Portal de Periódicos da CAPES

Life on the edge: rare and restricted episodes of a pan‐tropical mutualism adapting to drier climates

2011; Wiley; Volume: 191; Issue: 1 Linguagem: Inglês

10.1111/j.1469-8137.2011.03683.x

1469-8137

Michael McLeish, Danni Guo, Simon van Noort, Guy F. Midgley,

Botany and Plant Ecology Studies

New PhytologistVolume 191, Issue 1 p. 210-222 Full paperFree Access Life on the edge: rare and restricted episodes of a pan-tropical mutualism adapting to drier climates Michael McLeish, Michael McLeish Department of Botany and Zoology, DST-NRF Centre of Excellence for Invasion Biology, University of Stellenbosch, Private Bag X1, Matieland, 7602, South AfricaSearch for more papers by this authorDanni Guo, Danni Guo Climate Change and Bioadaptation Division, South African National Biodiversity Institute, Kirstenbosch Research Center, Private Bag X7 Claremont, 7735, Cape Town, South AfricaSearch for more papers by this authorSimon van Noort, Simon van Noort Natural History Department, Iziko South African Museum, PO Box 61, Cape Town, 8000, South Africa Department of Zoology, University of Cape Town, Private Bag, Rondebosch, 7701, South AfricaSearch for more papers by this authorGuy Midgley, Guy Midgley Climate Change and Bioadaptation Division, South African National Biodiversity Institute, Kirstenbosch Research Center, Private Bag X7 Claremont, 7735, Cape Town, South AfricaSearch for more papers by this author Michael McLeish, Michael McLeish Department of Botany and Zoology, DST-NRF Centre of Excellence for Invasion Biology, University of Stellenbosch, Private Bag X1, Matieland, 7602, South AfricaSearch for more papers by this authorDanni Guo, Danni Guo Climate Change and Bioadaptation Division, South African National Biodiversity Institute, Kirstenbosch Research Center, Private Bag X7 Claremont, 7735, Cape Town, South AfricaSearch for more papers by this authorSimon van Noort, Simon van Noort Natural History Department, Iziko South African Museum, PO Box 61, Cape Town, 8000, South Africa Department of Zoology, University of Cape Town, Private Bag, Rondebosch, 7701, South AfricaSearch for more papers by this authorGuy Midgley, Guy Midgley Climate Change and Bioadaptation Division, South African National Biodiversity Institute, Kirstenbosch Research Center, Private Bag X7 Claremont, 7735, Cape Town, South AfricaSearch for more papers by this author First published: 24 March 2011 https://doi.org/10.1111/j.1469-8137.2011.03683.xCitations: 10 Author for correspondence:Michael McLeishFax: +27 21 808 2405Email: mcleish@sun.ac.za AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinked InRedditWechat Summary • The fig tree–fig wasp obligate pollination mutualism has strong ancestral affinities with tropical communities, but is present in much drier contemporary biomes, especially at higher latitudes at the edge of their range. The extent to which adaptation to environmental variables is evolutionarily conserved and whether environmental differences function in ecological speciation of the mutualism are unknown. • Here we use climate models and phylogenetic reconstructions to test whether the Ficus–fig wasp mutualism has adapted and radiated into drier climates and led to ecological speciation in both plant and insect. • The results showed phylogenetic correspondence between closely related Ficus species with either savanna, forest, or riparian habitat categories, were most strongly explained by both climate and environmental variables. Rare episodes of adaptation to dry apotypic conditions have resulted in substantial radiations into savanna. • Inferences were consistent with predictions of niche conservatism and support the postulate that ecological speciation of the mutualism occurs, but under contrasting and intertwined circumstances among plant-pollinator adaptation and tolerance to the environment. Introduction Co-dependent diversification is more often discussed in terms of either reciprocal selective regimes between species (Ehrlich & Raven, 1964), co-speciation (Fahrenholz, 1913), or as an ecologically driven process (Jermy, 1976) where host diversification generates possibilities for ecological speciation (Schluter, 2000) in associates. Ecologically based selection occurs as a consequence of the interaction between populations and their environment, but understanding how environmental differences function in ecological speciation remains limited (Rundle & Nosil, 2005). This is especially true for synergistic interactions, such as obligate mutualisms, which predict parallel radiations (Kiester et al., 1984). For mutualisms, it is expected that changing environmental conditions will create different circumstances for each of the species in the association (Holt & Keitt, 2000; Toju, 2008). Associations exhibiting co-adaptation that are suspected to have co-evolved have been proposed to have arisen indirectly as a result of distribution patterns (van Noort & Compton, 1996) that accentuate ecological interactions. Conceivably, over time, changes in the distributions of the interacting species may occur by range expansion into a novel environment, or as an outcome of adaptive solutions to the changing conditions and subsequent speciation (Holt, 2003; Parmesan et al., 2005; Goldberg & Lande, 2007; Roy et al., 2009). Fragmentation or co-extinction ensues in those instances in which the changing environment exceeds the physiological tolerance thresholds of the organism (Dunn et al., 2009). Spatial variations in plant species distributions are predominantly explained in terms of climate and energy variables (Wright, 1983; Hawkins et al., 2003; Currie et al., 2004; Costa et al., 2007). Climate is therefore expected to strongly structure the distribution of Ficus (Moraceae) among environmentally distinct biomes in addition to the physiological tolerances (Warren et al., 2010) and dispersal capabilities (Ahmed et al., 2009) of its obligate fig wasp pollinator (Agaonidae). Ficus is a circumtropically distributed genus of c. 800 species including trees, epiphytes, hemi-epiphytes, lithophytes, and shrubs. There are in excess of 108 species distributed in Africa and its adjacent islands (Berg, 1989; Rønsted et al., 2007). About two-thirds of the African species are associated with humid tropical lowland forest, with higher species richness in West and Central African regions (Kissling et al., 2007). South Africa possesses 32 known Ficus species, with many belonging to the southern-most constituent of the East African coastal forests (http://www.figweb.org). At high latitudes at the boundaries of their range, South African species are distributed over distinct hydric to apotypic xeric biomes that vary in environmental stability (Hély et al., 2006). The evolution of climate tolerance and the biological invasion by the obligate mutualism into apotypic conditions present in more arid environments have not been empirically evaluated. One way to do this is to test the hypothesis that groups of Ficus species are associated with distinct habitat types and climates. Long-term field observations have shown that African Ficus species are subdivided into groups that are affiliated with habitats possessing contrasting climates. Unique combinations of biotic or abiotic variables associated with different environments have been used to describe a given niche (Hutchinson, 1957). Conservatism of such an environmental niche (e.g. Wiens & Graham, 2005; Pearman et al., 2008) predicts correspondence of range limits among closely related lineages (Roy et al., 2009). Whether habitat affiliation strongly predicts niche requirements that determine natural groups among Ficus, and whether these are congruent with phylogeny, remain unknown. Previous work (Godsoe et al., 2009) has shown that diversification in the yucca moth (Prodoxus quinquepunctellus)–Joshua tree (Yucca brevifolia)–mutualism is not dependent on habitat specialization. This work was conducted at a spatial scale measured in hundreds of kilometres and at a taxonomic scale between two species only. However, conservatism of ecological associations has been shown to vary between particular habitat types or environments (Vinson, 1976; Jaenike, 1990; Goldberg & Lande, 2006), but appears to be most apparent at the community level (Losos et al., 2003; Ackerly et al., 2006; Ricklefs, 2007; Losos, 2008). The fig wasp obligate pollination mutualism (Janzen, 1979) is characterized by a species-rich group that is widespread throughout the planet and presents a larger geographic and taxonomic scale by which to reveal abiotic mediated processes. Species presence data combined with historical information are useful for testing hypotheses linking environmental variables, species ecology, and clade evolution (Case et al., 2005; Thuiller, 2007). The aim of this study was to identify the most limiting combination of climate and habitat variables to which particular groups of Ficus are restricted and reconcile these with clade evolution. The intensity of the phylogenetic signal among species that share the same environmental niche or habitat category is investigated using ancestral reconstructions and divergence time estimates and related to the emergence of these habitats in Africa. We investigate the evolution of Ficus habitat affiliation by testing predictions (Wiens & Graham, 2005) that sister species have similar climate/habitat characteristics; that derived clades will tend to be distributed over apotypic climates; that the most ancestral character state for a habitat will reflect the pleisiotypic conditions of the tropics; and that species will track their climate/habitat regime where ecological relationships are conserved but where geographical affinities are not. Materials and Methods Species sample Estimates of Ficus species richness (Kissling et al., 2007) closely coincide with that of species richness of woody genera over sub-Saharan Africa (Lovett et al., 2000). Some species are more abundant, and it is for these that habitat categorization is possible, as a result of substantial observations made in Southern Africa over many years. Habitat category affiliation is uncertain for rare species. Only species for which reliable habitat records were available and that had DNA sequence data available online were used for phylogenetic reconstructions. Ficus cordata salicifolia is now recognized as a species, Ficus salicifolia. Ficus thonningii has been synonymized with Ficus burkei as Ficus burkei. We have elected to retain the old nomenclature to be consistent with herbarium records. Habitat category classification Classification of vegetation types is not without its problems (Lawesson, 1994). However, biome sensitivity to climate along coarsely defined forest–savanna vegetation boundaries is most accurately recovered at geological scales (Vincens et al., 2000; Hély et al., 2006). The general physical appearance and diversity of biomes that are dominated by angiosperm communities vary greatly among the regions of the world, but can be categorized using physiognomic criteria such as tree height, leaf size, and taxonomically dominant family presence/absence (Jacobs, 2004). The same criteria can be used to track the evolution of such biomes that existed in the past and are represented in the fossil record. We partitioned the presence data into six habitat categories: (1) riparian in forest, (2) forest, (3) forest and savanna, (4) riparian in savanna, (5) savanna and (6) Succulent Karoo and desert. Forest communities are defined by varying conditions of precipitation periods, diurnal temperature, and altitude, but more generally by the presence of tall trees with overlapping canopies. The composition of forest and savanna is more open, with smaller trees whose crowns touch intermittently and a grass understory, and again varies according to annual rainfall and temperature regimes. Savanna has a reduced woody component that varies along moisture gradients, and the succulent and desert vegetation is characterized by extensive arid and semi-arid rangeland with a grass component and shrubby assemblages (Yeaton & Esler, 1990; Vincens et al., 2000; Hély et al., 2006). Climatic and other landscape feature variables were optimized over each of the six habitat categories. Climatic variable optimization Presence data for 26 Ficus species (Table 1) located in South Africa (82% of known South African species), comprising a total of 2261 locality records, were compiled from our own field trip records obtained between 1998 and 2010, and from the SANBI and PRECIS databases (available at http://posa.sanbi.org/searchspp.php). The species at each locality collected during our field trips were identified using morphological criteria and from the fig wasp fauna which are extremely specific to each Ficus host species. No records were used in the analyses where species identity was not known. Several approaches were used to investigate habitat category, climate, and environmental niche (vegetation and morphology) variables. The performance of model-based methods for predicting species distributions remains poorly known, and instead, extrapolating from results over multiple approaches is recommended (Araújo & New, 2007). Principal component analysis (PCA), a statistical tool for analysing data, and maxent (Phillips et al., 2004, 2006), a program based on maximum entropy modeling of species geographic distributions, were both used to explore the relationships between Ficus species distributions with environmental and climatic variables and corresponding habitat categories. maxent uses a Bayesian approach in which the species probability distribution is statistically estimated by searching the family of probability distributions under the maximum entropy criterion subject to environmental constraints (Phillips et al., 2004, 2006). In this case, the results from the maxent model were used to help to explore Ficus habitat relationships statistically; maxent was thus not used purely as a species distribution mapping tool. Prior assumptions and beliefs and parameters were not used in this case study; the results from the statistical and spatial analyses were used to explore and reveal hidden aspects of the relationship between Ficus species distributions and the habitat categories. PCA was used to explore variable importance according to variance in the data; the maxent analysis estimated the contribution of the variables to species occurrence; and the Jackknife test examined the importance of the variables to species occurrence. Cluster analyses of both PCA and maxent results were used to infer the Euclidian distance relationships among habitat categories. Table 1. Sample number, Ficus species richness, and species names for each habitat category in South Africa Forest Riparian in forest Forest and savanna Riparian in savanna Savanna Succulent Karoo and desert Sample size 181 263 237 163 1292 125 Species richness 3 2 5 3 12 2 Habitat area (km2) 34 900 94 900 50 400 59 600 129 000 88 800 Species F. bizanae F. sur F. bubu F. capreifolia F. abutilifolia F. cordata cordata F. craterostoma F. trichopoda F. burtt-davyi F. sycomorus F. burkei F. ilicina F. polita F. lingua F. verruculosa F. cordata salicifolia F. lutea F. glumosa F. natalensis F. ingens F. petersii F. salicifolia F. sansibarica F. stuhlmannii F. tettensis F. thonningii F. tremula Area estimates are based on a per cell average length of 1.6 km and an area of 2.56 km2 and given rounded to the nearest 100 km2. Data for a range of seasonal variables, and vegetation and biome data (Mucina & Rutherford, 2006; Schulze, 2007) were used to investigate the contributions and relative importance of climatic and niche variables in explaining the presence of Ficus. The climatic and environmental variables comprised spring mean daily maximum temperature, spring mean daily minimum temperature, spring mean rainfall, spring mean solar radiation, spring mean temperature, summer mean daily maximum temperature, summer mean daily minimum temperature, summer mean rainfall, summer mean solar radiation, summer mean temperature, autumn mean daily maximum temperature, autumn mean daily minimum temperature, fall mean rainfall, fall mean solar radiation, fall mean temperature, winter mean daily maximum temperature, winter mean daily minimum temperature, winter mean rainfall, winter mean solar radiation, and winter mean temperature; and also biome, vegetation group, altitude, morphology, South African rainfall concentration and rainfall seasonality. Solar radiation, vegetation group, and morphology were included and excluded from different analyses to assess the relative importance of these variables in explaining species distributions. These variables were chosen in consideration of South Africa's unique environmental conditions (Guo et al., 2009, 2010). The temperature and rainfall variables were needed to calculate the climate niche of the Ficus species belonging to each of the habitat categories. All four seasons of the year were considered separately in climate analyses rather than using the annual mean, as seasonality is known to influence biota in South Africa (Hély et al., 2006). Rainfall concentration indicates whether the rainfall season is concentrated over a shorter or longer period (Schulze, 2007). The variable 'biome' refers to a broader division of the vegetation of South Africa (e.g. forest, desert, Nama-Karoo, Fynbos, and grassland), and 'vegetation group' to a finer division (e.g. Azonal Forests, Mopane Bioregion, and Sand Fynbos). These variables provide a benchmark for determining whether the presence of Ficus species is influenced by the ambient vegetation community. Altitude (height) and morphology (e.g. plains, hills, mountains, lowlands, and pans) are included as a constraint on the modeling of the climatic niche alone and provide an indication of landscape affinities (Guo et al., 2009, 2010). Phyloclimatic modeling Sequence data for up to 767 bp of a ribosomal internal transcribed spacer (ITS) and up to 479 bp of an external transcribed spacer (ETS) were retrieved from GenBank (Jousselin et al., 2003; Rønsted et al., 2007: Table S1). We also amplified ITS and ETS regions for new specimens of Ficus ilicina, Ficus cordata cordata, and a variety of Ficus sycomorus whose sequences have been lodged in GenBank under the accession numbers HM746955 to HM746960. We amplified DNA using the protocol detailed in previous work (Rønsted et al., 2007). Phylogenetic inferences incorporated 34 ingroup Ficus species that comprised representatives of each of the subsections present in Africa for which sequence data were available. No outgroup taxa were used and rooting the tree was achieved by a posteriori sister-clade matching to conform to the more comprehensively sampled inference of Rønsted et al. (2005). Phylogenetic reconstructions were implemented using MrBayes 3.1.1 (Huelsenbeck & Ronquist, 2001) and paup*4.0b10 (Swofford, 2002). The Bayesian analyses partitioned the sequence data into ITS and ETS. A general time-reversible DNA substitution model was used with gamma-distributed rates with a proportion of invariant sites. Posterior probabilities and mean branch lengths were derived from 35 000 trees sampled every 1000 trees from generations 5 to 40 million. The trees were derived from post-burnin generations that had reached apparent stationarity. Convergence was assessed by plotting the post-burnin generation log likelihoods to assess the point in the chain where stable values were reached and with the standard deviation of split frequencies of successive runs. All Bayesian reconstructions were run four times to verify consistency of phylogenetic inferences. Parsimony was used to test the robustness of the partitioned model-based phylogeny and implemented using paup*. The parsimony analyses consisted of 1000 bootstrap replicates using a full heuristic search, keeping best trees only with branch-swapping by stepwise addition using 100 random additional sequences, holding five trees at each step to calculate bootstrap support. Divergence dating and calibration To estimate a relaxed molecular clock chronogram, we used BEAST (Drummond & Rambaut, 2003), a Bayesian Markov chain Monte Carlo (MCMC) analysis, to test evolutionary hypotheses without conditioning on a single tree topology. This approach calculates node height (age) summary distributions that represent 95% upper and lower highest posterior density intervals around the mean node height. The Bayesian consensus tree was used as a reference to define monophyletic taxon subsets, using beauti v.1.4.8 (Rambaut & Drummond, 2007a), of nodes supported with > 95 posterior probabilities. The Yule pure-birth constant speciation rate per lineage was assumed. We specified a gamma + invariant sites DNA substitution model to conform to our Bayesian inference priors, and selected a relaxed clock uncorrelated lognormal model. Analysis was undertaken by sampling every 1000th generation of a 20 × 106 generations MCMC chain with a burnin of 1 × 106 generations. The consensus chronogram was generated using TreeAnnotator v.1.4.8 (Rambaut & Drummond, 2007b). To calibrate the chronogram, we used priors as reported by Rønsted et al. (2005). Ficus section Galoglychia, which is restricted to Africa and estimated to have originated c. 40 Ma, shares a common ancestor, dated at 60 Ma, with sections Sycomorus and Urostigma, which have a more extensive distribution encompassing both the Afrotropical and Indo-Australasian biogeographical regions (Rønsted et al., 2005). We set a 60 Myr normally distributed prior at the split between sections Galoglychia and Sycomorus/Urostigma (the root) to infer a maximum clade credibility chronogram. Alternative hypotheses for the origin of the most recent common ancestor of these sections were also considered: Machado et al. (2001) estimated this split at c. 82 Myr and for the pollinating fig wasps associated with these Ficus sections, Lopez-Vaamonde et al. (2009) dated the divergence at c. 114 Myr. Ancestral habitat reconstructions To explore alternative assumptions regarding character evolution, ancestral habitat states were inferred using parsimony-based reconstructions performed using MacClade v.4.0.6 (Maddison & Maddison, 2000) and Bayesian approaches implemented using BayesTraits (Pagel et al., 2004; http://www.evolution.rdg.ac.uk). A parsimonious model of character evolution was used to infer ancestral states that minimize parameterization. The parsimony-based approach does allow weighted analyses that assume different state transformation types, but is limited in the manner in which discrete vs continuous characters may be treated. Our habitat categories include 'intermediate' states that include elements of more than one habitat category, and this is in line with recoding discrete states to mimic continuous characters for analysis using MacClade. We chose to implement the 'unordered' transformation prior that assumes transformations between any two states can occur in a single step (Maddison & Maddison, 2000), as no information was available to suggest that a more complex model was appropriate here. The reconstruction of ancestral character states associated with Ficus is problematic because of phylogenetic uncertainty and the potential for frequent hybridization and substantial levels of ancestral polymorphism shared across recently diverged species (Machado et al., 2005; Renoult et al., 2009). Unlike MacClade, Bayesian approaches offer an opportunity to account for phylogenetic uncertainty and consider branch length differences among lineages that provide information about the expected amount of character change. The logic of the Bayesian approach for the model of trait evolution applies to both discrete and continuously varying traits (Pagel et al., 2004). We tested the hypothesis that each of the six categories was phylogenetically structured or conserved. Although Ficus species sampling was nonexhaustive (82% of known species) and slightly underrepresented the potential total number of South African species, we expected general patterns of ancestral character evolution to be informative, especially for the Afrotropical-restricted section Galoglychia. The BayesMultiState function in BayesTraits was used to reconstruct and test how habitat categories evolve on our Bayesian consensus phylogeny. BayesMultiState permits character states to vary their rate of evolution within and between branches and allows integration over different models among the transformation rates using the reverse jump hyperprior MCMC function. The marginal likelihoods associated with different states at each node are estimated as the model traverses the trees. The approach employs an acceptance rate that indicates the appropriateness of the model parameters. Several sets of combinations of priors were tested before the priors were chosen. Acceptance rates of the chosen prior values were all within the recommended 20 to 40% bounds. A rate deviation prior of 40 was used with the reverse jump hyperprior (RJHP) function with a gamma prior of 0, 10, 0, 10 (minimum and maximum of priors for both mean and variance parameters). Although our Bayesian consensus is well resolved and well supported over deeper divergences, uncertainty in node support was accommodated by evaluating ancestral character reconstructions over a sample of post-burnin phylograms (n = 500) from the MrBayes Markov chain. We used 50 × 106 iterations sampling every 1000th generation with a burnin of 40 × 106 iterations to estimate character state probabilities for selected nodes. Results Climatic niche modeling Climatic variables were optimized over six habitat categories to compare the intensity of phylogenetic habitat conservatism among Ficus species. Variable contributions were modeled using a Bayesian approach that calculated the per cent contribution of each variable and a jackknife estimation of the importance of the variables in explaining the habitat categories (Table 2). Only the first four variables contributing to any given habitat category were reported, after which a sharp drop-off in the importance of variables generally occurred. In all cases, each habitat category was explained by contrasting sets of climatic (Supporting Information Fig. S1–S3) and environmental variables (Table 2). Initial model runs indicated that solar radiation was generally the most important variable that influences the relationship between Ficus species distributions and habitat categories (Table 2). Excluding solar radiation variables in later model runs revealed that different habitats had different seasonal rainfall and temperature requirements. Solar variables were important in all habitat categories except desert. Without solar radiation variables, 'rain' became most important in explaining the presence of Ficus in wet/dry habitat extremes, but temperature variables tended to be more important to the 'intermediate' habitats, especially 'forest and savanna' and 'riparian in savanna'. Ficus species presence responded to different seasonality variables depending on which habitat category they fell into. Table 2. Percent contributions of variables derived from maxent that explain habitat categories with and without the inclusion of the variable solar radiation (upper table) and jackknife tests showing variable importance with and without the inclusion of the variable solar radiation (lower table) Forest Riparian in forest Forest and savanna Riparian in savanna Savanna Succulent Karoo and desert maxent percent contribution (%) With solar and vegetation group Spr srad 33.2 Spr srad 32.8 Spr srad 54.3 Spr srad 34.5 Sum srad 26.4 Veg group 22.8 Veg group 17.7 Sum srad 19.0 Aut min temp 13.9 Spr min temp 18.2 Spr min temp 24.8 Sum rain 22.1 Aut rain 15.3 Aut min temp 10.5 Veg group 5.3 Win min temp 14.5 Sum rain 19.0 Spr rain 12.6 Sum rain 12.4 Spr rain 7.5 Aut rain 4.8 Win srad 8.4 Morphology 6.9 Spr min temp 8.2 No solar and vegetation group Aut rain 27.9 Win min temp 28.5 Win min temp 28.0 Win min temp 39.5 Spr min temp 32.0 Sum rain 31.5 Spr rain 18.6 Spr rain 27.3 Spr rain 24.5 Spr min temp 19.6 Sum rain 23.3 Morphology 14.7 Sum rain 18.5 Sum max temp 12.4 Aut min temp 12.8 Sum rain 14.2 Spr rain 15.8 Spr rain 12.1 Win min temp 17.1 Aut min temp 9.9 Win rain 11.0 Spr max temp 9.2 Morphology 8.9 Win rain 10.6 Jackknife variable importance With solar and vegetation group Veg group Spr srad Spr srad Spr srad Veg group Veg group Spr rain Sum srad Veg group Veg group Sum srad Sum rain Spr srad Veg group Win min temp Spr min temp Spr min temp Spr rain Aut rain Spr rain Aut min temp Sum srad Sum rain Aut rain No solar and vegetation group Spr rain Spr rain Win min temp Spr min temp Sum rain Sum rain Aut rain Win min temp Aut min temp Aut min temp Spr min temp Aut min temp Sum rain Aut rain Spr rain Win min temp Spr rain Aut rain Win min temp Aut min temp Aut rain Win max temp Win max temp Spr rain Color coding indicates variables that are related to: temperature (orange); rainf