Study Region and Butterfly Sampling. The study was conducted on the Urato Islands in Matsushima Bay, Japan (Fig. 1). The diversity of butterfly species and their host plants was investigated on five islands: Mahanashi, Katsura, Sabusawa, Nono, and Hoh. The respective areas of these islands are 0.15, 0.76, 1.45, 0.56, and 0.15 km2. The locations of 4, 12, 22, 9, and 4 sampling points were randomly determined on Mahanashi, Katsura, Sabusawa, Nono, and Hoh, respectively, such that the number of sampling points per area was approximately the same (0.04–0.06 km2). A census of butterflies at all of the sampling locations was conducted three times, in August 2004 and in July and August 2005. To minimize the effect of the seasonal change in butterfly appearance, the census was conducted only in August and July. The number of individuals of each species observed and captured by sweep nets within 5 m either side of a 10-m line transect was recorded by two investigators during 12 min when weather conditions were suitable for butterfly activity.
The analysis of the diversity and abundance of host plants and butterfly species was conducted mainly at eight communities because these islands support different vegetation types (Fig. 1). At the five-community scale, each community corresponded to a single island, and the results for the 5 communities were shown in SI Table 3 and Fig. 5. At the eight-community scale, two and three communities were classified on Katsura Island and Sabusawa Island, respectively (Fig. 1). Sampling points 6, and 7, and 8, were included on Katura Island and Sabusawa Island, respectively. In the east of Katsura Island, a large area was occupied by deciduous forests and this was unique compared with the other areas on the island. On Nono Island, the northeast area includes vegetation types that are somewhat different than those found in other areas of the island; however, this area is a narrow peninsula and only one sampling point was included. Owing to the position of the randomly chosen sampling points, it was difficult to adequately divide Nono Island into two communities. Thus, Nono Island was not divided for the eight-community analysis. To remove the effects of different sampling effort on the different islands, 16 samplings were conducted on all of the islands, with the exception of Sabusawa where sampling was conducted once at all 22 sampling points. For instance, on Katsura, the sampling was conducted at all 12 sampling points and 4 random sampling points selected from the original 12. For the 8-community analysis, sampling was conducted twice on Mahanashi and Hoh islands in each census. Six to nine samplings were conducted in the communities. The numbers of individuals of each species per sampling point (100 m2) per sampling time were used as the butterfly densities for the analysis.
Throughout the sampling period, a total of 43 butterfly species were identified. Parantica sita niphonica and Dichorragia nesimachus were omitted from the analysis because P. sita niphonica does not reproduce on these islands, and the host plant of D. nesimachus could not be found. In addition, in the field, it was difficult to distinguish Limentis glorifira and Neope goschkevitschii from Limenitis camilla japonica and Neope niphonica, respectively. Therefore, the data for L. glorifira and N. goschkevitschii were combined with that of L. camilla japonica and N. niphonica, respectively. Thus, in total, 39 butterfly species were used in the study.
Estimation of Host Plant Biomass. Investigations of the distribution of resources for butterflies on the Urato Islands were conducted from 2003 to 2005. In this study, we defined resources for butterflies as host plants for the juvenile stages; resource distribution was estimated by using the following procedure. The host plants used by each species of butterfly were selected by comparing the floral composition based on preliminary field investigations and the host plant records listed in Fukuda et al. (30). Species belonging to the Satyridae and Hesperiidae (with the exception of Daimio tethys) are, however, largely polyphagous, and few accurate records of host plant species are available. In these cases, only the plant species on which the larvae of a given butterfly were frequently recorded were classified as host plants. For butterflies that subsist on bambusoid plants, their host plants were defined as all of the bambusoid plant species distributed in the study area. This was necessary because the host plants recorded in the literature were not identified to the species level. In the case of Parnara guttata, known as a pest of rice, all of the plants belonging to the Poaceae and Cyperaceae were considered as host plants. This is because a relatively wide range of host plants has been recorded for this butterfly, although food plant preference at the species level remains undetermined (30).
Vegetation surveys were carried out by using the following procedures. We divided the vegetation and land use into the following 12 types according to the results of preliminary field surveys and an aerial digital photograph (one pixel = 10 cm x 10 cm) taken in July 2003: (1) residential areas, roads, and other man-made structures; (2) rice paddy fields; (3) other crop fields; (4) grasslands with vegetation by annual grasses such as Trifolium spp.; (5) grasslands dominated by Miscanthus sinensis and other perennial tall grasses; (6) wetland vegetation mainly dominated by Phragmites communis; (7) bamboo or dwarf bamboo thickets; (8) Cryptomeria japonica afforested areas; (9) Pinus densiflora (P. thunbergii, in part) forests; (10) Machilus thunbergii evergreen forests; (11) deciduous forests mainly dominated by Quercus serrata; and (12) mixed secondary forest dominated by Juglans mandshurica var. sachalinensis, Celtis sinensis var. japonica, and Neolitsea sericea (Fig. 1). We then established 8–26 quadrants of 10 m x 10 m for each vegetation or land-use type. Measurements were taken of the number of individuals of each plant species in a quadrant, the height of each tree individual, and the degree of cover by herbaceous plants. From these data, the average number of individuals per unit area for each size class of each tree species and the average degree of cover per unit area for each herbaceous plant were calculated for each vegetation or land-use type. In this study, we assumed that the leaves of each plant species were the edible parts for the larvae of butterflies and estimated the wet weights of leaves per individual for tree species and those per unit area for herbaceous plants. For each plant species, we measured the wet weights of leaves and counted the number of leaves per individual or per unit area. We then estimated the unit biomass of the edible parts based on the average weights per leaves and the average number of leaves per individual or unit area. The available resource biomass of a given plant species in each local community was estimated from the unit biomass and the area of vegetation and land-use types in which the plant species was distributed. Hence, in the analysis, the biomass of a given plant species was used as biomass density (unit biomass per unit area).
Analysis of Butterfly and Host Plant Diversity. The relationships among the species richness of butterflies and host plants, the total biomass densities of butterflies and host plants, and the total biomass densities of butterfly species and species richness of butterflies were examined by using a linear regression analysis. The total biomass densities of butterflies were calculated as 
di wi, where di is the densities of species i, and wi is the average weights of adult butterflies of species i. The total biomass densities of the host plants were calculated as the sum of the biomass densities of all of the 58 host plant species.
We tested the effects of resource and geographic distance on butterfly biodiversity by applying an extension of the Mantel test for estimating the regression coefficients of independent distance matrices (23, 24). We considered a multiple regression equation of the form: aij = 
0 + AB.Cbij + AC.Bcij, where bij is the dissimilarity matrix of the relative resource abundance between communities i and j (Resource matrix), cij is the geographic distance (Distance matrix, both linear and log-transformed were considered), and aij is the dissimilarity matrix of butterfly species diversity (Butterfly matrix). AB.C and AC.B are the regression coefficients, and the significances were tested by randomization. The geographic distances were measured between the centers of gravity of two communities. To measure the similarity of butterfly relative abundance and host plant resource distribution, we used Odum's percentage difference method (31). The dissimilarity between communities 1 and 2 is represented as follows:
 |
where x1i and x2i are the numbers of individuals of species i in community 1 and community 2, respectively. For butterfly species, the average numbers of individuals per sampling time per sampling point were used as x. When resource similarity was considered, x1i and x2i are the biomasses of host plant species i in community 1 and community 2, respectively. The biomass was divided by the area of the community. D values increase with decreasing similarity. The number of individuals and the biomass of host plant species were log transformed because the evaluation of dominant and rare species were equivalent (32).
Effects of Host Plant Biomass on Butterfly Density. The effects of the biomass of the host plant on the density of each butterfly species were examined in eight communities. We used a linear regression in which the density of a butterfly species was a dependent variable, and the biomass density of its potential host plant species (SI Fig. 3) was used as an independent variable. If there was a high correlation coefficient between the biomass densities of the host plants, one of either host plant species was omitted in the analysis. We explored the set of predictors of the host plant species with a stepwise AIC procedure; independent variables were removed and added to the models to determine the set of predictors that yielded the lowest AIC by using the R software. Given the model with the lowest AIC, we obtained the significance of the regression coefficients. We conducted an analysis of 39 butterfly species, and thus, to control the errors of multiple comparison, we used the false discovery rate (FDR) procedure (33).
Answers to all your Biology Questions
