vol. 166, no. 3 the american naturalist september 2005 ൴

Notes and Comments Environment, Area, and Diversification in the Species-Rich Flowering Family

T. Jonathan Davies,1,2,* Vincent Savolainen,2,† Mark W. Chase,2,‡ Peter Goldblatt,3,§ and Timothy G. Barraclough1,2,k

1. Division of Biology and Natural Environmental Research The distribution of species richness among branches of Council Centre for Population Biology, Imperial College London, the tree of life is highly uneven, with a small number of Silwood Park Campus, Ascot, Berkshire SL5 7PY, United clades possessing a disproportionately large number of Kingdom; 2. Molecular Systematics Section, Jodrell Laboratory, Royal Botanic species (Marzluff and Dial 1991). Similarly, species rich- Gardens, Kew TW9 3DS, United Kingdom; ness varies markedly among different geographic regions 3. Missouri Botanical Garden, St. Louis, Missouri 63166 (Gaston 2000). To date, approaches for studying these two patterns have been largely separate. The causes of taxo- Submitted August 5, 2004; Accepted April 12, 2005; nomic imbalance have been explored using phylogenetic Electronically published June 28, 2005 approaches, focusing largely on finding biological traits that explain variation in species richness. Although such Online enhancements: appendix, tables. traits have been found (Hodges and Arnold 1995; Heilbuth 2000; Sargent 2003; Gianoli 2004; Ree 2005), the propor- tion of variation in species richness they explain tends to be low (Harvey and Purvis 2003; Ricklefs 2003). In con- abstract: Phylogenetic analyses provide a means to explore evo- trast, most work on geographic variation in species rich- lutionary explanations for regional variation in species richness. The ness has followed an ecological approach, for example, environment might also explain much of the previously unexplained looking at the effects of area on species richness or for imbalance of phylogenetic trees. We use data on geographic distri- bution and phylogenetic affinity to examine correlates of species regional differences in the abiotic environment that explain richness among genera of irises (family: Iridaceae). Irises display spatial variation of species richness (Willig et al. 2003; strong phylogenetic imbalance, with a few clades containing a dis- Currie et al. 2004; see also Currie and Francis 2004; Qian proportionate number of species, most notably those found in the and Ricklefs 2004). dry Mediterranean climate of the Cape of South Africa. The abiotic Combining these approaches can provide further in- environment and area are strong predictors of iris species richness, sights into both areas. Phylogenetics provides a means to but environment alone is insufficient to explain the high diversity evaluate the importance of evolutionary history in deter- of Cape clades. One possible explanation is that the interaction be- mining geographic patterns in diversity. For example, do tween biological traits and environment resulted in the unusually high diversification rates in the region. regions with more species have older clades or clades with faster net diversification rates (Cardillo 1999)? In addition, Keywords: species richness, environment, diversification rates, Iri- the environment might provide the missing explanation daceae, Cape of South Africa. for taxonomic imbalance in species richness: clades occupy regions with different areas or different abiotic environ- ments, thereby influencing net diversification rates (Rick- lefs 2003). For example, Davies et al. (2004) showed that * Present address: Department of Biology, Gilmer Hall, University of Virginia, area and measures of environmental energy, such as tem- Charlottesville, Virginia 22904; e-mail: [email protected]. perature, explained ∼40% and ∼20%, respectively, of the † E-mail: [email protected]. variation in species richness between sister families of ‡ E-mail: [email protected]. angiosperms. § E-mail: [email protected]. Here we use a generic-level phylogenetic analysis of Iri- k E-mail: [email protected]. daceae to examine environmental correlates of species Am. Nat. 2005. Vol. 166, pp. 418–425. ᭧ 2005 by The University of Chicago. richness. By comparing younger and more narrowly dis- 0003-0147/2005/16603-40565$15.00. All rights reserved. tributed taxa than Davies et al. (2004), it was hoped to Environment, Area, and Diversification 419 more precisely discriminate differences in environments Geographic Data between clades. We chose Iridaceae largely because a nearly complete -level phylogenetic analysis was already Generic distribution maps were assembled using a range available and also because they display important char- of revisionary and floristic accounts dealing with Iridaceae, acteristics relating to both taxonomic and geographic var- notably Mathew (1981), Goldblatt and Henrich (1987), iation in species richness. Taxonomically, Iridaceae are a and Goldblatt and Manning (2002). These were digitized highly diverse family with ∼65 genera. The family displays in ArcView (GIS 3.2, Environmental Systems Research In- a strong pattern of taxonomic imbalance: a few genera stitute). The total area of each distribution was calculated contain large numbers of species (e.g., Crocus [80 species], and log transformed. Iris [280], and Gladiolus [260]), whereas many genera con- For measures of abiotic environment, following broad tain only one or a few species (e.g., Isophysis [1], Devia ecological surveys of plant diversity (Currie 1991; [1], and Onira [1]; Goldblatt 2001). O’Brien et al. 2000; Francis and Currie 2003), we focused Geographically, Iridaceae favor dry Mediterranean-type on variables summarizing energy input, water availability, climates and are especially species rich in the Cape region and elevation in each region. We collated raster data sets of South Africa, defined here as the geographic region representing temperature (1931–1960, Global Ecosystem characterized by a Mediterranean climate and encom- database, http://www.ngdc.noaa.gov/seg/cdroms/ged_iia/ passing the Cape Folded Mountain Belt (Goldblatt and datasets/a03/lc.htm), ultraviolet (UV) radiation (1978– Manning 2002). The Cape is renowned for its high levels 1993, NASA/GSFC TOMS Team, http://jwocky.gsfc of plant species richness and endemism. As a small and .nasa.gov), actual evapotranspiration (AET; 1920–1980, GRID, UNEP, http://www.grid.unep.ch/data/data.php essentially temperate climate region, the Cape has been p highlighted as an unexplained outlier from global trends ?category atmosphere), elevation (GTOPO30, U.S. Geo- between the abiotic environment and species richness logical Survey’s EROS Data Center, http://edcdaac.usgs (Cowling et al. 1992; Simmons and Cowling 1996; Rich- .gov/gtopo30/gtopo30.html), and mean degrees latitude ardson et al. 2001; Goldblatt and Manning 2002; Linder from the equator. We calculated the mean value of each 2003). The two broad explanations for high species rich- variable within each generic distribution. To compare sister ness in the Cape are readily amenable to phylogenetic clades deeper in the phylogenetic tree, generic range dis- analyses: either the Cape represents an old, relatively un- tributions within each clade were combined to form a disturbed area able to accumulate species richness or the single polygon theme, and mean environmental values recent onset of its Mediterranean-type climate triggered were calculated as described above. The environmental data are resolved to between 0.25Њ and 0.5Њ, equating to rapid diversification. Here we ask two questions. First, are 2 2 measures of the abiotic environment sufficient to explain cell sizes of approximately 750 km and 3,025 km ,re- spectively, at the equator. The smallest geographic range the diversity of irises in the Cape and elsewhere? Although 2 the Cape departs from general trends between the envi- was that for Devia at around 12,000 km . ronment and flowering-plant species richness, it remains possible that those lineages that have radiated extensively Evolutionary Rates in the Cape fall within clades that follow a different func- tional response to the abiotic environment that can still Following Davies et al. (2004), we included a measure of explain the clades’ geographic variation in species richness. molecular evolution rates to attempt to distinguish Second, is high diversity of Iridaceae in the Cape due to whether any observed correlation between species richness rapid diversification or to the presence of old clades? and energy might be explained by the intermediate effect of environment on evolutionary rates (Rohde 1992). Max- imum likelihood branch lengths were estimated using the Material and Methods HKY85 model of DNA evolution (Hasegawa et al. 1985) Phylogenetic Data for synonymous and nonsynonymous changes. Further partitioning of the data was not supported by nested log- We use one of the two most parsimonious phylogenetic likelihood ratio tests (results not shown). A comparison trees from a recent molecular study by Goldblatt et al. with branch lengths estimated using a more complex (2005), sampling 58 of the 65 currently recognized genera model of DNA evolution (general time reversal using within Iridaceae. Species numbers were taken from Gold- a gamma distribution with an a shape parameter and a blatt (2001). Significant shifts in rates of diversification proportion of invariant sites) found strong correspon- were identified by contrasting species richness between dence between the two estimates (r 2 p 0.99 for the re- sister clades using the method of Slowinski and Guyer gression with both the synonymous and nonsynonymous (1993). partitions). 420 The American Naturalist

Construction of Linear Models Missing Taxa

We constructed linear models to explore the correlation Our analyses assume that all species within the lineages among contrasts in environmental measures, molecular compared have been sampled, although seven genera could rates (explanatory variables), and species richness (re- not be included in the phylogenetic tree reconstruction sponse variable) between sister clades. All procedures fol- because suitable tissue samples were unavailable. We there- low Davies et al. (2004). We calculated species richness fore used published statements about their most likely contrasts as phylogenetic affinities and collapsed clades in which it was thought they were likely to fall to one representative taxon (table B1 in the online edition of the American Naturalist). log (species in clade A) Ϫ log (species in clade B). Branch lengths and geographic distributions were esti- mated as described for nodes deeper in the tree. The com- To correct for the tendency of older clades to differ more plete analysis was then repeated, producing four additional in species richness, contrasts were divided by the relative minimum adequate models. We regarded the model de- age of the node subtending the two clades. The ages of rived from the nonnested sister contrasts (assuring phy- sister clades were estimated from the DNA sequence data logenetic independence), weighted by bootstrap percent- using the nonparametric rate-smoothing method of San- ages (minimizing the influence of phylogenetic error) and derson (1997) to adjust for nonconstancy of rates across for which we had accounted for the influence of missing lineages (further details provided in app. A in the online taxa, as described above, to be the most conservative. edition of the American Naturalist). Contrasts in the en- vironmental variables and molecular rates were calculated Ϫ as (XA) (XB), where X is molecular branch length or the Results mean of the environmental variable as derived from dis- All minimum adequate models derived from the linear tribution maps. We standardized the variance among regression were highly significant (P ! .001 ; tables C1 and branch length contrasts by dividing by the mean of the C2 in the online edition of the American Naturalist). Area branch lengths leading to both sister families. is the single best predictor of generic species richness, ex- Regression through the origin (Harvey and Pagel 1991) plaining 56% to 66% of the variation. Inclusion of envi- was performed using the statistical package R, version 1.6.0 ronmental variables and molecular rates into the model (R Development Core Team 2002). Model simplification increased its explanatory power by between 7% and 29%. was performed by removing nonsignificant parameters in Exactly which variables were retained within minimum a stepwise fashion to produce a minimum adequate model adequate models was sensitive to alternative treatments; (Crawley 2002). however, their direction of influence was consistent be- tween models (fig. 1). All models conformed to the as- sumptions of constancy of variance and normality of er- Model Sensitivity rors; hence, our results are not a product of a small number of contrasts of high influence but reflect a consistent pat- We performed three alternative treatments to examine ro- tern among Iridaceae lineages. bustness of our results and sensitivity to phylogenetic er- Of the measures of molecular rates, only synonymous ror. First, we constructed linear models including the max- changes were significantly correlated with species richness. imum set of sister clades for which there was no taxonomic If environment influenced diversification via its effect on overlap (nonnested contrasts), that is, no clades shared rates of molecular evolution, we would expect the latter between different contrasts, chosen to maximize the spe- to be the more immediate predictor of species richness. cies richness contrasts of included sister comparisons. Sec- However, we found that both environment and molecular ond, we included contrasts between sister clades for all rates were retained in the majority of our models, indi- nodes within the tree. For nodes deeper in the tree, mo- cating that both correlate with species richness indepen- lecular rates were estimated by successively averaging dently, consistent with previous findings among angio- branch lengths from the tips toward the root. Third, we sperm families (Davies et al. 2004). repeated models one and two but weighted contrasts by Sister clades were found to differ greatly in species rich- bootstrap support for monophyly of the respective clades; ness, indicating high variability in net diversification rates bootstrap percentages below 75% were given zero weight. among lineages in the phylogenetic tree. Ten clades were A total of four minimum adequate models were generated. identified as having a significantly greater number of spe- Model criticism was performed following Crawley (2002), cies than their sister clades (table D1 in the online edition as described in Davies et al. (2004). of the American Naturalist). The node subtending Isophysis Environment, Area, and Diversification 421

synium, Solenomelus, and Hesperoxiphion). Although a few South African genera in are also overpredicted by the model (e.g., Chasmanthe, Radinosiphon, and Du- thiastrum), their departure is more than an order of mag- nitude less than that observed among non-African clades. High species richness in Cape clades might simply be a consequence of these lineages having accumulated spe- cies over a greater time. However, differences in species richness reflect variation in net diversification rates be- cause model predictions were generated from contrasts between sister clades, which are by definition the same age. Clades with more species than predicted by the model have had higher diversification rates than expected. Fur- thermore, genera with southern African distributions tend to be younger than genera found elsewhere (median clade age of 18 million years and 22 million years, respectively). Hence, lineages in the Cape have diversified at a faster rate than non-Cape clades even when compared with lin- Figure 1: Range of t values for each of the parameters retained within eages found in regions of similar Mediterranean climates. one or more of the minimum adequate models. Boxes represent the first and third quartile. Values above the box plots indicate the number of models out of a total of eight within which the respective parameters Discussion were retained. Our model can explain up to 85% of the variation in species richness among Iridaceae clades. Whereas area is and the remainder of Iridaceae is the most imbalanced; the single most important variable, environmental factors the next most-basal nodes are also highly imbalanced, in- independently explain an additional ∼20% of the variation dicating that the ancestral state may have been a low net in net diversification rates between lineages. We find that diversification rate. However, only two of the significantly parameters associated with warm, topographically diverse imbalanced contrasts demonstrate a worse fit to the model habitats and low rainfall are the best predictors of species than the mean (mean magnitude of residual deviation for richness, confirming the family’s departure from more all contrasts: 0.07). In both cases the species-poor clades, general global trends of higher species richness in more Isophysis and Diplarrhena, are sister to lineages including productive environments (Davies et al. 2004). Adapted to a large number of clades, perhaps highlighting the limi- neither the intense competition for light nor the rapid tations of averaging environmental values for more inclu- growth required for gap colonization within dense vege- sive clades. Variation in species richness among the re- tation typical of tropical regions, Iridaceae instead favor mainder of the imbalanced nodes can be largely attributed seasonally dry environments (Rudall 1995). Although a to differences in environment included within our model. few genera, such as Neomarica (8) and Eleutherine (2), are Of genera with the greatest deviation from the model found in the Neotropics, they are relatively species poor. (table 1), all those that contain more species than predicted Hence, the environment including area can explain a large (Geissorhiza, Hesperantha, Ixia, and Therianthus)fall proportion of the variation in net diversification rates within Crocoideae, indicating a heritable component to among clades even in cases known to contradict global diversification not included within the model parameters. trends. Whereas a number of biological traits are characteristic of The richness of the Cape is consistently underpredicted Crocoideae, there are insufficient state transitions to test by general climate-energy models (O’Brien et al. 2000; for a general correlation between traits and species rich- Taplin and Lovett 2003). In principle, it is possible that ness. The most prominent distinction between genera with clades favoring the Cape, such as Iridaceae, might display fewer species than predicted by the model and the genera a different response to energy measures than responses identified above is in their geographic distribution. that characterize global biodiversity patterns, but that re- Crocoideae are a largely southern African clade, and we gional variation in their diversity is still predictable by a find our model consistently underpredicts species richness climate model. Despite the high explanatory power of our of clades in this region, in particular the Cape (fig. 2). By model, we find that environment is insufficient to explain comparison, non-Cape genera tend to contain fewer spe- the high diversity of irises in the Cape at the scale of our cies than predicted by the environment model (e.g., Ol- analysis. Our results also indicate that lineages in the Cape Table 1: Ten genera with the greatest deviation in species richness (residual deviation) from the most conservative model Species Residual Geographic Underground Genus number deviation distribution organ Pollinator Dispersal Iris (Ϫ) 280 .33 North Africa/Eurasia/ Rhizomea Actinomorphicb Bee/bird Ant?/passive North America Geissorhiza (ϩ) 85 .22 South Africa Corm Actinomorphic/ Bee/long proboscid fly/beetle/generalist Passive zygomorphic Olsynium (Ϫ) 12 .20 Americas Actinomorphic Bee/butterfly?/generalist Passive Hesperantha (ϩ) 79 .19 Southern/tropical Cormc Actinomorphicd Bee/long proboscid fly/moth/butterfly/beetle Wind/passive Africa Ixia (ϩ) 52 .16 Southern Africa Corm Actinomorphic Bee/long proboscid fly/beetle/generalist Passive Solenomelus (Ϫ) 2 .14 Rhizome Actinomorphic Bee? Passive Thereianthus (ϩ) 7 .12 South Africa Corm Actinomorphic/ Bee/long proboscid fly/butterfly?/beetle Passive zygomorphic Isophysis (Ϫ) 1 .11 Tasmania Rhizome Actinomorphic ? ? Hesperoxiphion (Ϫ) 4 .10 South America Actinomorphic ? ? Diplarrhena (Ϫ) 2 .09 Australia Rhizome Zygomorphic ? Wind? Note: Clades suffixed with (ϩ) have greater species numbers than predicted by the model; those suffixed by (Ϫ) have fewer species than predicted.? p uncertain character state. a Some with ; in one instance, a tuber-like corm. b Flower composed of three units, each of which is zygomorphic. c One species with . d One species with zygomorphic flower. Environment, Area, and Diversification 423

Figure 2: Spatial variation in the difference between predicted and actual generic species richness: overpredicting (a) and underpredicting (b). Values derived by summing the generic distribution maps weighted by the difference between the observed and predicted species richness from the most conservative model (tables C1 and C2 in the online edition of the American Naturalist). Dark shading represents high disparity between actual and predicted species richness; light shading represents low disparity. Note that although the model was derived from contrasts rather than genus values, Pagel (1993) showed that residual values for taxa can be calculated reliably using model parameters from contrasts analysis. have undergone more rapid diversification than those us from discriminating environmental parameters at fine found elsewhere. The results are consistent with molecular enough resolution to obtain a fully predictive model. The evidence for a recent origin of Cape endemics (Richardson geographic distribution of species richness within genera et al. 2001). is not uniform; for example, 75% of the species within What factors might explain the residual variation in Moraea are endemic to the Cape, but the clade’s total species richness, particularly in the Cape? First, the spatial geographic extent stretches to Ethiopia and from the Med- and taxonomic scale of our analyses may have prevented iterranean to the Middle East (Goldblatt et al. 2002). As 424 The American Naturalist geographic extent increases, so does the tendency for mean Acknowledgments values to tend toward a global average. In principle, this We are grateful for assistance from B. Mathew in compiling trend could lead to underrepresentation of the effects of distribution maps for Europe and M. van der Bank in the a narrowly distributed climate region such as the Cape DNA sequencing. P. Linder, S. Renner, and an anonymous and hence the underprediction of species richness. How- reviewer provided useful comments and suggestions. This ever, the fact that we could distinguish the Cape as an work was partly supported by the Darwin Initiative, Lev- outlier in our mapping of residuals indicates that there erhulme Trust, National Geographic Society, and the were a sufficient number of narrowly distributed clades to United Kingdom Natural Environment Research Council. define the region. Clearly, further spatial and phylogenetic resolution is desirable in such studies, once suitable phy- Literature Cited logenetic data become available. Second, some aspect of the environment absent from Bernhardt, P., and P. Goldblatt. 2000. The diversity of pollination mechanisms in the Iridaceae of southern Africa. Pages 301–308 our models might explain residual variation. For example, in K. L. Wilson and D. A. Morrison, eds. Monocots: systematics fire is thought to be a key factor explaining Cape diversity and evolution. CSIRO, Melbourne. (Cowling et al. 1992). If the frequency of fire is not pre- Cardillo, M. 1999. Latitude and rates of diversification in birds and dicted by the variables in our model, this might be an butterflies. Proceedings of the Royal Society of London B 266: additional factor that could explain variation in species 1221–1225. richness between Cape and non-Cape clades. Alternatively, Cowling, R. M., P. M. Holmes, and A. G. Rebelo. 1992. Plant diversity the Cape is hypothesized to have experienced less climatic and endemism. Pages 62–112 in R. M. Cowling, ed. The ecology of fynbos. Oxford University Press, Cape Town. variability than other regions during the Pleistocene, which Crawley, M. J. 2002. Statistical computing: an introduction to data might have lowered extinction rates and increased net di- analysis using S-plus. Wiley, Chichester. versification rates (Dynesius and Jansson 2000). More Currie, D. J. 1991. Energy and large-scale patterns of animal-species complex models could be devised that included these and and plant-species richness. American Naturalist 137:27–49. other environmental measures if suitable GIS data were Currie, D. J., and A. P. Francis. 2004. Regional versus climate effect available for these measures. on taxon richness in angiosperms: reply to Qian and Ricklefs. American Naturalist 163:780–785. Finally, biological traits might play a role in explaining Currie, D. J., G. G. Mittelbach, H. V. Cornell, R. Field, J.-F. Gue´gan, residual variation. Many of the Cape genera have a number B. A. Hawkins, D. M. Kaufman, et al. 2004. Predictions and tests of traits in common, including zygomorphic flowers, cor- of climate-based hypotheses of broad-scale variation in taxonomic mous rootstocks, and the absence of highly developed richness. Ecology Letters 7:1121–1134. mechanisms for long-distance seed dispersal (table 1). Davies, T. J., V. Savolainen, M. W. Chase, J. Moat, and T. G. Bar- These traits have all been argued to promote reproductive raclough. 2004. Environmental energy and evolutionary rates in flowering . Proceedings of the Royal Society of London B isolation (Goldblatt 1991; Bernhardt and Goldblatt 2000; 271:2195–2200. Goldblatt and Manning 2002). However, several species- Dynesius, M., and R. Jansson. 2000. Evolutionary consequences of poor genera in other regions, some of which have fewer changes in species’ geographical distributions driven by Milan- species than predicted by the model, share many of the kovitch climate oscillations. Proceedings of the National Academy above key traits with their species-rich counterparts. One of Sciences of the USA 97:9115–9120. interpretation is that such traits might provide the key for Francis, A. P., and D. J. Currie. 2003. A globally consistent richness- climate relationship for angiosperms. American Naturalist 161: clades to invade and prosper in the Cape region, thereby 523–536. exploiting the potential for rapid diversification conveyed Gaston, K. J. 2000. Global patterns in biodiversity. Nature 405:220– by the Cape environment. The sample size of evolutionary 227. transitions in the Iridaceae phylogenetic tree is too small Gianoli, E. 2004. Evolution of a climbing habit promotes diversifi- to explore the interaction between biological traits and cation in flowering plants. Proceedings of the Royal Society of environment further, but such analyses will be possible as London B 271:2011–2015. Goldblatt, P. 1991. An overview of the systematics, phylogeny and more phylogenetic trees become available for clades that biology of the African Iridaceae. Contributions from the Bolus have independently colonized the Cape region. Herbarium 13:1–74. In conclusion, phylogenetic and geographic approaches ———. 2001. Phylogeny and classification of the Iridaceae and the for studying species richness are mutually informative. relationship of Iris. Annali di Botanica 1:13–28. Phylogenetic analyses provide a means to explore in which Goldblatt, P., and J. E. Henrich. 1987. Notes on Cipura (Iridaceae) geographic regions diversification rates have been unusu- in South and Central America, and a new species from Venezuela. Annals of the Missouri Botanical Garden 74:333–340. ally high or low and to evaluate putative causes. Environ- Goldblatt, P., and J. C. Manning. 2002. Plant diversity of the Cape mental variables offer the potential to explain much of the region of southern Africa. Annals of the Missouri Botanical Garden previously unexplained imbalance of phylogenetic trees. 89:281–302. Environment, Area, and Diversification 425

Goldblatt, P., V. Savolainen, O. Porteous, I. Sostaric, M. Powell, G. for statistical computing. R Foundation for Statistical Computing, Reeves, J. C. Manning, et al. 2002. Radiation in the Cape flora Vienna. http://www.R-project.org. and the phylogeny of peacock irises Moraea (Iridaceae) based on Ree, R. H. 2005. Detecting the historical signature of key innovations four plastid DNA regions. Molecular Phylogenetics and Evolution using stochastic models of character evolution and cladogenesis. 25:341–360. Evolution 59:257–265. Goldblatt, P., T. J. Davies, V. Savolainen, J. C. Manning, and M. Richardson, J. E., F. M. Weitz, M. F. Fay, Q. C. B. Cronk, H. P. Linder, van der Bank. 2005. Phylogeny of Iridaceae subfamily Crocoideae G. Reeves, and M. W. Chase. 2001. Rapid and recent origin of based on plastid DNAs. In J. T. Columbus, E. A. Friar, J. M. Porter, species richness in the Cape flora of South Africa. Nature 412: L. M. Prince, and M. G. Simpson, eds. Monocots: comparative 181–183. biology and evolution. Rancho Santa Ana Botanic Garden, Clare- Ricklefs, R. E. 2003. Global diversification rates of passerine birds. mont, CA (forthcoming). Proceedings of the Royal Society of London B 271:2285–2291. Harvey, P. H., and M. D. Pagel. 1991. The comparative method in Rohde, K. 1992. Latitudinal gradients in species-diversity: the search evolutionary biology. Oxford University Press, Oxford. for the primary cause. Oikos 65:514–527. Harvey, P. H., and A. Purvis. 2003. Opportunity versus innovation. Rudall, P. 1995. Anatomy of the . VIII. Iridaceae. Nature 425:676–677. Oxford University Press, Oxford. Hasegawa, M., H. Kishino, and T. Yano. 1985. Dating the human- Sanderson, M. J. 1997. A nonparametric approach to estimating di- ape splitting by a molecular clock of mitochondrial DNA. Journal vergence times in the absence of rate constancy. Molecular Biology of Molecular Evolution 21:160–174. and Evolution 14:1218–1231. Heilbuth, J. C. 2000. Lower species richness in dioecious clades. Sargent, R. D. 2003. Floral symmetry affects speciation rates in an- American Naturalist 156:221–241. giosperms. Proceedings of the Royal Society of London B 271:603– Hodges, S. A., and M. L. Arnold. 1995. Spurring plant diversification: 608. are floral nectar spurs a key innovation? Proceedings of the Royal Simmons, M. T., and R. M. Cowling. 1996. Why is the Cape Peninsula Society of London B 262:343–348. so rich in plant species? an analysis of the independent diversity Linder, H. P. 2003. Radiation of the Cape flora, southern Africa. components. Biodiversity and Conservation 5:551–573. Biological Reviews 78:597–638. Slowinski, J. B., and C. Guyer. 1993. Testing whether certain traits Marzluff, J. M., and K. P. Dial. 1991. Life history correlates of tax- have caused amplified diversification: an improved method based onomic diversity. Ecology 72:428–439. on a model of random speciation and extinction. American Nat- Mathew, B. 1981. The iris. Timber, Portland, OR. uralist 142:1019–1024. O’Brien, E. M., R. Field, and R. J. Whittaker. 2000. Climatic gradients Taplin, J. R. D., and J. C. Lovett. 2003. Can we predict centres of in woody plant (tree and shrub) diversity: water-energy dynamics, plant species richness and rarity from environmental variables in residual variation, and topography. Oikos 89:588–600. sub-Saharan Africa? Botanical Journal of the Linnean Society 142: Pagel, M. 1993. Seeking the evolutionary regression coefficient: an 187–197. analysis of what comparative methods measure. Journal of The- Willig, M. R., D. M. Kaufman, and R. D. Stevens. 2003. Latitudinal oretical Biology 164:191–205. gradients of biodiversity: pattern, process, scale, and synthesis. Qian, H., and R. E. Ricklefs. 2004. Taxon richness and climate in Annual Review of Ecology and Systematics 34:273–309. angiosperms: is there a globally consistent relationship that pre- cludes region effects? American Naturalist 163:773–779. Associate Editor: Susanne S. Renner R Development Core Team. 2002. R: a language and environment Editor: Jonathan B. Losos