Content uploaded by Norman A Douglas
Author content
All content in this area was uploaded by Norman A Douglas on Apr 15, 2019
Content may be subject to copyright.
Recent vicariance and the origin of the rare, edaphically
specialized Sandhills lily, Lilium pyrophilum (Liliaceae):
evidence from phylogenetic and coalescent analyses
NORMAN A. DOUGLAS,* WADE A. WALL,* QIU-YUN (JENNY) XIANG,* WILLIAM
A. HOFFMANN,* THOMAS R. WENTWORTH,* JANET B. GRAY† and MATTHEW G. HOHMANN‡
*Department of Plant Biology, PO Box 7612, North Carolina State University, Raleigh, NC 27695, USA, †Directorate of Public
Works, Endangered Species Branch, United States Army, Fort Bragg, NC 28310, USA, ‡US Army Corps of Engineers, Engineer
Research and Development Center, PO Box 9005, Champaign, IL 618262, USA
Abstract
Establishing the phylogenetic and demographic history of rare plants improves our
understanding of mechanisms that have led to their origin and can lead to valuable
insights that inform conservation decisions. The Atlantic coastal plain of eastern North
America harbours many rare and endemic species, yet their evolution is poorly
understood. We investigate the rare Sandhills lily (Lilium pyrophilum), which is endemic
to seepage slopes in a restricted area of the Atlantic coastal plain of eastern North
America. Using phylogenetic evidence from chloroplast, nuclear internal transcribed
spacer and two low-copy nuclear genes, we establish a close relationship between
L. pyrophilum and the widespread Turk’s cap lily, L. superbum. Isolation-with-migration
and coalescent simulation analyses suggest that (i) the divergence between these two
species falls in the late Pleistocene or Holocene and almost certainly post-dates the
establishment of the edaphic conditions to which L. pyrophilum is presently restricted,
(ii) vicariance is responsible for the present range disjunction between the two species,
and that subsequent gene flow has been asymmetrical and (iii) L. pyrophilum harbours
substantial genetic diversity in spite of its present rarity. This system provides an
example of the role of edaphic specialization and climate change in promoting
diversification in the Atlantic coastal plain.
Keywords: coalescence, divergence, edaphic, Lilium, Pleistocene, rarity
Received 18 November 2010; revision received 28 April 2011; accepted 5 May 2011
Introduction
Molecular studies of rare plant taxa usually aim to
quantify the level and patterns of genetic diversity in a
particular species (Karron 1987; Hamrick & Godt 1990;
Ellstrand & Elam 1993; Gitzendanner & Soltis 2000).
Phylogeographic studies, on the other hand, often focus
on widespread species and try to discern continental-
scale patterns (Taberlet et al. 1998; Brunsfeld et al. 2001;
Soltis et al. 2006). However, the tools of phylogeogra-
phy, particularly coalescent-based analyses that provide
information about the age and historical demography of
species (Knowles 2009), have only rarely been applied
to investigate the history of rare species (Raduski et al.
2010; Whittall et al. 2010).
Of the ‘seven forms of rarity’ (Rabinowitz 1981), the
most extreme describes taxa that have a narrow geo-
graphic range, require specific habitats and maintain
only small local populations. Many edaphic endemics
(plants restricted to soils with unusual physical or
chemical properties) belong to this category. While the
textbook examples of edaphic endemic plants are
restricted to serpentine, various substrates support
edaphic endemics, including guano, alkali, salt, and
gypsum deposits, limestone, chalk, and granite out-
crops, oligotrophic bogs and deep porous sands (Orn-
duff 1965; Axelrod 1972; Parsons 1976; Kruckeberg &
Correspondence: Norman A. Douglas, Fax: (919) 515 3436;
E-mail: norman_douglas@ncsu.edu
2011 Blackwell Publishing Ltd
Molecular Ecology (2011) 20, 2901–2915 doi: 10.1111/j.1365-294X.2011.05151.x
Rabinowitz 1985; Kruckeberg 1986; Williamson & Baz-
eer 1997). Many aspects of the origin of edaphic ende-
mic species are poorly understood (Rajakaruna 2004).
For instance, such species often occur in close geo-
graphic proximity to their progenitor lineages (e.g.
Baldwin 2005), yet it is not usually known whether or
how strongly gene flow is interrupted. While taxa dis-
playing edaphic endemic syndromes often show
reduced genetic diversity compared with their close rel-
atives (Godt & Hamrick 1993; Baskauf et al. 1994; Ayres
& Ryan 1999; but see Raduski et al. 2010), this may
reflect genetic drift due to lower population sizes or the
effects of selection. Strong selection imposed by edaphi-
cally challenging soils could be sufficient to foster pop-
ulation divergence (Nosil et al. 2009; Freeland et al.
2010). Some edaphic endemics may represent vicariant
populations isolated in narrow parts of formerly wider
ranges and niches of their progenitors (e.g. Crawford
et al. 1985), which may themselves be able to grow on
the unusual substrate without being restricted to it.
Edaphic specialists (especially in bog and sand habi-
tats, Sorrie & Weakley 2001) are an important compo-
nent of the endemic-rich flora of the coastal plain of
eastern North America. Few coastal plain endemics
have been the subject of molecular analyses. Sand dune
habitats in Florida apparently served as Pleistocene
refugia for the genera Dicerandra and Conradina
(Edwards et al. 2006; Oliveira et al. 2007), and in gen-
eral, Florida has been proposed as a major Pleistocene
refugium for many taxa in eastern North America (Sol-
tis et al. 2006). Yet, recent phylogeographic work indi-
cates that some coastal plain endemic species likely
persisted in northerly latitudes throughout the Pleisto-
cene. For instance, the Atlantic coastal plain endemic
Pyxie Moss, Pyxidanthera (Diapensiaceae), shows long-
term range stasis (Wall et al. 2010).
The Fall-Line Sandhills of North and South Carolina
(which occur at the western boundary of the coastal
plain) provide one of the clearest examples of the
edaphic contribution to the botanical diversity of the
Atlantic coastal plain. This region is comprised of roll-
ing hills of open, fire-maintained longleaf pine (Pinus
palustris) savanna dissected by numerous blackwater
streams and wetlands, providing a diverse matrix of
habitats that support at least eight endemic plants (and
numerous near-endemics, Sorrie & Weakley 2001). In
the core of the Sandhills region in southern North Caro-
lina, the uppermost deposit is the Pinehurst formation,
which is characterized by loose coarse-grained sands
found along ridgetops. This formation was deposited in
a tidal environment (J. Nickerson, North Carolina Geo-
logical Survey, personal communication) and may date
to the Eocene (Cabe et al. 1992). Below the Pinehurst
formation (and exposed along drainages and slopes
throughout the region) lies the Cretaceous Middendorf
formation, which is of deltaic origin and thus has more
abundant clays (Sohl & Owens 1991). At the interface
between these (and similar formations in the Carolinas
and southeastern Virginia) occur Sandhills seep and
streamhead pocosin ecotone communities. When kept
open by frequent fires encroaching from the surround-
ing xeric pine savannas, these wetlands can support
extremely high local species richness, among the high-
est values ever recorded in North America (>102 spe-
cies per 1 ⁄100 ha, Schafale & Weakley 1990). The age of
the formations implies that endemic species have poten-
tially had a very long time to adapt to the unusual
edaphic conditions.
In this study, we consider the Sandhills lily, Lilium
pyrophilum (Liliaceae), a striking endemic of the Sand-
hills in the Carolinas and southeastern Virginia. For-
mally described only recently (Skinner & Sorrie 2002),
specimens of this species were previously identified in
herbaria as any of three similar species in the region
(L. superbum,L. michauxii or L. iridollae) that share the
distinctive ‘Turk’s cap’ morphology, in which flowers
are pendent with the tepals reflexed upward. Skinner &
Sorrie (2002) identified three specific plant communities
(Schafale & Weakley 1990; Sorrie et al. 2006) that sup-
port L. pyrophilum, including Sandhills seep and
streamhead pocosin ecotones. The third, small stream
swamps are affected by frequent flooding events in
addition to seepage and rarely support L. pyrophilum
(Sorrie et al. 2006).
Lilium pyrophilum is a very rare species. There are
fewer than 75 historical and extant locations in North
and South Carolina, and Virginia (North Carolina Natu-
ral Heritage Program 2007), and between 2007 and
2009, a survey of all known populations located
<500 stems across 35 populations (W. Wall, unpub-
lished data). Approximately half of the extant popula-
tions and a quarter of the individuals occur on Fort
Bragg Military Reservation in North Carolina, where
prescribed and ordnance-ignited fires maintain appro-
priate habitat.
In describing L. pyrophilum (Skinner & Sorrie 2002),
the authors outlined three phylogenetic hypotheses con-
cerning the origin of the species. First, they speculated
that L. pyrophilum may represent a peripheral isolate of
the Turk’s cap lily, L. superbum, which it most resem-
bles morphologically (albeit with significant differences,
Skinner & Sorrie 2002). Lilium superbum is distributed
throughout much of eastern North America (Fig. 1),
and in contrast to the edaphically specialized L. pyro-
philum, it is a generalist, occurring in rich woods and
oligotrophic wetlands from high elevation to sea level.
Especially in northern parts of its range (e.g. the Pine
Barrens of New Jersey), it can be found in saturated
2902 N. A. DOUGLAS ET AL.
2011 Blackwell Publishing Ltd
sandy habitats not unlike those preferred by L. pyrophi-
lum, but it is not restricted to them. However, it is
essentially absent from the Piedmont and Atlantic
coastal plain from the Carolinas southward. Thus, it is
disjunct from L. pyrophilum by at least 150 km every-
where except in southeastern Virginia (Fig. 1) where
the coastal plain narrows.
Second, they speculated that L. pyrophilum may repre-
sent a hybrid species, with the widespread Carolina lily
(L. michauxii) and L. superbum as progenitors. Homop-
loid hybrid speciation has been implicated in the origin
of other edaphic specialists, e.g. Helianthus paradoxus
(Rieseberg et al. 1990) and Hawaiian Scaevola (Howarth
& Baum 2005). Of the three potentially related species,
L. pyrophilum resembles L. michauxii least, differing in
leaf shape and producing fragrant flowers (Skinner
2002). While the range of L. michauxii does overlap the
range of L. pyrophilum (Fig. 1), they occur in contrast-
ing habitats, with L. michauxii favouring much drier
sites. Notably, L. michauxii and L. superbum co-occur
throughout much of their ranges (Fig. 1), yet natural
hybrids are apparently rare (Skinner 2002).
Finally, Skinner and Sorrie suggested the possibility
that L. pyrophilum may represent a disjunct population
of the Pot-o’-gold or Panhandle lily (L. iridollae), a nar-
row endemic of wet pine savannas in northwestern
Florida (where it is listed as endangered) and adjacent
Alabama. This hypothesis emphasizes similar habitat
requirements of the two species, but downplays consis-
tent morphological differences (e.g. details of rhizome
structure, Skinner 2002; Skinner & Sorrie 2002) and a
range separation of over 700 km (Fig. 1).
In this study, we report the results of a molecular
study focused on L. pyrophilum and its close relatives.
First, we investigated the phylogeny of the eastern pen-
dent species of Lilium to address whether L. pyrophilum
represents a peripheral isolate of L. superbum, a hybrid
between L. superbum and L. michauxii, or a disjunct
population of L. iridollae. Second, we analysed the dis-
tribution of genetic variation within and among the taxa
thought to be closely related to L. pyrophilum and used
coalescent-based methods to explicitly evaluate the pos-
sible timing of the divergence of L. pyrophilum. Our
results are interpreted in the context of the evolution of
rare, edaphically specialized lineages in the Atlantic
coastal plain.
Materials and methods
Sampling and molecular data
Samples were obtained from 50 populations spanning
the geographic range of each of the four focal species
(Fig. 1). We also sampled two populations of Lilium
km
0
km
20 40
Fort Bragg
Military
Reservation
L. iridollae
L. michauxii
L. superbum
L. pyrophilum
0150 300
Fig. 1 Distribution of populations included in this study and geographic ranges of the four focal species.
ORIGIN OF LILIUM PYROPHILUM 2903
2011 Blackwell Publishing Ltd
canadense, another pendant species that lacks the Turk’s
cap morphology. Sampling information is provided in
Table S1 (Supporting Information). Populations were
located in the field based on documented occurrences
from herbarium specimens, element occurrence records
from state Natural Heritage Programs and communica-
tion with local botanists. We endeavoured to sample a
similar number of populations of L. superbum and
L. michauxii spanning the geographic range of each spe-
cies. Our sampling of the rare L. iridollae was limited to
two populations. In general, one individual was taken
to represent each population. Genomic DNA was iso-
lated from fresh or frozen leaves, using the CTAB
method (Doyle & Doyle 1987). Nuclear ribosomal inter-
nal transcribed spacer (‘ITS’) sequences were obtained
with primers ITS4 and ITS5a (White et al. 1990; Stan-
ford et al. 2000). This locus was sequenced to facilitate
comparison with abundant existing data available in
GenBank to determine whether the species in this study
form a monophyletic group. We screened eight chloro-
plast markers from Shaw et al. (2007); of these, three
(the atpI-atpH,psbD-trnT and rpl32-trnL intergenic spac-
ers) consistently amplified and contained variable sites.
As the chloroplast behaves as a single nonrecombining
locus, sequences of these three regions were concate-
nated, and this marker is hereafter referred to as ‘CP’.
We developed single-copy nuclear markers for Lilium.
In general, we screened EST or complete CDS
sequences from Lilium against the Oryza sativa genomic
sequence at GenBank using SPIDEY (Wheelan et al.
2001) with the ‘divergent sequences’ and ‘use large
intron sizes’ options. Candidate sequences were down-
loaded and manually aligned in Se-Al (Rambaut 1996)
using amino acid translations. Homologous sequences
from GenBank were incorporated into the alignments.
When we were confident of the positions of the introns
in the rice genome, we then designed primers using Pri-
mer3 (Rozen & Skaletsky 2000), which were screened
against DNA extracted from L. longiflorum and an Asi-
atic hybrid cultivar (which served as positive controls
because nearly all of our candidate regions were based
on sequences from these cultivated lilies) and the four
taxa in our study. We were able to obtain single ampli-
cons for relatively few of these regions even after exten-
sive PCR optimization; it was often the case that
primers would amplify nontarget regions or that introns
would be small, invariant or missing entirely. The clo-
sely related L. canadense has a phenomenally large gen-
ome (1C = 47.90 pg, 46.9 Gbp; Zonneveld et al. 2005;
Peruzzi et al. 2009), which may have contributed to the
difficulty we encountered in obtaining single-copy
nuclear sequences. However, we were able to design
primers that amplified two novel regions. The first
includes two introns between exons 8 and 10 of the
L. longiflorum alkaline phytase gene, LlAlp (‘AP’, prim-
ers: AP8f, 5¢-TCTCCTTGGGCTCTTTCTTG and AP10r,
5¢-GAAAACCTCAAATGGGCAGAG), which is
involved in phytic acid metabolism (Mehta et al. 2006).
While GenBank contains sequences for two isoforms of
this gene, our PCR experiments are consistent with
these representing splice variants of a single locus. The
second region corresponds to a region between exons 5
and 10 of the AKT1-like potassium channel LilKT1
(‘AKT’, primers: AKT5f, 5¢-AGAGACTCTTGATGCACT
TCCTAAA and AKT10r, 5¢-AAGAGAACAACA-
CAACTTTCATTCC). This locus was more difficult to
amplify, and we were unable to generate sequences for
L. iridollae. Primers and PCR conditions for ITS and the
chloroplast loci followed White et al. (1990) and Shaw
et al. (2007). For AP and AKT, PCR contained 2.5 lL
10·PCR buffer, 1%BSA, 200 lMdNTPs, 2.5 mMMgCl
2
,
4lMof each primer and 0.5 U Taq DNA polymerase.
Cycling conditions were 95 C for 4 min, followed by
35 cycles of 95 C for 30 s, 58 C for 30 s, 72 C for
2.5 min, and a final extension step of 72 C for 4 min.
Amplicons were cleaned with Antarctic Phosphatase
and Exonuclease I (New England Biolabs, Ipswich, MA,
USA). Sequencing was performed on an Applied Bio-
systems 3730 capillary sequencer (Foster City, CA,
USA) using Big Dye chemistry. Chromatograms were
edited in Sequencher 4.1.2 (Gene Codes Corporation,
Ann Arbor, MI, USA). Heterozygous bases were easily
identified in the chromatograms for the three nuclear
regions and coded with standard IUPAC notation.
Because of the low levels of divergence among our
sequences, alignment was trivial and performed manu-
ally in Se-Al. The most likely haplotypic phases of AP
and AKT genotype sequences were ascertained with a
combination of cloning and the program PHASE 2.1
(Stephens et al. 2001; Stephens & Donnelly 2003) called
by the ‘Open ⁄Unphase genotype’ option in DnaSP v. 5
(Librado & Rozas 2009); the inferred alleles form the
basis for all further analyses involving these loci. The
preferred model of sequence evolution for each locus
(ITS: TIM3ef + I + G; CP: K81uf + I; AP: TVM + I; AKT:
TVM + I + G) was determined according to Akaike
Information Criterion (AIC) in jModelTest (Posada
2008). Sampling details, genotype information and Gen-
Bank accession numbers are provided in Tables S1 and
S2 (Supporting Information).
Phylogenetic analyses and descriptive population
genetics
For the ITS analysis, 44 new sequences were aligned
with 49 from GenBank to create a matrix of 93
sequences. Included were the four species in this study,
plus 37 other taxa including the pendent eastern North
2904 N. A. DOUGLAS ET AL.
2011 Blackwell Publishing Ltd
American species, L. michiganense,L. canadense and
L. grayi, and eight others from Lilium section Pseudoliri-
um, the monophyletic group of North American species
(Nishikawa et al. 1999) to which all taxa in this study
belong. Unweighted parsimony analysis for the ITS
locus was accomplished using PAUP* 4.0b10 (Swofford
2002) using 100 random-addition sequence replicates
with TBR branch swapping; owing to overall low
sequence divergence, parsimony bootstrapping was
conducted with 10
6
‘fast’ stepwise addition sequences
(Soltis & Soltis 2003). Maximum-likelihood (ML) analy-
sis for this locus was conducted in GARLI v. 1.0
(Zwickl 2006). Likelihood bootstrap values were
obtained with 1000 replicate searches. The statistical
parsimony haplotype network was computed for com-
plete sequences of the three chloroplast regions, atpI-
atpH,psbD-trnT and rpl32-trnL (38 sequences), using
TCS (Clement et al. 2000). The nuclear loci (AP: 82
haplotypes; AKT: 62 haplotypes) have a more compli-
cated evolutionary history than chloroplast sequences;
thus, network analyses for the two were conducted
using the geodesically pruned quasi-median network
algorithm (Ayling & Brown 2008) as implemented in
SplitsTree4 (Huson & Bryant 2006), which produces
pruned networks that connect all sequences (including
multistate characters) by at least one shortest path. ML
trees (not shown) were inferred for these sequences as
well; they were poorly resolved and showed few sup-
ported nodes. However, neither nuclear locus showed
phylogenetic evidence of paralogy. For L. michauxii,
L. superbum and L. pyrophilum, Arlequin v. 3.5 (Excof-
fier & Lischer 2010) was used to estimate haplotype
richness, number of segregating sites, nucleotide diver-
sity p(Nei 1987) and Watterson’s (1975) population
mutation parameter h, for the chloroplast and single-
copy nuclear loci.
Testing divergence between L. michauxii,
L. pyrophilum and L. superbum
As our data include a single individual per ‘popula-
tion’, we treated species as the main hierarchical level
for the purposes of these analyses. Pairwise F
ST
(Weir &
Cockerham 1984) and the exact test of population dif-
ferentiation (Raymond & Rousset 1995; Goudet et al.
1996) between L. michauxii,L. superbum and L. pyrophi-
lum were calculated in Arlequin v. 3.5 (Excoffier & Li-
scher 2010), with individuals and species used as the
hierarchical groupings. Significance was assessed with
10
3
permutations (F
ST
)or2·10
6
Markov chain steps
(exact test).
The nature of the divergence between L. superbum
and L. pyrophilum was further investigated using the
isolation-with-migration model (Nielsen & Wakeley
2001), implemented in IMa2 (Hey & Nielsen 2007). The
full model in the two-population case includes six
parameters (divergence time, hfor the ancestral and
two descendent populations and migration rates
between the descendent populations). This model
assumes no recombination within loci and free recombi-
nation between loci and that markers are selectively
neutral. Thus, several recombination detection methods
available in the program RDP3 (beta 40; Martin et al.
2005) were used to search for recombinant alleles. As
selection or demographic changes can cause departures
from neutral expectations, DnaSP v. 5 (Librado & Rozas
2009) was used to perform three different tests of neu-
trality: Tajima’s D(Tajima 1989), Fay and Wu’s H(Fay
& Wu 2000) and R
2
(Ramos-Onsins & Rozas 2002). Crit-
ical values for these statistics were obtained using 10
5
coalescent simulations. The chloroplast data set showed
no evidence of recombination; the AP and AKT data
sets were filtered with IMgc Online (Woerner et al.
2007) to create data sets that were free of detectable
recombination and infinite sites violations. Maximum
priors for the IMa2 analysis were based on recom-
mended starting values given in the program documen-
tation and refined after preliminary exploratory runs.
Priors ultimately selected were population mutation
rates (for L. pyrophilum,L. superbum and ancestral pop-
ulation) h
0
,h
1
and h
2
= 47, splitting time parameter
t= 3 and population migration rate m
1
and m
2
= 10.
Mutation rate priors (CP: 1.5 ·10
)9
, AP & AKT:
6.03 ·10
)9
) were specified based on values given by
Gaut (1998). Seventy geometrically heated chains (using
the heating parameters ha= 0.98, hb= 0.50) were run
for 750 000 generations beyond a 150 000 generation
burn-in and trees were sampled every 75 generations.
This process was repeated 10 times using different ran-
dom number seeds.
Because results from each replicate were similar, 10
5
trees were concatenated into a single run in load-trees
mode and the ‘test nested models’ option was activated.
This option evaluates the likelihood of 24 models sim-
pler than the full isolation-with-migration model by
constraining parameters (other than divergence time)
and rejecting those that are significantly worse than the
full model based on a likelihood ratio test. We also
compared models using an information-theoretic
method (Carstens et al. 2009), which allows the relative
performance of nested and non-nested models to be
compared using AIC. Compared with a hypothesis-test-
ing approach, which simply identifies models that are
rejected as significantly worse than the full model, the
information-theoretic approach provides model weights
that allow the relative performance of each of a given
set of models, including the full model, to be com-
pared directly with others given the data (Burnham &
ORIGIN OF LILIUM PYROPHILUM 2905
2011 Blackwell Publishing Ltd
Anderson 2002). We used the full model posterior prob-
ability and the 2(log-likelihood ratio) values, which
IMa2 estimates for each model under the assumption
that the model’s posterior probability is proportional to
its likelihood, to calculate the AIC for the full model
and each nested model. Subsequently, Akaike weights
and evidence ratios were calculated (Burnham &
Anderson 2002; Carstens et al. 2009).
Conversion of the IMa2 parameter estimates from
coalescent to demographic units was accomplished
assuming a generation time of 20 years. This is arbitrary
but conservative, based on what little is known about
the natural history of these species. Germination and
establishment is slow, taking two seasons, and plants
need 7 years to reach flowering size. Year-to-year survi-
vorship is relatively high (>0.95, Wade Wall, unpub-
lished data). Using the equation T=a+[s⁄(1 )s)],
where T= generation time, a= age of first reproduction
and s= adult survivorship (Lande et al. 2003), we
obtain a value of 26 years. Although estimates of survi-
vorship could be too high, the Lande equation does not
account for the fact that older plants are typically larger
and more fecund than younger ones. In either case, our
generation time should be considered a minimum esti-
mate.
Because isolation is implicit in the isolation-with-
migration framework, we tested this assumption with a
series of coalescent simulations. Briefly, we estimated
N
e
for each locus using BEAST (Drummond & Rambaut
2007). Because only L. pyrophilum and L. superbum
sequences were included, simpler ML models were uti-
lized (CP: HKY, AP: TnN + I + G, AKT: K81uf + I). We
then used Mesquite v. 2.73 (Maddison & Maddison
2010) to simulate 1000 data sets under each of several
simple divergence models (using estimated substitution
models for each locus). We treated each species as a
population such that L. superbum had a N
e
3·that of
L. pyrophilum (the total N
e
corresponding to the value
from BEAST). The two populations coalesced at times cor-
responding to 2.58 Ma (earliest Pleistocene), 126 ka
(upper Pleistocene) or 18 ka (last glacial maximum). We
then conducted parsimony searches using PAUP* 4.10b
(Swofford 2002) on each simulated data set saving 1000
consensus trees. Slatkin and Maddison’s s(i.e. the num-
ber of parsimony steps implied by a given topology
treating source population as a character, Slatkin &
Maddison 1989) was computed for each tree to create a
null distribution for each locus and divergence time.
This was compared with the value of sfor the empirical
data. When minimum empirical values for swere
higher than 95%of the simulated values, we rejected
the scenario. To evaluate the effect of the level of migra-
tion inferred by IMa2, we duplicated these analyses,
but allowing migration. Because Mesquite only allows
symmetrical migration, we specified a rate of 9.8 ·10
)6
migrants per individual per generation, which corre-
sponds to the estimated value of the parameter under
the ‘equal migration rate’ nested model in IMa2.
Finally, following Gugger et al. (2010), we evaluated the
no-divergence scenario by simulating 1000 data sets per
locus under a single population scenario. The resulting
parsimony consensus trees were contained within the
two-population model described previously, and the
null distributions of swere calculated. In this case, the
scenario was rejected if the maximum empirical values
of swere lower than 95%of the simulated values. As
coalescent parameter estimates based on single loci are
highly sensitive to stochastic error (Edwards & Beerli
2000), these simulations were conducted for both the
upper and lower 90%HPD estimates of N
e
from BEAST.
Table 1 Genetic diversity and results of neutrality tests
Species locus
Lilium michauxii Lilium pyrophilum Lilium superbum
CP AKT AP CP AKT AP CP AKT AP
Individuals
(haplotypes)
8 (8) 5 (10) 7 (14) 15 (15) 13 (26) 18 (36) 13 (13) 12 (24) 15 (30)
Aligned length (bp) 2361 1428 453 2360 1428 453 2361 1428 453
Segregating sites 7 10 13 7 24 8 9 30 18
Observed haplotypes 5 7 9 4 16 9 7 17 12
Nucleotide diversity p0.0010 0.0033 0.0098 0.0008 0.0024 0.0016 0.0009 0.0040 0.0053
Watterson’s theta h0.0011 0.0025 0.0090 0.0009 0.0044 0.0043 0.0012 0.0061 0.0100
Tajima’s D)0.4150 0.0487 0.3349 )0.4468 )1.7637* )1.8536** )1.0835 )1.2142 )1.6319*
Fay and Wu’s H1.7857 0.8000 2.2418 1.3429 )8.8862* )2.8794* )1.9615 )4.8333 0.6437
R
2
0.1577 0.2091 0.1597 0.1301 0.0625** 0.0495*** 0.1105* 0.0828 0.0692*
Sampling represents the number of individuals and the number of haplotypes (for phased nuclear loci). Significance of neutrality
tests was assessed with 10
5
coalescent simulations in DnaSP v. 5.1 (*P< 0.05, **P< 0.01, ***P< 0.001).
2906 N. A. DOUGLAS ET AL.
2011 Blackwell Publishing Ltd
Results
Phylogenetic analyses
In the analysis of ITS data, overall support is quite weak
at the level of intra- and interspecific relationships, with
no significant (‡70%) bootstrap support for the mono-
phyly of the North American section Pseudolirium or the
eastern pendent-flowered species (Fig. 2). However,
there is a relatively high level of support for the branch
uniting two accessions of Lilium iridollae, for that uniting
the eight samples of L. michauxii, and, finally, for the
branch leading to the 32 samples of L. pyrophilum and
L. superbum. Little divergence is evident among the
Fig. 2 Maximum-likelihood (ML) Phylogram of internal transcribed spacer sequences. Support values are ML bootstrap ⁄Bayesian
posterior probability.
ORIGIN OF LILIUM PYROPHILUM 2907
2011 Blackwell Publishing Ltd
accessions of each species (with the exception of the
GenBank sequences for L. superbum,L. canadense and
L. michiganense). The statistical parsimony network
(Fig. 3) computed for the chloroplast data revealed a
common haplotype (1) that was found in all four species,
plus 11 less common types. Overall, four of the six non-
singleton haplotypes occur in multiple species. Quasi-
median networks produced for the AKT and AP loci
(Fig. 4) showed that, while AP haplotype 8 is one muta-
tional step from the nearest L. michauxii haplotype
(m4a), most L. michauxii (and L. iridollae in AP) haplo-
types are separated from a cloud of L. pyrophilum and
L. superbum haplotypes, which are thoroughly inter-
mixed and frequently shared. No haplotypes were
shared between L. pyrophilum and L. michauxii.
Genetic diversity
Haplotype richness h, segregating sites S, nucleotide
diversity pand Watterson’s hare given in Table 1.
Nucleotide diversity is relatively low, with values
between 0.0008 and 0.00978 substitutions per site, and
average values for AP and AKT are nearly five times
the value for the chloroplast data set.
Tests of neutrality
Departures from neutrality were detected in the
nuclear data sets in L. pyrophilum and L. superbum,
where there were significant negative estimates of
Tajima’s Dand R
2
. Fay and Wu’s His significant in
L. pyrophilum only. Tajima’s Dis sensitive to both
demographic expansion and selection, and R
2
is
designed to detect population expansion (Ramos-
Onsins & Rozas 2002). While Fay and Wu’s His most
sensitive to recent positive selection, it may be sensi-
tive to particular demographic conditions involving
structured populations (Fay & Wu 2000). We believe
these loci are unlikely to be under positive selection,
because there is no obvious reason two loci should
deviate from neutrality more strongly in L. pyrophilum
than in the other two taxa. The chloroplast data also
show some demographic expansion in L. superbum
(weakly significant R
2
) without a significantly negative
D. Thus, while we cannot eliminate the possibility of
some background selection in the nuclear data sets
(which does not violate the assumptions of IMa2), it is
more likely that demographic factors explain the signif-
icant values for these statistics.
2: i3,i1,m8,s14
8: p3,p8,p11,p12,p13,s4,s16
L. iridollae
L. michauxii
L. superbum
L. pyrophilum
6: p21
3: m1
10: s11,s3,s15
12: s12
5: m5,m9
11: s5
9: p15 7: p1
4: m4,s6
1: i2,m7,m6,p16,p17,p18,
p2,p4,p5,p6,s17,s13,s7
CP
Fig. 3 Chloroplast haplotype network. Statistical parsimony network for CP haplotypes. Chart area reflects the frequency of the hap-
lotype; each slice reflects the frequency at which each haplotype was found in each species. Haplotype numbers (bold) and sample
abbreviations correspond to those in Tables S1 and S2 (Supporting Information). Edges represent mutations, black dots unsampled
haplotypes.
2908 N. A. DOUGLAS ET AL.
2011 Blackwell Publishing Ltd
Differentiation of L. michauxii
Pairwise F
ST
values (Table 2) revealed that L. michauxii
was significantly divergent from L. pyrophilum and
L. superbum for the AKT and AP data sets, whereas dif-
ferentiation between L. pyrophilum and L. superbum was
minimal and only significant in the AKT data set. No
significant differentiation was detected among any of
the three species for the CP data set. Conversely, all
pairwise exact differentiation tests (Raymond & Rousset
1995) were significant for the two nuclear loci; for the
cpDNA, a significant result was only obtained between
L. pyrophilum and L. michauxii.
Divergence between L. pyrophilum and
L. superbum
Under the isolation-with-migration model, estimates of
the mutation parameter theta (h) were L. pyrophilum:
3.736; L. superbum: 10.79; and ancestral population:
1.292, corresponding to effective population sizes (95%
highest posterior density interval, abbreviated ‘95%
HPD’) of 11 400 (2800–29 700), 32 900 (12 800–86 900)
and 3900 (0–14 400), respectively (Fig. 5a). The splitting
time between L. pyrophilum and L. superbum was esti-
mated as 0.7725 coalescent units, with the 95%HPD
being 0.3435–2.405 (Fig. 5b). This estimate corresponds
to a divergence time of 188 ka (95%HPD 84–586 ka)
with the assumed mutation rates and generation time.
The posterior distribution of splitting time did not reach
zero (nor did it for much higher prior values in preli-
minary runs), so 95%HPD intervals should be inter-
preted with caution. The coalescent migration rate m
from L. superbum into L. pyrophilum was highest at
zero, while the converse was 1.915. Thus, population
migration rates (2 NM, Hey & Nielsen 2004) are asym-
metrical and quite high from L. pyrophilum into L. su-
perbum (2 NM = 9.98, Fig. 5c). The model selection
procedure (Table 3) preferred a model that holds the
two species’ population sizes equal and the migration
rate from L. superbum to L. pyrophilum at zero (model
weight w= 0.32). The next best model (w= 0.22) also
fixed the L. superbum fiL. pyrophilum migration rate at
zero but allowed the population sizes to vary. The full
model (w= 0.19) had the next highest weight, and the
next three models differed in that they fixed the
population sizes as above (model 4), held migration
rates equal (model 5) and held the L. pyrophilum fi
L. superbum migration rate at zero (model 6). The six
best models are assigned 95.6%of the total weight. The
remaining 19 models had some combination of zero
migration, and one or both of the population sizes
Fig. 4 Quasi-median joining networks for the nuclear loci AP and AKT. Network representations of the relationships between
nuclear haplotypes (bold numbers and sample abbreviations correspond to Tables S1 and S2, in Supporting Information). In quasi-
median-joining networks, each haplotype is connected to the others by at least one shortest path. Mutational steps are indicated by
edges, and black dots represent potential unsampled haplotypes.
ORIGIN OF LILIUM PYROPHILUM 2909
2011 Blackwell Publishing Ltd
equal to the ancestral population size. For the sake of
comparison, likelihood ratio tests comparing each
nested model to the full model rejected 20 of 24 nested
models. The four that were not rejected, combined
with the full model, represent 94.2%of the cumula-
tive model weight from the information-theoretic anal-
ysis. Coalescent simulations under both the earliest
Pleistocene (129 000 generations, 2.58 Ma) and upper
Pleistocene (6300 generations, 126 ka) divergence sce-
narios were rejected (Table 4). However, divergence
during the last glacial maximum (900 generations,
18 ka) was not rejected, and neither was the single
population scenario under either the highest or lowest
credible estimates for N
e
. Inclusion of migration in
these simulations did not qualitatively change the
results.
Discussion
Three hypotheses
Our results do not favour two of the three hypotheses
concerning the relationships of Lilium pyrophilum
advanced by Skinner & Sorrie (2002). First, it is unlikely
that L. pyrophilum simply represents a disjunct popula-
tion of L. iridollae: the ITS phylogeny unambiguously
allies L. pyrophilum with L. superbum, whereas L. iridollae
is most closely related to L. michauxii. That L. pyrophilum
and L. iridollae are independent only heightens the con-
servation concern of each of these rare species.
Second, the hypothesis that the species originated as
a hybrid between L. michauxii and L. superbum is not
supported by network analyses (Fig. 4). If L. pyrophilum
represented a recent hybrid, single-copy nuclear loci
should be related to both parental species. Instead, most
L. pyrophilum and L. superbum haplotypes are closely
related to each other (and many are shared), while they
show less similarity to L. michauxii. The phylogenetic
analysis of ITS sequences placed the L. pyrophilum sam-
ples with L. superbum sequences only, to the exclusion
of the L. michauxii sequences.
Lilium pyrophilum appears to be a peripheral isolate
of L. superbum. Our results indicate that the overall
magnitude of divergence between the two lily species
is very low and that the origin of L. pyrophilum is
likely to have been very recent, i.e. during the latter
Pleistocene or Holocene. Our estimated divergence
date from the IMa2 analysis of 188 ka (Fig. 5b) would
fall within the Illinoian glacial period. The minimum
credible divergence time of 84 ka would seem to indi-
cate that L. pyrophilum is in fact isolated from L. super-
bum. In spite of low F
ST
values (Table 2), zero
probability is assigned to the most recent divergence
times in this analysis. The results of the simulation
Table 2 Pairwise Fst and exact test of
population differentiation
Lilium michauxii Lilium pyrophilum Lilium superbum
L. michauxii 0.109 ⁄0.393*** ⁄0.625*** 0.046 ⁄0.328*** ⁄0.567***
L. pyrophilum *⁄*** ⁄*** 0.007 ⁄0.021 ⁄0.057*
L. superbum –⁄*** ⁄** – ⁄** ⁄*
Loci: CP ⁄AP ⁄AKT. Above diagonal, pairwise F
ST
; below diagonal, exact test of
differentiation (Goudet et al. 1996; Raymond & Rousset 1995). Significance assessed in
Arlequin by either 10
3
permutations (F
ST
)or2·10
6
Markov chain steps (exact test);
*P< 0.05, **P< 0.01, ***P< 0.001.
0 20 40 60 80 100 120 140
0.0 0.1 0.2 0.3 0.4 0.5
Effective pop. size (Ne),
×
1000
Probability
Ancestral
L. pyrophilum
L. superbum
(a) (b) (c)
0 200 400 600
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
Divergence time (ka)
0 20406080
0.00 0.05 0.10 0.15 0.20 0.25
Migration (2NM)
L. pyrophilum
L. pyrophilum
→
→
L. superbum
L. superbum
Fig. 5 Posterior probability distributions for IMa2 model parameters under the full model. (a) Effective population size for both spe-
cies and the ancestral population. Both descendent taxa are inferred to have larger effective population size in this analysis. Esti-
mated values for N
e
are Lilium pyrophilum, 11 400 (95%HPD 2800–29 700); L. superbum, 32 900 (12 800–86 900); and ancestral, 3900
(0–14 400). (b) Divergence time. No probability is found for divergence times near zero; however, the distribution fails to reach zero
at the upper end. The peak corresponds to a value of 188 (84–586) ka. (c) Migration rate. Highest probability for migration from
L. superbum into L. pyrophilum is zero; there is, however, a higher probability of migration in the opposite direction (2 NM = 9.98).
2910 N. A. DOUGLAS ET AL.
2011 Blackwell Publishing Ltd
analysis lead us to interpret the IMa2 results with cau-
tion, however, because they reject divergence >6300
generations (126 ka) ago for each locus and fail to
reject the scenarios with divergence at 900 generations
(18 ka) and with no divergence (Table 4). The models
tested in this approach, however, were simplified with
respect to the full IMa2 model and treat each locus
separately rather than jointly. Regardless of whether
the IMa2 results or the coalescent simulation results
are preferred, the isolation between the two taxa is not
ancient. Mid- to late Pleistocene divergence times have
been found in surprisingly few studies of plants (e.g.
Strasburg & Rieseberg 2008; Bittkau & Comes 2009;
Cooper et al. 2010).
Our results provide insight into the demographic pat-
terns that have affected the two species. Deviations
from neutral expectation indicate population expansion
in both taxa (e.g. the average value for Tajima’s D
across three loci: L. pyrophilum =)1.35, L. super-
bum =)1.31, Table 1). This result is corroborated by the
IMa2 analysis, which demonstrates modern effective
population sizes higher than the ancestral, with the
widespread L. superbum having a larger value (N
e
2.7
times that of the endemic L. pyrophilum, Fig. 5a). It is
worth noting that the effective population size of
L. pyrophilum (11 000 individuals) is surprisingly high
considering the very small range of the species; in fact,
our estimate of N
e
is well in excess of the current cen-
sus population size estimated by a recent inventory.
Two factors may explain this discrepancy. First, our
estimated generation time may be too low, which
would cause us to overestimate effective population
size (and underestimate divergence time). Second, agri-
culture, timber harvesting and fire suppression have
Table 3 IMa2 analysis of nested models
Model description log(P) Terms AIC DAIC
Model
weight
Cum.
weight d.f. 2LLR
P-value,
LRT
h(pyrophilum)=h(superbum), mzero from superbum
to pyrophilum
)4.442 3 14.884 0 0.301 0.301 2 2.986 0.2247
mzero from superbum to pyrophilum )3.825 4 15.65 0.766 0.2052 0.5062 1 1.752 0.1856
Full IM model )2.949 5 15.898 1.014 0.1813 0.6875 — — —
h(pyrophilum)=h(superbum))3.972 4 15.944 1.06 0.1772 0.8647 1 2.045 0.1527
Symmetrical migration )4.803 4 17.606 2.722 0.0772 0.9419 1 3.707 0.0542
mzero from pyrophilum to superbum )6.29 4 20.58 5.696 0.0174 0.9593 1 6.681 0.0097
h(pyrophilum)=h(ancestral), mzero from superbum to
pyrophilum
)7.985 3 21.97 7.086 0.0087 0.968 2 10.07 0.0065
h(pyrophilum)=h(ancestral), mzero from pyrophilum to
superbum
)8.116 3 22.232 7.348 0.0076 0.9757 2 10.33 0.0057
h(pyrophilum)=h(superbum), symmetrical migration )8.408 3 22.816 7.932 0.0057 0.9814 2 10.92 0.0043
h(pyrophilum)=h(ancestral), symmetrical migration )8.424 3 22.848 7.964 0.0056 0.987 2 10.95 0.0042
All hequal, mzero from superbum to pyrophilum )9.858 2 23.716 8.832 0.0036 0.9906 3 13.82 0.0032
h(pyrophilum)=h(ancestral) )7.899 4 23.798 8.914 0.0035 0.9941 1 9.9 0.0017
h(superbum)=h(ancestral), mzero from superbum to
pyrophilum
)9.192 3 24.384 9.5 0.0026 0.9967 2 12.49 0.0019
All hequal )9.858 3 25.716 10.832 0.0013 0.9981 2 13.82 0.001
h(superbum)=h(ancestral) )9.192 4 26.384 11.5 0.001 0.999 1 12.49 0.0004
h(pyrophilum)=h(superbum), mzero from pyrophilum
to superbum
)10.63 3 27.26 12.376 0.0006 0.9997 2 15.36 0.0005
h(superbum)=h(ancestral), symmetrical migration )12.1 3 30.2 15.316 0.0001 0.9998 2 18.3 0.0001
All hequal, symmetrical migration )13.4 2 30.8 15.916 0.0001 0.9999 3 20.9 0.0001
h(superbum)=h(ancestral), zero migration )14.26 2 32.52 17.636 0 0.9999 3 22.63 0
Zero migration )13.35 3 32.7 17.816 0 1 2 20.8 0
h(superbum)=h(ancestral), mzero from pyrophilum to
superbum
)14.26 3 34.52 19.636 0 1 2 22.63 0
All hequal, mzero from pyrophilum to superbum )18.52 2 41.04 26.156 0 1 3 31.13 0
h(pyrophilum)=h(superbum), zero migration )24.86 2 53.72 38.836 0 1 3 43.83 0
h(pyrophilum)=h(ancestral), zero migration )29.23 2 62.46 47.576 0 1 3 52.57 0
All hequal, zero migration )30.93 1 63.86 48.976 0 1 4 55.97 0
Models include the full IM model and 24 simpler nested models for the two-population case. Information-theoretic statistics, based
on log(P), follow Burnham & Anderson (2002) and have been sorted by model weight. Models not rejected under traditional-
likelihood ratio tests (LRT) are included in the 95%confidence set of models selected by AIC.
ORIGIN OF LILIUM PYROPHILUM 2911
2011 Blackwell Publishing Ltd
dramatically transformed much of the landscape of the
Sandhills over the past few hundred years, which may
well have extirpated many populations. As these plants
are long-lived outcrossers, too few generations may
have elapsed for the impact of the current bottleneck to
be fully reflected in the estimated N
e
(Lande & Bar-
rowclough 1987). Although our results suggest that the
existing population has apparently been greatly
reduced recently, much of the original genetic diversity
remains and could be conserved, minimizing the impact
of the present-day population bottleneck.
Gene flow is inferred from L. pyrophilum to L. super-
bum. Models including symmetrical migration are not
weighted heavily compared with models that have zero
or nearly zero gene flow from L. superbum to L. pyrophi-
lum (Table 3). Presently, the two species are disjunct.
However, the plants are visited by strong-flying pollina-
tors, such as swallowtail butterflies and hummingbirds
(Skinner 2002), and the seeds are adapted for wind dis-
persal. Why migration would be asymmetrical is
unknown, but this could be explained by pollinator
behaviour, dispersal or intrinsic barriers to gene flow.
Edaphic endemism in the Sandhills
The Sandhills pre-date the Pleistocene and may be sub-
stantially older, raising the possibility that some ende-
mic taxa may have originated in the Pliocene or earlier
and maintained populations in the region continuously.
How might Pleistocene climate changes have affected
the distribution of Lilium spp. in the coastal plain and
effected the isolation of L. pyrophilum? While periods of
severe climate change may eliminate edaphic endemics
that are unable to migrate to areas with a suitable cli-
mate and substrate, edaphic endemics may in fact be
likely to endure climate change in their geographic
ranges. As their niches are defined more by soils than
climate, they are likely to remain the best competitors
on restrictive soils under a wide range of conditions. In
fact, the degree of edaphic restriction exhibited by a
species often varies with climate: populations may be
widespread in environments with low competition and
edaphically restricted in more favourable climates
(Brooks 1987; Harrison et al. 2009).
The edaphic conditions that currently support popu-
lations of L. pyrophilum have probably been relatively
stable, because the erosional process has no doubt con-
tinually exposed the interface between permeable and
impermeable soils, creating seeps. Yet, the divergence
between L. pyrophilum and L. superbum is compara-
tively recent. Genetic diversity of L. pyrophilum, while
lower than that of L. superbum, is still high, making a
vicariant scenario likely. Thus, the phenotypic diver-
gence described by Skinner & Sorrie (2002) probably
occurred in the context of large populations and sub-
stantial gene flow.
The combination of long-term persistence and recent
divergence of L. pyrophilum indicates that this species
descends from locally adapted populations that were
stranded in the Sandhills as L. superbum retreated to
higher elevations. It is not clear why the intervening
Piedmont region supports neither taxon; however,
many groups show a similar disjunction (Braun 1955;
Sorrie & Weakley 2001). This study indicates that for
these lilies, at least, the disjunction coincided with Pleis-
tocene climate oscillations; this may apply to other taxa
that share similar distributions. More in-depth studies
of the L. pyrophilum ⁄L. superbum system, using micro-
satellite markers, will quantify genetic structure within
L. pyrophilum, and gene flow within and between L. py-
rophilum and L. superbum. These more detailed analyses
will improve estimates of divergence time and gene
flow and identify populations of high conservation pri-
ority. Better understanding of this group will provide
further insight into the role of edaphic specialization,
possibly brought on by climate change, in promoting
diversification.
Acknowledgements
We thank Fort Bragg Military Reservation and the Endangered
Species Branch for logistic support and the Construction Engi-
neering Research Laboratory (US Army Corps of Engineers
Agreement #W9132T-07-2-0019) for funding. We also thank
Xiang Liu, David Thomas, Patrick Zhou, Esther Ichugo, Matt
Table 4 Results of coalescent simulation study
Simulation model
Marker
AKT AP CP
Divergence time (in generations) without migration
129 000 0.000 0.000 0.000
6300 0.023 0.008 0.036
900 0.992 0.379 0.394
Divergence time (in generations) with migration
129 000 0.001 0.001 0.006
6300 0.028 0.013 0.034
900 0.839 0.438 0.422
No divergence
High N
e
0.122 0.148 0.546
Low N
e
0.065 0.181 0.235
P-value for each model was obtained by comparison of either
minimum (divergence) or maximum (no divergence) empirical
svalue (Slatkin & Maddison 1989) with simulated distributions
of sunder coalescent scenarios to test whether observed data
were consistent with divergence times discussed in text.
Simulations were based on assumed 20-year generation time.
2912 N. A. DOUGLAS ET AL.
2011 Blackwell Publishing Ltd
Cleary and Jacob Hilton for assistance with laboratory work
and Marshall Wilson, Tom Phillips, Mac Alford, Rob Naczi,
Gary Shurette, Viola Walker, Emil Devito, Heather Sullivan,
Paul Manos, John Pogacnik, James Smith, Wayne Longbottom,
Misty Buchanan and others for assistance locating Lilium popu-
lations.
References
Axelrod DI (1972) Edaphic aridity as a factor in angiosperm
evolution. The American Naturalist,106, 311–320.
Ayling SC, Brown TA (2008) Novel methodology for
construction and pruning of quasi-median networks. BMC
Bioinformatics,9, 115.
Ayres DR, Ryan FJ (1999) Genetic diversity and structure of
the narrow endemic Wyethia reticulata and its congener W.
bolanderi (Asteraceae) using RAPD and allozyme techniques.
American Journal of Botany,86, 344–353.
Baldwin BG (2005) Origin of the serpentine-endemic herb Layia
discoidea from the widespread L. glandulosa (Compositae).
Evolution,59, 2473–2479.
Baskauf CJ, McCauley DE, Eickmeier WG (1994) Genetic
analysis of a rare and a widespread species of Echinacea
(Asteraceae). Evolution,48, 180–188.
Bittkau C, Comes HP (2009) Molecular inference of a Late
Pleistocene diversification shift in Nigella s. lat.
(Ranunculaceae) resulting from increased speciation in the
Aegean archipelago. Journal of Biogeography,36, 1346–1360.
Braun EL (1955) The phytogeography of unglaciated Eastern
United States and its interpretation. The Botanical Review,21,
297–375.
Brooks RR (1987) Serpentine and its Vegetation: A
Multidisciplinary Approach. Dioscorides Press, Portland, OR.
Brunsfeld SJ, Sullivan J, Soltis DE, Soltis PS (2001) Comparative
phylogeography of northwestern North America: a
synthesis. In: Integrating Ecology and Evolution in a Spatial
Context (eds Silvertown J, Antonovics J), pp. 319–340.
Blackwell Science, Oxford.
Burnham KP, Anderson DR (2002) Model Selection and
Multimodel Inference: A Practical Information-Theoretic
Approach. Springer, New York.
Cabe S, Nickerson JG, Hoffman CW, Farrell K (1992)
Cretaceous and Tertiary stratigraphy of the northern
Sandhills area, North Carolina. In: Geologic field guides to
North Carolina and vicinity (eds Dennison J, Stewart KG),
Annual Meeting of the Southeastern Section of the
Geological Society of America. Winston-Salem, NC:
Geological Society of America, Boulder, CO.
Carstens BC, Stoute HN, Reid NM (2009) An information-
theoretical approach to phylogeography. Molecular Ecology,
18, 4270–4282.
Clement M, Posada D, Crandall KA (2000) TCS: a computer
program to estimate gene genealogies. Molecular Ecology,9,
1657–1659.
Cooper EA, Whittall JB, Hodges SA, Nordborg M (2010)
Genetic variation at nuclear loci fails to distinguish two
morphologically distinct species of Aquilegia.PLoS ONE,5,
e8655.
Crawford DJ, Ornduff R, Vasey MC (1985) Allozyme variation
within and between Lasthenia minor and its derivative
species, L. maritima (Asteraceae). American Journal of Botany,
72, 1177–1184.
Doyle JJ, Doyle JL (1987) A rapid DNA isolation procedure for
small quantities of fresh leaf tissue. Phytochemical Bulletin,19,
11–15.
Drummond A, Rambaut A (2007) BEAST: Bayesian
evolutionary analysis by sampling trees. BMC Evolutionary
Biology,7, 214.
Edwards SV, Beerli P (2000) Perspective: gene divergence,
population divergence, and the variance in coalescence time
in phylogeographic studies. Evolution,54, 1839–1854.
Edwards CE, Soltis DE, Soltis PS (2006) Molecular phylogeny
of Conradina and other scrub mints (Lamiaceae) from the
southeastern USA: evidence for hybridization in Pleistocene
refugia? Systematic Botany,31, 193–207.
Ellstrand NC, Elam DR (1993) Population genetic consequences
of small population size: implications for plant conservation.
Annual Review of Ecology and Systematics,24, 217–242.
Excoffier L, Lischer HEL (2010) Arlequin suite ver 3.5: a new
series of programs to perform population genetics analyses
under Linux and Windows. Molecular Ecology Resources,10,
564–567.
Fay JC, Wu CI (2000) Hitchhiking under positive Darwinian
selection. Genetics,155, 1405.
Freeland JR, Biss P, Conrad KF, Silvertown J (2010) Selection
pressures have caused genome-wide population
differentiation of Anthoxanthum odoratum despite the
potential for high gene flow. Journal of Evolutionary Biology,
23, 776–782.
Gaut BS (1998) Molecular clocks and nucleotide substitution
rates in higher plants. Evolutionary Biology,30, 93–120.
Gitzendanner MA, Soltis PS (2000) Patterns of genetic variation
in rare and widespread plant congeners. American Journal of
Botany,87, 783–792.
Godt MJW, Hamrick JL (1993) Genetic diversity and
population structure in Tradescantia hirsuticaulis
(Commelinaceae). American Journal of Botany,80, 959–966.
Goudet J, Raymond M, de-Meeus T, Rousset F (1996) Testing
differentiation in diploid populations. Genetics,144, 1933–
1940.
Gugger PF, Sugita S, Cavender-Bares J (2010) Phylogeography
of Douglas-fir based on mitochondrial and chloroplast DNA
sequences: testing hypotheses from the fossil record.
Molecular Ecology,19, 1877–1897.
Hamrick JL, Godt MJW (1990) Allozyme diversity in plant
species. In: Plant Population Genetics, Breeding, and Genetic
Resources (eds Brown AHD, Clegg MT, Kahler AL, Weir BS),
pp. 43–63. Sinauer Associates, Sunderland, MA.
Harrison S, Damschen E, Going BM (2009) Climate gradients,
climate change, and special edaphic floras. Northeastern
Naturalist,16, 121–130.
Hey J, Nielsen R (2004) Multilocus methods for estimating
population sizes, migration rates and divergence time, with
applications to the divergence of Drosophila pseudoobscura
and D. persimilis.Genetics,167, 747–760.
Hey J, Nielsen R (2007) Integration within the Felsenstein
equation for improved Markov chain Monte Carlo methods
in population genetics. Proceedings of the National Academy of
Sciences,104, 2785–2790.
Howarth DG, Baum DA (2005) Genealogical evidence of
homoploid hybrid speciation in an adaptive radiation of
ORIGIN OF LILIUM PYROPHILUM 2913
2011 Blackwell Publishing Ltd
Scaevola (Goodeniaceae) in the Hawaiian Islands. Evolution,
59, 948–961.
Huson DH, Bryant D (2006) Application of phylogenetic
networks in evolutionary studies. Molecular Biology and
Evolution,23, 254–267.
Karron JD (1987) A comparison of levels of genetic
polymorphism and self-compatibility in geographically
restricted and widespread plant congeners. Evolutionary
Ecology,1, 47–58.
Knowles LL (2009) Statistical phylogeography. Annual Review
of Ecology, Evolution, and Systematics,40, 593–612.
Kruckeberg AR (1986) An essay: the stimulus of unusual
geologies for plant speciation. Systematic Botany,11, 455–463.
Kruckeberg AR, Rabinowitz D (1985) Biological aspects of
endemism in higher plants. Annual Review of Ecology and
Systematics,16, 447–479.
Lande R, Barrowclough GF (1987) Effective population size,
genetic variation, and their use in population management.
In: Viable Populations for Conservation (ed. Soule
´ME), pp. 87–
124. Cambridge University Press, Cambridge.
Lande R, Engen S, Sæther B-E (2003) Stochastic Population
Dynamics in Ecology and Conservation. Oxford University
Press, Oxford.
Librado P, Rozas J (2009) DnaSP v5: a software for
comprehensive analysis of DNA polymorphism data.
Bioinformatics,25, 1451–1452.
Maddison WP, Maddison DR (2010) Mesquite: A Modular
System for Evolutionary Analysis. http://mesquiteproject.org.
Martin DP, Williamson C, Posada D (2005) RDP2:
recombination detection and analysis from sequence
alignments. Bioinformatics,21, 260–262.
Mehta BD, Jog SP, Johnson SC, Murthy PPN (2006) Lily pollen
alkaline phytase is a histidine phosphatase similar to
mammalian multiple inositol polyphosphate phosphatase
(MINPP). Phytochemistry,67, 1874–1886.
Nei M (1987) Molecular Evolutionary Genetics. Columbia
University Press, New York.
Nielsen R, Wakeley J (2001) Distinguishing migration from
isolation: a Markov chain Monte Carlo approach. Genetics,
158, 885–896.
Nishikawa T, Okazaki K, Uchino T, Arakawa K, Nagamine T
(1999) A molecular phylogeny of Lilium in the internal
transcribed spacer region of Nuclear ribosomal DNA. Journal
of Molecular Evolution,49, 238–249.
North Carolina Natural Heritage Program (2007) North Carolina
Natural Heritage Program Element Occurrence Database.NC
Department of Environment and Natural Resources, Raleigh,
NC, 27699–1610.
Nosil P, Funk DJ, Ortiz-Barrientos D (2009) Divergent selection
and heterogeneous genomic divergence. Molecular Ecology,
18, 375–402.
Oliveira LO, Huck RB, Gitzendanner MA, Judd WS, Soltis DE,
Soltis PS (2007) Molecular phylogeny, biogeography, and
systematics of Dicerandra (Lamiaceae), a genus endemic to
the southeastern United States. American Journal of Botany,94,
1017–1027.
Ornduff R (1965) Ornithocoprophilous endemism in Pacific
basin angiosperms. Ecology,46, 864–867.
Parsons RF (1976) Gypsophily in plants-a review. American
Midland Naturalist,96, 1–20.
Peruzzi L, Leitch IJ, Caparelli KF (2009) Chromosome
diversity and evolution in Liliaceae. Annals of Botany,103,
459–475.
Posada D (2008) jModelTest: phylogenetic model averaging.
Molecular Biology and Evolution,25, 1253–1256.
Rabinowitz D (1981) Seven forms of rarity. In: The Biological
Aspects of Rare Plant Conservation (ed. Synge H), pp. 205–217.
Wiley, New York.
Raduski AR, Rieseberg LH, Strasburg JL (2010) Effective
population size, gene flow, and species status in a narrow
endemic sunflower, Helianthus neglectus, compared to its
widespread sister species, H. petiolaris.International Journal of
Molecular Sciences,11, 492–506.
Rajakaruna N (2004) The edaphic factor in the origin of plant
species. International Geology Review,46, 471–478.
Rambaut A (1996) Se-Al: Sequence Alignment Editor. Available at
http://evolve.zoo.ox.ac.uk/.
Ramos-Onsins SE, Rozas J (2002) Statistical properties of new
neutrality tests against population growth. Molecular Biology
and Evolution,19, 2092–2100.
Raymond M, Rousset F (1995) An exact test for population
differentiation. Evolution,49, 1280–1283.
Rieseberg LH, Carter R, Zona S (1990) Molecular tests of the
hypothesized hybrid origin of two diploid Helianthus species
(Asteraceae). Evolution,44, 1498–1511.
Rozen S, Skaletsky H (2000) Primer3 on the WWW for general
users and for biologist programmers. Methods in Molecular
Biology,132, 365–386.
Schafale MP, Weakley AS (1990) Classification of the Natural
Communities of North Carolina, Third Approximation. North
Carolina Department of Environment, Health, and Natural
Resources, Division of Parks and Recreation, Natural
Heritage Program, Raleigh.
Shaw J, Lickey EB, Schilling EE, Small RL (2007) Comparison
of whole chloroplast genome sequences to choose noncoding
regions for phylogenetic studies in angiosperms: the tortoise
and the hare III. American Journal of Botany,94, 275–288.
Skinner MW (2002) Lilium. In: Flora of North America North of
Mexico. Magnoliophyta: Liliidae: Liliales and Orchidales (ed.
Flora of North America Editorial Committee), pp. 172–197.
Oxford University Press, Oxford.
Skinner MW, Sorrie BA (2002) Conservation and ecology of
Lilium pyrophilum, a new species of Liliaceae from the
Sandhills region of the Carolinas and Virginia, USA. Novon,
12, 94–105.
Slatkin M, Maddison WP (1989) A cladistic measure of gene
flow inferred from the phylogenies of alleles. Genetics,123,
603–613.
Sohl NF, Owens JP (1991) Cretaceous stratigraphy of the
Carolina Coastal Plain. In: The Geology of the Carolinas:
Carolina Geological Society Fiftieth Anniversary Volume (eds
Horton JW, Zullo VA, Carolina Geological Society), pp. 191–
220. Univ. of Tennessee Press, Knoxville, TN.
Soltis PS, Soltis DE (2003) Applying the bootstrap in phylogeny
reconstruction. Statistical Science,18, 256–267.
Soltis DE, Morris AB, McLachlan JS, Manos PS, Soltis PS (2006)
Comparative phylogeography of unglaciated eastern North
America. Molecular Ecology,15, 4261–4293.
Sorrie BA, Weakley AS (2001) Coastal plain vascular plant
endemics: phytogeographic patterns. Castanea,66, 50–82.
2914 N. A. DOUGLAS ET AL.
2011 Blackwell Publishing Ltd
Sorrie BA, Gray JB, Crutchfield PJ (2006) The vascular flora of
the longleaf pine ecosystem of Fort Bragg and Weymouth
Woods, North Carolina. Castanea,71, 129–161.
Stanford AM, Harden R, Parks CR (2000) Phylogeny and
biogeography of Juglans (Juglandaceae) based on matK and
ITS sequence data. American Journal of Botany,87, 872–882.
Stephens M, Donnelly P (2003) A comparison of Bayesian
methods for haplotype reconstruction from population
genotype data. American Journal of Human Genetics,73, 1162–
1169.
Stephens M, Smith NJ, Donnelly P (2001) A new statistical
method for haplotype reconstruction from population data.
American Journal of Human Genetics,68, 978–989.
Strasburg JL, Rieseberg LH (2008) Molecular demographic
history of the annual sunflowers Helianthus annuus and H.
petiolaris—large effective population sizes and rates of long-
term gene flow. Evolution,62, 1936.
Swofford DL (2002) PAUP*: Phylogenetic Analysis Using
Parsimony (*and Other Methods). Sinauer, Sunderland, MA.
Taberlet P, Fumagalli L, Wust-Saucy AG, Cosson JF (1998)
Comparative phylogeography and postglacial colonization
routes in Europe. Molecular Ecology,7, 453–464.
Tajima F (1989) Statistical method for testing the neutral
mutation hypothesis by DNA polymorphism. Genetics,123,
585–595.
Wall WA, Douglas NA, Xiang QY, Hoffmann WA, Wentworth
TR, Hohmann MG (2010) Evidence for range stasis during
the latter Pleistocene for the Atlantic Coastal Plain endemic
genus, Pyxidanthera Michaux. Molecular Ecology,19, 4302–
4314.
Watterson GA (1975) On the number of segregating sites in
genetical models without recombination. Theoretical
Population Biology,7, 256–276.
Weir BS, Cockerham CC (1984) Estimating F-statistics for the
analysis of population structure. Evolution,38, 1358–1370.
Wheelan SJ, Church DM, Ostell JM (2001) Spidey: a tool for
mRNA-to-genomic alignments. Genome Research,11, 1952–
1957.
White TJ, Bruns T, Lee S, Taylor J (1990) Amplifications and
direct sequencing of fungal ribosomal RNA genes for
phylogenetics. In: PCR Protocols: A Guide to Methods and
Applications (eds Innis M, Gelfand D, Sninsky J, White T),
pp. 315–322. Academic Press, San Diego, CA.
Whittall JB, Syring J, Parks M et al. (2010) Finding a (pine)
needle in a haystack: chloroplast genome sequence
divergence in rare and widespread pines. Molecular Ecology,
19, 100–114.
Williamson PS, Bazeer SK (1997) Self-incompatibility in Abronia
macrocarpa (Nyctaginaceae). Southwestern Naturalist,42, 409–
415.
Woerner AE, Cox MP, Hammer MF (2007) Recombination-
filtered genomic datasets by information maximization.
Bioinformatics,23, 1851–1853.
Zonneveld BJM, Leitch IJ, Bennett MD (2005) First nuclear
DNA amounts in more than 300 angiosperms. Annals of
Botany,96, 229–244.
Zwickl DJ (2006) Genetic algorithm approaches for the
phylogenetic analysis of large biological sequence datasets
under the maximum likelihood criterion. http://code.google.
com/p/garli/.
N.A.D. is an evolutionary biologist who focuses on phylo-
genetics and phylogeography of arid-adapted and edaphic
endemic plants. W.A.W. is a plant ecologist interested in the
interplay between population genetics and population dyna-
mics of rare and endemic species. Q.Y.X.’s research explores the
mechanisms underlying biodiversity patterns and morphologi-
cal variation, with emphasis on the Cornales. W.A.H. is inter-
ested in ecology and conservation of savanna ecosystems.
T.R.W. is a vegetation scientist with research interests in vegeta-
tion ⁄environment relationships and biological diversity. J.B.G.
is a botanist working on conservation of rare plants in Coastal
Plain habitats. M.G.H. uses diverse approaches to inform rare
species conservation.
Data accessibility
Sample and haplotype information is found in Table S1 (Sup-
porting Information). DNA sequences: GenBank accessions
JF829316–JF829423 (Table S2, in Supporting Information). ITS
data and phylogenetic tree available at http://purl.org/phylo/
treebase/phylows/study/TB2:S11519.
Supporting information
Additional supporting information may be found in the online
version of this article.
Table S1 Sampling. Haplotype numbers correspond to sample
labels in Figs 3 and 4, and to names accessioned in GenBank
(Table S2, in Supporting Information). For AP and AKT
sequences, integers identify haplotypes recovered more than
once in this study and other identifiers refer to unique haplo-
types
Table S2 GenBank accession numbers. Haplotype names corre-
spond to samples in Table S1 (Supporting Information)
Please note: Wiley-Blackwell are not responsible for the content
or functionality of any supporting information supplied by the
authors. Any queries (other than missing material) should be
directed to the corresponding author for the article.
ORIGIN OF LILIUM PYROPHILUM 2915
2011 Blackwell Publishing Ltd
