Abstract
Species diversity varies greatly across the different taxonomic groups that comprise the Tree of Life (ToL). This imbalance is particularly conspicuous within angiosperms, but is largely unexplained 1. Seed mass is one factor that may help some lineages diversify more than others by influencing key life history traits, such as dispersal, colonisation, environmental tolerance and reproductive success 2. However, the extent and direction of these effects have not been assessed across the angiosperm ToL. Here, we show for the first time that absolute seed size and the rate of change in seed size are both associated with variation in diversification rates. Based on an unequalled phylogenetic tree that included 4105 angiosperm genera, we found that smaller-seeded plants had higher rates of diversification, possibly due to improved colonisation potential 3. The rate of phenotypic change in seed size was also strongly positively correlated with speciation rates, supporting emerging evidence that rapid morphological change is associated with species divergence 4. Our study now reveals that variation in morphological traits, as well as the rate at which traits evolve, contribute significantly to the extremely uneven distribution of diversity across the ToL.
Angiosperms are one of the most species-rich clades on Earth and have dominated terrestrial plant communities since the Late Cretaceous 5. The astounding diversity of flowering plants is, however, extremely unevenly distributed across the ToL. Each of the five most species-rich angiosperm families contains >10,000 species while more than 200 families contain <100 species each. An enduring pursuit in evolutionary biology is to explain this uneven distribution of biodiversity, not only in angiosperms, but also across the whole ToL 1.
Seed mass influences individual life history in ways that can ultimately shape angiosperm diversification 6. Along with adult plant size, seed mass affects survival, reproductive lifespan and dispersal 7. These traits contribute to fitness and adaptation, which are the ultimate determinants of whether lineages diversify or go extinct 8. In support of this idea, seed mass has been shown to correlate negatively with diversification in the Polygoneaceae 9, but this has not been investigated across large taxonomic scales. As seed mass varies over ten orders of magnitude in angiosperms, from the minute 1 µg seeds of some orchids to the >18 kg seeds of the sea coconut Lodoicea maldivica, this huge variation may coincide with variation in species diversity. Generalising the direction and magnitude of an effect of seed mass on diversification across taxonomic scales has however proved difficult. Some life history traits encapsulated by seed mass are expected to promote speciation or extinction, while others may simultaneously counteract such effects.
The rate of change in key life history traits such as seed size can be as important in driving macroevolutionary dynamics as the absolute values of the traits themselves 4. This is because phenotypic divergence may cause reproductive isolation that results in speciation 10. Nevertheless, few empirical studies have detected a correlation between rates of phenotypic evolution and lineage diversification 11. This correlation may be expected where a trait can change more rapidly in some species than others in response to selective pressures (i.e. high “evolvability” 12); this enables greater access to new ecological niches or quicker establishment of reproductive isolation, and thereby increasing the rate of speciation (λ) 13. In the case of seed mass, the ability to switch rapidly, for example, from small seeds with high dispersal ability to larger seeds with lower dispersal ability, might promote cycles of rapid colonisation and isolation or permit adaptation to new dispersal vectors in novel environments. Rapid evolution of new phenotypes may also allow individuals to escape harsh environmental conditions and competitive interactions 14, thereby decreasing extinction rates (µ). The net outcome of these processes on diversification (r = λ-µ) will ultimately depend upon which of these rates responds more strongly to phenotypic change.
Here we show for the first time that both seed mass and its phenotypic rate of evolution influence speciation and extinction across the angiosperm ToL. Our approach combined the most comprehensive phylogenetic timetree available 15 with an unparalleled dataset of seed mass measurements to obtain mean seed mass values for 24% of all described angiosperm genera (n = 4105). We estimated rates of speciation and extinction using Bayesian Analysis of Macroevolutionary Mixtures (BAMM) 16, which models rate heterogeneity through time and lineages, and accounts for incomplete taxon sampling. Similarly, we analysed the rates of change in seed size across the phylogeny. We then estimated the correlations of the macroevolutionary dynamics with absolute seed size and the rate of seed size evolution.
As expected, given the high degree of taxonomic imbalance observed in the angiosperm phylogeny, we found strong support for multiple shifts in the rates of diversification. The median number of rate shifts for the speciation/extinction analysis was 123 (95% confidence interval (CI): 95-155) and 160 (95% CI: 62-110) for the seed size evolution analysis (Extended Data Fig. 1). There was marked heterogeneity in the rates of seed size evolution (Fig. 1), which varied over two orders of magnitude (Extended Data Fig. 2). We then estimated whether shifts in macroevolutionary dynamics (λ, µ and r) were significantly correlated with absolute mean genus seed size and rates of seed size evolution by comparing the existing correlations to a null distribution of correlations using STructured Rate Permutations on Phylogenies (STRAPP), which is robust to phylogenetic pseudoreplication (see Methods for details) 17.
For the first time, we were able to explain major differences in diversity across the angiosperm ToL with a single trait that integrates multiple aspects of life history. We specifically found evidence that increased speciation was associated with smaller seed sizes (Spearman’s ρ = −0.22, p-value = 0.016; Fig. 2a). Increased extinction rate was similarly associated with smaller seeds (ρ = −0.20, p-value = 0.045), but given its relatively weaker effect, the net outcome of λ-µ was that diversification rates increased with decreasing seed size (ρ =−0.19, p-value = 0.049). We also identified a stronger positive association between the rate of seed size evolution and both speciation and extinction (ρ = 0.48, p-value < 0.0001 and ρ = 0.37, p-value = 0.003, respectively; Fig. 2b). Again, as the effect of speciation was greater than that of extinction, rates of diversification and phenotypic change were positively correlated (ρ = 0.46, p-value = 0.0002; Fig. 2b). Generally, the observed correlations arose from many phenotypically fast-evolving clades distributed across the phylogeny (Extended Data Fig. 2).
Our results were unaffected by intra-generic variation in seed mass for two reasons. First, there was no systematic bias in intra-generic variation across the phylogenetic tree. We detected no correlation between the mean and the coefficient of variation (CV) for seed mass of each genus (Extended Data Fig. 3, PGLS: F1,131 = 0.67, p-value = 0.416). Second, we could disregard intra-generic variation in rates of seed mass evolution as influencing our results as we found that genera with larger variation in seed size did not have different macroevolutionary dynamics (Extended Data Fig. 4, Spearman’s ρλ = −0.05, p-value = 0.114; ρµ = −0.03, p-value = 0.406; ρr = −0.067, p-value = 0.058).
Our study supports the idea that variation in seed mass can explain disparity in diversification across the angiosperm ToL by influencing dispersal and habitat interactions. We specifically found that smaller-seeded genera had faster speciation rates. This may be because they are capable of dispersing over larger distances 3 that can result in isolated populations and eventually lead to speciation 18. Large dispersal distances may be especially important for isolation to occur in continuous habitats.
However, the positive effect on speciation from long-distance dispersal may be dampened in highly fragmented landscapes where isolation is more common over shorter distances 19. Dispersal syndromes may also alter the effect of seed size on speciation. Species with larger seeds are generally associated with biotic dispersal that distributes seeds over greater distances than wind or gravity dispersal 7. However, broad scale predictions on the effects of dispersal syndromes on diversification may be inaccurate, since the former depend on landscape connectivity 20 and can sometimes be inconsistent, e.g. a wind-dispersed seed might be transported by an animal. Detailed contextual data will be necessary to expand upon the mechanisms underlying our findings. We also found that smaller-seeded genera had higher rates of extinction, possibly due to smaller nutritional reserves that constrain establishment, environmental tolerance and access to limiting resources 21.
Seed mass is associated with other traits that can affect diversification, but there is little evidence that these better explain our observed correlations. For example, genome size positively correlates with seed mass 22 and faster rates of genome size evolution have been linked to increased speciation in angiosperms 23. Shorter, smaller-seeded plants also tend to have faster life cycles, which may accelerate mutation rates 24 and promote diversification 25. By comparing the effects of genome size and life cycle across a subset of >900 genera in our dataset, we found that only the distinction between strictly annual versus perennial genera influenced macroevolutionary dynamics in a similar way as seed size (Supplementary Information, Extended Data Fig. 5). Unlike other traits 23 both absolute seed size as well as its rate of change were correlated with speciation and extinction. Thus, while we cannot exclude unobserved traits as drivers of diversification 26, we argue that seed mass plays a central role in plant life history both on its own and by integrating across traits that should predictably shape macroevolutionary dynamics (Extended Data Fig. 6).
Our finding that high rates of phenotypic change correlate with diversification (Fig. 2b) has recently been observed in other taxonomic groups 4,11, but never across the whole of the angiosperm ToL. Accelerated morphological evolution may allow radiating lineages to occupy more complex adaptive landscapes 27. Similarly, species with greater rate of change in their seed mass (i.e., higher evolvability) could shift between adaptive peaks or develop reproductive barriers more rapidly. However, current methods do not allow us to distinguish whether speciation is responding to morphological change or vice versa when reconstructing 250 million years of evolutionary history 4.
The approach applied here can help to unravel the processes responsible for generating large-scale asymmetries in biodiversity. It also offers the potential to test how widely-varying traits influence other aspects of the evolution and adaptation of flowering plants (e.g. 15). Clade-specific exceptions arising from local interactions with non-focal traits 29 and specific spatio-temporal contexts will undoubtedly interact with broad-scale macroevolutionary patterns and may modulate the effects of seed mass on diversification. Regardless, our results clearly demonstrate that seed size, and its rate of change, drive speciation and extinction and help to explain why some clades are much more species-rich than others.
Methods
Seed mass and phylogenetic dataset
Seed mass data for 31,932 species were obtained from the Royal Botanic Gardens Kew Seed Information Database 30. Species names were standardised with The Plant List (TPL) nomenclature 31 and cleaned using the Taxonstand R package 32. Further processing at the genus-level was carried out with the taxonlookup R package 33, which is a complete genus-family-order mapping for vascular plants that draws from TPL, the Angiosperm Phylogeny website 34 and a higher-level manually curated taxonomic lookup 15. Seed mass mean values for each genus were calculated for a total of 4763 genera.
We used the most comprehensive phylogenetic tree for land plants 15,35 that comprises 31,389 species. Taxonomic information for our phylogenetic tree was run through Taxonstand and taxonlookup as described above to make it as comparable as possible to the seed mass dataset. Monophyly of the genera in the tree was assessed using the Monophy package 36 The initial estimate was 16%, but we removed taxa classified as outliers by Monophy (i.e., taxa that lay outside of an established “core clade” for each genus) and this resulted in only 8% of the genera not being monophyletic. The Phyndr 33 package was then used to generate a genus-level tree with as much overlap as possible between the phylogeny and the trait data. The final phylogenetic tree included representatives from 303 plant families (Extended Data Fig. 7).
Diversification and phenotypic evolution analysis
Speciation, extinction and net diversification rates and rates of seed size evolution were estimated separately on the phylogeny using BAMM version 2.5.0 16. BAMM models shifts in macroevolutionary regimes across a phylogenetic tree using reversible-jump Markov chain Monte Carlo (rjMCMC) sampling. Initial prior settings were calculated with the setBAMMpriors function in BAMMtools 37, and the expectedNumberOfShifts parameter was set at 25 and 10 for the speciation/extinction and trait evolution analyses, respectively. We incorporated non-random incomplete sampling information following BAMM protocols by calculating the proportion of genera sampled inside each family and estimated the backbone sampling as the overall proportion of sampled genera. Taxonlookup was used as a reference for these calculations.
All analyses were run for 50 million generations. We verified convergence by plotting chain traces and ensuring that the effective sample sizes of all relevant parameters exceeded 200. The first 10 million generations were discarded as burn-in.
Correlation of diversification and trait evolution
We used STRAPP to test for multiple associations between macroevolutionary dynamics and each of seed mass (using genus mean values as character states) and seed mass rate of evolution (using seed mass evolutionary rates at the tips of the phylogeny as character states). STRAPP compares the correlation between a focal trait and a macroevolutionary parameter (λ, µ or r) to a null distribution of correlations. The null correlations are generated by permuting the evolutionary rates in the tips of the phylogenetic tree while maintaining the location of rate shift events in the phylogeny. In each case, we calculated the absolute difference between the observed correlation of the macroevolutionary rate and the trait state and the null correlation obtained by the structured permutations across 5000 samples from the BAMM posterior. The reported p-value was the proportion of replicates where the null correlation coefficient was greater than the observed correlation. We found a low type I error associated with our STRAPP correlation analysis (p-value = 0.11, Extended Data Fig. 8).
Code availability
Scripts used to carry out the analysis described in the paper and generate the figures will be deposited in Github.
Author contributions
J.I and A.S.T.P conceived the study. J.I. and E.F.M. performed the analysis. J.I. and A.J.T interpreted the analysis and wrote the manuscript. All authors edited the manuscript.
Author information
Scripts used to carry out the analyses described in the manuscript will be uploaded to Github. Correspondence and requests for materials should be sent to J.I. (ji247{at}cam.ac.uk)
Acknowledgments
We thank D. A. Coomes, A. J. Helmstetter, T. Jucker and W. G. Lee for useful comments that helped improve the manuscript. J.I. and A.J.T. thank the Gatsby Charitable Foundation, Wellcome Trust and Newton Trust for funding. E.F.M was funded by the BBSRC DTP at the University of Cambridge.