Oligocene niche shift, Miocene diversification – cold tolerance and accelerated speciation rates in the St. John’s Worts (Hypericum, Hypericaceae)

Background Our aim is to understand the evolution of species-rich plant groups that shifted from tropical into cold/temperate biomes. It is well known that climate affects evolutionary processes, such as how fast species diversify, species range shifts, and species distributions. Many plant lineages may have gone extinct in the Northern Hemisphere due to Late Eocene climate cooling, while some tropical lineages may have adapted to temperate conditions and radiated; the hyper-diverse and geographically widespread genus Hypericum is one of these. Results To investigate the effect of macroecological niche shifts on evolutionary success we combine historical biogeography with analyses of diversification dynamics and climatic niche shifts in a phylogenetic framework. Hypericum evolved cold tolerance c. 30 million years ago, and successfully colonized all ice-free continents, where today ~500 species exist. The other members of Hypericaceae stayed in their tropical habitats and evolved into ~120 species. We identified a 15–20 million year lag between the initial change in temperature preference in Hypericum and subsequent diversification rate shifts in the Miocene. Conclusions Contrary to the dramatic niche shift early in the evolution of Hypericum most extant species occur in temperate climates including high elevations in the tropics. These cold/temperate niches are a distinctive characteristic of Hypericum. We conclude that the initial release from an evolutionary constraint (from tropical to temperate climates) is an important novelty in Hypericum. However, the initial shift in the adaptive landscape into colder climates appears to be a precondition, and may not be directly related to increased diversification rates. Instead, subsequent events of mountain formation and further climate cooling may better explain distribution patterns and species-richness in Hypericum. These findings exemplify important macroevolutionary patterns of plant diversification during large-scale global climate change. Electronic supplementary material The online version of this article (doi:10.1186/s12862-015-0359-4) contains supplementary material, which is available to authorized users.


Phylogenetic inference
Figure S1 -Phylogeny of Hypericaceae produced from Bayesian analysis of the combined sequence data Branch length is given in units of expected substitutions per side. Accession and clade names are detailed. Bayesian posterior probabilities and ML bootstrap support is given per branch (pp|ML; or pp above and ML below the branches). Rooting follows Xi et al. [1].

Topological discordance
Topological discordance between chloroplast (petD+trnL-trnF) and nuclear (ITS) sequence inference is present only in two places, although without strong support. The first topological discordance concerns basal nodes within Hypericum: H. elodes, H. aegypticum, and H. calcicola are sister to the Triadenum+Myriandra+Brathys s.l. clade in the chloroplast topology, and in the rDNA tree in a grade, consisting of the earliest branchings in Hypericum.
This topological incongruence between chloroplast and rDNA inference has also been reported in other studies (for discussion see [2]). Sánchez Meseguer et al. [3] analyzing three chloroplast marker (psbA-trnH, trnL-trnF, trnS-trnG) and ITS and using a deep sampling report no such incongruence between chloroplast and rDNA sequence inference, suggesting that the 'chloroplast' topology is a sampling issue.
The second topological incongruence has not yet been reported (this is due to missing analyses in the groups of interest using nuclear rDNA) and concerns the position of the other tribes of the Hypericaceae; Vismieae sister to Hypericum (Hypericeae) in the chloroplast topology (a result also shown in other studies, e.g., [1]), and sister to the rest of the family in the rDNA tree. Reasons for this incongruence have not yet been investigated, and one might consider the specific mode of rDNA evolution (concerted evolution causing bidirectional homogenization) as possibly responsible processes. Moreover, Sánchez Meseguer et al. [4] studying low-copy nuclear genes (EMB2765 and PHYC) do also report discordance when compared to the chloroplast topology. Since none of the respective nodes received strong support, we concatenated the chloroplast and nuclear data set.

Age estimation: calibration
The following external calibration points were considered in the divergence time estimations, relying on six fossils which were constrained by hard minimum bounds (i.e. uniform calibrations), and two age estimates reported in other studies which were constrained by lognormal distributions.
1) The age of the root node (i.e. the crown node of the clusioid clade) estimated by Xi et al. .0 95% HPD) was constrained by a log mean of 2.4, a log standard deviation of 0.3, an offset of 78, and an initial value of 89.5.
2) The age of the crown node of the Hypericaceae estimated by Xi et al. [1] to 53.7 Ma  4-33.9 Ma) of West Siberia (Užaniha) [5], considered to be the oldest fossil remain of the genus [6], possess synapomorphies in the general seed morphology and the testa sculpturing with Hypericum. Especially, the characteristic meridional ribs on the testa resemble those of Hypericum sections Elodes, Brathys, Trigynobrathys, and Sampsonia. This fossil is certainly an extinct member of the Hypericum lineage.
Depending on weather the fossil is a member of the crown group or the associated stem group, the age estimation will differ between the assignment to the crown node or to the stem node. Following Magallón and Sanderson [7], a fossil that possesses all apomorphies of a particular (sub)clade within the study group should be assigned to the crown node of that study group since it can be unambiguously identified as a crown group member. If the fossil presents some, but not all apomorphies of particular (sub)clades within the study group, it is assigned to the stem node of the study group.
The reason is that the fossil could belong to a branch that diverged after the evolution of some apomorphies, but before the extant members of the study group diverged. In our case, assigning H. antiguum to the crown group of Hypericum, and thus assigning a time constraint to the crown node is legitimate if it is an extinct member of a certain (sub)clade within the genus. On the other hand, it is also appropriate to assign the fossil to the stem group of Hypericum, if the relationship to the genus but not the relationship to a certain clade within the genus is beyond doubt. Because no obvious relationship of H. antiguum to a certain (sub)clade within Hypericum can be unequivocally hypothesized, the conservative approach when using the fossil as a minimum time constraint (uniform calibration) is to assign the fossil to the Hypericum stem group. We designed two analyses: analysis A which uses the conservative approach and assigns the seed fossil to the stem node of Hypericum (i.e. constrains the stem node to a minimum age of 33.9 Ma) and analysis B which assigns the fossil to the crown node (i.e. constrains the crown node to a minimum age of 33.9 Ma) in order to enable comparison of the ages inferred in our study with the results of a recently published study [3], in which the fossil was assigned to the Hypericum crown node. 4) Hypericum tertiaerum Nikitin fossil seeds [6] from the Lower to Upper Miocene (23.0-5.3 Ma) of East Europe and Siberia were used to calibrate the stem node of Triadenum with a minimum age constraint of 5.3 Ma. 5) Seeds of H. virginicum (= Triadenum virginicum) from the Pleistocene [8], the earliest fossil record from North America [6], were used to calibrate the crown node of the North American part of the Triadenum clade with a minimum age constraint of 0.01 Ma.
8) Fossil Pliocene seeds [5], which show close affinity to the seeds of H. perforatum, were used to calibrate the crown node of 'core Hypericum' to a minimum age of 2.5

Ma.
Two approaches were conducted, which differed only in the assignment of the seed fossil Hypericum antiguum (see above, point 3): divergence time estimation A, which assigns the fossil to the stem node of Hypericum, and divergence time estimation B, which assigns the fossil to the crown node of Hypericum. All other calibrations remained unchanged in the two analyses. Fig. S2 gives the MCC tree produced in analysis A, and Table S1 lists the results of both analyses.

Figure S2 -Maximum clade credibility (MCC) chronogram of Hypericaceae (produced by divergence time estimation A)
Numbers in circles (1)(2)(3)(4)(5)(6)(7)(8) indicate the assignment of external time constraints to the respective node. Node bars indicate the 95% highest posterior density (HPD). Note that in analysis A the time constraint N o 3 is assigned to the stem node of Hypericum (using the "including stem" option; indicated by the black arrow), whereas it is assigned to the crown node in analysis B.

Historical Biogeography
It is necessary to re-analyze the biogeography of the study group, because the parametric likelihood approach employed, i.e. the dispersal-extinction-cladogenesis (DEC) model [12] implemented in the program Lagrange [12,13], takes into account the possibility of multiple states (the areas defined in the analysis) for the reconstructed distribution of ancestral populations. One of the advantages of the DEC model is its flexibility, as it can integrate temporal and paleogeographical information or species dispersal capabilities and ecological tolerance [14] by defining an instantaneous rate matrix (Q) of lineage dispersal between areas and extinction within an area [15]. Since there is no restriction for geographic ranges in the number of ancestral states (beside computational limitations), this approach compensates for the far distribution of several species of Hypericum.

Table S2. Summary statistics obtained by optimization of ancestral areas over 1000 posterior trees generated by age estimations A and B.
Bold font indicates the most favored biogeographic scenario per node reconstructed under the different approaches as indicated by composite Akaike weights (w i ) and evidence ratios (w i /w j ). Up to three alternative scenarios are given unless w i ≥ 0.2, or w i /w j ≥ 3. A, Afrotropical; WP, western Palearctic; EP, eastern Palearctic; IP, Indo-Pacific; NA, North America (Nearctic); SA, South America (Neotropic).

Diversification Rate Analyses
The summary across the entire posterior produced in analysis A and B is given in Fig. S4, together with the credible set of distinct shift configurations that accounts for >87% of the probability of the data. The 95% credibility set contained six (A) and five (B) distinct shift configurations, but all with a posterior probability of a magnitude smaller than the thirdprobable shift configuration.

Figure S4 -Diversification rate shifts obtained by analysis A and B
The mean 'phylorate' plots are given in the center detailing speciation rates. The first three shift configurations from the Bayesian credible sets of distinct shift configurations accounting for ≥87% of the posterior probability are given to the left (A) and to the right (B), respectively. In the figure top left/right the maximum a posterior configuration is given. Circle sizes indicate marginal probabilities of the respective rate shift (f = frequency = posterior probability).

Figure S5. Niche shifts obtained by analysis A and B
Circles on the trees locate shifts, and circle colors corresponding to the scales indicate the magnitude of shifts. Circle size indicates posterior probabilities (pp) of shifts. Arbitrarily chosen colors of branches highlight clades, which were estimated to possess distinctive adaptive optima (cutoff ≥0.2 pp).

Table S3 -Results of diversification rate and bioclimatic niche analyses under different age estimations (analysis A and B).
For the diversification rate analysis, detected shifts are marked by their probability. The mean speciation rate (species/Ma) per clade is detailed with the 5%, and 95% highest posterior density in brackets below. For the bioclimatic niche analysis, shifts are marked by their probability, and the new phenotypic optimum (PC1 score optimum) is detailed with the standard error in brackets below.