• José Gabriel Segarra-Moragues,

    1. ARAID-UZ. Departamento de Agricultura y Economía Agraria, Escuela Politécnica Superior de Huesca, Universidad de Zaragoza. Carretera de Cuarte, Km 1, E-22071 Huesca, Spain
    2. E-mail:
    Search for more papers by this author
    • Present address: Centro de Investigaciones sobre Desertificación (CIDE, CSIC-UV-GV), Camí de la Marjal s/n, E-46470 Albal (Valencia), Spain

  • Fernando Ojeda

    1. Departamento de Biología, Universidad de Cádiz, Campus Río San Pedro, E-11510 Puerto Real, Spain
    2. E-mail:
    Search for more papers by this author


Understanding the processes of biological diversification is a central topic in evolutionary biology. The South African Cape fynbos, one of the major plant biodiversity hotspots out of the tropics, has prompted several hypotheses about the causes of generation and maintenance of biodiversity. Fire has been traditionally invoked as a key element to explain high levels of biodiversity in highly speciose fynbos taxa, such as the genus Erica. In this study, we have implemented a microevolutionary approach to elucidate how plant-response to fire may contribute to explain high levels of diversification in Erica. By using microsatellite markers, we investigated the genetic background of seeder (fire-sensitive) and resprouter (fire-resistant) populations of the fynbos species Erica coccinea. We found higher within-population genetic diversity and higher among-population differentiation in seeder populations and interpreted these higher levels of genetic diversification as a consequence of the comparatively shorter generation times and faster population turnover in the seeder form of this species. Considering that genetic divergence among populations may be seen as the initial step to speciation, the parallelism between these results and the pattern of biodiversity at the genus level offers stimulating insights into understanding causes of speciation of the genus Erica in the Cape fynbos.

Biodiversity hotspots are mostly circumscribed to tropical latitudes (Myers et al. 2000). Out of the tropics, the most remarkable biodiversity hotspot is the South African fynbos biome, within the Cape Floristic Region (CFR). Its outstanding levels of plant species richness and large concentration of endemic taxa raise crucial questions regarding the causes of speciation and the maintenance of biodiversity (Linder 2003; Latimer et al. 2005; Barraclough 2006). The genus Erica, with ca. 700 species and a high level of narrow endemism, is the epitome of biodiversity in fynbos (Oliver et al. 1983; Linder 2003).

Fire has been traditionally invoked as a key element accounting for the high levels of biodiversity in fynbos woody plants (Cowling 1987; Cowling et al. 1996). Regarding their response to fire, two basic regeneration modes, “seeder” (fire-sensitive) and “resprouter” (fire-resistant), are recognized. Both of them confer resilience to recurrent wildfires at the population level, but they result in two contrasting population dynamics. Although resprouter populations are composed of multiple cohorts that survive following fires (i.e., overlapping generation dynamics), populations of seeder individuals tend to have even-aged cohorts (i.e., nonoverlapping generation dynamics; Bond and van Wilgen 1996; Ojeda et al. 2005). This is particularly true when germination from the seed bank is triggered by fire (fire-recruiting species; Bond and van Wilgen 1996), because the seed bank becomes depleted after the passage of fire and seedling recruitment is thus always preceded by the adults’ death. Thus, seeder populations have shorter generation times as well as distinct and faster population turnovers (each fire cycle) than resprouter populations (Wells 1969; Ojeda et al. 2005).

Wells (1969) suggested that higher levels of diversification in seeder lineages are to be expected in woody taxa from fire-prone ecosystems that contain both seeder and resprouter species. Incidentally, species diversity and narrow endemism in the speciose genus Erica are tightly associated with the seeder habit (Ojeda 1998). Higher frequency and intensity of natural selection in seeder populations, favored by their comparatively faster generation turnovers, would increase genetic differentiation among populations and, hence, speciation in seeder lineages (Wells 1969). Certainly, genetic differentiation among populations within a given species may be seen as the initial step to speciation (Avise 2000; Coyne and Orr 2004). Several studies have actually found a positive relationship between population genetic differentiation and species diversity across latitudinal gradients (Martin and McKay 2004; Eo et al. 2008).

Since Wells (1969), several authors have put forward more indirect evidence to support his hypothesis (e.g., Ojeda 1998; Cowling and Pressey 2001), or to challenge it (e.g., Bond and Midgley 2003; Lamont and Wiens 2003; Perry et al. 2009). Barraclough and Savolainen (2001) found a positive relationship in flowering plants between the number of species within a clade and the rate of neutral molecular evolution, and suggested that rapid speciation is linked to faster rates of molecular change. Following this argument, Verdú et al. (2007) could not find statistical support for either higher molecular evolutionary rates or higher diversification in seeder lineages than in resprouter from Mediterranean-type ecosystems, concluding that Wells’ hypothesis might not hold as a general rule. However, more recently, Smith and Donoghue (2008) highlighted the existence of a general, consistent relationship between short generation times and high rates of molecular change in flowering plants.

In any case, at the within-species level, high rates of genetic differentiation among populations are mainly driven by demographical stochasticity (Whitlock 1992) and local processes of extinction and recolonization (i.e., population turnover), so long as most individuals colonizing an extinct patch originate from a single source population (the propagule-pool model; Slatkin 1977; Wade and McCauley 1988; Pannell and Charlesworth 1999, 2000). In this sense, postfire recruitment could be considered as a particular case of recolonization from a single source (the seed bank; see Gotelli 1991) which, moreover, does not contribute to gene flow among sites, thus enhancing differentiation. Hence, dynamics of fire-recruiting seeder populations would fit a model of extinction and propagule-pool recolonization (i.e., rapid population turnover), whereas that of resprouter would resemble the case with no local extinction (i.e., slow population turnover). Accordingly, genetic differentiation among seeder populations is expected to be higher than among resprouter ones.

However, diversifying seeder populations—and potentially “incipient species”—would also be prone to extinction, particularly under unpredictably fluctuating environments, owing to their non- (or little) postfire adult plant survival and subsequent lack of reproductive storage potential over generations (Warner and Chesson 1985; Higgins et al. 2000). Such counteracting higher extinction probability would hence account for the apparent lack of higher species diversification in seeder lineages reported by Verdú et al. (2007; see also Perry et al. 2009). In fact, high rates of species proliferation also require the existence of low extinction risks (Dynesius and Jansson 2000; Mittelbach et al. 2007).

In this study, we used a microevolutionary approach to test whether genetic differentiation among populations in the genus Erica is more marked in seeder populations (short generation times, rapid population turnover) than in resprouter populations (long generation times, slow population turnover). We have focused on the fynbos species Erica coccinea L., which includes distinct seeder and resprouter populations (Ojeda 1998), and thus avoids possible confounding effects attributable to independent phylogenetic histories. Specifically, we have analyzed the genetic structure of seeder and resprouter groups of populations of this species by means of microsatellite markers. Our results provide evidence of a strong association between post-fire regeneration modes and genetic variation and differentiation among populations. In a more general context, they illustrate the role of contrasting population dynamics—rapid versus slow population turnover—in the development of genetic divergence among populations. Finally, this study offers stimulating insights into the causes of diversification, and hence speciation, in the highly speciose genus Erica in the CFR fynbos.

Materials and Methods


Erica coccinea (Ericaceae) is a relatively abundant and widespread heath species in the CFR fynbos under mesic conditions. Besides its conspicuous variation in flower color (Rebelo and Siegfried 1985; see Table 1), the most remarkable morphological feature of this species is the existence of distinct seeder and resprouter individuals (Ojeda 1998; Bell and Ojeda 1999), frequently (but not always) in disjunct populations. Both seeder and resprouter populations are common in fynbos communities of coastal mountains and hills of the south-western CFR, characterized by reliable winter rainfall and mild summer drought (Ojeda 1998; Ojeda et al. 2005). In more inland mountains, where winter rainfall becomes less reliable and summer drought somewhat more intense, seeder populations become less abundant owing to their comparatively lower success under such climatic conditions (Ojeda 1998; Ojeda et al. 2005). This accounts for an apparent north-south latitudinal partitioning of seeder and resprouter populations (Ojeda et al. 2005) although, apart from winter rainfall reliability and summer drought strength, most environmental conditions (e.g., soil, mean annual rainfall, fire regime) are similar across the CFR mesic mountain fynbos (Campbell 1986; Ojeda 1998).

Table 1.  Population data and genetic diversity indices in 11 seeder and 11 resprouter populations of Erica coccinea for eight microsatellite loci.
Habit andFlowerLatitudeLongitudeAltitude PopulationN1A1HO1HE1FIS1
populationscolor  (m) size     
  1. 1N=sample size; A=mean number of alleles per locus; HO, HE, observed and expected heterozygosity, respectively; FIS=inbreeding coefficient.

 S01-TMNP. Below Maclear's beaconYellow33° 58′ 35″S18° 25′ 15″E800>500307.880.6330.637+0.006ns
 S02-Kleinmond. Kogelberg NRYellow34° 19′ 13″S18° 57′ 43″E 65>500299.380.6410.758+0.154***
 S03-Caledon to Hermanus. Shaw's PassYellow34° 19′ 03″S19° 24′ 24″E226 100–200198.000.7110.800+0.112***
 S04-Hermanus. Camphill villageYellow34° 23′ 10″S19° 13′ 22″E123 150–300247.000.7030.714+0.015ns
 S05-De Hoop NR. PotbergYellow34° 22′ 19″S20° 32′ 33″E287>30000309.000.6420.720+0.109***
 S06-Cape Agulhas. SandbergOrangish34° 48′ 11″S19° 58′ 34″E 96>300003011.000.6960.757+0.080*
 S07-Cape Agulhas. SoetanysbergGreenish34° 45′ 18″S19° 51′ 31″E 80>20003011.500.7040.762+0.076*
 S08-Napier. Napierberg. FM antennaYellow34° 31′ 41″S19° 53′ 17″E558 500–10003112.250.7140.757+0.057ns
 S09-Hermanus. Vogelgat NRYellow34° 24′ 00″S19° 10′ 18″E125 200–300249.250.6460.741+0.128**
 S10-TMNP. Cape of Good HopeYellow34° 21′ 06″S18° 29′ 07″E105 100–150208.000.6810.674−0.012ns
 R05s-Bot Rivier. Honingklip 34° 16′ 53″S19° 08′ 52″E145 100–150107.750.7340.785+0.065*
 R01-TMNP. Blackburn ravine trailRed34° 03′ 21″S18° 22′ 19″E205 300–500308.630.6830.732+0.066***
 R02-TMNP. Devil's PeakRed33° 57′ 13″S18° 26′ 21″E953 200–300207.500.6380.661+0.036ns
 R03-Tulbagh. Tulbagh Waterval NRRed33° 22′ 43″S19° 06′ 48″E491 200–300316.750.5770.636+0.093*
 R04-Franshoek (Fh Pass) Mont RochelleRed33° 53′ 46″S19° 09′ 37″E897 200–300197.620.7700.729−0.056ns
 R05r-Bot Rivier. HoningklipRed34° 16′ 53″S19° 08′ 52″E145>500278.000.7040.759+0.072*
 R06-Stellenbosch. Swartboskloof h. trailRed33° 59′ 56″S18° 57′ 48″E558>500308.250.6540.705+0.072*
 R07-Caledon. SwartbergRed34° 13′ 10″S19° 25 46″E364 300–400176.880.6990.707+0.012ns
 R08-Greyton. Boesmankloof h. trailRed34° 02′ 07″S19° 37′ 23″E280 300–400167.380.7190.689−0.043ns
 R09-Swellendam. Marloth NRRed+yellow33° 59′ 56″S20° 26′ 47″E225>5000307.880.6830.726+0.058*
 R10-Heidelberg. Grootvaderbosch NRRed+yellow33° 58′ 52″S20° 51′ 08″E315 200–300156.380.7170.660−0.086ns
 R11-Riversdale. Krystal Kloof h. trailRed33° 57′ 43″S21° 15′ 15″E780 400–500277.130.6670.732+0.090ns

Resprouter adult plants are frequently multistemmed, indicating post-disturbance regrowth, whereas seeder ones are single stemmed. More importantly, resprouter individuals present a swelling modification of the stem base (lignotuber) bearing high bud activity on its surface and thick roots with abundant tissue for starch storage, whereas seeder ones lack these features (Fig. 1). These morphological, anatomical, and functional differences seem to have a strong genetic basis and are detected even from early seedling stages (Verdaguer and Ojeda 2002, 2005), making seeder and resprouter individuals readily distinguishable from each other, even when they co-occur. Otherwise, seeder and resprouter plants are similar in flower shape and reproductive ecology—bird-pollination and passive (wind) seed dispersal—although their flowering phenologies do not overlap. Flowering spans from late January to early June in resprouter plants (peak in February March), and from late August through November in seeder plants (peak in September–October; F. Ojeda, unpubl. data).

Figure 1.

Representative image of the stem base and root system of seeder and resprouter plants of Erica coccinea. Resprouter adult plants are frequently multistemmed, indicating postdisturbance regrowth, whereas seeder ones are single-stemmed. Note the conspicuous swelling modification of the stem base (lignotuber) in the resprouter individual, bearing high bud activity on its surface, and the thick roots, and the lack of these features in the seeder individual.

Both seeder and resprouter E. coccinea plants are fire-recruiters (sensu Bond and van Wilgen 1996). They set flowers and fruits every year after reaching sexual maturity but seeds remain dormant in a soil-stored seed bank until the passage of fire, which breaks seed dormancy and, hence, triggers germination and recruitment. Presumably, seed banks become depleted after post-fire recruitment in both regeneration forms (Ojeda et al. 2005), although this still needs empirical verification. Therefore, because a large proportion of resprouter adult plants survive fire and regenerate their aboveground biomass, they may resume sexual activity and again supply seeds into the seed bank, thus experiencing further postfire recruitment events. By contrast, offspring recruitment in seeder plants is preceded by the succumbing of adults to fire. Hence, it can be stated that seeder populations in this species have shorter generation times and faster population turnovers than resprouter populations.


Ten seeder populations (i.e., composed of only seeder individuals) and 10 resprouter ones (i.e., composed of only resprouter individuals) were sampled across a large part of the species’ geographical range (Fig. 2). They included 267 individuals from the seeder populations (S01–S10) and 235 individuals from the resprouter ones (R01–R04, R06–R11). Besides, one mixed population (R05) was sampled including 10 seeder and 27 resprouter individuals, but were considered as separate seeder (R05s) and resprouter (R05r) populations in all subsequent analyses. Therefore, the overall sampling included 539 individuals, 277 and 262 from 11 seeder and 11 resprouter populations, respectively (Table 1, Fig. 2). All populations except the R08 and R11 were from mature fynbos stands, that is, more than 10 years after the last fire, and in all instances only adult individuals were sampled (resprouting adults in R08 and R11).

Figure 2.

Geographical location of the seeder (white dots) and resprouter (black dots) populations of Erica coccinea sampled in this study. The gray dot (R05) corresponds to a location with both resprouter (R05r) and seeder individuals (R05s; see Material and Methods).


Fresh leaves were sampled from each of the 539 individuals, dried on silica gel (Chase and Hills 1991), and stored at −80°C until DNA extraction. Approximately 100 mg dry weight per sample was used for DNA extraction. Dry material was reduced to fine powder using 2.3 mm stainless steel beads on a Mini-beadbeater-8 cell disrupter (BioSpec, Bartlesville, OK). DNA was extracted using the SpeedTools plant DNA extraction kit (Biotools, Madrid, Spain) and eluted in 50 μl in Tris-EDTA 0.1× buffer. Working dilution for PCR amplification was 1:10 of the eluted DNA solution. Eight dinucleotide microsatellite loci were amplified following Segarra-Moragues et al. (2009). PCR products were analyzed on an ABI3730 automated sequencer (Applied Biosystems, Madid, Spain) using LIZ500 as internal lane size standard and fragments were assigned to allele classes using Genemarker version 1.80 software (Softgenetics, State College, PA).


Allele frequencies, mean number of alleles per locus (A), and observed (HO) and unbiased expected (HE) heterozygosities (Nei 1978) were calculated for each population using GENETIX version 4.05 (Belkhir et al. 1996–2004). Wright's F-statistics were estimated according to Weir and Cockerham (1984) using GENEPOP′007 (Rousset 2008) and tested for significance by Fisher's exact tests. This software was also used to check for departures from Hardy–Weinberg equilibrium at each locus and for genotypic linkage disequilibrium between pairs of loci within each population using Fisher's exact tests. Because multiple tests were carried out, sequential Bonferroni corrections were used to adjust P-values. To assess whether population diversity indices differed between seeder and resprouter populations, average allelic richness per locus (A*), applying the rarefaction method of Hurlbert (1971) adapted by El Mousadik and Petit (1996), HO, expected heterozygosity (HS) and inbreeding coefficient (FIS) within populations were compared between seeder and resprouter groups of populations using FSTAT version (Goudet 2001) and tested for significance using 10,000 permutations. Pairwise population differentiation estimates (FST) were also calculated by FSTAT for all pairs of populations and average values between seeder and resprouter groups of populations were compared in the same manner.

Population genetic structure was investigated through a Bayesian clustering method implemented in STRUCTURE version 2.1 (Pritchard et al. 2000; Pritchard 2002). The program allows the user to find the optimal number of genetic clusters (K), and assigns individuals to the different clusters based on allele frequencies at each locus. Our analyses were based on an admixture ancestral model with correlated allele frequencies (because of high FISs in some populations; see Results section), for a range of K values from 2 to the number of populations considered plus 2 (i.e., 24). In doing so, the proportion of membership of each individual and population to the inferred K clusters were calculated. We used a burn-in period and a run length of the Monte Carlo Markov Chain (MCMC) of 1 × 105 and 1 × 106 iterations, respectively. Ten runs were carried out for each K to quantify the amount of variation of the likelihood. The number of K present in the dataset was evaluated according to Evanno et al. (2005). This method uses an ad hoc parameter (ΔK) to estimate the rate of change of likelihood values between successive K values.

Analyses of molecular variance (AMOVAs; Excoffier et al. 1992) were performed to partition the total variance into variance components using ARLEQUIN version 3.11 (Excoffier et al. 2005). These analyses were conducted for E. coccinea s.l., and at different hierarchical levels. First, according to their regeneration mode: between regeneration modes (i.e., seeder vs. resprouter), among populations within regeneration mode, and within populations. Second, according to the K optimal genetic clusters detected with STRUCTURE: between genetic clusters, among populations within genetic clusters, and within populations. In all instances, the significance of the variance components was obtained using 1000 permutations.

To explore the contributions of mutation and genetic drift to differentiation between populations within each regeneration mode (seeder and resprouter), and between seeder and resprouter groups of populations, we used SPAGeDi version 1.2g (Hardy and Vekemans 2002) to compare FST and RST statistics computed on the same data (Hardy et al. 2003). Although FST is based on allele identity (Wright 1951), RST also considers allele size information (Slatkin 1995). SPAGeDi computes the statistic pRST by using a randomization procedure in which allele sizes at a locus are randomly permuted, while the allele identity information is kept intact. Because allele identity, but not allele size, is taken into account to estimate pRST, mean pRST values are equal to expected FST ones (for further details, see Hardy et al. 2003). Therefore, if RST is significantly larger than pRST (i.e., FST) it can be established that genetic differentiation is the outcome of a stepwise-like mutation process rather than a consequence of drift. Conversely, if both RST and FST share equal expectations, it may be concluded that differentiation is mainly caused by genetic drift (Hardy et al. 2003). To conduct these analyses, microsatellite allele sizes were converted to number of repeats, as recommended in the SPAGeDi manual, considering the number of repeats and sequence size of the sequenced clones reported in Segarra-Moragues et al. (2009).

Pairwise genetic distances between populations were calculated using Nei et al's. (1983)DA, that assumes the infinite alleles model (IAM, Kimura and Crow 1964), as implemented in POPULATIONS version 1.2.3 beta (Langella 2000). The statistical robustness of the groupings was assessed by bootstrap analysis over populations with 1000 replicates (Felsenstein 1985). Further genetic indices assuming the stepwise mutation model (SMM, Kimura and Ohta 1978) were also assayed to have a prudent range of evolutionary assumptions covered (Slatkin 1995). Nonetheless, they were later discarded following the SPAGeDi results (see below). The resulting distance matrix was used to construct neighbor-joining (NJ) midpoint rooted trees using MEGA 4.0 software (Tamura et al. 2007) and to compute eigenvectors and eigenvalues to perform principal coordinates analyses (PCO) using NTSYSpc version 2.11a (Rohlf 2002). A minimum spanning tree was also constructed with NTSYSpc and superimposed on the two-dimensional PCO.

Finally, isolation by distance was assessed separately for seeder and resprouter groups of populations by matrix correlation analyses between a matrix of pairwise linearized FST values [i.e., FST/(1 −FST); Slatkin 1995] computed with ARLEQUIN and a matrix of log-transformed pairwise geographical distances between populations. Log transformation of pairwise geographical distances assumes a two-dimensional stepping-stone model of population structure (Rousset 1997). Significance of the correlation was tested for each group with Mantel tests (1000 permutations) using NTSYSpc. Additionally, to search for possible differences between seeder and resprouter groups in the increase of genetic differentiation across geographical distance, the intercepts and slopes of fitted line regressions to the data were compared by means of standard analyses of covariance (ANCOVA).



All eight microsatellite loci were polymorphic in all 22 analyzed populations. Number of alleles amplified ranged from 12 (loci Ecoc117 and Ecoc115) to 44 (locus Ecoc446) with a mean of 21.63 ± 9.85 (± SD) alleles per locus (see Appendix S1). Mean number of alleles per locus ranged from 6.75 ± 2.49 in population R03 to 12.25 ± 5.77 in population S08 (Table 1). Total number of different SSR alleles scored was 173, 120 of which (69.3%) were shared by seeders and resprouters, whereas 47 (27.2%) were exclusive to seeders and only six (3.5%) to resprouters. Consequently, seeders had a moderately higher average number of alleles per locus (6.44 ± 0.74) than resprouters (5.61 ± 0.73), these differences being statistically significant (Table 2). Observed heterozygosities ranged from 0.577 ± 0.203 (R03) to 0.770 ± 0.309 (R04), and expected heterozygosities from 0.636 ± 0.152 (R03) to 0.800 ± 0.060 (S03) (see Table 1). Seeder and resprouter populations showed overall similar observed heterozygosity values and slightly higher expected heterozygosity values, although differences between the two groups were not statistically significant (Table 2). Eighteen of the 22 populations showed HW deviations toward heterozygote deficiency (i.e., positive mean FIS values). However, these deviations were only significant in 12 populations (Table 1) and were not consistent across loci or populations. This is suggestive of heterozygote deficiency being an artifact due to the size of populations and the low allele frequency for the highly variable loci and populations, rather than a reflection of inbreeding. Accordingly, seeders and resprouters showed overall FIS values close to zero, and differences between both groups were statistically negligible (Table 2).

Table 2.  Comparison of mean genetic polymorphism values between the seeder and resprouter populations of Erica coccinea.
  1. 1A=allelic richness calculated after the rarefaction method of El Mousadik and Petit (1996) and based on minimum sample size of 10 individuals (corresponding to the smallest R05s population).

  2. 2HO, HS, average observed and expected heterozygosity within populations, respectively.

  3. Significant values based on 10,000 permutations are indicated in bold. Seeders, N=11 populations; resprouters, N=11 populations.



Moderate to high levels of population differentiation were observed among populations, with all pairwise FST values being significantly different from zero (P < 0.001) and ranging from a minimum of FST= 0.026 (pair R07–R08) to a maximum of FST= 0.287 (pair S01–S10; data not shown). Average pairwise FST (mean, 95%CI) was almost twice as high between seeder populations (FST= 0.152, 0.095–0.221) than between resprouter ones (FST= 0.082, 0.060–0.106), this difference being statistically significant (Table 2).

The optimal number of K genetic clusters present in the data was difficult to estimate following the procedure described in STRUCTURE software (Pritchard et al. 2000) because likelihood values did not show a clear maximum, but increased with K until a plateau was reached at approximately K= 12 (Fig. 3A). The method of Evanno et al. (2005) showed a maximum modal value of ΔK= 528.76 for K= 2 (Fig. 3B). This K= 2 clustering separated fairly clearly seeder and resprouter populations with high proportions of membership of each population to cluster 1 (composed of resprouters) or to cluster 2 (composed of seeders), with the exception of seeder population S01, which showed a higher proportion of membership to cluster 1 (Fig. 4).

Figure 3.

(A) Log-likelihood of the eight microsatellite loci data for 22 E. coccinea populations given K clusters, obtained through 10 runs of the STRUCTURE analysis. (B) Corresponding ΔK estimation according to Evanno et al. (2005) showing maximum peaks of ΔK values at K= 2 and K= 4, indicating that those are the optimal solutions for K given the data.

Figure 4.

Bayesian analyses of genetic structure of 539 individuals from 22 populations of Erica coccinea. (A) Mean proportion membership of each population to the predefined, K= 2 and K= 4 clusters with the highest ΔK values obtained following Evanno et al. (2005). (B) Proportion of membership of individuals for K= 2 and K= 4 predefined clusters. Populations 1–11, seeders, 12–22, resprouters.

A further maximum modal value of ΔK= 146.52 was obtained for K= 4 (Fig. 3B). In this clustering most of the seeder populations were grouped into two exclusive clusters (cluster 1 and 2) with high proportion of membership, whereas most of the resprouter populations grouped into another exclusive cluster (cluster 3) with similar high proportions of membership. Some other resprouter populations grouped into a last cluster (cluster 4) that also included one seeder population (S01) with high proportion of membership (Fig. 4).

AMOVA analyses conducted for E. coccinea s.l. attributed 13.46% of the total variation between populations. Hierarchical AMOVA with populations arranged in two groups according to their morphological membership (i.e., regeneration modes: seeders and resprouters) revealed that only 3.23% of the total variation was attributable to differences between groups, whereas 11.56% of the variation was found among populations within groups and 85.21% within populations (Table 3). Two further hierarchical AMOVAs were conducted with populations grouped in two (K= 2) and four (K= 4) genetic clusters retrieved from STRUCTURE. For K= 2, the analysis resulted in 4.01% of the variance distributed between clusters 1 and 2 and 11.11% of the variance among populations within genetic clusters. The highest proportion of variance among groups (5.20%) was obtained in the hierarchical AMOVA with populations grouped according to four genetic clusters (K= 4). This analysis also revealed the lowest genetic variation among populations within each group (9.40%; see Table 3).

Table 3.  Analyses of molecular variance (AMOVA) of Erica coccinea populations.
Source ofSum of squareddfVariancePercentage of the
variation (groups)deviations (SSD) componentstotal variance
1. Erica coccinea s.l.
 Among populations 517.466  210.4457813.46
 Within populations3026.21810562.8657486.54
2. Morphological membership: seeders (S01–S10, R05s) versus resprouters (R01–R11)
 Among groups  81.627   10.10862 3.23
 Among populations within groups 435.839  200.3887511.56
 Within populations3026.21810562.8657485.21
3. Genetic membership: Two clusters of STRUCTURE analysis; cluster 1 (S01, R01–R11) versus cluster 2 (S02–S10, R05s)
 Among clusters  94.954   10.13541 4.01
 Among populations within clusters 422.512  200.3751211.11
 Within populations3026.21810562.8657484.88
4. Genetic membership: four clusters of STRUCTURE analysis; cluster 1 (S02–S05, S07–S09, R05s) versus cluster 2 (S06, S10) versus cluster 3 (R04–R05r, R07–R11) versus cluster 4 (S01, R01–R03, R06)
 Among clusters 191.755   30.17457 5.20
 Among populations within clusters 325.711  180.31536 9.40
 Within populations3026.21810562.8657485.40

Mean pairwise multilocus estimates of differentiation (95%CI in brackets) were FST= 0.152 (0.095–0.221), RST= 0.137 (0.063–0.154) between pairs of seeder populations and FST= 0.082 (0.060–0.106), RST= 0.068 (0.034–0.101), between resprouter ones, whereas differentiation between seeder and resprouter populations were of FST= 0.135 (0.095–0.182), RST= 0.115 (0.067–0.145) (see Appendix S2). All the observed RST values between seeder and resprouter populations were within the 95% confidence interval (CI) of pRST. For within-group pairwise comparisons, only one locus within seeder (Ecoc446) and two loci within resprouter (Ecoc142 and Ecoc431) showed significantly higher RST values than pRST. These results show evidence of a preponderant role of genetic drift in population differentiation in E. coccinea. They also suggest that the microsatellite mutation process does not follow an SMM and that, subsequently, FST is a better estimator of population differentiation than RST in this species.

Genetic distances based on the IAM clustered populations more congruently with geographical distribution and morphological types than those based on the SMM (results not shown). This agreed with the little relevance of the SMM in the mutation process reported by the SPAGeDi analysis (see above). Therefore, only the results from Nei et al. (1983)DA genetic distance will be commented on further. In the NJ tree constructed with this distance method, resprouter populations formed a somewhat homogeneous group, with populations showing shorter genetic distances between them than seeder populations (Fig. 5). Seeder populations, by contrast, grouped into different clusters, and two of them (S01 and R05s) were embedded into the cluster of resprouters. Genetic distances between seeder populations were larger than those between resprouter populations. Even the two seeder populations that were genetically close to resprouter populations (S01 and R05s) showed larger distance to their respective pairs than any of the other resprouter pairs (Fig. 5). The results of this NJ cluster analysis support the effect of exclusive alleles in increasing the genetic distance of seeder populations. Similar results were obtained in the PCO analysis (results not shown).

Figure 5.

Neighbor-joining midpoint rooted tree based on DA genetic distance (Nei et al. 1983) showing the relationships among seeder (white dots) and resprouter (black dots) populations. Bootstrap values obtained from 1000 permutations over populations are shown above branches of the NJ tree when higher than 40%.

A significant correlation between pairwise geographical distances and linearized FST values was found in both seeder (r= 0.43, P= 0.017) and resprouter (r= 0.45, P < 0.001) populations (Fig. 6), indicating that both morphological groups showed significant isolation by distance. The intercept of the regression of the seeder group (assuming standard line regression of the data) was significantly higher than that of the resprouter (P < 0.0001; ANCOVA), whereas no significant differences were detected between the slopes of both line regressions (P= 0.145; ANCOVA). This can be interpreted as a reflection of higher genetic distance values for seeder populations throughout the geographical range but similar patterns of increase in genetic differentiation with geographical distance in both seeder and resprouter groups of populations. However, despite the similarity in the slopes between both groups, values were more scattered for seeder populations throughout the geographical range (Fig. 6). This degree of scatter, measured as the average of the absolute residuals from a standard line regression, was in fact twice as large in seeder (0.053 ± 0.041) than in resprouter populations (0.021 ± 0.018), being these differences highly significant (P < 0.0001; Mann–Whitney U test).

Figure 6.

Isolation by distance analysis. Correlation between log-transformed pairwise geographical (x-axis) and linearized FST(Slatkin 1995) pairwise values (y-axis) of seeder (white dots) and resprouter (black dots) populations. Correlation between matrices was r= 0.436, P= 0.014, and r= 0.445, P= 0.001 for seeder and resprouter populations, respectively; P values reported after 1000 random permutations Mantel tests.



The role of population dynamics may be crucial to explaining levels of genetic differentiation in wild populations (Wade and McCauley 1988; Pannell and Charlesworth 1999; Vitalis et al. 2004). In this sense, pairwise genetic differentiation levels in E. coccinea were twice as high in seeder populations than in resprouter ones. Average genetic variance among seeder populations was even higher than across seeder and resprouter groups of populations, which accounted for only slightly more than 3% of the total variation (see Table 3). This challenges the possibility of seeder and resprouter groups of E. coccinea actually being two different cryptic taxa. It also reinforces the role of postfire regeneration modes in explaining genetic divergence among populations in this species, rather than possible selective factors associated with the apparent geographic partitioning of seeder and resprouter populations (Ojeda et al. 2005; see Fig. 2). However, with the data presented in this study, we cannot rule out this alternative explanation.

The above-mentioned low percentage of genetic variation between regeneration modes assigned by the AMOVA (Table 3) seems at odds with the nearly accurate distinction between seeder and resprouter populations provided by the STRUCTURE analysis (see Fig. 4). However, it shall be stressed that although the AMOVA, as well as the NJ analysis are distance-based approaches, STRUCTURE is a coalescent clustering method based on inferred allele frequencies between populations. Because both seeder and resprouter seem to be conspecific, no distinct genetic backgrounds, but just slight differences in frequencies of the most common alleles are expected, which STRUCTURE is particularly well suited to identify.

Genetic diversity in natural populations is shaped by a balanced effect of mutation and migration (gene flow), which generate within-population variation, and natural selection and genetic drift, which erode it and promote differentiation among populations (Hutchison and Templeton 1999; Hartl and Clark 2007). High levels of genetic differentiation among populations are frequently related to changes in allele frequencies and overall declines in within-population genetic variation, mainly caused by drift (Loveless and Hamrick 1984; Charlesworth 2003). Interestingly, our results revealed that not only genetic differentiation among populations, but also variation within populations (as measured by the average allelic richness per locus) was higher in seeder populations (see Table 2). These higher genetic diversity values within seeder populations might be interpreted as the outcome of presumably higher mutation rates in seeder populations owing to their shorter generation times (Smith and Donoghue 2008; but see Verdú et al. 2007), which would subsequently accelerate the generation of new alleles in populations.

Alternatively, higher gene flow rates (via seeds and/or pollen) in seeder populations might be invoked to explain their higher levels of (within-population) genetic diversity, assuming similar levels of total diversity (HT) between the seeder and resprouter group of populations. As a matter of fact, pairwise geographical distances between seeder populations (78.7 ± 47.2 km; mean ± SD) were slightly but significantly shorter than those between resprouter (110.6 ± 65.6 Km; P < 0.01; Mann–Whitney U test; see Fig. 2). However, our analyses revealed significant patterns of isolation by distance in both seeder and resprouter groups of populations, with similar, positive correlation values between pairwise geographical and genetic distances in the two groups (Fig. 6). This is not surprising because both seeder and resprouter plants share equal reproductive traits. They both produce small, oval seeds (F. Ojeda, pers. obs.) that are dispersed by ants or passively by wind. These short range means of dispersal (wind dispersal distances of Erica seeds may not exceed 100 m; Bullock and Clarke 2000) would not overcome genetic patterns of isolation by distance. Regarding pollen, both seeder and resprouter plants exhibit showy flowers with a distinct ornithophylous syndrome (Rebelo and Siegfried 1985) and are actively and apparently equally visited by Nectariniidae (sunbirds) across their geographical range (F. Ojeda, pers. obs.). In any case, should the higher genetic diversity in seeder populations be attributed to a putatively higher gene flow across populations, it would, in turn, cause differentiation among populations to be lower (Zhivotovsky 2001; Hartl and Clark 2007) in the seeder than in the resprouter group, while the opposite has been found in this study. Besides, total diversity values were slightly higher in the seeder (HT= 0.851) than the in resprouter (HT= 0.760) group of populations.

These somewhat unexpected results (i.e., both higher within-population genetic variation and higher among-population divergence in the seeder group) may indeed be reconciled by considering the contrasting population dynamics determined by the seeder and resprouter regeneration modes (Bond and van Wilgen 1996; Ojeda et al. 2005). Each fire cycle means a complete generation turnover in seeder populations, whereas generation times of resprouter are longer (several to many fire episodes). Hence, faster rates of new alleles arising in the populations are to be expected in seeder populations of E. coccinea as a consequence of their shorter generation times. At the same time, faster population turnover rates in seeder populations may accelerate genetic drift (Wade and McCauley 1988; Pannell and Charlesworth 1999), causing allele loss and subsequent changes in allele frequencies across populations. The wider scattered plot of the genetic versus geographical distances in the seeder group shown in Figure 6 (see also Results section) may be indicative of stronger genetic drift in seeder populations (e.g., see Hutchison and Templeton 1999; Johansson et al. 2006; Hamilton and Eckert 2007) compared to resprouter ones, despite the overall shorter pairwise geographical distances in the seeder group (see above). This random genetic erosion would somewhat counterbalance the above-stated allele gain boosted by the shorter generation times and would thereby enhance genetic differentiation among seeder populations. Hence, the contrasting population dynamics of seeder and resprouter populations of E. coccinea may satisfactorily account for both the higher within-population genetic diversity and the more marked levels of differentiation among seeder populations found in the present study.

The lower levels of differentiation among resprouter populations were clearly evidenced by the Bayesian analysis of genetic structure, where resprouter populations diversified into fewer genetic clusters (Fig. 4). Analogously, the NJ tree analysis clustered all resprouter populations together (Fig. 5). By contrast, the higher diversification among seeder populations is evidenced in these analyses (Figs. 4 and 5), as a consequence of their higher proportion of exclusive alleles and higher levels of genetic differentiation.


Genetic differentiation among populations, considered the initial step to speciation (Avise 2000; Coyne and Orr 2004), is frequently associated with marked reductions in variation within populations as a consequence of genetic drift (Loveless and Hamrick 1984; Young et al. 1996; Charlesworth 2003). This genetic erosion of individual populations may compromise their long-term persistence by increasing their risk of local extinction (van Treuren et al. 1991; Newman and Pilson 1997), and would ultimately undermine possible speciation. Notwithstanding, here we have suggested that genetic erosion in differentiation-prone seeder populations would be counterbalanced by the allele gain boosted by their shorter generation times, thereby sustaining or even increasing genetic divergence (Figs. 4 and 5).

However, despite the apparent strength of seeder populations against genetic erosion, they are still highly prone to extinction, particularly under harsh and highly variable environmental conditions that jeopardize postfire recruitment (e.g., severe summer drought and winter rainfall unreliability; Ojeda et al. 2005). Due to their nonoverlapping generation dynamics, seeder populations lack reproductive storage potential over generations (Warner and Chesson 1985; Higgins et al. 2000) and, unlike resprouter, they are extremely vulnerable even to a single postfire recruitment failure (Bond and Midgley 2001; Ojeda et al. 2005). In fact, the apparent north-south geographical partitioning of resprouter and seeder populations of E. coccinea in the CFR fynbos may be rather explained by the higher extinction risk of seeder populations under climatic conditions different from those found in coastal mountains of the south-western CFR (Ojeda 1998; Ojeda et al. 2005).

High speciation rates certainly require the existence of low extinction risks (Dynesius and Jansson 2000; Mittelbach et al. 2007). Consequently, higher genetic divergence of seeder populations might only translate into higher speciation when the risk of (local) extinction is reduced. In this regard, it must be again emphasized that both species diversity and endemism in Erica reach highest levels in coastal mountains of the south-western CFR (Oliver et al. 1983; Ojeda 1998) and are tightly associated to the seeder life form (Ojeda 1998; Cowling and Lombard 2002). Particular climatic conditions in this region reduce extinction rates of seeder populations (Ojeda et al. 2005), thereby fostering speciation in differentiation-prone seeder populations. Thus, we believe that the results presented in this study contribute significantly to our understanding of the high biodiversity levels of the genus Erica in the South-African CFR fynbos.

Associate Editor: M. Burd


We thank A. García and R. Marriner for helping with field sampling. SANPARKS and Cape Nature authorities provided permits for collecting plant material. We also thank managers of Vogelgat (Hermanus) nature reserve, and landowners of Brandfontein (Agulhas), Camphill (Hermanus) and Honingklip (Bot Rivier) farms for letting us collect samples in their private properties. FO is deeply indebted to F. Conrad and K. Roux (KRC, SANBI, South Africa) for hosting and facilitating deliverables for plant collection. S. Donat-Caerols for laboratory assistance. P. Linder, X. Picó, M. Verdú, R. Cowling, J. Pannell, M. Burd, and two anonymous referess provided valuable comments to previous versions of the manuscript. Fieldwork was financed by travel grants to FO from the University of Cádiz (Plan Propio UCA de Investigación) and the Junta de Andalucía (subvención de Incentivos de Actividades Científicas 2/2007). Financial support for laboratory work has been provided by project VAMPIRO (CGL2008-05289-C02-01/BOS; Spanish Ministerio de Ciencia e Innovación), project P07-RNM-02869 (Junta de Andalucía, Spain) and project ACOMP09/073 (Generalitat Valenciana, Spain). JGS-M was supported first by an ARAID (Agencia Aragonesa para la Investigación y el Desarrollo, Spain) postdoctoral contract and next by a “Ramón y Cajal” (MICINN-RYC, Spain) postdoctoral contract.