1Habitat fragmentation, which reduces the size and increases the isolation of populations, is a major threat to biodiversity. For Phyteuma spicatum, a self-incompatible, rare understorey herb in deciduous forests of north-western Germany, I tested the hypotheses that: (i) fitness (in terms of reproductive success) is reduced in small or isolated populations, (ii) reproduction in small populations is reduced by pollen limitation and (iii) genetic effects cause fitness reductions in small populations.
2I compared the reproductive success of plants of Phyteuma in 14 populations of different size and degree of isolation. Seed production was, as predicted, positively related to population size but was also influenced by plant size, although not by population isolation, density or habitat quality.
3I performed supplemental hand-pollinations in 10 of the 14 populations using pollen from the same population (test for pollen quantity) or from another large population (pollen quality). The proportional difference in seed production between hand-pollinated plants and open-pollinated controls increased with decreasing population size, indicating pollinator limitation of reproduction in small populations. There was no difference between the two hand-pollination treatments, suggesting that a sufficient number of cross-compatible mates was available even in the smallest populations.
4Progeny from the 14 populations were grown for 32 weeks in a common environment. There was no effect of population size on germination, but final seedling survival was positively related to population size, and this relationship was more pronounced in the glasshouse than under more favourable growing conditions in a common garden. Genetic effects may thus reduce fitness (here measured in terms of survival) in plants from small populations, making them more susceptible to environmental stress.
5The results suggest that both reproduction and offspring performance may be reduced in small populations even of long-lived species such as Phyteuma spicatum. Different processes, such as pollen limitation and genetic deterioration, may interact and affect local population dynamics and the persistence of species in fragmented landscapes.
Habitat fragmentation is considered to be one of the major threats to biological diversity (Eriksson & Ehrlén 2001; Oostermeijer 2003). In large parts of Central Europe deciduous forests have been highly fragmented for hundreds of years (e.g. Wilcove et al. 1986; Kelm 1994) as a consequence of habitat loss as natural landscapes have been converted into intensively used agricultural land or planted with conifers (Haila 1999). In such landscapes, obligate forest herbs are confined to the remnants of a formerly more or less continuous forest cover, making them particularly susceptible to edge effects and deteriorating environmental conditions (Murcia 1995; Jules 1998) and causing many plant populations to be small and isolated (Jacquemyn et al. 2002).
Habitat fragmentation may negatively affect plant fitness and lower population viability by a number of different mechanisms, as recently reviewed by Lienert (2004). In general, small populations are more vulnerable to the effects of environmental, demographic and genetic stochasticity than large populations (Shaffer 1987) and are thus expected to face an elevated risk of local extinction (Matthies et al. 2004). Small populations may be subject to increased inbreeding and to the loss of genetic variation due to genetic drift (e.g. Ellstrand & Elam 1993; Young et al. 1996). This may lead to a reduction in fitness of individuals, i.e. inbreeding depression (Fischer & Matthies 1998b; Lienert et al. 2002; Vergeer et al. 2003), and, in the long term, limit the evolutionary potential of populations to respond and adapt to changes in the environment (Huenneke 1991).
Biotic interactions, such as plant–pollinator mutualism or the dispersal of seeds, may be disrupted by fragmentation (Olesen & Jain 1994; Kwak et al. 1998; Santos et al. 1999). Genetic divergence between populations of many species implies that interpopulation gene flow via pollen or seeds is low, especially in human-dominated landscapes (e.g. Raijmann et al. 1994; Fischer & Matthies 1998b). On the one hand, this prevents the amelioration of the deleterious effects of genetic deterioration; on the other hand, however, the possibility for outbreeding depression is avoided. Small populations may also be less attractive to pollinators than large populations and thus suffer pollen limitation (Sih & Baltus 1987; Jennersten 1988); this effect may be especially severe in obligate outcrossing species that also depend on the presence of a sufficient number of different mating types. S-allele diversity and cross-compatibility may be reduced in small populations owing to genetic drift (e.g. Byers & Meagher 1992). Both insufficient pollen quantity and poor pollen quality have been suggested to be the cause of reduced reproductive success in plants of small populations (Jennersten 1988; Byers 1995; Ågren 1996; Tomimatsu & Ohara 2002; Waites & Ågren 2004). The effects of population size and isolation on pollination success may also interact: Groom (1998), for example, found that small populations of an annual herb suffered reproductive failure as a result of lack of effective pollination when critical levels of isolation were exceeded, whereas sufficiently large populations attracted pollinators regardless of their degree of isolation.
There is increasing evidence that forest plant species with specific biological attributes, including high habitat specificity, low seed production, restricted dispersal capacity or small size, are especially adversely affected by changes in land use, and particularly by habitat fragmentation, as demonstrated by landscape-scale analyses of patterns of patch occupancy (e.g. Dupré & Ehrlén 2002; Verheyen et al. 2003; Kolb & Diekmann 2004, 2005). Furthermore, it has been suggested that species lacking mechanisms to compensate for the effects of insufficient pollinator service, such as clonal propagation or self-pollination, may be particularly susceptible to fragmentation (Bond 1994). To date, there have been few attempts to study fragmentation effects on fitness components in herbs of temperate deciduous forests (e.g. Jacquemyn et al. 2002; Tomimatsu & Ohara 2002), but these support the notion that non-clonal and self-incompatible species may be most adversely affected by reductions in population size or increased isolation.
I studied the effects of forest fragmentation on plant fitness of Phyteuma spicatum, a self-incompatible, rare understorey herb in deciduous forests of north-western Germany, by combining experimental studies with detailed field records that were collected on a landscape scale. Fitness, commonly defined as the proportionate contribution of individuals to future generations, cannot be measured directly (Silvertown & Charlesworth 2001). However, the term is also used more generally to refer to characteristics that can be expected to be positively correlated with fitness, such as the number of seeds produced or the survival of offspring (e.g. Fischer & Matthies 1998a; Hooftman et al. 2003; Lienert 2004). For simplicity, these parameters are hereafter summarized as fitness. I compared the reproductive success of several populations to test the hypothesis that population fitness (recorded as average individual fitness) is reduced in smaller or more isolated populations. A pollination experiment was then used to test whether reproduction in small populations is reduced by pollen limitation and, if so, whether pollen quantity (i.e. insufficient pollen due to pollinator limitation) or quality (i.e. lack of cross-compatible pollen due to loss of S-allele diversity) is responsible. I also conducted a glasshouse and common garden experiment to assess the performance of progeny from the field populations and to test whether genetic effects cause fitness reductions in plants of small populations.
study area and study species
The study area is situated in the agricultural landscape of north-western Germany (Fig. 1), an area mostly covered by Pleistocene deposits of the Saale glaciation. The landscape is flat to slightly undulating, with elevations varying between 10 and 40 m a.s.l. Forests are highly fragmented and cover c. 13% of the landscape, of which only 25% are deciduous hardwood forests (Kolb & Diekmann 2004). Many obligate forest herbs are therefore restricted to relatively small and isolated forest patches.
Phyteuma spicatum L. (Campanulaceae) (hereafter referred to as Phyteuma), a species endemic to Central and Atlantic Europe (Wheeler & Hutchings 2002), is common in southern Germany but less so in the north (for a distribution map see http://www.floraweb.de). In the study area, Phyteuma is restricted to fresh and moist, base-rich deciduous hardwood forests and is relatively rare, only occurring in c. 10% of all such forests (Kolb & Diekmann 2004).
Phyteuma is a polycarpic, perennial, non-clonal hemicryptophyte that produces annual rosettes of basal leaves and one to many inflorescences on upright stems. Normally, Phyteuma remains a vegetative, non-reproductive plant throughout its first 2 years and reaches sexual maturity during the third year, although sometimes also later (Wheeler & Hutchings 2002). The species is mainly pollinated by bumblebees and the flowers are protandrous (Wheeler & Hutchings 2002; my personal observation). Spontaneous autogamy and geitonogamy have been observed, but only very low numbers or no seeds were produced (Huber 1988; my unpublished data). Gametophytic self-incompatibility has been presumed to be responsible for the low levels of seed production following self-pollination (Huber 1988). Phyteuma lacks aids to seed dispersal (Wheeler & Hutchings 2002), and most dispersed seeds land close to the parent plant, although a few seeds may travel up to 7 m (Maier et al. 1999).
population size and isolation
Within the study area, I selected 14 (of the 19 existing) populations to cover a wide range of population sizes and degrees of isolation (Fig. 1, Table 1). Populations with very few (≤ 3) individuals were excluded. Population size was determined as the number of flowering individuals during peak flowering (24 May–6 June 2003). In a few cases I encountered more than one population per forest patch. Because plants of Phyteuma growing about 90 m from larger concentrations of plants were found to produce no seeds (Wheeler & Hutchings 2002), I used a threshold distance of 100 m to separate individual populations from each other. However, populations were usually separated by several hundreds of metres (Table 1). Population isolation was quantified as the distance to the next population. I also tested the mean distance to the next three closest populations, but as the two isolation measures were significantly correlated (Pearson correlation, r = 0.630, P = 0.016, n = 14), this did not considerably alter the results.
Table 1. Population size, isolation, density and environmental variables of the 14 sampled populations of Phyteuma spicatum
Populations included in the pollination experiment are marked with an asterisk.
Number of flowering individuals.
Distance to the next population (in m).
Number of flowering individuals m−2 (see Methods).
Photosynthetic photon flux density of photosynthetically active radiation.
Population density was quantified as the number of flowering individuals m−2, based on the number of flowering individuals and the approximate area over which the population was distributed. The distribution of each population was delineated in the field and marked on 1 : 10 000 forestry maps. The area of the minimum convex polygon was then determined using a geographic information system (ArcView GIS 3.3, Environmental Systems Research Institute, CA, USA).
reproductive success in natural populations
Within each population, 25 flowering individuals (or all individuals when n < 25) were selected for the measurement of plant traits and marked with aluminium tags. In the small populations (n ranging between 35 and 217; Table 1), individuals were chosen completely at random. Here, plants were distributed over a relatively small area that was fairly homogeneous in terms of edaphic conditions and light environment. However, in the large populations habitat quality usually varied considerably. In order to obtain meaningful measurements of the environmental conditions, i.e. to obtain estimates representative for all sampled plants within a population, I randomly chose a homogeneous area for sampling similar in size to the area covered by plants in small populations, but excluding forest edges, gaps or other atypical habitats. The area usually varied between 25 m2 and 200 m2 in size, depending on population density. Within this area, individual plants were chosen at random.
I recorded a number of morphological and reproductive traits for each selected plant. Shortly after peak flowering (16–20 June 2003) I determined the number of rosette and cauline leaves broader than 1 cm and the width and length of the three largest leaves for an estimate of total leaf area [number of leaves × (mean width × length of the three largest leaves)] as well as the number and height of all inflorescences. During the first week of July, at the time of seed maturity, I collected the inflorescences of each selected plant. Because the time of seed maturity varied between the individual plants, I visited all populations several times in order to be able to collect inflorescences with ripe but still-closed seed capsules. In this way, all seeds of each plant were captured. Seeds were air-dried and stored in a dry and dark place at room temperature. For each plant, I determined the total number of seed capsules and seeds as well as mean seed mass (mean mass of 50 randomly chosen seeds per plant). The mean number of seeds per capsule was calculated as the total number of seeds per plant divided by the number of capsules per plant. The total seed mass per plant was estimated as the product of the number of seeds per plant and mean seed mass. A few plants could not be relocated or were predated by animals; these were excluded from analysis (Table 1).
In each population, five 4-cm-deep soil cores were collected from below the litter layer and pooled (18–19 July 2003). Each sample was air-dried to constant mass and passed through a 2-mm sieve. All samples were analysed for pH (determined from a solution of 10 g of soil and 25 mL of 0.01 m CaCl2 with a standard glass electrode) and carbon and nitrogen (in %; elemental analyser EuroEA 3000, HEKAtech, Germany). Light intensity was measured as photosynthetic photon flux density (PPFD) of photosynthetically active radiation (µmol s−1 m−2; LI-COR Quantum Sensor, USA) at 10 locations spaced regularly across the sampled area of each population and simultaneously outside each forest patch in the open (15-s averages each), on days with an overcast sky. The measurements were averaged and expressed as PPFDforest/PPFDopen (×100%), for a measure of relative light intensity.
To test the hypothesis that a positive correlation between population size and reproductive success is due to a higher degree of pollen limitation in small than in large populations, I performed supplemental hand-pollinations in 10 of the 14 study populations (Table 1) in 2004. In each population, up to 36 single-inflorescence plants were selected within an environmentally homogeneous area that harboured enough individual plants. The plants were marked and assigned to one of three treatments (n = 12 per treatment and population, unless fewer plants were available): (1) supplemental hand-pollination with pollen from at least two different donor plants located c. 3–15 m away (when possible) from the recipient plant within the same population (test for pollen quantity), (2) hand-pollination with pollen from at least two different donor plants from another large population in the study area (test for pollen quality) and (3) open-pollinated control. In one small population, only treatments (2) and (3) could be applied. Because the plants in a population begin to flower at different times, I adopted the following procedure for selecting the experimental plants: each time I visited a population, I selected those plants ready for pollination and assigned these to the different treatments by order of sets of replicates (i.e. the first three plants were each assigned to one of the three treatments, then the second group of three plants, etc.). The plants had to be in a similar stage of development and had to have inflorescences of comparable size.
For the hand-pollinations, pollen-covered styles were collected from the donor plants, stored in an Eppendorf tube and then wiped over the receptive stigmas of the recipient plants until the stigmas were visibly covered with pollen. Because the flowers open sequentially in the inflorescence from the base towards the apex, I hand-pollinated every plant at least 3–4 times, approximately every 2–3 days, usually using different pollen donors at each hand-pollination event. For each experimental plant, I determined the height of the inflorescence, the number of seed capsules and seeds per plant, the mean number of seeds per capsule and mean seed mass. Population density was quantified as the maximum number of inflorescences m−2 at the site where the experimental plants were located. I assumed that bumblebees, the most important pollinators for Phyteuma (Wheeler & Hutchings 2002; my personal observation), would be most attracted to the patch with the maximum density of flowering plants and that, once attracted to this patch of plants, they would also move to other plants in the same area. A number of plants were predated or destroyed by animals and had to be excluded from analysis, giving a total of 262 experimental plants.
germination and offspring survival
To analyse the effects of population size on offspring fitness, I randomly selected 20 mother plants (six in the smallest population) from each of the 14 populations. To analyse the effects of population size on germination, I randomly chose 100 seeds from each population using five seeds from each plant. For each population, mean seed mass was determined as the total mass of all seeds divided by 100. The seed samples were placed on filter paper in Petri dishes. The dishes were then watered with 8 mL of a solution of gibberellic acid (0.5 g GA3 L−1) to break dormancy and kept in a growth chamber at a temperature and light regime of 14 h light/20 °C and 10 h dark/10 °C. The number of germinated seeds (seeds with emerged radicle) was determined daily until no more seeds germinated (for c. 4 weeks). Total germination was expressed as the total number of germinated seeds (in %). The germination rate was expressed as the slope of the regression line through the germination curve from day 10 to day 20 after the start of the experiment.
To analyse the effects of population size on seedling survival, 40 seeds (or fewer when n < 40) were randomly chosen from each of the selected mother plants (n = 20), placed in Petri dishes (7 January 2004), treated with gibberellic acid and kept in a growth chamber as described above. Seeds from one mother plant are subsequently referred to as a seed family. After 7 days, 20 seeds of each seed family were transferred into each of two replicate 10-cm-diameter pots filled with a 1 : 1 mixture of potting soil and sand, giving a total of 531 pots. To avoid location effects, pots were arranged in a randomized block design in the glasshouse. There were 20 blocks, each with one seed family of each population, i.e. both pots of one seed family were placed in the same block. After 4 weeks, the number of emerged seedlings was recorded in each pot, for another measure of germination. The number of surviving seedlings in each pot was counted 6, 10, 14, 18, 25 and 32 weeks after the start of the experiment. In May 2004, after 18 weeks, one pot of each replicate was transferred to the experimental garden, while the other pot of the replicate was kept in the glasshouse. After 32 weeks of growth, at the end of the experiment, final survival was recorded; seedlings were still too small to harvest and determine above-ground biomass.
To analyse whether population size and isolation had an influence on reproductive success independent of effects of population density, environmental variation between sites and plant size, mixed-model regressions were used (Type III sums of squares), with the number of seeds per capsule and per plant, seed mass and total seed mass as the dependent variables. I included plant size, habitat quality and population density in the regression models, because all are known to affect fecundity (e.g. Kunin 1997; Jacquemyn et al. 2002; Vergeer et al. 2003; Brys et al. 2004). Population size, isolation, density, soil pH, C : N ratio, relative light intensity and plant size were entered as fixed effects, while population ID was entered as a random factor. For model selection, backwards elimination of predictor variables and Akaike's Information Criterion (AIC) were used. Models with the smallest AIC were considered to be best, meaning that the final models could include non-significant variables if their removal was associated with an increase in AIC.
To test whether the relationship between population size (as determined in 2003) and reproductive success differed between the open-pollinated controls and plants receiving supplemental pollen, I used a general linear model (Type I sums of squares) on treatment means for each population of number of seeds per capsule, number of seeds per plant and mean seed mass. Pollination treatment was treated as a fixed factor and population size as covariate. To account for possible confounding effects of population density and plant size (expressed as height of the inflorescence and number of seed capsules) on pollination and fecundity, I also included these as covariates. The number of seed capsules was used as a surrogate for the length of the inflorescence; both parameters were strongly correlated (tested across all pollination treatments; r = 0.906, P < 0.001). The effects of population size and density were tested against the residual variation among populations, all other effects were tested against the residual. Orthogonal contrasts were used to test for differences between (i) hand-pollinated vs. open-pollinated plants, and (ii) plants pollinated with pollen from the same population vs. plants pollinated with pollen from another population. To test whether the effect of supplemental hand-pollination on seed production decreased with increasing population size, I calculated a standardized index of pollen limitation (L) for each population and correlated L with population size. Following Larson & Barrett (2000), the index was calculated as L = 1 − (Po/Ps), where Po is the mean number of seeds produced per capsule or plant for open-pollinated controls and Ps is the equivalent mean measure for plants that received supplemental hand-pollination.
To test whether germination or final survival of seedlings differed between populations of different size, mixed-model regression was used (for details of analysis see above). Population size and seed mass were entered as fixed effects and population ID as a random factor. Seed mass was included to account for its potential effects on offspring performance. For each seed family, final survival was expressed as the number of seedlings that had survived after 32 weeks of growth divided by the number of seedlings that had germinated after 4 weeks.
Variables were transformed (arcsine square root for proportional data or log or square root for all other data) prior to analysis if necessary to achieve normality and homoscedasticity of residuals. All analyses were performed with SPSS 11.5 (SPSS, Chicago, IL, USA).
effects of population size and isolation on reproductive success in natural populations
Reproductive success was negatively affected by decreased population size (Table 2), consistent with the hypothesis that plant fitness would be reduced in small populations. The number of seeds per capsule, the number of seeds per plant and total seed mass were positively correlated with population size (Pearson correlations, each r > 0.7 and P ≤ 0.003; Fig. 2), but were also strongly determined by plant size (Table 2). Large individuals produced more seeds than small individuals. However, there were no significant effects of soil pH, C : N ratio, relative light intensity or population density on these population parameter values (Table 2). Mean seed mass, by contrast, was not affected by population size and tended to increase with decreasing light availability. Furthermore, plant size decreased with decreasing population size [r = 0.847, P < 0.001 (total leaf area), r = 0.654, P = 0.011 (mean height of the inflorescences), n = 14], but was largely unaffected by environmental conditions, population isolation or density [mixed-model regressions with only population size (total leaf area) or population size and density (mean height of the inflorescences) remaining as significant predictors in the final models]. Plants from dense populations were taller than plants from more sparse populations. Thus, the effect of population size on plant reproductive success appeared to be exerted also indirectly via a positive effect on plant size.
Table 2. Mixed-model regression analysis of population sizea, population isolation, population density, environmental variables (soil pH, C : N ratio, relative light intensity) and plant size (total leaf area and mean height of the inflorescences) on reproductive success in 14 natural populations of Phyteuma spicatum (n = 318 experimental plants). For each dependent variable, the best model is presented (see Methods for details of model selection). P-values < 0.05 are shown in bold type and < 0.1 in italics
Log10-transformed number of flowering individuals.
Variables were square root-transformed prior to analysis.
Contrary to the hypothesis that plant fitness would be reduced in isolated populations, reproductive success was not affected by distance to the next population (Table 2). The correlation between population size and isolation was not significant (r = −0.382, P = 0.177, n = 14), suggesting that there were no interacting effects of population size or isolation on fecundity. Furthermore, neither soil pH nor C : N ratio were significantly correlated with population size [r = −0.078, P = 0.791 (pH), r = −0.418, P = 0.137 (C : N), n = 14], indicating that these parameters did not differ between small and large populations (see also Table 1). However, population size decreased with increasing relative light intensities (r = −0.593, P = 0.025, n = 14).
Supplemental hand-pollination increased the number of seeds produced per capsule [General linear model, P (orthogonal contrast) = 0.055] and the number of seeds produced per plant (P = 0.013). However, there were no differences between the two pollination treatments [P (orthogonal contrasts) > 0.6 in both cases], indicating that pollen quality did not affect reproductive success. There was a (marginally) significant interaction between population size and the effects of pollination treatment on the number of seeds produced per capsule and per plant (Table 3). The standardized index of pollen limitation, reflecting the proportional difference in seed production between hand-pollinated plants and open-pollinated controls, was negatively correlated with population size, indicating that seed production was pollen-limited in small populations (Fig. 3). However, the relationship between population size and the magnitude of pollen limitation was only significant in terms of the number of seeds per capsule (Fig. 3a) and not or only marginally significant in terms of the number of seeds per plant (Fig. 3b). Supplemental hand-pollination did not affect mean seed mass (Table 3).
Table 3. General linear model of the effects of population sizea and pollination treatment (supplemental hand-pollination with pollen from the same population, hand-pollination with pollen from another large population, and open-pollinated control) on number of seeds per capsule, number of seeds per planta and mean seed mass. Population density and plant size (expressed as mean height of the inflorescences and mean number of seed capsules) were added as covariates to account for potential confounding effects on pollination and fecundity. See Methods for details of analysis. P-values < 0.05 are shown in bold type and < 0.1 in italics
Source of variation
Number of seeds per capsule
Number of seeds per plant
Mean seed mass
Mean height of inflorescences
Mean number of seed capsules
Pollination treatment × population size
germination and offspring survival
In the growth chamber, mean total germination across all populations was 95.8% (SE 1.8%). There was no significant effect of population size on total germination (Spearman rank correlation, rs = 0.231, P = 0.427, n = 14) or germination rate (r = 0.460, P = 0.098, n = 14). Likewise, there was no effect of seed mass on germination (P > 0.8).
Across all populations, 78.6% (SE 1.3%) of the seeds planted in the glasshouse and common garden experiment germinated and produced a seedling. During the course of the experiment, mean cumulative survival in the glasshouse decreased steadily (data not shown). Seedlings transferred to the common garden after 18 weeks of growth, however, showed little further mortality. In all populations, final survival was higher in the garden than in the glasshouse (Fig. 4), indicating that growing conditions were more favourable in the garden.
Again, neither the size of the population of origin nor seed mass affected germination (Table 4). Final survival after 32 weeks of growth, however, increased with increasing size of the population of origin (Table 4, Fig. 4), although this relationship was less strong under the more favourable growing conditions in the common garden (Table 4). In the glasshouse, the positive correlation between population size and survival steadily increased over time (data not shown). Seed mass had a marginally significant effect on final survival in the glasshouse, but did not affect seedling survival in the common garden (Table 4).
Table 4. Mixed-model regression analysis of population sizea and seed mass on germination and final survival of plants of Phyteuma spicatum grown from seed of 14 populations of different size (n = 266 seed families). All plants were grown in the glasshouse; after 18 weeks of growth one of two replicate pots of each seed family was transferred to the common garden. For each dependent variable, the best model is presented (see Methods for details of model selection). P-values < 0.05 are shown in bold type and < 0.1 in italics
Log10-transformed number of flowering individuals of the population of origin.
Variables were transformed to the arcsine of the square root prior to analysis.
Reproductive success was significantly reduced in small populations of Phyteuma, as also observed in a congener (Phyteuma nigrum, Boerrigter 1995) and numerous other plant species (e.g. Ågren 1996; Fischer & Matthies 1998a; Jacquemyn et al. 2002; Tomimatsu & Ohara 2002; Brys et al. 2004), and this pattern was not affected by environmental conditions (see below) or population density. Furthermore, plants of large populations were larger than plants of small populations, and large individuals produced more seeds than small individuals. Thus, there also appears to be an indirect positive effect of population size on plant reproductive success, namely via its effect on plant size. By contrast, population isolation did not influence reproduction, a result similar to that of Mavraganis & Eckert (2001). This is probably due to the fact that most populations are, in fact, truly isolated, so that it does not matter whether a population is several hundreds of metres or more than a kilometre away from the next closest population. In Phyteuma nigrum, the maximum distance between populations covered by foraging bumblebees was 230 m (Kwak et al. 1998), but in the study presented here, distances between populations were much larger, ranging from 300 to 4700 m (Table 1). Furthermore, gene flow through the dispersal of seeds is also very limited in P. spicatum (Maier et al. 1999; Wheeler & Hutchings 2002).
effects of environmental conditions
In principle, the reduced reproductive success in plants of small populations could be due to unfavourable environmental conditions, pollen limitation, genetic deterioration (i.e. increased inbreeding and loss of genetic variation due to genetic drift) or their interaction (e.g. Vergeer et al. 2003; Lienert 2004). The results of this study suggest that the reduced fecundity in plants of small populations was not caused by lower habitat quality. Neither edaphic conditions nor relative light intensity influenced reproductive success. Likewise, these parameters did not affect plant size, meaning that there was also no indirect effect of habitat quality on seed production. There was a weak negative effect of increased relative light intensity on mean seed mass, but this did not affect total seed mass. However, it is possible that important environmental variables have not been measured.
Soil pH and C : N ratio were not correlated with population size. However, population size appeared to be negatively affected by high relative light intensities. Lower habitat quality in forest fragments may thus negatively affect population size and potentially, albeit indirectly, plant reproductive success. Interpretations of the relationship between habitat quality and population size, however, need to be made with caution, as I measured habitat quality in large populations only in a limited area.
I found some evidence that the lower fecundity in plants of small populations may have been caused by pollen limitation. However, there was no difference between the two pollination treatments (pollen from the same or a different population). The increase in the number of seeds produced in plants of small populations following supplemental pollination indicates that seed production is limited by pollen quantity. Small populations are likely to be less attractive to pollinators than large populations and this may lower pollinator visitation, cause pollen to be limiting and reduce seed set (Sih & Baltus 1987; Jennersten 1988). The number of seeds per capsule is higher in larger inflorescences of Phyteuma (Wheeler & Hutchings 2002; my unpublished data), indicating that smaller concentrations of Phyteuma flowers may receive fewer insect visitors. Bumblebees, the most important pollinators for Phyteuma (Wheeler & Hutchings 2002; my personal observation), have been shown to pollinate more flowers and also to pollinate a higher proportion of flowers in large than in small populations (Sih & Baltus 1987). Furthermore, pollen limitation in small populations may reduce pollen competition and decrease the selectivity between gametes before and during fertilization, thereby causing higher seed abortion rates and lowering offspring fitness (see, for example, Colling et al. 2004 and references cited therein). The increased number of seeds in plants of small populations following supplemental hand-pollination may therefore have been partly due to reduced abortion. In several of the large populations, seed production was lower after supplemental pollination than after open pollination (Fig. 3), suggesting that more pollen donors, or pollen donors of higher quality, may have been involved in open pollination. However, a decline in seed production with increased pollen load may also be caused by pollen removal or stigma damage by pollen thieves (Young & Young 1992), which are likely to be more common in the larger populations.
In species with a self-incompatibility system, seed production may also be reduced by the absence of compatible mates. Reductions in population size may result in the loss of S-alleles (and thus in the number of different mating types) due to genetic drift, thereby reducing cross-compatibility and seed set (Byers & Meagher 1992; DeMauro 1993; Byers 1995; Vekemans et al. 1998). However, the fact that seed production did not differ between the intra- and interpopulation pollination treatments suggests that a sufficient number of cross-compatible mates was available, even in the smallest tested population (35 flowering individuals, Table 1). Mate availability tends to be large in species with a gametophytic incompatibility system [as suspected to be present in Phyteuma (Huber 1988)], even in populations with only 30 individuals (Vekemans et al. 1998). A breakdown of the self-incompatibility system in plants of small populations, i.e. a shift from predominantly cross-fertilization towards mixed mating, therefore appears to be relatively unlikely.
genetic effects of population size on plant performance
Although the reduced fecundity in plants of small populations may have been caused by pollen limitation, there is also some indication that this pattern is due to genetic effects. After 32 weeks of growth under common environmental conditions, I found a strong positive correlation between the size of the population of origin and the mean number of surviving seedlings, at least under the less favourable growing conditions in the glasshouse. Differences in performance of plants from different populations, when grown under common environmental conditions, most likely reflect genetic differences. However, non-genetic maternal effects may also influence offspring fitness, especially during the early stages of development (Roach & Wulff 1987). These effects may thus confound the ability to estimate the genetic basis of offspring performance accurately, even when plants were grown in the same environment. Large maternal effects have been demonstrated for seed size, which in turn may affect germination, seedling or even adult plant size (Roach & Wulff 1987). There is some evidence that maternal effects are less likely to be responsible for the observed differences in offspring performance between populations of different size. Seed size did not affect germination and was not significantly (or only marginally so) related to the final survival of seedlings. In addition, the negative effects of small population size on seedling survival became more pronounced with time, a result similar to that found by Fischer & Matthies (1998a) for Gentianella germanica. Although maternal effects may carry through to the early seedling stages, the genotype of the offspring begins to contribute significantly to the variation in performance at later seedling stages (Roach & Wulff 1987). However, I cannot completely exclude the possibility of maternal carry-over effects. Boerrigter (1995), for example, found for Phyteuma nigrum that progeny of a large meadow population from a site of good habitat quality performed significantly better when grown under controlled conditions than progeny of a small road verge population from a site of poor habitat quality.
The results thus give some support to the hypothesis that genetic effects may cause fitness reductions in plants of small populations. Negative genetic effects of small population size on offspring performance have also been shown experimentally for Gentianella germanica (Fischer & Matthies 1998a), Primula veris (Kéry et al. 2000), Carex davalliana (Hooftman et al. 2003) and Succisa pratensis (Vergeer et al. 2003). However, on the basis of these data, the effects of loss of genetic variation through genetic drift and increased levels of inbreeding or their interaction cannot be distinguished. Increased inbreeding due to matings among close relatives, for example, may contribute not only to the decreased survival of the offspring grown in the glasshouse, but also to the reduced plant size and seed production of the adult plants growing in natural populations (see Lienert 2004 and references cited therein). Given the long history of forest fragmentation in the study area (Kelm 1994), many of the Phyteuma populations have probably been isolated for several hundreds of years, and genetic deterioration may thus have occurred in the small populations.
The positive relationship between the size of the population of origin and the final number of surviving seedlings was less pronounced under the better growing conditions in the common garden. These results suggest that the negative effects of small population size become especially apparent when environmental conditions are unfavourable. Dudash (1990), for example, compared the relative fitness of self- and outcrossed progeny of Sabatia angularis between three different environments and found the magnitude of inbreeding depression to be greatest in the harshest environment. Offspring of plants of small populations thus may be less able to respond to changes in the environment and may be more susceptible to environmental stress. For example, Boerrigter (1995) and Kéry et al. (2000) observed a reduced plasticity in offspring derived from small populations of Phyteuma nigrum and Primula veris, respectively, as plants were less able to increase growth in response to increased nutrient availability. Similarly, Fischer et al. (2000) found a lower adaptive plasticity in response to competition for progeny of small populations of Ranunculus reptans.
The reduced reproductive success and offspring survival observed in plants of small populations can be explained by a combination of pollen limitation and genetic effects, although possible effects of deteriorating habitat quality and maternal influences cannot be completely excluded. However, irrespective of the underlying causes of the reduced performance of plants of small populations, it is clear that reductions in population size, in combination with a high degree of isolation, adversely affect population fitness of Phyteuma. Reduced seed production and offspring survival may be a first serious indication of reduced population viability, although the growth rates of populations may not necessarily be seed-limited (Oostermeijer et al. 1996). If local and regional population dynamics are negatively affected (e.g. by reduced seedling recruitment, reduced chances to colonize unoccupied habitat patches or decreased ability to respond to changes in the environment), populations may enter an extinction vortex (Gilpin & Soulé 1986), a downward spiral of ever-decreasing population size and plant fitness (via ever-increasing vulnerability to the effects of environmental and demographic stochasticity, loss of genetic variability, etc.) that eventually drives them to extinction. The fact that the negative effects of reduced population size are measurable even in relatively long-lived species such as Phyteuma should be of considerable concern for conservation management.
I thank Dirk Enters for research assistance, Werner Vogel, Angelika Trambacz, Maike Isermann, Christina Winter, Simone Böckmann, Tim Daake and Anne Schmidt for glasshouse support, Marion Ahlbrecht for glasshouse and laboratory support, Werner Wosniok and Thomas Hoffmeister for statistical advice and Martin Diekmann for discussion. Earlier versions of the manuscript were improved by the helpful comments of Peter Alpert, Martin Diekmann, Johan Ehrlén, Ove Eriksson, Michael Hutchings, Diethart Matthies and one anonymous referee.