The study was carried out in 2007 near the city of Göttingen (51.5°N, 9.9°E), Lower Saxony, Germany. In an area of about 25 × 30 km, we selected 67 study sites (33 calcareous grasslands and 34 oilseed rape fields belonging to four categories (ESM 1): (1) 16 grasslands were isolated by at least 230 m from the nearest oilseed rape field (mean distance from grassland to field edge ± SE: 481 ± 8.8 m); (2) 17 grasslands were within 1–15 m distance of oilseed rape; (3) 17 oilseed rape fields were within 1–15 m distance of the study grasslands; and (4) 17 oilseed rape fields were isolated by at least 570 m from calcareous grasslands (mean distance ± SE: 1,598 ± 59.7 m; ESM 2). Each of the 17 study grasslands of category 2 was in 1–15 m distance of one of the studied oilseed rape fields of category 3. Study grasslands and oilseed rape fields were at least 1 km apart from other study grasslands and oilseed rape fields, respectively. At 300 m distance from calcareous grassland, abundances of cavity-nesting bees have been shown to be reduced by 95 % compared to grasslands (Krewenka et al. 2011). All study sites where we recorded Osmia bicornis in our traps were included in the analyses of pollen contents and brood cell numbers (in all cases >9 brood cells per site): (1) 14 grasslands isolated from oilseed rape; (2) 12 grasslands adjacent to oilseed rape; and (3) 10 oilseed rape fields adjacent to grasslands. Category 4 (oilseed rape fields isolated from grasslands) was excluded from the analyses of pollen contents and brood cell numbers because only 2 sites had been colonized by O. bicornis. Instead, the 34 oilseed rape fields of category 3 and 4 were included in an analysis of O. bicornis incidence (see below).
Oilseed rape fields were sown with Brassica napus in the autumn of the previous year and were conventionally managed with usually one insecticide application during the flowering period. Isolated grasslands and grasslands adjacent to oilseed rape were similar in management, exposition, inclination and size (grasslands isolated from oilseed rape: mean size ± SE: 1.7 ± 0.3 ha, min: 0.2 ha, max: 6.8 ha; grasslands adjacent to oilseed rape: mean size ± SE: 1.7 ± 0.5 ha, min: 0.1 ha, max: 4.8 ha). Flower cover (% cover of flower corollas per area ground surface) and the number of plant species flowering were recorded once during oilseed rape flowering in a 0.1-ha plot per grassland, and did not significantly differ (all P > 0.12) between grasslands isolated from oilseed rape (flower cover: 0.10 ± 0.07 %, min: 0.0001 %, max: 1.0 %; species number: 4.07 ± 0.58, min: 1, max: 9) and grasslands adjacent to oilseed rape (flower cover: 0.46 ± 0.17 %, min: 0.0001 %, max: 2.12 %; species number: 4.58 ± 0.51, min: 3, max: 7).
Around each study site, oilseed rape fields and semi-natural habitats (calcareous grasslands, orchard meadows, old fallows, hedgerows) were mapped in landscape circles with radii of 250, 500, 750, and 1,000 m. The proportions of oilseed rape fields in the landscape circles were calculated with GIS software (ESRI ArcView 3.2). The proportion of oilseed rape spanned a gradient from 0 to 65, 29, 23, and 17 % in the 250, 500, 750, and 1,000 m radius, respectively, and was not correlated with any other habitat type (Spearman rank correlations, all P > 0.1, n = 34), but was highly negatively correlated with the distance to the next oilseed rape field at all scales (Spearman rank correlations, all P < 0.001, n = 34). The proportion of semi-natural habitats spanned a gradient from 0.3 to 43, 26, 15, and 13 %, respectively, and was positively correlated with the Shannon Index of habitat diversity in landscape circles with 750 and 1,000 m radius (Spearman rank correlations, all P < 0.01, n = 33).
Pollen and brood cells in trap nests
We established trap nests in the edge and the center of our 67 study sites to assess the effect of oilseed rape on pollen-collecting behavior and the number of brood cells of O. bicornis. This solitary, polylectic bee can nest in a variety of naturally pre-existing cavities and also colonizes artificial trap nests. O. bicornis is the most abundant cavity-nesting bee in the study area (Holzschuh et al. 2010). Between April and June, females of O. bicornis build nests with up to 30 brood cells (Westrich 1989) and collect pollen for larval provisioning nearby their nests within a radius of up to 600 m (Gathmann and Tscharntke 2002). A high number of brood cells per trap nest can reflect both a preference of females to nest at this specific location and a high reproductive output per female.
Trap nests consisted of four plastic tubes (20 cm long, 10.5 cm diameter), each filled with about 200 internodes (20 cm long) of common reed Phragmites australis (see Tscharntke et al. 1998) and fixed on a post at a height of 1.2 m. Internodes which contain brood cells of O. bicornis can be easily recognised by a clay cap, which the bee builds at the end of the internode. One female can occupy more than one internode. We placed trap nests in the centre and the edge of the 67 study sites (134 trap nests with 536 plastic tubes). In oilseed rape fields, edge trap nests were placed between the first and the second row of oilseed rape plants, center trap nests were placed at 20 m distance from the edge. In grasslands, edge trap nests were placed at 1 m distance from the habitat border and center trap nests were placed at 20 m distance from the edge trap nests. Trap nests were established in March and checked for O. bicornis nests at the beginning of oilseed rape flowering in April. No nests had been built by O.bicornis before the beginning of oilseed rape flowering. Directly after the end of oilseed rape flowering (26 days later in May), all reed internodes containing bee brood cells were collected, stored at 4 °C to stop larval development, and opened in the laboratory. The number of O. bicornis brood cells per trap was recorded (Tscharntke et al. 1998).
For the identification of forage plants, pollen was collected from the first and the last brood cell per reed internode. These two most separated brood cells within a reed internode are normally built with an interval of several days inbetween (Westrich 1989) and might thus mirror the pollen-collection behavior at two different time points during oilseed rape flowering. All brood cells of O. bicornis contained large amounts of pollen, because larval development had not made considerable progress by that time. After transferring a small sample of the pollen from the brood cell to a glass slide, the percentage pollen of Brassicaceae was determined under a light microscope for 300 pollen grains, which were located in a randomly chosen cluster within the sample. We assume that all pollen of the Brassicaceae pollen came from oilseed rape, because no other Brassicaceae were abundantly flowering at that time in grasslands, crop fields, or field margin strips. Pollen analyses were conducted on 843 brood cells from the 36 sites where O. bicornis occurred. The numbers of brood cells were summed over the four plastic tubes of a trap nest per post and the two trap nests per site, and the percentage oilseed rape pollen per brood cell was averaged over all brood cells of a site, because the position of the trap nests in the field neither affected the number of brood cells (t test; F
1,55 = 1.3, P = 0.25) nor the percentage oilseed rape pollen (t test; F
1,55 = 1.4, P = 0.24).
The effect of grassland on the incidence (presence or absence) of O. bicornis in oilseed rape was assessed in a generalized linear model with quasibinomial errors and the predictor presence of adjacent grassland (oilseed rape adjacent to grassland vs. oilseed rape isolated from grassland).
To assess whether the number of O. bicornis brood cells in grasslands is higher adjacent to oilseed rape than isolated from oilseed rape and increases with the proportion of oilseed rape in the landscape, we performed ANCOVAs (type I sums of squares) with the dependent variable number of O. bicornis brood cells per grassland and the predictors presence of adjacent oilseed rape (grassland adjacent to oilseed rape vs. grassland isolated from oilseed rape), proportion of oilseed rape in the surrounding landscape circle and their interaction. Separate models were calculated for the different landscape circles (250, 500, 750, 1,000 m radius).
The effects of local and landscape-scale availability of oilseed rape pollen on the percentage oilseed rape pollen in larval food were assessed in ANCOVAs with percentage oilseed rape pollen as the dependent variable and the predictor site type (oilseed rape adjacent to grassland vs. grassland adjacent to oilseed rape vs. grassland isolated from oilseed rape) and their interaction. Separate models were calculated for the different landscape circles (250, 500, 750, 1,000 m radius). In the pollen models, we considered the 36 sites where trap nests had been colonized by O. bicornis. These were 14 of the 16 sites of category 1 (grasslands isolated from oilseed rape) and 22 of 34 sites of categories 2 and 3 (grasslands adjacent to oilseed rape and oilseed rape adjacent to grassland). All colonized sites of category 1 and eight colonized sites of categories 2 and 3 were spatially separated from all other sites; in seven cases, the two sites of categories 2 and 3 were directly adjacent to each other. For these seven cases, we additionally conducted a t test for paired samples (sites of category 2 vs. 3) and compared the result to the result of the ANCOVA to check whether neglecting the partly nested structure of the model affected the outcome of the ANCOVA model. Furthermore, we assessed—for grasslands isolated from oilseed rape—the relationship between percentage oilseed rape pollen in larval food and the distance from the nearest oilseed rape field in a linear regression model.
To assess whether the number of O. bicornis brood cells increased with increasing percentage of oilseed rape pollen in larval food, we calculated linear regression models with the dependent variable number of brood cells per site and the predictor percentage of oilseed rape pollen in larval food. Again, we considered those 36 sites where trap nests had been colonized by O. bicornis. The hypothesis that the positive impact of oilseed rape pollen decreases with increasing availability of alternative food resources at the local and landscape scale was tested in separate linear regression models for grasslands and oilseed rape fields with the dependent variable number of brood cells per site, the predictors percentage of oilseed rape pollen, proportion of semi-natural habitats in the surrounding landscape circle, flower cover, and diversity of flowering plants in the grassland (or in the adjacent grassland in case of oilseed rape fields), and the two-fold interactions between percentage oilseed rape pollen and the other predictors. Separate models were calculated for the four landscape circles. All models were computed in R (v.2.11.1; R Development Core Team 2011). Maximal models were simplified in a manual stepwise backward selection on the basis of F tests (Crawley 2007). Predictors with p > 0.05 were removed from the maximal models. Tukey’s post hoc test for multiple comparisons of means were performed with heteroscedastic consistent covariance estimation, which is a robust method for comparing means of groups with unbalanced group sizes (Herberich et al. 2010; packages multcomp and sandwich).