For species-rich lineages, phylogenetic reconstructions from published studies were used to gather information on the character states of fruit type and growth habit inferred for ancestors of these lineages, as is typically included in this type of study . For the few radiating lineages for which phylogenetic information has not yet been generated, character states of ancestors were either inferred from taxonomic treatments of island and mainland species or excluded from the analyses . In the case of population genetic studies, data on the distribution of genetic variation among hierarchical levels for a given species were extracted from each publication, typically presented as analysis of molecular variance results, or reanalysed with this method from information provided by the corresponding author on request . Data collection in this case was restricted to woody island taxa to avoid potential biases in the distribution of genetic variance within groups as a result of the inclusion of different growth forms . Only species with widespread distributions were selected to investigate the effect of multiple island colonization on population genetic structure. To reach a reasonable sample size for each type of lineage, we broadened our search to other oceanic archipelagos, plastic pots for planting with a few examples from the Azores and Cape Verde Islands being included for this analysis.Our study focused almost entirely on angiosperms as they represent the most diverse plant group of oceanic island floras.

Nevertheless, three gymnosperm lineages were also considered for the Canary Islands because of their wide distributions. Because the aim of the study was the analysis of factors related to speciation, only endemics were considered, thus avoiding potentially recent introductions . For the Canary Islands, monotypic lineages also included those for which populations have been documented from other Macaronesian islands, but are not found on the mainland. One of the potential biases of our approach is that the number of endemic species might not accurately represent the pattern of diversification of a given lineage, for example if extant monotypic lineages were more diverse on the islands in the past, but experienced extinctions; or if species with former island–mainland distributions went extinct in mainland areas and thus are not island endemics that originated by in situ speciation. Because fossil records are scarce for oceanic islands, and the potential bias caused by species extinctions is thus difficult to address, we followed the simple approach adopted in most analytical studies of assuming that monotypiclineages are more parsimoniously explained by anagenetic events of speciation rather than cladogenesis followed by multiple extinctions.Two qualitative characters associated with diversification rates and species richness in several studies were chosen for the present study. We chose qualitative rather than quantitative traits because the former are readily available from the large number of phylogenetic studies of island lineages including this type of data , and thus can be generated for large taxon sample sizes.

In addition, qualitative data provide a straightforward way to test for associations between particular trait combinations . To assign a character state to extant species, we followed the taxonomic descriptions compiled in the most comprehensive island floras of each archipelago: Wiggins & Porter for Galápagos; Wagner, Herbst & Sohmer for the Hawaiian Islands; and the compilation by Bramwell & Bramwell for the wild flora of the Canary Islands. For ‘growth form’ , the ‘woody’ condition included trees, shrubs, subshrubs, woody vines and other arborescent or suffrutescent plants. Previous studies have employed a phylogenetic context to analyse evolutionary shifts in growth habit between extant island endemics and their ancestors, but such cases were restricted to particular lineages or to a given island system, e.g. Macaronesia . In this study, we aimed to compare the information available from phylogenetic analyses among the three selected archipelagos. We followed the rationale of earlier studies analysing the evolution of growth form: woodiness is associated with relatively long generation times and other factors that could have an effect on speciation patterns . Scoring of character states in each case followed the descriptions provided in each phylogenetic study, where suffrutescent plants are typically regarded as woody . For ‘fruit type’, two character states were considered: fleshy fruit vs. dry fruit . FFs included drupes and berries, drupaceous and berry-like structures , and fruits enclosed by fleshy receptacles or calyces . In a few cases, seeds attached to fleshy tissues, such as arils, were also regarded as FFs . DFs, however, included achenes, capsules, pods and nutlets. Fruits were classified into these two categories on the understanding that not all fruits of each character state are alike, and may vary in physical characteristics and dispersal vectors.

For example, DFs of some species include propagules that may be dispersed widely and frequently by abiotic means , whereas those of other species have features favouring vertebrate dispersal . By grouping fruits into the two broad categories of DF and FF, we primarily aimed to test the hypothesis that FFs are generally related to species cohesion and limited diversification, notwithstanding important exceptions.Potential for island colonization was estimated as the number of islands on which each endemic species occurred , following the most comprehensive bibliographic information on species distribution for each archipelago: AcebesGinovés et al. for the Canary Islands; Jaramillo Díaz et al. for Galápagos; and Wagner et al. with modifications from Wagner et al. for Hawai‘i. This variable accounts for the occurrence of at least one successful event of inter-island dispersal for the establishment of new populations, with self-incompatible or dioecious plants representing simultaneous or separate dispersal of multiple individuals. We appreciate that the lack of occurrence of some species on particular islands may have as much or more to do with historical or extrinsic factors, such as niche pre-emption by earlier colonists, than with dispersal-related traits of the plants in question , and that differences in timing of arrival on islands will be important to include in future studies, as such data, which are especially lacking for monotypic lineages, become more widely available. For Canarian monotypic lineages with populations on other Macaronesian islands, analysis of distributional data was restricted to populations found in the Canary Islands. Other Macaronesian islands show clear floristic affinities with the Canary Islands, but this latter archipelago is considered as a biogeographical unit for the purpose of the present study. Because monotypic lineages with populations in other Macaronesian areas are widespread in the Canaries , the omission of populations outside the Canaries should not affect our analyses oncolonization ability . To analyse potential differences between lineage and fruit types in colonization ability, a mean range size across constituent species was calculated for each species-rich lineage, following Price & Wagner .Analyses of data obtained from phylogenetic reconstructions and population genetic studies support theidea that certain phenotypic traits are related to the diversification of island lineages. The analysis of character states obtained from published phylogenetic studies revealed that some trait combinations were particularly frequent among ancestors of extant species-rich lineages on oceanic archipelagos. Thus, the phenotypic syndrome ‘herbaceous-DF’ was the most common among colonizers that subsequently underwent diversification . In addition, population genetic and distributional data indicate that geographical isolation in lineages with only limited dispersal among islands has been an important pattern of diversification, as suggested by previous phylogenetic studies . In a recent review of the genetic diversity of the Canarian flora, Pérez de Paz & Caujapé-Castells found that levels of genetic differentiation in species-rich lineages were significantly higher than those obtained for monotypic lineages. Similarly, we found that the colonization of several islands by DF species of radiating lineages typically leads to strong genetic differentiation among these geographically separated populations . Oceanic barriers often impose limits to gene flow, particularly when seeds are not efficiently dispersed , drainage for plants in pots and ultimately provide the conditions for allopatric speciation to occur . Distributional ranges of species illustrate the importance of geographical isolation in lineage diversification, insofar as most species of radiating lineages are confined to one or a few islands and sister species are often allopatric .

Interisland colonization of species with poor dispersal abilities may thus have promoted allopatric speciation in some cases, although multiple factors, including major ecological shifts, have been implicated in both among and within-island differentiation of extensively diversified lineages , as is probably reflected by frequent parapatry of sister species in Hawai‘i . Thus, mechanisms of speciation acting at the island level may be of greater or lesser importance in the complex pattern of differentiation observed in most radiating island plant groups . Although the available evidence seems to suggest a significant association between particular traits and speciation events, it often proves difficult to infer causality, because shifts in a given trait could be either the consequence or the cause of speciation . Stasis in a given character with a suspected effect on speciation could indicate a causal relationship, because the possibility of phenotypic shift as an evolutionary consequence can be ruled out. In the present case, each of the analysed life history characters showed a different pattern. Growth habit was shown to be labile during the course of lineage diversification. For instance, many island radiations illustrate that the herbaceous condition of the first colonizers shifted to woodiness early in the process of diversification, as all phylogenetically derived species display a woody condition in these groups [e.g. silverswords , Silene L. and Geranium L. in Hawai’i or Sideritis L. and Argyranthemum Webb in the Canary Islands]. The high proportion of extant woody species , when compared with the inferred herbaceous condition of most putative ancestors, supports the idea that woodiness was strongly selected for in the island setting . In contrast, phylogenetic studies of some plant groups suggest that the herbaceous condition is derived in some species of radiating lineages, thus showing that reversals in habit also occur during the evolution of species-rich lineages . In the case of growth form, it is therefore difficult to discern whether shifts in character states were consequences of speciation or, alternatively, were involved in lineage divergence. A different pattern is shown by fruit type, as broadly defined here to include two character states: FF and DF. With relatively few exceptions, including the Hawaiian lobelioids and mints and the Bencomia alliance in the Canary Islands , fruit types have not undergone dramatic change between these states during the evolutionary history of most radiating island lineages. Unlike growth form, comparison of inferred character states in ancestors and those observed in the descendant species-rich lineages suggests that relative stasis in fruit type since island colonization has been the rule for most of these lineages. Carlquist noted that Hawaiian species often show modifications in fruits and seeds when compared with their mainland relatives, however, with changes related to quantitative characters apparently involved in dispersibility . Notwithstanding the generality of our trait categorization, high levels of diversification in certain oceanic island lineages appear to be mainly associated with remarkable phenotypic differentiation in vegetative rather than fruit characters , probably because the genetic architecture of vegetative characters makes them more easily adaptable to environmental shifts, and phylogenetic constraints for these characters are therefore weak . Our estimates of ancestral characteristics, however, should be considered with caution. Because ancestral states of species-rich lineages were obtained from inferences of phylogenetic analyses, these estimates are subject to the same caveats that potentially affect any phylogenetic approach. The extinction of closely related species and the limited sampling of mainland relatives might affect the accuracy of ancestral character state reconstructions typically reported in phylogenetic studies. For example, increased sampling of mainland taxa in a molecular phylogenetic analysis of the endemic Pericallis D.Don in Macaronesia suggested that the ancestral life form was woody rather than herbaceous, although the heterogeneity in growth form in this genus complicated an accurate inference of the ancestral state for the island group . However, fruit type represents a different case for character state reconstructions. Because this character has apparently been phylogenetically constrained since island colonization in most lineages, limitations imposed by phylogenetic inferences probably do not have a substantial effect on our conclusions in this case . Furthermore, population genetic data provide an alternative line of evidence to that obtained from phylogenetic inference , which also suggests that fruit type, in addition to other factors, is relevant to the diversification of extant species-rich lineages.Our analyses showed that FFs are generally associated with a high potential for inter-island colonization .