The tension between dispersal and persistence regulates the current distribution of rare palaeo-endemic rain forest flora: a case study


Maurizio Rossetto (tel. + 61 29231 8337; fax: + 61 29251 4403; e-mail:


  • 1Eidothea hardeniana (Proteaceae) is a narrow endemic representative of an ancient lineage restricted to a single site in northern New South Wales (Australia). This study aims to identify the life-history traits most likely to have influenced the current distribution pattern of this rain forest tree.
  • 2Using ecological and molecular analyses we found that its range is limited by the absence of efficient dispersal mechanisms rather than by habitat availability, or as a result of a recent bottleneck or of reproductive failure.
  • 3Resprouts can take over the position of senesced plants, even in the absence of disturbance, allowing local persistence and the maintenance of genetic diversity. Long-term persistence and preferential outcrossing further favour relatively high levels of diversity (HE = 0.542) despite a small effective population size (Ne = 21.6).
  • 4We used eight life-history traits to assess if the E. hardeniana findings were valid across 258 local taxa. Current distribution patterns of rain forest species within a species-rich community were accounted for by linking two important components of community assembly theory: dispersal and niche assembly.
  • 5The interaction of the seed-based dispersal dimension and resprouting potential best explains the current distribution of rare local taxa. Major dimensions of life-history trait variation were identified among local plant species, suggesting that a range of interacting traits contribute to a species’ response to environmental variables and mitigate the influence of potentially adverse circumstances.
  • 6The benefits of merging ecological and genetic approaches to interpret species distribution and population structure are applicable across a broader range of studies. Our findings highlight how currently constrained palaeo-endemic species with small populations in refugial habitats may retain the capacity to both persist and expand in response to changing circumstances and opportunities. This has important implications for species conservation, habitat management and reserve design.


The rapidly changing climatic conditions that characterized the ice ages caused significant adjustments to species distributions worldwide. Some species adapted to the new conditions or had the opportunity to disperse to more suitable locations, while others became extinct or persisted only in suitable sections of their former range, which were sometimes reduced to small refugia (Hewitt 2000). These refugial populations had the opportunity to expand into newly available habitat as the climatic conditions improved during the interglacial periods. The success of these post-glacial expansions depended on a range of factors such as recruitment and dispersal potential, and played an important role in defining modern-day distribution patterns.

Geographically disjunct rain forest areas such as those of northern New South Wales (NSW, Australia) are likely to have remained relatively isolated even during more favourable climatic conditions. Such isolation, fragmentation and loss of connectivity have undoubtedly resulted in numerous range contractions and extinctions across the local flora and fauna (see Kershaw et al. 1991; Hill 2004). Despite this the rain forests of northern NSW are part of an important biodiversity region that includes representatives of both tropical and temperate floras (Burbidge 1960), with over 50 endemic genera and more than 200 species at their most southern or northern limits (McDonald & Elsol 1984). The overlap of temperate and tropical elements has significant implications for the patterns of distribution of rare, relictual and palaeo-endemic species and is discussed in more detail by Webb & Tracey (1981) and Webb et al. (1984).

The rare rain forest tree Eidothea hardeniana P.H. Weston & R.M. Kooyman (Proteaceae) is a palaeo-endemic species restricted to a single population and location in the upland cool simple notophyll vine forest (SNVF after Webb 1978) of northern NSW. This forest type is more generally known as warm temperate rain forest, and is often dominated by just a few canopy species from families such as Cunoniaceae, Lauraceae and Myrtaceae. The forest structure is simple, with only two tree strata (canopy and subcanopy), and features such as large vines and buttressed roots are rare or absent.

Palaeo-endemic species are usually described as ancient vestiges of previously widespread taxa, with current distributions that reflect increased constriction of their specialized habitat through time (Kruckerberg & Rabinowitz 1985). Eidothea has only two remaining species. The genus occupies a basal position in the subfamily Proteoideae and is likely to have originated over 100 million years ago (Hoot & Douglas 1998; Weston & Kooyman 2002). The fossil record shows that Proteaceae is one of the early Gondwanan families and that only moderate morphological change has occurred in this family during the last 55 million years (Hill 2004). This is despite the fact that during that time the Australian continent travelled through 20° in latitude and underwent major climatic changes. The morphological adjustments that did occur were gradual and not necessarily in response to changes in climatic conditions. For instance, it has been suggested that even the appearance of scleromorphy/xeromorphy in Proteaceae during the middle Eocene was a response to nutrient-deficient soils and/or higher light levels, rather than to reduced rainfall (Hill 2004).

A number of questions have arisen on the ecological and evolutionary processes affecting the unique flora of this region. These include questions related to the role and relative influence of dispersal potential, resprouting and clonality on the distribution, population structure and diversity present among some of the local palaeo-endemic rain forest trees (Rossetto et al. 2004a; Rossetto et al. 2004b). Because of its evolutionary origin, history and highly localized distribution, E. hardeniana represents a unique opportunity to investigate the factors affecting the current range and long-term survival of palaeo-endemic taxa. In this study we aim to use a range of approaches to investigate whether the current distribution of E. hardeniana relates to highly specialized habitat requirements or to a combination of life-history traits. This information can then be used to assess if the features and circumstances identified for this species are unique or if they are more broadly evident in the local flora. Such an approach should identify major dimensions of life-history trait variation that are useful for interpreting ecological patterns in rain forest species distribution.

Materials and methods

study species

Eidothea hardeniana has only been discovered, accurately identified and named recently (Weston & Kooyman 2002). It is a grey-barked tree up to 40 m high, often producing basal subsidiary shoots. Juvenile leaves are in characteristic false-whorls with spinose-dentate margins, while adult leaves have entire margins. Inflorescences are axillary or ramiflorous and occur singly or in groups of up to five. The 7–11 flowers in an inflorescence consist of a single hermaphrodite surrounded by male flowers. Flower to fruit ratios are low, with generally only one fruit produced from any group of inflorescences, and most inflorescences producing no fruits. Early results of pollination studies suggest that self-incompatibility mechanisms or severe inbreeding depression are likely to limit fruit production. Pollen movement appears to be mediated by small nocturnal beetles with a limited pollen dispersal range. Fruits are broad-ovoid drupes, up to 4 cm in diameter when mature, with a thin dull golden yellow exocarp and an extremely bitter 2–4 mm thick mesocarp. The hard-shelled endocarp is 2.5–4 mm thick. Fruits take over 1 year to fully develop and fall in late summer. Seed predators (rats) generally feed on and destroy > 90% of fallen seed (personal observation).

The species is restricted to a single population in notophyll vine forest on acid (silicic) volcanic soils within the Nightcap Range in northern NSW (hence the common name ‘Nightcap Oak’). The only known population covers an area of just over 2.5 km in length, divided into two sites (Fig. 1). Leaf material from all known individuals of E. hardeniana, including mature trees (106 stems > 10 cm d.b.h), seedlings and saplings (85) was sampled, and DNA was extracted using Qiagen DNeasy kits. As the age of individual stems is unknown, size classes have been used to allocate stems to mature and recruit (seedling and sapling) categories.

Figure 1.

Distribution of the Eidothea hardeniana plants (distance in metres). The open triangles represent adult individuals, the closed squares juveniles. Site A represents (and is surrounded by) an extensive area of relatively homogenous forest (SNMVF). Site B represents a more heterogeneous mix of forest types but still with large areas of apparently suitable habitat (contiguous and adjacent). The sites are separated by distinct topographical features (very steep land and cliffs).

habitat surveys and floristic data analysis

A total of 9.2 hectares (92 quadrats, each 50 × 20 m) was sampled in this study. A geographically representative but random subset of all known location records for E. hardeniana provided 22 quadrats. Habitat-based area searches and sampling associated with the study of another taxon that occurs in the same habitat type (Uromyrtus australis) provided a further 43 quadrats (R. M. Kooyman, unpublished observations). Extensive area searches, based on habitat matching with the known location records for the species, provided the remaining 27 quadrats. Available vegetation type maps were used to derive criteria for the searches and habitat matching, which were improved using data generated by the 22-quadrat sample. The searches targeted the specific rain forest types (Table 4) and adjacent ecotones (areas of interaction and overlap with adjacent vegetation communities), covered the whole of the available habitat area (154 km2), and included > 103 kilometres of walked traverses over 4 years of integrated studies.

Table 4.  Forest type descriptions representing broad floristic groupings derived from PATN analysis and classification dendrogram
Forest typeiiiiiiivvvivii
  • *

    Based on Webb (1978): NVF = notophyll vine forest; SNVF = simple notophyll vine forest; SNMVF = simple notophyll-microphyll vine forest; SNMVT = simple notophyll-microphyll vine thicket.

Ratio of plots with Eidothea0/88/227/130/87/250/70/9
Typology*NVFSNVFSNMVF to SNMVTSNVF- SNMVF disturbedSNVF- SNMVFSNVF with Caldcluvia and SchizomeriaEucalyptus spp. with SNVF- SNMVF
DescriptionComplex forest with emergents Ficus sp., Sloanea woolsiiDominated by Ceratopetalum/Endiandra/ CanariumIncludes Ceratopetalum/Endiandra with emergent Tristaniopsis collinaIncludes Acacia oritesWith or without Lophostemon confertus, Callitris macleayana or Syncarpiaglomulifera as emergentsDisturbed, with Acacia orites, gullies with Sloanea spp.Transitions and overlaps between forest communities. Fire influence

All vascular plant species that occurred within a plot were identified and recorded to species level. Cover codes, based on a cover abundance scale (modified Braun-Blanquet) were estimated for each species within the plot. Each site was ascribed to a structural/physiognomic community type based on Webb (1978). The eight environmental variables used for ordination analyses were derived from the field-collected environmental data. The relative values and allocated rankings of the environmental variables are described in Table 1.

Table 1.  Environmental variables and rankings used in data collection and subsequent analyses. Fire frequency was ranked using floristic and visual indicators and known fire history
Topographic positionFire frequency (year)DisturbanceSoil depthSoil typeAltitudeSlope*Aspect
  • *

    Aspect refers to the position of the plot relative to the localized relief of the site, and is measured by taking a compass bearing that indicates the direction of site drainage. Values were divided by 2 and arcsin transformed before ranking.

1. Crest1. 0–1001. None1. Skeletal1. Clay/loamActualActualActual
2. Upper slope2. 100–2502. Light2. Shallow2. Peaty/clay   
3. Mid-slope3. 250–5003. Moderate3. Medium3. Alluvium   
4. Lower slope4. > 5004. Heavy4. Deep    
5. Creek/gully5. No fire      

The quadrat data were entered into a matrix consisting of 92 sites (objects) and 258 species (attributes). Species cover codes (modified Braun-Blanquet) were entered as a cover abundance scale (1–6), where 1 = cover < 5% of site and rare, 2 = cover < 5% of site and common, 3 = cover 6–20% of site, 4 = cover 21–50% of site, 5 = cover 51–75% of site, and 6 = cover 76–100% of site. The floristic data were classified by the grouping of similar plots using a numerical hierarchical agglomerative classification process, utilizing the Bray–Curtis association measure and a flexible unweighted pair-group method using arithmetic averages (upgma) sorting strategy with beta = 0–0.1 (FUSE module of PATN 3.03; Belbin & Collins 2004). Broad floristic groupings were recognized at the 0.6 and 0.8 dissimilarity levels (Faith et al. 1987). Other interpretable groupings at lower dissimilarity levels were generally recognized as reflective of the influence of environmental factors on floristics. The ordinations of sites were calculated using hybrid multidimensional scaling (HMDS default; Belbin & Collins 2004). Ordination was used to investigate the patterns of floristic composition in relation to the environmental variables (Table 1) and to provide a visual representation of the relationship between sites, species and environmental correlates. All analysis evaluation tools provided in PATN 3.03 were selected.

genetic diversity and structure

Primer pairs amplifying five microsatellite (SSR) loci developed by Rossetto & Porter (2005) were used for the genetic study on E. hardeniana (Table 2). Forward primers were labelled with fluorescent dyes and polymerase chain reactions (PCRs) were performed as previously described in Rossetto & Porter (2005). Sizing of all products was obtained by capillary electrophoresis on an ABI 3700 Genetic Analyser (SUPAMAC, University of Sydney, Sydney, Australia) and the analysis of genotyping data was performed using Genotyper 3.7 (Applied Biosystems, Foster City, USA).

Table 2.  Locus name, expected size (in bp), repeat type and frequency (italicised repeat is imperfect), total number of alleles observed (AN), observed (HO) and expected (HE) heterozygosity and inbreeding coefficient (f) across E. hardeniana
Eh182bgt157(AG)2340.5180.558 0.070
Eh251bgt238(AG)1880.7960.801 0.007

Overall-mean number of alleles and observed/expected heterozygosity were calculated using GDA 1.1 (Lewis & Zaykin 2002). Weir & Cockerham's (1984) F statistics equivalents were calculated using FSTAT 2.9 (Goudet 1995), with the probability of f (FIS equivalent) being greater than zero determined after 10 000 permutations and 95% confidence interval determined after 15 000 bootstraps. Hardy–Weinberg and linkage disequilibria were assessed using GENEPOP 3.2a (Raymond & Rousset 1995) using the exact test and significance levels were determined after 500 batches of 5000 iterations each. An estimate of the effective population size (Ne) of E. hardeniana was obtained using temporal changes (comparing adults with seedlings) in allele frequencies (as described by Waples 1989). This estimate relies on stable populations (with changes potentially leading to underestimates of Ne) and assumes no mutations (a reasonable assumption for the two generations in the small E. hardeniana population) and no migration (there are no other E. hardeniana populations). This measure is more suited to small populations, although potential bias can occur if Ne ≤ 4 (Luikart et al. 1999).

In order to investigate potential differentiation between the two sites, the genetic variance of population structure was determined after 2000 permutations using RSTCALC 2.2 (Goodman 1997) to calculate ρ (an unbiased version of RST). Significance levels were determined after 2000 bootstraps. Spatial genetic structure analysis throughout the species’ distribution was tested using the microspatial autocorrelation analysis method of microsatellite data developed by Smouse & Peakall (1999). This multivariate approach strengthens the spatial signal by combining alleles and loci and producing a proper correlation coefficient (r) similar to Moran's I. Tests for statistical significance are performed by random permutations (1000) to define upper and lower bounds of the 95% confidence interval. Spatial autocorrelation as well as a Mantel test on pairwise genetic and geographical distance matrices were performed using GenAlEx 5.31 (Peakall & Smouse 2004).

life-history trait analyses

A matrix was developed to investigate the relationship between the 258 recorded species and various combinations of eight life-history traits derived from published information (Floyd 1989; Ilic et al. 2000) and personal observations (Table 3). Measurements and rankings were based on Webb (1978) and Floyd (1989) for leaf size, Ilic et al. (2000) for wood density, Floyd (1989) for fruit type, fruit and seed size and dispersal mode, Floyd (1989) and Kooyman (personal observation) for shade tolerance, and Kooyman (unpublished data) for resprouting. Resprouting was ranked highest for plants that replace adult stems through time (regardless of the presence or absence of disturbance), next highest for plants that simply have basal suckers or display some suckering following disturbance and lowest for those where the trait is absent. This was done to avoid the obstacles identified by Bond & Midgley (2001) in relation to plant classification schemes (see for example Westoby 1998), and the problems associated with ranking both juvenile and mature plants of species on the basis of their resprouting behaviour in response to variable levels of disturbance (see Bellingham & Sparrow 2000). We ranked resprouting as an adult stem replacement mechanism (sensu-stricto) rather than as a continuum of potential responses to variable levels of disturbance.

Table 3.  Life-history traits and rankings used in data collection and subsequent analyses
Leaf sizeFruit typeFruit size (mm)Seed size (mm)Dispersal modeShade toleranceRe-sproutWood density (kg m−3)
1. Nanophyll1. Fleshy1. < 11. < 11. Frugivore1. Yes1. YesActual
2. Microphyll2. Aril2. 1–62. 1–62. Wind2. Intermediate2. Intermediate 
3. Notophyll3. Non-fleshy3. 6–153. 6–153. Scatter-horde3. No3. No 
4. Mesophyll4. Spores4. 15–304. 15–304. Ants   
5. Macrophyll5. Other5. > 305. > 305. Other   

Wood density data were available for 91 of the 258 species and the relationship of this trait to leaf size was tested using a two-factor anova, as was the relationship between leaf size and seed size for the 154 woody species in the data set. A modified matrix based on 258 species and the six traits identified statistically and comparatively as the most influential (seed size, fruit size, fruit type, dispersal mode, leaf size and resprouting), and for which data were available for all species, was subsequently analysed using PATN 3.03 (Belbin & Collins 2004) and an ordination diagram was generated. All available analysis evaluation tests were run to determine the statistical significance (influence) on target species and broader species groupings, of both the individual and correlated combinations of species life-history traits. We have not attempted to position or compare species in our data set on the basis of phylogeny and trait emergence (refer to Westoby 1998). The emphasis here is on functional ecology and the influence of life-history traits and trait combinations on the persistence and current distribution patterns of rare species in a rain forest habitat type.


floristics, environmental variables, species distribution and habitat indicators

The classification of the quadrat data (as a column fusion dendrogram generated by PATN analysis; not presented) identified broad patterns representing the relationship between geographical location and floristics, and the influence of environmental variables on these patterns. The numerous subgroupings show how (in decreasing order of importance) fire frequency, disturbance, altitude, topographical position and soil depth influence floristics in the 92 sites analysed (Fig. 2). A similar pattern was detected for the 22 sites with E. hardeniana present: these sites were closely grouped in the column fusion dendrogram (not presented) based on the relationship between geographical distance (location) and floristics in ‘core’ rain forest habitat (groups ii and iii in Table 4). The outlier sites with the species present (two groups of two in the dendrogram) represented the influence of environmental variables on floristics at the margins of the species’ identified ‘core’ rain forest habitat (groups iv and v in Table 4). The ordination diagram shows that there are substantial areas of potential habitat where the species is not present (Fig. 2). This was confirmed in the classification where 60 of the 92 quadrats sampled fell within the known habitat parameters of the species (Table 4).

Figure 2.

Constrained ordination for the 92 sites in this study. Sites with Eidothea hardeniana (22) are shown as solid black squares. The five dominant environmental variables and their relative values (Kruskal–Wallis) are: fire frequency (46.900), disturbance (46.861), altitude (33.766), topographic position (24.212), soil depth (24.046). Sites with E. hardeniana present show a negative association with increasing disturbance and fire frequency, and a positive association with decreasing soil depth (shallow soils). Overall stress in the ordination (0.2272).

The ordination of quadrats showed that the scatter of plots (Fig. 2) and species were consistent with the broad community/floristic patterns described in the classification. The distribution of E. hardeniana was found to be negatively associated with the broadly correlated variables of fire and disturbance, with 68% (15 from 22) of the sites including E. hardeniana showing no evidence of fire (Table 4, ii and iii), and the remaining seven occurring at the margins of the fire-free rain forest habitat (Tables 4, v). The fire interval for the seven sites was estimated at > 500 years based on the size and distribution of fire-reliant species such as Lophostemon confertus and Syncarpia glomulifera. E. hardeniana occurred predominantly on shallow soils in association with specific topographic positions and landscape features. The influence of altitude was primarily related to its correlation with geographical location and floristics.

The current realized habitat niche of E. hardeniana is therefore described by the convergence of specific environmental factors, including topographic positions and features within or at the edge of SNVF, on shallow rhyolite soils (including sandy alluvium adjacent to creeks), in cool, moist, fire-free upland habitats within a constrained (geographical) area and altitudinal range. This conforms to, and can be characterized as, refugial rain forest habitat.

genetic diversity, inbreeding and effective population size

The five AG-repeat loci used in this study were polymorphic and produced a total of 23 alleles across the entire species (191 individuals). Gene diversity (HE) ranged across the loci from 0.80 (Eh251bgt, which also produced the highest number of alleles, eight) to 0.31 (Eh94bgt, which also produced the lowest number of alleles, two) with no significant allelic fixation for any of the loci (Table 2). Pair-wise tests of genotypic disequilibrium produced four significant locus pairings from 10 possible comparisons after Bonferroni corrections. As E. hardeniana is capable of resprouting, a number of above-ground shoots (ramets) were tested to assess if they were part of the same genet. Four such individuals were confirmed as being young shoots connected underground to a larger tree, and were less than 2 m away from the larger plant. This confirms that, although capable of resprouting, the species ability to spread vegetatively is limited in extent. These four individuals were removed from further genetic analyses to avoid bias.

Five alleles (22%) were rare (i.e. frequency < 5%), overall levels of gene diversity and heterozygosity were relatively high (HE = 0.542 and HO = 0.572, respectively) and the inbreeding coefficient was not significant (f =−0.057) across E. hardeniana, suggesting that mating occurs randomly across the population. When comparing different demographic cohorts, heterozygosity levels were marginally higher in juvenile plants than in mature trees (Table 5), with neither showing significant levels of inbreeding. Distribution and frequency of alleles was surprisingly similar across generations (Fig. 3b). The effective population size measured using temporal changes in allele frequencies across one generation was Ne = 21.6.

Table 5.  Number of plants (N), mean number of alleles per locus (A), observed (HO) and expected (HE) heterozygosity, and inbreeding coefficient (f) for the two sites described in the text and for mature and juvenile plants
Site A1004.40.6040.561−0.078
Site B 914.20.5410.523−0.034
Juvenile 854.60.5860.577−0.015
Figure 3.

Allelic frequencies across the two sites (above), and within the two age cohorts (below).

spatial autocorrelation of genetic structure

Population genetics values were similar across the two sites, with slightly higher diversity (allele numbers and heterozygosity, Table 5) being recorded at the larger site. Two and one alleles were unique to sites A and B, respectively (Fig. 3a), with low but significant differentiation being measured among the two sites: ρ = 0.04 (P < 0.001). Two locus/site comparisons were outside Hardy–Weinberg equilibrium, both showing significant (P < 0.01) heterozygosity excess. A Mantel test detected a very low but significant correlation (R2 = 0.035, P = 0.001) between genetic differentiation and geographical distance among individuals. Figure 4 shows the level of genetic correlation (r) across increasing distance classes. The combined values show significant positive correlation steadily declining until a sampling unit of 600 m (Fig. 4a), thus across the only E. hardeniana population, positive spatial genetic structure extends to around 400 m.

Figure 4.

Correlation for increasing distance class sizes across whole population and selected groups. 95% CI about the null hypothesis of a random distribution of genotypes (U and L represent the upper and lower confidence limits respectively) and 95% confidence error bars about r (the genetic correlation as a function of the spatial distribution of individuals) are determined by bootstrapping. (a) Combined correlation; (b) comparison between sites A and B; (c) comparison between juvenile (J) and adult (T) individuals.

There was a notable difference in patterns of positive spatial genetic structure between the two separate sites, with r-values being higher and significantly positive for larger distances in site B (up to 400 m) than in site A (up to 200 m, Fig. 4b). This between-site difference is likely to be a consequence of the linear spatial configuration of the individuals in site B (Fig. 1) increasing the chance of positive autocorrelation. Graphical comparison of overlap in the error bars suggests that r is significantly larger in site A than in site B for distances of 10, 20 and 50 m. Differences in r-values were not as evident across age cohorts (Fig. 4c) and were not significant. In both age cohorts they remained significantly positive across all distance classes, with older plants’ values being higher up to the 100 m distance class, and juvenile values being higher from the 400 m distance class. Overall, the autocorrelation analyses detected consistent patterns of genetic structuring across the E. hardeniana population, suggesting that the null hypothesis of random distribution of genotypes can be rejected. As the measured inbreeding coefficient levels suggest that random mating is occurring, it is likely that such structure is a consequence of outcrossed offspring germinating in close proximity to their maternal plant. However, the fact that r-values are generally smaller than those expected for half-sibs (0.25) suggests that there is overlap among individual seed shadows (particularly at larger distance classes, Fig. 4).

life-history traits

The ordination of species by traits (Fig. 5) showed a consistent correlation between seed size, fruit size and fruit type, and a strong association between these traits and dispersal mode. Therefore, the relationship between seed mass and dispersal biology was identified as an important trait dimension in this data set. The other significant but less influential dimension identified involved leaf size, resprouting and wood density. This dimension was independent from seed mass and dispersal biology. The anova results for leaf size by wood density detected a positive relationship between smaller leaf size and increasing wood density and showed a statistically significant relationship between the traits (F2,88 = 3.929, P < 0.024) based on the 91 woody species for which data were available. However, the anova results for leaf size by seed size did not show a statistically significant relationship (F2,150 = 2.289, P > 0.1) for the 154 woody species used in the analysis. Based on the traits used in this study, this suggests the underlying factors that influence the study species’ current distribution and ‘fitness’ relate primarily to the interaction of the two major trait dimensions reflective of persistence (resprouting) and seed size and dispersal, and then to the relative level of influence of each of the secondary traits and trait combinations/dimensions.

Figure 5.

Constrained ordination of 258 species in relation to six life history traits. Kruskal–Wallis values for top five are: fruit type (200.732), dispersal mode (182.304), fruit size (178.005), resprout (153.293), seed size (151.926). To provide a less congested graph, only Eidothea hardeniana and the other rare (woody) rain forest species are shown (black triangles). Corokia whiteana (1), Elaeocarpus sp. Rocky Creek (2), Endiandra hayesii (3), Endiandra introrsa (4), Hicksbeachia pinnatifolia (5), Neimeyera whiteana (6), Syzygiumhodgkinsoniae (7), Uromyrtus australis (8) and Symplocos baeuerlenii (9). The influence of the fruit-size/seed-size/ dispersal dimension is apparent on the ordination of the rare species, with most of these positioned towards the larger seed end of the size gradient. Resprouting is the other important trait dimension influencing the ordination of the rare taxa. Overall stress in the ordination (0.1065).

Figure 5 shows E. hardeniana and the other rare tree species in the region positioned predominantly at the large-seed end of the range, with six of the nine rare species having seeds > 1 cm (and three having seeds ≥ 5 cm). Rare species account for > 66% of the largest seeded species in the full set of 154 woody species included in the analyses. The relative positions of Eidothea and Elaeocarpus sp. ‘Rocky Creek’ in the ordination also reflect the strong influence of the resprouting dimension, and their shared mode of seed dispersal (terrestrial mammal).


defining environmental boundaries for e. hardeniana

The first task in relation to understanding the current distribution of E. hardeniana was to identify possible habitat preferences and limitations. The results from the surveys and classification (Fig. 2, Table 4) show that E. hardeniana has preference for rain forest (SNVF) habitats in what can be defined as ‘cool moist mountain refugia’. These are areas likely to maintain suitable conditions for rain forest survival during the characteristically drier and more fire-prone glacial periods. This is not surprising as, on floristic grounds, the Nightcap area is regarded as one of the most important rain forest refugia in Australia and retains a high number of locally endemic species that have their closest relatives in both northern and southern Australia as well as New Caledonia (Adam 1987).

The habitat surveys show that habitat availability is not a limiting factor for E. hardeniana. A considerable number of suitable sites (> 60%) are not occupied by E. hardeniana despite supporting other rare, endemic species with similar requirements (R. M. Kooyman, unpublished observations). For instance, barriers to fruit dispersal were shown to be a more important limiting factor to between-population genetic exchange than geographical distance alone in another local rarity (the tree Elaeocarpus sp. ‘Rocky Creek’, M. Rossetto, unpubished data). Accordingly, we sought to test if reduced dispersal potential also played a significant role in restricting the distribution of E. hardeniana.

Wright (2002) reviewed a range of mechanisms that influence species coexistence and allow species-rich assemblages to persist through time. He identified a lack of competition between suppressed understorey plants (including seedling recruits) as a potentially important factor for species coexistence. This is one of several factors he identified that could also contribute to the persistence of E. hardeniana. Hubbell (2001) provides a detailed investigation of niche and dispersal assembly theories and suggests that a reconciliation of these theories is required to adequately explain coexistence and the maintenance of species richness in communities. Our study has identified factors related to both niche and dispersal assembly as significant for the persistence of the species.

genetic diversity and population size

Molecular genetics can uncover the history of small populations, and consequently the potential influence of human disturbance, as recent bottlenecks are likely to leave a measurable genetic signal (see Cornuet & Luikart 1996). Allelic diversity and temporal variance (adults vs. offspring) in allelic frequencies have been experimentally shown to be the most sensitive measure of genetic change caused by a reduction of effective population size (Spencer et al. 2000). However, the robustness of this approach relies on the suitability of the data (Leberg 2002). In the case of E. hardeniana, the availability of multigenerational data from all known individuals enabled us to interpret allelic diversity and temporal variance in allelic frequencies safely without risking bias caused by differences in sampling intensity. Differences in allelic frequencies and distributions did exist among sites (Fig. 3a), but were not apparent among age cohorts (Fig. 3b). Also, neither inbreeding (Table 5) nor autocorrelation values (Fig. 4c) were higher in seedlings than in adults of E. hardeniana, although both would be expected to increase within the progeny of bottlenecked populations (Aldrich et al. 1998; Rossetto et al. 2004b). Despite the small Ne therefore, the genetic data for E. hardeniana do not support the hypothesis of a recent population decline followed by a loss of equilibrium.

defining genetic boundaries and dispersal potential of e. hardeniana

The current distribution of this taxon is not explained either by a recent bottleneck or by reproductive failure, as viable seeds are produced and the sizeable juvenile element of the population is highly heterozygous. Observations to date indicate that in the early stages of development seedlings grow relatively quickly from the large seeds, tolerate some canopy shading, and are persistent (at a scale of years) once established (personal observation). This suggests that E. hardeniana seedlings are able to compete with other species for both space and resources (long-term demographic studies are currently underway). As a result, the remaining major hypothesis that can explain the current distribution of E. hardeniana is that of limited dispersal potential. If dispersal is limited, significant spatial structuring of genotypes can be expected, even in an outcrossed species, as prevalent seed germination near the maternal parent will result in a clustering of closely related individuals. Mantel tests and positive autocorrelation indicate that, despite the prevalence of random breeding, significant patterns of fine-scale genetic structure do exist in E. hardeniana, and genotypes are not randomly distributed in space. In other words, proximate plants are more genetically similar, and low levels of spatial genetic structure extend to 400 m (Fig. 4a). This was true across age classes and within both sites (Fig. 4b,c) and is consistent with the presence of overlapping seed shadows originating from related tree cohorts.

Preliminary observations (P. Weston, personal communication) suggest that pollen movement is mostly mediated by small nocturnal beetles and generally results in autogamous pollinations. Nevertheless, the molecular data show that most viable seeds are outcrossed, presumably due to severe inbreeding depression or to self-incompatibility mechanisms. As a result, the short range of pollen movement is unlikely to be the main cause of fine-scale genetic structuring. The bitter flesh of the E. hardeniana fruit is thought to be unpalatable to the local dispersal organisms (Weston & Kooyman 2002). Generally, once the fruit falls to the ground, bush rats (Rattus fuscipes) do not eat the flesh but chew through it and the hard endocarp to feed on the large nutritious seed. As a result, dispersal away from the maternal tree is likely to be uncommon, at least beyond the distance travelled by rats to cache their food (R. fuscipes generally have linear home ranges of around 200 m, Robinson 1987). Similar mechanisms were suggested for another local palaeo-endemic rain forest tree that produces apparently unpalatable fruits (Elaeocarpus sp. ‘Rocky Creek’, M. Rossetto, unpublished data). Both larger seeds and allocation to resprouting suggest that these species might produce low seed numbers per m2 of canopy per year, a further factor that could contribute to their rare status (refer to Murray & Westoby 2000). This contrasts with a common local rain forest tree, Elaeocarpus grandis, which produces large numbers of highly palatable fruits, and has low levels of genetic structure across a much greater geographical spread (Rossetto et al. 2004b). For E. hardeniana, both the molecular and observational data support the hypothesis that relatively low seed production and the absence of efficient dispersal mechanisms contribute significantly to the range limitations currently observed for the species.

persistence: holding onto suitable habitat while retaining viability

A small number of comparative SSR-based studies on common and rare local non-rain forest and rain forest trees are available (for example Rossetto et al. 1999; Rossetto et al. 2004b). While it is acknowledged that due caution needs to be taken when comparing SSR data based on different loci, it appears that confinement to a limited geographical area and small population size has not excessively depleted diversity within E. hardeniana. Endemic and rare plants are generally expected to have lower diversity levels than more widespread taxa (Hamrick & Godt 1989). If we use Σ(1 − pi)2N to calculate the expected number of alleles lost within one generation (where pi is the frequency of the ith allele and N the population size; Frankham et al. 2002), then a loss of as much as 4.1 alleles per generation could be expected. Such a significant allelic loss is not apparent for E. hardeniana. An important compensatory factor could be that very long generation times due to resprouting and slow adult replacement reduce generational overlap and stabilize family size variance close to equalization. Theoretically, if all parents contribute equally to the next generation and family size (the lifetime production of offspring per individual) variance (Vk) is equalized, then Ne∼2N.

Direct observations show that as a mature E. hardeniana stem decays, the gap is often filled in by shoots resprouting from the base of that same tree, thus ensuring local persistence and preserving the same genotype. Although apparently reducing the opportunities for the establishment of sexually produced seedlings, this strategy prevents the formation of even-aged, closely related cohorts of juveniles, encourages the persistence of rare alleles, and provides a good competitive strategy against seedlings from other species. However, successful cross-pollination events are likely to be rare, particularly considering that most neighbouring individuals are genotypically similar (Fig. 4). This would explain why, despite the long generation times and slow adult replacement, Ne is still considerably smaller than N in E. hardeniana.

Interestingly, the reduction in effective population size could contribute to the maintenance of diversity across the progeny and the population overall as it can potentially reduce the over-contribution of similar neighbouring genotypes. The long-term persistence of single individuals would then ensure that lengthy periods between successful seed establishment do not cause a rapid population crash. Thus the apparent limitations of this unconventional demographic cycle could in fact contribute to the maintenance of diversity and evolutionary potential within small, refugial populations. The occurrence of multiple stems on non-injured trees (and the capacity for in-situ stem replacement) could describe a major mechanism ensuring the long-term survival of palaeo-endemic species within small, isolated rain forest remnants. In that context, these life-history attributes would seem to favour persistence rather than spread of the species. These hypotheses are currently being tested through in-depth reproductive and demographic studies on a range of taxa.

Aside from external gene flow and demographic processes, balancing selection mechanisms could also be involved in diminishing the decline of diversity within smaller populations. Heterozygote advantage (when heterozygotes have higher fitness than homozygotes) is a possible factor; however, this is generally regarded as being more influential within large populations (Frankham et al. 2002). A similar mechanism, associative over-dominance, is more likely to operate within small populations. In this process, linked loci (detectable via linkage disequilibrium) are the source of heterozygote advantage as they prevent the potentially deleterious effects of chromosomal homozygotes. This is a selective mechanism that develops across many generations and could play a role in the preservation of heterozygosity in E. hardeniana, especially in view of the relatively high levels of linkage disequilibrium measured. However, such a process would not explain the persistence of a relatively high number of alleles as it does not prevent the loss of allelic diversity. As a result, it is unlikely to be the only contributing factor here.

significant life-history traits for rain forest species

This study suggests that a combination of current and ancient factors limit the distribution of E. hardeniana. Originally the distribution of this species is likely to have been reduced as a consequence of habitat attrition during the Quaternary. From the end of the Miocene the Australian continent became drier, and with the subsequent onset of ice ages the habitat of E. hardeniana and other rain forest taxa was placed under increasing pressure. In northern NSW, and around the world (Hewitt 2000), this resulted in increased levels of extinction and drastic range reductions for numerous taxa. While the milder interglacial climatic conditions offered an opportunity for rain forest expansion, in some cases such expansions are likely to have been made difficult because of the combination of large fruit/seed size and the extinction of some fruit/seed dispersers.

As stochastic events resulted in population extinctions, E. hardeniana was unable to recolonize newly available habitat because of its limited dispersal capacity. The local persistence of E. hardeniana through resprouting has extended its ability to contribute to future generations via potentially rare but highly outcrossed seedling establishment events. The long-term persistence of individual genets has contributed to the exceptionally long-term survival of small populations. Major environmental changes are likely to have been important drivers of selective pressures on other large-fruited species. It can be assumed that while large-seeded plants could establish more efficiently, the lower number of seeds produced per unit area of canopy and the absence of dispersal organisms resulted in reduced competitiveness in recolonizing newly available habitat. As a result, the large-fruited species that survived were the ones that could increase their competitiveness through local persistence.

To test this hypothesis, we examined the potential for generalizing these findings across the local rain forest flora (or at least across the 258 taxa recorded within our 92 survey quadrats). We found that the major underlying life-history traits that explain the current distribution of the rare rain forest species are those associated with the dispersal/recruitment dimension and those associated with local persistence. To our knowledge this is the first study that has incorporated and investigated the importance of the resprouting habit (albeit restricted to adult stem replacement) across a rain forest community, and in relation to the evolutionary implications for a range of taxa. While previous studies recognized the apparent relevance of this trait, they were hampered by the inherent difficulties associated with classifying levels of resprouting according to disturbance types. We circumvented this problem by allocating species to only two resprouting categories based on the taxa for which resprouting is purely a disturbance response, and those for which a resprout can replace a senescent adult plant in the absence of disturbance (for example E. hardeniana and Endiandra introrsa, Fig. 5).

Wright et al. (personal communication) identified seed size and fruit size as the most tightly related trait-pair among 10 pair-wise trait relationships in the neo-tropics. This was consistent with the findings from our data set. Our data also confirmed the connection between seed mass and dispersal biology, and (in broader terms) growth form. However, an anova on the data for the 154 woody species recorded confirmed that no correlation between leaf size and seed size could be detected within our data set. Our study did not use leaf economics data (representing the ratio of leaf area to dry mass) or twig cross-sectional area measures to determine the relationship between leaf size and seed size (see Westoby & Wright 2003). However, information on wood density was available for 91 tree species, and this trait was shown to correlate with leaf size, consistent with the findings of Wright et al. (personal communication). Thus, within our study area, the detected trait dimensions and relationships confirmed those detected in the larger study in the neo-tropics but included the additional trait of resprouting (as adult stem replacement). Future studies will position species relative to trait evolution and emergence and further explore the relationship between growth form and traits such as seed size (consistent with Moles et al. 2005 for example).

Beyond the correlations and relationships between traits identified in previous studies, our data show that the interaction of the two major trait dimensions identified (re-sprouting/persistence, and dispersal/seed size) can compensate for environmental limitations and variations (Higgins et al. 2000; Bond & Midgley 2001) as well as pressures such as the absence of effective dispersers or climate change. Therefore, despite being restricted in distribution and limited to small effective population size, palaeo-endemic species have not necessarily reached their evolutionary ‘use-by-date’. Rather, they remain as important representatives of unique lineages that may yet persist for great lengths of time and potentially expand from their current refugia when new opportunities arise. Although the focus of this study is local, its implications are pertinent to a broad range of ecosystems and environmental conditions, and emphasize the importance of identifying and protecting even the smallest refugial areas.


This project was partially funded by BGT, NPWS (DEC), Rain forest Rescue and Andrew Hall, and SCU. C. Porter and R. Jones are acknowledged for their technical support. M. Westoby and I. Wright are acknowledged for providing access to pre-publication data from the Neo-Tropics and insightful discussions related to life-history trait analysis. M. Westoby, P. Weston, R.O. Makinson and two anonymous reviewers are acknowledged for their useful comments on the manuscript.