Vincetoxicum hirundinaria Med. (Asclepiadaceae) is native to Europe and western Asia, and has been introduced to North America (Hultén & Fries 1986; DiTommaso et al. 2005). In Sweden, V. hirundinaria occurs along the north-western fringe of its wide palaearctic range (Hultén & Fries 1986; Donadille 1965). It is abundant on the large Baltic islands Öland and Gotland, and has a patchy distribution along the Baltic coast and inland along Lake Mälaren. The plant typically forms a short, branched rhizome, from which a dense tussock of above-ground shoots develop. Individual plants may produce from a few to more than 100 flowering shoots [median (range), 12 (1–120), n = 765 plants sampled in the study area in 1999]. In a given plant, the shoots are of roughly equal length and flower synchronously. The plant grows in a variety of conditions from sun-exposed rock ledges to shaded deciduous woods below cliffs, and along forest margins.
Flowering usually peaks in June, but may continue even into early September. The flowers are creamy white, hermaphroditic and arranged in cyme-like inflorescences, which develop from leaf nodes on the above-ground shoot. The anthers are fused to the gynoecium to form a gynostegium, and the pollen is packed in five pairs of pollinia. The results of controlled crosses suggest that V. hirundinaria has a late-acting self-incompatibility system, but that the frequency of self-fertile individuals may be relatively high (Leimu 2004). High inbreeding coefficients have been documented in V. hirundinaria populations in south-western Finland, which is consistent with a mixed-mating system (Leimu & Mutikainen 2005). The flowers are visited by a multitude of insects, the vast majority of which are nectar thieves. The large empidid fly Empis tesselata, which has an extended flight period and is likely to be one of the main pollinators in the area (C. Solbreck, pers. obs.), is a very good flier and can fly long distances (Chvála 1994). Most fruits mature in July and August, and fruit production varies strongly between years (Solbreck & Sillén-Tullberg 1986; Solbreck & Ives 2007). Each fruit contains about 20 wind-dispersed seeds.
The larva of the tephritid fly E. connexa (Fabr.) may destroy close to 100% of the seeds (Solbreck 2000; Solbreck & Ives 2007). The adult fly oviposits on developing fruits, and is a strong flier that can easily move between patches of its host plant in the study area (Solbreck 2000). The lygaeid bug, Lygaeus equestris (L.), is both a pre- and a post-dispersal seed predator and a good flier (Solbreck 1995). It is sometimes abundant, but its impact is small compared to that of E. connexa. The plant contains toxic substances (Tullberg et al. 2000) and is not fed upon by grazing mammals in the study area.
This study was conducted at Tullgarn (58°57′ N, 17°36′ E) on the coast of the Baltic Sea, about 50 km south-by-south-west of Stockholm, Sweden. This area receives relatively little rain. During the 3 years of study (1999–2001), precipitation during June–August ranged from 130 to 232 mm (Table 1). The presence of V. hirundinaria has been mapped in the 3 × 4 km study area since the late 1970s. The species is distributed patchily in the landscape, forming 51 discrete populations with a mean size of 237 flowering plants (median 62 plants, range 1–3100 plants in 1999) and a mean plant coverage of 25.3 m2 (range 0.25–371 m2) (Fig. 1).
Table 1. Weather data collected at weather stations close to the study area at Tullgarn in 1991–2001, and 30-year means from the reference period 1961–90. Data on number of sunshine hours from Stockholm (50 km north-by-north-east of Tullgarn), temperature from Hårsfjärden (25 km east-by-north-east of Tullgarn) and precipitation from Åda (4 km south-west of Tullgarn)
| Mean 1961–90||276||292||260||221||773|
|Mean temperature (°C)|
| 1999|| 8.2|| 14.5||1 8.1|| 15.6|| 16.1|
| 2000|| 10.4|| 13.2|| 15.6|| 15.4|| 14.7|
| 2001|| 9.6|| 14.0|| 18.1|| 16.2|| 16.1|
| Mean 1961–90|| 8.8|| 13.9|| 15.9|| 15.1|| 14.9|
| 1999|| 21|| 47|| 33|| 50||130|
| 2000|| 31|| 45||114|| 73||232|
| 2001|| 25|| 15|| 65|| 62||142|
Figure 1. Map of the study area showing Vincetoxicum hirundinaria populations (circles) classified according to population size (number of flowering individuals). Populations included in the present study are indicated with filled circles; other populations are indicated with open circles. Grey areas denote sea (to the east and south-east) and lakes, and the thick line shows the eastern limit of the investigated area.
Download figure to PowerPoint
To determine how flower and fruit production were related to habitat variables and population characteristics, we gathered data in 39 populations over 3 years. The study populations were chosen to represent a wide range in size, density and isolation. A population was defined operationally as a group of plants separated by a distance of at least 25 m from the closest conspecific. In most cases, among-population distances were considerably larger than this (mean 152 m, median 100 m, range 25–1223 m; Fig. 1). We quantified the size of the study populations as the total number of flowering plants. There is wide variation in size, morphology and colour of flowers, making the delimitation of plant individuals straightforward during the flowering period, even in dense populations. To quantify population connectivity, we calculated for each population the total area (in m2) of other V. hirundinaria populations within a radius of 500 m. Other radii and other isolation measures, examined in preliminary analyses, were related less strongly to plant reproductive success. To characterize sun exposure, soil depth and local population density, we took measures in direct connection with the plants whose reproductive success was quantified. In each population, up to 20 individuals were selected randomly, marked and mapped in 1999 [except in two of the largest populations, in which 94 (population T18) and 29 (T23) plants were included for study]. For each marked plant, we recorded sun exposure, soil depth and local population density. To quantify sun exposure, we determined the number of hours the plant was reached by direct sunlight at midsummer. A transparent plastic sheet was mounted vertically along the periphery of a horizontal semi-circular plate mounted on a tripod. The path of the sun at midsummer (as viewed from the circle centre of the base plate) was delineated on the sheet and the sun's position for each hour marked. The device was placed at the top of the plant and aligned with the aid of a compass and a spirit level. Looking from the circle centre, the number of hours (between 0600 h and 1800 h) the plant was hit by direct sunlight (not shaded by vegetation) could then be recorded. To quantify soil depth, a steel probe was pushed into the ground until it reached rock at 10 regularly spaced locations within a 1 m2 circular area with the focal plant in the centre, and the mean depth was calculated. To quantify local plant density, we recorded the number of V. hirundinaria individuals within the same 1 m2 area.
The study populations varied considerably in size [mean 284 flowering plants, median (range) 80 (1–3100) plants], mean density [2.6 (1–4.5) flowering plants per m2] and connectivity [153 (0–429) m2 of V. hirundinaria cover within 500 m of the focal population; n = 39; Fig. 1]. Population size and density, in terms of the number and density of flowering plants, did not vary during the 3-year period.
Each year, we quantified flower and fruit production of the marked plants. We counted the number of flowering shoots and estimated the number of flowers per shoot. At the end of the flowering season, the number of leaf nodes with an inflorescence (x1) and the number of scars left by pedicels in the inflorescence at the lowermost node (x2) were recorded for up to three flower-producing shoots per plant. The number of flowers per shoot (y) was estimated using the expression y = 5.51x1 + 1.95x2. This formula was obtained from a sample of 190 shoots collected at many different localities at Tullgarn in 1999, on which x1, x2 and y were all measured (R2 = 0.93). For each plant, we estimated the total number of flowers produced as the number of flowering shoots multiplied by the mean number of flowers per shoot. We recorded the number of fruits that had initiated development, the number of initiated fruits that aborted and the number of fruits that reached full size. We quantified fruit initiation as the number of initiated fruits divided by the total number of flowers produced, and fruit abortion as the number of aborted fruits divided by the number of initiated fruits. In addition, we counted the total number of full-size fruits in each population, and determined how many of these were attacked by E. connexa. For each population, we quantified fruit predation as the proportion of full-size fruits attacked by E. connexa.
To determine whether water availability limits fruit production, we conducted a watering experiment in population T18, the largest V. hirundinaria population in the area. This population is subdivided into two distinct subpopulations: T18A (which is located in an open, rocky environment on the top of a hill) and T18B (which is located in a shaded environment at the base of the hill). Plants were marked in early June 1999. In subpopulation T18A, we marked a total of 80 plants, of which 30 plants were assigned to the watering treatment. In subpopulation T18B, we marked a total of 60 plants, of which 30 plants were assigned to the watering treatment. Plants in the watering treatment were watered 12 times during the period 16 June to 18 August. Water was added to a circular area of 0.5 m2 with the plant in the centre. The total amount of water added to each plant corresponded to 324 mm of rain (cf. Table 1). At the end of the flowering period, we recorded the number of flower-producing shoots, and estimated the number of flowers produced by up to 10 flower-producing shoots per plant, and total flower production using the procedure outlined earlier. At fruit maturation, we scored the number of flowers that had initiated fruit development and the number of full-size fruits on all experimental plants.
To determine whether watering in one year would influence flower and fruit production in ensuing years, we recorded flower and fruit production of the experimental plants in 2000 and 2001. Some plants lost their tags, which reduced the final sample somewhat.
To examine whether fruit set was limited by insufficient pollination, we conducted a hand-pollination experiment in subpopulation T18A in 2000. For this experiment, we marked four inflorescences, each on a separate shoot, on each of 20 plants. Flowers in two of the inflorescences received supplemental hand pollination, while the flowers in the other two inflorescences served as open-pollinated controls. Pollinia were transferred with the aid of a fine straw of grass, and the pollen source was a plant at least 5 m away from the focal plant. Supplemental hand pollinations were conducted during peak flowering. At fruit maturation, we recorded the number of flowers and full-size fruits produced by experimental inflorescences. To examine whether supplemental hand pollination reduced the likelihood of fruit set in open-pollinated control inflorescences on the same plant (cf. Zimmerman & Pyke 1988), we compared their fruit production to that of open-pollinated inflorescences on 40 unmanipulated plants that grew in the same area of the population and that were included in the survey of among-population variation in flower and fruit production.
We used two-way analysis of variance (anova) to quantify how much of the total variance in number of flowers, fruit initiation and number of full-size fruits and number of intact mature fruits (i.e. those that had escaped seed predation) per plant that could be attributed to variation among populations and variation among years, respectively (REML estimates using the software JMP version 5.01, SAS Institute Cary, NC, USA). Corresponding estimates could not be obtained for fruit abortion and fruit predation because fruits were not produced in all populations in all years.
We used multiple regression models that were based on population means and that explored all possible combinations of predictor variables to assess the effects of sun exposure, soil depth and population size, density and connectivity on number of flowers, number of full-size fruits and number of intact mature fruits per plant. We used the same approach to explore causes of variation in fruit survival during different stages of fruit development, i.e. variation in fruit initiation, fruit abortion and fruit predation. Relationships were examined separately for 1999, 2000 and 2001. We ranked models using the Akaike information criterion (AIC), and selected the model with the lowest AIC (Quinn & Keough 2002). Collinearity of independent variables may complicate the interpretation of partial regression coefficients in multiple regression (Graham 2003). In the present study, variance inflation factors (VIF) were low (VIF < 1.6 for all predictor variables), suggesting that multi-collinearity was weak and should not represent a problem (cf. Quinn & Keough 2002).
To examine further the causal relationships between environmental predictors, population characteristics, mean number of flowers and fruit production, we compared different path models of measured variables, as in the first step of structural equation modelling (Shipley 2000). In addition to environmental predictors and population characteristics, these models included the mean number of flowers, initiated fruits, full-size fruits and intact mature fruits produced per plant. Because of the limited sample sizes (n = 39 populations), we chose not to fit latent variables. The module ‘Build’ in the software TETRAD III (Scheines et al. 2006) was used to determine possible paths with significance set to 0.1. The software uses prior knowledge and covariation of supplied variables to determine conditional independence of these variables. The result is a series of possible path diagrams. We assigned variables to different tiers, where variables in higher tiers could not be causes of those in lower tiers (1: sun exposure, soil depth and connectivity; 2: population size and density; 3: number of flowers; 4: number of initiated fruits; 5: number of full-size fruits; 6: number of intact mature fruits). The degree of fit between the observed and expected covariance structures of the final path models was tested statistically with the ‘sem’ function in the free software R (R Foundation for Statistical Computing, Vienna, Austria) (R Development Core Team 2005). A significant goodness-of-fit chi-square statistic indicates that the model does not fit the data.
Because of lack of variation in some variables, multiple regression models and path models differed slightly between the three study years. In 1999, almost all fruits were consumed by seed predators and very few intact fruits were produced; in 2000, very few initiated fruits aborted. It was therefore not meaningful to analyse among-population variation in fruit predation and output of intact mature fruits in 1999 or in seed abortion in 2000 with multiple regression models. Similarly, we did not include number of intact mature fruits in path models based on data collected in 1999, or the number of full-size fruits in models based on data from 2000.
Population size, number of flowers per plant, number of initiated fruits per plant (+0.5), number of full-size fruits per plant (+0.5) and number of intact mature fruits per plant (+0.5) were log-transformed, sun exposure was square-root transformed, and fruit initiation, fruit abortion and fruit predation were arcsine square-root transformed before statistical analyses.
Two-way anova was used to examine the effects of subpopulation and watering treatment on flower production, fruit initiation, fruit abortion, fruit set and number of full-size fruit in the watering experiment.
We used paired t-tests to compare fruit initiation, fruit set and number of full-size fruits in open-pollinated control inflorescences and in inflorescences receiving supplemental hand-pollination on experimental plants. We used one-way anova followed by Tukey tests to determine whether fruit production varied among hand-pollinated inflorescences, open-pollinated control inflorescences on the same plants and open-pollinated inflorescences on plants that did not receive supplemental hand pollination (second control). Analyses were conducted on plant means.