ORIGINAL ARTICLE doi:10.1111/j.1558-5646.2010.01131.x PLANT RESPONSES TO CLIMATE IN THE CAPE FLORISTIC REGION OF SOUTH AFRICA: EVIDENCE FOR ADAPTIVE DIFFERENTIATION IN THE PROTEACEAE Jane E. Carlson,1,2 Kent E. Holsinger,1 and Rachel Prunier1,3 1Department of Ecology and Evolutionary Biology, University of Connecticut, U-3043, Storrs, Connecticut 06268 2E-mail: jane.carlson@uconn.edu 3Kellogg Biological Station, Michigan State University, 3700 E. Gull Lake Dr., Hickory Corners, Michigan 49060 Received April 9, 2010 Accepted August 24, 2010 Local adaptation along environmental gradients may drive plant species radiation within the Cape Floristic Region (CFR), yet few studies examine the role of ecologically based divergent selection within CFR clades. In this study, we ask whether populations within the monophyletic white protea clade (Protea section Exsertae, Proteaceae) differ in key functional traits along environ- mental gradients and whether differences are consistent with local adaptation. Using seven taxa, we measured trait���environment associations and selection gradients across 35 populations of wild adults and their offspring grown in two common gardens. Focal traits were leaf size and shape, specific leaf area (SLA), stomatal density, growth, and photosynthetic rate. Analyses on wild and common garden plants revealed heritable trait differences that were associated with gradients in rainfall seasonality, drought stress, cold stress, and less frequently, soil fertility. Divergent selection between gardens generally matched trait���environment correlations and literature-based predictions, yet variation in selection regimes among wild populations generally did not. Thus, selection via seedling survival may promote gradient-wide differences in SLA and leaf area more than does selection via adult fecundity. By focusing on the traits, life stages, and environmental clines that drive divergent selection, our study uniquely demonstrates adaptive differentiation among plant populations in the CFR. KEY WORDS: Adaptation, adaptive radiation, selection���natural. Natural selection along environmental gradients can result in phe- notypic differentiation and local adaptation among wild popula- tions (Endler 1986 Linhart and Grant 1996 Ackerly et al. 2000 Geber and Griffen 2003) and can even drive speciation (Kawecki and Ebert 2004 Schluter 2009). An expected outcome of local adaptation is that traits and environments co-vary, as is commonly observed in both plant and animal taxa (e.g., Clausen et al. 1940 Dobzhansky 1947 Reich et al. 2003). Local adaptation can also contribute to species diversification into adaptive radiations, with each species specializing on only a portion of the natural variation in climate, habitat type, or interacting species (e.g., Hodges 1997 Schluter 2000 Friar et al. 2006). Yet not all among-population divergence is adaptive, and by extension, neither are all evolu- tionary radiations generated or maintained through adaptive pro- cesses (Schluter 2000 Comes et al. 2008). Among-population and among-species trait differences may also reflect environmen- tal plasticity or nonadaptive historical processes, such as disper- sal limitation and random genetic drift, yet efforts to distinguish among these alternatives are rare (but see e.g., Thorpe et al. 2005 Ellis et al. 2006 Comes et al. 2008). If local adaptation cur- rently promotes trait diversity within a species or clade, trait- environmental correlations should not only exist, they should also 1 0 8 C 2010 The Author(s). Evolution C 2010 The Society for the Study of Evolution. Evolution 65-1: 108���124

ADAPTIVE DIFFERENTIATION IN PROTEAS be promoted and maintained by ecologically based divergent se- lection (Ackerly et al. 2000 Kawecki and Ebert 2004). The Cape Floristic Region (CFR) of South Africa is char- acterized by high plant species richness���linked to extensive evolutionary radiations���and by steep environmental gradients (Schulze 1997 Goldblatt and Manning 2000 Linder 2003). The degree to which these characteristics are associated has attracted significant research interest (e.g., Cowling et al. 1997 Goldblatt et al. 2002 Linder 2003), but the role of environmental adapta- tion in generating diversity has been investigated for only a few Cape species (Verboom et al. 2004 Ellis and Weis 2006 Latimer et al. 2009). Much of the plant diversity in the CFR occurs in the fynbos, which are semi-arid, nutrient-poor scrublands that burn every ���12���30 years (Van Wilgen 1982). This relatively small re- gion (90,000 km2) has experienced numerous bouts of uplift and erosion over its geological history, resulting in over a dozen moun- tain ranges of varying elevations (up to 2300 m) that are dissected by arid regions where fynbos species do not occur (Partridge 1997). Associated with the mountain ranges are steep gradients in aridity, temperature, length of growing season, and soil fertility. Furthermore, the timing and amount of rainfall varies across the CFR: in the extreme Western Cape, rainfall is very heavy in the wintertime (May���August), becomes more aseasonal toward the eastern limit of the CFR, and shifts to a summer time peak (November���February) toward the eastern coast of South Africa. Plants respond to environmental gradients like those in the CFR through morphological and physiological traits affecting lo- cal performance and survival, known as functional traits (Geber and Griffen 2003 Reich et al. 2003). Examples of vegetative functional traits include leaf size and shape, stomatal density, photosynthetic capacity, height, and specific leaf area (SLA), that is, leaf area per unit mass. Among these, SLA, or its inverse leaf mass per area, contributes to a major axis of differentiation in leaf functional strategies (i.e., the leaf economics spectrum Wright et al. 2004). Low SLA leaves tend to be longer-lived, have lower nutrient concentrations, and be less prone to wilting. It follows that SLA and associated traits should differ along drought, cold, and soil fertility gradients (reviewed in Reich et al. 2003). For example, across species, populations, and even biomes, SLA, leaf size, and growth rate generally decrease with increasing drought and nutrient limitation (Fonseca et al. 2000 Li et al. 2000 Lamont et al. 2002 Knight and Ackerly 2003 Ordo�� nez et al. 2009), as do growth rate and leaf area with altitude/cold stress (Grime 1977 Woodward 1986 Oleksyn et al. 1998). In contrast, light-saturated photosynthetic capacity tends to increases with altitude or drought stress (Gurevitch 1992 Ashton and Berlyn 1994 Benowicz et al. 2000 but see Jonas and Geber 1999). For a few of these traits, changes in selection regimes along drought and elevation gradi- ents have been shown to reinforce trait���environment correlations within species (e.g., Etterson 2004 Byars et al. 2007). Tests of such relationships are rare, however, and they have yet to be per- formed within the CFR. Most studies of adaptive trait variation in the CFR have fo- cused on gradients in aridity and soil fertility, because these are considered the most important environmental stresses in the re- gion (Cowling et al. 1997). For example, species that occur in more arid environments in the CFR produce smaller or narrower leaves (Thuiller et al. 2004 Yates et al. 2010), have lower stom- atal density (Richardson and Kruger 1990), or have lower SLA (Latimer 2006) than species in less-arid environments. Although various studies have linked trait variability to key environmental gradients in the CFR, it remains poorly understood how these traits and gradients contributed to within- and among-species diversi- fication (Stock et al. 1997). Reciprocal transplant experiments and common gardens provide some evidence for local adapta- tion along rainfall gradients (Latimer et al. 2009) and to different soil characteristics (Verboom et al. 2004 Ellis and Weis 2006) however, little work has been done to identify which traits me- diate these adaptive responses, and which are most important in population divergence along these clines. Our overall objective is to determine if among-population differences in vegetative traits are related to natural selection and local adaptation in a representative evolutionary radiation in the CFR: the monophyletic white protea clade (Protea sect. Exser- tae, Proteaceae). Our tests of adaptive differentiation in the white proteas follow several predictions. First, trait differences among populations should be genetically based, such that they may re- spondtoselection.Second,functionalvegetativetraitsshouldvary along environmental gradients in ways that broadly match well- established trends from the literature. Third, selection gradients should change in strength and direction along key environmental gradients, and in doing so, they should match ecotypic variation among wild populations. To test these predictions, we measured trait���environment associations and selection gradients in relation to the environment using adult plants in 35 wild populations and their offspring grown in two common gardens. Our specific re- search questions are as follows: (1) Do white protea populations differ in key vegetative func- tional traits when grown in a common environment? that is, do traits have a strong genetic component? (2) Are among-population trait differences associated with strong CFR environmental gradients such as drought and cold stress, aridity, and soil fertility? that is, do traits in wild populations or in common gardens correlate with source environments? (3) Are trait���environment correlations maintained by ecolog- ically based divergent selection as inferred from (a) 35 populations of wild adults that span steep environmen- tal clines and (b) the two common gardens with different EVOLUTION JANUARY 2011 1 0 9

JANE E. CARLSON ET AL. Figure 1. Locations of the 35 white protea populations sampled in South Africa between February 2008 and May 2009. The black ar- row points to the Kirstenbosch garden, and the gray arrow points to the Jonaskop garden. environments? that is, do local selection gradients in wild populations co-vary with environment, and do selection regimes differ between gardens? Materials and Methods STUDY SPECIES We investigated trait���environment correlations and local selection gradients across all six species in the white protea clade, including both subspecies of Protea aurea: Protea aurea (Burm. f.) Rourke ssp. aurea, P. aurea ssp. potbergensis (Rourke) Rourke, P. lacti- color Salisb., P. mundii Klotzsch, P. punctata Meisn., P. subvestita N.E. Br., and P. venusta Compton. A recent phylogenetic anal- ysis of South African members of the genus Protea found 99% posterior support for monophyly of white proteas (Valente et al. 2010). White proteas are broad-leaved, sclerophyllous shrubs, and all are endemic to the CFR except P. subvestita, which occurs in the Eastern Cape, Kwa-Zulu Natal, and Lesotho (Rebelo 2001). Species distributions are largely allopatric and differ partially in altitude and climate (Latimer et al. 2009 Fig. 1), although over- lap along environmental axes is significant (Latimer 2006). For example, P. mundii and P. aurea occur at relatively low altitudes (up to 1300 m), P. lacticolor is intermediate (600���1500 m), and P. punctata, P. venusta, and P. subvestita all occupy relatively high altitude sites (1200 to over 2000 m Rebelo 2001). Among the white protea taxa, only P. punctata and P. venusta co-occur. These two species hybridize readily in the wild, and most others hybridize readily in cultivation. The seven white protea taxa share a suite of key life-history traits. They are all long-lived, evergreen perennials and most grow upright to over 4 m in height. Only P. venusta has a lateral growth habit, forming low-lying, dense mats. All species take at least 3 years to reach reproductive maturity, after which annual flower production is linked to their sympodial growth pattern. White proteas are pollinated mainly by sugarbirds and sunbirds (Rebelo et al. 1984 Rebelo 2006), and their seed set tends to be low, with fewer than 15% of florets in an inflorescence produc- ing plump, seed-containing achenes (henceforth seeds Carlson and Holsinger 2010). Seeds are stored aboveground in serotinous infructescences (henceforth seed heads) until they are released following fire, and they usually germinate during the next rainy season (Rebelo and Rourke 1986). Because fire kills adults, pop- ulations re-establish from seed into roughly even-aged stands. SAMPLING DESIGN Starting in February 2008, we sampled a total of 35 populations, consisting of five populations each of P. mundii, P. lacticolor, P. aurea subsp. aurea, and P. venusta, six each of P. subvestita and P. punctata, two of P. aurea subsp. Potbergensis, and a single hybrid population (Fig. 1 Table S1). Sites were chosen to cover most of the range of each species, using the extensive database of Proteaceae localities compiled by the Protea Atlas Project (Rebelo 2001, 2006). Within each population, we sampled 11���40 adult plants (mean n = 21 plants total N = 688). Individuals were selected approximately 10 m apart along transects through or near the pop- ulation center. On each plant, we measured the total height, the number of seed heads, and the number of branching events on the tallest stem, which serves as a proxy for plant age (��2 years T. Rebelo, pers. comm.). We estimated the mean annual growth increment by dividing plant height by our proxy for age. We also collected five to eight intact seed heads (���2 years old), and two fully expanded, mature leaves from the plant���s most recent growth. On one of the fresh leaves, we measured leaf length, width, and area using a LiCor 3100 leaf area meter (Lincoln, NE), dried the leaf for two weeks at 60���C, and then weighed it. We used these data to calculate leaf length:width ratio (LWR) and SLA (leaf area divided by leaf mass in cm2/g). We estimated stomatal den- sity (number of stomata per mm2) on the other fresh leaf using a light microscope and cellophane tape leaf peels from the abax- ial leaf surface (following Dunlap and Stettler 2001). Finally, we measured the wood density on a subset of sampled white protea plants (n = 504 including all populations except LW and JK), using ���1 cm3 sections of 3-year-old sapwood that were air-dried, peeled, and rehydrated under a vacuum. The mass of water dis- placed by the sample was converted to wet volume, and wood density was calculated as the mass of the dried sample (at 75���C for ���48 h) divided by its wet volume (following Martinez-Cabrera et al. 2009). Seed heads were dried at low humidity until opening, after which we removed and counted the number of plump, intact seeds per seed head. Many seed heads were infested with seed-eating 1 1 0 EVOLUTION JANUARY 2011