Study Sites And Species
A total of 276 individuals were sampled from 34 species of Cyrtandra across 19 sites on six islands (Table 1). Sampled elevations ranged from 20–1270 m, and included four habitat types: lowland rainforest, montane rainforest, montane mesic forest, and cloud forest (sensu Mueller-Dombois & Fosberg 1998). The mean number of observed Cyrtandra species per site was 3 (min = 1; max = 5), and the mean number of individuals sampled per site was 15 (min = 3; max = 48).
Functional Traits
Across all 34 species examined, maximum plant height ranged from sprawling shrubs as low as 115 cm (e.g., C. hispida) to small trees as tall as 6 m (e.g., C. richii) (Table S2). Mean values for leaf traits varied across all species as follows (abbreviations as in Table 2): SLA 14–41 m2 kg−1, LDMC 79–224 mg g−1, petiole length 2–14 cm, leaf size 29–776 cm2, leaf pubescence 0–362 trichomes per cm2. Mean values for floral traits varied across all species as follows: flower length 15–46 mm, flower width 5–14 mm, corolla lobe size 22–570 mm2, and floral display size 1–15 flowers (Table S2). Within species, most traits exhibited low levels of variation (Table S2, Fig. 2A), although high intra-specific variation was observed in height and leaf size (Table S2). The PCA explained 71% of the variation across the first four axes. The first PCA axis was largely defined by corolla tube length and width followed by SLA and leaf size (26% of the variation; Fig. 2B, Table 3), the second PCA axis was defined by flower number, LDMC, leaf size and leaf pubescence (19% of the variation; Fig. 2B, Table 3), the third PCA axis was defined by petiole length, corolla lobe size, flower number and SLA (15% of the variation; Table 3), and the fourth PCA axis was defined by height, LDMC and leaf pubescence (11% of the variation; Table 3).
Table 3. Principal components analysis (PCA) loadings for 10 functional traits across the first four axes (71% of variation). The strongest contributors to each principal component are in bold (based on a cutoff of √(1/10)= +/- 0.316).
Functional trait | PC1 | PC2 | PC3 | PC4 |
Corolla tube length | 0.511 | -- | 0.259 | -- |
Corolla tube width | 0.412 | -0.374 | -0.169 | -0.249 |
Leaf size | 0.325 | 0.400 | -0.261 | -0.120 |
Specific leaf area | 0.325 | 0.259 | 0.321 | 0.161 |
Flower number | 0.265 | 0.432 | -0.339 | -- |
Leaf dry matter content | -0.214 | -0.416 | -0.123 | -0.464 |
Leaf pubescence | 0.307 | -0.372 | -0.274 | 0.355 |
Petiole length | -- | -0.210 | -0.528 | 0.224 |
Corolla lobe size | 0.247 | -0.283 | 0.499 | -- |
Maximum height | 0.281 | -- | -- | -0.744 |
Phylogenetic Signal
Estimates of phylogenetic signal in the ten functional traits resulted in a range of K values, from 0.34 for petiole length to 0.89 for flower width (Table 4). Five traits (SLA, LDMC, petiole length, leaf pubescence, and maximum height) had K values that were significantly different from one, suggesting absence of phylogenetic signal in these traits. In contrast, four traits (leaf size, flower length, flower width, and floral display) had K values significantly different from zero, indicating the presence of phylogenetic signal. Corolla lobe size was not significantly different from one or zero, although there was a trend towards this trait exhibiting phylogenetic signal (K = 0.58, P = 0.09).
Table 4
Estimates of phylogenetic signal in functional traits using Blomberg’s K (Blomberg et al. 2003). Values of K < 1 indicate trait divergence, K ≈ 1 indicates random trait evolution (Brownian motion), and values of K > 1 indicate trait conservatism. Significant differences from the null expectation are indicated by the following: ‘*’ p < 0.05, ‘**’ p < 0.01; NS indicates a result not significantly different from the null expectation.
Trait | Blomberg’s K | Different from 0 | Different from 1 |
Maximum height | 0.45 | 0.233 NS | 0.037 * |
SLA | 0.50 | 0.150 NS | 0.046 * |
LDMC | 0.40 | 0.285 NS | 0.013 * |
Petiole length | 0.34 | 0.682 NS | 0.008 ** |
Leaf pubescence | 0.36 | 0.603 NS | 0.006 ** |
Leaf size | 0.67 | 0.003 ** | 0.292 NS |
Corolla tube length | 0.63 | 0.003 ** | 0.217 NS |
Corolla tube width | 0.89 | 0.025 * | 0.763 NS |
Corolla lobe size | 0.58 | 0.089 NS | 0.148 NS |
Flower number | 0.63 | 0.006 ** | 0.236 NS |
Phylogenetic Community Structure
Cyrtandra communities on all six islands were phylogenetically clustered; four were significantly more clustered than null communities under the MPD metric (Viti Levu, Vanua Levu, Kaua’i, and Hawai’i Island; mean Zscore = −2.75, P = 0.007), and/or the MNTD metric (Viti Levu, Vanua Levu, Savaiʻi, and Hawaiʻi Island; mean Zscore = −2.53, P = 0.02) (Table S3). The Fijian Island of Taveuni was the only island that did not exhibit significant clustering as estimated by both metrics (MPD: Zscore = −1.22, P = 0.14; MNTD: Zscore = −1.11, P = 0.14). Inter-island community structure inferred from the MNTD metric revealed several patterns (Table S4). The smallest phylogenetic distances were between island communities within the same archipelago. Within the Fijian archipelago, Viti Levu and Vanua Levu were most similar to Taveuni (geologically the youngest, and also the smallest island of the three), while Taveuni was most similar to the neighboring island of Vanua Levu. Between archipelagos, phylogenetic distances were smallest between island communities that were separated by the shortest geographic distance (e.g., Savaiʻi, Samoa was most similar to Taveuni, Fiji; Kauaʻi and Hawaiʻi Island were most similar to Viti Levu, Fiji).
Within islands, a total of 18 sites were examined for phylogenetic structure, with one site from Vanua Levu omitted due to the presence of only a single species at the site (C. longifructosa). Of the 18 sites, 15 were clustered and three were over-dispersed (Table S5). Only four sites were significantly clustered relative to null communities under the MPD metric (Komave, Waisali, Voma, and Kokeʻe; mean Zscore = −1.67, P = 0.03), while all other sites were not significantly different from the null expectation (mean Zscore = −0.74, P = 0.27). Similar results were obtained under the MNTD metric, with 16 clustered sites and two over-dispersed sites. Four communities were significantly clustered (Komave, Waisali, Matavanu, and Tomanivi; mean Zscore = −1.75, P = 0.03), while all others did not vary from the null expectation (mean Zscore = −0.79, P = 0.25).
Trait-based Community Structure
Analyses of island community structure based on the combination of all ten functional traits revealed that five islands were phenotypically clustered and one was over-dispersed (Hawaiʻi Island), although none of the communities were significantly different from the null expectation under the MPD metric (mean Zscore = −0.54, P = 0.32) or the MNTD metric (mean Zscore = −0.57, P = 0.31) (Table S6; Fig. 3). Island community structure based on individual traits suggested that four islands were significantly clustered for one or two traits under the MPD metric (Viti Levu, Vanua Levu, Savaiʻi, Kauaʻi). Clustered traits included maximum height, SLA, flower length, corolla lobe size, and flower number. In contrast, two islands were not significantly different from null communities in any of the traits examined (Taveuni and Hawaiʻi Island, the youngest islands in each respective archipelago). Under the MNTD metric, two islands were significantly clustered for maximum height (Vanua Levu) or flower length (Kauaʻi), while all others did not differ significantly from the null. Inter-island community structure based on functional trait distances corroborated the patterns seen in the phylogenetic inter-community analysis, with the single exception of Taveuni communities being most phenotypically similar to Viti Levu communities as opposed to Vanua Levu in the phylogenetic analysis (Table S7).
The community structure of sites within islands based on all functional traits combined revealed ten clustered sites and eight over-dispersed sites under the MPD metric (Table S8). Only three sites exhibited significant clustering (Colo-i-Suva, Korobaba, and Lavena; mean Zscore = −2.09, P = 0.03), while all other sites did not differ from null communities (MPD mean Zscore = 0.08, P = 0.52). Similar results were seen using the MNTD metric, with ten clustered sites and eight over-dispersed sites. Only one site was significantly clustered (Lavena, Zscore = −2.51, P = 0.03), while all other sites did not vary from the null (MNTD mean Zscore = −0.08, P = 0.48). Analyses based on individual traits revealed six communities that were significantly clustered under the MPD metric (Komave, Waisali, Des Voeux, Aʻopo, Matavanu, and Kokeʻe), while all other sites did not differ significantly from null communities. The MNTD metric revealed similar results, with six communities being significantly clustered (Komave, Korayanitu, Waisali, Des Voeux, Aʻopo, and Kokeʻe), and all others not differing from the null expectation. Traits that exhibited significant clustering under one or both metrics included maximum height, SLA, LDMC, petiole length, flower length, and floral display.