Darwin review: angiosperm phylogeny and evolutionary radiations

Abstract

Darwin's dual interests in evolution and plants formed the basis of evolutionary botany, a field that developed following his publications on both topics. Here, we review his many contributions to plant biology—from the evolutionary origins of angiosperms to plant reproduction, carnivory, and movement—and note that he expected one day there would be a ‘true’ genealogical tree for plants. This view fuelled the field of plant phylogenetics. With perhaps nearly 400 000 species, the angiosperms have diversified rapidly since their origin in the Early Cretaceous, often through what appear to be rapid radiations. We describe these evolutionary patterns, evaluate possible drivers of radiations, consider how new approaches to studies of diversification can contribute to our understanding of angiosperm diversity, and suggest new directions for further insight into plant evolution.

1. Darwin: phylogeny and plants

Although best known for articulating the concept of evolution by natural selection, Darwin explored a host of other topics, laying the foundations for subsequent scientific inquiry in fields as diverse as plant structure/function, carnivorous plants, and phylogenetics. In fact, scientists in this last field openly acknowledge Darwin's major contributions, from the sketch of evolutionary relationships in his notebook in 1837 to the single figure (of a bifurcating tree) in On the origin of species clearly depicting extant species and the historical connections that preceded them to his beautiful description of the great Tree of Life [1, pp. 129–130]:

‘The affinities of all the beings of the same class have sometimes been represented by a great tree. I believe this simile largely speaks the truth. The green and budding twigs may represent existing species; and those produced during former years may represent the long succession of extinct species. … and this connexion of the former and present buds by ramifying branches may well represent the classification of all extinct and living species in groups subordinate to groups. … As buds give rise by growth to fresh buds, and these, if vigorous, branch out and overtop on all sides many a feebler branch, so by generation I believe it has been with the great Tree of Life, which fills with its dead and broken branches the crust of the earth, and covers the surface with its ever-branching and beautiful ramifications.’

This clear representation of phylogeny set the stage for rapid advances in understanding and visualizing the Tree of Life, for example, in the writings and illustrations of Haeckel and those who followed.

Far less well known than Darwin's explanation of evolution by natural selection or his concept of the Tree of Life is his immense contribution to both basic and evolutionary botany; indeed, Darwin was the original ‘evolutionary botanist’ ([2]; http://www.bgci.org/education/article/0659/), having applied evolutionary principles to botanical questions even as he wrote On the origin of species [1]. He published six books and over 75 papers on plants, with topics ranging from the fossil record of angiosperms to reproductive biology (especially of orchids [3]), insectivorous plants, plant domestication, and even plant movement (see [4] for further discussion of Darwin and orchid pollination and [5] on carnivorous plants). His botanical contributions, as described briefly below, both strengthened his evolutionary arguments by providing specific examples of principles presented in On the origin of species and laid the foundation for evolutionary botany.

Best known among his botanical writings is the ‘abominable mystery’ in reference to ‘…the rapid development as far as we can judge of all the higher plants within recent geological times…’ as described in a letter to his close friend J. D. Hooker, Director of the Royal Botanic Gardens, Kew, in July, 1879 (from [6, p. 5]). Darwin was specifically referring to the fossil record available at the time, which did little to identify the closest relatives of the earliest angiosperms, the early branching patterns of angiosperm phylogeny, and the relative timing of angiosperm features. However, the ‘abominable mystery’ term has been applied to a multitude of ongoing gaps in our understanding of angiosperm origins [6], from the sister group of the angiosperms, to the morphological and structural attributes of the first flower, to the genetic underpinning of floral morphology, and more. Darwin remained perplexed for decades about the apparent radiation that gave rise suddenly and mysteriously to exquisite angiosperm diversity in the mid-Cretaceous, in apparent violation of his view of gradual evolutionary change. In fact, Darwin hypothesized that angiosperms may have endured a long, gradual existence in a remote and undiscovered region, perhaps an island in the Southern Hemisphere, to account for their greater diversity than he thought their age warranted. The possibility of an earlier origin of ‘ecologically restricted angiosperms’, perhaps in the Jurassic, may be consistent with both molecular and fossil evidence (cf. [7]). Friedman [6, p. 5] interprets, however, that the most abominable issue for Darwin was not that the evolutionary origins of angiosperms were unclear but that ‘…evolution could be both rapid and potentially even saltational.’

Three of Darwin's books focused on plant reproductive biology, and his formal observations on the topic trace back to his early days on the H. M. S. Beagle. In fact, decades later, Darwin hypothesized that one possible explanation for an evolutionary radiation associated with the origin of angiosperms (if such a radiation must be accepted) was an elevated evolutionary rate driven by coevolution with insect pollinators (see [6]). Darwin's contributions to plant reproductive biology—Fertilisation in orchids (1862) [3], Effects of cross- and self-fertilisation in the vegetable kingdom (1876) [8], and Different forms of flowers on plants of the same species (1877) [9]—illustrated specific principles from On the origin of species, captured decades of observations, and produced new hypotheses to explain topics ranging from pollinator attraction to geitonogamy to heterostyly, all topics that remain central to understanding plant biology and evolution (e.g. [10–13]).

Darwin's fascination with plants did not end with their flowers. His Insectivorous plants (1875) [14] gives detailed observations of carnivorous plants in nature and describes his experiments on the mechanisms of insect capture by plants. Woven through the book are hypotheses to explain these amazing plants—such as the origins of carnivory as an adaptation to poor soil—and illustrations by himself and his sons George and Francis. Intrigued by the trapping movements of these plants and by plants' ability to climb [15], Darwin focused further on movement as an adaptation, culminating in The power of movement in plants (1877), and of course, Darwin's experiments on plants contributed immensely to The variation of animals and plants under domestication, published in 1868 [16] and further illustrating many of the evolutionary principles in On the origin of species.

Despite clearly articulating his views on the Great Tree of Life in On the origin of species and producing extensive writings on diversity in plant morphology, structure, and reproduction, Darwin never really put the themes of phylogeny and plants together himself, other than to share his opinion of an early angiosperm radiation as abominable. However, he did believe that ‘the time will come… when we shall have very fairly true genealogical trees of each great kingdom of nature’ (letter to T. H. Huxley 1857; Darwin Correspondence Project, www.darwinproject.ac.uk/letter/DCP-LETT-2143.xml). But others—the new ‘evolutionary botanists’ who followed in his footsteps—soon began viewing plant diversity through the lens of evolution and began to shape a new view of angiosperm phylogeny.

2. Patterns of angiosperm phylogeny

(a). Historical overview of the study of angiosperm phylogeny

Haeckel's beautifully rendered illustrations of organismal diversity and his hypotheses of phylogeny as early as the 1860s (e.g. [17]) set the stage for further development of depictions of phylogenetic relationships (see [18]). The general method of using hand-drawn illustrations to depict intuitively determined evolutionary relationships influenced researchers studying all branches of the tree of life. In the early 1900s, a well-known depiction of angiosperm relationships was provided by the American botanist Charles Bessey, whose summary scheme is still referred to as Bessey's [19] ‘cactus’ because of its resemblance to the branching of a cactus. In the middle to latter decades of the twentieth century, the tradition of depicting putative angiosperm relationships via such diagrams continued.

Several prominent angiosperm experts provided widely used classifications along with their intuitive portrayals of evolutionary relationships: Cronquist [20,21], Takhtajan [22,23], Thorne ([24,25]; with revisions to 2008), and Dahlgren [26,27]. Despite the development of phylogenetic theory and explicit methodology in the mid-1900s by Willi Hennig [28,29] and Warren ‘Herb’ Wagner (in [30]), and the origins of many components of Hennig's ideas in the work of the botanist Walter Zimmermann [31,32], botanists were slow to accept this revolution, and their work remained largely intuitive. The broad assessments of angiosperm relationships proposed by Cronquist, Takhtajan, Thorne, and Dahlgren emphasized different sets of morphological features, but there were notable similarities among all of these systems and their implications for angiosperm evolution. All believed that flowering plants were monophyletic and derived from a gymnosperm ancestor. Another common theme in the works of these investigators was that a woody magnoliid group (often termed Magnoliidae, or Magnolianae) represented the ‘most primitive’ (their words) living angiosperms. Furthermore, they all recognized two major subdivisions of angiosperms, monocots and dicots, with monocots derived from within dicots, making dicots paraphyletic, although neither the concept of paraphyly nor its application to dicots was considered by these authors. Also, an Alismatales group was suggested to be an early monocot lineage by all of these botanists. Although the evolutionary relationships suggested by these authors exhibited some broad similarities, there were also striking differences in the major subgroups recognized, as well as the evolutionary relationships among groups recognized as orders and families—the treatments also differed, sometimes dramatically, in the circumscriptions of these groups.

Of these four botanists, Dahlgren was more ‘phylogenetic’ in his thinking and his impact (see e.g. [33]; Dahlgren & Bremer [34] was perhaps the first application of cladistic methods to angiosperm-wide phylogeny). Dahlgren's use of chemical characteristics as possible features uniting groups of flowering plants was novel. He was the first, for example, to point to a close relationship among the families that produce mustard oil glucosides (Brassicales sensu Angiosperm Phylogeny Group (APG) classifications (see below); [35–38]); other authors of his time considered many of these mustard-oil-producing families to be distantly related. His branching diagrams of relationships (called ‘Dahlgrenograms’) were also highly influential.

Cronquist and Takhtajan had the longest duration of influence and also the broadest worldwide impact. Their views on relationships and angiosperm evolution were very similar—they actually worked closely together for many years—and dominated the field from the 1960s through to the 1990s until molecular phylogenetic treatments began to become more prevalent and ultimately dominate evolutionary thinking (e.g. [39,40]). Their systems [21,23] were very similar in content (although differing in names and ranks) and ultimately became the most widely used of all the systems available at that time. They divided angiosperms into two classes (Magnoliopsida and Liliopsida) and these in turn into subclasses, orders, and families. Cronquist further developed a branching scheme of orders. Much like Bessey's [19] dicta, a set of hypothesized evolutionary changes based on Bessey's inference of relationships, Cronquist also provided a number of guiding phylogenetic principles that long dominated evolutionary thinking. In modern phylogenetic terms, many of these statements were essentially hypotheses of character polarity. For example, Cronquist hypothesized that the earliest flowering plants were probably large shrubs or small trees. Using the terminology of the time, he provided a list of many additional so-called ‘primitive’ features (today interpreted as ‘ancestral’): simple leaves are ancestral to compound leaves; reticulate venation is ancestral to parallel venation; laminar (leaf-like) stamens are ancestral to stamens with well-defined anther and filament; spirally arranged floral parts are ancestral to whorled floral arrangement; pollen with one aperture is ancestral to pollen with three apertures. These hypotheses can now be explicitly tested with current estimates of phylogeny and large-scale morphological datasets (e.g. [41]).

The application of phylogenetic theory and explicit methods to molecular data, especially DNA sequences, in the late 1900s dramatically altered the process of phylogenetic inference. Collaborative efforts by plant systematists generated datasets of hundreds of species (e.g. [39]), pushing the limits of algorithms and computation available at the time. These forays into botanical phylogenetics eventually led to the development of improved methods and their application to thousands of plant species, resulting in phylogenetic trees—for angiosperms (e.g. [42]), land plants (e.g. [43]), and most recently, all seed plants [44]—that exceeded in size those for any other group of organisms. Both the continued development of methods and the availability of such large phylogenetic hypotheses are now permitting explicit tests of diversification and of the possible drivers of radiations (see below).

The past two and a half decades have yielded a firm understanding of the major lineages of angiosperms and their relationships, yet much work is needed for more complete resolution of a clade of this size (perhaps 400 000 species per Govaerts [45], but other estimates have yielded different numbers, e.g. Christenhusz & Byng [46] estimated 295 383 based on counts of accepted species names). Although aspects of our current understanding of angiosperm phylogeny align well with the views of Cronquist, Takhtajan, Thorne, and Dahlgren, many surprises have emerged. Moreover, as we continue to gather data from both the plastid and nuclear genomes, we find cases of incongruence not only at shallow phylogenetic levels, resulting from introgression or incomplete lineage sorting (or both!), but also much deeper in the tree, suggesting ancient episodes of hybridization and introgression (see discussion of rosids, below). However, because the backbone of angiosperm phylogeny is largely consistent among analyses based on different genes and different methods, a new synthetic and collaborative classification now summarizes these major groups and forms a framework for more in-depth studies of specific clades. The APG, comprising dozens of plant systematists over the years, is an international consortium dedicated to community-based classification, a dramatic departure from the single-authored classifications of decades past. The APG focuses on clades classified at the ordinal and familial levels, but the classifications also include a number of rank-free supraordinal groups. The APG has published four versions of its classification [35–38], each generally representing only incremental improvements over previous versions. The Angiosperm Phylogeny Website (www.mobot.org/MOBOT/research/APWeb/), developed and maintained by Stevens [47], provides details and additional updates on angiosperm phylogeny. Additional information is compiled in Soltis et al. [48]. Here, we refer readers to figure 1, which summarizes the major clades of angiosperms recognized as orders and families by APG IV [38] and as supraordinal clades by Cantino et al. [49]. In the section that follows, we do not summarize all aspects of angiosperm phylogeny—a task that exceeds the scope of this review—but instead focus on those regions of the angiosperm tree that appear to result from radiations.

Figure 1.

Ordinal-level phylogeny of angiosperms based on APG IV [38]. Branch colours show species richness, based on the Angiosperm Phylogeny Website ([47] onwards). Yellow starbursts refer to major radiations discussed in the text. Around the border are typical representative species. (Online version in colour.)

(b). Radiations in angiosperm phylogeny

A recurrent pattern in angiosperm phylogeny is radiation, as indicated by clustering of short branches in a phylogenetic tree and, in many cases, the result of an increased rate of diversification. In fact, angiosperm evolution is punctuated repeatedly by radiations [50,51], ranging in time from near the origin of the angiosperms to recent events associated with colonization of new habitats, such as volcanic islands and recently glaciated areas in both arctic and alpine areas. Darwin's ‘abominable mystery’, in reference to the recent rise and rapid diversification of angiosperms based on inferences from the fossil record of his day, led to long-held views that the origin of the angiosperms themselves was followed by a rapid radiation. But what are radiations, and how are they detected? We tend to think of radiations as adaptive, the classic case being one in which the ancestor of a group found itself in a new, open environment (such as an island archipelago), and subsequent rapid evolution produced diverse descendants, each adapted to a variant of this new environment. Unfortunately, such obvious cases seem to be quite rare; instead, we tend to identify radiations as those regions in phylogenetic trees with numerous short branches that are difficult to resolve, but the adaptive perspective is typically not clear. Fortunately, recent innovations in phylogenetic and comparative analyses have generated methods by which statistically significant upshifts in rates of diversification—which we argue can be viewed as radiations, whether or not they are adaptive—can be identified and specific organismal features can be evaluated as correlates, and possible drivers, of diversification.

(i). Evidence for radiations

There have been many attempts to estimate diversification rates directly from fossils [52–55]; these efforts have primarily focused on taxa with dense fossil records, many of which are marine molluscs and microscopic algae. The lack of an adequately sampled historical record for the vast majority of life has inspired methods that do not require an extensive (or any) fossil record, most commonly using phylogenetic data [56] but occasionally focusing on simple species count data (e.g. [57]). Phylogenetic approaches rely on the fact that different patterns of speciation and extinction leave historical information in phylogenetic data, in the form of topology and branch lengths, that can be used to distinguish among them [58–60].

The methods available for detecting radiations have exploded in the past 10 years. Numerous statistical methods are now available to detect radiations associated with traits (‘key innovations’; [61–63]) or historical climate [64,65] and to test whether radiations have experienced density dependence [66,67]. It is also possible to reconstruct diversification patterns in the absence of a priori hypotheses (e.g. [67,68]) or even to reconstruct patterns of phenotypic or ecological radiation [69,70]. Such statistical approaches to diversification have seen intense recent discussion as these methods continue to be refined (e.g. [71–76]). Despite uncertainty about the best approaches, diversification models continue to see intensive use (in Google Scholar, as of 25 December 2018, there are 3090 papers with the word ‘diversification’ and one of these key words referring to popular methods: ‘BAMM’, ‘MEDUSA’, ‘RPANDA’). An optimistic future for detecting radiations is suggested by both this interest and the continual appearance of new approaches to the study of diversification.

(ii). Examples of radiations

Despite the availability of the methods described above, they are only just now being applied to specific cases of angiosperm phylogeny (see below for examples). However, inferences of radiations based on topologies and branch lengths are frequent features of angiosperm phylogeny.

Early angiosperms. Darwin's view of the ‘rapid rise and early diversification of angiosperms’ has long engendered the view of a radiation associated with the origin of angiosperms, or in modern terms, an ‘early burst’. The fossil record, with its limited representatives in the Early Cretaceous and comparative bounty by the Late Cretaceous, certainly supports the concept of an early angiosperm radiation. However, phylogenetic trees for extant species published over the past 20 years have demonstrated that the root of the angiosperm tree and succeeding nodes are characterized by a grade of depauperate branches rather than a burst of numerous species-rich evolutionary lineages (that is, early angiosperm branching is highly asymmetrical sensu [77]; e.g. [78–80]; see figure 1). Amborella trichopoda, the sole extant sister to all other extant angiosperms, Nymphaeales (water lilies), and Austrobaileyales (a small clade of Austrobaileya, Schisandraceae, and Trimeniaceae) together comprise only 175 species and represent the earliest divergences in angiosperm phylogeny.

Following this grade, however, is the radiation of Mesangiospermae (per [49]), a clade of five main branches comprising all remaining angiosperms (all clades above the ordinal level from [49]): Magnoliidae (magnoliids), Monocotyledoneae (monocots), Eudicotyledoneae (eudicots), Chloranthaceae, and Ceratophyllum. Magnoliids, monocots, and eudicots represent 3%, 22%, and 75% of all angiosperm species [81], and Chloranthaceae and Ceratophyllum have only 75 and eight species, respectively. Relationships among these clades have been surprisingly difficult to resolve (e.g. [82]), despite sampling of thousands of species in some analyses and hundreds of genes in others. Note that the traditional monocot–dicot dichotomy is not supported by molecular phylogenetic analyses (or even by most analyses of morphological data); the paraphyletic ‘dicots’ comprise magnoliids, eudicots, Chloranthaceae, and Ceratophyllum, plus Amborella, Nymphaeales, and Austrobaileyales. Instead, topologies with just about all possible sister-group relationships have been reported. Among recent studies, an analysis based on nearly complete plastid genomes and approximately 2000 species found that monocots and (eudicots + Ceratophyllum) were sister groups, and magnoliids + Chloranthaceae their sister, but support for these relationships was weak [83]. Nuclear genes seem to tell a different story, supporting magnoliids and monocots as sister groups with moderate support; however, the key studies reporting this topology included only 59 genes and 61 species of angiosperms [84] and 852 genes but only 103 species of green plants and 37 angiosperms [85]. The recent publication of the genome sequences of two magnoliids, Liriodendron chinense (tulip tree, yellow poplar; Magnoliaceae; [86]) and Cinnamomum kanehirae (stout camphor tree; Lauraceae; [87]), provides hope that additional sequence data may help resolve this phylogenetic conundrum, but until adequate taxon sampling accompanies massive gene sampling, this problem will probably remain [88]. Regardless of relationships among clades of mesangiosperms, it is clear that the clade itself represents an ancient radiation that diversified shortly after the origin of the angiosperms themselves, with these five branches diverging within a span of fewer than 5 million years ([82]; 143.8–140.3 Ma). This early pattern of radiation sets the stage for many nested radiation events within the component clades of mesangiosperms; in the paragraphs that follow, we describe patterns of radiation in monocots and in the rosid and asterid clades of eudicots as examples of patterns seen more generally throughout angiosperms [89].

Monocots. The monocots, formally Monocotyledoneae [49] and long recognized as a major group of angiosperms at some taxonomic level (often class Liliopsida, e.g. [21]), comprise approximately 85 000 species [90] classified in Linnaean taxonomy in 77 families in 11 or 12 orders [38], with each family and order representing a clade [38]. As one of the five extant clades of the early mesangiosperm radiation, the monocots themselves show repeated patterns of radiations throughout the clade (figure 1). In fact, methodological difficulties caused by apparent rapid radiations have vexed monocot phylogeneticists as they try to tease apart relationships and infer patterns of character evolution (e.g. [91,92]). The most comprehensive analysis of monocot phylogeny, diversification, and character evolution to date [93], based on nearly complete plastome sequences and 545 monocot species representing all 77 families, clearly identifies major radiations (as significantly increased rates of diversification) in the monocots and, in some cases, ties putatively adaptive character-state changes to these events. Based on this analysis, monocots diverged from the common ancestor of the eudicots + Ceratophyllum 136.1 Ma, and the monocot crown group dates to 132.4 Ma. All 12 clades recognized as orders of monocots were present by 114 Ma, and 45 of 77 families were present by the end of the Cretaceous (the Cretaceous-Palaeogene (K-Pg) boundary, 65 Ma).

Four large-scale accelerations in the rate of species diversification were detected in the common ancestors of the following clades of monocots: (i) PACMAD-BOP clade of Poaceae, (ii) the clade of Asparagales that is sister to Doryanthaceae, (iii) the Orchidoideae-Epidendroideae clade of Orchidaceae, and (iv) the clade of Araceae that is sister to Lemnoideae. Interestingly, all of these upshifts occurred shortly before or shortly after the K-Pg boundary, 65 Ma. Although the sizes of monocot families have long been recognized as unequal, ranging from two species in families such as Acoraceae to nearly 28 000 species in Orchidaceae (The plant list; http://www.theplantlist.org/), Givnish et al. [93] reveal that the more species-rich families, such as orchids and grasses, did not necessarily undergo accelerated diversification in their common ancestors, but instead, increased species richness and radiations occurred in specific clades within these families, such as the PACMAD-BOP clade of Poaceae (which includes most, but not all, species of grasses) and the Orchidoideae-Epidendroideae clade of Orchidaceae (which includes well-recognized and horticulturally significant genera such as Cattleya and Dendrobium). Moreover, these radiations were associated with morphological or functional transitions [93]: the origins of C4 photosynthesis and its attendant anatomical, morphological, and physiological shifts in the PACMAD-BOP clade; bulbous, cormous, or xeromorphic adaptations in the Asparagales clade that is sister to Doryanthaceae; the evolution of pollen packaged in pollinia in the common ancestor of the Orchidoideae-Epidendroideae clade, plus the emergence of epiphytism in part of this clade [94]; and, in Araceae, epiphytism and the origins of fleshy fruit, both of which may have limited dispersal and promoted diversification in tropical forests. Although some of these findings are themselves new hypotheses for further testing, it is clear that the application of new analytical methods to vastly improved phylogenetic hypotheses is pinpointing more precisely the location and extent of species radiations in the angiosperms and permitting explicit tests of associated morphological, ecological, or other features that may have driven the radiations.

Eudicots–Asterids. Deep relationships in the asterids, which comprise approximately 25% of all angiosperm species, have been relatively well resolved for nearly two decades (e.g. [95]), and patterns of phenotypic variation have been inferred across this clade (e.g. [96,97]). Asterids comprise four major clades (figure 1): Cornales are sister to all other asterids, and Ericales are sister to the ‘core asterids’, which in turn comprise campanulids and lamiids. Branches are long in the early asterids; at least in terms of extant clades, radiations at deep levels seem to be less common in asterids than in rosids, with a few major exceptions (see below). Instead, most apparent radiations tend to be relatively recent, for example, within clades recognized as families, such as Asteraceae, Lamiaceae, and Apiaceae. However, some key ancient radiations seem to have spawned subsequent radiations, most notably in Ericales and Lamiales, two of the most phylogenetically notorious clades of asterids, defying efforts at phylogenetic resolution.

Ericales represent more than 4% of all angiosperm species diversity (more than 11 000 species in 22 families) and include economically important plants such as tea (Theaceae), blueberries and cranberries (Ericaceae), persimmon and ebony (Ebenaceae), kiwi (Actinidiaceae), Brazil nuts (Lecythidaceae), sapote (Sapotaceae and Ebenaceae), fine woods including ebony (Ebenaceae), and many ornamental species including primroses (Primulaceae), and rhododendrons and azaleas (Ericaceae). Ericales are often ecologically dominant in diverse habitats ranging from boreal and alpine peatlands to tropical rainforests, and certain clades (e.g. Polemoniaceae, Primulaceae, and Ericaceae) have apparently moved in and out of tropical regions through montane corridors that share cool temperatures and perhaps other climatic similarities with temperate regions. However, although certain aspects of such movements are apparent in published phylogenies, more apparent is the lack of resolution among clades of Ericales recognized as families (e.g. [98,99]). Ericales phylogeny appears to result from an ancient radiation that generated most major lineages within this clade, at approximately 104–106 Ma [99].

As with Ericales, the phylogeny of Lamiales (approx. 20 000 species in approx. 25 families) has been very difficult to reconstruct, with most of the order tracing to an ancient radiation that is itself nested within radiations of lamiids [100,101], a clade that includes 15% of all angiosperm species. The hyper-diverse Lamiales exhibit global patterns of radiation that can potentially be linked to shifts in distribution, ecology, and phenotypic evolution. Unfortunately, such comparative studies remain intractable without greater confidence in phylogenetic relationships. Plastid [102] and nuclear [103] trees conflict, and additional work is needed to resolve this incongruence and set the stage for further global analysis of Lamiaceae (mints), which themselves comprise nearly 7900 species (The plant list, www.theplantlist.org) and remain poorly understood phylogenetically.

Asteraceae, with approximately 28 000 species, rival Orchidaceae for the largest family of angiosperms, and like Orchidaceae, Asteraceae have also undergone accelerated rates of diversification. Although Asteraceae have 460 times the number of species of their sister group, Calyceraceae, the species diversity is not evenly distributed throughout the family. Instead, radiations are nested within the clade, such that rates of diversification during the early evolution of Asteraceae are similar to those in Calyceraceae and Goodeniaceae [104]. Neither the evolutionary innovations of racemose capitulum or pappus, nor the whole-genome duplication (WGD) event that occurred early in the evolution of the family is significantly associated with increased rates of diversification. Asteraceae originated near the end of the Cretaceous, and from 69.5 Ma, before the K-Pg boundary, until the Eocene Climatic Optimum (approx. 50 Ma), diversification rates increased, but these increased rates are not clearly associated with any specific drivers of diversification. Much of the species diversity of Asteraceae actually originated relatively recently, with radiations in the Americas (vernonioid clade in the Early Eocene and a subclade of the Heliantheae alliance with phytomelanic fruit in the Late Oligocene) and Africa, where a radiation in the last 6.5 Myr expanded species diversity in six clades recognized as subfamilies. Despite progress in understanding radiations in Asteraceae, and hypotheses about possible causes of accelerated rates of diversification (e.g. [104,105]), the clade is under-represented in phylogenetic analyses in terms of species; recent analyses using up to approximately 800 nuclear loci [106] and increasing numbers of species will help to clarify patterns of radiation and stasis and will eventually allow for rigorous analyses of drivers of diversification.

Estimation of divergence times within asterids has long been difficult, owing in part to a poorer fossil record than is available for rosids and other major clades. Although many asterid lineages are woody (dogwoods, several clades of Ericales, some clades of lamiids), many of the most species-rich clades are herbaceous (e.g. Lamiaceae (mints), 7900 species; Asteraceae (composites), 28 000 species; The plant list, www.theplantlist.org) and therefore tend to lack a good fossil record. However, the recent discovery and application of more reliable fossils as calibrations in phylogenetic dating (see [107]) are improving the chronological picture of asterid evolution and our understanding of the patterns and processes of diversification.

Eudicots–Rosids. The rosid clade comprises nearly 90 000 species, approximately 25% of all angiosperm species. This megadiverse clade includes many dominant forest trees (e.g. Betulaceae (alder, birch), Casuarinaceae (Australian pine), Fabaceae (legumes), Fagaceae (oaks), Juglandaceae (walnuts, hickory), Moraceae (figs), Salicaceae (willows), Ulmaceae (elms), Rutaceae (citrus), Meliaceae (mahogany), Sapindaceae (maples, buckeye), Malvaceae (linden), Dipterocarpaceae (dipterocarps), and Myrtaceae (eucalypts); [108,109]). Rosid herbs and shrubs are also prominent components of arctic/alpine and temperate floras (e.g. Salicaceae, Rosaceae, Brassicaceae) and comprise aquatics (riverweed, water starwort), desert plants (euphorbs), and parasites (e.g. Rafflesia).

The rosid clade perhaps best exemplifies a pattern of nested radiations (figure 2). The clade originated in the Early to Late Cretaceous (115–93 Ma), followed by rapid diversification of two major subclades, the Fabidae (fabids) and Malvidae (malvids) crown groups, with all major lineages arising about 112-91 Ma and 109-83 Ma, respectively [108,111]. This early diversification of rosids occurred over possibly as few as 4–5 Myr and no more than 15 Myr [108]. Within both Fabidae and Malvidae is evidence for additional radiations. An early radiation in malvids generated Geraniales, Myrtales, Crossosomatales, Picramniales, Huerteales, Brassicales, Malvales, and Sapindales, with most of the species diversity of malvids in the latter three clades. Likewise, an early radiation in fabids produced Zygophyllales, Celastrales, Oxalidales, and Malpighiales (the latter three together forming the ‘COM’ clade), and the ‘nitrogen-fixing clade’, which includes all angiosperms with symbioses with nodule-forming nitrogen-fixing bacteria (Fabales, Rosales, Fagales, and Cucurbitales). Although strongly supported by plastid data as part of the fabid clade, the COM clade appears within the malvids in analyses based on mitochondrial and nuclear data [112,113], perhaps signifying an ancient episode of hybridization followed by chloroplast capture. Still further radiations are evident within rosid subclades as well, most notably in Malpighiales, a clade of 36 families that all diverged 90–102 Ma [108]. In all, rosids comprise clades recognized as 17 orders and 140 families [38].

Figure 2.

Family-level phylogeny of rosids based on Sun et al. [110]. Branch colours show species richness, based on the Angiosperm Phylogeny Website ([47] onwards). Around the border are typical representative species. (Online version in colour.)

The rosid radiations represent the rapid rise of angiosperm-dominated forests and associated co-diversification events that profoundly shaped much of current terrestrial biodiversity [108]. The timing of the earliest rosid radiations roughly corresponds to fossil leaf assemblages (104–97 Ma) remarkable for their diversity, including many new rosid groups [114]. The early rosid radiation also corresponds to radiations in major groups of herbivorous insects, such as ants [115], beetles and hemipterans [116,117], which have been suggested to have diversified in response to the ‘rise in angiosperm-dominated forests’ [115]. The early splits in placental mammals also seem to correspond in time to the early rosid radiations [118], and subsequent radiations in amphibians [119] and ferns [120] have been explained by responses to the angiosperm forests that formed through the diversification of rosids.

3. Drivers of angiosperm radiations

The huge diversity of angiosperms has long been attributed to characteristics unique to this clade, many of which are associated with reproduction and may have served as key innovations for the entire clade. These traditional explanations for both species richness and abundance of angiosperms relate to the flower—with its closed carpel—fruit (the mature ovary wall), and process of double fertilization, with presumable built-in advantages in reproductive assurance, rapidity of reproduction, opportunity for gametic competition, availability of endosperm as a nutritive source, and the role of fruit as a dispersal unit. These attributes of angiosperms were considered to provide reproductive superiority over non-angiospermous seed plants, leading to a burst of evolution following the origin of the flower. Subsequent diversification has long been considered the result of interactions with animals, especially insect pollinators.

(a). Co-diversification

Close associations with pollinators are thought to have played an early role in angiosperm diversification [121–123], possibly co-opted from ancient associations between gymnosperms and insect pollinators (e.g. [124,125]). In fact, many groups of insects diversified shortly after some of the early angiosperm radiations (ants, beetles, hemipterans), but these insect radiations did not appear to be related to pollination and modifications of flowers. Radiations of bees, however, do seem to show ancient patterns of co-diversification with angiosperms and floral diversity [126]. More recent diversifications owing to associations between plants and pollinators are clear in certain groups, for example, Iochroma of Solanaceae [127] and Polemoniaceae (the phlox family) [128]. Such cases of radiation on these smaller scales certainly are responsible for much of angiosperm species diversity; however, they fail to capture those features that may have served as key innovations for ancient, large-scale radiations.

(b). Phenotypic associations

The only angiosperm-wide analysis of potential drivers of diversification is now 15 years old, and although new methods, new tree topologies, and new phenotypic data are available, the study by Davies et al. [129] remains the most comprehensive such analysis. Molecular dating of a supertree of the angiosperms showed a radiation among ‘basal angiosperms’ [129], consistent with the pattern noted by Darwin as the ‘abominable mystery,’ as well as numerous other nodes that deviated from equivalent evolutionary rates in daughter clades. At several of these nodes, which range in age from 133 Ma (Acoraceae versus all remaining monocots) to 41 Ma (Ecdeiocoleaceae versus Poaceae), one daughter clade is significantly larger than its sister group, indicative of substantial variation in diversification rate and possible radiations. A set of biotic attributes—pollination mode, dispersal mode, habit, dioecy, chromosome number, geographical distribution, and lifestyle—was compared among the 10 most imbalanced nodes in the tree to determine if certain traits were consistently associated with radiations. Although all of these traits have been suggested as key innovations for one or more clades, no pattern emerged among these 10 nodes. The only trait consistently found for the large clade but not the small was a cosmopolitan distribution, but this trait is probably a consequence of the radiation rather than a driver of it. In nearly all other trait comparisons, a given trait characterized the large clade in some sister groups and the small one in others; for example, although many traits were polymorphic in the large clades, both alternatives characterized certain small clades: biotic and abiotic dispersal, woody and herbaceous habit, no dioecy and polymorphism for dioecy, distributions on nearly any continent, and annual and perennial lifestyle. Thus, the drivers of diversification in angiosperms are not consistent across radiations; instead of a single factor promoting diversification, it appears likely that combinations of traits may have been more important than single traits (see also [130]). Moreover, these and other traits should not be considered in the absence of environment. Unlike Darwin, who sought a single driver for the origin of the angiosperms and found the results ‘wretchedly poor’ [129,131], Davies et al. [129] concluded that diversification in the angiosperms is clade-specific, most likely owing to a combination of factors, and a ‘complex process in which propensity to diversify is highly labile’. Given new methods now available (see above), the time seems right to re-investigate the hypotheses tested by Davies et al. [129] using a larger tree, more comprehensive coverage of traits, and methods that explicitly test for diversification-associated traits.

(c). Whole-genome duplications

Although no single phenotypic feature has emerged as a consistent driver of diversification, at least one genomic attribute seems to be associated with many radiations across the angiosperms: WGD. It has long been recognized that WGD, or polyploidy, has played a major role in plant evolution, particularly angiosperms [132–134]. Recent genomic studies have illustrated that the frequency of WGD in angiosperms is much greater than long thought. All living angiosperms are descendants of an ancient WGD event [135,136]. Furthermore, genomic studies have shown that ancient WGD events are widespread across the angiosperm tree of life [113]. Towards the tips of the tree, approximately 35% of all angiosperm species are the result of more recent WGD [137].

Despite the prevalence of polyploidy, the role of WGD in promoting diversification has remained unclear. A heuristic approach of placing WGD events on an angiosperm tree and assessing the extent of speciation after the event suggests that some ancient WGD events appear to be associated with major radiations [138]. The WGD event shared by all living angiosperms, for example, suggests that polyploidy may have been associated with the rise of the angiosperms. Other key WGD events are associated with major clades, including monocots, eudicots, Brassicales, and Poaceae. Other studies suggest that key innovations may accompany WGD ([139]; see also [140]), serving as possible drivers of diversification.

Several studies have attempted to rigorously assess the association between WGD and diversification using large phylogenies [141,142]. Landis et al. [142] examined 106 ancient WGDs that could be placed with some certainty on a large angiosperm tree [43] and found that 46 of these events were significantly associated with increased rates of diversification. Moreover, sister-group comparisons showed that 70 of 99 WGD events were associated with an increase in species richness in one of the daughter clades. Thus, evidence suggests that, although not all WGDs generate radiations (see also [137,143]), many polyploidy events are associated with increased rates of diversification.

(d). The ecological backdrop of diversification

Despite longstanding interest in ecological dimensions to the diversification process, the use of diversification methods in ecology remains rare [56,144]. Because adaptive radiation in particular has often been understood in terms of responses to ecological opportunity [145–147], ecology is clearly a critical potential driver of evolutionary radiations. Such a role for ‘ecology’ in diversification has often been thought of as a contrast between change in niche (divergence) or a failure to change (conservatism) as lineages diversify [148]—for instance, asking whether a radiation represents ecological speciation or is niche-neutral. The ultimate role of niche conservatism (the idea that closely related species also tend to be similar in niche) in diversification may vary and be scale-dependent [149,150]; nevertheless, profound evidence of niche conservatism exists in many groups of plants [148,151].

(i). Diversification of biomes

Several studies have sought to understand present-day ecological communities in terms of the evolutionary past of their members. Much attention in diversification studies has been devoted to iconic hyper-diverse areas of the globe, such as the Amazon [152] and, especially in plants, the Andes [153–155], as well as to global diversity gradients [156,157]. These studies typically ask whether increased diversification alone explains species richness patterns, or whether they are better explained by time elapsed since colonization [152], climatic niche [158], or other causes. Among the remarkable outcomes of this work has been that neither diversification rate ([152,157], but see [159]) nor age of a community ([160,161], but see [152,162]) can alone account for species richness in contemporary communities. Indeed, even for relatively ancient clades, much of species diversity is of recent origin (e.g. [163]; the same is also true for cycads, [164]), and diversification rates do not always correlate with species richness. Many of these studies of diversification focus on globally distributed groups, largely tetrapods, that have their greatest diversity in the tropics; fewer results are available for temperate regions and plants (but see e.g. [165,166]).

(ii). Spatial distribution of the diversification process

Macroevolutionists often frame the diversification process in terms of finding the role of ecology or other factors in shaping present-day diversity. The opposite framing—the role of diversification in ecology—has recently spawned a large number of methods under the class of spatial phylogenetics. Rather than explicit models of the diversification process like those summarized above, these approaches seek to use summary phylogenetic statistics of ecological communities (cumulative tree length, mean divergence time, etc.) to understand the spatial distribution of evolutionary history and shed light on community assembly via the diversification process. Compared to probabilistic diversification methods, it is challenging to convincingly integrate spatial phylogenetics with historical environmental data (e.g. historical mean annual temperature), distinguish among alternative diversification hypotheses, or understand the historical timing of evolutionary events. Nevertheless, one key advantage of these approaches is the ability to explore fine geographical scales. While explicit geographical diversification models have been proposed [167,168], these often rely on coarse categorical descriptors of regions or biomes (‘Africa’, ‘chaparral’, etc.). Additionally, it is straightforward to incorporate contemporary environmental data such as climate and soil predictors to test a relationship between diversification and alpine biomes [169]. Finally, it is possible to use diversification models in a community context by directly calculating rates of lineage diversification in present-day communities via ‘tip rates’ (rates of diversification estimated for extant clades); such methods have seen substantial recent interest for understanding biodiversity gradients [156,157]. Thus, new avenues are emerging for examining possible associations between the environment and species diversification.

(iii). Niche or rates of niche?

When considering the role of ecology in diversification, an important distinction lies in whether habitat is itself associated with species diversification, or rather the association is with rates of niche evolution (‘niche lability’) (figure 3). The latter association has been found in several cases [170,171], sometimes in the absence of a relationship with niche itself [172]. Such an association may capture ecological divergence itself (rather than particular ecologies) as a driver of diversification [170,171]. However, this coupling may not necessarily be simultaneous. In literature on trait-associated diversification, it has often been observed that there can be lag times among trait innovation and diversification (reviewed in [130,173]; for genomic data (WGDs), see [138,141,142,174]). We have found similar evidence of a lag between diversification and subsequent evolutionary rates of ecological niche occupancy in Saxifragales [172]. In fact, a similar ‘lag’ pattern is observed across the rosids as a whole (H. Sun, R.A. Folk, M.A. Gitzendanner, S.A. Smith, C. Germain-Aubrey, R.P. Guralnick, P.S. Soltis, Z. Chen, D.E. Soltis 2019, unpublished) and in Ericales (C. Fu, L. Gao, D.E. Soltis, P.S. Soltis 2019, unpublished); in Saxifragales, rosids, and Ericales, increased rates of species diversification preceded shifts in ecological niche, suggesting perhaps a common pattern for species groups that are most species-rich in temperate and cold climates (see [172]) (figure 3). The presence of lags could suggest that either these are simply uncoupled [69,172,175] or that innovations are necessary but not sufficient for subsequent diversification [130,141,142].

Figure 3.

Hypothetical diversification-through-time plots demonstrating various types of radiation patterns observed in empirical datasets: (a,b) demonstrate differences in the tempo of radiations; (c,d) show the detection of potential drivers of diversification; and (e,f) show potential relationships between ecological divergence and diversification. (a) Early burst, (b) recent burst, (c) environment-associated diversification, (d) trait-associated diversification, (e) niche divergence with diversification, and (f) niche divergence follows diversification. (Online version in colour.)

(iv). Density dependence

It is commonly observed in empirical datasets that diversification slows down over time ([176,177], among many others), a phenomenon often attributed to a density-dependent diversification process (that is, that the diversification rate depends on the number of lineages). Density-dependent models can be used to capture ecological limits on diversification because ecological limits such as niche filling are commonly invoked to explain observed early bursts followed by diversification declines [178–181], although not all models for declining diversification have been ecological in nature [182]. There are numerous empirical exceptions to the diversification-slowdown pattern (e.g. [156,172]) as well as more complex patterns [118], suggesting both that the pattern may be dependent on phylogenetic scales [56] and that alternative processes may maintain diversification rates into the present. In particular, continuing ecological opportunity, as opposed to a discrete ecological shift, is one mechanism that may enable an escape from declining diversification [172].

4. Conclusion

Although more than 150 years have passed since Darwin wrote about phylogeny and plants, plant phylogeneticists are just beginning to reveal the patterns among those ‘beautiful ramifications’ and understand the processes that generated them. Even as we learn more about plant reproductive biology, the role of isolation in promoting diversification, the frequency and consequences of whole-genome duplication, and other population-level processes, how these microevolutionary processes produce macroevolutionary patterns still remains a mystery. Recent innovations in high-throughput data generation for phylogeny reconstruction, coupled with development of theory and methods for evaluating patterns of diversification (see discussion and references above), are beginning to reveal significant shifts in rates of species diversification across plant phylogeny. Moreover, in a few cases, such patterns can be associated statistically with potential drivers of diversification, from specific phenotypic traits to physiological shifts to genome-wide events. However, further progress requires greater species sampling in phylogenetic trees, improved accounting for missing data, better incorporation of fossils and fossil calibrations into trees of modern species, and increased trait data. The huge size of the angiosperm clade is itself a stumbling block for application of many current methods; thus, analyses are typically confined to much smaller clades or to exemplar-based trees with only a few representatives each from key clades. Although this approach is useful for revealing patterns and potential processes within these clades, it necessarily prevents a holistic view of angiosperm radiations—whether or not there were synchronous radiations in multiple groups, whether specific geographical regions have spawned multiple nested radiations, and other possible large-scale patterns and their potential drivers. As we move forward and attempt to fill current gaps in our sampling and knowledge, the confluence of phylogenomic methods and phenomics promises to shed new light on patterns of angiosperm radiations.

Data accessibility

This article has no additional data.

Competing interests

We declare we have no competing interests.

Funding

This work was supported in part by NSF grants DBI-1458640 to D.E.S. and P.S.S. and DBI-1523667 to R.A.F.