Relative frequency of sympatric species influences rates of interspecific hybridization, seed production and seedling performance in the uncommon Eucalyptus aggregata


*Correspondence author. E-mail:


  • 1Habitat fragmentation can alter the relative frequency of cross-compatible species within an area, which can affect the levels of interspecific hybrid production and reduce the viability of small populations through genetic and demographic swamping. For 18 populations of Eucalyptus aggregata, we examined the effects of absolute and relative population size (compared with its congeners E. rubida, E. viminalis and E. dalrympleana) on hybrid production, genetic diversity and subsequent seed production and seedling performance.
  • 2Relative population size was strongly negatively correlated with rates of hybrid seed production, suggesting increased hybridization when the potential sources of interspecific pollen outnumber the sources of intraspecific pollen for E. aggregata trees.
  • 3Genetic diversity was negatively correlated with relative population size which suggests that hybridization may maintain diversity lost through bottlenecks and drift following reductions in population size. However, the presence of fertile hybrid adults, and introgressed leaf traits within populations exhibiting high hybridization rates, suggests that small E. aggregata populations may be vulnerable to genetic swamping by common congeners.
  • 4Amongst an array of population parameters (population sizes, genetic diversity and inbreeding), seed production was only positively correlated with relative population size, whereby sites with low relative population sizes tended to produce fewer seed. This could be due to the action of pre-zygotic barriers which removes inviable hybrid genotypes as levels of interspecific pollen flow increase.
  • 5Germination and survivorship displayed a similar positive correlation with relative population size, suggesting post-zygotic hybrid breakdown may also contribute towards to demographic swamping of remnant populations.
  • 6Synthesis. Our results suggest that relative population size is an important parameter determining rates of hybrid production, seed production and seedling performance. Furthermore, relative population size has stronger effects on population fecundity than absolute population size, genetic diversity and levels of inbreeding. Relative population sizes > 0.5 (i.e. at least equal frequencies of parentals) may be required to avoid the deleterious effects of genetic and demographic swamping on the viability of rare species.


Habitat fragmentation, which can reduce the size and increase the isolation of populations, is considered to be one of the major threats to biological diversity and the long term persistence of rare species (Young & Boyle 2000; Hobbs & Yates 2003). The genetic and demographic effects of habitat fragmentation on plant population viability are well-documented and include increased inbreeding (Dudash & Fenster 2000), loss of genetic diversity (Oostermeijer et al. 1994), the disruption of plant × pollinator interactions (Mustajärvi et al. 2001) and reduced reproductive success (Cunningham 2000). For cross-compatible species, one potential effect of habitat fragmentation is alteration of both the rate and the spatial scale of interspecific hybridization. In some cases, this occurs through the anthropogenic introduction of foreign species within close proximity to compatible native species, which enables gene exchange to occur due to the disruption of geographical barriers (Abbott et al. 2003). Increased hybrid production has also been observed in small plant populations of sympatric congeners within fragmented landscapes (Potts & Wiltshire 1997; Butcher et al. 2005). However, few studies have empirically explored the extrinsic ecological and population conditions influencing hybrid production in fragmented landscapes. Such knowledge on the conditions promoting hybridization and the subsequent effects on population fitness could be useful for predicting rates of hybridization in fragmented landscapes and managing the demographic and genetic viability of remnant populations.

Elevated rates of hybridization between congeners has important implications for the demographic and genetic viability of small plant populations and ecological interactions within remnant populations (Ellstrand & Elam 1993). For a rare species, increasing hybridization is expected to directly influence the time to local extinction (Wolf et al. 2001) due to reductions in the number of pure-bred genotypes (demographic swamping) and the dilution of the gene-pool through introgression (genetic swamping) with the more abundant congeners (Levin et al. 1996). In several cases the demographic and genetic effects of hybridization have been identified as a major cause of plant population decline (Rieseberg & Gerber 1995; Levin et al. 1996; Rhymer & Simberloff 1996). Moreover, an increase in the number of hybrids in plant populations can have important implications for ecological communities through alteration of the composition of insect and bird communities (Whitham et al. 1999) and disruption of parasite resistance in plant species (Fritz et al. 1999). Despite the potential adverse effects, hybridization is also considered to be a substantial adaptive force in the evolution of many plant species (Rieseberg & Carney 1998; Barton 2001). In some circumstances hybridization may be considered to be beneficial to small populations due to the influx of novel genotypes. This may counteract the deleterious effects of genetic drift and inbreeding (Savolainen & Kuittinen 2000) and impart new adaptive variation required by remnant populations to adapt to changing environments (Ellstrand 2003). Considering the potential for both beneficial and adverse outcomes of hybridization, knowledge on both the potential adverse declines in population fitness and beneficial effects on genetic diversity would be useful for managing remnant populations.

The distribution and size of the parent populations affect pollinator behaviour and subsequent plant mating patterns (Kunin 1997), and are therefore likely to be major factors influencing hybrid production in remnant populations. The absolute and relative population sizes (i.e. relative frequencies of hybridizing taxa) of parentals within a site can easily be altered by habitat fragmentation, and may reflect preferential clearing of one parental species when species exhibit landscape scale environmental preferences (e.g. for slopes or valleys). Source-sink dynamics predict that these population parameters are probably important determinants of the amount of immigrant/interspecific pollen received by populations (Ellstrand & Elam 1993). For example, hybridization rates would be expected to be greater in small plant populations as they receive more pollen than they export. Furthermore, reduced population size of a focal species relative to congeners would increase the probability of interspecific pollen flow due to a greater relative availability of conspecific pollen within the dispersal area of pollinators (Ellstrand & Elam 1993). This is expected to result in a decrease in pollinator fidelity, especially if the floral structure of the hybridizing species are similar (Kunin 1993).

Our understanding of hybrid promotion in fragmented landscapes is predominately based on theoretical models (Pederson et al. 1969) and extrapolations from distance and density studies which examined intraspecific pollen flow within experimental arrays (e.g. Ellstrand et al. 1989; Richards et al. 1999). Although these studies suggest a negative correlation between hybrid production and both absolute and relative population size, direct studies of natural hybrid systems are sparse. A number of studies have documented increased hybridization or introgression in small populations (Rieseberg et al. 1989; Gallo et al. 1997; Kennington & James 1997; Burgess et al. 2005), but there have been no explicit attempts to examine the rate of hybrid seed production across a range of absolute and relative population sizes in a natural system. Testing this relationship in natural interspecific systems is important because there is growing evidence of hybrid promotion in fragmented landscapes (Potts & Wiltshire 1997; Buerkle et al. 2003; Butcher et al. 2005) but empirical evidence that these population parameters are important in complex plant distributions plants found in natural populations is limited. Related declines in demographic parameters would be expected in populations with more frequent interspecific gene flow due to the action of pre-zygotic barriers and hybrid breakdown on the performance of hybrid genotypes (Rieseberg & Carney 1998). However, the potential for hybridization to reduce important population demographic parameters such as plant fecundity and offspring performance remains untested. This information is vital in determining the processes promoting hybridization and the ability of remnant plant populations of hybridizing species to persist and re-establish across fragmented landscapes.

The genus Eucalyptus (Myrtaceae) is a particularly good system with which to explore hybrid promotion in fragmented landscapes. This is because many species are known to hybridize freely due to weak reproductive barriers (Griffin et al. 1988) and hybrid swarms have often been associated with the results of human activities such as habitat modification and fragmentation (Potts et al. 2003; Butcher et al. 2005). Eucalyptus species are of high conservation priority as they are a diverse and highly important vegetative component of the Australian continent (Hill 1994). However in modern times over half of the Eucalyptus forests have been cleared or highly modified (Young et al. 1990), with many species now inhabiting only small fragmented and disturbed remnant populations.

The primary objective of this study was to examine the relationship between population size (absolute and relative) and hybrid seed production. This also provided the opportunity to test the potential benefits of hybridization through the maintenance of genetic diversity. In addition, we examined the power of population sizes, hybridization rates and mating system parameters estimated from a current open-pollinated seed cohort, to explain variation in seed production, seed germination and seedling performance.

These questions were examined using the long-live tree species Eucalyptus aggregata, which hybridizes with the more common sympatric species, E. rubida, E. viminalis (Field et al. 2008) and possibly E. dalrympleana. The pollination system of these Eucalyptus species is probably entomophilous due to floral foraging by a diversity of insect species, in particular the introduced Honeybee (Apis mellifera), and various native bees (Leioproctus Colletidae, Lasioglossum/Homalictus Halictidae). Populations of E. aggregata range from small remnants where it is out-numbered by a number of more common species, through to large relatively intact woodland populations where E. aggregata is numerically dominant. This range provides a good opportunity to examine rates of hybrid production in relation to absolute and relative population size. The pre-zygotic barriers and hybrid breakdown documented in Eucalyptus (Potts et al. 2003) are expected to result in reduced fecundity and seedling performance for populations with more frequent interspecific gene flow. However, this association has not been tested in Eucalyptus or compared to other well-documented demographic and genetic consequences of small population size.

We asked three specific questions: (i) Does the absolute and relative population size of E. aggregata (compared to common sympatric species) affect hybridization rates in progeny arrays? (ii) Do population sizes (absolute and relative) and hybridization rates affect genetic diversity and levels of inbreeding in progeny arrays? (iii) Do hybridization rates (and parameters associated with hybrid production) have a greater affect on population fecundity and seedling performance than population size, levels of inbreeding, outcrossing rates and genetic diversity?


study sites and seed sampling

This study was conducted on the southern and central tablelands of New South Wales, in South-Eastern Australia (Fig. 1). Eighteen mixed sites for E. aggregata were selected with populations ranging from small road verge remnants dominated by compatible congeners through to larger woodland populations where E. aggregata is numerically dominant (Table 1). Five reference populations of each of E. rubida and E. viminalis were selected on the basis of high or complete dominance of the respective species and the absence of putative adult hybrids. Reference populations of E. aggregata were not used because previous work found hybrids were still recorded at these sites and their inclusion made little difference to hybrid assignments (Field et al. 2008). Reference populations of E. dalrympleana were also not used as this species co-occurred with E. aggregata at only one of the mixed sites and has little genetic differentiation from E. viminalis and E. rubida (Cayzer 1993).

Figure 1.

Map of the location of 18 Eucalyptus aggregata (black circles), five E. rubida (open circles), and five E. viminalis (grey circles) populations sampled for open pollinated seed for the assessment of hybridization rates.

Table 1.  Site details, including the abbreviation code, geographic location and population sizes (APS, RPS) of three Eucalyptus species sampled for open pollinated seed
  1. APS, absolute population size reproductive adults; RPS, the relative population size of E. aggregata to congeners (see Methods).

E. aggregata
 Manar CreekMC35°17′30″ S, 149°41′45″ E10.25
 Willandra laneWL35°07′20″ S, 149°36′00″ E40.31
 NorongoNG35°42′00″ S, 149°25′00″ E60.21
 Sth TogganoggraST35°39′00″ S, 149°36′00″ E70.15
 MedwayMW34°29′50″ S, 150°17′00″ E150.75
 Mozart RoadM33°48′00″ S, 149°47′15″ E300.75
 RosevalleyRS35°39′00″ S, 149°25′30″E300.33
 BerrimaBR34°29′13″ S, 150°18′50″ E350.64
 Duck FlatDF35°09′00″ S, 149°34′20″ E420.51
 Grabben GullenGG34°32′00″ S, 149°23′00″ E500.77
 HollowH35°40′00″ S, 149°27′00″ E900.69
 Bendoura SthBS35°31′50″ S, 149°40′00″ E800.67
 Reedy CreekRC35°21′30″ S, 149°32′45″ E1000.83
 NevilleN33°42′00″ S, 149°12′00″ E1500.87
 RiversideRV35°16′00″ S, 149°54′30″ E2000.95
 FairviewFV33°24′00″ S, 150°04′00″ E3000.83
 Bendoura TSRBT35°30′10″ S, 149°42′00″ E4000.75
 Captains FlatAP33°21′00″ S, 150°05′30″ E7000.90
E. rubida
 Captains FlatCF35°23′19″ S, 149°21′40″ E~100
 Kings HwyK35°18′15″ S, 149°45′00″ E~100
 Sth Captains FlatSCF35°37′00″ S, 149°26′15″ E~100
 TaragoT35°10′00″ S, 149°39′45″ E~100
 Reed CreekRE34°21′00″S, 149°33′00″ E~100
E. viminalis
 BallalabaB35°34′16″ S, 149°37′45″ E~100
 Faegans CreekFC35°24′57″ S, 149°58′05″ E~100
 Parker GapPG35°37′30″ S, 149°29′00″ E~100
 Warri BridgeW35°20′39″ S, 149°44′15″ E~100
 Wattle FlatWF33°09′00″ S, 149°41′00″ E~100

For each of the E. aggregata sites, two variables were estimated by counts of mature reproductive individuals: (i) absolute population size of E. aggregata (APS), a count of mature E. aggregata individuals, (ii) relative population size of E. aggregata (RPS) a measure of the relative frequency of E. aggregata compared to the sympatric species, defined as the absolute population size of E. aggregata divided by the sum of the absolute population sizes of all compatible species at the site [RPS = APS values for E. aggregata/(APS of E. aggregata + E. rubida + E. viminalis + E. dalrympleana)]. The relative population size ranged from 0.15 to 0.95, where a value of 0.50 indicates a site where the frequency of E. aggregata individuals is equal to that of the sympatric species. The sites ranged from 50 × 50 m to 900 × 600 m with the boundary selected on the basis of a buffer of at least 1 km to the nearest E. aggregata or compatible congeners. For a few sites, compatible trees > 1 km from the site boundary may contribute to interspecific pollen flow although they are not included in population size estimates. However their contribution is expected to be relatively small compared to trees within the sites because the majority of pollen dispersal is expected to come from within the first 600 m of target trees (Barbour et al. 2005; Byrne et al. 2008).

Between November 2002 and January 2003, 50 to 200 open-pollinated capsules were collected from the canopies of one to 20 haphazardly selected E. aggregata within each of the mixed populations, and from five trees at each of the E. rubida and E. viminalis reference populations. Trees of each parental species were selected on the basis of distinguishing morphological features including bud size and number, leaf size and bark persistence according to Brooker & Kleinig (1999). Capsules were dried for over 2 weeks at room temperature and the seed released was bulked within trees. A total of 3120 seeds consisting of 20 to 30 seeds from each sampled tree were individually sown in a 33% river sand: 33% peat: 33% compost soil mix in separate pots (50 mm wide, 100 mm deep) in a randomised block design in an unheated glasshouse. These samples were used to examine germination, seedling survivorship and performance, and estimate hybrid production rates at 13 of the E. aggregata populations. Due to low numbers of capsules, seed and germinated seedlings at the remaining five E. aggregata populations, further seed were collected and planted in a separate trial to estimate hybrid production. Due to the offset timing of these plantings, these five populations were excluded from the germination, seedling survivorship and performance analyses. The smallest mixed population (MC) which consisted of a single E. aggregata and four E. viminalis was removed from all following regression analyses because the tree produced no hybrids and had high leverage, however it was used as a single data point to infer the possible selfing capability of an isolated tree.

genetic analysis

A total of 2559 seedlings representing 15–25 seedlings per tree were genotyped in order to identify hybrids, estimate hybrid production rates for each population, and to determine population genetic diversity and mating system parameters. Allozyme genotypes at six loci Gpi-2, Gput, Pgm-1, Pgm-2, 6pgd-1, 6pgd-2 were assayed from fresh leaf material of each seedling following the techniques of Moran & Bell (1983). Previous work indicated that allele frequencies were highly skewed or near diagnostic between E. aggregata and both E. rubida and E. viminalis at the Pgm-2, 6pgd-2, and the Gpi-2 locus (Field et al. 2008). The genotypes consisted of two data sets, the first a seedling data set from seedlings grown from seed arrays of individual trees that were classified on the basis of distinguishing morphological traits of each parental species (Field et al. 2008), and the second, an adult maternal genotype data set inferred from the progeny arrays using the method of Brown & Allard (1970) in the MLTR program (Ritland 2002). The adult genotypes were used to identify hybrid maternal parents. This data was used to detect hybrid individuals among the adult maternal genotypes, and progeny from these were removed from population estimates of hybrid production.

hybrid identification

Hybrid adults and seedlings were identified from admixture proportions (q) using the Bayesian methods implemented in the program structure 2.1 (Pritchard et al. 2000; Falush et al. 2003). This method uses population allele frequencies to assign individuals (as represented by their multi-locus genotypes) to K groups/clusters by minimizing within group linkage-disequilibrium and simultaneously assuming within group Hardy Weinberg Equilibrium. In this way, individuals can be assigned to a single group (e.g. q1 = 0.99, q2 = 0.01) or jointly to two or more groups if their multi-locus genotype indicates admixture due to hybridization (e.g. q1 = 0.5, q2 = 0.5). We used structure with the admixture model, no prior population information, a burn-in period of 20 000 generations and 200 000 MCMC's (Monte Carlo Markov Chain) to calculate the admixture proportion and the 90% probability interval (default value recommended; Pritchard et al. 2000) for each individual with respect to the E. aggregata cluster. For all STRUCTURE analyses, two genetic groups were assigned (K = 2), with E. aggregata in cluster 1 (q1) the combined E. rubida and E. viminalis in cluster 2 (q2). This was because E. aggregata was highly differentiated from both E. rubida (FST = 0.59, P < 0.01) and E. viminalis (FST = 0.58, P < 0.01). However the latter two species were relatively poorly differentiated at these allozyme loci (FST = 0.13, P < 0.01) (Field et al. 2008). Considering the low differentiation between E. dalrympleana and E. viminalis previously described at many of these allozyme loci (Cayzer 1993), and the close taxonomic relationship with E. rubida we assumed that E. dalrympleana would similarly group into cluster 2.

Critical thresholds of admixture proportions in the E. aggregata cluster (q1) were used to assign seedlings from pure-bred maternal trees and adults into one of the following genomic groups; (i) pure-bred E. aggregata for q1 > 0.9, (ii) pure-bred E. rubida or E. viminalis for q1 < 0.1; (iii) F1 hybrid (E. aggregata × E. rubida/E. viminalis/E. dalrympleana) for q1 0.7 to 0.3; (iv) backcross hybrid (generation unknown) q1 0.7 to 0.9 or q1 0.1 to 0.3. We grouped the different hybrid combinations together due to the low genetic differentiation between E. rubida and E. viminalis, however we roughly separated the cross-type on the basis of morphology (Field et al. 2008) and on congener species present at the each particular site. The admixture thresholds for each of the four groups were selected on the basis of simulated mating among pure-bred and hybrid individuals (Field et al. 2008). Previous modelling work by Boecklen & Howard (1997) would suggest that the assignment of hybrid classes with the number of markers in our study should be sufficient as a coarse classification of pure-breds, F1 and backcrossed hybrids. However, hybrid production is probably underestimated due to the difficulty in distinguishing more advanced backcrosses from pure-bred individuals if these exist within E. aggregata populations. Therefore, the classification of hybrid classes is used here cautiously, especially for the adult populations, as some F1 hybrids and pure-bred individuals may be more advanced backcrosses.

Considering these caveats for hybrid classification we used three approaches to estimate hybridization rates for populations. First, a ‘mean F1 hybrid production’ was calculated as the mean percentage of F1 hybrid progeny across pure-bred families within each population. A second estimate was also calculated across pure-bred families, but with the percentage of F1 and backcrossed hybrids combined. A third estimate of hybridization rate was calculated by the ‘mean admixture’ of q1 as an alternative measure that has no thresholds for assignments.

hybridization rates and population characteristics

Univariate nonlinear regressions were used to test the relationship between the two predictor variables, APS (log transformed) and RPS with the three response variables; (i) mean F1 hybrid production, (ii) mean F1 and backcross hybrid production, and (iii) mean admixture. A variety of nonlinear models were examined to fit the data including, negative exponential, power, Gaussian and Weibull (genstat v 8). A leverage analysis was used to identify outlying data points that exhibited disproportionate influence on the relationship (genstat v 8) and the analyses were run with and without outliers to examine their influence on the relationships. This analysis was also conducted separately for E. aggregata populations that were sympatric with E. rubida and those sympatric with E. viminalis. Populations with both E. viminalis and E. rubida were classed as sympatric with E. viminalis due to this species’ greater dominance at those sites. Seedlings from maternal parents identified as hybrids or backcrosses were excluded from the above analyses as we were only interested in hybrid production from pure-bred E. aggregata.

genetic diversity, allele frequencies and outcrossing rates

Genetic diversity measures were estimated for each population separately, including the percentage of polymorphic loci (P), mean number of alleles (A), allelic richness (Ar), unbiased genetic diversity (He) and the inbreeding coefficient (f), averaged across loci using the program fstat (Goudet 1995). Allelic richness was estimated for each locus separately with rarefaction to control for variable sample sizes between populations (Goudet 1995), and presented as an average across loci. Multi-locus outcrossing rates (tm) for each population were estimated with maximum-likelihood procedures implemented in the program MLTR (Ritland 2002) using 1000 bootstraps. In order to examine the influence of population size and hybridization rates on genetic diversity, the APS, RPS and the average hybridization rate of each population was compared with the average Ar and He and tm using nonlinear regressions fitted by a power function with a negative exponent (y = αx−β).

seed production, germination, survival and seedling performance

The following counts and measurements were made for seed arrays from each E. aggregata tree sampled: (i) seed weight (of 20 randomly selected seed), (ii) number of seeds (weighed by dividing seed number by capsule number), (iii) percentage weight of seed compared to total capsule contents, (iv) total debris weight (excluding seed) as a percentage of capsule weight. Fertile seeds were clearly distinguishable under low magnification from the remaining capsule contents which consisted of sterile particles of chaff and aborted seed. Viable seed were dark and filled while chaff and aborted seed were pale and flat. A preliminary trial comparing the germination of putatively fertile and aborted seeds, placed in petri dishes on wet filter paper, indicated that none of the seed identified as aborted were viable whereas > 50% of apparently fertile seed germination.

Starting one week after the 3120 seeds were planted, observations were made every 3 days for 6 weeks, recording when individuals germinated. Eleven months after planting, the number of seedlings that survived was recorded, and two traits were scored as surrogates of individual performance/vigour on all surviving seedlings (n = 1028) from E. aggregata mothers. These were plant height, measured to the tip of the apical bud on the main stem, and total number of leaf pairs (expanded and un-expanded).

Separate linear regressions provided the best fit (genstat v 8) to data used to examine the relationship between each of the seven parameters (APS, RPS, H%, Ar, He, f, tm) and population averages (across families) of each of the seed production characteristics (seed weight, seed number, seed weight percentage, debris weight percentage), germination and survival rates and seedling performance (height, leaf pairs). Seed numbers, weight and germination were also combined into a single measurement of the number of viable seeds per 10 g of capsule contents (e.g. Burrows 2000). Population RV was removed from regression analyses because it exhibited consistently lower seed production, the percentage weight of seeds, germination, survival, and seedling height in comparison to the other large populations. This may be due to a past bottleneck or unusually high inbreeding, as this population previously exhibited the lowest outcrossing rates, despite large population size. This experimental design does not explicitly control for the effects of inbreeding and genetic variation on seed production and early seedling performance parameters, but rather explores the amount of variation explained by each population and mating system parameter.


hybrid detection

Of the 1838 seedlings from pure-bred E. aggregata, the majority 1752 (89%) had a high affinity to the E. aggregata cluster q1 > 0.9 (Fig. 2a). Implementing the critical thresholds of q1 for individual classifications indicated 89% were pure-bred, 4.4% were F1 hybrids, and 6.3% were hybrid backcrosses. A higher proportion of seedlings from pure-bred E. rubida and E. viminalis had a high affinity to cluster 2 due to q1 < 0.1 (Fig. 2b,c). Applying the classification thresholds to the 361 seedlings from E. viminalis indicated 92% were pure-bred, 1.6% were F1 hybrids and 6.4% were hybrid backcrosses. Of the 343 seedlings from E. rubida 98.6% were pure-bred, zero F1 hybrids and 1.4% were backcrossed hybrids.

Figure 2.

Ranked admixture proportions (q1) of allozyme genotypes of seedlings from the open-pollinated progeny arrays of pure-bred (a) E. aggregata, (b) E. viminalis and (c) E. rubida. Grey bars indicate the 90% posterior probability intervals. Dashed lines indicate admixture thresholds used to classify individuals into each of four genomic groups including (i) pure-bred E. aggregata, (ii) E. viminalis/E. rubida, (iii) F1 hybrids and (iv) hybrid backcrosses.

The proportion of hybrid offspring in progeny arrays from all of the pure-bred E. aggregata trees sampled (n = 124) ranged from 0% to 44.4% and hybrids were present in progeny arrays at 15 of 18 sites (83.3%). The average hybridization rate across all E. aggregata populations was 4.2% for (F1) hybrids only and 8.9% for F1 hybrids and backcrosses. Within each population the percentage of hybrids in progeny arrays for individual trees varied markedly (e.g. site M, 0% to 5.4%, ST, 10% to 44.4%; Table 2). The mean percentage of hybrid progeny per population ranged from zero to 24.4% for F1 hybrids and up to 31% for F1 and backcross hybrids (Table 2). At the four smallest population that produced hybrids (WL, NG, ST, MW) all trees sampled produced some hybrid offspring (100% H-families; Table 2). In contrast, for the largest populations that produced hybrids, fewer of the sampled trees produced hybrid offspring (e.g. 50% population BT, 58% population AP; Table 2).

Table 2.  Details of population size parameters (APS, RPS), sampling size of open-pollinated seedlings (n) and families (n − f), from 18 E. aggregata populations sampled for assessment of mean hybridization rates. For each population, hybridization rates estimated as the mean admixture proportion in the E. aggregata cluster (q1) and the mean percentage, standard error (± SE) and range for F1 hybrids only and F1 hybrids and backcrosses combined
PopulationAPSRPSsnn − fMeanF1 hybrid %F1 hybrid + backcross %H-families
  • APS, absolute population size of E. aggregata; RPS, the relative population size of E. aggregata to congeners (see Methods).

  • Sympatric species at population, r = Eucalyptus rubida; v = E. viminalis; d = E. dalrympleana.

  • Percentage of sampled families with hybrid progeny (F1 hybrid + backcross).

MC10.25v2410.990.0(0.00) − 0.0(0.00)00
BR350.64r, v5950.951.4(0.01)0–5.65.0(0.02)0–11.180
N1500.87r, v5950.930.0(0.00)00.0(0.03)0–16.760
FV3000.83r, v198160.940.8(0.01)0–8.38.0(0.02)0–26.756
AP7000.90v, r249170.915.4(0.01)0–14.312.0(0.04)0–53.858
Mean     0.934.2  8.9   

For adult plants putatively assigned as E. aggregata (140 trees from 18 populations), classification based on the admixture proportions resulted in 129 (92.1%) assigned as pure-bred E. aggregata, 8 adults as F1 hybrids (5.7%), and 3 (2.1%) as hybrid backcrosses. Within reference populations of the congeners, all 25 E. rubida trees were classified as pure-bred, and of the 25 E. viminalis trees, 1 (4%) was classified as a backcrossed hybrid.

population sizes and hybridization rates

Nonlinear regressions of all E. aggregata sites showed that the mean (F1) hybridization rate was moderately and negatively correlated with absolute population size (APS) (R2 = 0.35, P = 0.007; Fig. 3a) and strongly negatively correlated with relative population size (RPS) (R2 = 0.59, P < 0.001; Fig. 3b), with the latter best modelled with a power function with a negative exponent (y = 1089.89x−1.47). When limiting the analysis to E. aggregata sites sympatric with E. viminalis, there remained a strong and significant negative correlation between hybridization rates and RPS (R2 = 0.88, P = 0.004; Fig. 3c) which was best modelled with a power function with a negative exponent (y = 730.80x−1.26). However, for these same sites we did not detect a significant correlation with APS (P = 0.09). Nonlinear regressions of all E. aggregata sites sympatric with E. rubida showed that the mean hybridization rate was not significantly correlated with APS (P > 0.05) or RPS (P > 0.05; Fig. 3d). Using the mean admixture as an alternative measure of hybridization rates revealed similar trends, with mean admixture negatively correlated with APS (R2 = 0.27, P = 0.02) and RPS (R2 = 0.41, P < 0.003). Regressions of absolute and relative population sizes of E. aggregata combining both F1 and backcrossed hybrids, and run separately for sites sympatric with E. viminalis and E. rubida showed similar trends to the above analysis for all comparisons (data not shown).

Figure 3.

Relationship between the average percentage of F1 hybrid production within open-pollinated seed arrays from 17 E. aggregata populations, against the absolute population size of E. aggregata adults (APS) and the relative population size of E. aggregata adults to its congeners (RPS). Separate nonlinear regressions were used for; (a) APS for all populations, (b) RPS for all populations, (c) RPS for populations sympatric with E. viminalis, and (d) RPS for populations sympatric with E. rubida. Populations sympatric with E. viminalis are indicated with filled circles (inline image), E. rubida with open circles (inline image), and E. dalrympleana with open squares (inline image). Bars indicate standard errors of the mean at each population. n.s. P > 0.05 (not significant); *P < 0.05.

genetic diversity and outcrossing rates

A number of genetic diversity measures displayed trends with population parameters and hybridization rates. The mean number of alleles (A) for E. aggregata ranged from 1.16 in a small remnant population consisting of a single tree to 3.8 in one of the largest populations (Table 3). Adjusting for difference in sample sizes, the allelic richness (Ar) displayed no relationship with APS (P > 0.05), but there was a linear relationship with average hybridization rate (R2 = 0.29, P = 0.01) and a negative power relationship with RPS (R2 = 0.52, P < 0.01; y = 2.66x−0.0035). Similarly, genetic diversity (He) displayed a significant linear relationship with average hybridization rate (R2 = 0.42, P = 0.01) and a negative power relationship with RPS (R2 = 0.42, P < 0.01; y = 0.42x−0.054), and there were no relationships with APS (P > 0.05).

Table 3.  Adult population size parameters (APS, RPS), genetic diversity estimates (P, A, f, Ar, He) and outcrossing rates (tm) estimated from six allozyme loci of progeny arrays collected from Eucalyptus aggregata populations (n = 18)
  1. APS, absolute population size of E. aggregata adults; RPS, the relative population size of E. aggregata to congeners.

  2. Genetic diversity estimates, P = number of polymorphic loci; A = mean number of alleles per locus; Ar = mean allelic richness; He = unbiased expected heterozygosity (gene diversity); f = inbreeding coefficient; tm = multi-locus outcrossing rate; SE = standard error of multi-locus outcrossing rate.

Mean  89.002.902.140.310.060.89 

The inbreeding coefficient (f) ranged from −0.16 to 0.23, with linear regressions indicating no relationship with either APS, RPS or hybridization rates (P > 0.05). The majority of E. aggregata populations were highly outcrossed with an average level of multi-locus outcrossing (tm) for E. aggregata of 0.91, with populations ranging from 0.59 to 1.2 (Table 3). However, no significant relationships were detected between outcrossing rate and either APS or RPS (P > 0.05).

seed production

Among populations of E. aggregata, there was a substantial variation in several seed production parameters, with the average seed weight ranging from 4.3 to 7.2 mg and the number of seeds ranged from 0.8 to 2.8 seeds per capsule. There was evidence of substantial seed abortion as seed comprised less than half the capsule contents by weight (16.9 to 38.4%), with the remaining contents consisting of chaff and likely aborted seed. Among seven population variables, only relative population size was significantly correlated with seed production (Table 4). On average, populations with higher relative population sizes (E. aggregata numerically dominant) tended to produce more seed (R2 = 0.29, P = 0.03; Fig. 4a) and seed comprised a greater percentage of the total weight of capsule contents (R2 = 0.30, P = 0.03; Fig. 4b). However, there were no significant linear relationships between any population variables and either seed weight or debris weight percentage (P > 0.05; Table 4). There was an almost significant negative relationship between the inbreeding coefficient and seed weight percentage (R2 = 0.29, P = 0.06), however there were no significant relationships between any seed production variables and outcrossing rates (tm; P > 0.05), allelic richness (Ar; P > 0.05), or genetic diversity (He; P > 0.05).

Table 4.  Summary of separate linear regressions between population components and measures of seed production and seedling performance. Seed production and seedling performance calculated from population averages (across n = 2 to 20 families) at each of 14 E. aggregata populations
Population componentNumber of seedsSeed weight %Germination %Survival %Plant heightLeaf pairs
  • n.s. P > 0.1 (not significant), (*) P < 0.1, *P < 0.05, **P < 0.01.

  • The direction of the slope (+/−) is given for pairs with significant terms. APS = absolute population size of adult E. aggregata; RPS = relative population size of E. aggregata compared to its congeners; H% = average percentage of F1 hybrid production in progeny arrays; (Ar) mean allelic richness; (He) unbiased expected heterozygosity (genetic diversity); f = inbreeding coefficient; tm = multi-locus outcrossing rate.

  • Additional measures of seed weight and seed debri % were not significantly related to any population component.

APS n.s. n.s. n.s.n.s. n.s. n.s.
RPS+0.29*+0.30*+0.27*+0.33** n.s. n.s.
H% n.s. n.s.0.18(*)0.16(*) n.s.+0.27*
Ar n.s. n.s. n.s. n.s. n.s.+0.20(*)
He n.s. n.s. n.s. n.s. n.s.+0.22*
f n.s.0.22(*) n.s. n.s.0.29* n.s.
tm n.s. n.s. n.s. n.s. n.s. n.s.
Figure 4.

Relationship between the relative population size of E. aggregata to its congeners (RPS) and the average seed production of seed arrays (n = 13 populations). Linear regressions were fitted between RPS and; (a) the average number of seeds, (b) the average weight of seed as a percentage of the weight of total capsule contents. *P < 0.05.

germination, survivorship and seedling performance

Seed germination, mortality and the number of viable seed (per 10 g) varied substantially between E. aggregata populations. Germination ranged from 41% at NG to 80% in population M, survivorship from 51% at RS to 94% at MW, and the number of viable seed per 10 g of capsule contents ranged from a population mean of 11 660 ± 8246 at WL to 4520 ± 3040 at RS. Among the seven population variables, only relative population size was significantly correlated with seed germination and seedling survival (Table 4). There were significant and positive linear relationships between RPS and germination rates (R2 = 0.27, P = 0.02; Fig. 5a), seedling survival (R2 = 0.33, P = 0.02; Fig. 5b), and the number of viable seeds per 10 g (R2 = 0.28, P = 0.02; data not shown).

Figure 5.

Relationship between the relative population size of E. aggregata to its congeners (RPS) and germination and survivorship of seed arrays (n = 14 populations). Linear regressions were fitted between RPS and; (a) percentage of seeds germinated, (b) the percentage of seedlings survived to 11 months of age. *P < 0.05.

There was substantial variation in average measures of seedling performance amongst populations. Seedling height ranged from 419 mm (RV) to 578 mm (RC), and the number of leaf pairs ranged from 35.0 (FV) to 67.3 (ST). Only the inbreeding coefficient (f) was significantly related to plant height (Table 4), because populations with more heterozygotes tended to have taller seedlings (R2 = 0.27, P = 0.02). Populations with higher hybridization rates (H%) and gene diversity (He), tended to have more leaf pairs (H%, R2 = 0.27, P = 0.03; He, R2 = 0.22, P = 0.05).


Reductions in absolute and relative population size of plant species within an area can have implications for hybridization rates and the risk of local extinction of many rare species. Our results indicated a consistent negative relationship between relative population size and hybridization rates, and associated declines in seed production, germination and seedling survival when E. aggregata was less numerous than its congeners. As a consequence of increased hybridization, both demographic and genetic swamping is likely to be occurring in E. aggregata sites with low relative population sizes (i.e. RS, WL, ST).

Some Eucalyptus species have a high rate of hybrid production (E. aggregata here 4.2–8.9%) compared to a number of other genera (e.g. Iris (< 1%), Phlox (< 1%), Senecio (< 1%), and Brassica (1.5%)) (Rieseberg & Carney 1998), which may suggest weaker reproductive barriers exist in this genus. In comparison to other Eucalyptus species, the average percentage of hybridization for E. aggregata is similar to those described in E. argutifolia (6%; Kennington & James 1997). However, they are greater than the rates in natural populations reported in a review of 13 Eucalyptus species, where F1 hybrid production ranged from 0.03% to 3.5% (Potts et al. 2003). Furthermore, the two largest E. aggregata populations exhibited hybridization rates (AP 5.4%; BT 3.3%) higher than those reported in the largest population of E. argutifolia which similarly used allozymes to identify hybrids (< 0.1%; Kennington & James 1997). This could reflect weaker pre-mating barriers to hybridization in E. aggregata or that the sampled populations exhibited lower relative population sizes compared to Eucalyptus populations sampled in earlier studies.

hybridization and relative population size

Our results indicate that reduction in population size may facilitate hybridization in the uncommon E. aggregata. Relative population size was strongly correlated with rates of hybrid seed production for E. aggregata populations when examining all sites, and for those sites sympatric with E. viminalis. Importantly, these relationships were consistent for alternative methods of estimating hybridization rates with the inclusion of backcrossed hybrids with F1 hybrids or by using mean admixture rate. For all tests, relative population size proved to be more informative than the absolute population size as a predictor of hybridization rates, which likely reflects the importance of relative population size as a measure of the potential ratio of interspecific to intraspecific pollen. As such, pollen swamping is likely occurring within the sites with low relative population sizes (RS, WL, ST), because (i) the potential sources of interspecific pollen (E. viminalis, E. dalrympleana, E. rubida) outnumber the sources of intraspecific pollen for E. aggregata trees, and (ii) the dominant insect foraging at the flowers of these species was the generalist Honeybee (Apis mellifera). The occurrence of pollen swamping within the small hybrid producing sites is further evident with all of the trees sampled at the small sites (WL, NG, ST, MW) producing hybrid offspring, in contrast to the largest populations (FV, BT, AP) where 50% to 58% of the trees at these sites produced hybrids.

The increase in hybridization rates as relative population size declines follows theoretical model predictions (Pederson et al. 1969) and is in line with empirical expectations from two source-sink experimental arrays (Klinger et al. 1992; Richards et al. 1999). While few natural population studies have explicitly measured relative population size, similar patterns have been observed in Nothofagus, with higher hybridization rates for isolated trees surrounded by compatible species (hybrid rate; 80%) compared to trees within dense conspecific stands (0%; Gallo et al. 1997; Marchelli & Gallo 2001). In an allozyme study of E. argutifolia, the small populations exhibited the highest hybridization rates in progeny arrays (47%; Kennington & James 1997). Similarly, estimates of hybrid rates with morphological identification have reported increased hybridization in progeny arrays from small populations compared to large populations for E. agygdalina (small 14% vs. large 3%) and E. morrisbyi (small 17% vs. large 1.6%) (Potts & Wiltshire 1997).

Considering the similar flower size and structure of each of the Eucalyptus species studied here, pollinator behaviour is unlikely to be a strong pre-mating barrier to interspecific pollen flow. Behavioural studies have found Honeybees forage as generalists (Goulson 2003), exhibiting low fidelity in mixed species arrays, especially when co-occurring plant species have similar flower size and structure (Kunin 1993). In other Eucalyptus species, evidence of past hybridization has been identified from extensive chloroplast sharing across geographic regions (McKinnon et al. 2001). In our study, the presence of large hybrid adults and likely advanced backcrossing suggests that hybridization probably predates the arrival of the introduced Honeybee to Australia in the early 19th Century (Goulson 2003). Interestingly, some studies have reported less interplant movements by Honeybees compared to other bee species (Goulson 2003), which suggests that the introduction of Honeybees could alter patterns of interspecific pollen flow. Therefore, it would be interesting to examine the foraging behaviour of insect visitors to these Eucalyptus species to ascertain the importance of Honeybees and native bees for current interspecific pollen dispersal patterns in remnant populations.

Correlations were detected between relative population size and hybrid production separately for sites sympatric with E. viminalis but not for sites with E. rubida. This could be due fine-scale effects of local plant density and plant size on pollinator behaviour and plant mating patterns (Thomson 1981; Field et al. 2005), a lack of E. rubida sympatric sites with lower relative population sizes (RPS < 0.5), or stronger pre-mating barriers to hybridization with E. rubida compared to E. viminalis. The latter could be confirmed with experimental crosses, and if pre-mating barriers were stronger with E. rubida then optimal conservation strategies would depend on which species were present within a given area.

genetic diversity and outcrossing rates

Absolute population size was not a good predictor of genetic diversity in seed cohorts, suggesting that small E. aggregata remnants may not have undergone reductions in genetic diversity following habitat fragmentation. However, the positive correlation between hybrid production and two genetic diversity measures (Ar and He) indicates that hybridization could be playing a role in maintaining genetic diversity through the introduction of new genetic material from sympatric species. Considering the parallel negative correlations detected between relative population size and both genetic diversity and hybrid production, elevated hybrid production and introgression could be masking declines in genetic diversity in small remnants. Nevertheless, a lack of correlation between measures of population size and genetic diversity has been found in the majority of Eucalyptus studies (Potts & Wiltshire 1997). In contrast to the traditional model of reduced genetic diversity in small populations, Potts & Jackson (1986) argue that in Eucalyptus, hybridization may play a role in maintaining high diversity in remnant populations. This has been suggested as an explanation for high genetic diversity in small populations of rare species such E. rameliana (Sampson et al. 1995) and E. benthamii (Butcher et al. 2005). While our study was based on a limited number of loci (six), our findings suggest hybridization can play an important role in remnant populations by not only maintaining genetic diversity, but through increasing diversity above levels observed in large populations.

Outcrossing rates in E. aggregata were moderate to high across populations and are similar to rates described with allozyme markers across a range of Eucalyptus species (Potts & Wiltshire 1997). Many of the populations with slightly depressed outcrossing rates (e.g. M, H, FV, RV) also exhibited deficits in heterozygotes (f > 0) indicating a departure from panmictic mating and some level of inbreeding in the seedling cohort. In contrast, the populations with the highest hybrid rates exhibited random mating or a slight excess of heterozygotes (ST, WL, RS) that would be expected from the influx of alleles from E. viminalis, E. rubida and E. dalrympleana. The lack of relationship between population size and either inbreeding (f) or outcrossing rates (tm), suggests that even small populations are able to avoid potentially deleterious inbreeding and maintain the output of high numbers of outcrossed progeny. While relationships between outcrossing rates and population size have been reported previously in Eucalyptus (Sampson et al. 1989; Hardner et al. 1996), as with our study, others have also found no relationship (Kennington & James 1997; Butcher et al. 2005). The lack of relationship may be due to variation in the frequency of male sterile individuals in a population (Ellis & Sedgley 1992b), or abortion of increased numbers of selfed seeds in smaller populations (Kennington & James 1997). The maintenance of substantial numbers of outcrossed progeny is important for the viability of remnant populations as the risk of genetic swamping may be lessened due to hybrid offspring competing with potentially superior outbred seedlings (Lopez et al. 2000).

seed production

Among an array of population parameters, relative population size of E. aggregata compared to its congeners had the strongest influence on seed production, with fewer seed set as relative population size declined. The strong negative correlation between hybridization rates and relative population size, suggests that elevated interspecific pollen flow onto E. aggregata trees may be responsible for depressed seed production. While these species are clearly cross-compatible and the identification of fertile hybrid adults at these sites indicates some proportion are viable, the declines observed in our study could be due to hybrid breakdown. Previous work in Eucalyptus indicates that in some cases, the success of interspecific pollinations can be lower than intraspecific crosses (Griffin et al. 1988; Ellis et al. 1991). Declines in hybrid seed production are often attributed to pre-zygotic barriers to interspecific pollen in the form of structural and physiological barriers in the style that reduce the frequency of pollen tubes penetrating the ovule (Gore et al. 1990; Ellis et al. 1991). The role of hybrid breakdown in seed production in E. aggregata could be confirmed by comparing the success of controlled crosses between E. aggregata and its congeners with outcrossed pollinations between E. aggregata individuals. If a higher number of hybrid pollinations fail compared to outcrossed pollinations, this would suggest that increased interspecific pollen flow followed by early hybrid breakdown is responsible for declines in seed production as the relative population size decreases.

The lack of significant relationships between seed production and levels of inbreeding, genetic diversity and absolute population size are surprising since such relationships have been inferred in a number of systems (Oostermeijer et al. 1994; Kunin 1997; Kery et al. 2000) including Eucalyptus (Burrows 2000, but see; Butcher et al. 2005). Declines in seed set with population size are often attributed to reductions in pollen quality (selfing) or pollen quantity due to overall reductions in the number of pollinators attracted to smaller populations. While these factors may play a role in our study, the high outcrossing rates within all populations except the single isolated tree (population MC) suggests pollinator limitation is unlikely a cause at these populations. Alternatively, declines in seed set could be due to increased levels of self pollen, but the production of substantial seed set through selfing with single isolated trees (population MC) suggests that E. aggregata is to some extent self-compatible. However, variation in self-compatibility among individual trees has been reported in Eucalyptus (Ellis & Sedgley 1992a) suggesting that experimental pollinations may be required to determine if pollen quality is contributing to depressed seed production for small E. aggregata populations. In our case, it could be that declines in seed set due to increased interspecific hybridization and potential hybrid breakdown may be more severe than the effects of pollen quantity and quality on seed production.

germination, seedling survival and performance

Germination and survivorship showed a similar negative correlation with relative population size which paralleled the relationship with seed production. Reduced germination rates and survivorship in populations of low relative size suggests that these trees are producing a greater number of inviable seeds and the remaining seedlings have poor vigour. Populations of E. aggregata out-numbered by congeners tended to have less than 5000 viable seeds per 10 g (exception WL) compared to over 8000 for most of the populations when E. aggregata was dominant. Similar declines in viability have been detected for several other Eucalyptus species, for example, E. melliodora trees exhibited averages of 7162 viable seed (per 10 g) for in large woodland populations which declined to 3192 for isolated trees (Burrows 2000). Such declines in seed viability in Eucalyptus and other genera have been attributed to increased geitonogamous pollinator movements (Kunin 1997) and subsequent depressed seedling growth and survival due to inbreeding depression (Borralho & Potts 1996; Hardner et al. 1996).

Declines in germination and survivorship for remnant populations of E. aggregata could be due to a number of factors including post-zygotic selection against increased numbers of hybrid individuals (Lopez et al. 2000), or increased deleterious inbreeding (Borralho & Potts 1996). Considering relative population size was the most consistent parameter related to germination and seedling survival, this indicates that increases in the number of inviable hybrids could have an important role in reducing seedling viability and performance. Hybrid breakdown can be caused by a range of mechanisms including the deleterious and complementary action of genes (Orr 1995), cytoplasmic epistasis (Tiffin et al. 2001), and in later generations the disruption of co-adapted gene complexes (Rieseberg & Carney 1998). In Eucalyptus, reports have described both slower germination and poorer survival of hybrid individuals compared with out-bred individuals (Lopez et al. 2000). However there are also cases of hybrid germination rates as successful as those of intraspecific seed, at least in the first generation (Tibbits 1988; Ellis et al. 1991; Potts et al. 1992). In E. aggregata, the almost significant negative correlation between hybrid production and seedling germination and survival suggests hybrid breakdown could be an important factor, and could be confirmed with experimental fitness trials of controlled crosses.

There were two trends in seedling performance: a negative correlation between height and inbreeding, and a positive correlation between leaf pairs, hybridization rates and genetic diversity. The first trend is likely due to the effects of inbreeding depression on plant growth, as a number of studies in Eucalyptus have similarly reported declines in the fitness of open-pollinated progeny as outcrossing rates declined (Borralho & Potts 1996; Hardner et al. 1996), and there are direct reports of depressed growth of inbred compared to outbred progeny (Griffin & Cotterill 1988; Hardner & Potts 1995; Lopez et al. 2000). The positive correlation between the number of leaf pairs and both hybridization rate and genetic diversity is probably due to the effects of hybridization and introgression from E. viminalis, which has significantly greater numbers of leaf pairs than E. aggregata (Field et al. 2008). This has important implications for the viability of remnant populations and implies that E. aggregata populations of small relative size have been subject to introgression of E. viminalis traits.

implications for conservation

These results demonstrate elevated levels of hybrid production in fragmented populations, and that for some species pairs, the relative frequencies of the parental species can be a strong predictor of hybridization rates. The negative correlation between relative population size and hybrid production, the substantial numbers of backcrossed hybrids and declines in fecundity suggest that hybridization could represent a significant threat to population viability through both demographic and genetic swamping. Despite the potential deleterious effects of demographic and genetic swamping, the potential adaptive benefits of hybridization should not be ignored. The observed increases in genetic diversity in remnant populations through hybridization have the potential to rapidly provide new multi-locus genotypes that may be required by remnant populations to survive and re-establish in changing environments.

Our study suggests that the maintenance of pure-bred in situ populations or collection of mostly pure-bred ex situ seed for reforestation should focus on preservation and sampling from higher relative population sizes > 0.5 (i.e. at least equal frequencies of parentals). If seed are required to be collected from sites with lower relative population sizes, our results suggest twice as many capsules may need to be collected to offset the losses due to depressed seed production, seedling germination and survival.


Authors thank Linda Broadhurst, Tony Brown, Leanne Cox, Robert Godfree, Liz Gregory, and Melinda Pickup for assistance. This work was conducted while (D.L.F.) was receiving a University of Wollongong Post-graduate Research Award and ‘top-up’ scholarship from CSIRO Plant Industry.