Conservation Genetics

, Volume 13, Issue 4, pp 899–911

Variable rates of random genetic drift in protected populations of English yew: implications for gene pool conservation

  • Igor J. Chybicki
  • Andrzej Oleksa
  • Katarzyna Kowalkowska
Open Access
Research Article

DOI: 10.1007/s10592-012-0339-9

Cite this article as:
Chybicki, I.J., Oleksa, A. & Kowalkowska, K. Conserv Genet (2012) 13: 899. doi:10.1007/s10592-012-0339-9

Abstract

Protecting populations in their natural habitat allows for the maintenance of naturally evolved adaptations and ecological relationships. However, the conservation of genetic resources often requires complementary practices like gene banks, translocations or reintroductions. In order to minimize inbreeding depression and maximize the adaptive potential of future populations, populations chosen for ex situ conservation should be selected according to criteria that will result in a reduction of global coancestry in the population. Generally, large populations should reveal lower coancestry and higher genetic variation than small populations. If detailed knowledge about coancestry is lacking, census population number (Nc) can be used as a proxy for required characteristics. However, a simple measure of Nc may be misleading in particular cases as genetic processes rely on effective population size (Ne) rather than Nc and these two measures may differ substantially due to demographic processes. We used an example of English yew to address whether Nc can be a good predictor of genetic parameters when used in conservation programs. Using microsatellite markers, we estimated allelic richness, inbreeding and coancestry coefficients of six relatively large yew populations in Poland. Each population was characterized by Ne using the linkage disequilibrium method. Our results showed that populations of English yew were subject to substantial divergence and genetic drift, with both being inversely proportional to the effective subpopulation size (Ne). Additionally, allelic richness appeared proportional to Ne but not to Nc. However, the Ne/N ratio differed greatly among populations, which was possibly due to different population histories. From the results we concluded that choosing source populations based only on their census size can be fairly misleading. Implications for conservation are briefly discussed.

Keywords

Population size Allelic richness Inbreeding Ex situ conservation Taxus baccata 

Introduction

The conservation of endangered and threatened plant species is focused on in situ practices. Protecting populations in their natural habitat allows the maintenance of natural ecological and evolutionary processes, as well as evolved adaptations and ecological relationships. However, conservation of genetic resources usually requires additional forms of conservation, so called ex situ practices such as gene banks, translocations or reintroductions (Pierson et al. 2007). If possible, choosing populations for ex situ conservation should follow some objective criterions which include genetic parameters such as inbreeding coefficients and levels of genetic polymorphism (Mistretta 1994). High genetic variation allows a population to better respond to unpredictable environmental changes in the future (Frankham et al. 2002; Willi et al. 2006). In addition, inbreeding depression can be avoided by maintaining low levels of inbreeding (Charlesworth and Willis 2009 and the literature therein). The generally acknowledged strategy for ex situ conservation is to minimize the global coancestry in the population (Montgomery et al. 1997; Caballero and Toro 2000). This requires kinship structures among individuals to be known in advance before their contribution to the next generation can be predicted (Saura et al. 2008). If the kinship structure is unknown, census population size may be the only proxy available to determine required characteristics. Generally, small populations should reveal higher coancestry and lower genetic variation than large populations. However, as genetic processes rely on effective (Ne) rather than census (Nc) population size and as these two measures may differ substantially due to demographic processes (e.g. biased reproductive success or sex ratios) (Frankham 1995), a simple measure of Nc may be misleading in particular cases.

English yew (Taxus baccata L) serves as a good example of a species that needs both in situ and ex situ conservation. The present-day occupancy of this understory species is limited mainly to small and highly discontinuous remnants (Thomas and Polwart 2003). At a metapopulation scale English yew exhibits typically substantial levels of genetic differentiation (Hilfiker et al. 2004; Myking et al. 2009; Zarek 2009; Dubreuil et al. 2010; González-Martínez et al. 2010) driven by low population size and restricted gene flow (Chybicki et al. 2011). Due to similar reasons, increased inbreeding and kinship structure can build-up within yew populations (Lewandowski et al. 1995; Myking et al. 2009; Dubreuil et al. 2010; Chybicki et al. 2011).

English yew is slow growing, dioecious, late maturating conifer. It produces airborne pollen and fleshly fruits that are often dispersed by birds (Thomas and Polwart 2003). Although high shade tolerance allows young yew recruits to outcompete other, less shade-tolerant plants (Iszkuło and Boratyński 2006), it is known to regenerate poorly within a site. This is primarily attributed to grazing and damage by fungi or insect pests (Hulme 1996; Svenning and Magård 1999; Mysterud and Østbye 2004; Thomas and Polwart 2003). Poor regeneration results in ageing of natural populations that puts them under risk of extinction. Therefore, long-term in situ conservation of the natural genetic resources may fail in the case of English yew, at least in some populations. For example, the largest Polish reserve “Cisy Staropolskie” in Wierzchlas (see Materials and methods) has been exhibiting continuing decline over the last 100 years, with less than 3,000 individuals remaining from an initial population of roughly 5,500 individuals. Given the late maturity of this species (ca. 70–120 years; Thomas and Polwart 2003), the rate of reduction of population size amounts to approximately 35 % per generation. To curb such a rapid decline, ex situ conservation employing gene bank collections and restitution programs is required. Restitution of English yew is especially important as it can play a crucial role in the conservation of genetic resources of the species. Firstly, restituted populations could constitute a natural long-term repository of genetic variation. Secondly, a net of reintroduced populations could form natural corridors for gene exchange between fragmented populations. As a consequence, the destructive process of genetic drift occurring within small yew fragments could potentially be counterbalanced by increased gene immigration.

The long-term goals of ex situ programs will be achieved given appropriately chosen source populations. Nevertheless, in many cases census population numbers are the only information available prior to administrative decisions. Because knowledge about genetic structure and demography of English yew is only starting to accumulate (Lewandowski et al. 1995; Hilfiker et al. 2004; Dubreuil et al. 2010; González-Martínez et al. 2010; Chybicki et al. 2011), it remains unclear whether Nc reflects Ne and other genetic structure parameters enough to make a decision about the usefulness of particular populations for ex-situ conservation. Our study aimed to address this issue using SSR data collected for six populations of English yew in Poland. Specific goals included the assessment of levels of genetic polymorphism, inbreeding and coancestry, as well as Ne. We also tested if a bottleneck had occurred in the recent population history. Finally, we verified whether census population size correlates with genetically derived parameters.

Materials and methods

Study sites and sampling

Studies were conducted within six selected populations of English yew (Fig. 1). The criterion of choice was the potential utility of the populations in ex situ conservation programs. Therefore, the study sites represented the largest yew populations in a region. The selected sites could be considered virtually independent remnants of the ancestral population as they are separated by large distances (ca. 260 km on average). All the sites are located within nature reserves established to protect yew populations. Except for Cisy Staropolskie, the study populations are characterized by a low density (as for a temperate tree species), reaching about 20 individuals per hectare (Table 1). Cisy Staropolskie is a unique site because of its large size and relatively high density. Nonetheless, it can be considered as a declining population as there is an absence of natural regeneration (absolute mortality among seedlings within two successive seasons) as well as a continuous reduction in population number. On the contrary, successful regeneration is particularly evident in Cisy nad Czerską Strugą as well as Cisy w Czarnem, with the latter seeming the most stable out of all six populations. It should be mentioned that little is known about the origin of the study sites. In consequence, we a priori assumed that all the populations are natural remnants of the larger population, with a possibility of bottleneck in the recent history. Nonetheless, other scenarios (incl. artificial origin) cannot be definitely excluded.
Fig. 1

Map of location of the study populations of English yew in Poland (populations names coded as in Table 1)

Table 1

Summary characteristics of the study populations

Population

Code

Longitude, latitude

Area (ha)

Population sizea

Sex ratio (% females)

Sample size

Cisowa Góra

CG

50°32′N, 16°42′E

18.6

708

53.1b

60

Cisy nad Czerską Strugą

CS

53°45′N, 17°55′E

17.2

299

57.4a

60

Cisy nad Liswartą

LI

50°46′N, 18°47′E

21.2

466

54.9a

60

Cisy Staropolskie

ST

53°30′N, 18°7′E

18.5

2,856

33.1b

293c

Cisy w Czarnem

CZ

53°44′N, 16°58′E

26.3

437

46.4a

216c

Radomice

RA

50°44′N, 20°39′E

27.2

543

No data

60

aBased on data gathered by Kostrzyca Forest Gene Bank (available at http://www.lbg.jgora.pl/dzialalnosc/inw_cisa.htm)

bData from Iszkuło et al. 2009

cThe same sample of individuals as in Chybicki et al. 2011

For the analysis of DNA polymorphism, 10 cm long twigs were sampled from selected individuals at each site. In order to reflect stabilized characteristics of genetic structure, the oldest possible individuals were sampled at a particular site.

Molecular methods

A total genomic DNA was extracted using a slightly modified CTAB protocol (Doyle and Doyle 1990). For genetic typing five nuclear SSR loci were analyzed: Tax26, Tax31, Tax36 (Dubreuil et al. 2008), TS09 (Huang et al. 2007) and Tax362 (Chybicki et al. 2011) according to the protocol described elsewhere (Chybicki et al. 2011).

Statistical methods

Genetic structure

To describe genetic structure, standard per-locus genetic structure parameters were computed (observed and effective number of alleles, expected and observed heterozygosity). Because sample sizes differed among populations (Table 1), which affected levels of detected polymorphism, an added measure of allelic richness was computed using the rarefaction method (Petit et al. 1998). To study concordance with Hardy–Weinberg proportions, the MCMC approximation of the exact test was employed as implemented in GENEPOP 4.0 software (Rousset 2008). To test the overall homogeneity of genetic structure parameters across populations (a population as a factor) we used the Friedman rank test (Sheskin 2000), a nonparametric analog of a repeated measure ANOVA.

To assess the contribution of the study populations to the within- and between-population allelic richness, we used the rarefaction-based method of partitioning of allelic diversity introduced recently by Caballero and Rodríguez-Ramilo (2010). Decomposition of allelic richness into within- (AS) and between-population (DB) measures allows the computation, by analogy to FST, of a global allelic diversity AST = DB/AT, where AT = AS + DB. The relative contribution of a given population to a global within- and between-population allelic diversity was estimated by disregarding this population in the re-estimation of AS, DB and AT parameters (Petit et al. 1998; Caballero and Rodríguez-Ramilo 2010).

Inferences about inbreeding levels were conducted using the method (IIM) implemented in INEst software (Chybicki and Burczyk 2009), which allows the calculation of unbiased estimates for a multilocus average inbreeding coefficient (FIS) in the presence of null alleles (pn) (that are typically present in SSR markers used in this study; Dubreuil et al. 2008; González-Martínez et al. 2010; Chybicki et al. 2011).

Each population was also characterized by the coancestry coefficient (θ), which corresponds to the probability that two random genes in a subpopulation are identical by descent due to shared ancestry within a subpopulation (Reynolds et al. 1983). In this study θ was estimated using a model introduced firstly by Balding and Nichols (1995) and then adapted to various applications (e.g. Falush et al. 2003; Beaumont and Balding 2004). The model used here, which follows that in Beaumont and Balding (2004), allows subpopulations to diverge from a common ancestral population with different rates of genetic drift. The rates of genetic drift are measured with the subpopulation-specific parameters that can be interpreted as the coancestry coefficients θ (Balding 2003). The estimation procedure was based on the multinomal-Dirichlet likelihood for the allele counts, which for the i-th subpopulation at the j-th locus equals.
$$ L(a_{ij1} ,\ldots,a_{{ijK_{j} }} |p_{j1} ,\ldots,p_{{jK_{j} }} ,\theta_{i} ) = \frac{{(\sum\limits_{k} {a_{ijk} } )!\Upgamma (1 /\theta_{i} - 1)}}{{\Upgamma (1/\theta_{i} - 1 + \sum\limits_{k} {a_{ijk} } )}}\prod\limits_{k} {\frac{{\Upgamma (a_{ijk} + (1/\theta_{i} - 1)p_{jk} )}}{{a_{ijk} !\Upgamma ((1/\theta_{i} - 1)p_{jk} )}}} $$
(1)
where aijk denotes the count of the k-th allele at the j-th locus in the i-th population, pjk denotes the frequency of the k-th allele at the j-th locus in the ancestral population, θi denotes the coancestry coefficient for the i-th population (Balding 2003) and Γ() is the standard gamma function (Abramowitz and Stegun 1965). To estimate θi we used Metropolis–Hastings algorithm, choosing a uniform (0,1) distribution as a prior for {pjk}, while a truncated exponential distribution X ∈ (0,1) with a mean μ as a prior for {θi}. We subsequently set μ = 0.01, 0.1, and 0.25. Among these values μ = 0.1 corresponds roughly to the average FST reported for Taxus species (e.g. Senneville et al. 2001; Myking et al. 2009; Mohapatra et al. 2009; González-Martínez et al. 2010), while two remaining values may represent two possible extremes for the species. The algorithm (described in Appendix) used to estimate θi was implemented in the Pascal/Delphi computer program BayeF (available from IJC).

It is worth noting that the across-populations average coancestry coefficient estimated as described above is closely related to a global FST parameter (Holsinger 1999), whereas under the scenario of completely isolated subpopulations, each population-specific θ has a useful interpretation in terms of population-scaled divergence time (t) through −ln(1 − θ) ≈ t /2Ne (Reynolds et al. 1983). Although the study populations were assumed to be highly isolated, to verify whether gene flow occured among the study populations we used a Bayesian clustering method. Generally, such methods provide estimates for the number of distinct populations and their genetic homogeneity (through the identification of immigrants) based genetic structure alone. In this tudy we used an algorithm implemented in GENELAND ver. 3.2.4 (Guillot et al. 2008). This method was chosen because, unlike similar methods, it treats a number of clusters (K) as an estimable parameter, enabling straightforward hypothesis testing. The estimation procedure was based on the F-model (assuming dependency of allele frequencies among sub-populations). Because markers are affected by null alleles, the Null Allele model was used. Estimates were obtained after 100,000 iterations (saving every 50th).

Effective population size

To test whether a bottleneck had occurred in the recent population history, BOTTLENECK software (Cornuet and Luikart 1996) was used. Although the step-wise mutation model (SMM) seems theoretically the most relevant for microsatellites, recent studies suggested that the exact SMM may be not optimal in the case of Taxus baccata (González-Martínez et al. 2010). Therefore, the bottleneck test was performed under the assumption of the infinite allele model (IAM), the step-wise mutation model and the intermediate model (30 % multiple-steps mutations and 70 % single-step mutations), so-called the two-phase mutation model (TPM).

Rates of genetic drift were scaled with effective population size (Ne). Ne was estimated using two different approaches, both being the single-generation estimators. First we use the estimator based on a linkage disequilibrium measure (r2, a coefficient of determination reflecting a level of nonrandom associations among alleles from different loci) using the methods implemented in LDNe software (Waples and Do 2008). We denoted the LD-based estimator by Ne(LD). As Ne(LD) can be biased due to low-frequency alleles (Waples and Do 2010), the values of r2 were estimated after omitting alleles with a sample frequency lower than three copies. Approximate confidence intervals around the estimated Ne were obtained after jackknifing over pairs of loci. Obtained Ne estimates are valid under the assumption that linkage disequilibrium is solely an effect of genetic drift. However, because LD can arise due to physical linkage or epistatic selection, marker pairs were checked for homogeneous r2 across different populations using the Friedman rank test across pair-wise r2 values (a pair of markers as a factor). Ne was also assessed according to the method proposed by Wang (2009), based on the reconstructed proportion of full-, half-sib and unrelated pairs among a cohort of genotyped individuals (hereafter referred to as Ne(SA)). These proportions were estimated with a simulated annealing algorithm implemented in Colony 2 (Jones and Wang 2009). Unlike Ne(LD), Ne(SA) accounts for non-random mating and genotyping errors (Wang and Santure 2009). Ne(SA) was estimated with the following settings: the full-likelihood-based analysis (medium likelihood precision and medium length of run), the assumption of polygamy and dioecy. Rates for allelic dropout were substituted with locus-specific estimates of null allele frequencies provided with INEst software (see Genetic structure section). Rates of other error types were arbitrarily fixed at 0.025. Allele frequencies were updated during optimisation. No pair of individuals was a priori excluded as a possible full- or half-sib. Also, sampled individuals were assumed to form a single generation, therefore parent-offspring relations were not allowed. The analysis was initiated without a prior distribution for the paternal and maternal sib-ship sizes of the offspring.

Finally, we studied the relationship between census and effective population sizes and genetic structure parameters: allelic richness, inbreeding and coancestry. The expected number of alleles in a population generally increases nonlinearly with a population size under both the step-wise mutation model (Kimura & Ohta 1975) and the infinite alleles model (Watterson 1975). However, because our estimates of Ne were within a relatively narrow range (15–175, see Results), we assumed direct proportionality between a population size and allelic richness. In the case of both coancestry and inbreeding we used Nc1 or Ne1 as a predicting variable, because both quantities are inversely proportional to population size (Wright 1931; Reynolds et al. 1983).

Results

Genetic structure

The average number of alleles detected at five SSR loci equaled 14 per population. However, alleles were mostly detected in low copy numbers, resulting in deflated effective numbers of alleles of only 5.1 on average (Table 2). As expected, the number of alleles detected correlated with sample size. Therefore, to better reflect polymorphism levels in the study populations, allelic richness (AR) values were of primary interest. AR values spanned from 7.8 to 14 with the average of 10.9. The Friedman test revealed that single-locus AR values were not homogeneously distributed across populations (\( \chi_{r}^{2} \) = 15.17, df = 5, p-value = 0.010). When treating single-locus AR values as independent replicates, the approximate 95 % jackknife confidence bounds did not include the average AR value for two populations (i.e. Cisy nad Liswartą and Radomice). Thus, both Cisy nad Liswartą and Radomice can be considered as negatively outlying populations in terms of polymorphism levels. Yet using the same procedure, no population could be identified as a positive outlier.
Table 2

Population genetic parameters estimated for the study populations (averaged across loci): A—number of alleles, Ae—effective number of alleles, AR—allelic richness (based on 66 genes), He—expected heterozygosity, Ho—observed heterozygosity, FIS—multilocus inbreeding coefficient

Population

A

Ae

AR

He

Ho

FIS

CG

13.4

5.3

11.9

0.813

0.480

0.082

CS

14.0

5.9

12.8

0.831

0.487

0.090

LI

8.4

3.8

7.8

0.737

0.405

0.116

ST

16.6

5.3

10.9

0.813

0.547

0.037

CZ

22.2

7.4

14.0

0.865

0.679

0.017

RA

9.2

4.2

8.3

0.764

0.555

0.057

Mean

14.0

5.1

10.9

0.804

0.526

0.067

The contribution of the study populations to within- and between-population allelic richness is shown on Fig. 2. Among studied populations, only Cisy nad Liswartą and Radomice had a negative contribution to the total allelic diversity. This was mainly because of a negative contribution to the within-population component (AS). Cisy w Czarnem had, by far, the largest contribution, of which both between- and within-population components exceeded the estimates for remaining populations. Cisowa Góra and Cisy Staroploskie also contributed positively to both components of the overall allelic diversity. However, the latter population had a minor impact only. Interestingly, in spite of its positive contribution to AT, Cisy nad Czerską Strugą had a slightly negative impact on between-population allelic diversity. This suggests that the population had relatively small pair-wise differences in allelic composition with respect to the others. Finally, the partitioning method allowed the estimation of the global allelic diversity at the level AST = 0.297. Thus, about 30 % of the total allelic diversity is between populations.
Fig. 2

Partitioning of allelic diversity into within- (AS) and between-population (DB) components (note that a total allelic diversity AT = AS + DB). Bars show relative contribution (in %) of each population into the global AS and DB, while open circles—to the global AT. Positive values indicate a loss of global allelic diversity at a particular level when disregarding the population, and vice versa

Generally, genotypic proportions deviated significantly (p < 0.05) from panmictic expectations, except in three single-locus cases, namely for Tax31 in Cisowa góra, Tax362 in Cisy staropolskie, and Tax26 in Radomice (data not shown). In all cases a deficiency of heterozygotes was found (Table 2). Using INEst software, the analysis revealed that two phenomena contributed unevenly to the excessive homozogosity: null alleles and inbreeding. The average null allele frequency was about 13.3 %. Tax31 appeared the least affected by null alleles (4.2 % on average), while for the others the average null allele frequency ranged between 12.1 and 17 %. ‘Null-free’ estimates of FIS laid within 0.017–0.116 (Table 2), with the grand average 6.7 %. Among the study populations, Cisowa Góra, Cisy nad Czerską Strugą and Cisy nad Liswartą were characterized by higher than average FIS. It is worth mentioning that multilocus FIS estimates obtained with INEst appeared much lower than the classically derived values (i.e. on average 1−Ho/He = 0.33). Hence, inbreeding explained about 20 % of the large heterozygote deficiency observed in the study populations suggesting that it resulted primarily from the presence of null alleles.

The estimates for coancestry coefficients were only weakly dependent on the assumed prior distribution (Table 3), though the results were generally lower for the prior with a mean of 0.01. Individual θ values varied much, with the coefficient of variation between 53.8 and 63.4 %. For the most representative prior (μ = 0.1, see Materials and Methods) θ spanned between 0.04 and 0.222, with the average equal 0.115. Again, two populations were clearly distinguishable from the rest, namely Cisy nad Liswartą and Radomice, both having θ ≈ 0.2 (twice the average).
Table 3

Estimates of the coancestry coefficient (θ) obtained for different prior distributions (characterized by a mean μ)

Mean (μ) of the prior distribution for θ

Coancestry coefficient (θ)

CG

CS

LI

ST

CZ

RA

0.01

0.069 (0.050,0.093)

0.050 (0.034,0.070)

0.152 (0.116,0.197)

0.078 (0.060,0.101)

0.037 (0.028,0.048)

0.137 (0.104,0.178)

0.10

0.086 (0.062,0.117)

0.061 (0.042,0.089)

0.217 (0.154,0.291)

0.091 (0.068,0.116)

0.042 (0.031,0.054)

0.193 (0.140,0.259)

0.25

0.086 (0.059,0.116)

0.063 (0.042,0.088)

0.222 (0.161,0.299)

0.092 (0.070,0.119)

0.042 (0.030,0.054)

0.196 (0.139,0.265)

95 % credible intervals are given in parentheses

Using GENELAND we found that six study populations form K = 5 distinct genetic pools. The posterior probability that K > 5 was P = 0.072, while for K < 5 the probability was P ≈ 0. Based on individual ancestry probabilities (Fig. 3) for each population we estimated the overall frequency of genes belonging to one of five clusters. Except for Cisy nad Czerską Strugą, the study populations appeared clearly distinguishable from each other, with >84 % genes belonging to one of five clusters. Cisy nad Czerską Strugą, on the other hand, had the most ambiguous genetic pool with 64 % genes belonging to the cluster most likely referring to the Cisowa Góra population.
Fig. 3

Population genetic structure estimated with GENELAND software. Each individual is represented by a vertical line, which is partitioned into five segments reflecting the individual’s assignment to a given cluster. The study populations are separated with black thick lines

Effective population size

The deviance of the expected heterozygosity (He) levels from their theoretical values under the drift-mutation equilibrium depended on the mutation model assumed. Under IAM all populations revealed significant heterozygosity excess (the Wilcoxon test; p-value < 0.05). In contrast, neither TPM- nor SMM-based tests showed heterozygosity excess. Under TPM model, however, some signatures of bottleneck were noted for Cisy nad Liswartą and Radomice, where three out of five loci were characterized by excessive He levels, yet in these cases the overall He excess was not significant.

Pair-wise linkage disequilibrium coefficients across populations did not deviate from homogeneity as revealed by the Friedman test (\( \chi_{r}^{2} \) = 10.95, df = 9, p-value = 0.279). Hence, no marker combination was specifically suspected for tight linkage or epistatic selection. The LD-based effective population sizes (Ne(LD)) ranged from 15.1 to 175.7 (Table 4), with the extremes significantly different from the (harmonic) overall mean 42.5. As compared with Ne(LD), the Ne(SA) values estimated from reconstructed pedigrees were on average ca. 20 % lower and revealed more homogeneity among populations, ranging between 32 (LI) and 113 (CZ). Nonetheless, Ne(LD) and Ne(SA) were significantly positively correlated (0.94, p-value = 0.002). Generally, the lowest Nes (averaged for the two estimators) were estimated for those populations, which also exhibited more pronounced bottleneck signatures (RA and LI). The study populations differed strikingly in Ne/Nc ratio, ranging from 0.03 to 0.4 (with an average of 0.13) and from 0.02 to 0.256 (with an average of 0.11), for Ne(LD) and Ne(SA), respectively. Thus, association between Nc and Ne was poor in the study populations regardless the Ne estimator.
Table 4

Estimates of effective population size: Ne(LD)—LD-based effective population size, Ne(SA)—pedigree reconstruction-based effective population size, Ne()/Nc—a ratio of effective and census population size for a given Ne estimator

Parameter

CG

CS

LI

ST

CZ

RA

Ne(LD)

85.1 (50.3,198.9)

44.6 (32.8,64.2)

15.1 (10.1,22.8)

95.5 (76.6,120.7)

175.7 (129.2,254.8)

40.4 (25.5,72.2)

Ne(LD)/Nc

0.120

0.149

0.032

0.033

0.402

0.074

Ne(SA)

44 (29–68)

50 (34–76)

32 (20–51)

68 (48–96)

113 (88–149)

45 (30–68)

Ne(SA)/Nc

0.062

0.167

0.069

0.024

0.259

0.083

Approximate 95 % confidence intervals are given in parentheses

Population sizes were associated with estimated genetic structure parameters. The majority of variation in the genetic structure parameters was explained by Ne (from 51 to 90 %, Fig. 4). A significant positive correlation was found between number of alleles and estimates of effective population size (r = 0.95, p-value = 0.002 and r = 0.94, p-value = 0.003 for Ne(LD) and Ne(SA), respectively). A similar, yet not significant, relationship was found between Ne and allelic richness (r = 0.76, p-value = 0.081 and r = 0.71, p-value = 0.123 for Ne(LD) and Ne(SA), respectively). We also found a systematic increase in genetic identity, measured with coancestry (θ) and inbreeding (FIS), as 1/Ne increased. In the case of Ne × θ relationship correlation coefficients equalled r = 0.81 (p-value = 0.052) and r = 0.79 (p-value = 0.063) for Ne(LD) and Ne(SA), respectively. In the case of Ne × FIS relationship correlation coefficients equaled r = 0.78 (p-value = 0.070) and r = 0.92 (p-value = 0.005) for Ne(LD) and Ne(SA), respectively. In contrast, census population size (Nc) poorly explained all the above genetic structure characteristics with the values of R2 ranged from 0.002 to 0.151.
Fig. 4

The relationship between census population size (ab) or effective population size (cf) and levels of polymorphism (A, average number of alleles; AR, average allelic richness) and genetic identity (FIS, inbreeding coefficient; θ, coancestry). Lines show the best fitting linear regression functions

Discussion

The study populations of English yew were found to differ in genetic structure parameters. For example, allelic richness spanned from 3.8 to 7.4, while inbreeding coefficient and average coancestry reached values from 0.017 to 0.116 and from 0.04 to 0.222, respectively. We found that all these features were poorly related with census numbers. In contrast, marker-assisted characteristics such as effective population size and bottleneck signatures were found to explain the observed genetic structure.

Up to now several studies showed that some inbreeding might be present in yew populations, although FIS estimates varied much among populations (Lewandowski et al. 1995; Myking et al. 2009; Dubreuil et al. 2010; Chybicki et al., 2011). Our results with the average FIS = 0.067 were close to those reported for Norway populations (Myking et al. 2009) but lower than those reported by Dubreuil et al. (2010). This suggests that southern yew populations, at least in the Montseny Mountains, might be subject to more intense inbreeding. However, FIS estimates reported for populations in the Montseny Mountains did not account for null alleles and thus might be severely overestimated (Chybicki and Burczyk 2009). Generally, given dioecy of the species, even low inbreeding levels suggest that strong bi-parental inbreeding occurs routinely in this species. A clear kinship structure found within populations may facilitate mating between closely related individuals (Chybicki et al., 2011), given that pollen dispersal is limited. Nonetheless, to our knowledge, rates of pollen-mediated gene flow and mating patterns have not yet been directly studied in English yew. The only results are probably those by Lewandowski et al. (1995), based on seeds collected from 15 female trees in the Cisy Staropolskie reserve. However, they do not provide a strong support for the hypothesis of non-random mating, highlighting the need for additional studies that address the issue more directly.

The study populations appeared to have elevated coancestry coefficients. It is generally acknowledged that coancestry results from limited gene immigration, when the probability of sharing a common ancestry evolves over generations (Reynolds et al. 1983). Results of GENELAND analysis revealed that all the study populations, except Cisy nad Czerską Strugą, are characterized by distinct genetic pools. This finding supports the hypothesis about strong isolation of yew populations. In the case of Yew, the isolation is directly related with the fact that their populations in Poland are typically separated by large distances. Pollen might experience difficulty in traveling large distances in spite of being wind-dispersed and possesing a relatively low terminal velocity (Dyakowska 1959). For pollen to be dispersed over long distances, movement to higher altitudes is generally required (Di-Giovanni et al. 1996). This poses a significant obstacle for typical understory species, such as English yew, as the presence of a closed canopy can substantially reduce wind speed and convective currents needed for long distance pollen dispersal. Moreover, one can predict that together with an increase of forest density, wind speed can decrease enough to severely restrict pollination within a population. Cisy Staroploskie is a particularly good example of such a situation. The study on pollen dispersal conducted there with the use of pollen traps revealed that pollen grain density decreases rapidly together with distance from a male individual (Noryśkiewicz 2006). This phenomenon certainly facilitates a development of inbreeding by reduction of effective population size, as discussed in our previous paper (Chybicki et al., 2011).

In contrast with remaining populations, Cisy nad Czerską Strugą had a relatively low estimate of coancestry, given that it has the lowest census number among all study populations. However, GENELAND analysis showed that individuals belonging to Cisy nad Czerską Strugą did not cluster as a separate genetic pool, and instead they formed a mixture of different populations. Interestingly, about 64 % of genes clustered with Cisowa Góra located about 370 km away. Such a high admixture would be difficult to explain in this case unless we assume that Cisy nad Czerską Strugą was established artificially with a use of seedlings of various origin. In this way we could address the issues of low coancestry in the population coupled with relatively high allelic richness, which has support in neither census nor even effective population size. High heterogeneity of the genetic pool of Cisy nad Czerską Strugą was also in agreement with the partitioning of allelic diversity. The contribution of between-population allelic diversity was in this case negative because the population shares many alleles with the others.

In the face of strong fragmentation, high isolation of yew populations allow them to maintain many unique alleles. Our results showed that four out of six study populations had a positive contribution to the total allelic diversity (see Fig. 2). Interestingly, the two populations that had a negative contribution (RA and LI) also displayed signatures of a possible bottleneck in their recent history. This rapid reduction in population size could explain these partitioning patterns. However, using genetic markers alone, we could not verify whether bottlenecked populations experienced a natural process of rapid reduction in size or rather they were established with human assistance from seeds derived from a few female individuals in the recent past. Unfortunately, the history of these populations before the present generation is not known precisely, making factors of a putative bottleneck effect rather speculative.

Wind-pollination can generally facilitate the attainment of a large population size, though it can be influenced by a level of population fragmentation (Bucci et al. 1997; De-Lucas et al. 2009) interfering with a subpopulation density (Grivet et al. 2009). Nonetheless, the values of effective population size estimated in this study were generally lower than those observed for trees forming large continuous populations (e.g. Bucci et al. 1997; Chybicki et al. 2008). We also found a poor correspondence between census and effective population size suggesting that different populations have different demographic processes, or different demographic histories. For example, incidental bottlenecks and different sex proportions might be responsible for Ne/Nc heterogeneity. Under random mating the expected Ne/Nc ratio equals 4rf(1−rf), where rf is the proportion of females. For the study population, one might then expect Ne/Nc between 88.6 and 99.6 % (based on data in Table 1), while our Ne/Nc estimates were within 3.2–40.2 % or even 2.4–25.9 %, depending on the estimation method. This clearly shows that biased sex proportions alone cannot explain the poor relationship between census and effective number. One reason for such discrepancy would be some variation in a fraction of sexually active individuals in populations. It is known that the production of reproductive organs increases strongly with tree size and canopy openness (Svenning & Magård 1999). Also, femaleness is dominant in the open, while maleness in shade. This could explain relatively high Ne/Nc for Cisy w Czarnem, which occurs at the most open habitat compared with the others. For comparison, Cisy Staropolskie represents the opposite extreme, with a high density and a well developed canopy layer. Among additional factors, non-random mating patterns (as discussed earlier), fluctuating population size, population substructure (see Dubreuil et al. 2010, Chybicki et al. 2011) or overlapping generations could be attributed to deflated Ne values. Overlapping generations needs particular attention, as this feature might potentially influence both a demographic parameter (i.e. Ne) and its estimate (both Ne(LD) and Ne(SA) relied on the assumption of a single-generation cohort). English yew is know for longevity, that makes a possibility of mixing generations in natural populations. Therefore, although our study is based on samples of adult individuals, we cannot definitely exclude a possibility that, though to limited degree, our samples are a mixture of two generations. Although the two methods gave well correlated estimates, the values of Ne(SA) were on average lower than those of Ne(LD). Generally, the above-discussed factors could be responsible for discrepancies between the two Ne estimators. Nonetheless, it should be noted that Ne(SA) was found to be negatively biased when information content from genetic markers is low (here only five loci) (Wang 2009). On the other hand, low frequency alleles could cause overestimated Ne(LD) (although we applied the heuristic correction, as suggested by Waples and Do 2010, see M&M). Interestingly, similar trends were also noted in the recent study of ranid frogs (Phillipsen et al. 2011), suggesting that further attention is needed to determine susceptibility of a particular method to an information content of genetic markers.

We found high correlation between coancestry and 1/Ne estimates. It is worth mentioning that the correlation is not explicitly obvious. Actually, these two measures were estimated from very different data features, with θ being of single-locus (standardized) variances of allele frequencies, while Ne—a function of between—loci allelic correlations (LD) or a function of inter-individual relationships (SA) (however, all these features are related to the probability of identity by descent, Sved 1971, Balding 2003). Furthermore, both Ne estimators refer to inbreeding effective population size, while θ has affinity with variance effective population size (Lindgren and Mullin 1998). These two differently defined Nes can vary to a large extent unless a population is of constant size (Crow and Kimura 1970). Thus, the lack of complete correlation between 1/Ne and θ might be accountable (besides the low precision of estimates) to differences in demographic history of the populations, including migration rates, divergence times and even initial coancestry and linkage disequilibrium of the isolates.

Finally we would like to address the result of allelic diversity through AST. Recently, simulation studies have revealed that allelic diversity and gene differentiation measures can provide complementary information about demographic processes (Caballero and Rodríguez-Ramilo 2010). In particular, AST and FST behave differently in the case of low and high migration rates. For the study populations AST was about three times higher than the average coancestry coefficient (which can serve as an estimate of global FST, Holsinger 1999). Such a pronounced difference was due to the presence of many private alleles at low frequencies, which have a strong impact on AST while negligible on FST. Asymptotically, an AST greater than FST is predicted for high migration rates (Caballero and Rodríguez-Ramilo 2010). In our case, however, it is a questionable scenario, because populations seem almost completely disconnected due to separation by large distances. One reasonable explanation could be related to the observation, that AST approaches stationarity much faster than FST (Caballero and Rodríguez-Ramilo 2010), especially when migration is incidental. Another possible explanation could be high mutation rates of microsatellite markers (μ > 0.0001), that would strongly enhance allelic diversity but only weakly gene differentiation.

Perspectives for conservation

Currently, conservation biologists in Poland face the problem of the restitution of English yew as a component of the forest understory. However, because existing populations are strongly fragmented and typically small, the question remains, which populations should serve as main seedling sources? Two factors should be considered here. Firstly, population characteristics can determine present-day seedling viability, which is important for the effectiveness of re-establishment. Secondly, population characteristics can determine genetic quality of seedlings in terms of polymorphism, inbreeding, and genetic diversity levels. All these features are crucial from the perspective of long-term conservation of species’ gene pools (Mistretta 1994; Ellstrand and Elam 1993; Frankham et al. 2002; Willi et al. 2006; Charlesworth and Willis 2009).

Our study suggested that choosing source populations for ex situ conservation based on their size only can be fairly misleading. For example, Radomice and Cisy nad Liswartą, two relatively large yew populations, both had a negative contribution to the total allelic richness, which means that excluding these populations in ex situ conservation programs would cause a gain in the total allelic richness in the next generation. Interestingly, the largest Polish population Cisy Staropolskie had almost null contribution to all components of allelic diversity. In contrast, the relatively small population of Cisy nad Czerską Strugą had a positive contribution to the total allelic richness. These conclusions could also be extended to coancestry or inbreeding. It is evident that inbreeding would not be minimized efficiently when seedlings were derived from populations chosen based on the census number only. It is worth noting that current-generation coancestry coefficients can shed light on future inbreeding levels within a population. It is known that, under random mating, the total (or group) coancestry of a population becomes inbreeding of the progeny. Thus, based on the results we can predict that all the study populations, except Cisy nad Czerską Strugą, will probably be subject to an increase of inbreeding in the next generation, but the rates of increase will be poorly related with census population numbers.

Expectedly, allelic richness and inbreeding were functions of effective and not census population size. Although, generally the Ne/Nc ratio was about 0.1 and hence corresponded well with the findings for natural populations (Frankham 1995), Ne/Nc appeared to vary much from population to population, making indirect inferences about Ne problematic. Incorporating data on sex ratio or population density was ineffective and did not improve inferences, as discussed above. Further studies are required to resolve whether more precise predictions of the Ne/Nc ratio could be obtained with a number of additional characteristics, such as age structure and fecundity distribution. However, one can anticipate that information on the current demographic structure would be useful if the population is demographically stable. If not, historical factors could dominate over present-day processes making predictions invalid. Our study demonstrated that the use of neutral genetic markers would be helpful to determine demographic parameters of conservation units. However, neutral markers could not enable direct insights into adaptive potential of populations, which might do not correspond with polymorphism and inbreeding levels. Hence, ex situ conservation based solely on the predictions drawn from neutral markers might lead to negative consequences. For example, if isolated populations are well adopted through specific gene combinations, progeny populations derived as a mixture might suffer from outbreeding depression. This need to be taken into account in conservation programs. Nevertheless, the use of neutral genetic markers in ex situ programs should not be underestimated.

Acknowledgments

This study was supported by the research grant from the Polish Ministry of Science and The Higher Education (NN304 4216 33) to IJC. The authors would like to thank Maria Jennifer Sjölund as well as two anonymous reviewers for making valuable suggestions that improved the manuscript.

Open Access

This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.

Copyright information

© The Author(s) 2012

Authors and Affiliations

  • Igor J. Chybicki
    • 1
  • Andrzej Oleksa
    • 1
  • Katarzyna Kowalkowska
    • 1
  1. 1.Department of GeneticsKazimierz Wielki UniversityBydgoszczPoland

Personalised recommendations