Introduction

Hybridisation is considered to be widespread among plants and has been estimated to be involved in the differentiation of up to 70% of the Angiosperms (Whitham et al, 1991). As recently reported (Rieseberg and Wendel, 2004), phylogenetic studies conducted over the past decade in several plant genera have shown that the ‘marks’ of past hybridisation are considerably more frequent than previously believed (eg, Cronn and Wendel, 2004; Doyle et al, 2004). Hybridisation is therefore a prominent factor in plant evolution which may trigger the differentiation of new lineages (Arnold, 1997; Rieseberg, 1997). However, natural hybridisation does not seem to be distributed equally among plant taxa, and some taxonomic groups may be particularly inclined to gene exchanges (Ellstrand et al, 1996). Moreover, most reported cases of hybrid speciation based on molecular markers have involved polyploid complexes. There are few convincing examples of natural homoploid hybrid speciation in which the relative genetic contribution of parental species to hybrids has been documented (Rieseberg, 1997; Cronn and Wendel, 2004).

Oaks (Quercus genus, Fagaceae) consist of several hundred diploid species (2n=2x=24), which are famous for their high propensity to interbreed. However, the occurrence, frequency and geographical extent of hybridisation and parental introgression vary greatly among oak species and locations (Craft et al, 2002; Dodd and Afzal-Rafii, 2004). Despite weak interspecific barriers, oak species involved in interspecific crosses usually remain distinct at the morphological, physiological, genetic and ecological levels (Whittemore and Schaal, 1991). In previous studies, hybrid oak individuals or populations have been reported to combine the morphological traits of both parental species (eg Bacilieri et al, 1996), and countless oak species have been described as hybrids, exclusively on the basis of morphological observations (Castroviejo et al, 1990). Their hybrid origin should in most of these situations be ascertained through genetic analysis, which, together with ecological studies, should also ensure that the assumed hybrid species is not a transient population. In this respect, the only convincing case is Quercus crenata Lam., a natural hybrid between two closely related parental species, Q. cerris and Q. suber (Bellarosa et al, 1996).

In the present work, we focus on Q. afares Pomel, an endemic North African deciduous species which occurs in one population in Northern Tunisia (about 550 trees; L. Toumi, personal observations), as well as in several populations in Northern Algeria covering about 12 000 ha (Figure 1a) (Boudy, 1950; Quézel and Santa, 1962). Q. afares possesses 24 chromosomes (A. Zine El Abidine, personal communication) as other oak species, and is therefore diploid. It has been assigned to the Cerris Spach Quercus subsection, because of morphological similarity with two other species from this subsection, namely Q. cerris L. (southern Europe) and Q. castaneifolia Mayer (Caucasus and Iran) (Boudy, 1950; Maire, 1961). However, closer examination indicates that Q. afares combines morphological, physiological and ecological traits of the semideciduous Q. canariensis Willd. and the evergreen Q. suber species (Boudy, 1959; Maire, 1961; Tutin et al, 1993), from the genetically very distant subsections Quercus Örsted (European white oaks) and Cerris Spach, respectively (Manos et al, 1999). The fossil record shows that these two subsections were already very distinct at the end of the Tertiary period (Kvacek and Walther, 1989).

Figure 1
figure 1

(a) Geographic distribution of Q. afares (grey), Q. suber (north of the dotted line) and Q. canariensis (north of the interrupted line) in North Africa from Quézel and Santa (1962). The location of the studied Tunisian and Algerian contact zones are indicated by arrows. (b) and (c) Topography and sampling transects (indicated by arrows) of Q. afares (Qa), Q. suber (Qs) and Q. canariensis (Qc) across the Tunisian (b) and Algerian (c) contact zones. Bold characters indicate the locally predominant species.

Full size image

Since Q. afares combines morphological, physiological and ecological traits from Q. suber and Q. canariensis, it could result from Q. suber × Q. canariensis hybridisation. To examine this possibility and clarify the genetic relationships among the three species, a genetic analysis was conducted in two areas, located in Algeria (Tizi-Oufe-Llah) and Tunisia (Aïn Zena), where Q. afares, Q. suber and Q. canariensis trees were either mixed or geographically close (contact zones). Populations of both Q. suber and Q. canariensis growing as monospecific stands outside the contact zones were used as genetic references for the assumed parental species. Polymorphism at five allozyme loci and RFLP variation of chloroplast DNA (cpDNA), maternally inherited in oaks (Dumolin et al, 1995), were used to address five questions: (1) Does Q. afares originate from a Q. suber × Q. canariensis hybridisation? (2) If so, what is the relative contribution of each parental species to the nuclear genome of Q. afares? (3) Which species served as mother in the original hybridisation event? (4) Can Q. afares be considered as a stabilised species? (5) Do the Tunisian and Algerian Q. afares have a common origin? We have also used our results to examine how Q. afares may have maintained itself over time and even spread across relatively large areas.

Materials and methods

Species and sampling

Q. suber and Q. canariensis are sympatric with Q. afares over most of its geographical distribution (Figure 1a). Q. suber and Q. afares share a biennial reproductive cycle, a trunk covered with corky bark (although thinner in Q. afares) and similar fruit morphology (Maire, 1961). Trees of both Q. afares and Q. canariensis can reach 25–30 m in height and constitute dense stands. Leaf morphology in Q. afares is intermediate between that of Q. suber and of Q. canariensis (Maire, 1961; Tutin et al, 1993). With regard to ecological requirements, Q. afares and Q. suber do not occur on limestone, whereas Q. canariensis is indifferent to substrate. Both Q. afares and Q. canariensis grow in humid, and even cold (−10°C), Mediterranean areas whereas Q. suber is restricted to the warmer humid and subhumid Mediterranean areas. Q. afares has been reported to grow from 200 m (in mixed stands with Q. suber) to 1200–1600 m of elevation where mixed stands with Q. canariensis are found, particularly above 700 m. Q. afares also constitutes monospecific populations, especially above 1200 m and on soils damaged by fire (Boudy, 1959; Maire, 1961).

A total of 25–53 individuals were sampled in seven sites located in North Africa for allozyme polymorphism (see Table 1 for geographic location, sample size and population abbreviation). One population of each species (Q. afares, Q. suber and Q. canariensis) was sampled in each of two large contact zones involving the three species in Tunisia (Aïn Zena) and Algeria (Tizi-Oufe-Llah) (referred to as ‘contact populations’; Figure 1b and c and Table 1). In Aïn Zena, all the Q. afares individuals were mixed with Q. suber and especially Q. canariensis individuals, whereas the larger Tizi-Oufe-Llah area included a monospecific stand of Q. afares individuals and, at one edge, a mixture of Q. afares and Q. canariensis individuals (Figure 1b and c). In addition, four monospecific populations of Algerian and Tunisian Q. suber and Q. canariensis were also sampled (referred to as ‘pure populations’, Table 1). We also included one pure population of Q. canariensis sampled in Morocco (Table 1), and four pure populations of Q. suber previously characterised by Toumi and Lumaret (1998) using the same allozymes: three are from Tunisia (Aïn Draham: 30 individuals, Nefza: 31 and Tabarka: 36) and one from Morocco (Maâmora forest: 25). These pure populations were used as genetic reference for each species. For cpDNA analysis, five individuals were randomly sampled from each pure Algerian and Tunisian Q. suber and Q. canariensis populations, and their chlorotypes were subsequently used as reference. All the Q. afares, Q. suber and Q. canariensis individuals from the mixed populations were also analyzed. A three-letter code will be used below for designating populations referring to country (A=Algeria, T=Tunisia, M=Morocco), population status (c=contact, p=pure) and species (a=Q. afares, s=Q. suber, c=Q. canariensis).

Table 1 Geographical location, code and sample size (N) of the contact and/or pure Q. afares, Q. suber and Q. canariensis populations studied in Algeria, Tunisia and Morocco
Full size table

Species identification was based on morphological characters detailed in Maire (1961) for Q. afares, and in Tutin et al (1993) for Q. suber and Q. canariensis. A few leafy branches per individual were collected in situ, except for the three mixed Algerian populations for which branches were sampled from trees maintained in Montpellier. These trees originate from acorns collected on separate mother trees and sown in Montpellier about 15 years ago.

Allozyme analysis

Proteins were extracted from fresh leaves in a Tris-HCl buffer (pH 7.6) and the total protein extracts were then stored at −80°C until analysis, as described in Toumi and Lumaret (2001). Five presumably neutral loci from different enzyme systems were analysed: PGI-1 (phosphoglucose isomerases, EC 5.3.1.9.), LAP-1 (leucine aminopeptidases, EC 3.4.11.1.), AcPh-1 (acid phosphatases, EC 3.1.3.2.), ADH-1 (alcohol dehydrogenases, EC 1.1.1.1.) and IDH-1 (isocitrate dehydrogenases, EC 1.1.1.42.). Horizontal starch-gel electrophoresis and staining were performed as described in Toumi and Lumaret (2001).

Mendelian transmission and homology of electrophoretic bands at these five unlinked loci were previously studied in controlled crosses involving a large number of oak species and several interspecies hybrids (eg in Michaud et al (1995), Samuel et al (1995), Toumi and Lumaret (2001) and references therein). In addition, normal allele segregation was observed in intraspecific open pollinated Q. afares progeny. In the present study, bands with similar migration distances at a given locus in all oak species studied under identical experimental conditions were therefore considered as the same allele. Each allele was numbered as a function of migration distance relative to allele 100 (the most frequent allele observed in Q. ilex, originally used as reference species). Electrophoreses were run using Q. suber and Q. ilex standards.

Analysis of cpDNA

Chloroplasts were isolated from 4-g aliquots of freeze-dried powder, and cpDNA was extracted from chloroplasts as described in Lumaret et al (2002). Total cpDNA was cut with four endonucleases (HhaI, AvaI, DraI and BamHI), which provided clear restriction patterns characterised by a large number of fragments (usually over 45). DNA restriction, electrophoresis and DNA staining techniques were similar to those described in Lumaret et al (2002). Note that no statistical analyses were performed because of very limited variation.

Data analysis

Allele frequencies at allozymic loci were calculated for each locus and population. Two reference pure populations, representative of allelic composition in pure North African Q. suber and Q. canariensis, were constituted by pooling data from Aps, Tps and the four populations studied by Toumi and Lumaret (1998) for the former, and from Apc, Tpc and Mpc for the latter.

The respective contributions of Q. suber and Q. canariensis to Q. afares genome were evaluated based on a hybrid index (HI), defined as the number of observed Q. suber alleles. The per-locus HI is zero in Q. canariensis and two in Q. suber, and the multilocus HI is the sum of HI over all loci. Alleles diagnostic of a given species were pooled in two synthetic alleles (S for Q. suber and C for Q. canariensis) per locus. Alleles common to both species were assigned according to their coordinates on the first axis of a correspondence analysis realised on the pure Q. suber and Q. canariensis populations (see Daguin et al, 2001). HI was estimated per locus and over all loci for each Q. afares individual, and then averaged per population. The difference among loci was tested within and among Q. afares populations using an analysis of variance (ANOVA, conducted using SAS; SAS Institute Inc., 2001). The multilocus HI distribution was compared between the two Q. afares populations using an ANOVA.

In each Q. afares population, we estimated the pairwise associations between the Q. suber and Q. canariensis genomes within locus (K1,1, corresponding to an averaged Hardy–Weinberg (HW) disequilibrium over all loci) and between loci within each genome (K0,2, corresponding to an averaged parental linkage disequilibrium over each pair of loci), after encoding genotypes using synthetic alleles (Barton, 2000). Calculations were made using N Barton's package available at the web site http://helios.bto.ed.ac.uk/evolgen/ and Mathematica 3.0 (Wolfram, 1996). The likelihood was estimated assuming that the Q. afares population is at HW linkage equilibrium (HWLE), and was compared to those of models allowing within-genome pairwise associations (K0,2, linkage disequilibrium), between-genome pairwise associations (K1,1, HW disequilibrium), both associations and higher-order associations. These models were also tested allowing different contributions to the associations between loci. The best model was determined based on Akaike information criterion (see Barton, 2000). The maximum theoretical disequilibria (Kmax) was estimated by mixing pure Q. suber and Q. canariensis populations as described in Bierne et al (2003). When a locus distorted the global disequilibrium estimation, it was removed from the genotype matrix. Departure from HW disequilibrium was then tested at this locus by calculating Fis. As HW disequilibrium can result from forces other than hybridisation (eg, inbreeding), we also estimated Fis using only those Q. afares homozygotes for alleles of a single mother species (‘species-Fis’). All Fis were estimated following Weir and Cockerham (1984) using GENEPOP 3.2 (Raymond and Rousset, 1995).

In each Q. afares population, the frequencies of particular parental genes combinations, reflecting per-locus respective contribution and particular associations of Q. suber and Q. canariensis genomes, were calculated and compared to those of other genotypes.

Nei's genetic distance was calculated between all pairs of populations, and used in a nonmetric multidimensional scaling analysis (or proximity analysis; Escoufier, 1975). The same distances were also used in a hierarchical cluster analysis (UPGMA), with the ‘average distance’ used as the clustering criterion (Roux, 1985). Clusters at different levels of agglomeration (total ranging from 0 to 100%) were mapped onto the diagram obtained from the multidimensional scaling analysis (see Toumi and Lumaret (2001) and references therein). These analyses were performed using BIOMECO 3.7 (Lebreton et al, 1987).

Results

In the reference pure populations, distinct alleles were found in Q. suber and Q. canariensis at the PGI-1, AcPh-1 and ADH-1 loci, which were considered as diagnostic markers (Table 2). The LAP-1 and IDH-1 loci were semidiagnostic since one allele was shared between the two species (LAP-194 and IDH-1100). Three diagnostic Q. canariensis alleles were observed in the Tcs population and, reciprocally, six diagnostic Q. suber alleles were recovered in the Acc population (Table 2). Five local alleles (ie observed in a single country) were identified in the Q. suber and Q. canariensis populations (see Table 2). All were rare except PGI-180 in Tpc.

Table 2 Allelic frequencies at the five allozyme loci studied in pure reference, pure and contact populations of Q. suber, Q. canariensis and Q. afares
Full size table

The two Q. afares populations showed very similar allelic compositions, both combining alleles from the Q. suber and Q. canariensis populations that occur in the corresponding contact zones. However, at most loci, Q. suber alleles were more frequent than those of Q. canariensis (Table 2). On the contrary, some alleles with low to medium frequency in Q. canariensis (PGI-1120 and LAP-1103) were not observed in Q. afares populations. PGI-180 was the only local allele (found in Tpc) that was detected in Q. afares, and exclusively in the Tca population. At IDH-1, the common allele 100 was largely predominant in both Q. afares populations, whereas the Q. suber allele 110 was rare (Table 2).

In the reference pure populations, the diagnostic alleles were pooled in two synthetic alleles per locus (S and C). LAP-194 was common and considered as a Q. suber allele, according to results of the CA (data not shown). IDH-1100 could not be clearly assigned to a species, and IDH-1 was discarded when estimating the HI. The multilocus HI was high in both Q. afares populations (Table 3), and the estimates did not significantly differ between these populations. Q. suber alleles were predominant at the four loci (HI>1). HI per locus was not significantly different between the two populations, except at LAP-1 (lower in Tca than in Aca; Table 3). Most multilocus HI were higher than 6 (Figure 2) indicating again a predominance of Q. suber alleles. Note that no individual exhibited an HI of 4.

Table 3 Mean individual hybrid index in the Algerian and Tunisian Q. afares populations for each locus and over all of loci
Full size table
Figure 2
figure 2

Distribution of individual multilocus HI values (ie the number of Q. suber alleles per individual over the PGI-1, LAP-1, AcPh-1 and ADH-1 loci) in the Tunisian (filled bars) and Algerian (open bars) Q. afares populations.

Full size image

The multilocus disequilibria were also estimated once discarding IDH-1. A significant HW disequilibrium (K1,1), corresponding to a global deficit of heterospecific heterozygotes (SC), was found in both Q. afares populations. A significant linkage disequilibrium was found in both Tca (K0,2, indicating global within-genome pairwise associations) and Aca (K0,3, indicative of global lack of within-genome associations involving triplets of loci) (Table 4). In both populations, the observed values of HW and linkage disequilibria were much smaller than their maximum values (Kmax; Table 4). No significant improvement was achieved by allowing both interspecific (K1,1) and intraspecific (K0,2 or K0,3) allele associations, higher-order associations or different contributions among loci. In both populations, a deficit of heterozygotes was also found at IDH-1 (Aca: 0.730, P<0.001; Tca: 0.360, P=0.165). Species-Fis, estimated over the three most polymorphic loci (PGI-1, LAP-1 and ADH-1) in Q. afares individuals homozygous for Q. suber alleles (not enough data were available in other situations), did not significantly differed from zero (Aca: −0.188, P=0.151; Tca: 0.268, P=0.07).

Table 4 Estimates of multilocus disequilibria significantly differing from zero in the Algerian and Tunisian Q. afares populations
Full size table

In Aca, ‘combination 1’ (PGI-1, LAP-1, AcPh-1 and ADH-1 homozygous for the S allele and IDH-1 homozygous for the allele 100) was predominant over the 17 observed combinations, standing for 34.1% of the individuals (26.7% expected at HWLE). It also reached the highest frequency in Tca (30.0%, HWLE: 16.8%), together with ‘combination 2’ (PGI-1, AcPh-1 and ADH-1 homozygous for the S allele, LAP-1 homozygous for the C allele and IDH-1 homozygous for the allele 100; 23.3%, HWLE: 14.3%). In all, 10 combinations were observed in this population.

Results of the multidimensional scaling analysis based on Nei's distances are reported in Figure 3. Populations of Q. canariensis were clearly separated from other populations on axis 1 (clustering level: 8%), while Q. afares and Q. suber populations were discriminated on axis 2 (5%). The largest interspecific genetic distance was indeed found between pure Q. suber and Q. canariensis populations (1.698 on average). Q. afares was much more distant from Q. canariensis (0.935) than from Q. suber (0.098). On the other hand, low intraspecific genetic distances were observed among the six pure Q. suber populations (0.027), as well as among the three Q. canariensis (0.009) and the two Q. afares (0.060) populations.

Figure 3
figure 3

Position of contact and/or pure Q. suber, Q. canariensis and Q. afares populations from Algeria and Tunisia in the multidimensional scaling analysis based on Nei's genetic distances. The codes refer to country (A=Algeria, T=Tunisia) population status (c=contact, p=pure) and species (a=Q. afares, s=Q. suber, c=Q. canariensis). Populations were gathered at the 8 and 5% levels of the hierarchical cluster analysis.

Full size image

In pure populations, each endonuclease produced two very distinct cpDNA restriction patterns, one in Q. suber and one in Q. canariensis individuals. This allowed defining a single reference chlorotype per species over the four restriction enzymes used. All Q. afares individuals exhibited the Q. suber reference chlorotype. In mixed populations, all Q. suber and Q. canariensis individuals exhibited the reference chlorotype of their own species. No within-species variation was observed between the Algerian and Tunisian pure Q. suber, pure Q. canariensis and mixed Q. afares populations.

Discussion

Q. afares probably results from Q. suber × Q. canariensis hybridisation

Q. afares exclusively combines Q. suber and Q. canariensis alleles at each locus (Table 2), which suggests a hybrid origin. This is consistent with the fact that the range of both Q. suber and Q. canariensis includes that of Q. afares (Figure 1a) and that the latter species exhibits intermediate morphological, physiological and ecological traits between the other species (Boudy, 1959; Maire, 1961; Tutin et al, 1993). As underlined by Rieseberg (1997) in other situations, the probability of common ancestral (symplesiomorphic) or convergent origin is strongly reduced since Q. afares exclusively displays diagnostic alleles from the two presumed parents, and Q. suber and Q. canariensis belong to very differentiated clades (Manos et al, 1999). Moreover, most alleles observed in Q. afares are specific to either Q. suber or Q. canariensis and were not observed in Q. ilex or Q. coccifera, two evergreen species broadly distributed in Algeria and Tunisia (Toumi and Lumaret, 2001). Therefore, the possible contribution of these two species to the original current Q. afares populations can be ruled out. Q. castaneifolia and Q. cerris have initially been proposed as potential ancestors of Q. afares (see Introduction). The strongest argument against this hypothesis is that both species have never been found in North Africa.

Q. afares nuclear and chloroplastic genomes are dominated by Q. suber alleles

No first-generation hybrids between Q. suber and Q. canariensis were detected in the two Q. afares populations. Moreover, Q. afares genome was clearly dominated by Q. suber alleles at three of the five loci studied (Figure 2 and Table 3), even when Q. afares did not occur in mixed stands with Q. suber (ie in the Algerian contact zone; see Figure 1c). We also did not detect strong parental linkage disequilibrium or excess of heterospecific heterozygotes (Table 4), indicating the rupture of parental associations. These results suggest that the Q. suber × Q. canariensis hybridisation that produced Q. afares is not recent, which is consistent with the occurrence of pure Q. afares populations out of contact zones in Algeria (Boudy, 1959).

After the initial hybridisation event(s), several processes can account for the observed pattern. (i) Probably the most likely explanation is introgressive hybridisation, a process which has been considered as a potent force in differentiation of new species (Rieseberg and Wendel, 1993). Indeed, backcrossing may have been favoured by synchronous flowering phenology of early hybrids and Q. suber individuals, by a demographic superiority of Q. suber, or by better gamete compatibility (potentially linked to cytonuclear incompatibilities, see below), as reported previously in other studies (Cruzan and Arnold, 1993; Arnold, 1997). (ii) Recurrent mating between hybrids after the initial hybridisation is another possibility. However, this idea can hardly be reconciled with the predominance of Q. suber alleles at most loci in Q. afares, without the action of the other processes cited below. (iii) Genetic drift and bottlenecks, especially acting at the early stages of the hybridisation process when populations had small size, could explain the observed pattern. (iv) Selection acting on genes closely linked to neutral loci (see Rieseberg, 1997) could also explain the predominance of the Q. suber alleles. Previous work on Helianthus (Rieseberg et al, 1996) has already demonstrated that the parental contributions to hybrid species are determined by endo-selection against negative interactions between genes from different species (Dobzhansky, 1937; Muller, 1942). Exo-selection could have influenced the Q. afares genomic composition as well (see Gross and Rieseberg (2005) for a review and Cruzan and Arnold (1993) for an example in the genus Iris). Genotype–habitat correspondence has been considered as indicative of effective exo-selection pressure (Arnold, 1997). The species studied here may indeed be found as monospecific populations with strong habitat preferences, with Q. afares, Q. suber and Q. canariensis occupying deteriorated siliceous soils at high altitude, siliceous soils at low altitude and chalky soils, respectively (Boudy, 1959; Maire, 1961). Selection though cannot easily account for the pattern found at LAP-1, where we observed both a lower HI than at other loci and unequal parental contributions when comparing the two populations studied (Table 3). This locus might rather have evolved through genetic drift. (v) The predominance of Q. suber alleles in Q. afares may reflect an altered expression of Q. canariensis genes (gene silencing), resulting from epigenetic phenomena linked to the union of divergent genomes, as previously suggested in other plant hybrids (see Liu and Wendel (2003) for a review). However, all allelic combinations were observed on zymograms at all loci, which would imply incomplete gene silencing. Moreover, this hypothesis is expected to result in a general deficit of heterozygotes which was not observed. (vi) Gene conversion might also explain genome homogenisation of hybrid species (Sperisen et al, 1991), including hybrid oaks (see Bellarosa et al (2005) for Q. crenata). Here, it could have favoured the replacement of Q. canariensis genes by Q. suber ones in Q. afares, especially in the early stages of its differentiation. However, gene conversion (except if incomplete) does not account for the maintenance of variability at the loci studied. Moreover, it is a slow process, presumably acting at time scales much longer than the lifespan of species. In conclusion, several processes may account for the current structure of Q. afares genome, and further insight could be gained only by studying more populations and, especially, more loci.

A further important result is that all Q. afares individuals displayed the Q. suber chlorotype. Since chloroplast DNA is maternally inherited in oaks, this indicates that Q. afares differentiated from crosses where Q. suber was the maternal species. However, as already reported above, such crosses may have occurred either at the initial hybridisation stage, or later, during early backcrosses with Q. suber. In experimental studies, Boavida et al (2001) demonstrated that prezygotic barriers linked to pollen–pistil interactions strongly determined the direction of interspecific crosses between Mediterranean oak species. The role of Q. suber as maternal species has some empirical backing in the latter study, since the authors demonstrated that Q. suber could successfully be pollinated by Q. faginea, of which Q. canariensis was long considered as a subspecies. Unfortunately, reciprocal crosses and interspecific crosses including Q. canariensis were not tested by these authors.

Q. afares: a stabilised hybrid species

Arnold (1993) suggested that morphology, genetic characteristics and ecological exigencies must be taken into account to determine whether a hybrid species is stabilised. Q. afares is characterised by a specific morphotype (Maire, 1961) with little interindividual variation. Such morphological homogeneity contrasts with the great variability observed among transient hybrid oaks or introgressed individuals (Rushton, 1993; C Mir, personal observations). Q. afares also presents unique ecological requirements and habitat preference, combining those of its parents and idiosyncratic new traits (Boudy, 1959; Maire, 1961). Our genetic study further demonstrates that Q. afares corresponds to a genetic entity, clearly distinct from Q. canariensis and Q. suber populations (Figure 3). Moreover, Q. afares is reproductively autonomous. One reason is that the flowering periods of the three species (chronologically Q. canariensis, Q. afares, Q. suber; L Toumi, personal observations) usually do not overlap, favouring preferential mating within species. These considerations support the idea that Q. afares is a stabilised species. Further arguments include the high production of acorns in mixed populations (Toumi et al, unpublished data), a high density of seedlings on forest ground, an equivalent competitive ability of Q. afares when compared to its parents in contact zones (no species replacement has been reported), and even better fitness at high altitude and on impoverished soils (Boudy, 1950). We should note though that the presence of Q. afares over 1000 m might have been favoured by forestry treatment replacing Q. canariensis (less adapted to such altitude) to produce firewood (Boudy, 1950).

It remains unclear whether the genome of Q. afares will fix a new parental gene combination, as has been observed for various hybrid species (Rieseberg et al, 1990; Arnold, 1993; Rieseberg et al, 1996). Some gene combinations, one in Algeria and two in Tunisia, actually occurred at high frequencies. As excesses of homospecific homozygotes were not due to inbreeding (see species – Fis) and were not preserved across generations, these combinations could reflect preferential assortative mating between Q. afares individuals with similar genomic composition (Cruzan and Arnold, 1993), which would further favour genome fixation. This process could also be influenced by the introgression of parental genes through current gene flow. Further experimental information is necessary to clarify that point.

Relationships between Tunisian and Algerian Q. afares remain unclear

The Tunisian and Algerian populations of Q. afares may share a common hybrid origin, or result from independent hybridisation events. The first hypothesis implies the expansion of Q. afares after hybridisation, followed by population splitting (see Figure 1a). This cannot be evaluated in the absence of palaeobotanical data about the past distribution of Q. afares. In our study, some alleles (ADH-184/W and PGI-180; Table 2) were found in a single Q. afares population and in parental species of the same country exclusively, which might be indicative of independent origin. However, these alleles occur at low frequency, and their absence may be due to limited sampling size or genetic drift.

The cpDNA analysis was not helpful in distinguishing between the two hypothesis, since a single chlorotype was detected when pooling all Q. afares and Q. suber populations. Chlorotype differentiation between Algerian and Tunisian Q. afares, reflecting geographical chlorotype differentiation of Q. suber, could indeed have indicated double-hybrid origin. Further analysis of cpDNA using another technique (PCR-RFLP and eight combinations of chloroplast primers/endonuclease) also failed to reveal any variation in Q. suber (Lumaret et al, unpublished data) suggesting that the lack of variation is not due to the technique used here. The multiple origin hypothesis cannot therefore be evaluated. Note that our observations are not incompatible with a multiple origin: the selective forces involved in the differentiation of a new lineage might work well in a single direction under homogeneous ecological conditions (Rieseberg et al, 1996). The morphological and general genetic similarity of both Q. afares populations for nuclear and cytoplasmic markers could result from identical selective pressures.