Does the seed bank contribute to the build-up of a genetic extinction debt in the grassland perennial Campanula rotundifolia?

Abstract

Background and Aims

Habitat fragmentation threatens global biodiversity. Many plant species persist in habitat fragments via persistent life cycle stages such as seed banks, generating a species extinction debt. Here, seed banks are hypothesized to cause a temporal delay in the expected loss of genetic variation, which can be referred to as a genetic extinction debt, as a possible mechanism behind species extinction debts.

Methods

Fragmented grassland populations of Campanula rotundifolia were examined for evidence of a genetic extinction debt, investigating if the seed bank contributed to the extinction debt build-up. The genetic make-up of 15 above- and below-ground populations was analysed in relation to historical and current levels of habitat fragmentation, both separately and combined.

Key Results

Genetic diversity was highest in above-ground populations, though below-ground populations contained 8 % of unique alleles that were absent above-ground. Above-ground genetic diversity and composition were related to historical patch size and connectivity, but not current patch characteristics, suggesting the presence of a genetic extinction debt in the above-ground populations. No such relationships were found for the below-ground populations. Genetic diversity measures still showed a response to historical but not present landscape characteristics when combining genetic diversity of the above- and below-ground populations.

Conclusions

The fragmented C. rotundifolia populations exhibited a genetic extinction debt. However, the role of the seed banks in the build-up of this extinction debt is probably small, since the limited, unique genetic diversity of the seed bank alone seems unable to counter the detrimental effects of habitat fragmentation on the population genetic structure of C. rotundifolia.

INTRODUCTION

Earth has entered the Anthropocene, an era in which the human imprint on the Earth is measurable (Hoekstra et al., 2005; Steffen et al., 2011). The disruption of biogeochemical nutrient cycles, climate change, land cover changes and habitat destruction as a result of anthropogenic actions are all driving global biodiversity losses (Sala et al., 2000). Habitat destruction and accompanying fragmentation is considered one of the most pervasive factors (Sala et al., 2000; Hoekstra et al., 2005). When formerly continuous habitat becomes divided into small and isolated fragments, many plant species experience a strong decrease in population size (Eriksson, 2000; Honnay and Bossuyt 2005) and reduced dispersal (Lindborg et al., 2012). As well as having direct effects on demographic stability and on pollination success (Hesse and Pannell, 2011), these changes may be accompanied by erosion of genetic diversity (Leimu et al., 2006; Jacquemyn et al., 2012) and inbreeding (Keller and Waller, 2002), as a result of genetic drift and reduced among-population gene flow (Young et al., 1996; Honnay and Jacquemyn, 2007). These genetic processes further reduce population viability (Spielman et al., 2004), leading to an increased risk of local extinction (Aguilar et al., 2008).

Still, many plant species initially persist beyond the point when their local environment has become unsuitable, by means of resistant life cycle stages such as persistent seed banks (Lindborg et al., 2012; Plue and Cousins, 2013) or through perenniality and clonality (Eriksson, 2000; Honnay and Bossuyt, 2005). These various means enabling population persistence, combined with reduced seed dispersal between isolated habitat patches (Lindborg et al., 2012), alter the balance between failing immigration and slow extinction processes in response to a changing landscape configuration (Jackson and Sax, 2010). This results in the delayed extinction of plant species in fragmented plant communities, referred to as the landscapes’ extinction debt (Helm et al., 2009; Vellend et al., 2006; Kuussaari et al., 2009). When relating present biodiversity patterns to both historical and contemporary landscape characteristics (e.g. Helm et al., 2009; Vellend et al., 2006; Kuussaari et al., 2009; Krauss et al., 2010), the observation of a stronger relationship between species diversity and historical landscape patterns, rather than between species diversity and contemporary landscape patterns, provides evidence for such an extinction debt (Kuussaarri et al., 2009; Krauss et al., 2010). This extinction debt implies that some species will become extinct in the near future due to current landscape conditions. Fragmentation-induced extinction debts are increasingly uncovered across taxa [plants and butterflies (Krauss et al., 2010); birds (Szabo et al., 2011)], ecosystems [temperate forest (Vellend et al., 2006); tropical forest (Wearn et al., 2012); grassland (Saar et al., 2012); heathland (Piessens and Hermy, 2006) and geographical regions [North-west Europe (Krauss et al., 2010); North-America (Vellend et al., 2006); Australia (Szabo et al., 2011)]. The mechanisms underlying species extinction debts in fragmented landscapes, however, remain poorly understood (Hylander and Ehrlén, 2013).

As genetic diversity is expected to decline with habitat fragmentation before species diversity starts to decline (Aguilar et al., 2008; Helm et al., 2009), temporal delays in the loss of genetic diversity, in seemingly persistent species, may be an indicator of a delay in their extinction. Yet, testing for the existence of a conceptually similar genetic extinction debt, i.e. expected future losses of genetic diversity which is still present among fragmented populations today, has only rarely been undertaken. There is strong evidence that historical landscape changes and shifts in habitat continuity affect current population genetic structure (Jacquemyn et al., 2004; Vellend, 2004; Prentice et al., 2006; Vandepitte et al., 2007; Rosengren et al., 2013). Yet, the few studies that directly tested whether current population genetic diversity is best predicted by historical habitat configuration rather than present habitat configuration are mixed. Honnay et al. (2006) and Helm et al. (2009) found no evidence of a genetic extinction debt in the grassland species Anthyllis vulneraria and Briza media, respectively. However, Münzbergová et al. (2013) found genetic diversity in Succisa pratensis to be related to both current and historical landscape structure, implying a temporal delay in genetic erosion following habitat fragmentation.

Plant life history traits enabling prolonged population persistence such as perenniality, clonality or persistent seed banks have been suggested to contribute to delayed plant species extinctions (Honnay and Bossuyt, 2005; Plue and Cousins, 2013). Each of these may therefore also mediate temporal delays in the response of genetic diversity to habitat fragmentation. In this study, we will focus specifically on the role of persistent seed banks, as recurrent seedling recruitment from seed bank genotypes into fragmented populations may increase effective population size and thus mitigate random genetic drift (Nunney, 2002; Honnay et al., 2008; Ayre et al., 2009; Roberts et al., 2014). Though the magnitude may be dependent upon seed longevity (Honnay et al., 2008), seed bank species may keep their population genetic diversity at the pre-fragmentation level for a longer time, extending the relaxation time (Krauss et al., 2010). Still, the role of the seed bank in mediating population genetic diversity following habitat fragmentation is poorly understood (Honnay et al., 2008; McCauley, 2014).

Here, we studied patterns in population genetic diversity of both above-ground plants and below-ground seed bank individuals in a grassland species. The aim was to detect a genetic extinction debt and assess the role of the persistent seed bank in generating this extinction debt. We selected the grassland species Campanula rotundifolia (Stevens et al., 2012). The species declines rapidly after habitat fragmentation but persists as remnant populations (Lindborg et al., 2005) and invests in a persistent seed bank (Milberg, 1995; Plue and Cousins, 2013). Our study area is a landscape which has been progressively fragmented over the last 150years (Cousins and Eriksson, 2008), rendering it a suitable landscape to test for the presence of a genetic extinction debt (Kuussauri et al., 2009). Our main objective is to test the hypothesis that the persistent seed bank and its genetic diversity may slow down genetic erosion to create a genetic extinction debt. To achieve this, we first established the presence of a genetic extinction debt for the above-ground populations of C. rotundifolia. Subsequently, we test the potential seed bank’s role in generating this genetic extinction debt. We hence address the following questions: (1) are patterns in above-ground population genetic diversity best predicted by the historical rather than the present landscape configuration? (cf. Kuussauri et al., 2009) and (2) to what extent does the seed bank hold a unique component of genetic variation in the below-ground population, which may act to buffer genetic erosion and differentiation in the above-ground populations? Genetic processes act on the entire population gene pool, including both the above- and below-ground component, which are linked through seedling recruitment and incorporation of new seeds to the seed bank. We can therefore address the role of below-ground genetic diversity in mediating genetic responses to habitat fragmentation by questioning: (3) does the combined genetic diversity of the above- and below-ground populations remain best predicted by the historical rather than current landscape structure, compared with above-ground genetic diversity only? If so, the role of the genetic diversity of the below-ground populations in generating the genetic extinction debt may be limited.

MATERIALS AND METHODS

Study species

Campanula rotundifolia L., common harebell, is a slender, rhizomatous and evergreen perennial herb (Stevens et al., 2012) which is largely self-incompatible and bee pollinated. Campanula rotundifolia has a wide ecological amplitude and is found in grassland, heath and dune communities across a wide soil pH gradient. Campanula rotundifolia spreads by seed and by slow vegetative growth from rhizomes. It typically produces 10–100 seeds per flower. Seeds are small (0·69–1·22mm long by 0·38–0·53mm wide) and lightweight (56–81μg), enclosed in a capsule which releases seeds after being shaken by wind or grazing (Stevens et al., 2012). Their limited weight enables wind dispersal (Bonde, 1969), yet most seeds are deposited near the parent plant, with median and 99th percentile dispersal distances of 0·07 and 0·35 m, respectively (Soons and Ozinga, 2005). With mean distances of 248 m in 2011 (see ‘Study area and population sampling’), this implies a low chance of successful seed dispersal between habitat patches. Campanula rotundifolia is known to form a persistent seed bank, ranging in size between 5 and >1000 seeds m⁻² (Stevens et al., 2012). Stevens et al. (2012) note that C. rotundifolia’s seed longevity may be<5years, but the authors suspected that seed longevity may be>5years based on Milberg (1995). Indeed, both Milberg (1995) and Plue and Cousins (2013) recorded C. rotundifolia seeds in the absence of parent plants in northern Europe. Seed longevity is indeed a plant trait subject to climatic and environmental control (Pakeman et al., 1999; Abedi et al., 2014). Overall, we thus assume a mean longevity of 5years, ranging between 2 and 10years (Julia Wilson, pers. comm.), i.e. seed viability declining after 2years, though some seeds may survive up to 10years under environmental conditions promoting longevity. The C. rotundifolia seed bank thus consists of a mix of seeds of varying age, descent and genetic make-up. In warmer months, seed banks drive light-dependent germination of C. rotundifolia (Koutsovoulou et al., 2014), being further enhanced by gap creation, which is essential to seedling survival (Stevens et al., 2012). Campanula rotundifolia is a habitat specialist dependent upon grassland management, though it may persist in remnant populations, ensuing fragmentation and abandonment, recovering quickly as grassland management is restored (Lindborg et al., 2005). Campanula rotundifolia reproduces by both sexual and clonal recruitment in our study area, with seedlings forming a tapering root, while clonal offspring do not (Stevens et al., 2012). We thus estimate that individual C. rotundifolia ramets may only have a relatively limited mean longevity of 5years, potentially doubled if the ramet has produced and banked long-lived seeds. Above-ground C. rotundifolia populations thus are a mix of individuals of varying age, descent and genetic make-up, similar to those below-ground. Campanula rotundifolia has a circumpolar boreo-temperate distribution (Hultén and Fries, 1986), with a complex cytotype distribution, dominated by diploid and tetraploid forms (Laane et al., 1983). In central Sweden, only the tetraploid form occurs (Böcher, 1966). Campanula rotundifolia is considered an autotetraploid species (Laane et al., 1983). For a detailed species’ ecology, see Stevens et al. (2012).

Study area and population sampling

The study site covered a 25km² area of the island of Selaön (59°24'N, 17°10'E) in Lake Mälaren, central Sweden (Fig. 1). Selaön is an open agriculture–forest landscape (56 % agriculture and 36% coniferous forest) with little remaining semi-natural grassland. Based on historical land cover maps, Cousins and Eriksson (2008) reconstructed grassland fragmentation on Selaön. Grassland cover declined gradually from 60 % in 1854 to 35 % in 1897 due to agricultural expansion. The 1930s forest grazing ban and the 1940s rise of commercial forestry led to rapid habitat loss to approx. 10 % grassland cover by 1954. This landscape has seen a loss in spatially continuous grassland patches and a strong increase in grassland patch isolation. Further grazing abandonment has slowly reduced semi-natural grassland area to 2 % of the study area in 2011, with increasing grassland patch isolation (from 138 m in 1954 to 248 m in 2011) and a decrease in grassland patch size (Cousins and Eriksson, 2008).

Fig. 1.

The fragmentation of semi-natural grasslands (black) in the Selaön study area (central Sweden) over the last 57years, including the location of the sampled Campanula rotundifolia populations (crosses). Dotted areas are forest; white areas are crop-fields (for detailed information, see Cousins and Eriksson, 2008).

Within the landscape, 20 extant C. rotundifolia populations were arbitrarily selected from a set of 25 populations, for which a persistent seed bank was known to be present based on an earlier assessment (Plue and Cousins, 2013). Each population was visited in July 2011. We sampled the soil seed bank, taking 50 core samples (3·5cm diameter, 5cm deep) in a 2 × 2 m plot (cf. Plue and Hermy, 2012), located in the centre of each extant population. The 50 soil core samples, without litter layer, were pooled per plot. The concentrated samples (Ter Heerdt et al., 1996) could germinate under a 16h day, 8h night regime, with a constant temperature of 21°C for 3months. A range of similar temperature and photoperiod germination conditions has led up to 80–89 % germination of C. rotundifolia seeds (Kew Seed Information database; http://data.kew.org/sid/). This germination regime captures the majority of sampled C. rotundifolia seeds of each population. All C. rotundifolia seedlings were counted as a proxy for seed bank size. Leaf material was collected from 25 seedlings per population for genetic analysis. Although seedlings emerged in all 20 populations, five populations had <10 seedlings, which were deemed inadequate to quantify the genetic diversity in the seed bank, creating bias in further analyses. These populations were hence omitted. The remaining 15 above-ground populations were revisited in July 2012. The number of ramets was counted in the same 2 × 2 m plot used for seed bank sampling, providing a comparable estimate of population size. Collection of the leaf samples for genetic analysis was done on 25 arbitrarily chosen individuals from the entire area occupied by a population, both inside and outside the seed bank sampling plot. Care was taken to avoid sampling clonal offspring by leaving at least 50cm between sampled individuals. When spaced closer together due to limited availability of individuals, we checked for the absence of physical connections between ramets. All leaf material was dried and stored in silica gel. For summary statistics of the population variables, see Supplementary Data Table S1.

Microsatellite analysis

Total DNA was extracted from 10–20mg of silica-dried leaf tissue of 674 individual samples following the NucleoSpin Plant II protocol (Macherey-Nagel, Düren, Germany). DNA quality and concentration were estimated using a NanoDrop ND-2000 spectrophotometer (Thermo Scientific, Wilmington, DE, USA). DNA amplifications were carried out in two multiplexes of four microsatellites (Plue et al., 2015) in a 10μL reaction volume containing 2–10ng of DNA (2μL), 1·2μL of one of the two multiplexed primer combinations (both forward and reverse primers, 0·3 μm), 5μL of Qiagen HotStarTaq Master Mix and 1·8μL of double-distilled water. The following cycling protocol for the PCR was performed on a TC-412 Programmable Therma Controller (Techne): an initial denaturation step at 95 °C for 15min, followed by 35 cycles of 30s at 94 °C, 90s at 56 °C and 60s at 72 °C, with a final extension step of 30min at 72 °C. Fragment lengths of the PCR products were determined with a ECO500 loading dye using an ABI 3730 genetic analyser. Fragments were scored in GeneMapper v4.0 (Applied Biosystems). As all individuals were tetraploid, every individual was scored for 1–4 bands per marker. The validity of each allele call after an automated analysis procedure of the electropherograms was confirmed by a visual inspection of the automated calls. A second similar visual check of the allele calls was then carried out by a second trained technician, ensuring scoring consistency for each marker, throughout the data set. Regardless, weak and ambiguous allele callings were conservatively removed to avoid creating false genetic variation. Repeats on 15 individuals (one per above-ground population) confirmed highly consistent allele calling and replicability. Across 309 original allele calls, three original calls were not found in the repeats (error rate <1 %), whereas seven allele calls were discovered across the 313 allele calls in the repeat samples. The uncertainty rate was hence 3·16 % [total number of uncertain allele calls (3 + 7)/total number of unique allele calls (309+7)], validating data quality and reliability. Sixteen individuals (2·4 %) were removed from the data, as they had missing data in four loci or more.

Present and historical landscape

Although a 1854 land cover map was available, the map only covered 12 of 15 sampled populations. In order not to decrease the number of population replicates, we restrict ourselves to the 1954 and 2011 land cover maps to test for the presence of a genetic extinction debt. In the preceding decade, the 1954 landscape had lost spatially continuous grassland patches and saw a strong increase in grassland patch isolation (see ‘Study area and population sampling’). Assuming a genetic extinction debt to be present, we expect that genetic diversity patterns would only be related to the 1954 landscape configuration. For each of the 15 sampled populations, habitat patch area and connectivity were calculated for the 1954 and 2011 time steps in ArcGis 9.3.1 (ESRI 2009). Connectivity was approximated by S, an index considering the distance (d) of an occupied habitat patch (i) to all other occupied habitat patches (j) as well as their habitat patch area (Aj), with α and b as weighting parameters of the effect of distance on migration and the effect of habitat patch size respectively (Moilanen and Nieminen, 2002):

$$S_i={\displaystyle \sum e^{-\alpha d_{ij}}}{A}_j^b$$

The weighted α and b parameters were set at 1 and 0·5 respectively, as in Honnay et al. (2006). Besides the 15 sampled populations, all other habitat patches with C. rotundifolia were identified from Plue and Cousins (2013) to calculate connectivity. Distances were measured as the shortest distance between the edges of two habitat patches.

To calculate historical habitat patch area and connectivity, current habitat fragments containing C. rotundifolia populations were cross-referenced to the aerial photograph-based 1954 map to (1) establish that all currently occupied fragments were indeed semi-natural grasslands in 1954 and (2) identify all other fragments which qualified as semi-natural grassland in 1954. Next, due to lack of data on the historical distribution of C. rotundifolia, we assumed (1) that all current occupied fragments were also occupied in 1954 and (2) that all other semi-natural grassland fragments in 1954 contained C. rotundifolia, since it is currently present on all semi-natural grasslands in the study area. These assumptions suggest that C. rotundifolia decline follows the loss of semi-natural grassland (Cousins and Eriksson, 2008), as a prerequisite to test the hypothesis of a genetic extinction debt. This assumption is likely given C. rotundifolia’s reported decline across Europe due to agricultural intensification and forest encroachment (Stevens et al., 2012), also two dominant drivers of landscape change in our studye area (Lindborg et al., 2005; Cousins and Eriksson, 2008). Still, our assumptions may overestimate C. rotundifolia’s historical distribution, having used its potential historical distribution to calculate connectivity. Even so, how connectivity was calculated remains ecologically sensible, reflecting the fact that C. rotundifolia’s 1954 meta-population was better connected than the 2011 meta-population, due to more available and suitable grassland fragments (Fig. 1), even if some may have been temporarily unoccupied. Moreover, C. rotundifolia’s historical distribution was probably broader compared with its current distribution given continued habitat fragmentation, regardless of whether some habitat fragments were momentarily unoccupied in 1954 or not (Fig. 1; cf. Stevens et al., 2012). The 1954 connectivity metric can thus be confidently used to interpret any significant relationship to current genetic diversity or composition, as a true effect of changing meta-population connectivity with landscape change over the past 57years. For detailed summary statistics of 1954 and 2011 habitat patch area and connectivity, see Table S1.

Data analysis

As C. rotundifolia is autotetraploid, we used the AUTOTET software to estimate genetic diversity metrics (Thrall and Young, 2000). For each vegetation (above-ground; n=15), seed bank (below-ground; n=15) and the combined above-ground+below-ground population (n=15), we calculated the total number of alleles (sum of all alleles across all loci, A_T) and allelic richness (mean number of alleles per locus, A). We expected a genetic extinction debt to be most visible in patterns of allelic richness in rare alleles. We therefore calculated the total number of rare alleles (sum of rare alleles across loci, A_R), found in less than 5 % and 2·5 % of all individuals. Expected heterozygosity (H_E) and the inbreeding coefficient (F_IS) were calculated based on random chromosomal and random chromatid segregation, as C. rotundifolia’s segregation pathway at meiosis is unknown. Under the latter, sister chromatids segregate into the same gamete during meiosis, leading to lower heterozygosity than expected under random chromosomal segregation (Thrall and Young, 2000).

To test for a genetic extinction debt separately in the above- and below-ground populations, we used Generalized Linear Models (GLMs) with their respective genetic diversity measures as dependent variables. The tested predictor variables were (1) population size (number of individuals or seedlings); (2) 1954 habitat patch area; (3) 2011 habitat patch area; (4) 1954 connectivity; and (5) 2011 connectivity. Due to predictor variable collinearity (see Supplementary Data Table S2 and Fig. S1), our modelling approach involved adding a single predictor variable into a null model (De Frenne et al., 2011). Error structures across the GLMs used a Poisson (count data) or Gaussian distribution (normally distributed data). Population size and habitat patch area were log(X)-transformed to meet the assumption of normality and homoscedasticity. To evaluate the significance of adding a predictor to the null model, χ² statistics compared the null model with the model including one single predictor variable (‘anova()’ in base R functions; Zuur et al., 2009). This step was repeated for each of the five predictor variables. In cases where population size proved significant, we investigated the additive effects of the remaining predictor variables on a model containing population size as a main effect. This meant including a single predictor variable in a model that contained population size, using χ² statistics to compare models with and without the added predictor variable for the additive significance of that variable.

Before exploring the seed bank’s contribution to the genetic extinction debt, we tested the extent to which below-ground populations store genetic diversity. First, paired t-tests compared genetic diversity between above- and below-ground populations. This comparison was carried out on the allelic diversity metrics calculated (1) using the raw data and (2) after rarefaction to correct for above- and below-ground population size differences (rarefied to ten individuals; SPAGeDI software, Hardy and Vekemans, 2002). Then, to test whether the below-ground populations hold genetic diversity not present in the above-ground populations, we partitioned the total number of alleles across all loci into three categories per population: (1) alleles unique to below-ground individuals; (2) alleles unique to above-ground individuals; and (3) alleles shared between above- and below-ground populations. These three variables were compared using paired t-tests, to uncover where most unique genetic diversity was present. We ran a randomization procedure to test whether the genetic diversity partitioning was significantly different from random, establishing whether the number of unique alleles was higher or lower than expected by random per allele category (i.e. shared, unique above-ground and unique below-ground). All above- and below-ground individuals of a population were 999 times randomly re-assigned as being either an above- or a below-ground individual of that population. At each run, the total number of alleles across all loci was partitioned as before. This randomization procedure was run for each population, and the mean numbers of unique and shared alleles from the 999 draws served as the randomized partitioning of alleles. Paired t-tests compared the randomized and observed partitioning of alleles. Finally, a MANOVA using these three variables as multiple non-independent response variables (‘manova()’ in base R functions) tested how the observed genetic diversity partitioning may change with population size, 1954 and 2011 habitat patch area and 1954 and 2011 connectivity.

To test whether the below-ground populations contributed to the genetic extinction debt via their private genetic diversity, the above GLM procedure was repeated using the combined above-ground–below-ground genetic diversity measures. A changed GLM outcome compared with the above-ground population patterns would identify if the private, below-ground genetic diversity alters the genetic extinction debt pattern, hinting at the seed banks’ importance for population genetic dynamics following fragmentation. Population size was approximated by the sum of the number of seedlings and ramets per plot.

Finally, pairwise genetic differentiation (F_ST) among above- and below-ground populations was calculated using the ‘polysat’ R package (Clark and Jasieniuk, 2011). A redundancy analysis (RDA; ‘rda()’, ‘vegan’ package) used this pairwise genetic distance matrix to explore population genetic variation constrained by five environmental predictor variables, adding above-/below- (population) as a sixth predictor. An automatic forward selection procedure (Blanchet et al., 2008; ‘ordistep()’, ‘vegan’ R package) reduced the number of environmental predictors while maximizing explained variation, re-running the RDA with the most important predictors. Restricted Monte Carlo randomization tests (i.e. 999 restricted permutations so that observations of one predictor variable are only permuted within levels of another predictor variable; ‘how()’ ‘vegan’ R package) were then used to test the significance of remaining predictors in the reduced RDA in explaining compositional genetic variation. Finally, we tested for multivariate homogeneity of variances (‘betadisper()’, ‘vegan’ R package; Anderson et al., 2006) to quantify differences in the size of variation in genetic composition between above- and below-ground populations. This procedure is the multivariate analogue of the Levene test for homoscedasticity and uses the F_ST distance matrix to evaluate whether the above- and below-ground populations have the same level of multivariate variation in genetic composition.

Despite multiple testing, no Bonferroni corrections were applied since it was unnecessarily conservative, following the arguments by Gotelli and Ellison (2004). All analyses were performed in R-3.3.1 (R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

Genetic extinction debt

All above-ground simple sequence repeat (SSR) genotypes were unique, suggesting that the sampling procedure avoided sampling clonal ramets. Even below-ground SSR genotypes were unique, suggesting an adequate level of resolution with the microsatellite set to capture each population’s genetic structure. Across all eight loci, we uncovered 8–26 alleles per locus in above-ground populations vs. 7–22 alleles per locus in below-ground populations, averaging 14 alleles per loci overall (for summary statistics, see Table S1).

None of the measures of genetic diversity for the above- or below-ground C. rotundifolia populations was related to population size (Table 1). Although connectivity in 1954 had little measurable influence on above-ground genetic diversity, habitat patch area in 1954 was significantly and positively related to average number of alleles, the total number of rare alleles (both A_{R5 %} and A_{R2·5 %}) and expected heterozygosity (H_E). The inbreeding coefficient (F_IS) was not affected by any of the landscape variables (habitat patch area and connectivity in 1954 and today). In the below-ground populations, none of the measures of genetic diversity displayed a significant relationship with any of the landscape variables (Table 1).

Table 1.

Effects of population and landscape characteristics on genetic diversity in above-ground (Vegetation), below-ground (Seed bank) and combined above- and below-ground (Vegetation + Seed bank) populations

													Chromo						Chromd
n = 15	A t			A			A r (2·5 %)			A r (5 %)			H E			F IS			H E			F IS
	χ2			χ2			χ2			χ2			χ2			χ2			χ2			χ2
	Deviance	P		Deviance	P		Deviance	P		Deviance	P		Deviance	P		Deviance	P		Deviance	P		Deviance	P
Vegetation
Population size	–			0·28	0·56		1·49	0·22		1·39	0·24		0·001	0·09		7·31 × 10−4	0·17		0·001	0·08		6·67 × 10−4	0·17
Area 1954	–			6·21	<0·001	↑	5·52	0·02	↑	4·51	0·03	↑	0·002	0·02	↑	6·57 × 10−4	0·20		0·002	0·02	↑	6·72	0·17
Connectivity 1954	–			0·48	0·44		0·31	0·58		0·11	0·74		4·21 × 10−4	0·34		0·001	0·07		4·95 × 10−5	0·27		9·17 × 10−4	0·10
Area 2011	–			2·38	0·06		2·85	0·09		1·12	0·29		9·17 × 10−4	0·15		9·13 × 10−4	0·12		0·001	0·09		0·001	0·07
Connectivity 2011	–			0·11	0·73		2·02	0·16		1·21	0·27		6·96 × 10−5	0·71		6·94 × 10−4	0·08		1·19 × 10−4	0·61		6·81 × 10−4	0·09
Seed bank
Population size	–			1·87	0·08		2·55	0·11		1·11	0·29		0·002	0·08		8·71 × 10−5	0·76		0·002	0·06		6·61 × 10−6	0·94
Area 1954	–			0·41	0·45		2·5	0·11		3·68	0·06		0·001	0·22		0·001	0·26		0·001	0·2		0·001	0·26
Connectivity 1954	–			0·02	0·88		0·01	0·90		1·50	0·22		2·61 × 10−4	0·58		6·35 × 10−4	0·4		5·58 × 10−5	0·78		7·86 × 10−4	0·4
Area 2011	–			0·37	0·47		0·91	0·34		1·63	0·2		4·21 × 10−4	0·48		0·002	0·16		3·62 × 10−4	0·47		0·002	0·19
Connectivity 2011	–			0·11	0·70		0·05	0·82		0·02	0·89		9·98 × 10−5	0·74		1·10 × 10−4	0·73		1·76 × 10−4	0·64		1·01 × 10−5	0·92
Vegetation + seed bank
Population size	0·01	0·93		3·42 ×10−4	0·98		0·91	0·34		1·29	0·26		3·18 × 10−5	0·70		0·001	0·009	↓	2·45 × 10−5	0·71		0·001	0·01	↓
Area 1954	4·86	0·03	↑	5·87	<0·001	↑	5·13	0·02	↑	5·91	0·02	↑	6·69 × 10−4	0·04	↑	5·71 × 10−4	0·13		5·90 × 10−4	0·04	↑	6·83 × 10−4	0·12
Connectivity 1954	1·05	0·31		1·33	0·18		6·17 × 10–4	0·98		1·31	0·25		2·40 × 10−4	0·27		5·36 × 10−6	0·89		2·04 × 10−4	0·27		5·85 × 10−6	0·9
Area 2011	2·41	0·12		2·99	0·03	↑	1·76	0·18		1·72	0·19		3·07 × 10−4	0·2		3·17 × 10−4	0·28		2·70 × 10−4	0·20		3·61 × 10−4	0·28
Connectivity 2011	0·43	0·51		0·50	0·44		2·02	0·16		1·11	0·29		3·57 × 10−4	0·18		4·82 × 10−4	0·19		3·31 × 10−4	0·16		5·20 × 10−4	0·2
Population size +	–			–			–			–			–			6·92 × 10−4	0·03	↓	–			8·23 × 10−4	0·03	↓
Area 1954
Population size +	–			–			–			–			–			2·77 × 10−6	0·91		–			2·99 × 10−6	0·91
Connectivity 1954
Population size +	–			–			–			–			–			2·19 × 10−4	0·28		–			2·51 × 10−4	0·29
Area 2011
Population size +	–			–			–			–			–			1·60 × 10−7	0·98		–			1·10 × 10−6	0·95
Connectivity 2011

The results are shown for individual predictor variables (comparison with null General Linear Model) and additive predictor variable effects (comparison with the model with Population size as the fixed term).

Significance of model comparison is based on χ² test statistics following likelihood ratio testing.

Deviance represents the amount of explained variation compared with the null model.

A _t, total number of alleles in vegetation + seed bank, A, total number of alleles, H_E expected heterozygosity, F_is inbreeding coefficient, A_r, total number of rare alleles; Chromo, assumption of chromosome segregation at meiosis; Chromd, assumption of chromatid segregation at meiosis.

Arrows indicate the direction of the relationship.

Differences in genetic diversity below- and above-ground

Raw and rarified allelic diversity metrics (A, A_R) and expected heterozygosity (H_E) were significantly higher in the above-ground populations compared with the below-ground populations (Fig. 2), but there was no significant difference in the inbreeding coefficient (F_{IS, chromosome}, t = 0·19, P = 0·86; F_{IS, chromatid}, t = 0·12, P = 0·91). The number of alleles unique to below-ground individuals was consistently and significantly lower than the number of alleles unique to above-ground individuals (mean difference in number of unique alleles between above- and below-ground population across populations of 8·87 alleles, t = 5·21, P ≤ 0·001). Most alleles were shared between above- and below-ground populations [mean 72·75 % (± s.e. of 8·33)] compared with the number of alleles unique to above-ground individuals [mean 19·52 % (± 7·38); mean difference of 39·2 alleles, t = 14·28, P ≤ 0·001] and below-ground individuals [7·73 % (± 4·00); mean difference of 48·07 alleles, t=23·00, P ≤ 0·001] (white boxplots, Fig. 3). In other words, the observed seed bank populations on average contained approx. 80 % (72·75 % shared alleles+7·73 % unique seed bank alleles) of the genetic diversity present in the total population. The comparison of the randomized vs. observed genetic diversity partitioning confirmed the observed genetic diversity partitioning to be significantly different from random (Fig. 3). The observed numbers of shared alleles and those unique to the below-ground populations were lower than expected under random genetic diversity partitioning. The observed numbers of alleles unique to the above-ground populations was higher than expected under random partitioning. The MANOVA showed that the genetic diversity partitioning across above- and below-ground populations was best explained by the 1954 landscape features (Pillai’s trace=0·69, P=0·003). The number of shared alleles (z = 0·57, P=0·46) and alleles unique to below-ground populations (z = 0·06, P=0·80) were not significantly affected by 1954 habitat patch area. The number of alleles unique to above-ground populations, however, significantly declined with 1954 habitat patch area (z=9·62, P=0·008; Fig. 4).

Fig. 2.

Boxplots comparing genetic diversity metrics (number of alleles, number of rare alleles and expected heterozygozity) between the sampled above-ground (Vegetation) and below-ground (Seed bank) Campanula rotundifolia populations, using both the raw data [white boxes; median, 25th and 75th percentile (box) and 1·5 × interquartile range (whiskers)] as well as the data corrected for differing above- and below-ground sampling sizes via rarefaction (grey boxes) for the allelic richness metrics. Results (t- and P-value) of the paired t-tests are shown per diversity metric. Note that the lines in the boxplot on Number of rare alleles (<2·5 %) are true boxplots, though with very limited variation around their means.

Fig. 3.

Partitioning of genetic diversity (total of number of alleles) between above-ground (Vegetation) and below-ground (Seed bank) Campanula rotundifolia populations. White boxes [median, 25th and 75th percentile (box) and 1·5 × interquartile range (whiskers)] indicate the observed partitioning of genetic diversity. The grey boxes represent the randomized partitioning of genetic diversity, i.e. the mean genetic diversity partitioning per population after randomly reassigning all above- and below-ground individuals of a population 999 times as being either an above- or below-ground individual of that population, calculating how the total number of alleles across all loci was partitioned per run. Results of the paired t-test (Δalleles = mean difference in number of alleles, t- and P-value) show the significant differences between the randomized and observed genetic partitioning.

Fig. 4.

Partitioning of genetic diversity (total of number of alleles) between above- and below-ground Campanula rotundifolia populations as a function of 1954 patch area. Grey circles indicate the number of shared alleles. Filled and open circles show the number of alleles unique to above-ground and below-ground populations, respectively. The regression line indicates the only significant relationship between 1954 patch area and number of alleles unique to above-ground populations (z = 3·81, P < 0·001).

Genetic extinction debt in the combined above- and below-ground gene pool

For the combined above- and below-ground populations, F_IS was related to population size, with both F_IS coefficients decreasing with increasing population size. All genetic diversity measures were not explained by connectivity but were significantly and positively related to 1954 habitat patch area (Table 1). Average number of alleles was positively related to 2011 habitat patch area, though 2011 habitat patch area explained only half of the deviance explained by 1954 patch area (Table 1).

Genetic composition as a function of historical and present landscape structure

Overall genetic differentiation between populations was low, averaging 0·03 (range: 0·01–0·06; full F_ST matrix, see Supplementary Data Table S3). The full RDA significantly explained 34·58% of the compositional genetic variation, though forward selection identified above-/below- and 1954 connectivity as the optimal set of environmental predictors for a reduced RDA. The reduced RDA significantly explained 27·98 % of the variation in genetic composition (Fig. 5), with restricted Monte Carlo permutation testing establishing the significant predictive power of above-/below- (F = 7·98, P < 0·001) and 1954 connectivity (F = 2·51, P = 0·014). The difference in the level of compositional genetic variation, quantified as multivariate homogeneity of variance, was significantly larger in the below-ground populations compared with the above-ground populations (F = 31·8; P < 0·001; see also group spread in Fig. 5).

Fig. 5.

Reduced redundancy analysis (RDA) of the F_ST pairwise genetic distance matrix based on genetic composition of the 30 above- and below-ground Campanula rotundifolia populations. Restricted Monte Carlo permutation tests indicate that the constraining predictor variables (above-/below-ground population and Connectivity 1954) significantly explain compositional genetic variation in the sampled Campanula rotundifolia populations. Group centroids of the above- and below-ground populations are marked with a cross.

DISCUSSION

Genetic extinction debt

In contrast to earlier studies (Honnay et al., 2006; Helm et al., 2009), our results indicate a genetic extinction debt in fragmented above-ground populations of C. rotundifolia. Similar to the findings of Münzbergová et al. (2013), the 1954 landscape configuration in terms of habitat patch area and connectivity proved important in explaining contemporary population genetic diversity and composition, respectively. Both allelic richness and expected heterozygosity in the above-ground populations were significantly related to the 1954 habitat patch area, with the historical landscape pattern reflected in the frequency of alleles unique to the above-ground populations. Historically, higher levels of gene flow between larger and better connected habitat patches probably maintained high levels of genetic diversity and variation (Leimu et al., 2006; Honnay and Jacquemyn, 2007), possibly in part having limited population differentiation to the low levels still observed today (Matter et al., 2013). In the current landscape, habitat fragmentation is expected to erode genetic diversity and compositional variation through both progressively reducing population size (Young et al., 1996; Honnay et al., 2006), and limiting seed and pollen flow between populations (Bossuyt, 2007). The relationship between current population genetic diversity and habitat patch area in 1954 thus indicates that genetic diversity is greater than expected given current habitat patch sizes. Consequently, extant levels of both population genetic diversity and genetic composition in our study area might reflect habitat loss, isolation levels and associated genetic bottlenecks prior to 1954, including the rapid loss of semi-natural grassland in the 1930s following the outfield grazing ban (Cousins and Eriksson, 2008), rather than recent landscape changes.

Seed banks and the genetic extinction debt

To explore the role of the seed bank in building up a genetic extinction debt, we first established whether seed banks stored a significant share of the genetic diversity present in the above-ground populations. Our data support the storage of a relatively high amount of genetic diversity in the seed bank, averaging approx. 80% of the total gene pool, with 8% private alleles. Overall, genetic diversity was significantly lower in the below-ground populations, with the difference becoming smaller after rarefaction. This lower overall and private below-ground diversity may partially be attributed to spatial sampling scale differences. The wider above-ground sampling scale led to individuals being sampled both inside and outside of the focal plot. As the focal plot hence only sampled a small spatial sub-set of the seed bank, this would cause both a lower overall below-ground diversity and an overestimation of private above-ground diversity, the latter proven by the randomization procedure. The question remains, however, of whether the observed private seed bank alleles are true numbers of private alleles or if they might be a potential overestimation or even an artefact of the sampling scale differences. Private seed bank alleles may indeed be present above-ground, yet they may have been missed as not all above-ground individuals in the focal plot were sampled. The randomization procedure also tentatively suggests that our sampling design may possibly underestimate private below-ground diversity. This result seems logical as only a sub-set of all potentially emerging seedlings across the entire extent of the population was sampled. A last argument underlining our sampling procedure to have adequately captured the genetic structure of above- and below-ground populations is the higher observed variation in genetic composition in below- vs. above-ground populations. This is hard to reconcile with an over-riding effect of sampling scale differences on the observed partitioning of genetic diversity. As above-ground populations were sampled at a wider scale than the focal plot sampling in below-ground populations, strong sampling artefacts would result in exactly the opposite pattern of higher above-ground vs. below-ground variation in genetic composition.

Below-ground populations thus contained an average of 8 % of alleles not present in the above-ground populations, in line with Honnay et al. (2008). The 8 % private below-ground diversity represents a conservative, lower estimate of the potential seed bank contribution to above-ground genetic processes. The presence of private alleles is plausible, given the persistent nature of C. rotundifolia seeds, with private alleles coming either from extinct individuals, from individuals selected under different environmental conditions or from mating events when the population was larger (Ottewell et al., 2011). Thus, although the C. rotundifolia seed bank may only contain approx. 80 % of the total gene pool, the seed bank may potentially re-introduce an average 8 % of alleles unique to the seed bank into above-ground populations. The seed bank may as such buffer minor genetic diversity losses, as observed earlier by Ottewell et al. (2011) and Roberts et al. (2014), potentially safeguarding genotypes selected for under different environmental conditions.

No clear relationship between genetic diversity and either the historical or current landscape was found in the below-ground populations. This was surprising, given that they are descended from past and/or current above-ground populations. Multiple processes may explain the absence of a landscape pattern in the below-ground genetic diversity. First, not all genetic diversity from above-ground populations is introduced into the persistent seed bank due to variation in seed production, primary and secondary seed predation and differential herbivory between sampled populations and individuals. Such variation among individuals in their seed bank contribution accords well with the observed genetic divergence between the above- and below-ground populations. Secondly, this genetic divergence between above- and below-ground populations may also point towards differential recruitment from the seed bank (Honnay et al., 2008), favouring the presence of specific genotypes in the above-ground populations, giving rise to the observed above-ground patterns. Thirdly, genetic diversity may be lost from the seed bank through differential seed senescence rates due to abiotic differences between sites (Abedi et al., 2014), at rates faster than the generation of genetic diversity through local sexual recombination and/or gene flow from adjacent populations. A final explanation may be that C. rotundifolia’s mean 5year seed longevity results in one of the following: (1) too few long-lived seeds (>5 years) are present either to capture or to shape the below-ground extinction debt pattern; or (2) a quick loss of seed viability (<5years) may rapidly erase the pattern, even though the above-ground extinction debt pattern is regularly imprinted on the seed bank via the seed rain. The absence of a historical landscape pattern in the below-ground gene pool does not, however, exclude the possibility of a minor seed bank contribution to the genetic extinction debt, as the above-ground re-introduction of private belowground alleles and genotypes at a rate of one individual per generation may already suffice to mitigate the population genetic consequences of habitat fragmentation (Young et al., 1996; Živković and Tellier, 2012).

To test the latter directly and address the seed bank’s role in causing the genetic extinction debt, we related the combined, total genetic diversity of the above- and below-ground populations to the historical and present landscape configuration. The inbreeding coefficient (F_IS) only responded to historical habitat patch area when the entire gene pool was considered, suggesting that in the long term, small populations will experience increased levels of inbreeding. Regardless, the impact of the historical landscape configuration on population genetic diversity increased only slightly relative to the patterns observed in above-ground populations. This suggests that the C. rotundifolia seed bank alone cannot explain the observed genetic extinction debt. Yet, a persistent seed bank may extend the populations’ gene pool (Lundemo et al., 2009), increase the effective population size (Nunney, 2002; Falahati-Anbaran et al., 2011) and small recruitment rates of long-lived seeds buffer downward demographical responses to loss of genetic diversity (Young et al., 1996; Živković and Tellier, 2012). In sum, the seed bank may potentially mitigate the loss of alleles through genetic drift to a minor extent via its limited number of private alleles. However, this process appears insufficient to give rise to the genetic extinction debt. Other, additional mechanisms are thus probably also at play.

Other characteristics of C. rotundifolia that probably contributed to the build-up of a genetic extinction debt include its relatively long generation time (Stevens et al., 2012) and its tetraploidy (Böcher, 1966). Perenniality and limited clonal growth in C. rotundifolia result in longer generation times and few elapsed generations since fragmentation (Young et al., 1996; Aguilar et al., 2008; Münzbergová et al., 2013), delaying reductions in genetic variation due to random genetic drift (Aguilar et al., 2008). Polyploidy delays genetic diversity losses and inbreeding (Moody et al., 1993; Barringer and Geber, 2008). This variety of life history traits, rather than the presence of a genetically diverse seed bank alone, probably explains why neither genetic diversity nor genetic composition mirrors the current landscape, and why C. rotundifolia successfully survives in strongly fragmented landscapes (Lindborg et al., 2005). The varied trait set seems to make our study species well equipped to deal with the consequences of habitat fragmentation.

CONCLUSIONS

To understand the mechanisms behind species extinction debts in fragmented landscapes is a crucial aim within conservation biology. Still, temporal delays in genetic processes as a mechanistic base to postponed species extinctions are rarely considered, although genetic erosion contributes to species extinctions. Our results underline that these temporal delays in genetic processes do exist, uncovering the existence of a genetic extinction debt in C. rotundifolia. Thus, C. rotundifolia may experience further losses of genetic diversity, becoming ever more susceptible to local extinction (Spielman et al., 2004), despite genetic buffering mechanisms such as polyploidy, long generation times via perennial, clonal individuals and a genetically diverse seed bank containing some private alleles. Our estimates of below-ground genetic diversity suggest that persistent seed banks may be one of the potential mechanisms slowing down genetic erosion and differentiation to create a genetic extinction debt. However, the impact of the seed bank on genetic processes is probably too limited to counter the effects of habitat fragmentation on the population genetic structure of C. rotundifolia.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Table S1: complete overview of the population and landscape data as well as the summary statistics of the raw genetic diversity metrics for each of the 15 sampled populations. Table S2: Spearman rank correlation and associated level of significance (P-values) between the main GLM predictor variables. Table S3: pairwise F_ST distance matrix indicating the pairwise level of genetic differentiation between both above- and below-ground populations as well as different patches. Figure S1: visual representation of the correlations between 1954 and 2011 habitat patch area and connectivity, respectively.