Journal list menu
Flowering time evolution is independent of serpentine tolerance in the California flora
Abstract
Comparative phylogenetic tests have been only recently applied to the many hypotheses about the role ultramafic (serpentine) soils play in the evolution and ecology of regional floras. An association between serpentine tolerance and early flowering has been observed at microevolutionary scales but not yet tested broadly across angiosperms. I used both hypothesis testing (phylogenetically independent contrasts) and model-selection approaches to compare published data of serpentine tolerance with flowering time using phylogenetic trees representing 24 clades of plants representing 27 genera and 17 families. A total of 126 independent contrasts revealed no significant difference in flowering times across all clades or within any one clade except for the Thelypodieae, in which flowering time of serpentine-tolerant lineages was later. Similarly, an Ornstein-Uhlenbeck model with one optimal flowering time was preferred over a model with separate optima depending on serpentine tolerance in nearly all genera. A phylogenetically un-corrected ANOVA found that serpentine-tolerant species have an earlier distribution of flowering times compared to both endemic and non-tolerant species, and median flowering times of non-tolerant species show a bimodal distribution. Therefore, I conclude that the long-term evolutionary responses of flowering time in a lineage that becomes tolerant of serpentine are variable across genera, with no significant overall bias toward earlier or later flowering. However, given that close relatives of serpentine-tolerant lineages tend to also flower relatively early, early flowering time may serve as an exaptation for serpentine tolerance. In combination with previously published ecotype studies, these results illustrate the eco-evolutionary scale dependence of flowering time.
Introduction
Varied and contrasting edaphic regimes have long been understood as having a major influence on biodiversity and trait evolution at both large and small scales, particularly in plants. Continental-scale phylogenetic turnover and community structuring can be partly explained by soil gradients (Fine and Kembel 2011). At the population level, differences in soil type have been shown to be an ecological filter and a strong isolating mechanism of plant lineages even at short distances (MacNair and Christie 1983, Gardner and Macnair 2000, Moyle et al. 2012). Consequently, edaphic specialization has long been cited as a major factor governing plant distribution, reproductive isolation, and lineage splitting within genera (Wallace 1895, Rajakaruna 2003, Baldwin 2005, von Humboldt and Bonpland 1805, Yost et al. 2012).
One of the starkest edaphic regimes is derived from ultramafic rock formations. These soils, broadly referred to as serpentine soils, are generally characterized by low calcium-to-magnesium ratios, low concentrations of plant essential macronutrients (N, P, K), and high concentrations of heavy metals (e.g., Mg, Fe, Ni). In addition to the harsh chemistry, serpentine soils show elevated soil temperatures and erosion compared to nearby non-serpentine areas (Walker 1954, Kruckeberg 2002). Serpentine soils are also thought to be more xeric (Walker 1954, Kruckeberg 2002, Harrison 2013), though this has been disputed by Raven and Axelrod (1978) and others (Eskelinen and Harrison 2015). In any case, the challenging serpentine-associated conditions for plants have led to distinctive floras on six continents (Kruckeberg 1985, Rajakaruna et al. 2009). In the tropical floras of Cuba and New Caledonia, 27% and 50% of endemic taxa are also endemic to serpentine soils. Among temperate regions, California has the most species-rich serpentine flora. Approximately 10% of California's endemic plant taxa are endemic to serpentine soils although these habitats represent less than 1.5% of the state's surface area (Kruckeberg 1985, Safford et al. 2005). These factors, in combination with a robust history of botanical study, have contributed to the region's prominence as a model system for studying the evolutionary ecology of serpentine floras (Anacker 2011).
Given the intense selective regime imposed by serpentine soils, botanists and ecologists have long sought to identify morphological, physiological, and phenological adaptations that permit colonization of serpentine soils (e.g., Wulff 1943, Kruckeberg 1954, Kruckeberg 1985; Wright et al. 2006, O'Dell and Rajakaruna 2011), as well as their genetic basis (e.g., Gailing et al. 2004). Studies in California and elsewhere have found that serpentine plants have increased tolerance to drought and heavy metal accumulation, reduced leaf size, shorter stature, elevated root-to-shoot ratios, greater tolerance and lower absorption of Mg, lower Ca levels but higher absorption, and earlier-flowering phenology (Kruckeberg 1954, Schmitt 1980, Lee et al. 1997, Tyndall and Hull 1999, reviewed by Brady et al. 2005). However, very few of these hypotheses have been tested in a comparative phylogenetic framework. Independently comparing species on and off of serpentine soils does not account for their shared evolutionary history (Felsenstein 1985). Recent efforts have begun to address these deficiencies, especially in regard to macroevolution and diversification in California (e.g., Anacker et al. 2011, Anacker and Harrison 2012). However, more research is required to better understand trait evolution in these systems.
The focus of the present study was to test the hypothesis that the flowering time of plant lineages that invade serpentine soils is consistently earlier compared to non-serpentine plants. This has been a long-standing claim based mostly on observations of ecotypes that flower earlier on serpentine soils than do their non-serpentine conspecifics (Schmitt 1980, Brady et al. 2005). Serpentine soils are generally rocky and low in clay and organic matter, and therefore may desiccate more rapidly than non-serpentine soils. Accordingly, plants on California serpentines may preempt the earlier onset of summer drought conditions by flowering earlier, with a stronger shift expected among serpentine-endemic taxa, which persist only on serpentine, than serpentine-tolerant taxa, which are found in both serpentine and non-serpentine habitats. Here, I tested this flowering-time hypothesis using phylogenetically independent contrasts (PICs) and comparisons between three models of flowering time evolution: (1) a Brownian motion (BM) model representing only genetic drift, (2) an Ornstein-Uhlenbeck (OU) model, which generalizes the BM model to include the effects of stabilizing selection on a given lineage, and (3) and a multiple-optimum OU model, sometimes called a Hansen model, in which different trait optima (i.e., selective regimes) are permitted on different branches of the phylogeny. In the context of this study, the third model allowed for different flowering time optima between lineages growing on and off of serpentine soils. Finally, I use non-phylogenetic comparisons among flowering times of California native taxa to see whether the ecotypic-scale observations of earlier flowering time can be generalized to broad patterns across a regional flora, irrespective of evolutionary history.
Materials and Methods
Clade selection
Clade selection was based on three criteria: (1) At least one taxon of each clade must be endemic to serpentine soils in California, (2) a published molecular phylogeny including both plastid and nuclear sequence data and extensive taxonomic sampling must already exist, and (3) clades must represent independent origins of serpentine tolerance, such that no two sampled clades are sister to, or nested within, each other. Twenty-four clades were found to meet these criteria, including four of the top-five most important and top-five most diverse serpentine genera in California (Safford et al. 2005). The fifth clade, Hesperolinon, has unclear taxonomic boundaries and phylogenetic relationships as a result of a recent, rapid radiation, making reliable ancestral-state reconstructions impractical (Schneider et al. 2016). Twenty-two of these clades were reported by Anacker et al. (2011) as part of a large analysis of diversification onto serpentine soils. Phylogenies for the remaining clades were published subsequently: the Eriogonoideae (Kempton 2012) and Streptanthoid complex (=Thelypodieae; Cacho et al. 2014). Tree topologies and branch lengths used from the studies of Anacker et al. (2011) and Kempton (2012) reflect a majority-rule consensus tree, whereas the Thelypodieae phylogeny of Cacho et al. (2014) is the maximum clade credibility tree from a BEAST analysis (see the original studies for full details).
I made several modifications to these phylogenies: Conspecific (or convarietal) terminals resolved as a clade were pruned to a single exemplar per taxonomic species, subspecies, or variety to match the resolution of trait data. Phylogenetic trees were made ultrametric using Sanderson's (2002) semi-parametric penalized likelihood algorithm implemented in the “ape” package version 3.0-7 (Paradis et al. 2004) of the statistical program R version 2.15.1 (R Core Team 2012). The smoothing parameter, λ, was selected for each clade using a cross-validation algorithm. Outgroup taxa and ingroup tips for which no flowering time data were available (or in the case of a few Eriogonoideae, flower year-round) were pruned following the rate smoothing but prior to comparative analyses.
Trait data
Serpentine tolerance and serpentine endemism were treated as binary characters. Serpentine endemism was defined as taxa being naturally restricted to ultramafic soils, whereas serpentine tolerance was defined as any persistence on ultramafic soils regardless of where else the plants may be found and therefore also includes all serpentine-endemic taxa. Character determinations followed a simplification of the ternary classification of Anacker et al. (2012) and Cacho et al. (2014), who relied on a database of serpentine affinity based on peer-reviewed and gray literature, expert assessment and observations, and herbaria records (Safford et al. 2005). Taxa not included in previous phylogenetic analyses were assessed for serpentine tolerance following Safford et al. (2005) and taxon descriptions in the revised Jepson Manual (TJM2; Baldwin et al. 2012).
Flowering time was treated as a continuous character by determining the midpoint of the flowering range as described by experts in published floras, similar to Bolmgren et al. (2003). Flowering time onset or conclusion are two other common ways to measure flowering phenology. However, flowering time midpoint is probably a more stable measure of phenology than either onset or end of flowering time, which can be affected by population size (CaraDonna et al. 2014), and better reflective of the central tendency in the absence of detailed surveys of field populations or museum specimens. Flowering times of species native or naturalized to California, representing over 91% of study taxa, were taken from TJM2. Most of the remaining flowering time data came from the Flora of North America North of Mexico (Flora of North America Editorial Committee 1993+), with <1% from other sources such as the Flora of Baja California (Wiggins 1980), Flora of China (Zhengyi et al. 1994+, Brach and Song 2006), SEINnet (www.swbiodiversity.org), or the Lady Bird Johnson Wildflower Center (www.wildflower.org). Taxon nomenclature was standardized throughout. The resolution of flowering time ranges was usually month-to-month, resulting in semi-monthly (occasionally weekly) resolution of flowering time midpoint. All taxa included in this study were reported to have a unimodal distribution of flowering time.
Acknowledging the limitations inherent in this approach, including the coarseness of time record, and inability to control for latitude or elevation at the population level, flowering ranges from regional floras were used instead of mining herbarium record data for four reasons. First, substantial spatial, temporal, and taxonomic collection biases may exist in collections data, particularly across the geographic scales in this study. Using published floras allows for standardization across taxa, and expert understanding, which may offset some of these biases. Second, in some lineages, floral morphology is not taxonomically diagnostic. For example, many Apiaceae are distinguished by fruit characters, with this phenophase likely overrepresented in herbaria collections. If not filtered before analysis, this would bias records toward having a later flowering time. Third, published floras generally report longer flowering time durations than direct analysis of herbarium data supports, perhaps due to the coarseness of the data or temporal sampling bias in herbaria collections (Bolmgren et al. 2003, Bolmgren and Lönnberg 2005). Finally, the taxonomic and geographic scope of this study (1205 taxa across North America) is much greater than those for other studies that have effectively used herbarium records to estimate phenological trends, generally from one or several well-curated herbaria (e.g., Primack et al. 2004, Lavoie and Lechance 2006, Calinger et al. 2013). Though most herbarium records in California have been digitized, records from neighboring states range from very low to moderate (Taylor 2014). However, rapid progress is being made in digitizing collection data, which hopefully will overcome this limitation in the future to the extent that it is not driven by collection bias.
Two species of Ericameria included in the analysis are spring flowering, while all other sampled taxa in the genus are fall flowering; only the fall-flowering species were included. Whether those two species were scored as earlier or later did not qualitatively affect the results of independent contrasts; however, these taxa were pruned from the phylogenies and omitted from the model-testing analyses. Several species in the Eriogonoideae that flower year-round were also removed from analysis. Complete trait data are available in Data S1, with detailed descriptions in Metadata S1.
Phylogenetic tests
The effect of serpentine tolerance on flowering time was tested using both PICs and likelihood comparisons between models that account for shifts in flowering time and those that do not. Blomberg's K statistic was also calculated for each genus to assess phylogenetic signal in flowering time at the within-clade level using the R package phytools (v. 0.5-64; Revell 2012). Significance testing was performed using a nonparametric tip randomization (1000 replicates).
Independent contrasts were performed using the “aot” module in the program Phylocom 4.2 (Webb et al. 2008). This program allows for comparisons between a discrete independent, predictor variable (serpentine tolerance) and a continuous dependent variable (flowering time). Significance testing was done using a Wilcoxon signed-rank test on contrasts from each clade individually, as well as all contrasts together because of low sample sizes in the within-clade analysis, and because when pooled the data are not normally distributed (P < 0.001, Shapiro–Wilk normality test). To test whether lineages with a stronger relationship to serpentine were more likely to have different flowering time, analyses were repeated with taxa scored in two different ways: (1) serpentine endemics vs. non-endemics, and (2) serpentine-tolerant taxa (which includes all endemics) vs. non-serpentine taxa. In both cases, trait values for serpentine tolerance were coded such that a positive contrast represents later flowering of the serpentine-tolerant lineage, and a negative contrast represents earlier flowering. Although PICs assume trait evolution follows a BM model, this method has been found to be quite robust to violations of the BM assumption and branch length (Diaz-Uriarte and Garland 1996, Ackerly 2000) and my results from these analyses are consistent with the model-based approach also presented below.
Evolutionary model testing was conducted in R using the OUCH package version 2.8-2 (Butler and King 2004). Log-likelihoods were calculated for three different models of flowering time evolution: a one-parameter BM model, in which trait evolution follows a random walk; a two-parameter OU model with a single evolutionary optimum for flowering time; and a three-parameter OU model that estimates separate optimal flowering times for lineages on and off of serpentine. Because I was only interested in the shift of mean flowering time and not changes in other parameters caused by edaphic shifts (i.e., for attraction, α, and drift, σ2), I did not use the more general model of Beaulieu et al. (2012). Ancestral character states of serpentine tolerance were reconstructed using a maximum-likelihood analysis implemented in the R package “picante” (Kembel et al. 2010) using either a symmetric or asymmetric rate matrix depending on the results of a likelihood ratio test. A preliminary study found that the ancestral-state reconstructions were relatively robust to the smoothing parameter (data not shown). The genus Orthocarpus was omitted from this analysis because it included only one serpentine-tolerant taxon.
Non-phylogenetic comparisons
In order to test the generalizability of the observation that serpentine floras flower earlier than non-serpentine floras, regardless of evolutionary history, non-phylogenetically corrected comparisons of serpentine-endemic, serpentine-tolerant, and non-tolerant species were made using a one-way ANOVA. Bonferroni-corrected Wilcoxon signed-rank tests were used to determine significant differences in average flowering time among soil regimes. All 24 clades used in the phylogenetic analyses were used, but with complete sampling of all species, subspecies, and varieties native to California (i.e., minimally ranked taxa). Taxa not native to California were excluded so that a single source (TJM2) could be used for all flowering time data and to geographically constrain the area of comparison. Two additional genera that show substantial diversity on serpentine but lack suitable phylogenetic data were also included (Lomatium [Apiaceae] and Packera [Asteraceae]), resulting in a total of 1088 minimally ranked taxa, or 20% of California's native angiosperm flora.
Results
Independent contrasts
A total of 126 independent contrasts of flowering time from 24 clades of angiosperms containing 896 operational taxonomic units were identified. Lineages with serpentine tolerance were found to have flowering times slightly later than their non-serpentine sister lineages (+0.15 ± 0.10 months later, mean ± standard error; Wilcoxon signed-rank test P = 0.04; Fig. 1A). However, this difference was driven mostly by a strong shift to later flowering time in serpentine-tolerant lineages within the Thelypodieae (0.9 ± 0.2 months; P = 0.002), as flowering time shifts in serpentine-tolerant lineages from all remaining clades were not significantly different (+0.1 ± 0.1 months; P = 0.16). At 42 nodes, the serpentine-tolerant lineages flowered earlier than their non-serpentine sister; at 67 nodes, the serpentine-tolerant lineages flowered later; and at 17 nodes, there was no difference. Considered individually, no clade besides Thelypodieae showed a significant difference in flowering time (Table 1), although sample sizes (transitions between serpentine and non-serpentine states) were generally low. Two genera (Layia and Orthocarpus) had only one contrast apiece, and therefore, significance testing was not possible.
| Clade | No. OTUs | Blomberg's K (Flowering time) | Serpentine tolerant vs. non-serpentine | Serpentine endemic vs. not | ||||
|---|---|---|---|---|---|---|---|---|
| K | P | No. contrasts | Flowering time difference | P | No. contrasts | Flowering time difference | ||
| Allium | 52 | 0.39 | 0.001 | 9 | Later | ns | 5 | Later |
| Aquilegia | 19 | 0.95 | ns | 2 | Later | ns | 1 | Later |
| Arctostaphylos | 46 | 0.99 | ns | 4 | Later | ns | 3 | None |
| Balsamorhiza | 18 | 0.54 | ns | 2 | Later | ns | 1 | Later |
| Calochortus | 42 | 0.20 | ns | 8 | Later | ns | 5 | Later |
| Calycadenia | 16 | 0.72 | 0.004 | 3 | Earlier | ns | 2 | Later |
| Ceanothus | 52 | 0.64 | 0.001 | 9 | Later | ns | 3 | Earlier |
| Cirsium | 46 | 0.34 | 0.086 | 6 | Later | ns | 2 | Later |
| Collinsia | 17 | 0.52 | 0.074 | 6 | Later | ns | 1 | Later |
| Ericameria a | 22 | 0.54 | ns | 4 | Earlier | ns | 1 | Earlier |
| 22 | 0.72 | ns | 4 | Later | ns | 1 | Earlier | |
| Eriogonoideae | 129 | 0.20 | 0.001 | 9 | Earlier | ns | 3 | Later |
| Erythronium | 20 | 0.47 | 0.076 | 3 | Earlier | ns | 1 | None |
| Iris | 19 | 0.67 | 0.009 | 4 | Later | ns | 2 | Earlier |
| Layia | 10 | 0.46 | ns | 1 | Later | ns | 1 | Later |
| Lessingia | 24 | 0.56 | 0.057 | 2 | Later | ns | 2 | Later |
| Mimulus s.l. | 71 | 0.16 | ns | 13 | Later | ns | 1 | None |
| Navarretia | 35 | 0.33 | ns | 4 | Earlier | ns | 3 | Earlier |
| Orthocarpus | 5 | 0.73 | ns | 1 | Earlier | ns | 1 | Earlier |
| Perideridea | 16 | 0.27 | ns | 4 | Later | ns | 2 | Later |
| Sanicula | 14 | 0.25 | ns | 4 | Earlier | ns | 1 | Later |
| Sidalcea | 40 | 0.51 | 0.053 | 6 | Earlier | ns | 1 | Earlier |
| Thelypodieae | 51 | 0.72 | 0.019 | 8 | Later | 0.002 | 4 | Later |
| Trichostema | 11 | 1.22 | 0.028 | 3 | Later | ns | 1 | Later |
| Trifolium | 51 | 0.42 | 0.001 | 12 | Later | ns | 1 | Later |
Notes
- OTU, operational taxonomic units. For each of 24 clades, the following is shown: Blomberg's K statistic for flowering time and P-value (ns = P ≥ 0.1), the number of PICs, direction of average difference in flowering time of serpentine-tolerant lineages compared to non-serpentine sister lineages, and associated P-values from the Wilcoxon signed-rank test. Phylogenetically independent contrasts were calculated using the “aot” module of Phylocom 4.2. Significance testing was not performed on the serpentine-endemic dataset due to a dearth of transitions.
- a Two taxa in Ericameria have spring flowering times, while the rest flower in the fall. Analyses were repeated by coding this as a shift to earlier flowering time (top values) or a later flowering time (bottom values).
Considering only serpentine endemics, I found a similar stasis in flowering time (Fig. 1B). Only 48 contrasts were recovered: 13 in which the serpentine-endemic lineage is earlier flowering, 27 in which it is later flowering, and eight with no difference. Taken together, serpentine endemics have a non-significantly later flowering time (0.27 ± 0.13 months later; P = 0.067). For most clades, within-group analyses were not possible due to a dearth of transitions to serpentine endemism (<2). One exception was in the Thelypodieae, in which serpentine-endemic lineages flower nearly a month later than their non-endemic sister lineages (0.86 ± 0.25 months; P = 0.04).
Model comparisons
Of the three models of flowering time evolution tested, the preferred model for 18 of 23 clades was the single-optimum OU model, based on Akaike Information Criterion (AIC) scores (Fig. 2, Table 2). Using the Schwartz Information Criterion, which gives a higher penalty for overparameterization, the single-optimum OU model was preferred over the Hansen model in all but three clades. Among those for which the two-optimum Hansen model was preferred over the single-optimum OU model, the direction in flowering time shift was variable. Only in Cirsium and Thelypodieae were the ΔAIC scores greater than 3, but the trait shifts were in opposite directions: Serpentine-tolerant lineages show an earlier trait optimum in Cirsium but later in Thelypodieae (Fig. 2). Likewise, when only the two-optimum model is considered, the estimated parameters support an earlier flowering time optimum among serpentine lineages in nine of 23 genera and a shift to later flowering time among serpentine lineages in the remaining 14. The BM model was preferred for Trichostema, Layia, and Arctostaphylos, though this may be an artifact of limited phylogenetic resolution or small phylogenetic trees. Estimated parameters for each clade are available as supporting information with the online version of this article (Appendix S2).
| Clade | One-Optimum OU | Two-optima OU (Hansen) | BM | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| θ | ln(L) | AIC | SIC | θnserp. | θserp | ln(L) | AIC | SIC | ln(L) | AIC | SIC | |
| Allium | 6.09 | −70.76 | 147.51 | 153.37 | 6.34 | 5.87 | −70.51 | 149.01 | 156.82 | −73.18 | 150.36 | 154.26 |
| Aquilegia | 7.25 | −20.88 | 47.77 | 50.60 | 7.26 | 7.24 | −20.88 | 49.77 | 53.54 | −21.10 | 46.19 | 48.08 |
| Arctostaphylos | 3.08 | −60.52 | 127.05 | 132.53 | 2.91 | 3.44 | −59.09 | 126.18 | 133.49 | −59.90 | 123.80 | 127.46 |
| Balsamorhiza | 6.01 | −13.42 | 32.85 | 35.52 | 6.10 | 5.80 | −12.79 | 33.59 | 37.15 | −16.55 | 37.11 | 38.89 |
| Calochortus | 6.23 | −51.93 | 109.85 | 115.06 | 6.33 | 6.18 | −51.79 | 111.58 | 118.53 | −63.69 | 131.38 | 134.85 |
| Calycadenia | 7.63 | −8.97 | 23.95 | 26.26 | 7.68 | 7.54 | −8.93 | 25.85 | 28.94 | −10.68 | 25.35 | 26.90 |
| Ceanothus | 4.67 | −67.90 | 141.79 | 147.65 | 4.66 | 4.69 | −67.90 | 143.79 | 151.60 | −69.10 | 142.20 | 146.11 |
| Cirsium | 7.61 | −42.75 | 91.50 | 95.89 | 7.86 | 6.83 | −39.77 | 87.54 | 93.40 | −47.48 | 98.95 | 101.89 |
| Collinsia | 5.78 | −20.09 | 46.17 | 48.67 | 5.72 | 5.86 | −20.05 | 48.09 | 51.43 | −22.57 | 49.14 | 50.80 |
| Ericameria | 9.53 | −24.39 | 54.78 | 57.77 | 9.53 | 9.61 | −24.39 | 56.78 | 60.76 | −26.97 | 57.94 | 59.93 |
| Eriogonoideae | 5.22 | −214.94 | 435.88 | 444.46 | 7.26 | 7.35 | −214.94 | 437.87 | 449.31 | −231.22 | 466.43 | 472.15 |
| Erythronium | 5.33 | −28.28 | 62.55 | 65.54 | 5.42 | 5.22 | −28.23 | 64.46 | 68.44 | −30.49 | 64.98 | 66.97 |
| Iris | 5.62 | −16.38 | 38.75 | 41.59 | 5.60 | 5.64 | −16.37 | 40.74 | 44.52 | −18.70 | 41.40 | 43.29 |
| Layia | 5.66 | −20.98 | 47.97 | 51.10 | 5.65 | 7.60 | −20.94 | 49.88 | 54.06 | −19.41 | 42.82 | 44.91 |
| Lessingia | 8.78 | −29.87 | 65.73 | 69.27 | 8.61 | 9.37 | −29.25 | 66.51 | 71.22 | −31.82 | 67.64 | 70.00 |
| Mimulus s.l. | 6.00 | −106.04 | 218.07 | 224.86 | 6.00 | 6.02 | 106.03 | 220.07 | 229.12 | −128.97 | 261.94 | 266.46 |
| Navarretia | 6.31 | −39.57 | 85.15 | 89.81 | 6.24 | 6.55 | −39.07 | 86.15 | 92.37 | −49.07 | 102.15 | 105.26 |
| Perideridea | 7.35 | −20.27 | 46.53 | 48.85 | 6.46 | 7.61 | −19.21 | 46.42 | 49.51 | −23.09 | 50.17 | 51.72 |
| Sanicula | 4.81 | −14.78 | 35.57 | 37.48 | 5.10 | 4.67 | −14.16 | 36.33 | 38.88 | −20.95 | 45.89 | 47.17 |
| Sidalcea | 6.39 | −56.93 | 119.86 | 124.93 | 6.36 | 6.49 | −56.89 | 121.78 | 128.54 | −60.80 | 125.60 | 128.98 |
| Thelypodieae | 5.52 | −78.42 | 162.84 | 168.64 | 5.34 | 6.73 | −75.46 | 158.92 | 166.65 | −81.51 | 167.01 | 170.88 |
| Trichostema | 8.12 | −11.83 | 29.66 | 30.85 | 8.03 | 9.69 | −11.65 | 31.29 | 32.88 | −10.88 | 25.77 | 26.56 |
| Trifolium | 6.03 | −72.98 | 151.95 | 157.75 | 6.32 | 5.06 | −70.49 | 148.99 | 156.71 | −76.49 | 156.98 | 160.84 |
- θ, flowering time optimum, reported in months of the calendar year (1.0 = January 1); ln(L), log-likelihood; AIC, Akaike Information Criterion; SIC, Schwartz Information Criterion. Two spring-flowering species of Ericameria were omitted from the analysis because all other sampled taxa in the genus are fall flowering. The best-supported OU model for each support criterion for each genus is printed in boldface. Underlined values indicate that the BM model was preferred over both OU models. For additional free parameters, see Appendix S2. Analyses were performed using the OUCH package (v. 2.8-2) of the statistical program R (v. 2.15.1).
Phylogenetic signal in flowering time
Phylogenetic signal in flowering time among genera, as measured by Blomberg's K statistic, was variable, but consistently less than 1, and significantly so for eight of the 25 genera (Table 1). Blomberg's K was not correlated with the number of tips (R2 = 0.46; F = 2.18, P = 0.15).
Non-phylogenetically corrected comparisons
Specialization to serpentine soils was found to affect flowering time (one-way ANOVA, F = 3.28, df = 2, P = 0.04). Taxa that are serpentine tolerant but not endemic flower earlier than both endemic taxa (Bonferroni-corrected Wilcoxon signed-rank test P < 0.02) and non-tolerant taxa (P < 0.01; Fig. 3), whereas serpentine endemics flower at the same time as their congeners found off of serpentine. For clade-specific results, see Appendix S1.
Discussion
Natural selection can lead to directional evolution (Franks et al. 2007) or stasis (Evans et al. 1989) in flowering time. My results are consistent with selection acting upon flowering time, as evolutionary models that accounted for selection (OU models) were generally supported over non-adaptive BM models of evolution (Table 2). However, I found little to no evidence of selection causing a shift in flowering time of lineages on serpentine soils at the species level and above (Tables 1, 2, Figs. 1, 2). Depending on the clade, serpentine habitats can select for either earlier or later flowering in serpentine-tolerant lineages, but most commonly neither (Table 2, Fig. 2). Although lineages that are “serpentine tolerant” (i.e., with populations on and off of serpentine soils) flower significantly earlier than both serpentine endemics and species excluded from serpentine (Fig. 3), these differences disappear when comparisons are controlled for shared evolutionary history.
While these data do not support the hypothesis that movement of a lineage onto or off of serpentine results in a unidirectional shift toward earlier or later flowering time in angiosperms, they should not be interpreted as evidence against the effect of edaphic factors on flowering time, just that patterns across scales and across clades are complex, with many processes acting independently, as evidenced by several genera in which a two optimum model is strongly favored (Fig. 2) but with opposing shifts. Flowering time is a relatively labile trait, with low phylogenetic signal. Blomberg's K statistic is lower than 1 in all 24 clades except for a single case (Trichostema, Table 1), indicating elevated trait evolution relative to what would be expected under a BM model. This is consistent with other regional-level studies of flowering plants (Du et al. 2015). Consequently, a real shift at very fine-scale phytogeographic levels, such as those observed in numerous ecotype studies of serpentine-tolerant species such as Gilia capitata, Leptosiphon androsaceus, Linanthus bicolor, the Mimulus guttatus complex, and Collinsia sparsiflora (Schmitt 1980, 1983, Brady et al. 2005, Wright et al. 2006), would not be observed, or evolutionarily relevant, at the deeper phylogenetic levels studied here. If this is the case, then my results present another example of discordance between microevolutionary process and macroevolutionary patterns (Jablonski 2007).
Four alternative explanations consistent with these data are discussed below. While it may be relatively straightforward to test them in particular cases, the large diversity across angiosperms makes it unlikely that a single explanation will apply universally. It remains a difficult task to determine the most important factors involved in the evolution of flowering time on serpentine, and to tease apart their relative effects so extrapolating the trends seen in fine-scale studies should be done with extreme caution.
Countergradient variation
Countergradient variation is a pattern in which genetic influences on phenotype negatively co-vary with environmental effects on phenotype, resulting in minimal change in a given trait value over the gradient (Conover and Schultz 1995). In other words, phenotype shows minimal change because the differences in environmental and genetic contributions to phenotype between two ecotypes counteract each other. This process has been documented in over 60 species including serpentine and non-serpentine ecotypes (Conover et al. 2009). A full factorial study of riparian and serpentine ecotypes of the annual Helianthus exilis grown in potting soil and serpentine soil found that although serpentine genotypes took less time to flower compared to riparian genotypes grown in the same soil, both genotypes flowered earlier when grown on riparian soil compared to serpentine soil (Sambatti and Rice 2007). In this case, within a genotype, the environmental factors promoted later flowering when grown on serpentine, but within a phenotype, genetic differences caused earlier flowering of serpentine-adapted races. Curiously, a reverse example of countergradient variation was found in a reciprocal field-transplant study of C. sparsiflora: Serpentine genotypes flowered significantly later than non-serpentine genotypes when grown together but, regardless of genotype, transplants in serpentine soils flowered earlier (Wright et al. 2006). This finding is consistent with other reports of earlier-flowering serpentine ecotypes or populations (e.g., Dyer et al. 2010). One key difference between the H. exilis and C. sparsiflora studies is that plants in the former study were kept well-watered with distilled water throughout the duration of the experiment. The delay in flowering observed by Sambatti and Rice may have resulted from delayed ontogeny due to edaphic factors such as low soil fertility (Walker 1954, Kruckeberg 1985, Cooke 1994). It is unclear the role that countergradient evolution plays at deeper taxonomic levels, but this phenomenon may contribute to the large number of serpentine/non-serpentine sister lineages that show no difference in flowering time (Fig. 1).
Environmental heterogeneity within serpentine outcrops
This present study is limited to clades of plants important to the California flora, and it is important to point out that serpentine soils (and consequently serpentine-tolerant species) are not evenly distributed across the landscape. Serpentine soils are most abundant in northern and central montane California, including the Sierra Nevada, Klamath Ranges, and Coast Ranges (Kruckeberg 1985), and at elevations between 300 and 2300 m (Burge and Salk 2014). However, aside from their parent material, serpentine soils and outcrops show great variability, even edaphically—worldwide, 11 of the 12 major soil orders include some ultramafic soil types (Rajakaruna and Bohm 1999, Jurjavcic et al. 2002, Alexander et al. 2007). Even within California, variation in aspect, topography, and hydrology can provide a number of different environments in a single patch of serpentine (Rajakaruna and Bohm 1999). This variation may be partially responsible for the conflicting information in the literature regarding water availability on serpentine compared to nearby non-serpentine soils, and likely responsible for variability in flowering time response (Raven and Axelrod 1978, Brady et al. 2005, Alexander et al. 2007). While serpentine soils generally have similar water holding capacity compared to non-serpentine soils (Burt et al. 2001), in western North America such soils are usually much rockier compared to other areas. Consequently, they may be better drained, with less soil to hold water per unit volume, factors that may contribute to the bareness of habitats associated with serpentine soils (Alexander et al. 2007, Cacho and Strauss 2014). However, Raven and Axelrod (1978) noted that some intermittent streams persist longer in the dry season in serpentine areas compared to elsewhere, and recent studies have demonstrated greater variance in both soil water capacity and plant cover on serpentine compared to non-serpentine soils (Harrison et al. 2014, Eskelinen and Harrison 2015). This is generally because such streams often have a fracture-driven hydrology rather than the more common pore-driven hydrology (Alexander et al. 2007). In this diversity of habitats, it is possible for a clade to invade serpentine but not necessarily encounter a drier habitat. For example, the serpentine-tolerant H. exilis and serpentine-endemic Cirsium fontinale are restricted to moist seeps that may remain wetter longer than sites in adjacent habitats. Therefore, even accepting the hypothesis that water availability plays a large role in flowering time differences, heterogeneity in hydrological regimes would be expected to result in heterogeneous responses in flower phenology such as those observed in this study.
Influence of biotic factors
Water availability is a crucial abiotic factor in shaping serpentine soil communities and plant evolution (Kruckeberg 1985, Brady et al. 2005, Anacker and Harrison 2012). However, in some systems flowering phenology may respond more strongly to other biotic or abiotic factors. Increasing attention has been given to the role of pollinators, herbivores, and other biotic agents in affecting flowering phenology (Elzinga et al. 2007). Biotic interactions may put constraints on flowering time evolution or counteract selection of abiotic conditions (Evans et al. 1989, Pilson 2000, Levin 2006). Other times, biotic and abiotic factors can independently lead to convergence in plant traits. For example, sclerophylly, glaucousness, level of pubescence, decrease in specific leaf area, shrubbiness, and increase in root-to-shoot biomass ratios are associated with both herbivore resistance and xeric or low-nutrient environments, like serpentine soils (Brady et al. 2005, Alexander et al. 2007). The ways in which interactions between abiotic and biotic factors affect plant phenology in serpentine systems remain poorly studied and it may be that the evolutionary factors that drive plant phenology at deeper phylogenetic timescales studied here differ from those at fine (e.g., ecotypic) scales.
Earlier flowering time: An exaptation?
The difference in the distribution of flowering times between tolerant and both endemic and non-tolerant lineages noted in Fig. 3 is not independent of phylogeny. That is, species with earlier flowering are marginally more likely to persist on serpentine, although this is variable by clade. Therefore, instead of serpentine acting as a key selective agent following colonization, an earlier-flowering species may simply be more successful at initially colonizing serpentine. The key drivers of both species and trait composition in an area vary over spatial, temporal, and phylogenetic scales (Swenson et al. 2006, Cavender-Bares et al. 2009), and it is possible that earlier flowering time may have an ecological role and a short-term evolutionary role by isolating divergent populations, but little long-term evolutionary role at deeper timescales. This may explain why the distribution of serpentine-endemic flowering times is similar to that of non-tolerant plants. Except in the case of founder-event speciation, a serpentine endemic can be thought of as a serpentine-tolerant lineage that has been extirpated from non-serpentine soils (Anacker et al. 2011).
Among the clades studied, flowering times of non-tolerators formed a bimodal distribution (Fig. 3), consistent with reported patterns in the eastern United States, the Rocky Mountains, and Japan (Kochmer and Handel 1986, Aldridge et al. 2011, CaraDonna et al. 2014). It is possible that non-tolerant lineages that show earlier peak flowering are more likely to give rise to serpentine-tolerant lineages than those that flower later in the season. This hypothesis is supported by the phylogenetically corrected results, which suggest no difference between serpentine-tolerant lineages and their closest relatives (Table 1, Fig. 1).
Conclusions
Phylogenetically independent contrasts and comparisons between three models of flowering time evolution both suggest that serpentine tolerance is not correlated with a unidirectional shift in flowering time. Comparing the results of phylogenetic methods with non-phylogenetic methods can provide stronger insight into the evolutionary ecology of unique edaphic conditions more than either can alone. Although serpentine-tolerant plants show a slightly earlier distribution of flowering times than non-tolerant plants, this pattern is likely a result of ecological filtering, and not post-colonization adaptation. The contrast between these results and finer-scale ecotype studies suggest scale dependence of the major evolutionary factors (biotic or abiotic) governing flowering time phenology and evolution in this and other systems.
Acknowledgments
The author thanks D. Ackerly, B. Baldwin, W. Freyman, C. M. Guilliams, S. Harrison, and B. Mishler for discussion and suggestions at various stages of this research, and several anonymous referees for their evaluation and critiques. B. Anacker, I. Cacho, and S. Strauss graciously provided access to phylogenetic trees from their published studies. Open access publication of this article was made possible by support from the Berkeley Research Impact Initiative sponsored by the UC Berkeley Library. The author has no conflicts of interest to declare.
