Evolutionary winners are ecological losers among oceanic island plants

Adaptive radiation, in which successful lineages proliferate by exploiting untapped niche space, provides a popular but potentially misleading characterization of evolution on oceanic islands. Here we analyse the respective roles of members of in situ diversified vs. non‐diversified lineages in shaping the main ecosystems of an archipelago to explore the relationship between evolutionary and ecological ‘success’.


| INTRODUC TI ON
Oceanic islands are model systems for dissecting the interaction between evolutionary and ecological determinants of community assembly and composition (Emerson & Gillespie, 2008;Gillespie, 2004;Whittaker & Fernández-Palacios, 2007;Whittaker et al., 2017). The extreme isolation of most oceanic archipelagos limits the functional and phylogenetic variety of species able to colonize, opening evolutionary opportunities to diversify for those that do (Buira et al., 2020;Whittaker et al., 2017). Typically, in plants, a high proportion of species derives from the diversification of just a few colonists (Price et al., 2018;Price & Wagner, 2011). Such evolutionarily successful radiations are disproportionately derived from herbaceous colonists and include Hawaiian Bidens, silverswords (both Asteraceae) and lobeliads and Macaronesian Aeonium (Crassulaceae) and Sonchus (Asteraceae) Knope et al., 2012;Whittaker & Fernández-Palacios, 2007). These clades can be considered evolutionary winners, often exhibiting speciation rates per unit area far above those observed in mainland settings . Such radiations are conventionally interpreted as a process of repeated specialization to the island environment, with species inhabiting novel niches and new geographic areas: implying that ecological and evolutionary success go hand in hand. But do these radiations really dominate the ecosystems of their constituent islands ( Figure 1)?
Local communities, whether continental or insular, are typically dominated by just a few species, 'ecological winners' that impose their influence, in terms of demography, architectural structure, energy allocation, biomass production, water and nutrient cycling, on the majority of the species present (Lomolino et al., 2017;Ricklefs, 2009). The remaining species are rare in the community and mostly have less ecological and structural importance, although they can be fundamental to certain plant-animal interactions (Jain et al., 2014).
Species rarity/success has been categorized by Rabinowitz (1981) into three main axes: (a) geographic range size, (b) abundance within communities locally, and (c) habitat breadth within the geographic range (which in turn influences site occupancy within the range). We might consider those species successful in at least two axes to qualify as ecological winners (Figure 1).
An early macroecological generalization is that large-ranged species tend to be locally abundant, while small-ranged (locally endemic) species are locally rare (Andrewartha & Birch, 1954;Lomolino et al., 2017) and associated with marginal habitats (Gaston et al., 2007): thereby placing these two groups of species at opposite are abundant only in marginal habitats and could, therefore, be considered ecological losers. Within this particular oceanic archipelago, and we posit within at least some others, evolutionary success in plants is accomplished predominantly at the margins.

K E Y W O R D S
Canary Islands, colonization, community assembly, diversified lineages, ecological losers, island evolution, Rabinowitz rarity, vascular plants, vegetation plots F I G U R E 1 Characteristics of ecological and evolutionary winners and losers. Circles represent single species. Left: all combinations of "winner" and "loser" states for the three performance categories in the table above provide for eight states, seven of which are forms of rarity sensu Rabinowitz (1981), the most extreme of which are "loser" (yellow sphere) in all three senses and "winner" (green) in all three. Right: evolutionary success might be considered marked by combinations of high rates of diversification, the occupancy of novel habitats and the development of novel adaptations corners of Rabinowitz's system. However, sometimes small-ranged endemics are locally abundant (Lesica et al., 2006), outperforming much larger-ranged congenerics. Additionally, species of limited geographical range are often found to be habitat specialists and thus may dominate in limited habitat patches within landscapes (Lavergne et al., 2004). Notwithstanding that different forms of rarity may not therefore always covary in a simple way, in general, the distribution of species along each rarity axis is strongly skewed toward rarity: this is also true of oceanic island biotas, including the Canary Islands (Martín, 2009).
The Canaries provide a uniquely suitable laboratory for analysing the relationships between evolutionary and ecological success, because (a) their flora is well studied and (b) a set of 2001 relevés provides standardized community composition data across islands and Major ecosystem types (METs). This has enabled us to distinguish members of non-diversified from diversified lineages (NDL vs DL) and to assess them quantitatively by life form (Table 1) and MET against Rabinowitz's (1981) three rarity axes: spatial range attributes (which we assess three ways: number of islands, spatial occupancy and maximum geographic extent of occupied grid cells), habitat breadth (hypervolume in standardized temperature/precipitation space) and local species cover (abundance in the six MET).
Here, we use a unique eco-evolutionary dataset of phylogenetic and halved the archipelago's area, with consequences such as the fusion of the islands of Lanzarote, Fuerteventura and satellite islets in a single landmass called Mahan, the emersion of several submarine banks and the diminution of the distance to the Africa. These profound changes have doubtless influenced colonization probabilities and the distribution of endemic species (Fernández-Palacios, 2016;Rijsdijk et al., 2014;Weigelt et al., 2016). Within the Canaries, there are more than 12,500 terrestrial and 5500 marine species in, or around, a land area of only 7500 km 2 , from which about 3800 species and 113 genera are endemic (Arechavaleta et al., 2010). The endemics include many examples of spectacular plant radiations (e.g. within Aeonium, Argyranthemum, Cheirolophus, Echium, Limonium, Lotus, Pericallis, Sideritis and Sonchus).
The Canaries are affected by a Mediterranean-type climate and exhibit an outstanding ecosystem diversity. Steep climatic gradients occur, especially on the northern, more humid, slopes of the five higher islands (>1450 m), whereas the two eastern islands (Lanzarote, Fuerteventura) are lower and exhibit less environmental variation. Six major terrestrial ecosystem types (METs) have been described for the Canary Islands. Five form a zonal cline from coast to high mountain Fernández-Palacios, 1992;Fernández-Palacios & de Nicolás, 1995): sub-desert coastal Euphorbia scrub (found on all seven islands); thermophilous woodlands (seven islands); laurel forest (five islands); pine forest (five islands); summit scrub (three islands), and rock communities that can be found at any elevation. Other azonal vegetation types, such as coastal halophytic communities, sand dune habitats or wet habitat can be distinguished, but are too localized to be distinguished as separate entities for the present analyses. Short descriptions of each ecosystem can be found in del  and Zobel et al. (2011). The full array of community types that comprise each MET is provided in Table S1. Natural vegetation has been severely disturbed by human activities, beginning with the first colonization of the archipelago by Berber tribes from North Africa ca. 2400-2200 yr BP TA B L E 1 Life forms (%) of species in diversified lineages (diversification) and in non-diversified lineages (colonization). We use a modified Raunkiaer (1934)

| Community inventories
To obtain information about species contributions to ecosystem assembly and structure, we collated 2001 vegetation plots covering all major habitats and islands, extracted from the online database SIVIM (Information System of the Iberian and Macaronesian Vegetation: http://www.sivim.info/sivi/). This database includes most of the published phytosociological surveys on the Canary Islands in the last 50 years. Some additional plot data from other published sources have been used (Cabrera, 2020;von Gaisberg, 2005;Méndez, 2010;Santos-Guerra, 1983;Stierstorfer, 2005). The common sampling protocol refers to the standard phytosociological methodology (Braun-Blanquet, 1964). Only the most natural vegetation types (phytosociological associations of potential vegetation series), but not secondary or substitution communities, were included, following the criteria of del . Only associations with at least five plots were included in the analyses. Surveys of a number of different phytosociological associations were merged to the corresponding six METs (see Table S1). Other marginal azonal vegetation types, such as dunes or coastal marshes, were not taken into account. Following this approach, we obtained the mean proportion being of low elevation and relatively degraded they are also ecologically simpler than the other islands.
Plots for phytosociological surveys (relevés) are subjectively selected in representative patches of the target communities (Braun-Blanquet, 1964), which leads to a certain bias towards wellpreserved sites. However, this possible bias is countered, in our case, by the large number of plots representing accurately the composition and structure of the main vegetation types. The data for each site included: geographical characteristics (geographical coordinates, elevation, slope and area sampled), plus the cover for each species present, using a semi-quantitative index of its contribution.
The so-called "abundance-dominance" index used in data collection, is based on the coverage, defined as the percentage of plot area covered by the projection of the aerial organs of a specific species within the plot. The index varies from r to 5, where: r = one individual with negligible coverage; + = more individuals, but a very low cover; 1 = up to 5% cover; 2 = 5-25% cover; 3 = 25-50% cover; 4 = 50-75% cover and 5 >75% cover). The phytosociological plots analysed vary in size (average: 260.5 m 2 , range: 4-2000 m 2 ) depending on the type of vegetation and size of its species pool, although the proportion of species cover is not influenced by plot size or species richness.
The different METs have different overall levels of vegetation cover. For example, the laurel forest typically reaches 100% cover, but mature communities of coastal and summit shrublands, depending on the site, may have total cover values of 30-50% of the plot area. We calculated the relative cover of species of diversified lineages with respect to the total vegetation cover in the plot, which can exceed 100% in the case of sites with tree and shrub layers.
For each phytosociological relevé and habitat type, we calculated separately the relative cover of all species within diversified and non-diversified lineages by substituting the relevé abundance code by the mean cover value of the abundance range (+, r = 1%; 1 = 5%, 2 = 15%, 3 = 37.5%, 4 = 62.5% and 5 = 87.5%). The rarest species are arguable slightly overrepresented by this approach, which therefore makes our analysis relatively conservative towards low-scoring species.

| Species data
In compiling our data, we used the most recent checklist of the

| Rarity dimensions
When analysing the rarity of diversifying vs. non-diversifying lineages we distinguished among three dimensions of rarity, following Rabinowitz (1981): (a) spatial or geographic rarity, that is, wide vs. narrow distribution; (b) local community abundance or demographic rarity, that is, abundant vs. scarce species, according to the cover which the species contribute to a given community and (c) habitat breadth or ecological rarity, that is, generalist vs. specialist species, here represented by the climatic niche width.
Spatial rarity was assessed by three different indices: the number of islands occupied, the geographic extent and the spatial occupancy across the Canaries, although in most analyses only the latter two were used. From the Canary Islands checklist of vascular plants (Acebes Ginovés et al., 2010), we extracted the number of islands a species is present on. Geographical extent and spatial occupancy were based on occupancy data across a grid of 500 × 500 m cells from the Canarian Government Atlantis 3 Biodiversity bank (Martín et al., 2005). The geographical extent for each species was calculated as the maximum linear Euclidean distance between the grid cells occupied by that species within an island for those species restricted to a single island, or, for all other species, within the archipelago, including the distance between islands, based on the UTM zone 28 coordinates of the mass centroids of the grid cells. Spatial occupancy was expressed by the total number of grid cells a species has been reported in. We used only those species occurrences with geo-referenced UTM coordinates or other accurate spatial information about the specific geographic location, in order to guarantee a high level of data quality, that is, the accuracy levels 1 and 2 in Martín et al. (2005).
The species community abundance value was defined as the maximum relative cover of the target species, expressed as the highest mean cover value recorded among all plots in a phytosociological association belonging to one of the selected six MET on any of the Canary Islands. First, we calculated the mean cover value, that is, the sum of cover values divided by the number of total plots in the association  (Blonder et al., 2014). The band width was fixed at the mean of all band widths estimated for the individual species based on the estimate_bandwidth function (Blonder et al., 2014).
The four rarity parameters (those listed above with the exception of number of islands) were standardized to a scale of 0-1 for purposes of comparison. Furthermore, a novel overall rarity index for each species was defined as the mean value of the four standardized rarity parameters, with the two spatial metrics (geographic range and spatial occupancy) each weighted by 0.5 and the rest by 1.

| Data analyses
Kernel density plots of species distributions were calculated for all four rarity parameters (geographic range, spatial occupancy, community abundance and climatic niche) and for overall rarity using bootstrap approach (Crawley, 2007). We tested if the mean proportion of the relative cover of species belonging to diversified lineages in each MET on a certain island was significantly different from its proportion in the island's species pool, that is, if they are under-or overrepresented in the local community. We took 10,000 random samples with replacement from the total number of cover values for each species group and plant assemblage. Then, we calculated 95, 99 and 99.9% confidence intervals with lower and upper bounds of the cumulative probability distribution of the 10,000 bootstrapped mean values of the random samples. If the confidence interval lay below or above the value of the proportion of the island species pool, then the species group is deemed to be under-or overrepresented in the local community, respectively.
The test was performed for each MET at the island and the archipelago level using the R-package "bootstrap." To test for linear relationships between the proportion of relative cover of diversified species in community plots and plot elevation, slope and community tree cover, we applied generalized linear mixed models (GLMM) with a binomial error distribution, as recommended for continuous proportional data (Crawley, 2007).
We added geological age of the island as a proxy of evolutionary time, which is hypothesized to be a key determinant of overall island richness and endemism by the "General dynamic model of oceanic island biogeography," which postulates a humped relationship of these metrics with island age (GDM) (Whittaker et al., 2008). To visualize the effect of each variable on the distribution of diversified species in local communities we present partial residual plots of GLMMs on the scale of the original response after applying the inverse link function.
The proportion of plot cover of radiated species was introduced in the model as the dependent variable and elevation, slope, tree cover and island age (including a quadratic term) as independent variables. Island was treated as a random effect. Out of the total of 2001 plots, 1839 plots (92%) with information about slope, elevation and tree cover were used to perform the model. Model performance statistics such as generalized R 2 -value and explained deviance are reported. Analyses were performed using the R-libraries "lme4," "MuMin" and "visreg."

| Diversification and life forms
The Canarian native vascular flora comprises 939 species (

| Species rarity
For each rarity metric, whether considering the entire flora, or just perennials, a consistent pattern of greater rarity of DL species emerges ( Figure 2, Table 2 ). The greatest differences are for aspects of geographical distribution, especially geographical range, which is strongly constrained by the geographical distribution of species among islands (Figure 2a,e,i). Notably, 11% of NDL species are single-island endemics, contrasting with 60% of DL species (44% for moderately diversified lineages, 71% for highly diversified lineages) ( Figure 2i). The NDL group shows two peaks of island occupancy, for five islands (mostly species present in the five ecologically similar, topographically high, central/western islands) and seven islands (whole archipelago). Combining mean values of geographic range (weight 0.5), spatial occupancy (0.5), local abundance (1), and climatic niche (1) into a composite rarity index, we find that overall rarity also increases with the degree of diversification (Figure 2j,k,

| Community dominance
To further evaluate Rabinowitz's (1981) community abundance axis,  Table S2). The GLMM also showed a (weak) humped relationship with the geological age of the island (i.e. time for speciation).
To test the "refuge hypothesis" that the high number of DL species in rock communities is ecologically equivocal, being the result of human disturbance, we applied non-parametric Wilcoxon tests which showed that non-native species are not F I G U R E 2 Canarian plant species distributions among five rarity parameters: geographic range (a, e), spatial occupancy (b, f), community abundance (c, g), climatic niche (d, h) and distribution among islands (i)) and overall rarity (mean value a-d and e-h, while a, b and e, f weighted 0.5). All species (panels a, b, c, d, i, j), and perennials only (e, f, g, h, k). Density plots generated with Gaussian kernel smoothing, where the total area under the curve integrates to one. Community abundance and occupancy values were transformed (log, square root) to improve visual comparison. Community abundance refers to the maximum cover a species obtains among all plots of one of the six selected METs. See Table S2 (and Methods) 2010), with the exception of the rock communities, which may be found at any elevation. The proportion (%) of area for each MET across the whole archipelago is indicated in the legend. Boxplots represent the relative cover of diversified species in vegetation plots of each community. All communities revealed significantly lower or higher levels than expected by species pool proportions (see Table 3). Data for Lanzarote and Fuerteventura were merged because they formed a single island (Mahan) during extensive low-sea-level stands during the Pleistocene glacial periods perceptibly different from European plants, others constantly, and these are less numerous and less common. Then come the distinct species; of these some would be called varieties by many botanists, and others good species by all; these are still less common." Later in the article he noted: "The lofty mountains of the Canaries, though upwards of 11,000 feet high, contain no alpine plants, and as in the case of the Madeiran group, many of the most peculiar forms are extremely rare and local." Here, for "peculiar" we would today write "endemic." Summing up for the Atlantic islands as a whole Hooker stated "As a general rule, the species of the mother continent are proportionally the most abundant, and cover the greatest surface on the islands. The peculiar species are rarer, the peculiar genera rarer still." These excerpts serve to highlight that Hooker was aware of and thought noteworthy the existence of relationships between biogeographical distribution and local rarity that we quantify here and which we show to be related to in situ cladogenesis.

| Is rarity cause or consequence of diversification?
While the localized distribution of many island endemic plant species is well known (e.g. Barquín & Voggenreiter, 1988;Brochmann et al., 1997;Hooker, 1867;Price & Wagner, 2004), this is the first empirical demonstration that evolutionary winners are ecological losers using multiple dimensions and quantifications of rarity for the entire flora of an oceanic archipelago. This finding raises the further question: is diversification a cause or a consequence of rarity? If most DL species (evolutionary winners) are comparatively young, evolving within a rich flora, the new species may have had less time and opportunities to spread and become abundant compared to the early colonizers: implying species rarity as a consequence of diversification. Rare species inhabiting stressful habitats (ecological losers) may face low competition from other species and/or may diversify because they occur in patchy habitats with small, isolated populations that have limited between-patch genetic flux: implying diversification as a consequence of species rarity. A reciprocal interplay, whereby diversification is both a cause and consequence of different forms of rarity may reinforce the effects of these processes, as has been posited for the Hawaiian flora (Price & Wagner, 2004) and demonstrated for Samoan and Hawaiian National Parks, where local abundance is lower for species in larger lineages (Ibanez et al., 2021). In the Canaries, the direction of causality appears to vary among zones.

| Diversification and habitats
The lowland METs (Figure 3) have been filled mostly by colonization of NDL species from nearby NW Africa (including the so-called Rand Flora palaeoendemics: Campylanthus, Ceballosia, Justicia, Kleinia, Neochamaelea, Plocama (Pokorny et al., 2015)) or more recent colonizers, such as Euphorbia balsamifera or E. canariensis and, in some cases but mostly within the understory, by in situ radiation   Figure 3).
Finally, the summit scrub seems an exception since it is the only zonal ecosystem dominated by DL endemics (Figure 3). Insular high mountains (sky islands), are known to be hotspots of endemism . These habitats are ephemeral, very scarce and isolated in time and space with respect to environmental conditions and source regions . It is not colonization by pre-adapted species from similar environments from the mainland, but in situ speciation from sources originating at lower elevations that has dominated community assembly of the summit scrub. Although an azonal ecosystem and thus located across the elevational gradient, the rocky cliffs MET is consistently (and across all seven islands) characterized by DL species such as Aeonium and Sonchus, which dominate both species lists and community cover.
These habitats, products of local erosion processes, are characterized by geomorphological dynamism and lack of soil: they are environmentally harsh and typically feature low levels of vegetation cover.

| Biogeographic and ecological aspects of diversification
Our findings challenge previous claims of niche pre-emption (Silvertown, 2004) or priority effects (Shaw & Gillespie, 2016), which posit that in which competition is relatively low (Buira et al., 2020;Kontopanou & Panitsa, 2020;Lavergne et al., 2004). This supports the idea that the availability of patchy ecological opportunities encourages diversification (Stroud & Losos, 2016). In the Canaries, rock communities are patchy in space and time and are distributed among all main ecosystems, covering the whole elevation gradient. They thus provide a network of ephemeral spaces across topographically complex landscapes encompassing diverse climatic conditions, permitting geographic and reproductive isolation, which drives speciation (Otto et al., 2016).
The successful diversification in the Canarian flora has mostly involved shrub life-forms (Lens et al., 2013). Moreover, several of the largest radiations involve the classic island pattern of continental herbaceous ancestors producing clades containing several derived (cladogenetic) woody descendants, often involving multiple evolutionary transitions (Lens et al., 2013). The origin of the secondary woodiness within the Canary flora is still debated, but might be related to the rocky, open, relatively competition-free habitats and the dominant Mediterranean climate (see e.g. Frankiewicz et al., 2020;Lens et al., 2013;Whittaker & Fernández-Palacios, 2007).
An alternative explanation is that the high number of DL species in rock communities is ecologically equivocal, being the result of human disturbance (refuge hypothesis). However, further analysis (Table S3) did not support this interpretation, as the core zonal ecosystems did not have higher cover of non-native species or fewer threatened plant species than steep rock habitats (Otto et al., 2016).

| CON CLUS IONS
We have shown that DL species, endemic products of withinarchipelago diversification, are mostly rare in all three rarity axes (sensu Rabinowitz, 1981) and typically do not play an important role in structuring plant communities on the Canaries. It appears to be the rocky, high elevation, marginal and ephemeral habitats that have fostered much of the in situ diversification. Compared to NDL species, cladogenetic endemics show higher levels of rarity in geographic range, habitat affinity and local community abundance, which alongside other island plant syndromes (Burns, 2019) make them more susceptible to human extinction drivers such as habitat loss and alterations or climate change (Otto et al., 2017).
Only further research can establish the generality of the patterns we describe here. However, we consider that they will not be unique to these Macaronesian archipelagos and we postulate: evolutionary winners are typically ecological losers among island plants.

DATA AVA I L A B I L I T Y S TAT E M E N T
Vegetation inventory data of the phytosociological associations assigned to the six major ecosystem types for each of the Canary Islands were obtained from the webpage "Sistema de Información de la Vegetación Ibérica y Macaronésica" (http://www.sivim.info/sivi/).
Finally, data of the Canarian meteorological stations were obtained from the "Agencia Española de Meteorología" webpage (http:// www.aemet.es/es/portada). The maps of interpolated climate data and the input files for all statistical analyses are deposited in the DRYAD repository (https://doi.org/10.5061/dryad.qrfj6 q5fs).Data will be released on 31th of May 2022.