Several plant traits are known to evolve in predictable ways on islands. For example, herbaceous species often evolve to become woody and species frequently evolve larger leaves, regardless of growth form. However, our understanding of how seed sizes might evolve on islands lags far behind other plant traits. Here, we conduct the first test for macroevolutionary patterns of seed size on islands. We tested for differences in seed size between 40 island–mainland taxonomic pairings from four island groups surrounding New Zealand. Seed size data were collected in the field and then augmented by published seed descriptions to produce a more comprehensive dataset. Seed sizes of insular plants were consistently larger than mainland relatives, even after accounting for differences in growth form, dispersal mode and evolutionary history. Selection may favour seed size increases on islands to reduce dispersibility, as long-distance dispersal may result in propagule mortality at sea. Alternatively, larger seeds tend to generate larger seedlings, which are more likely to establish and outcompete neighbours. Our results indicate there is a general tendency for the evolution of large seeds on islands, but the mechanisms responsible for this evolutionary pathway have yet to be fully resolved.
Seed size varies greatly among plant species, from tiny wind-dispersed orchid seeds to the massive double coconut (Lodoicea maldivica) , and has important consequences for reproductive success [2–4]. Seedling survival is directly influenced by seed size [4,5], and many functional traits covary with the size of seeds. These traits include dispersal mode, growth form, specific leaf area and seed number [6,7]. Many of these other traits evolve predictably on islands (e.g. growth form and leaf area; [8,9]); however, detailed quantitative investigations of how seed size is affected by insularity are lacking.
A reduction in the dispersal ability of seeds is a common evolutionary pathway for plants on islands [9,10]. For example, wind-dispersed members of the family Asteraceae typically display a reduction in pappus size relative to achene size on islands [9,10]. Animal-dispersed taxa (such as Bidens), which produce structures that promote ectozoochory, illustrate a similar pattern, with a reduction in the size of hooks and awns relative to achene size . Furthermore, fleshy-fruited members of the family Araliaceae tend to produce larger fruits and seeds on islands . The evolution of reduced dispersibility has also been detected in as few as five generations, suggesting strong selection pressures .
One explanation for potential changes in seed size on islands is that the small size and isolation of islands may select against dispersal to reduce propagule mortality at sea [9,11]. Selection acting to increase seed size may reduce wind dispersal distances in anemochorous species . The same may be true for fleshy-fruited plants, although tests for directional changes in seed size of fleshy-fruited species have yet to be conducted.
Selection may favour larger seed size on islands for reasons other than dispersal ability. For example, most islands house fewer species than comparable communities on the mainland. Therefore, a germinating seedling on an island is more likely to be adjacent to a conspecific, leading to greater levels of intraspecific competition [13,14]. Larger seeds are more competitive than small seeds, all else being equal [2,4,5]. Therefore, higher levels of intraspecific competition on islands may also select for increases in seed size.
Here, we conduct the first macroevolutionary test for increased seed size on islands. By collecting specimens in the field and using published seed descriptions, we compiled a diverse dataset consisting of 40 island–mainland taxonomic pairings from four island systems surrounding mainland New Zealand. To test for overarching changes in seed size on islands, we first compared seed sizes on islands with seed sizes on the mainland using reduced major axis (RMA) regression. Second, we used a mixed effects modelling approach to test for effects of evolutionary history, growth form and dispersal mode on the island–mainland seed size relationship.
New Zealand has a long history of geological isolation. It separated from Gondwana 80 Ma and has been isolated in the southwest Pacific since. The three main islands (North, South and Stewart Islands) are encircled by numerous smaller islands , the flora of which consists mainly of taxa that have dispersed overwater from New Zealand. Although many of these islands were once connected by land bridges, we focused on four island groups that remained isolated from the main islands during the Pleistocene (Kermadec, Three Kings, Chatham and sub-Antarctic Islands; see ). In particular, we focused on the Chatham group (176° W, 44° S), situated 850 km east of the main islands (figure 1).
To maximize the number of species in the dataset, we visited the Chatham Islands twice, at different times of year. This allowed the inclusion of species with different fruiting phenologies. Seed sizes were measured on Chatham Island taxa in January 2008  and March 2012. Searches were made in Henga Scenic Reserve (43°51.0′ S, 176°33.2′ W), Nikau Forest Reserve (43°45.7′ S, 176°34.8′ W) and Rangaika Scenic Reserve (44°3′37.0182″ S, 176°26′6.0792″ W). Mainland samples were collected from Otari Walton's Bush (41°14′ S, 174°45′ E), Moa Point (41°20′ S, 174°49′ E) and Nelson Lakes National Park (41°48′ S, 172°50′ E). Fruits were randomly selected from each individual, collecting five or more from at least five individuals (following ). This was not always achievable; therefore, a variable number of seeds were used to characterize seed size (table 1). Seed size was estimated as the product of seed length (length of the longest axis) × seed width (maximum distance perpendicular to the length measurement at the widest point of the seed).
To allow for the greatest possible number of island–mainland comparisons, we supplemented field data (25 taxa) with seed measurements contained in Seeds of New Zealand gymnosperms and dicotyledons (hereafter ‘seed atlas’; ). Seed descriptions in the seed atlas result from the examination of at least 10 seeds from each of 10 collections of fruiting material for each species. We obtained the median value from seed dimension ranges when compiling data from within the seed atlas. To justify the use of seed atlas data and to promote accuracy, we ran ordinary least-squares (OLS) regression of field measures against data from the seed atlas. Measurements for two Hebe species were not contained in the seed atlas, and seed sizes were obtained from An Illustrated Guide to New Zealand Hebes . In many cases (42 taxa), the literature provided only length dimensions for seed size. We therefore ran OLS regression of area (length × width, as described above) against length, for the 36 taxa where both length and width parameters were available. Regression parameters were then used to estimate area for those taxa where only a length dimension was available. ANOVA of observed versus predicted values was carried out to test the robustness of seed area predictions. Variables were log-transformed prior to analysis.
Determining mainland relatives for insular species was simple when taxa were undifferentiated (table 1). For Chatham Island endemics, the recent molecular analysis by Heenan et al.  was used where possible. In this study, DNA sequence data were used to identify the closest relatives for 35 taxa endemic to the Chatham Islands. For taxa from other island systems, phylogenetic analyses were used where available [21–26]. When multiple mainland taxa were identified as being equally related to an island endemic, the average seed size of the mainland taxa was used (Leptinella plumosa and L. lanata—see  and Myrsine chathamica—see ). Where taxa were differentiated at species, but not genus level the mainland taxon chosen is the most likely relative based on morphological similarities (e.g. Macropiper melchior and M. excelsum; Coprosma acutifolia and C. tenuifola—see ). In other cases, insular taxa were a variety or subspecies of well-known mainland species.
To test for differences in seed size among island and mainland taxa, a variety of statistical methods could be used. Regressing mean values for insular taxa against mainland taxa is one option. The slope and intercept parameters providing information on the relationship (slope > 1 and intercept > 0 = island taxa with larger seeds; slope < 1 and intercept < 0 = mainland seed size larger). However, the use of OLS regression minimizes the sum of squared variation in the Y (in this case, island) direction and is not appropriate when measurement error in X and Y variables is likely. To avoid these confounding sources of bias, we ran RMA regression. RMA was used to obtain slope and intercept parameters along with 95% confidence intervals.
Phylogenetic relatedness between island–mainland taxonomic pairs creates a lack of independence in our dataset. To overcome this, we ran a mixed effects model treating seed size as the dependent variable, location (island or mainland) as a fixed factor, and ‘species pair’ (island–mainland taxonomic pairing) as a random factor. Many of the taxonomic pairings are separated at species level, potentially indicating more evolutionary divergence than conspecific pairings. We therefore included a second random effect of nested taxonomy (with two levels: undifferentiated at species level or differentiated at species level) in the model structure. Seed dispersal mode and plant growth form were included as fixed factors to test for effects they may have on the island–mainland seed size relationship.
The majority of taxa in our dataset were either wind-dispersed or fleshy-fruited plants. We therefore defined dispersal modes as either ‘dry-fruited’ (wind, water, ballistic/wind) or ‘fleshy-fruited’. Only two island–mainland taxonomic pairings were vines, and we therefore removed these taxa from the growth form analysis. We are specifically interested in whether dispersal mode or growth form influences seed size differences between ‘locations’ (island or mainland). Therefore, the two-way interactions between location and these two factors were maintained in the model structure. We calculated the ratio of island seed size to mainland seed size to visualize whether seeds tended to be larger on islands for each level of dispersal mode and growth form. Ratios greater than one indicate a tendency for larger island seeds. Kruskal–Wallis tests were then used to compare ratios between dispersal modes and growth forms. Robust tests for the effect of island system were not possible owing to uneven sample sizes between islands.
All analyses were conducted in the R environment for statistical computing . RMA analyses were conducted with the SMATR package , and the CAR package  was used to carry out a likelihood ratio test for the mixed effects model. Seed size data were logarithm transformed to conform to normality assumptions.
Field measurements scaled positively with those contained in the seed atlas (OLS regression: R2 = 0.912; p < 0.001). Slope and intercept parameters were not significantly different from one and zero, respectively (slope: 0.971, 95% CI = 0.811–1.131, p = 0.698; intercept: −0.054, 95% CI = −0.373–0.266, p = 0.721), indicating that published seed sizes accurately reflect field measurements. OLS regression showed that seed lengths scaled strongly with seed surface area (R2 = 0.928; p < 0.001; see the electronic supplementary material, figure S1). Seed area estimates calculated using the regression equation did not differ significantly from observed values (F1,70 = 0.086, p = 0.770), indicating that estimates accurately reflected real seed sizes.
Island seed sizes scaled positively with mainland seed sizes (RMA analysis: R2 = 0.854, p < 0.001; figure 2). However, the island–mainland seed size relationship had a slope greater than one (1.18; 95% CI = 1.041–1.338) and an intercept marginally less than zero (−0.021; 95% CI = −0.166 to 0.123). This indicates a tendency for insular taxa to produce larger seeds than mainland relatives, particularly at the larger end of the seed size spectrum.
The likelihood ratio test from the mixed effects model showed no significant effect of plant growth form on the island-to-mainland seed size relationship (χ2 = 7.32, d.f. = 4, p = 0.120). However, there was a significant effect of dispersal mode (χ2 = 14.328, d.f. = 2, p < 0.001). Seed sizes did differ significantly by location (island or mainland; χ2 = 12.562, d.f. = 1, p < 0.001), consistent with results of the RMA analysis, and all dispersal modes and growth forms displayed a tendency for larger seed sizes on islands (figure 3a,b).
Seeds of insular taxa were consistently larger than those of their mainland relatives. Furthermore, this result was consistent, regardless of dispersal mode and growth form. Increasing seed size reduces dispersibility in anemochorous species, which is intuitive, as large seeds are likely to disperse over shorter distances than small seeds [1,10,12,32,33]. In the case of fleshy-fruited plants, this is the first time, to the best of our knowledge, that an increase in seed size on islands has been demonstrated quantitatively. Larger seeds in fleshy-fruits may limit the range of dispersal vectors available to a plant, as well as reducing the number of fruits eaten during a feeding period [34,35]. Furthermore, seed number scales negatively with seed size [36,37] and producing fewer seeds reduces the likelihood of long-distance dispersal .
RMA regression of island seed size against mainland seed size produced a scaling relationship that differed from isometry. In particular, a slope parameter of greater than one suggests that the size of insular seeds increases disproportionately with increasing seed size. The majority of small-seeded taxa in the dataset rely on wind dispersal, while species with larger seeds generally produced fleshy fruits. Anemochorous species may experience stronger constraints on maximum seed size owing to the aerodynamics of wind dispersal , while in fleshy-fruited species (and zoochorous species in general) seed size may be less tightly constrained. Birds often display size increases on islands and size coupling between fruits and frugivores is common [39,40]. Therefore, putative selection pressures for increased seed size may be less constrained in fleshy-fruited taxa. Analysis of dispersal mode reflected this, as fleshy-fruited taxa showed greater size increases on islands than dry-fruited taxa. However, dry-fruited taxa still displayed a tendency for larger seeds on islands. Furthermore, patterns in seed size persisted even after accounting for differences in taxonomic distance and phylogeny between island–mainland pairings, suggesting that selection for increased seed size is strong on islands (see ).
Work on animals suggests the depauperate nature of island communities may increase intraspecific competition and promote insular size changes [13,14,41]. This situation may also apply to plants. Larger seed sizes may promote a competitive advantage owing to the increased nutrient reserves, which produce larger, more competitive seedlings [2,4]. Evidence also suggests that larger seed sizes increase plant survival at later-life stages (e.g. sapling stage; ).
Selection early in ontogeny may also influence size patterns evident at later-life stages. In animals, adult body size is strongly influenced by size at birth and a recent investigation suggests that body size patterns on islands may reflect selection acting on birth size . A parallel situation may be occurring in plants. The probability of seedling establishment increases with seed size . Seed size is strongly correlated with traits evident later in ontogeny, such as plant height and stem size [2,7,43]. Many herbaceous lineages develop woodiness on islands, and recent research suggests that an increase in leaf size is common [8,9]. As a result, it could be that selection first acting on seeds may facilitate evolutionary changes at later-life-history stages.
Our results suggest that selection favours increased seed size on islands, regardless of dispersal mode, growth form and evolutionary history. Several processes may explain this macroevolutionary trend. First, increasing seed size may reduce propagule mortality associated with unfavourable dispersal into the ocean (see [9,11]). Second, it may provide competitive advantages post-dispersal and increase the likelihood of establishment. The sizes of plant traits are also known to scale allometrically with one another , so selection acting on seeds may facilitate size changes in other traits and this deserves further attention. Direct investigations of potential processes are now needed, in addition to global analyses of seed size to establish whether the observed pattern is universal.
For data accessibility, see the electronic supplementary material, S2.
The Wellington Botanical Society and the Arnold and Ruth Dench New Zealand Botanical Award provided funding.
We thank the Hokotehi Moriori Trust for granting ethical approval for data collection on Chatham Island. We also thank Rafael Barbieri for statistical advice and two anonymous referees for their comments on previous versions of this manuscript.
- Received March 20, 2014.
- Accepted April 17, 2014.
- © 2014 The Author(s) Published by the Royal Society. All rights reserved.