis in Macroe-

cology. Ecography, 33, 46–50.

Ratnasingham, S. & Hebert, P.D.N. (2007) BOLD: The Bar-

code of Life Data System (www.barcodinglife.org). Molecu-

lar Ecology Notes, 7, 355–364.

Richardson, J.E., Costion, C.M. & Muellner, A.N. (2012) The

Malaysian floristic interchange. Biotic evolution and envi-

ronmental change in Southeast Asia (ed. by D. J Gower, J.E.

Richardson, B.R. Rosen, L. R€uber and S.T. Williams), pp.

138–163. Cambridge University Press, Cambridge.

Rodrigues, A.S.L. & Gaston, K.J. (2002) Maximising phyloge-

netic diversity in the selection of networks of conservation

areas. Biological Conservation, 105, 103–111.

Rodrigues, A.S.L., Brooks, T.M. & Gaston, K.J. (2005) Inte-

grating phylogenetic diversity in the selection of priority

areas for conservation: does it make a difference? Phylogeny

and Conservation, 8, 101–119.

Sechrest, W., Brooks, T.M., da Fonseca, G.A.B., Konstant,

W.R., Mittermeier, R.A., Purvis, A., Rylands, A.B. & Gittle-

man, J.L. (2002) Hotspots and the conservation of evolu-

tionary history. Proceedings of the National Academy of

Sciences USA, 99, 2067–2071.

Silvestro, D. & Michalak, I. (2012) raxmlGUI: a graphical

front-end for RAxML. Organisms Diversity & Evolution, 12,

335–337.

Sniderman, J.M.K. & Jordan, G.J. (2011) Extent and timing

of floristic exchange between Australian and Asian rain for-

ests. Journal of Biogeography, 38, 1445–1455.

Truswell, E.M., Kershaw, A.P. & Sluiter, I.R. (1987) The Aus-

tralian-Southeast Asian connection, evidence from the pal-

aeobotanical record. Biogeographical evolution of the Malay

archipelago (ed. by T.C. Whitmore), pp. 32–49. Oxford

University Press, Oxford.

UNESCO (1988) Decision – 12COM XIV.A – Inscription:

Wet Tropics of Queensland. World Heritage List. Available

at: http://whc.unesco.org/en/list/486 (accessed 15 July

2012).

Webb, C.O. & Donoghue, M.J. (2005) Phylomatic: tree

assembly for applied phylogenetics. Molecular Ecology

Notes, 5, 181–183.

Webb, L.J. & Tracey, J.G. (1981) Australian rainforests, pat-

terns and change. Ecological biogeography of Australia (ed.

by A. Keast), pp. 605–694. Junk, The Hague.

SUPPORTING INFORMATION

Additional Supporting Information may be found in the

online version of this article:

Appendix S1 Complete list of all genera represented in phy-

logenetic tree and their corresponding species, sample IDs,

Voucher and GenBank Accession number.

Figure S1 Summary tree, including the outgroup, of the

10 Diversity and Distributions, 1–11, ª 2014 John Wiley & Sons Ltd

C. M. Costion et al.

maximum clade credibility tree from our APG III con-

strained Bayesian analysis. Cladograms for subtrees with pos-

terior support values for each node displayed are provided in

subsequent pages.

Figure S2 Randomly selected tree from the APG III con-

strained Bayesian analysis with branch lengths. Subtrees for

each major clade displayed are provided in subsequent pages.

Figure S3 Phylogenetic diversity (A) and richness of genera

(B) indicated by colour-coded scale bar at 0.0625° grid cell

resolution.

Figure S4 Moran’s I correlogram for phylogenetic diversity

(PD) (solid symbols) and the residuals of multiple regression

with predictors and spatial filters for best fitting model (open

symbols) for plant species community composition in all 238

1 Ha sites.

Figure S5 Per cent Sunda species per site plotted against

elevation.

Figure S6 Per cent Sahul species per site plotted against

elevation.

Figure S7 Per cent Sunda species per site plotted against

unresolved species per site.

Figure S8 Per cent Sahul species per site plotted against

unresolved species per site.

BIOSKETCH

Craig M. Costion is a postdoctoral fellow at the Australian

Tropical Herbarium at James Cook University. His research

is focused on plant biodiversity patterns and conservation in

the Australasia and Pacific region.

Author contributions: C.C., A.L. and D.C. conceived the

study. C.C., A.F. and D.M. collected the data. C.C., H.C. and

M.H. performed the laboratory work. D.M., A.F., J.E.R. and

D.H. provided essential data for the analysis. C.C. and W.E.

performed the analysis. C.C. led the writing.

Editor: Kerrie Wilson

Diversity and Distributions, 1–11, ª 2014 John Wiley & Sons Ltd 11

Phylogenetic diversity and rain forest refugia in NE Queensland