Enhancing the survival and expansion potential of a founder population through clonality


Author for correspondence:
Maurizio Rossetto
Tel: +61 2 9231 8337
Email: maurizio.rossetto@rbgsyd.nsw.gov.au


  • Although founder populations often have low diversity, they can potentially serve as stepping stones for further colonization, as refugia during nonoptimal times and as a source of specialized adaptive potential. The demonstration of such potential within natural plant populations has proven to be particularly difficult. Our investigation into a geographically disjunct population of a heterostylous shrub, Erythroxylum pusillum, aims to explore the evolutionary and ecological consequences of being an isolated founder population.
  • Microsatellite-based analyses were used to find evidence for, and trace the origins of, a severe founder effect. Molecular and spatial evidence was used to quantify clonality and to discover proof of somaclonal mutations.
  • We describe the unprecedented case of an isolated population that persisted through historical environmental fluctuations and in marginal habitat through vegetative spread, and is counteracting the lack of sexual recombination and gene flow through somatic mutation.
  • Our findings advance our understanding of how founder populations survive, differentiate and evolve. They also have implications for how conservation agencies should perceive and manage previously considered ‘dead-end’ populations.


Whether small isolated populations are the product of dispersal or range contraction, they have been subject to the stochastic genetic processes evident of founder effects. In general, founder populations (populations that originated from a small sample of individuals as a result of either a bottleneck or a dispersal event) have a reduced amount of genetic variation relative to the ancestral population, and both selection and drift can shape their genetic make-up (Frankham et al., 2002). The loss of alleles and polymorphisms caused by drift and isolation can, over time, lead to increased inbreeding rates and potentially culminate in a complete loss of heterozygosity. In small populations, inbreeding can have a severe impact on fitness, whereas the complete absence of migration can increase the risk of stochastic extinction and reduce the potential for adaptive response (Oostermeijer et al., 1995; Young et al., 1996; Frankham et al., 2002). Genetic diversity and vigour can be regained in plant populations through seed and pollen dispersal, sexual recombination (including hybridization) and random mutations. Although migration and recombination have been the subject of countless studies, random mutation events are difficult to detect and, consequently, are often neglected in natural population studies in spite of their theoretical relevance (Kingman, 1982).

Despite the risks associated with being small and disconnected, founder populations can contribute significantly to the spatial dynamics of species by serving as stepping stones for further successful colonization or by persisting in refugia prior to range re-expansion. Genetically isolated founder populations may also lead to speciation through divergence from source lineages (Wright, 1932; Mayr, 1954; Carson, 1968; Templeton, 2008). Such evolutionary successful outcomes from a limited starting pool present interesting questions about the mechanisms involved in species’ survival and expansion following severe founder effects.

A remarkable example of a founder population is illustrated by Erythroxylum pusillum ms. at Gove in Northern Territory (NT, Australia). A single population, comprising < 350 plants, was discovered within a mining lease at Gove, c. 500 km from the core distribution of this Australian endemic species on the Cape York Peninsula, Queensland (CYP). Lying between the two disjunct distributions is a shallow epicontinental sea, known as the Gulf of Carpentaria (Fig. 1). On the basis of the recent geological history of the region, it is unclear whether the current disjunct distribution of E. pusillum ms. is a result of vicariance (as a consequence of range contraction) or long-distance dispersal. Palaeoenvironmental studies have indicated that the Gulf region underwent extended periods of connectivity to CYP between 36 000 and 11 000 bp when Lake Carpentaria experienced periods of subaerial exposure and transgression to saline environments (Smart, 1977; Torgersen et al., 1988; Chivas et al., 2001). During the height of the arid period corresponding to the Last Glacial Maximum (18 000 to 14 000 bp), the lake dried-up, forming almost continuous vegetated areas between NT and CYP (Chivas et al., 2001). With the rise in sea level at the start of the Holocene (c. 12 000 bp), the Gulf flooded again, and present sea levels have probably existed since 7000 bp (Smart, 1977).

Figure 1.

 Distribution map for Erythroxylum pusillum ms. and location of study sites (inline image, species’ distribution; inline image, genetic sample point).

According to Baker’s law, successful colonization after dispersal is more likely with self-compatible than self-incompatible species (Baker, 1955; Stebbins, 1957). This is because fewer propagules need to be dispersed successfully by self-compatible species to ensure successful establishment after the initial event. However, plants can overcome the absence of a suitable sexual partner through vegetative reproduction or apomixis (Baker, 1955, 1967). Within Erythroxylum, a range of breeding systems, from self-compatibility to complete morph incompatibility, and including apomixes, have been recorded (Berry et al., 1991; Pailler et al., 1998; Ganders, 2008; Rosas & Domínguez, 2008). Although little is known about the specific inter- and intramorph compatibilities within E. pusillum ms., we know that it is heterostylous, producing pin and thrum flowers, and thus likely to be a preferential outcrosser. The distribution patterns of Erythroxylum species on La Réunion have led researchers to conclude that heterostyly did not necessarily hinder successful colonization after dispersal (Pailler et al., 1998); however, the search for compatible mates can be a challenge in small populations of obligate outcrossers (Elam et al., 2007). As only thrum flowers and no fruits, as well as root suckering (not detected at CYP), have been observed at the Gove site, low diversity and reduced sexual reproduction can be expected. In support of this expectation, in a preliminary genetic survey, we found evidence of clonality at Gove (van der Merwe et al., 2009).

Clonality has often been associated with marginal populations (Kawecki, 2008; Silvertown, 2008), suggesting that vegetative spread could play a role in the long-term persistence and survival of E. pusillum ms. at Gove. Optimal range structure is primarily determined by a species’ niche and its interspecific interactions (Holt & Keitt, 2005), and it is these factors that ultimately determine the fate of peripheral populations. For example, it has been posited that the benefits of sexual reproduction may be overridden by the pressures of geographical and genetic isolation, and that the suboptimal environmental conditions experienced by marginal populations can cause a shift towards asexual reproduction and clonality (Ouborg et al., 1999; Eckert, 2001; Billingham et al., 2003; Arnaud-Haond et al., 2006; Beatty et al., 2008). Consequently, it can also be expected that populations will be genetically impoverished when found within habitats that are both geographically and ecologically marginal (Johannesson & Andre, 2006).

In this study, we aim to investigate the ecological and evolutionary consequences of being a highly disjunct founder population by taking advantage of the unique opportunity presented by E. pusillum ms. at Gove. Although several species from different families (that include a range of breeding systems and life forms) show distribution patterns on either side of the Gulf of Carpenteria, none of these exhibit such a restricted distribution on one side of the Gulf. In this study, we aim to gain important empirical data on the evolution of disjunct populations of dimorphic plants, to track the evolutionary trajectory of a highly localized, natural founder event, and to use this information in support of relevant management strategies.

In particular, this study focuses on the following questions. Can we find genetic evidence of a founder effect? What are the genetic consequences of drift in a small population of an obligate outcrosser? How has such a small population survived in isolation on marginal habitat?

Materials and Methods

Study species

Erythroxylum (the primary genus in Erythroxylaceae) is a large genus with over 230 species of tropical and subtropical trees and shrubs, mostly distributed across America (180 species) and Madagascar (Mabberley, 2008). The South American E. coca, from which cocaine is extracted, is probably the best known member of the genus. The six species recognized in Australia are mostly confined to tropical and subtropical areas. Erythroxylum sp. ‘Cholmondely Creek’ (J. R. Clarkson 9367) is the smallest and also the only species in Australia in which heterostyly has been recorded. We refer to the species as Erythroxylum pusillum ms., as proposed in a revision by J. R. Clarkson (unpublished). This small subshrub barely reaches a height of 30 cm, has inconspicuous white flowers and bright red fruit. Apart from the small population discovered in 2007 as part of a floristic survey at Gove in north-eastern Arnhem Land (NT), the species is only found in the Cape York Peninsula in Queensland (CYP; Fig. 1). As the Gove population is highly disjunct and the only one known to be present within the Northern Territory (extensive searches in surrounding areas failed to discover further populations), it was deemed to be of conservation value, and the development of a sound management plan was instigated. The soils at Gove are rich in bauxite, and much of the area forms part of a mining lease agreement, placing this site under pressure from anthropogenic development.

Study site, spatial mapping and sampling

The population of E. pusillum ms. at Gove covers a very small area (0.068 ha). At this site, a pilot study found evidence of clonality (but not in the CYP populations). As a result, the collection of material was designed to ensure the maximum sampling of genetic diversity with even spatial sampling across the population. The overall objective was to identify the level of genetic differentiation between individuals, as well as their relative spatial distribution. A 75 m × 75 m plot was established across the site, approximately centred on the core of the population. This was divided into 375 contiguous 1.5 m × 1.5 m quadrats. Searches were undertaken for E. pusillum ms. within each quadrat. In quadrats that contained the species, the most central ramet was tagged using pin tags placed to the immediate west of the stem, and its location was marked using a Leica (Heerbrugg, St. Gallen, Switzerland) GPS1200 high precision GNSS receiver, which utilized a real-time kinematic system. The accuracy for the pick-up points averaged 3.70 ± 0.10 cm. The population was found to occupy an area of c. 680 m2 centred on the coordinate −12°15′0.0504″S, 136°50′38.2014″E. The species was recorded in 317 of the 1.5 m2 cells (Fig. 2) at the Gove site. From each tagged plant, one to four leaves were collected and used for DNA extraction. DNA was extracted from all samples using the Qiagen Tissue Lyser and Qiagen DNeasy 96 Plant Kit following the manufacturer’s protocols.

Figure 2.

 Spatial distribution of the multi-locus genotypes observed with six microsatellite loci at Gove (NT). The inset represents a network showing the relationship between the multi-locus genotypes observed with six microsatellite loci at Gove.

Collections from CYP aimed at sampling representative populations from the entire distribution of the species. The inclusion of samples from the core distribution of the species enabled us to investigate the relationship between the two disjunct distributions, to understand the population structure and gene flow across the entire species and to place Gove within its evolutionary context. At CYP, E. pusillum ms. occurs as a large metapopulation around the Weipa area, with more scattered populations occurring further away (Fig. 1; Table 1). As there were no previous indications of clonality at CYP, collections were less spatially structured than at Gove.

Table 1.   Distribution and genetic diversity details for the Erythroxylum pusillum ms. populations investigated in this study, including sample size (N), number of alleles (Na), number of effective alleles (Ne), observed and expected heterozygosities (Ho, He), inbreeding coefficient (FIS), number of genotypes (G) and genotypic diversity (R)
  1. aValues in parentheses after correcting for multi-locus lineages (MLLs).

  2. bSignificant heterozygous deficit.

  3. cSignificant heterozygous excess.

Heathlands area−11.7863142.48236156.1673.4480.590.640.060151
Weipa 1−12.6731141.8883175.8332.6970.480.550.108171
Weipa 4−12.61531142.12757136.3333.4530.590.660.077131
Weipa 5−12.48291141.89474165.0003.0320.560.610.046161
Batavia−12.86367142.73773155.3333.4860.450.610.228b130.86 (0.71)
Archer−13.41943142.93638205.6673.4830.630.64−0.003201 (0.95)
Gove genotypes  12 2.3180.970.57−0.765c  

Molecular analyses

Six microsatellite loci (ery1, ery2, ery3, ery6, ery7, ery8), previously shown to be variable within the sexual population of E. pusillum ms. at Archer (van der Merwe et al., 2009), were used to investigate genetic diversity within the sampled populations. These markers have been shown to be suitable for population-level studies of clonal plants (Halkett et al., 2005; Arnaud-Haond et al., 2007a,b).

Two sets of three microsatellite loci each were amplified simultaneously using the Qiagen Multiplex PCR Kit in 10 μl volumes, each containing 5 μl 2× Qiagen Multiplex PCR Master Mix, 0.05 μm of each forward primer, 0.2 μm of each reverse primer, 0.3 μm of M13 forward primer (with fluorescent dye attached) and c. 9 ng of DNA. Reactions were carried out under the following conditions: one cycle of 95°C for 15 min; 30 cycles of 94°C for 30 s, 60°C for 90 s, 72°C for 60 s; eight cycles of 94°C for 30 s, 53°C for 90 s, 72°C for 60 s; one cycle of 60°C for 30 min. PCR multiplexes were combined, enabling the genotyping of all loci in a single run on an ABI 3730 Capillary Sequencer (Applied Biosystems) after the addition of GeneScan™ 500 LIZ™ Size Standard (Applied Biosystems, Carlsbad, California, USA) at the Ramaciotti Centre for Gene Function Analysis (University of New South Wales). GeneMapper version 3.7 (Applied Biosystems) was used to investigate and score peak sizes amplified at all loci.

To verify the genotypes of ramets at Gove, PCRs for 30 randomly selected individuals were run twice. In addition, for those samples with a genotype assigned to four or less ramets, PCRs were also duplicated. Duplicate PCRs were also conducted for 12 randomly selected samples from CYP. If results differed between runs, samples were run a third time for confirmation. All but two duplicates were identical to the initial run.

In order to further investigate the long-term connectivity between CYP and Gove, the trnL–trnF region of the chloroplast was sequenced for four samples from each of the CYP populations and one sample from each of the multi-locus genotypes (MLGs) observed at Gove. Primers c and f (Taberlet et al. 1991) were used in PCRs carried out in 10 μl reactions with 2 mM of each deoxynucleoside triphosphate (dNTP), 25 mM MgCl2, 10 μm of each primer, 2 μg bovine serum albumin, 0.5 U of BIOTAQ™ (Bioline, Alexandria, Australia), 10 ng of each DNA and 10× PCR buffer (supplied with the polymerase).

Measurement of genetic diversity

The number and effective number of alleles, expected and observed heterozygosity, and fixation index were calculated using FSTAT2.9.3 (Goudet, 1995). FIS was calculated per locus and per sample, and significant deficit/excess of heterozygotes was also estimated using FSTAT. GenClone2.0 (Arnaud-Haond & Belkhir, 2007) was used to assign ramets to genotypes. We estimated the genotype diversity, R = (G−1)/(N−1), where G is the number of genotyopes and N is the number of ramets, following the method of Arnaud-Haond et al. (2007a,b). We calculated Hardy–Weinberg (HWE) and linkage equilibria for each locus and each population using GenePop1.2 (Raymond & Rousset, 1995). A strong deviation from HWE would indicate the absence of segregation, whereas linkage disequilibrium would indicate the absence of recombination. Segregation and recombination are both important consequences of sexual reproduction (Tibayrenc et al., 1990).

Multi-locus lineages and the spatial distribution of genets

To fully understand the extent of genetic diversity and the extent of clonality in an organism capable of vegetative spread, it is important to distinguish between genotypes that have arisen through sexual recombination and those that are the product of somatic mutation. Arnaud-Haond et al. (2007a) distinguished between these two using the term ‘multi-locus lineage’ (MLL) for MLGs that share the same common parent material, but differ from each other as a result of somatic mutations. Establishing the number of MLLs in a population can help us to understand the extent to which somatic mutations contribute towards total genetic variation within a population. In the absence of sexual recombination and migration, somatic mutation is the only means of introducing new genetic combinations into a population.

MLGs belonging to the same MLL will differ from each other by a very small number of alleles, and a very small number of repeats would differentiate these alleles. We used GenClone2.0 to investigate the frequency of the number of allelic differences between genotypes within a population (i.e. for six loci, the maximum number of allelic differences between two genotypes is 12). GenClone2.0 was then used to calculate the probability (Psex) that two samples with the same MLG in a population originated from a distinct reproductive event using a binomial expression (Tibayrenc et al., 1990; Parks & Werth 1993; see Arnaud-Haond et al., 2007a,b). Psex uses the probability of occurrence of a given genotype, given the number of loci, the frequency of each allele and the number of heterozygous loci in the sample, to estimate the probability that the repeated genotype originated through a distinct sexual event (thus a low probability score, Psex < 0.01, supports clonal origins for genetically identical ramets).

To determine whether small genotypic differences between MLGs were a consequence of sexual recombination or somatic mutation, we first screened all populations for MLGs that differed by a single allele. We then removed the distinguishing locus across the individuals belonging to these unusually similar MLGs (so that their genotypes could be scored as identical) and re-estimated Psex. A Psex score of < 0.01 would indicate that the null hypothesis of these ramets originating through distinct sexual recombination events could be rejected and, consequently, that these MLGs could be grouped within the same MLL (Arnaud-Haond et al., 2007a). The grouping of MLGs into MLLs using this method supports the occurrence of clonality.

Measurement of genetic differentiation

GenAlEx6.1 (Peakall & Smouse, 2005) was used to perform a principal co-ordinate analysis in order to investigate the clustering between individuals and populations. In order to investigate the underlying genetic structure in E. pusillum ms., a model-based clustering algorithm, implemented in Structure2.3.1 (Pritchard et al., 2000; Falush et al., 2003), was used with and without the Gove population (as the inclusion of clonal individuals negates the assumption of equilibrium). A series of independent runs for each value of K (the number of populations) between one and five was conducted with 4 × 106 iterations, following a burn-in of 150 000 iterations. The lengths of the burn-in and iterations were determined by running a series of trial runs using different increasing numbers of iterations. We used the default parameter settings in Structure2.3.1 for the ‘admixture’ and ‘no admixture’ models. The optimal number of clusters was verified using the ΔK statistical approach suggested by Evanno et al. (2005). Finally, an analysis of molecular variance (AMOVA) was performed to test the significance of the genetic structure identified with the other tests (Excoffier et al., 1992).

A Mantel test was performed in GenAlEx6.1 to investigate the relationship between genetic and geographical distances between the populations with and without the inclusion of Gove.

Environmental envelope of E. pusillum ms.

We used Maxent3.32 (Phillips et al., 2006; Phillips & Dudik, 2008) to generate a predictive distribution model based on a suite of climatic variables and locality data from CYP to ascertain whether conditions in the Gove peninsula fall within the environmental envelope of E. pusillum ms. at CYP. We selected Maxent as our modelling platform because comparative studies have found it to be one of the best performing models (Tognelli et al., 2009), consistently outperforming those using traditional linear methods (Elith & Leathwick, 2009), and it provides robust performance with smaller sample sizes and presence-only data (Pearson et al., 2006; Elith & Leathwick, 2009; Kumar & Stohlgren, 2009; Leachéet al., 2009), such as the dataset available for our study.

Our location data consisted of Queensland Herbarium records (= 26) and CYP genetic samples (= 7). Sample coordinates from Gove were excluded from the dataset in order to test whether the model predicted the locality as having conditions similar to those found at the core of the species’ distribution. We used seven environmental coverage variables to model the distribution of E. pusillum ms. The layers were related to precipitation and temperature. These variables were downloaded from WorldClim (http://www.worldclim.org), and represent a set of global climate layers derived from interpolated climatic averages (Hijmans et al., 2005). We used the 30 arc-seconds database, which equates to c. 1 km2 cells.

For the optimization of the model, all 19 Bioclim variables from Wordclim’s dataset were initially considered. In order to assess the repeatability of the results, 29 runs were conducted using all variables; all variables that did not contribute an average of > 1% across all models were excluded. Seven variables were used for the final prediction (Supporting Information Table S2). We also performed runs utilizing Maxent’s jackknife feature in order to assist in assessing the individual importance of each variable. For our final model, we used Maxent’s default settings, except for setting aside a test percentage, in order to provide an assessment of the accuracy of niche model predictions. Locality data were randomly partitioned and 60% were set aside for training, with the remaining 40% being used to test the model.

The area under the curve (AUC) of a receiver operating characteristic (ROC) plot was used to evaluate model performance. This threshold-independent statistic has been widely used for model evaluation (Fielding, 2002; Elith & Leathwick, 2009), and is part of the Maxent output. When used with presence-only data, the maximum theoretically achievable value is < 1.


A total of 74 alleles was detected at the six loci assessed among the 428 samples of E. pusillum ms. The number of alleles per locus ranged between nine (ery8) and 22 (ery1), and the total allelic richness per population (sum of the allelic richness of each locus divided by the number of loci), for corrected sampling, ranged between 3.67 (Gove) and 6.27 (Weipa 4). Pairwise tests of genotypic disequilibrium showed significant linkage between ery1 and ery6 at Batavia. At Gove, we found significant linkage between all pairs of loci, as expected for a highly clonal population (Tibayrenc et al., 1990; Halkett et al., 2005; De Meeûs et al., 2006). Excluding the population at Gove, all loci were in HWE after Bonferroni corrections. At Gove, the extreme negative FIS value approaching −1 (Table 1) gives further support to extensive clonality (Balloux et al., 2003).

Genetic diversity: Gove vs CYP

The number of alleles, effective number of alleles, genotypic diversity (R) and fixation index were all found to be lowest at Gove (Table 1). Although the observed heterozygosity at the CYP populations ranged between HO = 0.45 and HO = 0.63, at Gove it was HO = 0.99 (with HE = 0.54). This is because over 315 of the genotyped ramets at Gove were heterozygous at all six loci, with only two of the individuals being homozygous at only one locus (ery1). Gove was the only population in which a significant excess of heterozygotes was observed, with Batavia the only population with a significant excess of homozygotes.

In five of the eight populations, a genotypic diversity of unity was observed. The Gove population contrasted dramatically with CYP, with only 12 genotypes identified from the 317 samples using six microsatellite loci. Locus ery1 showed most variation at the CYP and Gove populations, whereas three of the loci that were variable at the CYP populations (ery3, ery6 and ery7) showed no divergence amongst the Gove samples. Two genotypes dominated at Gove (A with 60 ramets and E with 214 ramets) with the other genotypes occurring in numbers of < 10 (Table S2; Fig. 2). Three genotypes consisted of single sampled ramets. No genotypes were shared between Gove and CYP.

Extent and distribution of clonality at Gove

The genotypes at Gove differed from each other by between one and three alleles (Table S2; Fig. 3). A dramatic contrast was observed between the CYP populations and Gove, as the genotypes at CYP differed from each other, on average, by seven alleles (varying between one and 11; Fig. 3). We used the GPS coordinates of each sampled ramet at Gove to investigate the spatial distribution of genotypes at the site (Fig. 2). The fine-scale mapping revealed that the two dominant genotypes (A and E) were adjacent to each other, with A occupying the south-western corner of the site and E occupying most of the northern and eastern areas. The other genotypes were more or less nested amongst A and E. The three small ‘islands’ of plants towards the north-east and west of the population all belong to genotype E. Each genotype differs from its neighbour by only one allele, and none of the nondominant genotypes has more than one neighbouring genotype. Using the allelic information of each genotype and the spatial distribution of the genotypes, we constructed a simple network showing the most likely genetic relationship amongst the genotypes (inset to Fig. 2). Accordingly, the observation of allele sizes indicated that there was no specific direction (increase or decrease) of mutation in the size of the microsatellite.

Figure 3.

 Frequency of step-based allelic distances between genotypes observed at Gove (black) and CYP (white) based on six microsatellite loci.

The probability that the repeated genotypes at Gove originated through distinct sexual reproductive events in a population under HWE was estimated using Psex. The probability that genotype A (which occurred 66 times in the population) occurred more than five times as a result of distinct sexual reproductive origins was significantly low (Psex ≤ 0.004 with every encounter after the first five). For genotype E (which was found in 214 ramets), separate sexual recombination events could potentially explain the occurrence of nine of the re-occurrences, but the null hypothesis of separate sexual events could be rejected for further occurrences (Psex ≤ 0.008). For the genotypes with < 10 but more than two assigned samples (B, C, D and G), only two occurrences could be a result of separate sexual events, with a third sexual origin being highly unlikely (Psex < 0.01). Because of the low overall diversity within the population, the removal of ery1 from the dataset (which resulted in only three genotypes remaining, with 305, eight and two ramets, respectively) could not fully reject the possibility of sexual origins.

In summary, several lines of evidence support Gove as a clonal population (FIS, linkage between loci, absence of segregating genotypes and a high HO). In addition, the spatial clustering of observed MLGs (Fig. 2) appears to clearly track the spread and persistence of genotypes. Importantly, those that differed from each other by more than one allele were spatially well separated, enabling single-allele mutations between MLGs to be tracked on the ground. Together, this information suggests that all 12 MLGs may belong to one MLL. The nonsignificant Psex probabilities are most probably a consequence of low statistical power when dealing with a limited set of alleles, low clonal diversity and high clonal dominance (Arnaud-Haond et al., 2007a).

Clonality at CYP

At Batavia, we could reject the hypothesis that the two pairs of samples with similar genotypes originated from distinct sexual events (Psex = 0.007 and Psex = 0.00001). Three pairs of CYP samples shared genotypes and three genotypes (2.7%) differed from each other by only one allele. The elimination of distinguishing loci also provided no support for a sexual origin of these identical pairs (Psex = 0.0001). Thus, the 13 observed MLGs at Batavia could be grouped into 12 MLLs. A similar scenario was uncovered for the two identical samples at Heathlands (Psex = 0.0006), and the two genotypes differing by one allele at Archer (Psex = 0.007), suggesting that clonality is not completely restricted to Gove.

Divergence within and amongst CYP and Gove

Principal co-ordinate analysis indicated independent clustering between the Gove and CYP samples, but provided no evidence for any clustering amongst samples from single CYP locations. Pairwise FST values (Table 2) ranged between FST = 0.004 and FST = 0.091 (mean FST = 0.036) amongst CYP populations, indicating high levels of gene flow between geographical locations at CYP. Divergence between Gove and all CYP populations was considerably higher (mean FST = 0.204). No correlations between FST or genetic distance and geographical distance were found for the CYP populations, but a significant correlation was found with the inclusion of Gove in the Mantel test. The Bayesian assignment procedure implemented by Structure2.3.1 supported the entire CYP area as one single population. The inclusion of the 12 genotypes from Gove suggested the existence of two groups. This result is potentially affected by the violation of the assumption of equilibrium at Gove; however, interestingly, one of the two populations identified included all the Gove genotypes as well as a few samples from Batavia. AMOVA partitioned 69% of the variation within populations, with the remaining 31% partitioned amongst populations. Excluding Gove from the AMOVA, 95% of the genetic variation was partitioned to within-population variation.

Table 2. FST pairwise comparisons between populations of Erythroxylum pusillum ms. estimated from six microsatellite loci
 ArcherWeipa 1Weipa 4Weipa 5Heathl. AHeathlandsBataviaGove
Weipa 10.0450      
Weipa 40.0060.0350     
Weipa 50.0100.026−0.0050    
Heathl. A0.0260.0970.0160.0270   

TrnL–trnF sequence data revealed two chlorotypes for E. pusillum ms. One chlorotype was present at Gove and in two of the 28 individuals tested at CYP (both from the Batavia population). The second chlorotype was found across the remaining CYP individuals tested. The two Batavia samples that shared the Gove chlorotype were also included in the samples grouped with the Gove population by the Bayesian analysis of nuclear microsatellites.

Environmental conditions at CYP and Gove

Although the area covered by the Gove population is too small to unequivocally define its microhabitat, the regional environmental envelope was found to be different from that defined from the CYP dataset (Fig. S1). The probability of occurrence within the Gove Peninsula ranged from 0.007 to 0.009, which is considered to be very low. By contrast, the area of CYP in which E  pusillum is distributed exhibited a probability of occurrence of 0.495–0.759.

The calculated AUC values were 0.998 for training data and 0.997 for test data, indicating that the modelled distribution had a good ability to distinguish between occupied and unoccupied sites (a random AUC would be 0.5). The variable which made the highest percentage contribution to the model was the precipitation of the wettest quarter (Table S1). Using Maxent’s jackknife feature, this variable also usually displayed the highest predictive value or gain when used in isolation. Precipitation seasonality also made a significant contribution to the model output (Table S1). Interestingly, a small area around Melville Island and Darwin in NT was found to have an environment similar to that occupied by E. pusillum ms. on CYP, and may warrant future investigation.


Gove: isolation after an extreme founder effect

Our data present evidence of a population that has undergone a severe loss of diversity after a founder effect. Shared nuclear and chloroplast polymorphisms suggest relatively recent connectivity between the two distribution areas of E. pusillum ms., and that Batavia (which is also found in a drier habitat compared with the core distribution of the species, and where clonality appears to be present) could be a possible ancestral source for Gove. However, the genetic data also confirm that connectivity between these two distant regions has now been lost. On the basis of the available evidence, it is plausible that this relictual population originated before the complete flooding of the Gulf of Carpentaria which created a biogeographical barrier between NT and CYP. Assuming the presence of a single MLL at Gove, using the method described by Ally et al. (2008) and pertinent mutation rates (Cloutier et al., 2003), this MLL could date back to such a time (see Methods S1 for details on the dating method).

However, it is not possible to unequivocally characterize the founding mechanism. The high levels of gene flow between populations at CYP (mean FST = 0.036) indicate that genetic connectivity can be maintained across a large geographical area, and that E. pusillum ms. could have occupied an essentially continuous distribution between NT and CYP during the Last Glacial Maximum. With the rise in sea levels and flooding of the Gulf of Carpentaria during the Holocene, these two regions have become separated and, as environmental conditions changed dramatically within NT, the regional distribution of E. pusillum ms. was gradually eroded to one single individual at Gove. Yet, the hypothesis of a long-distance dispersal event occurring at any stage of landscape connectivity between NT and CYP cannot be excluded. This is particularly true when faster mutation rates are taken into account, and when considering that a single propagule can successfully establish a viable population.

The clonal strategy: surviving in marginal habitat

The absence of fruit and of segregating genotypes at Gove supports the expectation that this heterostylous species is an obligate outcrosser. The microsatellite data show that, although clonality may be present, sexual reproduction is the dominant form of reproduction within CYP, and that it is only at Gove that E. pusillum ms. is exclusively clonal. Furthermore, the spatial distribution of these clonal genotypes, the type of allelic differentiation and the absence of segregating genotypes reveal that they may all be part of as little as one MLL. A decrease in population size can skew the morph ratio in populations of self-incompatible species, such as heterostylous species (Brys et al., 2003; Elam et al., 2007), and have negative consequences on pollination success (Byers, 1995). The subsequent loss of sexual reproduction will increase the relative fitness of individuals capable of vegetative reproduction, and further reduce the opportunities for sexual reproduction. Ultimately, this will result in a population in which sexual recruitment is absent because of a lack of sexually compatible individuals (Rossetto et al., 2004; Bossuyt & Honnay, 2006).

A similar mechanism is likely to have resulted in the dominance of clonality at Gove. By excluding admixture with individuals unfit for the local environmental conditions, clonality could have simultaneously facilitated a genotype that is well adapted to a marginal habitat to breed true (Haag & Ebert, 2004). The combination of these two factors would have enabled E. pusillum ms. to overcome the sexual limitations of a dimorphic plant after a severe bottleneck or founder event. Clonal spread of a locally fit MLL is now critical for the persistence of this population, as opportunities to disperse and track suitable environmental conditions or to adapt through recombination are all but lost at Gove.

The predominance of clonal growth in peripheral populations has been described in several other species distributed across wide environmental gradients (e.g. Vaillancourt et al., 2001; Johannesson & Andre, 2006; Beatty et al., 2008; May et al., 2009). Compared with Gove, the Weipa area (which represents the core distribution of E. pusillum ms.) experiences significantly higher rainfall in its wettest quarter (paired t-test P = 0.01) and has an even more marked pattern of precipitation seasonality with a slightly drier dry season. The results from the modelling data suggest that several factors, including climatic, have interacted to constrain the niche space in which sexual reproduction occurs in E. pusillum ms., suggesting that clonality at Gove might have been an initial response to atypical environmental conditions.

Restoring diversity for long-term survival: persistence and somatic mutation

Regardless of the severity of the founder effect at Gove, we found evidence that, over time, E. pusillum ms. has persisted through vegetative growth and has increased its genetic diversity through somatic mutations. Although other studies have described the prevalence of clonality in peripheral populations, we provide unique empirical evidence that, even in the absence of sexual recombination, completely isolated individuals can accumulate genetic diversity through time. Although a purely clonal population has limited potential for the accumulation of adaptive genetic diversity, environmentally adaptive mutational changes have been recorded for clonal micoorganisms (Loxdale & Lushai, 2003). Furthermore, competition between somatic mutants can lead to diplontic selection and accelerate the accumulation of favourable alleles (McKey et al., 2010).

Although clonality has been said to be an evolutionary ‘dead-end’ (Maynard Smith, 1978; Kawecki, 2008), the ability of the Gove population to reproduce vegetatively has ensured the survival of E. pusillum ms. outside its current core distribution range for an extended period of time. In small isolated populations, one advantage of clonal spread over sexual recombination is the complete absence of random genetic drift. Furthermore, the two alleles from each locus act independently, and thus new mutations at both alleles are maintained. Over time, this combination will ensure complete heterozygosity (as shown at Gove), with the two alleles within one genet often more distant from each other than those found within two distinct genets (Meselson’s effect; Birky, 1996).

Although clonality is rarely thought of as a method of speciation because of the lack of sexual recombination (Coyne & Orr, 2004), it could be hypothesized that, over time, mutations could re-establish sexual compatibility, thus allowing for a combination of sexual and vegetative reproduction, and, eventually, evolutionary differentiation.

Conclusion: one individual can be enough for survival and expansion

After the initial loss of diversity (either through a severe bottleneck or a rare dispersal event), the Gove population has persisted through historical environmental changes, whilst maintaining complete heterozygosity and even increasing allelic diversity through somatic mutation. This study suggests that the entire population could have originated from as little as a single individual, and that clonal persistence is an important survival strategy for peripheral populations. We show that populations at the range edge are evolutionarily important as they may preserve rare alleles, and feature unique life-history traits that can be critical to establish long-term resilience to changing environmental conditions. From both an ecological and genetic standpoint, marginal populations should therefore be considered an important biological resource and thus integrated into conservation measures, even when very small. Dedicated management strategies must, however, consider the fact that, in spite of their capacity for long-term persistence, small clonal populations still remain highly vulnerable to catastrophic events.


We acknowledge John Westaway and John Clarkson for the collection of leaf material, Rohan Mellick and Joëlle Catherine for DNA extractions, and Sophie Arnaud-Haond and two anonymous reviewers for useful comments. This work forms part of an ongoing investigation into the Gove population, and is funded by RioTinto Alcan Gove.