Very little is known about the nature and strength of reproductive isolation (RI) in Quercus species, despite extensive research on the estimation and evolutionary significance of hybridization rates. We characterized postmating pre- and postzygotic RI between two hybridizing oak species, Quercus robur and Quercus petraea, using a large set of controlled crosses between different genotypes. Various traits potentially associated with reproductive barriers were quantified at several life history stages, from pollen–pistil interactions to seed set and progeny fitness-related traits. Results indicate strong intrinsic postmating prezygotic barriers, with significant barriers also at the postzygotic level, but relatively weaker extrinsic barriers on early hybrid fitness measures assessed in controlled conditions. Using general linear modelling of common garden data with clonal replicates, we showed that most traits exhibited important genotypic differences, as well as different levels of sensitivity to micro-environmental heterogeneity. These new findings suggest a large potential genetic diversity and plasticity of reproductive barriers and are confronted with hybridization evidence in these oak species.
Focusing on plant species, studies on RI estimates from the past two decades have been reviewed recently and led to several important conclusions (Rieseberg & Willis, 2007; Lexer & Widmer, 2008; Lowry et al., 2008; Widmer et al., 2008): (i) prezygotic barriers are on average twice as strong as postzygotic barriers, (ii) postmating barriers are often asymmetrical, thus allowing gene exchange preferentially in one crossing direction and (iii) there is, however, a large diversity in barriers types and strength in different plant systems, consistently with different numbers, effects or types of genes potentially involved in RI.
Lexer & Widmer (2008) further argue that plants are likely to fit the genic view of the speciation process (Wu, 2001), which recognizes that RI can be controlled by a moderate number of genes, allowing higher hybridization rates in the rest of the genome. This view, along with ecological speciation models and case studies (Schluter, 2001; Rundle & Nosil, 2005), emphasizes the primary role of selection in speciation (Charlesworth & Charlesworth, 2000; Coyne & Orr, 2004; Gavrilets & Harrison, 2005; Via, 2009) and revisits the original idea of Darwin (1859) about the importance of natural selection via ecological interactions for the evolution of divergent lineages. Furthermore, ecologically based divergent selection is a necessary component of speciation in sympatry, because RI can be directly or indirectly linked to habitat adaptation and allow divergence despite gene flow (Via, 2001).
Evidence of selection driving ecological divergence despite hybridization thus confirms the Q. robur and Q.petraea species complex as a good case for studying ecological speciation with gene flow. However, very little is known about the nature and strength of reproductive barriers in these species. Previous large crossing experiments showed that both pre- and postzygotic isolation mechanisms operate (Steinhoff, 1993, 1998), and reproductive barriers were inferred in several studies using nuclear markers for paternity analyses (Curtu et al., 2009; Lepais & Gerber, 2011). We need to better understand which traits influence RI, what are their strength and relative contribution to gene flow limitation among species, and how much of their variation is explained by genetic and/or ecological factors.
To start tackling these issues, the objectives of this study were (i) to characterize and quantify RI between Q. petraea (sessile oak) and Q. robur (pedunculate oak) across various stages of the reproduction process, and (ii) to assess the variation across different genotypes for traits related to RI. To fulfil these objectives, we performed a large number of interspecific controlled crosses on adult trees from a common garden with clonal replicates and compared them with intraspecific crosses. We focused on postmating barriers, prezygotic mechanisms in particular by developing original cytological observations and postzygotic mechanisms by analysing seed production, germination and early hybrid fitness data. RI was quantified according to Ramsey et al. (2003) and from linear models of variation. Correlations between traits were estimated and genotypic effects among individuals for these traits were tested. Particular care was taken for better controlling and interpreting micro-environmental variation using biological covariates in linear models of variation.
Material and methods
Plant material and controlled crosses
Quercus petraea and Q. robur are monoecious species (i.e. both male and female flowers are on the same tree) with a wind-pollinated and almost completely outcrossing mating system (Bacilieri et al., 1996). Quercus robur individuals were chosen from 207 full-sib genotypes belonging to a mapping pedigree (3PxA4, Barreneche et al., 1998). All genotypes were vegetatively propagated 5–12 times and planted in year 2000 at the INRA experimental unit of Bourran (latitude 44.30°N, longitude 0.40°E). The male parent (A4) was from Arcachon (latitude 44.40°N, longitude 1.11°W), and the female parent (3P) from Pierroton (latitude 44.44°N, longitude 0.46°W). Interspecific crosses were performed between 87 Q. robur individuals from this pedigree, used as female parents, and the same mixture of four Q. petraea male parents (Qs11, Qs27, Qs31 and Qs32) sampled in a natural population (latitude 47.83°N, longitude 1.91°E, Orléans, France). Crosses could only be performed reasonably in that direction because offspring are much more difficult to obtain when Q. petraea is used as the mother, resulting in a strong asymmetry of hybridization success (Steinhoff, 1993; Bacilieri et al., 1996; Petit et al., 1997; Steinhoff, 1998; Jensen et al., 2009; and our own unpublished data over the last 20 years).
The 87 Q. robur female individuals included 27 genotypes with three clonal replicates (10 in 2008, 19 in 2009 with two genotypes crossed both years, Fig. 1). Due to their young age (8 years), trees were chosen from preliminary data on their ability to flower and produce acorns and so that their variance in height and bud burst phenology was representative of the entire pedigree (Derory et al., 2010). The four male parents had been previously shown to span a priori a large set of crossing compatibility behaviours among different oak species (Lepais, 2008). We also used natural crosses resulting from open pollination and compared them with interspecific crosses. Those crosses included 68 Q. robur female individuals from the same 27 genotypes (10 genotypes in 2008 with three replicates, 19 genotypes in 2009 with two replicates, Fig. 1). They were assumed to result from within Q. robur matings only, given the exclusive environment of Q. robur pollen from surrounding plantations of thousands of trees.
To perform crosses, male inflorescences were collected from 1999 to 2005 just before anther dehiscence (Steinhoff, 1993), kept at 20 °C until complete anther maturation, dried in a ventilated room at 30 °C during 12 h inside sieve bags to isolate the pollen, which was finally conserved at −18 °C in the dark (G. Roussel, ‘Controlled cross techniques for European oaks’, see http://www4.bordeaux-aquitaine.inra.fr/biogeco_eng/People/Q-Z/Roussel-Guy). Pollen samples were rehydrated overnight at 4 °C and then stained in a solution of fluorescein diacetate at 0.1% (Heslop-Harrison & Heslop-Harrison, 1970) to assess their viability under fluorescent microscopy (UV 450–490 nm, zoom 40 ×, see Fig. 2b). Viability rates were scored for approximately 100 pollen grains per individual, and pollen quantities were adjusted to reach similar viable grain proportions per genotype in pollen mixtures.
Despite trees height (3.75–7.45 m, see Fig. S1 and Table 1), we managed to include 155 individuals in the crossing scheme (Fig. 1). We followed a crossing protocol that has been improved over the last 20 years at INRA (Roussel, 2009), and that consisted in entirely bagging the mother trees to isolate them from the surrounding pollen cloud, with tissue that was permeable to light and humidity. This was done in February 2008 and 2009 before the start of vegetative growth, tissue bags being closed at the top and attached around the trunk at the bottom of the trees (Fig. S1). Then, in April, as pistils showed a shiny red colour (Fig. 2a), the pollen mixture was injected inside bags using an air compressor and a syringe linked to an air chamber that ensured mixing and homogenizing of the air and pollen inside the bag and around the tree. Injections were performed twice a week during 3 weeks in April of both years. The bags were removed 10 days after the last pollen injection.
Table 1. Description of reproductive isolation traits (including range, number of observations and sample size per genotype).
Underlying reproductive barrier mechanism
Total number of measures
*2009 measures only.
Unaborted inflo rate
1 − (aborted number of female inflorescences/total number of female inflorescences), recorded one day in June
Flowers abortion prevents fertilization
Pollen tubes nb
Number of pollen tubes in the female flower style (two sampled flowers per tree)
Pollen grain adhesion and germination depends on interactions between pollen exine layer and papillar stigma cells
Pollen tube progress
Score indicative of pollen tube length in the style, between 0.5 (tube observed just below the stigma surface) and 8 (tube observed close to one ovary) (see also Fig. S4)
Mediated by continuous interactions with transmitting tissue cells
Seed number per mother tree
Results from embryo fertilization success
Seed unit weight
Mean weight of one seed (g) per mother tree
Link to resources available for progenies
Seed production ratio
Number of seeds divided by number of inflorescences (inflo nb in Table 2)
Results from embryo fertilization success, taking into account the initial number of inflorescences
Seed germination rate per mother tree
Progeny fitness component
Germinated seed survivorship rate per mother tree
Progeny fitness component
Height of progeny in June the year following the cross (mm)
Progeny fitness component
Progeny leaves nb
Number of leaves of progeny, In June the year following the cross
Progeny fitness component
Choice of RI traits
We measured nine different traits related to cross-compatibility and postmating RI that could be reasonably assessed on tall trees (Table 1). Prezygotic traits included first the rate of female inflorescences that did not abort relative to the total number of inflorescences recorded (unaborted inflo rate), which was expected to be positively correlated with reproductive success (see Table 1 and Fig. S2). Indeed, female flower abortion during pollen tube growth has been observed as a prezygotic barrier for self-incompatibility (SI) in the closely related species Quercus ilex (Yacine & Bouras, 1997), and there are well-known links between SI and interspecific compatibility in several species (e.g. Hancock et al., 2003; Swanson et al., 2004). Cytological observations of sampled flowers were performed for tracking and measuring pollen tubes progression in styles, which required preliminary experiments to optimize tissue fixation, fluorescent microscopy protocols, and flower sampling dates [see Fig. S3 and Data S1 for details on preparing slides of flower longitudinal sections (Fig. 2a)]. The crossing experiment was carefully designed to ensure that pollen tubes for all trees would be at a similar stage of development for cytological observations: (i) controlled pollinations series and flower sampling of both types of crosses were performed at the same time for all trees, (ii) pollinations were performed when male inflorescences on Q. robur in the surrounding environment were observed, and (iii) the choice of flower sampling date was optimized from previous time series data on pollen tube growth (Fig. S3). At the time of observation, most pollen grains had already germinated, so the total number of pollen tubes observed in the stigma and style was recorded as the direct result of pollen adhesion and germination (pollen tubes nb, Fig. 2d and Table 1). Interactions with the transmitting tissue cells mediate the pollen tube growth and directionality within the style, which may constitute prezygotic RI mechanisms (Swanson et al., 2004). We therefore initially considered the pollen tube length on fixed tissues as a relevant trait to estimate pollen tube growth (Boavida et al., 2001). However, because oak female flowers were lignified, the total tube length was not always visible (Fig. 2d). Thus, having observed and assuming that most pollen grains germinated on the stigma surface, we defined the pollen tube progress as a score indicative of the tube length between 0.5 (tube observed just below the stigma surface) and 8 (tube observed close to one ovary), which gives a good surrogate of the tip of the pollen tube distance from the stigma, and thus of its length in the transmitting tissue (Table 1 and Fig. S4). This score was recorded for all significant pieces of pollen tube observed in the pistil that could unambiguously be attributed to distinct tubes.
Four postzygotic traits were assessed for each mother tree to quantify postzygotic isolation that was expressed during germination and seedling development (Table 1): in September, all acorns were harvested before their fall and counted for each tree (seed nb, Table 1). Seed unit weight was calculated per tree by dividing the total seed set weight by seed nb. However, the initial resources allocated to the reproductive female function by the tree will impact seed nb independently from crossing success. So the number of inflorescences (inflo nb) was also counted on each mother tree (Table 2). The correlation between seed nb and inflo nb was indeed significant (R =0.57, P <0.002), thus we eventually reported and analysed further the seed production ratio defined as the ratio of seed nb divided by inflo nb. To prevent parasite infections after seed collection, seeds were treated in a bleach solution (0.07% of active chlorine) for 15 min, dried for 8 days at room temperature, treated with the Benlate fungicide (100 g hL−1, DuPont®, Wilmington, DE, USA), and conserved at 1°C during autumn and winter. For genotyping purposes, a cotyledon slice (1 mm large) was sampled for each acorn, which was given a unique identification before being pasted on a polyethylene sheet. In May, these sheets were covered with humid vermiculite to trigger germination in a greenhouse. In June, germinating acorns were counted (germination rate in Table 1) and transferred in pots filled with compost that were randomized and watered as required. In July we measured the proportion of surviving progenies (germinated survival) and recorded progeny height and progeny leaves nb (Table 1).
Table 2. Description of biological covariates (including their range, number of observations and sample size per genotype).
Link with microenvironmental variation
Total number of measures
*2009 measures only.
Number of female inflorescences on the mother tree, in June 2008 and 2009
Initial potential for seed production and resources allocation to female function
Height of mother trees in year 2010 (cm)
Available acquired resources of mother trees
Trunk diameter at 1 m of mother trees, in year 2010 (cm)
Available acquired resources by mother trees
Score of leaf development stage of mother trees after bud burst, in 2007
Phenological development of mother trees that can be affected by soil fertility
Height residuals mean (cm) across the eight trees surrounding mother trees in the field (year 2005) (see statistical analyses)
Soil and environmental heterogeneity (hydromorphy, nutriments, etc.)
Neighbors bud burst
Leaf development score residuals mean across the eight trees surrounding mother trees in the field (year 2007) (see statistical analyses)
Soil and environmental heterogeneity (hydromorphy, nutriments, etc.)
Choice of covariates
To account for soil and microclimate variation, we first tested a block effect as an independent factor that corresponded to the whole field partition in six blocks of similar area. We also used other traits as covariates in the linear models described below. These traits were assumed to account for microsite environmental heterogeneity in atmospheric and fertility conditions that would affect available levels of resources acquired by the mother trees, and thus potentially the expression of environmentally dependent RI traits (Table 2). Height and diameter of each pollinated tree were measured in 2010. Bud burst, previously recorded in 2007 according to Derory et al.’s (2006) notation scale, was assumed to provide information on the level of phenological development of each tree that could affect inflo nb, which in turns affects seed nb (see above). Indeed, preliminary tests showed that a large part of inflo nb micro-environmental variation was explained by bud burst variation within genotypes (Table S1). We further assumed that 2007 bud burst would be highly correlated with bud burst of the same trees in 2008 and 2009, due to its high heritability (Jensen, 2000; Scotti-Saintagne et al., 2004a). Neighbours bud burst scores for each mother tree were also computed from the residuals of linear models after excluding genotypic effects, as potential surrogates of micro-environment heterogeneity (Table 2 and see Statistical analyses below). Additional height data (recorded in 2005) were used to compute neighbours height in the same way (Table 2). Finally, seed unit weight was also used as a covariate as it is linked to the level of resources available for seed germination and progeny growth, and likely to be a component of progeny fitness (Ramsey et al., 2003).
Progenies from the interspecific crosses and their potential parents were genotyped at microsatellite loci to check for potential pollution events. DNA was extracted from (i) three leaf discs (1 cm diameter) from the 27 mother trees, (ii) 10 μL of pollen from the four sessile father trees and (iii) a basal cotyledon slice on the 1151 harvested acorns. We used Nucleospin® ready-to-use system for fast purification of nucleic acids according to the manufacturer protocol (Macherey-Nagel, Duren, Germany). Individuals were then genotyped at six polymorphic microsatellites (A11, A3, AB, AK, F, G) using the protocols described in Guichoux et al. (2011) for obtaining data, determining alleles real size and manual binning of alleles. All trees were successfully genotyped at the six loci, whereas 494 (42.9%) of harvested seeds were genotyped for at least four loci, the others had poor-quality profiles due to the low DNA quality extracted from acorn material.
To estimate pollution rates integrating experimental genotyping errors (potential triploid tissues extracted from the acorns, null alleles, microsatellites mutations and reading errors), mother–offspring mismatches were assessed with cervus software version 3.0.3 (Kalinowski et al., 2007). To avoid undue exclusions (Christie, 2010), the estimated error rate (0.099%) allowed for two mismatches between offspring and potential fathers. We used R programs available at http://sites.google.com/site/parentagemethods/genotyping-error. Forty-two seeds resulted from natural pollen pollution (8.5%), of which nine germinated and grew (1.3% of the progenies) and were removed from analyses. A male parent was assigned to the 452 remaining seeds, based on best likelihood ratios (LOD scores) from paternity analyses with the cervus software.
All traits were assessed on three clonal replicates per genotype for interspecific crosses, on two or three replicates for natural crosses. For traits where several observations per individual tree were made (pollen tubes nb, pollen tube progress, progeny height and progeny leaves nb), a tree within genotype effect was tested using one-way analyses of variance (anova). As none were significant, all measurements were considered as independent for subsequent analyses. Interspecific and natural crosses were first compared for all traits. For traits measured in both years, a year effect was first tested in a one-way anova within each cross. When this effect was significant, data from year 2 were kept unchanged and those from year 1 were corrected in reference to year 2 using least squares means of both years. The corrected data of individual trees were then used for testing differences among crosses in one-way anova models, checking normality of residuals and homoscedasticity of variances with Shapiro & Wilk (1965) and Levene (1960) tests, respectively. For traits departing from those assumptions (i.e. all traits except pollen tube progress and progeny height), various nonparametric tests including Kruskal–Wallis and Kuiper tests were applied (Kuiper, 1962; Sokal & Rohlf, 1995). We also quantified RI by computing the ratio , with SSQB being the sums of squares (SSQ) of the cross type effect, and SSQW that of genotypes within crosses in a simple linear model (model m1 in Table 3). therefore represents the contribution of the type of cross in the variation of both the crosses and the genotypic variability within crosses.
Table 3. General linear models used to partition the variation in reproductive isolation traits.
Model principal objective
*Indices are as follows: i for cross types, j for mother tree genotypes, k for block, l for individual tree, m for father genotype; E, E′, E″ are the various residual random variables, Y are the phenotypic data variables.
Testing the cross type Cirelative contribution to total variation between and within crosses
Yijk = μ + Ci + Gj/i + Eijk
Testing a mother tree genotype effect Gj within interspecific crosses
Yjkl = μ + Gj + bk + Ejkl
Testing both Gj and covariate effects with ancova
Yjkl = μ + Gj + bk + β (covjkl) + E′jkl
Testing both Gj covariate effects with Nested ancova
Yjkl = μ + Gj + bk + βj (covjkl) + E″jkl
Testing a father effect Fm within mother tree genotypes
Yjmkl = μ + Gj+ Fm+ bk + Ejmkl
Reproductive isolation was further quantified across traits by comparing interspecific and natural crosses (Fig. 1). We applied Ramsey et al. (2003) method where each RI component RIn corresponding to a particular trait n was defined as RIn = 1 − (interspecific crosses mean/natural crosses mean), which quantifies the relative decrease in the traits performances between natural and interspecific crosses. Absolute contributions (ACn) of each RI component n that were ordered from their sequential occurrence of barriers during the tree life cycle were computed as , with AC1 = RI1. The total RI (T), ranging from 0 to 1, was therefore defined as for the m traits considered. The relative contributions of each trait n (RCn) were then computed as RCn= ACn/T. To assess relationships between traits, Pearson’s correlations coefficients were computed on data from interspecific crosses at different levels (raw data, individual and genotypic means).
Before analysing the variation across interspecific crosses, year effects were first tested and corrected as above if necessary. In addition, we checked the consistency of year effect magnitudes estimated on two genotypes (113 and 131) that were measured in both years (Fig. 1), with those on all genotypes. General linear models of increasing complexity (in number of factors and covariates) were then used on RI traits from interspecific crosses to both test for differences among genotypes and explore the origin of micro-environmental variation (Table 3, models m2 to m4). Model m2 was also applied on height 2005 and bud burst 2007 data in the whole field trial, to compute the covariates neighbours height and neighbours bud burst, by averaging residual values for the eight neighbours surrounding each mother tree (see Table 2). Computing mean of residuals rather than original data ensured that we focused on trees’ average response to local genotype by micro-environment variation without being biased by strong effects from particular genotypes. We previously checked that mean values that could be linked to competition effects among trees were not more useful than mean residuals (data not shown).
Given potential associations between target variables environmental variation and covariates (Table S2), these were integrated either at the individual tree level applying an analysis of covariance (ancova) (Sokal & Rohlf, 1995; model m3 in Table 3) or at the genotypic level applying a nested ancova (e.g. Petit et al. 2005; model m4. in Table 3). Imbalance in the experimental design was accounted for with type III adjusted sums of squares (SSQ), but as they are not additive, the proportions of variation for the genotype effect () are reported with type I SSQ. The best models for each trait were chosen according to their coefficients of determination () and type III anovaP-values for genotype and covariates effects. We tested for residual normality and homoscedasticity as above. For traits deviating from these assumptions, an approximate permutation test was performed on mother tree individuals with 5000 randomizations to generate simulated P-values distributions for genotype effects. The new test P-value was given by the 95% quantile of the distribution (Anderson, 2001), and Kruskal–Wallis tests were also performed to test differences between genotypes. All tests were nevertheless reported for each trait, given the robustness of anova models to limited departures from normality (Sokal & Rohlf, 1995; Stratton, 1995; White & Bennetts, 1996) and as they were used to compute coefficients of determination. For pollen tubes nb and pollen tube progress, analyses were performed on individual means (corrected from a potential year effect) consistently with most other traits. The unaborted inflo rate trait was not included in the tests among genotypes because of missing data for around half of the genotypes. Finally, on progenies that were assigned a father, differences in father’s contributions across mother trees were tested with χ2 homogeneity and Fisher’s exact tests. A father effect was also tested in progeny fitness-related traits, adding either a principal effect Fm for father m (model m5 in Table 3) or a nested father effect Fm/j within mother genotypes j in the previous models m2, m3or m4. All statistical analyses were performed with the R software version 2.12.0 (R Development Core Team, 2010).
Crosses comparison and RI quantification
Mean performances for all traits were compared between interspecific and natural crosses using common sets of genotypes (Fig. 1). Significant differences were observed at one or more of the various tests performed, for six traits linked to both prezygotic and postzygotic mechanisms (unaborted inflo rate, pollen tubes nb, pollen tube progress, seed production ratio, germination rate and progeny height) (Fig. 3a–i, Fig. 4, Table S3). In general, we observed more flower abortions, less pollen tubes and smaller pollen tube progress in interspecific than in natural crosses (Fig. 3a–c). Seed production ratio and germination rate were on average lower for interspecific than for natural crosses (Fig. 3d–f). Reproductive barriers between species were also very clear when comparing interspecific and intraspecific controlled crosses that were performed in 2010 on Q.robur genotypes (Table S7). Interspecific crosses showed significantly higher inflorescence abortion rates (+16%) and a lower seed production ratio (−63%) (Fig. S10). In the progeny fitness-related traits, plants from the interspecific crosses were slightly shorter (Fig. 3h), whereas the mean number of progeny leaves (progeny leaves nb) was similar for all crosses (Fig. 3i). No differences were detected among crosses for seed unit weight and germinated survival (Fig. 3e,g, Table S3), although values for the interspecific crosses were slightly higher. Most traits exhibited a large range of variation that depended on cross types, traits or genotypes [e.g. germination rate ranging from 0% to 87% across interspecific cross genotypic means (Table S6 and Fig. S5d)], the highest variance often observed was for natural crosses. This could be due to a slightly lower sample size, to a larger environmental stochasticity for cytological observations from natural pollination, but also to a possible larger genetic background of male parents for progeny traits (Fig. 3a–i). To test that bagging the trees did not lead to any strong bias in the comparisons between interspecific and natural crosses, intraspecific controlled crosses were also performed on 10 among the 27 genotypes studied (Supporting Information and Fig. S6). Their performances were consistent with natural crosses, confirming the within Q. robur origin of natural crosses (Table S3 and Fig. S7a–i).
Individual RI values (RIn) ranged from 3.7% (unaborted inflo rate) to 40.1% (pollen tube progress) for traits showing significant barriers (Fig. 4). The strongest reproductive barriers were for pollen tube progress, germination rate (24.9%), pollen tubes nb (23.9%) and seed production ratio (19.3%) (Fig. 4). The total RI value (T) of 0.719 (0.752 when keeping only significant values) was consistent with a reasonable number of viable interspecific offspring. The relative contributions of each trait studied to T with the Ramsey et al. (2003) method were the highest for pollen tube nb (RCn = 32.1%) and pollen tube progress (RCn = 40.8% for), and the smallest for unaborted inflo rate and progeny height (RCn = 3.5%) (Fig. 4). These values were consistent with the importance of the barriers revealed by values for pollen tube progress (, Fig. 3c) and germination rate (, Fig. 3f). However, all the other traits showed much lower values (<5%, Fig. 3 and Table S3) because of a large variation between genotypes within crosses that this ratio accounts for, in contrast to Ramsey’s method. This was the case in particular for pollen tube nb and seed production ratio.
Genotypic means correlations between traits
Correlations between mean genotypic values of traits are reported here (Table 4) because they better account for genotypic covariation, but similar trends were observed when computing correlations between individual means or raw data (Tables S4 and S5). No significant correlation was observed between pollen tubes nb and pollen tube progress (Table 4), or between these two traits and seed production ratio, suggesting that they represent different mechanisms in pollen–pistil interaction until the embryo formation. The correlation was not significant between germination rate and germinated survival (r =0.13, P >0.05), thus high survival rates of germinated seeds could occur with either high or low germination rates. For seedling traits, progeny height and progeny leaves nb were correlated (r =0.56, P <0.01 in Table 4 and Fig. S8d), but only approximately 31% of the variation of one trait was explained by the other, suggesting different mechanisms in their expression at the genotypic level. Significant positive correlations were also observed between traits linked to various types of postzygotic events (seed production, germination and progeny fitness): seed unit weight and germination rate (r =0.62, P <0.001 in Table 4 and Fig. 8b), high seed weight being therefore associated with high germination rate. Offspring with higher survival among those who germinated also showed a higher number of leaves at the time of observation (r =0.76 between germinated survival and progeny leaves nb with P <0.001 in Table 4 and Fig. S8c).
Table 4. Pearson’s correlation coefficients between adjusted genotypic means of all studied traits in interspecific crosses. The group to which each trait belongs is indicated in brackets.
PRE, pre-zygotic; SEE, seed production; GER, germination; PRO, progeny.
*P < 0.05; **P < 0.01; ***P < 0.001.
(1) Pollen tubes nb [PRE]
(2) Pollen tube progress [PRE]
(3) Seed production ratio [SEE]
(4) Seed unit weight [SEE]
(5) Germination rate [GER]
(6) Germinated survival [GER]
(7) Progeny height [PRO]
(8) Progeny leaves nb [PRO]
Partitioning phenotypic variation of RI traits
For interspecific crosses, significant genotypic effects were observed across mother trees for most traits except for germinated survival (Table 5, Pgen values). For pollen tubes nb, there were departures from linear models assumptions, so permutation tests were performed and resulted in P-values at best below 10% (Pperm in Table 5). This was probably due both to a lack of power (despite a proportion of genotypic variation explained () above 47%), and to the sensitivity of this trait to micro-environmental conditions and the developmental stage of the trees (see below). For prezygotic, seed production and germination traits, genotype effects explained a fairly high part of the variation, values ranging from 40.9% for germinated survival to 66.6% for pollen tube progress (Table 5, see also Table S6, and Fig S5a–f). In contrast, values were only 14.5% and 7.7% for progeny height and progeny leaves nb, respectively (Table 5 and Fig. S5g,h), probably because of the additional and large genotypic variance component between unknown half-sib genotypes (within each mother tree).
Table 5. Best linear models for each reproductive isolation trait studied.
PRE, pre-zygotic; SEE, seed production; GER, germination; PRO, progeny.
‡anova, simple analysis of variance model without covariate; ancova, analysis of covariance; nested, nested analysis of covariance.
§Degrees of freedom of the genotype effect.
¶Proportion of variation explained by the genotype effect (in sums of squares %).
††Model coefficient of determination (in %).
‡‡Sign of the covariate regression coefficient(s) (when a nested analysis of covariance was used, ‘−’ means that all significant regression coefficients were negative, ‘+’ means that they were all positive, and ‘+/−’ means that they were either positive or negative).
§§Fisher’s test F-value of the covariate effect.
¶¶P-value of the covariate effect.
†††Fisher’s test F-value of the genotype effect.
‡‡‡P-value of the genotype effect.
§§§P-value of the genotype effect using the approximate permutation test.
¶¶¶P-value of the genotype effect using Kruskal–Wallis test.
Pollen tubes nb
Pollen tube progress
Seed production ratio
Seed unit weight
Neighbors bud burst
Seed unit weight
Progeny leaves nb
Block effects were detected for pollen tubes nb and seed unit weight only, illustrating the sensitivity of these particular variables to micro-environmental conditions at the experimental field scale (around 360 trees were located within each block of around 1000 m2, among which 15 trees on average were sampled for this experiment). Significant effects linked to particular covariates were also detected for pollen tubes nb and the other traits (Table 5, Pcov values), using either simple Analysis of covariance (ancova) or nested ancova models, where the covariate effects are tested as nested within genotypic effects (see ‘Statistical analyses’). The use of covariates increased the power for testing genotype effects for the traits pollen tube progress, inflo nb, and germination rate, which were otherwise not significant. ancova models are reported for the two best covariates per trait because they might give different and complementary information on the origin of micro-environmental variation. Remarkably for pollen tubes nb, height and neighbours height covariates nearly fully explained the variation observed, showing the sensitivity of this trait to a tree’s resource acquisition and competition with its neighbours. This sensitivity differed among genotypes, with both significant positive or negative relationships (Table 5, Sign column). The genotype effect explained more variation for pollen tube progress, but the combination of genotype and covariate effects explained around 97% of the total variation (see values in Table 5). Significant regression coefficients within genotypes in the nested ancova indicated that mother tree heights were either positively or negatively correlated with the pollen growth in the styles (Table 5, Sign column).
As expected, a significant effect of the seed unit weight covariate was also observed when testing genotype effects for germination rate, illustrating that bigger seeds triggered higher germination rates in general. All significant regression coefficients were indeed positive but with different magnitudes (Table 5, Sign column). Significant differences among genotypes were found for progeny height and progeny leaves nb (Kruskal–Wallis tests with , P <0.001 and , P <001 respectively in Table 5). For progeny leaves nb only, neighbours height and diameter covariates increased the proportion of variation already explained by approximately 10%, suggesting that tree diameter had a weak positive effect whereas height of neighbour trees had a weak negative impact on the number of leaves counted at a particular date (Sign column, Table 5).
No significant differences in Q. petraea fathers contributions
Assigned fathers for the 452 successfully genotyped seeds from the interspecific crosses showed that 31.2% of the mother trees were fertilized on average by the sessile father Qs11, 31.4% by Qs27, 24.5% by Qs31 and 12.8% by Qs32. These proportions of different male donors were not significantly different among mother trees genotypes (Fig. S9) based on chi-square and Fisher’s exact tests (, P <0.36 and P <0.34 respectively). Moreover, no significant father effect was detected when tested together with the mother effect (either as a principal or a nested factor, see ‘Statistical analyses’) in linear models for progeny height or progeny leaves nb variation.
The large crossing experiment reported here allowed us to quantify postmating reproductive barriers between Q. robur (as a mother) and Q. petraea (as a father) at different life history stages and to gain new insights into their evolution. First, reproductive barriers were stronger in the earliest reproduction steps (pollen–pistil interactions) than in postzygotic events (seed germination and progeny fitness-related traits). Second, most traits linked to pre- or postzygotic stages were not correlated, suggesting multiple and distinct mechanisms. Third, we demonstrated important genotypic differences among individuals for RI traits and a large sensitivity to micro-environmental variation for some of them.
Strength of RI between Q. robur and Q.petraea
Significant differences between interspecific controlled crosses and intraspecific natural crosses revealed the existence of reproductive barriers between Q. robur and Q. petraea, at different life history stages: from prezygotic pollen–pistil interactions (less pollen tube progress in interspecific crosses) that represented the strongest barriers to seed production (reduced seed set), germination (reduced germination rate) and progeny fitness (slightly reduced interspecific progeny growth). Between Q. petraea and Q. robur, the total RI quantitative estimate (T =0.719) was lower than most values reported recently in flowering plants (>0.950 for 15 of the 19 species pairs studied, Lowry et al., 2008). Other possible reproductive barriers such as flowering asynchrony or pollen competition could not be studied here, because pollen mixtures from a single species (Q. petraea) were brought directly onto mother trees from the other species. However, competition against heterospecific pollen is likely an important reproductive isolating barrier in oaks as in other species (Rieseberg et al., 1995; Howard, 1999; Klips, 1999; Campbell et al., 2003; Ramsey et al., 2003). The pollen competition absolute RI strength value is high indeed (68.4%) and shows the strongest asymetry, when recomputed using Table 1 in Lowry et al. (2008) for 11 species pairs that showed lower performances in interspecific crosses. A proper pollen competition experiment would be needed to test whether conspecific pollen tubes would not grow faster than heterospecific pollen tubes in mixed-species pollen mixtures brought on the same pistil, as previously observed in Betula species by Williams et al. (1999).
Seed production ratio exhibited another significant barrier (Fig. 4). Despite being affected by the last phases of pollen tube navigation towards the ovary, seed formation results mostly from early postzygotic mechanisms and seed developmental processes prior to germination. In these oak species, prezygotic barriers therefore appear stronger than postzygotic barriers in the set of barriers included in this study. Moreover, these barriers are typically intrinsic because they are mostly linked to pollen–pistil interactions. They represent close to 80% of the cumulated relative contributions (RCn) to the total RI measured in this experiment. This is consistent with the Lowry et al. (2008) review where prezygotic isolation is on average twice as strong as postzygotic isolation, despite very different case studies and underlying speciation models. However, the individual RI strength of germination rate in oaks was higher than the average value reported in Lowry et al. (25% vs. 4% if all reviewed cases are included, and vs. 18% when excluding F1 seeds showing heterosis for germination). When accounting for the variation between genotypes within crosses with the ratio, germination rate was the strongest barrier after the pollen tube growth, clearly above all the other traits. The mechanisms involved in the expression of both seed production ratio and germination rate could result in part from the evolution of between-locus incompatibilities (based on the Bateson-Dobzhansky-Muller model; Orr & Turelli, 2001), whose effects can accumulate across multiple loci, and for which there can be standing variation (Coyne & Orr, 2004, and see below).
In contrast, extrinsic barriers revealed by progeny fitness-related traits were either small or not significant, thus showing a very low contribution to total RI value (3% in Fig. 4 for progeny height, below 1% with the ratio in Table S3). Indeed, around 40% of progenies from the interspecific crosses showed a higher height than the natural progenies average (data not shown). Hybrid survival based on growth are likely underestimated because our observations were limited to early developmental stages and were obtained under optimal glasshouse conditions. In natural stands, selection could act against intermediate hybrids that would be maladapted to the contrasted ecological niches of parental species (Via, 2001). Such extrinsic postzygotic isolation mechanims have been suggested in oaks from reports of lower proportions of first generation hybrids at the adult stage compared to the seeds stage or to later-generation hybrids (Curtu et al., 2007, 2009; Lepais et al., 2009). However, there is also evidence from studies based on controlled crosses and on inferences from field observations that once formed, hybrids are fertile and produced pollen grains or offspring that are as viable as those sired by purebreds (Olrik & Kjaer, 2007; Lepais & Gerber, 2011). So overall, our results suggest that in the set of postmating barriers studied, intrinsic RI barriers between Q. robur and Q. petraea are the strongest, at the prezygotic stages in particular, with significant but weaker postzygotic barriers.
Diversity of reproductive barriers across traits, genotypes and micro-environments
We have shown that different traits contribute to RI between Q. robur and Q. petraea, similar to the great diversity of barrier types that can be encountered in plants (Coyne & Orr, 2004; Rieseberg & Willis, 2007; Widmer et al., 2008). This diversity is also illustrated in oaks by the absence of significant correlations between genotypic means that were detected among the traits showing a significant reproductive barrier (see Table 4). This would suggest the involvement of relatively distinct mechanisms for the expression of pre- and postzygotic barriers in oaks, as judged from their statistical linear independence.
Clonal replicates also allowed the partition of the large phenotypic variation observed in RI traits among trees into genotypic and micro-environmental components. The first remarkable result is that a large proportion of this variation is due to genotypic effects among female parents. This proportion ranged from 41% to 67% for traits linked to pollen growth, heterospecific seed production and germination, but was lower (15%) for progeny height. This illustrates that pollen grain genotypes or recognition phenotypes might control the initial stages of pollen–pistil interactions (capture, germination and elongation) via intrinsic mechanisms (McClure, 2004; Swanson et al., 2004). Based on previous QTL detection studies (Saintagne et al., 2004), the pedigree that was used here is considered to represent a fairly large genetic background for different adaptive traits. However, all mother trees are full-sibs from parents that were sampled in two geographically close populations (40 km) at the European scale. It is thus reasonable to expect an even higher genetic variability of crossing behaviours and RI traits between and within natural populations of both species across their large geographical range and various types of habitats. Indeed, field trials of different geographical populations have revealed significant differences for many adaptive traits within both species (Ducousso et al., 1996; Jensen, 2000; Jensen & Hansen, 2008; Vitasse et al., 2009).
In contrast, differences between the genotypes used as pollen parents were not significant, either in their contribution to progeny genotypic frequencies or in that to early growth performances. This may be due to the low power of the tests applied on pollen mixtures given the amount of offspring harvested, but also to the relatively narrow genetic background of pollen parents (from a single population) compared to that expected across the species range. Alternatively, reduced variability for crossing compatibility of the species used as pollen parent, Q. petraea, could be linked to its general ability to pollinate the pioneer Q. robur species, as a mean to colonize new environments (Petit et al., 1997, 2004). Increasing the geographical sampling scale of both parental species in controlled crosses could bring useful information on the extent of genetic diversity in traits associated with their RI barriers and on the variation in the strength of those barriers.
We also provided strong evidence that several traits are sensitive to micro-environmental variation, suggesting a large plasticity in the expression of RI between these oak species. This evidence comes from the use of covariates in general linear models. These covariates are particularly useful not only for increasing the power to detect genotypic effects in RI traits expression but also for refining the biological interpretation of these traits whose expression can depend on the micro-environment or, as previously shown, on the genetic or taxonomic backgrounds (Garnier-Géréet al., 2002; Petit et al., 2005). In this experiment, adding a micro-environmental component to the genotypic influence of the mother tree for pollen tube growth allowed much better prediction of the total variation observed. This means that genotypic effects in oaks interact with developmental and external factors, because tree height or possible response to the local micro-environment (via their neighbours’ height) could indeed be surrogates for resource acquisition. These covariates could thus be indirectly linked to soil moisture, temperature and atmospheric conditions that could affect pistil receptivity at the time of pollen availability in natural populations. Besides, how these levels of resource acquisition affect the targeted trait within each genotype depends on the genotypes themselves, as illustrated by their different regression coefficient signs in nested ancova models (see Table 5). One possible explanation is that depending on each genotype average phenological development at the time of pollination, higher growth (for a clonal replicate of this genotype) could be either a positive factor for flower development or a negative factor due to a potential trade-off between resources allocated to growth or reproduction. Another one is that the heterogeneity of local environment effects and responses across genotypes would produce these observations. More clonal replicates across a set of different genotypes for their development and shape would help to better estimate and understand these patterns.
Compared with pollen–pistil interaction traits, none of the tested covariates was significant for seed production (at constant inflorescence number). This could mean that intrinsic and genotypic factors have a relatively greater impact on the last stages of pollen–pistil interaction (i.e. pollen navigation towards the ovary) and on early postzygotic mechanisms affecting embryo and seed formation. However, more than 40% of the total variation in seed production is still unexplained by genotypes, the rest being in the residual. This variation might therefore still be due to unexplained micro-environmental variation or genotypic variation (via the male parents for example). The germination rate of hybrids also depends significantly on mother tree genotypes and is correlated (at the genotypic mean level) with seed unit weight, but its variation is additionally much better predicted when the micro-environmental component of seed unit weight is tested in the model within each genotype (see Table 5).
Developmental plasticity or genotype × micro-environment interactions are not often the focus of speciation research, so their importance might have been underestimated for both early and final stages of speciation (West-Eberhard, 2005; Härdling et al., 2009; Levin, 2009). In different tree species, however, there is evidence that temperature, interannual changes or environmental stress can affect pollen performance and competition, but also interact with genotypes for fertilization success (Pasonen et al., 2000; Williams et al., 2001; Hedhly et al., 2005; Nakanishi et al., 2005). These and our results call for a better integration of plasticity when studying the evolution of reproductive barriers among tree species.
Reproductive barriers vs. hybridization in Q. robur and Q. petraea?
We may ask how new results showing fairly strong prezygotic reproductive barriers can be reconciled with previous evidence of hybridization between these oak species in Europe (e.g. Lepais et al., 2009; Arnold, 2006). In the last few years, hybridization patterns have been shown to vary with geography, species pairs, individuals’ reproductive success, life stages, levels of environmental disturbance and species relative occurrence (Curtu et al., 2007, 2009; Jensen et al., 2009; Lepais et al., 2009; Lepais & Gerber, 2011). Despite this recent progress, we still only have a very partial view of the hybridization dynamics and history in any of these natural stands. Among the stands studied for genetic assignments or parentage analyses, only three included Q. petraea and Q. robur as either dominant species or in similar proportions to other species (Q. freinetto and Q. pubescens): Petite Charnie, central France (PC), Velling, Denmark (VE), Bejan Forest, Central Romania (BE). Overall, a very low occurrence of robur × petraea hybrids were observed in France and Romania, in contrast to a much higher proportion in Denmark. In PC, 5.7% of adults were assigned as F1 or later-generation hybrids between Q. petraea and Q. robur using the structure software (Lepais et al., 2009). In BE, only one adult was assigned as an F1 robur × petraea hybrid (Fig. 4 in Curtu et al. (2007)), less than with other species pairs. Parentage analyses performed on a restricted number of trees also revealed one Q. robur individual producing F1 hybrids seeds with Q.petraea. Our evidence for strong prezygotic isolation barriers is thus clearly consistent with results from these two stands, considering also their geographical proximity to the material that we used (for both female and pollen donors, see Material and methods), in contrast with the very large distribution of these species in Europe. However, in the Danish stand further North, VE, where Q. petraea was twice as frequent as Q. robur, F1 hybrids at the seedling and acorns stages ranged from 48% to 55% using 11 Q. robur as females, and from 15% to 17% using 29 Q. petraea as females (Jensen et al., 2009). Also, 15–20% of adult trees were assigned as intermediates with Simple Sequence Repeat data, despite belonging to clear morphological clusters. The contrast in hybridization rates between the three oak forests also means that both pre- and postzygotic barriers could be highly variable across latitude. This variation could further be related to diverse ecological conditions between Southern and Northern Europe and also to different ages of stand establishment during oaks recolonization history after the last glaciations.
We demonstrated strong postmating intrinsic RI barriers between Q. robur and Q. petraea at the prezygotic stages in particular, and significant but weaker postzygotic barriers that might be stronger in natural conditions. However, we still need to better understand how these barriers can be reconciled with the general sharing of polymorphisms between these two species, and how they evolved in different natural stands. In a context of ecological speciation in sympatry, divergent selection to different habitats could lead to assortative mating within each habitat (Bridle & Ritchie, 2001; Via, 2001, 2009). We therefore clearly lack data on premating isolation and flowering traits that we could not assess here with controlled crosses. Another prediction of divergent selection between habitats is that extrinsic postzygotic isolation by selection on maladapted migrants or hybrids could be strong (Rice & Hostert, 1993; Via, 2001). More data are thus also needed to assess hybrids’ fitness and characterize the efficiency of selection pressures at a young age with in situ experiments. Our results revealed both a large genotypic diversity and an important plasticity in the expression of most reproductive barriers. This further stresses the important role of ecological conditions in the evolution of those barriers, and the need for more information on their variation across mixed populations with different environmental and demographic characteristics. Finally, important insights will certainly be gained on the nature and evolutionary rate of genetic changes that cause speciation under divergent selection by focusing on RI genetic architecture, and in particular by attempting to characterize ‘speciation genes’, i.e. those that contribute to RI [see Rieseberg & Blackman (2010) and Nosil & Schluter (2011) for recent reviews]. More population genomics approaches in oaks would also greatly help in quantifying the amount of genes or regions that show reduced gene flow (higher divergence) between species (Via, 2009; Butlin, 2010).
We thank V. Castric, A. Ducousso, O. Lepais, A. Ressayre and E. Porcher for advices and discussions when preparing the 2009 crossing campaign; Personnel from BIOGECO and from the Bourran experiment unit (INRA Bordeaux-Aquitaine), M.-L. Greil and D. Monty in particular, for their help in sampling and field traits assessments; The ANR TRANSBIODIV (Contract no. 06-BDIV-003-04) for funding minor equipment for performing crosses experiments; A. Ressayre and the Plateau Technique ‘Imagerie, Cytologie’ (INRA Bordeaux-Aquitaine) for their help in realizing cytological observations; E. Guichoux and L. Lagache for their advice in microsatellite design and quality checking; R. Petit, S. Mariette, M. Keatley and two anonymous reviewers for constructive comments on a previous version of the manuscript. Microsatellite genotyping was performed at the Genotyping and Sequencing facility of Bordeaux (supported by grants from the Conseil Régional d’Aquitaine no.20030304002FA and 20040305003FA and for the European Union, FEDER no.2003227). P. Abadie received a Ph.D. grant from the ‘Ministère de l’Education Nationale, de l’Enseignement Supérieur et de la Recherche’ of France, and additional funding from the EVOLTREE network of excellence (EU Contract no. 016322), and from the French Bureau of Genetic Resources (Now FRB, http://www.fondationbiodiversite.fr/, Contract no. P03002).