Island woodiness underpins accelerated disparification in plant radiations

Summary The evolution of secondary (insular) woodiness and the rapid disparification of plant growth forms associated with island radiations show intriguing parallels between oceanic islands and tropical alpine sky islands. However, the evolutionary significance of these phenomena remains poorly understood and the focus of debate. We explore the evolutionary dynamics of species diversification and trait disparification across evolutionary radiations in contrasting island systems compared with their nonisland relatives. We estimate rates of species diversification, growth form evolution and phenotypic space saturation for the classical oceanic island plant radiations – the Hawaiian silverswords and Macaronesian Echium – and the well‐studied sky island radiations of Lupinus and Hypericum in the Andes. We show that secondary woodiness is associated with dispersal to islands and with accelerated rates of species diversification, accelerated disparification of plant growth forms and occupancy of greater phenotypic trait space for island clades than their nonisland relatives, on both oceanic and sky islands. We conclude that secondary woodiness is a prerequisite that could act as a key innovation, manifest as the potential to occupy greater trait space, for plant radiations on island systems in general, further emphasizing the importance of combinations of clade‐specific traits and ecological opportunities in driving adaptive radiations.

Methods S1 Clade specific tree and trait data -Phylogenetic reconstruction and age estimation (BEAST), fossils used for calibration, full references for trait data.
Notes S1 Hypericum traits -details on potential biases using mean plant height in the group. Table S1. Voucher -Information on taxa included in the study detailing species names per study group, clade assignment, mean plant height, coding of distribution (i, island; n, non-island) and life history traits (h, herbaceous; w, woody), and ENA/NCBI accession numbers of molecular marker used for phylogenies newly reconstructed in this study.

Methods S1 Clade specific information, phylogenetic reconstructions and trait data
All data used in this study (DNA alignments, xml BEAST input files, phylogenetic trees, trait data tables) are available on Dryad (https://doi.org/10.5061/dryad.rt530k9).
Echium -The genus Echium L. comprises ca. 68 species native to North Africa, mainland Europe and the Macaronesian islands. Early phylogenetic work demonstrated the monophyly of the Macaronesian species and characterised this island radiation as a classical example of insular woodiness and disparification of insular growth forms (Böhle et al., 1996). More recently published phylogenies of Echium either contain too few non-island taxa (García-Maroto et al., 2009) or lack a timeline (Romeiras et al., 2011). In order to provide a time-calibrated comparative framework encompassing the island and mainland diversity for evolutionary analyses we generated a new timecalibrated phylogenetic tree for all major lineages of Boraginaceae with as complete sampling as possible of the genus Echium, using data for two chloroplast markers (rbcL and trnLF) and two nuclear markers (ITS and D6DES) downloaded from GenBank (GenBank accession numbers are provided in table S1). Sequences were aligned using MUSCLE v3.6 (Edgar, 2004) and alignments were visually crosschecked and ambiguous fragments were excluded using Geneious v6.0.5 (Kearse et al., 2011). To test on supported topological conflict between the individual gene trees phylogenies were inferred for each individual marker using RAxML v8 (Stamatakis, 2006) with a GTR + Γ substitution model and rapid bootstrapping with 100 iterations. Since no supported conflicts were present, we concatenated the alignments prior to estimating the topology and divergence times simultaneously using BEAST v1.8.2 (Drummond, AJ et al., 2012) under an uncorrelated lognormal relaxed-clock model, using the general time-reversible (GTR) substitution rate model and Γdistributed rates among sites with a proportion of invariant sites to describe the rate heterogeneity among sites, with a Yule model as tree prior. We used critically selected fossils for calibration: the genus Ehretia P.Browne has a good fossil record from the early Eocene to the Neogene and represents the earliest fossil record of Boraginaceae, therefore we used (1) this set of fossils to set a minimum age of 33.9 Myr (Chandler, 1962;Mai & Walther, 1991;Ozaki, 1991). Other fossils used were: (2) a fossil of Cryptantha Lehm. ex G.Don (Darnell & Thomasson, 2007), and (3) a fossil of Lithospermum L. (Gabel, 1987), both from the late Miocene Ash Hollow Formation in the Ogallala group from c. 10.3 to c. 13.6 Ma; (4) a fossil nutlet and pollen records of Tournefortia L. from the Oligocene (Dorofeev, 1963); and (5) a fossil of Cordia L. from the Eocene (Huzioka & Takahasi, 1970).
The fossils were assigned to clades on the basis of their morphology and the respective nodes were constrained with uniform priors with the minimum age set by the age of the fossil, and the maximum age of 88 Myr. We ran four independent Markov Chain Monte Carlo (MCMC) runs of 50 × 10⁶ generations and checked for convergence between runs, and effective sample sizes of >200 for all parameters in Tracer 1.5 (Rambaut & Drummond, 2007). We combined the four runs in LogCombiner after discarding the first 10% generations (burnin) in each run and we selected and annotated the Maximum Clade Credibility (MCC) tree using TreeAnnotator. The MCC tree was used in all subsequent comparative analyses after pruning the outgroup taxa.
In our time-calibrated phylogeny of Echium all Macaronesian taxa form a monophyletic group, concordant with the topologies of Böhle et al. (1996) and García-Maroto et al. (2009), with all three Macaronesian clades recovered by the latter also recovered in our own phylogeny (Fig. S1).
Our crown age estimate for the Macaronesian clade is slightly younger, ca. 4.2 Ma vs 6.9 Ma, however, the age estimate recovered by García-Maroto et al. (2009) falls within our 95% HPD. Our phylogeny also recovers the Cape Verde Islands relationship of E. stenosiphon as sister to E. hypertropicum + E. vulcanorum as presented by Romeiras et al. (2011).
Hypericum -While the main centre of species diversity for Hypericum L. with around 330 species is in the Old World (Nürk et al., 2015), ca. 170 species occur in the Americas, thereof 120 in Latin America, of which ca. 70 are Andean páramo endemics (NB ca. 99 species in the 'Andean radiation' [clades G, H, I; Fig. S2] due to 'secondary out-of-páramo' dispersals; Nürk et al., 2018), with the 'sky island' here defined as the high elevation páramo ecosystem above the treeline (Nürk et al., 2018). To ensure as complete sampling as possible for the páramo endemics and related taxa, we used a previously published time-calibrated phylogeny, which has extensive sampling of the New World lineages (Nürk et al., 2018). Previous work has demonstrated a close association between arborescence and tropical montane occurrences in African mountains and the Andes (Nürk et al., 2013a) and elevated species richness associated with the very recent Pliocene / Quaternary radiation in the Andes, a radiation encompassing a disparate array of growth forms typical of tropical alpine sky island plant groups (Nürk et al., 2013b).
Lupinus -The legume genus Lupinus L. comprises ca. 290 species, of which ca. 190 occur across Mexico to South America, thereof ca. 85 across the Andes mountain range (Hughes & Eastwood, 2006;Drummond, CS et al., 2012;Nevado et al., 2016). We used the best currently available time-calibrated phylogeny (Drummond, CS et al., 2012), pruned to the robustly supported large western New World clade containing c. 220 species found in western North and South America (i.e. clades I to S of . Previous work demonstrated significant associations between shifts from annual to perennial life histories, lowland to montane habitats and accelerated rates of species diversification nested within this clade, with these three coincident shifts subtending a large western New World montane 'super-radiation' within which is nested the Andean radiation (Drummond, CS et al., 2012). Like the Andean radiation of Hypericum, Andean Lupinus also includes diverse growth forms (Hughes & Eastwood, 2006;Hughes & Atchison, 2015). For Lupinus the western New World 'super-radiation' sensu Drummond, CS et al. (2012) and Hughes and Atchison (2015) is defined as the montane sky island clade and the set of lowland western North American annual species that subtend the montane radiations are the non-island relatives.
Trait measurements were obtained from field observations, herbarium specimen data and literature (Gladstones, 1974;McVaugh & Anderson, 1987;Barneby, 1989;Baldwin et al., 2012).  (Baldwin et al., 1991;Baldwin & Sanderson, 1998;Baldwin & Wessa, 2000). Like Echium, the Hawaiian silversword radiation shows evidence of insular woodiness and disparification of insular growth forms including shrubs, treelets, lianas, acaulescent and candelabra rosettes (Baldwin, 1997) with a smaller radiation with parallel evolution of insular woodiness on the Californian islands in the genus Deinandra Greene (Baldwin, 2007), while the vast majority of the continental tarweeds are ephemeral low elevation herbs occupying seasonally dry mainland habitats. Although a suite of phylogenies that sample members of the Madiinae have been published (Baldwin et al., 1991;Baldwin, 1997;Baldwin & Sanderson, 1998;Barrier et al., 1999;Baldwin & Wessa, 2000;Baldwin, 2007) none included the comprehensive sampling of both mainland and island species needed for comparative analysis of species and trait diversification rates (but see: Landis et al., 2018). We generated a time-calibrated phylogenetic tree for the Silverswords-Tarweeds using the ITS marker for as many members of Madiinae with available data in GenBank (GenBank accession numbers are provided in table S1). Sequences were aligned using MUSCLE v3.6 (Edgar, 2004). Alignments were visually crosschecked and ambiguous fragments were excluded using Geneious v.6.0.5. We simultaneously estimated the topology and divergence times using BEAST v1.8 under an uncorrelated lognormal relaxed-clock model, using the general time-reversible (GTR) substitution rate model and Γ-distributed rates among sites with a proportion of invariant sites to describe the rate heterogeneity among sites. In the absence of available fossil calibrations, we used a calibration constraint of 15 Ma for the most recent common ancestor of the silversword alliance and its Californian sister group, as justified by Baldwin and Sanderson (1998) and later used by Baldwin (2007). While this approach may not provide accurate estimates of absolute divergence times (but cf. Landis et al., 2018), this will not affect our downstream rates analyses as our inferences are based on relative divergence times. We ran four independent MCMC runs of 50 × 10⁶ generations and checked for convergence between runs, and effective sample sizes of >200 for all parameters in Tracer 1.5. We combined the four runs in LogCombiner after discarding the first 10% generations (burnin) in each run before annotating the MCC tree using TreeAnnotator. The MCC tree was used in all subsequent comparative analyses after pruning all subspecies.

Silverswords-Tarweeds
Our time-calibrated phylogeny of the Madiinae is concordant with previous phylogenetic hypotheses in both topology and age estimates, with a crown age estimate for the silverswords of ca. 3.6 Ma and for the tarweeds from CA silands of ca. 1.2 Ma (table 1, Fig. S4). We recovered the Hawaiian silversword taxa as monophyletic matching the phylogenies of Baldwin and Sanderson (1998) and Barrier et al. (1999) and also recovered the CA island clade (Fig. S4) presented by Baldwin (2007). Our finding that the two species of Anisocarpus Nutt. are sister to the silversword clade is concordant with the phylogeny of Baldwin and Wessa (2000) and Landis et al. (2018). We find that within the silversword clade Wilkesia is nested within Dubautia, which together are sister to Argyroxiphium, matching the previously most comprehensively sampled topology for the silverswords of Baldwin and Sanderson (1998), but differs from the groupings recently recovered by Landis et al. (2018), in which Argyroxiphium is nested within Dubautia.
Trait data were obtained from the Monograph of the Hawaiian Madiinae (Carr, 1985), the Jepson manual (Baldwin et al., 2012), the manual of the flowering plants of Hawai'i, (Wagner et al., 1999) and the monograph of tarweeds and silverswords (Carlquist et al., 2003).
Notes S1 Hypericum traits: We are relaying on mean plant height as proxy for growth form evolution, which might introduce bias into comparisons of overall disparity between Hypericum clades because not all species descriptions contain measures of min and max plant height (Robson, 1987;1996). For example, the North American non-island species H. chapmanii W.P.Adams is descried as "up to 4 m tall" in Robson (1996: p 112), whereas the Andean sky island species H. laricifolium Juss. is described as "(0.1-) 0.3-3 (-6) m tall" (Robson, 1987: p 47