Patterns of seed dispersal and pollen flow in Quercus garryana (Fagaceae) following post-glacial climatic changes


*Jessica J. Hellmann, 107 Galvin Life Sciences, Department of Biological Sciences, University of Notre Dame, Notre Dame, IN 46556, USA. E-mail:


Aim  We examined the genetic structure of Quercus garryana to infer post-glacial patterns of seed dispersal and pollen flow to test the hypotheses that (1) peripheral populations are genetically distinct from core populations and from one another; (2) genetic diversity declines towards the poleward edge of the species’ range; and (3) genetic diversity in the chloroplast genome, a direct measure of seed dispersal patterns, declines more sharply with increasing latitude than diversity in the nuclear genome. We address our findings in the context of known historical oak distribution from pollen core data derived from previously published research.

Location  Oak–savanna ecosystems from southern Oregon, USA (core populations/non-glaciated range) northward to Vancouver Island, British Columbia, Canada (peripheral populations/glaciated range).

Methods  We genotyped 378 trees from 22 sites with five chloroplast and seven nuclear microsatellite loci. For both sets of markers, we estimated genetic diversity and differentiation using an analysis of molecular variance and generated Mantel correlograms to detect genetic and geographical distance correlations. For the nuclear markers, we also used a Bayesian approach to infer population substructure.

Results  There was a large degree of population differentiation revealed by six chloroplast haplotypes, with little (≤ 3) or no haplotype diversity within sites. Peripheral island locations shared the same, maternally inherited chloroplast haplotype, whereas locations in mainland Washington had greater haplotype diversity. In contrast, genetic diversity of the nuclear markers was high at all locations sampled. Populations clustered into two groups and were significantly positively correlated over large spatial scales (≤ 200 km), although allele richness decreased significantly with latitude. Population substructure was observed between core and peripheral populations because rare alleles were absent in peripheral localities and common allele frequencies differed.

Main conclusions  The observed pattern of chloroplast haplotype loss at the northern periphery suggests restricted seed dispersal events from mainland sites to peripheral islands. This pattern was unexpected, however, as refugial oak populations remained near the current post-glacial range even during the Last Glacial Maximum. Using nuclear markers, we found high within-population diversity and population differentiation only over large spatial scales, suggesting that pollen flow is relatively high among populations.


Colonization from elsewhere is the main reason that previously glaciated landscapes have floristic diversity (Hewitt, 1999). Organisms that have undergone such a post-glacial range shift usually show a genetic signature of that migration (Taberlet et al., 1998). We studied Quercus garryana Dougl. ex Hook. (section Quercus, Oregon white oak or Garry oak) in the Pacific Northwest of North America in order to determine the genetic effects of colonizing peripheral island locations after the Last Glacial Maximum (LGM). In contrast to post-glacial colonization in Europe, much of the Q. garryana range was not glaciated at the LGM (Waitt & Thorson, 1983; Fig. 1), providing an opportunity to investigate the signature of post-glacial colonization for a taxon that did not shift large distances after the LGM.

Figure 1.

Quercus garryana species range map inset into study area in the Pacific Northwest. Numbered dots are sampling sites (= 22). Capital letters are pollen sediment sampling sites from the literature (= 12). Estimate of ice sheet extent at the LGM drawn from Waitt & Thorson (1983). Range map inset reprinted from Burns & Honkala (1990). Base map created using GPSVisualizer (

For most angiosperm taxa, the path of post-glacial colonization has been examined using maternally inherited chloroplast markers because they provide information only on the dispersal of seed (Birky, 1976; Dumolin et al., 1995). Patterns of population differentiation for oak species observed using chloroplast markers have typically been attributed to population responses during and following the period of highly variable climates of the Pleistocene (Petit et al., 1993b; Dumolin-Lapègue et al., 1997; Grivet et al., 2006). Typically, the highest genetic diversity is found within the refugial range (Hewitt, 2000). Genetic diversity is often lost along the migratory path, permitting inference of the path and source populations through genetic analysis of the descendant populations. Diversity is initially lost because post-glacial recolonization is not from a random draw of seed (Feng et al., in press). Instead, dispersing seeds are drawn most frequently from the poleward edge of the shifting distribution. Once a post-glacial region is colonized, newly established populations often limit establishment from other sources by filling the niche space (Taberlet et al., 1998; Hewitt, 2000). Further, it is likely that a few colonizing individuals contribute the bulk of seed to newly established populations because colonization events are infrequent (Petit et al., 1997). Under this model, one expects low genetic diversity within post-glacial populations and an overall loss of genetic diversity with increasing geographical distance from the original source (Cwynar & MacDonald, 1987; Dumolin-Lapègue et al., 1997).

Nuclear markers, in contrast, provide a biparentally inherited view of biogeographical patterns, incorporating both pollen flow and seed dispersal. Wind pollination in trees can be effective over great distances, and pollen donors may frequently come from outside a recipient population (Liepelt et al., 2002; Bacles et al., 2005; Craft & Ashley, 2006, 2007). High within-site nuclear diversity is a common finding for oaks, and it is usually attributed to long-distance pollen flow (Dow & Ashley, 1996; Streiff et al., 1999; Aldrich et al., 2003b). For wind-pollinated species, there is also evidence that pollen flow can be enhanced in a fragmented or open landscape, as would be found in the wake of a retreating glacier (Bacles et al., 2005; but see Sork et al., 2002). For example, in an open, fragmented landscape, Craft & Ashley (2007) could not detect population structure for Quercus macrocarpa over 200-km geographical distances. In addition, Muir et al. (2004) found little nuclear differentiation among peripheral populations of Quercus petraea in sites in Ireland that were island colonizations and now in a highly fragmented landscape. This high within-population genetic diversity and low population differentiation has the effect of obscuring source populations of post-glacial locations while mitigating potentially deleterious genetic consequences of small founding populations.

In many systems, species are thought to have travelled great distances from their glacial refugia to colonize their post-glacial ranges (Brewer et al., 2002). European oaks were relegated to refugia south of the Alps, a major migratory barrier (Hewitt, 1999; Petit et al., 2002), and chloroplast genetic signatures reflect the loss of diversity with northward migration in distinct latitudinal lineages (Grivet et al., 2006). The pattern in eastern North America was thought to be similar to that in Europe, with temperate species restricted to southern refugia, but gaps in fossil data made determining glacial distributions difficult (Jackson et al., 2000). Evidence is emerging that oak and other temperate species may have maintained refugial populations relatively close to the ice sheet, potentially allowing some species to persist at low densities much further north than the pollen or macrofossil records suggest (McLachlan & Clark, 2004; Magni et al., 2005; McLachlan et al., 2005; Loehle, 2007; Birchenko et al., in press). In the Pacific Northwest of North America, species distributions were greatly influenced by the Laurentide ice sheet much further east. By the time the Cordilleran ice lobes were reaching their greatest extent on Vancouver Island and in the Puget Trough, temperate communities were advancing towards the ice (Whitlock, 1992). Maximum ice extent clearly did not restrict taxa to distant refugia. In addition, the Cordilleran ice sheet did not reach its maximum extent in the lowlands until alpine glaciers were retreating (Waitt & Thorson, 1983). The varied topography of the Pacific Northwest, including its north-to-south-trending mountain ranges, probably allowed small populations to wait out unfavourable conditions of the LGM.

We used pollen records to complement our phylogeographical study because they provide a temporal and spatial context for interpreting results (Cruzan & Templeton, 2000; Petit et al., 2002; López de Heredia et al., 2007). Pollen record data indicate that oak populations were present at low densities in Oregon and Washington before and during the Fraser glaciation (25–10 kyr bp, Fig. 1; Table 1), a period when lowland ice lobes reached their maximum extent (15 kyr bp). These data suggest that during what would appear to be region-wide unfavourable conditions (15–14 kyr bp), Q. garryana was not extirpated from even the northern parts of its mainland geographical range (compare Fig. 1 and Table 1). For example, high pollen levels were found at Mosquito Lake Bog in northern Washington (48.77°N, 122.12°W, 198 m) from 15–9 kyr bp, providing evidence that Q. garryana maintained populations close to the ice sheet (Hansen & Easterbrook, 1974). The only portion of the current range of Q. garryana that was glaciated during the LGM is the Puget Sound region and San Juan Islands of Washington northward into British Columbia (Brubaker, 1991; Brown & Hebda, 2002). As the Juan de Fuca lobe of the Cordilleran ice sheet retreated, marked increases in oak at most peripheral sites occurred from 10–4 kyr bp due to regional warming and drying (Barnosky, 1985a; Sea & Whitlock, 1995). Quercus garryana probably colonized Vancouver Island at this time from nearby locations, as the oldest oak pollen record from Vancouver Island is from just under 11 kyr bp.

Table 1.   Presence of Quercus garryana pollen in lake sediments by 500-year interval. Thumbnail image of

One previous study has examined the genetic diversity of Q. garryana across glaciated and unglaciated regions. This research used allozyme markers and suggested a weak but statistically significant increase in genetic divergence among populations with increasing physical distance (Ritland et al., 2005). Allozyme polymorphisms indicate that populations on Vancouver Island and the Gulf Islands are more genetically similar to each other than they are to the populations from the species core, and that both allele richness and heterozygosity decline with increasing latitude.

In this study we investigated the biogeography of Q. garryana using microsatellite markers in the chloroplast and nuclear genomes. Microsatellites have the advantage of high levels of polymorphism at many loci, allowing for a fine-scale analysis of local and regional patterns of genetic diversity (Degen et al., 1999), and maternally inherited chloroplast markers have the potential to reveal patterns of seed dispersal since the LGM. Thus, this study expands on that of Ritland et al. (2005) in that we include the biogeographical distribution of chloroplast haplotypes and have greater statistical power for revealing additional mechanisms for the patterns observed.

Our motivation for studying the biogeography of Q. garryana was to infer seed dispersal and pollen-flow patterns to understand the post-glacial colonization of peripheral regions, particularly where refugial populations were in close proximity to the post-glacial expanded range. Based on previous research and population models, we had different expectations for each marker system. We expected the genetic differentiation across sampling sites to be higher in the chloroplast markers than in the nuclear markers due to the strictly maternal inheritance of the chloroplast markers and the difference in pollen flow and seed dispersal (Petit et al., 1993a; Ennos, 1994; Grivet et al., 2008). Because known glacial refugia were close to the post-glacial colonization sites, we expected this difference to be less than is typically found for oak species, resulting from relatively high seed dispersal to areas that now represent the furthest reaches of the species’ range. However, the unstable climatological history of the region made it difficult to predict how many recognizable regional clusters we would find.

Specifically, we sampled Q. garryana populations along a latitudinal gradient from the species’ core to the poleward periphery, focusing on three questions: (1) Are peripheral populations genetically distinct from core populations and from one another? (2) Is genetic diversity reduced in northern peripheral populations relative to core locations? (3) Do the nuclear and chloroplast data indicate different patterns of pollen flow and seed dispersal? We examine our results in the context of the pollen sediment record for Q. garryana in order to understand how post-glacial colonization results in current spatial genetic patterns.

Materials and methods

Study species and study region

Quercus garryana is the sole oak species in oak–savanna ecosystems of the Pacific Northwest region of North America. The current distribution of Q. garryana ranges from isolated mountain areas in the Sierra Nevada, California, to the coastal regions and valleys of northern California and Oregon, along the Columbia River region of Oregon and Washington, the Puget Trough, Washington, and southern British Columbia (Fig. 1 inset). It is the only native oak species north of southern Oregon (Glendenning, 1944). At the northern range boundary on Vancouver Island, British Columbia, populations occur in fragmented patches surrounded by coniferous forest. At its extreme northern limit, Q. garryana occurs on rocky outcrops and on southerly and westerly facing slopes where Douglas fir (Pseudotsuga menzesii) grows poorly. Towards the centre of the species’ range, populations of Q. garryana are less isolated and fragmented, sometimes forming extensive woodlands.

These oak habitats harbour high botanical diversity and are threatened by coniferous encroachment, non-native species and anthropogenic disturbance (Thysell & Carey, 2001). Quercus garryana and its associated communities are of conservation concern, particularly in the northern portion of its geographical range (Ward et al., 1998). Under current predictions of climate change, oak–savanna ecosystems are expected to expand their geographical ranges northward on Vancouver Island into areas currently occupied by coniferous forest (Hamann & Wang, 2006).

Sample collection

Twigs of Q. garryana were collected from 22 sampling sites throughout the northern half of the species range in May 2005 and February and April 2007 (Fig. 1). Sampling locations included Siskiyou Mountain and Rogue River regions, Willamette Valley, Puget Trough, the San Juan Islands, Vancouver Island and Hornby Island. At each sample site, twigs were collected from nine to 22 trees at least 20 m apart. Twigs collected in 2005 were shipped to the laboratory on water ice and frozen at −80°C until DNA extraction. In 2007, most twigs were cut and dried on silica gel to maintain integrity of the DNA, although a small subset of the 2007 samples were shipped fresh to the laboratory.

DNA extraction and genotyping

We ground fresh, frozen and dried twigs under liquid nitrogen using a Freezer/Mill 6850 (Spex CertiPrep Group, Metuchen, NJ, USA). Genomic DNA was extracted using the DNeasy Plant Mini Kit (Qiagen, Valencia, CA, USA) according to the manufacturer’s protocol with the following modification: twig powder in extraction buffer was shaken in a tube with a 1/4-inch cylindrical ceramic bead (Qbiogene, Montreal, Canada) in a FastPrep FP 120 (Qbiogene) for three cycles of 15 s each to increase cell lysing.

We used five chloroplast microsatellite markers developed for Quercus petraea (Mattuschka) Liebl. and Quercus robur L. (Deguilloux et al., 2003) to reveal chloroplast haplotypes. Seven nuclear microsatellite markers, five developed from Quercus rubra L. (Aldrich et al., 2002, 2003a) and two from Q. petraea (Steinkellner et al., 1997), were used to amplify microsatellite loci in Q. garryana (see Appendix S1 and Table S1 in Supporting Information). We performed the polymerase chain reaction (PCR) for both chloroplast and nuclear markers in a total volume of 25 μL containing 5 U GoTaq Flexi (Promega, Madison, WI, USA), 1× colourless GoTaq Flexi buffer, 2.5 mm MgCl2, 200 nm each dNTP (Fermentas, Burlington, Canada), and 5 pm forward and reverse primer. The PCR protocol consisted of a 2-min initial denaturing step at 95°C followed by 35 cycles of 30 s at 95°C, 30 s at marker-specific annealing temperatures (Table S1), and 30 s at 72°C, then a 10-min final elongation at 72°C. Microsatellite fragment lengths were detected using forward primers labelled with WellRED dye (Sigma-Aldrich, St Louis, MO, USA) on a CEQ 8000 Genetic Analysis System (Beckman Coulter, Fullerton, CA, USA). For each marker, we ran samples once on plates with positive and negative controls, to determine if the fragment lengths were accurate and free of contamination. If amplified fragments were difficult to score, PCR was performed again for the given sample and marker combination. Samples that did not amplify or that remained too difficult to score accurately were discarded. Over all repeated sample and marker combinations we had a genotyping error rate of 0.5%.

Statistical analyses

We used Arlequin 3.1 (Excoffier et al., 2005) to examine population substructure through an analysis of molecular variance (AMOVA) (10,000 permutations) for the chloroplast microsatellite data. Also, we determined within-population diversity (HS), overall diversity (HT), and population differentiation (GST) using the program cpSSR ( following Pons & Petit (1995, 1996).

For the nuclear data, we used Arlequin 3.1 to test for departures from Hardy–Weinberg (HW) equilibrium, generate observed (Ho) and expected (He) heterozygosity, and examine population substructure using AMOVA (10,000 permutations). Micro-checker (van Oosterhout et al., 2004) was used to investigate the possibility of null alleles at each locus for each sample location. This program randomly assigns alleles to genotypes and compares the random assignments with the observed allele frequencies. If the marker shows HW disequilibrium, and an excess of homozygotes is spread evenly across the alleles, Micro-checker makes a probabilistic determination of null alleles. In addition, we obtained null allele frequency using the Brookfield 1 estimate (Brookfield, 1996) in Micro-checker.

For the nuclear data, we used Structure 2.2 (Pritchard et al., 2000) to investigate population substructure using a Bayesian approach that calculates the likelihood of the data given K groups. We evaluated K as= 1 to = 10, and the value of K with the highest log-likelihood probability was selected as the most likely to represent real groups. We ran a set of models in which individuals could have admixed ancestry and independent allele frequencies (λ = 1). We replicated runs for each of the ten models (= 1–10) a minimum of three times to ensure consistent natural log probabilities, and ran each model with a 100,000-step burn-in phase and a 1,000,000-step Markov chain Monte Carlo data collection. We then ran AMOVA with the regional groupings defined in Structure to investigate the proportion of genetic variation explained between the discernable groups.

We generated FST values for pairwise population comparisons in Arlequin 3.1, and using the program PASSAGE (Rosenberg, 2001) we performed a Mantel test (Mantel, 1967; Smouse et al., 1986) to correlate the FST values with geographical distance. To determine a more spatially refined relationship between genetic and geographical distance, we conducted Mantel correlograms for both nuclear and chloroplast data using 100-km interval geographical distance classes. We calculated a pollen seed-flow ratio based on FST values following Ennos (1994).


In five chloroplast microsatellites, we detected two to three alleles at each marker and a total of six haplotypes over 22 sampling locations (Fig. 2; Table 2). Three of the haplotypes were common and were found at multiple locations; the remaining three were found at only one sample site each (Fig. 2; Table 3). Within-population diversity (HS) was low (0.08 ± 0.03 SE), overall diversity (HT) was higher (0.67 ± 0.06 SE), and population differentiation (GST) was high (0.88 ± 0.05 SE; Fig. 2; Table 4). At 16 out of the 22 chloroplast sampling sites, all individuals within the population shared a single chloroplast haplotype.

Figure 2.

 Chloroplast haplotypes of Quercus garryana from 22 sampling sites. Haplotypes A = white, C = light grey, F = black are the predominant haplotypes over the study region. Haplotypes B (one individual at site 22 only), D (two individuals at site 1 only), E (17 individuals at site 1 only) are less common and are marked on the map.

Table 2.   Six chloroplast haplotypes of Quercus garryana described by the amplified fragment length of each of the five repeat loci.
Table 3.   Distribution table of the six chloroplast haplotypes found for Quercus garryana by sample site.
Sample siteHaplotype
  1. Sites 1–22 correspond to those in Fig. 1.

 1  2215 19
 2  19   19
 3  15   15
 4  17   17
 517     17
 6     99
 7     2222
 8     1515
 97    1522
10  16   16
11  9  1019
1220 2   22
1316     16
1418     18
1519     19
1613     13
17     1616
1812     12
1918     18
2018     18
21161    17
2218 1   19
Table 4.   AMOVA for both chloroplast and nuclear microsatellite data for populations of Quercus garryana.
 Source of variationd.f.Sum of squaresVariance componentsPercentage of variationFSTP
  1. Significance of variance components assessed with 10,000 permutations.

ChloroplastAmong populations21104.70.28786.070.861< 0.0001
Within populations35616.60.04713.93  
NuclearAmong populations21113.30.1094.940.049< 0.0001
Within populations6461352.72.09495.06  

The seven nuclear microsatellite loci ranged from six (quru-GA-1M17) to 17 alleles (ssrQpZAG 9) (Table S1). For the 22 sites genotyped with nuclear microsatellite markers, the overall allelic richness ranged from 27 alleles (sites 16 and 18) to 43 alleles (site 3) (Fig. 3). The allelic richness decreased significantly with increasing latitude (R2 = 0.63, < 0.001) (Fig. 3). This reduction in the number of alleles at the range periphery was due to a lack of alleles that are rare in the species core. Forty-two of the 61 total alleles were found across the entire study region. Of the remaining 19 alleles, 11 were unique to southern Oregon. We found five alleles that were not represented in southern Oregon samples: two alleles further north in both mainland and peripheral island locations, two alleles further north confined to the mainland, and one allele limited to the peripheral island populations. An additional three alleles were found in all regions except the peripheral islands. Pairwise FST values ranged from 0.000 to 0.184 among all 22 sampling sites, with a mean FST of 0.049 (< 0.0001).

Figure 3.

 Correlation between latitude and allele richness combined over all seven nuclear microsatellite loci. R2 = 0.63, < 0.001.

Observed heterozygosity from each sample site averaged over all loci ranged from 0.44 to 0.66 (Table 5). Significant departures from HW equilibrium (< 0.05) occurred in at least one sampling site for each marker. Markers quru-0G12 and quru-1M17 showed significant departures at one sampling site each; ssrQpZAG 36 and ssrQpZAG 9 showed significant departures at two sampling sites; quru-1G13 showed departures at three sites; quru-0C19 showed departures at four sites; and quru-0M05 showed departures at six sites (Table S2). Departures from equilibrium caused by homozygote excess were commonly found to be due to null alleles. Micro-checker analyses showed that short allele dominance or scoring errors due to stutter are not implicated in our markers. Instead, we found that marker quru-0M05 exhibited evidence for null alleles at six of the 22 sampling sites, and markers quru-0C19 and ssrQpZAG 9 exhibited null alleles at two sites each (Table S2). At the site and marker combinations showing null alleles, frequencies ranged from 0.11 to 0.29 (Table S2). Site 8 showed evidence for null alleles in two markers, but the other eight sites with null alleles had them only in one marker each. The presence of null alleles has the potential to bias results, but because of their low number (10 of the 154 marker and site combinations show a tendency for null alleles with frequency > 0.1) and haphazard distribution across markers and sample sites in our study, we interpret our results without modification. Departures from HW equilibrium at markers and sites not exhibiting null alleles could be due to the relatively small sample number per location. It is not likely that departures from equilibrium are the result of inbreeding.

Table 5.   Summary statistics of nuclear markers at 22 Quercus garryana sampling sites.
Sample sitenMean He/HoMean A
  1. Sample site numbers refer to sites shown in Fig. 1. Mean A is the number of alleles averaged over all seven loci. Mean He and Ho per locus are shown in Table S2.


Bayesian analyses in Structure indicated the data are most likely when = 2 (> 0.999) (Fig. 4). In the two-cluster case, all populations north of approximately 48° north latitude grouped together. This northern cluster also contained one Oregon population (7) and one southern Washington population (12) (Fig. 4). All other populations group together in a single southern cluster. For the AMOVA given two groups, significant levels of variation were explained within sampling site, within regions among sampling sites, and between the two regions (Table 6). Yet over 90% of the genetic variation was captured within sampling sites, indicating that all sites had a high level of genetic diversity (Table 6). Very little (c. 6%) of the nuclear diversity is explained among populations. The pattern driving the differences between regions was due, in part, to lower allelic diversity at the range periphery (Fig. 3). A weaker, but similar pattern was observed for levels of heterozygosity (not shown).

Figure 4.

 Degree of admixture per individual across a latitudinal gradient (site 1–22; refer to Fig. 1). Model = 2 with admixture (shown as assignment probabilities for the proportion grey to white per sample). Sampling locations with black arrows (7 and 12) denote sites that are located in the species core but have a greater proportion of the individuals assigned to the northern cluster.

Table 6.   AMOVA for Quercus garryana nuclear microsatellite data including a two-region grouping selected from groups assigned in the Bayesian cluster approach (Structure 2.2).
Source of variationd.f.Sum of squaresVariance componentsPercentage of variationFSTP
  1. Significance of variance components assessed with 10,000 permutations.

Between regions126.70.0683.030.0300.0001
Among populations within region2086.60.0743.300.034< 0.0001
Within populations6461352.72.09493.670.063< 0.0001

Geographical distances were significantly correlated with both chloroplast (Mantel test: 10,000 permutations, rM = 0.402, = 0.0012) and nuclear genetic distance (Mantel test: 10,000 permutations, rM = 0.622, = 0.0001). Mantel correlograms showed a significant positive genetic and geographical relationship within the first 200 km for both chloroplast and nuclear data (Fig. 5). Likewise, for both marker systems a significant negative correlation was detected between 700 and 900 km (Fig. 5). The only discrepancy in the relationship with geographical distance between the two genomes is at the 600–700-km range. The nuclear data show a negative correlation with geographical distance, whereas the chloroplast data show a positive correlation. This positive correlation with chloroplast data is driven by a shared haplotype from a site in southern Oregon (haplotype 1, site 5) with sites on Vancouver Island. From our data we estimated pollen to flow at a rate of c. 120 times that of seed dispersal.

Figure 5.

 Mantel correlograms from chloroplast (a) and nuclear (b) data. Filled squares indicate significant correlations between genetic and geographical distance at < 0.05. Points above the zero line are positively correlated; those below the line are negatively correlated.


The chloroplast markers harboured very little genetic diversity within a sampling location. Genetic differentiation among sampling sites was high, as has been found for oak by other investigators (Dumolin-Lapègue et al., 1997; Romero-Severson et al., 2003; Deguilloux et al., 2004; Grivet et al., 2006). Only a single sampling location had three haplotypes represented, and the majority of sites had only a single haplotype. The regional pattern of haplotype distribution includes the lack of mainland haplotypes on the peripheral islands, providing a signal consistent with post-glacial expansion. All haplotypes are represented towards the core of the distribution, whereas populations at the northern periphery retain a subset of haplotype diversity. In addition, our data support the ‘leading-edge hypothesis’ of Soltis et al. (1997) because the single predominant haplotype found on the glaciated island locations is the one that is represented in a subset of nearby unglaciated mainland localities. The low level of haplotype diversity also plays an important role in determining distances at which populations show significant genetic correlations. Virtually all peripheral populations share a single haplotype. In addition, over the whole study region a given site usually shares the same predominant haplotype with its nearest neighbour. Therefore, sites closest to one another (even over relatively large spatial scales of up to 200 km) exhibit positive genetic correlations with geographical distance.

The peripheral island samples are homogeneous for chloroplast haplotypes, with two exceptions. At the second most northern site (site 21), one individual possessed a unique haplotype. The Bayesian assignment test groups this individual with the others in its sample site, indicating that it might be an in situ mutation. At the most northerly site (site 22), however, another single individual had a haplotype found only in more southerly mainland populations. The nuclear assignment test for this individual groups it with the core cluster with a higher probability than with the northern cluster (Fig. 4). We postulate that this tree is a more recent colonist to this site, and could have been planted by humans since this location is managed as a community park of a rural residential development.

Unique haplotypes also occur at the most southerly site in Oregon, and we postulate that these are Q. garryana var. fruticosa (Engelm.) Govaerts. Quercus garryana var. fruticosa (formerly and occasionally called var. breweri) is a high-elevation, shrubby variety of Q. garryana, long recognized as being taxonomically separate from Q. garryana var. garryana (Watson, 1880). Because we do not have other Q. garryana var. fruticosa samples for comparison, this is only a suggestion.

Similarly to the patterns observed for the chloroplast microsatellites, we determined from nuclear microsatellite markers that allelic richness was reduced at the range periphery. We also observed latitudinal shifts in the frequency of common alleles. This reduction of alleles and allele frequency change at the range periphery is a signature of founder events during post-glacial colonization (Tomaru et al., 1997). Also similarly to the chloroplast data, we found that genetic distance was positively correlated with geographical distance within the first 200 km. This provides evidence that gene flow may be counteracting the effects of genetic drift over a reasonably large spatial scale, a different interpretation from a similar pattern from the chloroplast markers. Genetic and geographical distances were negatively correlated beyond 600 km, indicating that at spatial scales far larger than the expected size of a pollination neighbourhood, genetic drift has a stronger influence than migration. Still, the greatest amount of genetic diversity remains at the population level, and all predominant alleles present on the mainland were also present at the periphery of the species' range and were dramatically different from the chloroplast data. Thus, the level of pollen flow within and across sites has been sufficiently high to blur the genetic identity of founder populations over geographical distances less than 600 km. These findings of weak differentiation in the nuclear markers over large spatial scales are in agreement with the general patterns of allozyme diversity observed by Ritland et al. (2005).

Populations in the periphery are recognizably distinct from core populations, but are similar to each other. Our data support the supposition that colonization to the peripheral islands came from few mainland colonists and from one or few source populations. The apparently low-level of cross-colonization by seed among glacial source populations, given the age of these populations, provides evidence that effective seed dispersal in this species is limited. High nuclear diversity and little genetic differentiation due to high levels of pollen flow within our peripheral study region, however, are not unexpected, given that oaks persisted very near the ice sheet in unglaciated peripheral sites throughout the glacial period (e.g. site H in Table 1).

Even with the palynological evidence that Q. garryana persisted in populations near the ice sheet during the last glacial period, we still observe the loss of haplotypes in the glaciated range and high levels of chloroplast haplotype genetic differentiation. Our pollen-seed flow ratio of c. 120 is slightly less than (but of the same order of magnitude as) has been found for European oaks (Ennos, 1994), but it is higher than for Q. lobata in California (Grivet et al., 2008). We suggest that water between the mainland and island sites prevented greater seed dispersal to the peripheral islands, effectively increasing the distance between the two regions. Taking all our analyses into account, we found evidence for large pollination neighbourhoods (two clusters with considerable admixture among individuals within and across sites) and smaller seed-dispersal neighbourhoods with early, infrequent colonization of the post-glacial peripheral island sites.

It is important to recognize that low population differentiation determined from nuclear markers and high differentiation from chloroplast markers can occur even in situations of low pollen flow for tree species. Firstly, the effective population size of the chloroplast genome is one-quarter that of the nuclear genome (Muir et al., 2004). In addition, founding populations may no longer be small at the time of seed reproduction. For example, Austerlitz et al. (2000) modelled tree colonization and found that when the time to first reproduction is taken into account (often c. 50 years for oaks), founder effects were reduced by the accumulation of migrants over time, forming a large founding population with much of the genetic diversity of the source. In our study system, if founding events on the islands were common, we might expect to see more mainland haplotypes represented on the islands (those found at sites 9–12 and 17 in Fig. 2). We do not. Therefore, founding events were probably rare, and pollen flow is the likely mechanism for high nuclear diversity within peripheral populations. In addition, it has been shown with stochastic modelling that in situations with minimal seed dispersal, pollen flow has a strong homogenizing effect among populations (Austerlitz & Garnier-Géré, 2003).

We have low chloroplast haplotype richness across our entire study region, even with sample sizes larger than is typical for chloroplast haplotyping. We found very little within-population diversity and high levels of differentiation (in agreement with other studies) with approximately four times the number of samples per site than in many previous studies (Dumolin-Lapègue et al., 1997; Romero-Severson et al., 2003; Deguilloux et al., 2004; Grivet et al., 2006). We observed 14 total alleles across our five loci, indicating that chloroplast mutation rates are high enough to detect population-level differences.


In Q. garryana, chloroplast haplotype and nuclear allele richness in the glaciated island periphery was lower than that found in unglaciated areas where oak persisted. Low chloroplast diversity on the mainland was reduced to a virtual monomorphism in the island periphery. Although nuclear allele richness was reduced in peripheral populations due to the loss of mainland rare alleles, nuclear diversity within populations remains high. This suggests that regionally high pollen flow counteracts the effects of drift due to founder effects. As the current climate changes, a poleward geographical range shift may be possible for Q. garryana if seed dispersal limitation is overcome. After successful colonization of new localities, long-distance pollen flow to the new populations is likely to restore regional genetic diversity.


The following landowners allowed twig collection on their properties: Department of National Defence Canada, BC Parks, Capital Regional District Parks, Nature Conservancy Canada, Government House, City of Eugene, City of Roseburg, US Fish and Wildlife Service, Washington Department of Fish and Wildlife, The Nature Conservancy, Skagit Land Trust, Larrabee State Park, United States Army and the United States Bureau of Land Management. Inna Birchenko, Mary Kate Claiborne, Jason Dzurisin, James Giordano, Sean Hoban, Katie Marsico, Kevin McAbee, Tim McCleary, Carsten Nowak and Evgueni Zakharov helped in the laboratory. Jim Duemmel and Dan and Linda Dzurisin provided some of the samples. Vince Hellmann, Larry LaTarte, Jill Mueller, Derrick Parker, Kirsten Prior, Shannon Pelini and Neil Vargas assisted in field collections. David Severson allowed the use of laboratory equipment. T.D.M. was funded by the Arthur J. Schmitt Fellowship. This research was supported by the Office of Science (BER), US Department of Energy, Grant No. DE-FG02-05ER64023 and by the Indiana Center for Insect Genomics (funded by the 21st Century Fund for Research and Technology, the State of Indiana). Jason Dzurisin, Andrew Forbes, Katie Marsico, Jill Mueller, Carsten Nowak, Shannon Pelini and anonymous reviewers commented on the manuscript.


Travis D. Marsico researches dispersal and colonization to inform predictions regarding future responses to environmental change and limitations to geographical ranges using experimental and genetic approaches. Species’ peripheries are of particular interest due to the fragmented nature of the habitat and often marginal environmental conditions.

Jessica J. Hellmann studies the responses of populations and species to global change with a focus on ecological and evolutionary processes that limit range expansion. Her research considers insects, plants and their interactions, and management strategies for reducing negative effects of global change.

Jeanne Romero-Severson studies the role of genetic drift and selection in shaping the biogeographical distribution and population substructure of forest trees. Her current research interests include the effects of Pleistocene glaciations on the genetic structure of North American red oaks (Lobatae) and the effects of epidemic disease on the introgression of Japanese walnut into butternut, a related species native to eastern North America.

Editor: Brett Riddle