Evolutionary and plastic changes in a native annual plant after a historic drought

Abstract Severe droughts are forecast to increase with global change. Approaches that enable the study of contemporary evolution, such as resurrection studies, are valuable for providing insights into the responses of populations to global change. In this study, we used a resurrection approach to study the evolution of the California native Leptosiphon bicolor (true babystars, Polemoniaceae) across populations differing in precipitation in response to the state's recent prolonged drought (2011–2017). In the Mediterranean climate region in which L. bicolor grows, this historic drought effectively shortened its growing season. We used seeds collected both before and after this drought from three populations found along a moisture availability gradient to assess contemporary evolution in a common garden greenhouse study. We coupled this with a drought experiment to examine plasticity. We found evolution toward earlier flowering after the historic drought in the wettest of the three populations, while plasticity to experimental drought was observed across all three. We also observed trade‐offs associated with earlier flowering. In the driest population, plants that flowered earlier had lower intrinsic water‐use efficiency than those flowering later, which was an expected pattern. Unexpectedly, earlier flowering plants had larger flowers. Two populations exhibited evolution and plasticity toward smaller flowers with drought. The third exhibited evolution toward larger flowers, but displayed no plasticity. Our results provide valuable insights into differences among native plant populations in response to drought.

Plasticity may further promote survival during drought events.
Phenotypical shifts in development rate and intrinsic water-use efficiency have been exhibited under drought conditions (Gianoli, 2004;Gugger, Kesslering, Stöcklin, & Hamann, 2015;Kenney et al., 2014;Maherali, Caruso, Sherrard, & Latta, 2010). While it is hypothesized that natural selection may favor genotypes that exhibit plasticity in these traits, there is limited supporting evidence (Aspelmeier & Leuschner, 2004;Franks, 2011;Maherali et al., 2010). Therefore, to understand how droughts associated with global climate change affect plant populations, it is essential to study both natural selection and plasticity of drought-coping traits.
Rapid evolution in response to conditions linked to climate change has been detected for numerous taxa in as few as a couple of generations (reviewed in Parmesan, 2006;Franks, Hamann, & Weis, 2018). One approach used to detect this rapid evolution in plants is the resurrection study, in which seeds collected at one point in time (ancestral) are grown and compared with those collected later (descendant; Etterson et al., 2016;Franks, 2011;Franks et al., 2018;Franks, Sim, & Weis, 2007;Gómez, Méndez-Vigo, Marcer, Alonso-Blanco, & Picó, 2018;Hamann et al., 2018;Sultan, Horgan-Kobelski, Nichols, Riggs, & Waples, 2012;Thomann, Imbert, Engstrand, & Cheptou, 2015;Vigouroux et al., 2011). For plants, this approach is dependent on the collection and storage of seeds over a period in which an environmental change occurs.
Using this approach, evolution toward earlier flowering has been identified in crop species (Nevo et al., 2012;Vigouroux et al., 2011) and, in the common, non-native field mustard (Brassica rapa L.) following drought (Franks et al., 2007). However, later studies with the same populations of B. rapa observed no plasticity in flowering time associated with experimental drought treatments (Franks, 2011;Hamann et al., 2018).
Between 2012 and 2016, California experienced exceptionally warm and dry weather, resulting in its worst drought in at least 1,200 years (Griffin & Anchukaitis, 2014;NOAA, 2019). As the climate continues to change, forecasts suggest that California may expect more of these extreme drought events (Swain et al., 2018).
The recent prolonged drought led to the well-documented deaths of hundreds of millions of native trees, shrubs, and forbs, including exceptionally long-lived species like giant sequoia (Sequoiadendron giganteum Lindl. [J. Buchholz]) and notably drought-tolerant chaparral shrubs, like big berry manzanita (Arctostaphylos glauca Lindl.; Copeland et al., 2016;Jacobsen & Pratt, 2018;Paz-Kagan et al., 2017;Prugh et al., 2018;Stephenson et al., 2018;Venturas et al., 2016;Young et al., 2016). Studies have identified several traits, such as deep roots and shifts in stomatal regulation, that have enabled plants to survive and/or recover from the drought (Choat et al., 2018;Jacobsen & Pratt, 2018;Pivovaroff, Cook, & Santiago, 2018;Venturas et al., 2016). This drought also provides the opportunity to study evolution and plasticity of native populations in response to an extreme event.
The goal of our study was to test for evidence of evolution and plasticity in a native California annual along a naturally occurring precipitation gradient during the state's recent historic drought.
Our study species was Leptosiphon bicolor Nutt. (True babystars, Polemoniaceae), which is a highly selfing winter annual found in grasslands and woodlands throughout California and the far western United States and Canada (Figure 1). In these locations, which experience a Mediterranean-type climate, drought shortens the growing season for winter annuals. Our previous studies with this species have been conducted in populations found along a natural precipitation gradient in the interior coast range of California between 2005 and 2014. In this region, L. bicolor germinates with winter rains, flowers in April, and sets seed and senesces by late May. We have found both spatial and temporal variation in plant traits associated with moisture availability, suggesting both heritable and plastic responses (Lambrecht, 2013). Generally, leaves and flowers are smaller in drier populations and years than in moister ones. Fortuitously, during our years of work with this species, we collected seeds from several field populations along the precipitation gradient around the onset of the prolonged drought. We returned to these populations in 2017 to collect postdrought seeds from the same populations. In the current study, we used these seeds to address the following questions: (a) Is there evidence of evolution toward earlier flowering in these populations in response to the drought? (b) Do the plants from the different populations exhibit differences in evolutionary and plastic responses to drought? (c) Is flowering time associated with other traits, such as intrinsic water-use efficiency or flower size? To answer these questions, we conducted a multiphase resurrection study in a greenhouse on three populations of L. bicolor that vary in moisture availability.

| Study system
The populations of L. bicolor used in this study were selected from among those we have previously studied in Henry W. Coe State Park, Morgan Hill, CA (Lambrecht, 2013;L. T. Rosengreen, unpublished data). These populations, located at least 5 km from one another, span a range of moisture availabilities that this species occupies in central California. In order, from wettest to driest, the populations are Domino, Kingbird Pond, and Pacheco Creek. While we were unable to measure precipitation in the populations during the drought, we previously measured precipitation in or adjacent to these locations from 2005 to 2008, which included two wet and two dry years (Lambrecht, 2013). During that time, the 4-year precipitation average was 71.9 cm for Domino, 38.3 cm for Kingbird Pond, and 26.4 cm for Pacheco Creek (Lambrecht, 2013). Furthermore, the soils at Pacheco Creek are more coarse and sandy than those at Kingbird Pond, causing them to dry more quickly. During the recent prolonged drought, Coe Park experienced overall ~25% reduction of its average precipitation (Figure 2; Henry Coe State Park, 2019). All populations had >1,000 L. bicolor individuals. Each plant produces one to several flowers, which have five corolla lobes that are fused into a long corolla tube. The only known pollinator, the long-tongued fly, has never been observed in these populations, suggesting they are entirely selfing (L. T. Rosengreen, unpublished data). Flowers produce capsules before plants senesce, with each capsule containing ~2-10 seeds. Mature capsules were collected haphazardly from 150 to 200 plants within each population, with seeds separated by maternal line. Collections took place over several days during the peak seed set period of all years; however, this may have led us to miss the earliest and latest seed cohorts. The initial capsule collection year differed across populations, because we did not foresee the drought or this resurrection study at the time of collection

| Refresher generation
Following the protocol of Franks et al. (2018), we produced a refresher generation of F1 seeds to reduce any possible consequences due to prolonged seed storage or maternal effects. For this generation, we grew field-collected seeds in a greenhouse at San Jose State University, San Jose, CA. We planted 80 seeds from each population and year, for a total of 480 seeds, using only one seed per maternal line. In November of 2017, one seed was planted in each 18.4 cm deep Cone-tainer (Stuewe and Sons) filled with moistened Ciardella Mix soil (equal parts redwood compost, horticultural red lava, sand, and peat moss, along with a small amount of fertilizer; Ciardella's Garden Supply). Racks of randomly arranged Cone-tainers were placed in a seed germinator (Percival Scientific GR-36L Seed Germinator Chamber) set to a 12 hr day of 10°C and 12 hr night of 5°C. Cone-tainers were watered daily to keep the seeds moist. After 7 days, the racks were removed from the germinator and placed in a greenhouse away from direct sunlight for 3 days, to allow for acclimation. Then, the racks were removed to an open-air shade house covered with 50% shade cloth, where plants had access to ambient rainfall and light levels. Additional water was added as needed. Seed germination rates were similar across populations and years (overall mean = 76%; χ 2 = 0.91, p = .34 and χ 2 = 0.93, p = .33, respectively). Potential pollinators are unable to access this greenhouse, so the plants were left to self-pollinate. Capsules were collected from plants in spring 2018.

| Experimental generation
We planted a total of 150 F1 seeds collected from the refresher generation from each population and year combination, for a total of 900 seeds. This total included a minimum of 2 seeds from each of at least 60 F1 maternal lines, with the fully selfed siblings divided between the watering treatments (see below). Our seed germination protocol was the same as that for the refresher generation of the study, with the exception that we planted seeds on three separate dates, due to the capacity of the seed germinator. Seeds from each population and year were divided equally among the planting dates.
During this generation, germination rates were ~72%. Following removal from the germinator, the Cone-tainers were placed in an enclosed greenhouse out of direct sunlight on a bench for 3 days. Then, they were moved under 400 w daylight metal halide lamps (Hortilux Blue) set to a 12 hr day that provided a photosynthetic photon flux density of ~300 µmol m −2 s −1 . Greenhouse temperatures were set to 24°C during the day and 18°C during the night. Relative humidity during the experiment was ~50%. We recorded flowering time as the number of days between germination and flower opening.
Our watering treatment was initiated once plants had acclimated to the greenhouse. For the first two weeks in the greenhouse, all plants were watered daily to saturation. In the third week, we reduced watering to two times per week. Finally, in the fourth week, we ceased all water for the low water treatment, while the well-watered treatment continued to receive water twice a week.
This watering regime was designed to mimic the dry-down period experienced in field populations during spring, based on our preliminary greenhouse studies. This timeline also enabled us to initiate the drought treatment prior to any flowering.
We observed plants daily to record germination and flowering date. On the date the first flower opened, we measured the width and length of the corolla lobes, as well as the diameter of the corolla face, which are measurements we have previously shown to vary with moisture availability (Lambrecht, 2013). We then collected the aboveground vegetative portion of each plant for stable carbon isotope analysis (δ 13 C) to assess intrinsic WUE. Collected plants were placed in a 60°C drying oven for 72 hr, before being weighed and ground to a fine, uniform sample. Due to the small size of L. bicolor, we were only able to analyze 436 of the 499 plants that flowered (87%), since the remainder were too small. A 3.0 ± 1.0 mg subsample was analyzed for δ 13 C on a Delta-V Isotope Ratio Mass Spectrometer fitted with a Costech Elemental Analyzer at the Facility for Isotope Ratio Mass Spectrometry (FIRMS, University of California). For those plants that died before flowering, we recorded their date of death so we could assess survival across the generations and watering treatments. The experiment was terminated 11 weeks after initial planting. By that time, 63 plants had failed to flower.

| Data analyses
To test for evidence of earlier flowering time, we used a Cox proportional hazards model (Fox, 2001). These and all subsequent tests were analyzed using SPSS (v.25, IBM Corp.). Our analyses included generation (ancestral or descendant), population, watering treatment, and block (planting date), as well as all 2-and 3-way interaction terms between generation, population, and treatment. The generation term indicates evolution, while watering treatment indicates plasticity. The interaction of generation × watering treatment estimates whether the scale of plasticity changed after the drought.
The Cox model is semiparametric and does not assume a normal distribution, which is essential in testing flowering time, because some plants die before they flower (Fox, 2001). The test generates a Wald chi-squared test statistic, which is a one-tailed test with α = 0.10.
After our initial analysis indicated significant differences between populations (χ 2 = 9.535, df = 2, p = .008), we ran post hoc pairwise models to compare populations, using the Sidak-corrected α = 0.034 (Tripathi & Pandey, 2017). Finally, we ran individual analyses of flowering time for each population. Rates of evolutionary change in flowering time were calculated as Haldanes, which calculates unit change per generation, expressed as standard deviations (Gingerich, 2001).
We used mixed models to examine how physiological and morphological traits (intrinsic WUE, flower size, and aboveground vegetative mass) varied among populations over time and with watering treatment using MIXED in SPSS. MIXED uses restricted maximum likelihood methods. Additionally, it only generates type III sums of squares for fixed effects, while random effects contribute only to the covariance structure of the model. Denominator degrees of freedom are calculated using Satterthwaite approximations and are, therefore, not integers. Our initial models included population, generation, and watering treatment as fixed factors, and block as a random factor. For these and the tests described below, α = 0.05.
We first ran the full model including all 2-and 3-way interactions between population, generation, and treatment. We then ran subsequent models, eliminating nonsignificant interaction terms, one at a time, selecting the model with the lowest value of Akaike's information criteria. When a significant population effect was detected, we used estimated marginal means with a Bonferroni correction to make pairwise comparisons between populations. Finally, intrapopulation models were run to determine the effect of generation and watering treatment within each population.
We used pairwise Pearson correlations to examine the relationship between flowering time and the other measured traits.
Correlations were analyzed within each population, generation, and treatment combination.
We assessed survival of plants through the study using the nonparametric chi-squared test of independence. We compared survival both across the generations and between the watering treatments.
We followed these tests with an analysis that compared survival in the low water treatment between generations of each population.  (Table 3; Figure 3e). In contrast, evolution of WUE was not detected (Tables 2 and 3; Figure 3b). However, populations did exhibit differences in WUE (Table 2). Domino, which flowered later than the other populations, had higher WUE than both Kingbird Pond (p = .006) and Pacheco Creek (p = .001).

| Plasticity in response to an experimental drought
All three populations exhibited phenological plasticity by flowering earlier in the low water as compared to the well-watered treat-  producing smaller plants, indicated by reduced biomass, and flowers in the low water treatment. (Table 3; Figure 3). Kingbird Pond plants were smaller in the low water treatment, but floral size traits exhibited no plasticity (Table 3; Figure 3).

| Relationship between flowering time and other traits
Flowering time was more strongly correlated with morphological than physiological traits. Generally, flowers and plants were larger in earlier as compared with later flowering plants (Figure 4). For Domino, these negative correlations were observed more in the low water treatment and in the ancestral generation. For Kingbird Pond, these correlations were more frequent in the low water treatment and the descendant generation. In contrast, Pacheco Creek plants exhibited consistent correlations across generations and water treatments. Furthermore, while it was expected that plants that flowered earlier had lower intrinsic WUE than those that flowered later, this was true only in Pacheco Creek (Figure 4).

| Survival
Survival rates were affected by the watering treatment, but did not change between generations. The low watering treatment reduced Note: Values shown are F-statistics and denominator degrees of freedom. See Figure 3 for mean values.
*p < .05, **p < .01, ***p < .001, Bold values are significant following the Holm's sequential Bonferroni correction. NS indicates nonsignificant interaction terms that were removed because their inclusion did not improve the model fit.

| D ISCUSS I ON
Our results demonstrate how populations of a native plant species differed in their responses to California's recent historic drought.
In the Mediterranean climate region in which L. bicolor grows, the recent drought effectively shortened its growing season. However, only the wettest population (Domino) showed evolution toward earlier flowering, although all three populations showed plasticity toward earlier flowering in response to the experimental drought.
These results contrast with those of a resurrection study of Brassica rapa, which demonstrated evolution, but no plasticity in flowering time in response to natural and experimental drought (Franks, 2011;Franks et al., 2007;Hamann et al., 2018). We also observed trade-offs

| Evolution in response to a historic drought
In our resurrection study, the population with the greatest moisture availability, Domino, was the only one to exhibit evolution toward Aspects of our study may have limited our ability observe evolution in flowering time. First, seed collection in the field took place over several days during peak seed set for each population. We may, therefore, have missed the earliest and latest seed cohorts, which will have affected our estimates of flowering time. Secondly, our ancestral collection from Kingbird Pond was in 2014, after the worst two years of drought had passed. Therefore, we missed any evolution that may have occurred in the preceding years. This was unavoidable, since we had collected and stored our seeds for other projects and had not anticipated the drought. Finally, due to somewhat low germination rates in our refresher generation, we may have introduced an invisible fraction bias because the phenotypes of the 25% of seeds that did not germinate could not be recorded or used in our estimates of evolution (Weiss, 2018). The bias occurs when seed mortality is nonrandom. However, when we planted sibs in our ancestral generation, we observed no family-wise pattern in germination, suggesting that failure to germinate was random.
Evolution toward earlier flowering has been demonstrated in a wide array of annual taxa in response to experimental drought treatments, indicating that a drought escape strategy may be favored in annuals so that reproduction and the life cycle may be completed while soil water remains available (Brouillette, Mason, Shirk, & Donovan, 2014;Heschel & Rignios, 2005;Ivey & Carr, 2012;Kenney et al., 2014;Manzaneda et al., 2015;Sherrard & Maherali, 2006). The trend toward earlier flowering is supported by life history theory, which suggests that there are greater fitness costs with flowering later than flowering early (Austen, Rowe, Stinchcombe, & Forrest, 2017;Weis, Wadgymar, Sekor, & Franks, 2014). However, interannual variability in weather, which is predicted to increase with global change, may reverse the selection for early flowering that occurs during dry years and lead to maladaptive responses during intermittent wet years . Therefore, rapid evolution in flowering time may not be sufficient to maintain plant fitness under climate change.
While we did not detect evolution in WUE, we did observe that plants that flowered earlier in Pacheco Creek had a lower WUE, which was consistent with a drought escape strategy. Studies have shown flowering time and WUE tend to be more highly correlated under dry conditions (Edwards et al., 2012;Juenger, 2013  reflect the overall condition of the plant. In a recent analysis, 24 of 28 studies surveyed showed a negative correlation between flowering time and plant biomass, suggesting that those that are in better condition (e.g., larger) can flower early and produce larger flowers (Forrest, 2014).
Water loss from flowers may also influence flower size. Work with L. bicolor and the closely related L. androsaceus has demonstrated that flowers of Leptosiphon lose substantial amounts of water, and plants may produce smaller flowers in drier locations or conditions in order to limit that loss (Lambrecht, 2013;Lambrecht, Morrow, & Hussey, 2017). Floral water loss of Leptosiphon and other plant species has been shown to affect leaf physiology and can lead to reduced gas exchange, particularly in dry conditions (Dudley, Arroyo, & Fernandez-Murillo, 2018;Galen, Sherry, & Carroll, 1999;Lambrecht, 2013). taxa in moister or more variable environments (Gianoli, 2004;Kenney et al., 2014;Manzaneda et al., 2015). However, increased plasticity is not universally favored. A study of flowering time plasticity in Arabidopsis thaliana (thale cress, Brassicaceae) identified selection against plasticity in warmer environments where costs of plasticity were too high to maintain (Stinchcombe, Dorn, & Schmitt, 2004a).
Furthermore, long-term studies with B. rapa have found no plasticity in flowering time in response to drought (Franks, 2011;Hamann et al., 2018). Given the different responses across our populations, it is possible that plasticity in flowering time was sufficient to cope with drought in the two drier populations. However, it should be noted that we may have overestimated plasticity in flowering time, due to the invisible fraction bias introduced by the high mortality in the low water treatment. If late-flowering genotypes died before flowering, it would appear that populations exhibited plasticity toward earlier flowering.
As with flowering time, plasticity in WUE is hypothesized to be favored in variable environments (Kenney et al., 2014). In our study, all populations exhibited plasticity toward increased WUE in the low watering treatment. Therefore, the direction of plasticity of WUE was opposite that expected given the plasticity for earlier flowering time. While such patterns might suggest that the direction of plasticity is maladaptive, similar results have been observed in other taxa Kenney et al., 2014). The time at which the drought treatments were applied may have affected these results, where early droughts have been shown to favor low WUE and early flowering, while later droughts favor increased WUE Heschel & Rignios, 2005). Furthermore, selection analyses reveal that plasticity in WUE, per se, is favored to cope with periodic droughts and enable plants to make use of water while it is available (Kenney et al., 2014). However, few studies have directly tested for the adaptive value of plasticity. Moreover, epigenetics (Zhang, Fischer, Colot, & Bossdorf, 2013) and gene deletions (Stinchcombe, Weinig, et al., 2004b) influence flowering time and its response to environmental conditions, potentially disrupting its relation with other plant traits.

| How did populations along a moisture gradient differ in drought response?
Studies of species across broad geographic ranges have revealed differentiation in drought responses, where adaptations favor different combinations of traits associated with both drought escape and drought tolerance across populations, indicating that species do not rely strictly upon one or the other strategy (Bibee, Shishido, Hathaway, & Heschel, 2011;Brouillette et al., 2014;Kooyers et al., 2015;Paccard et al., 2014). Furthermore, differential amounts of plasticity have also been observed across populations in response to drought, with reduced levels of plasticity typically found in more stressful environments (e.g., warmest or driest) or at the margin of a species' range (Gugger et al., 2015;Paccard et al., 2014;Stinchcombe, Weinig, et al., 2004b). In accord with these studies, we observed the least plasticity in the driest of our populations, Pacheco Creek, and the most in the wettest, Domino. Our results suggest that wild populations may have divergent responses to short-and long-term droughts.
In conclusion, our study has demonstrated how rapidly drought can affect responses in a native plant species. We detected both plasticity and evolution in flowering time and flower size in just a few generations, although the magnitude and direction of these responses varied broadly across our three populations. These results provide insight into the variability of evolution across populations, where the role of plasticity versus evolutionary changed varied along the precipitation gradient. Furthermore, extreme drought events and interannual fluctuations in precipitation are predicted to increase as the climate changes. Given that intervening wet years may reverse the selection experienced during drought years , there is the need for multiyear studies in a broader array of taxa and across species' ranges in order to understand how populations and species respond to global change.

ACK N OWLED G M ENTS
We thank the following people who helped set up and maintain

CO N FLI C T O F I NTE R E S T
All authors declare there are no conflicts of interest.

AUTH O R CO NTR I B UTI O N S
SCL designed the study and contributed to greenhouse work, data analysis, and manuscript preparation. AKG and LJR contributed to greenhouse and laboratory work and manuscript preparation. LTR contributed to study design, greenhouse work, data analysis, and manuscript preparation.

O PE N R E S E A RCH BA D G E S
This article has earned an Open Data Badge for making publicly available the digitally-shareable data necessary to reproduce the reported results. The data is available at https://doi.org/10.5061/ dryad.3tx95 x6c3.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data are archived on Dryad (https://doi.org/10.5061/dryad.3tx95 x6c3).