Evolutionary divergence of potential drought adaptations between two subspecies of an annual plant: Are trait combinations facilitated, independent, or constrained?

Abstract

Premise

Whether drought‐adaptation mechanisms tend to evolve together, evolve independently, or evolve constrained by genetic architecture is incompletely resolved, particularly for water‐relations traits besides gas exchange. We addressed this issue in two subspecies of Clarkia xantiana (Onagraceae), California winter annuals that separated approximately 65,000 years ago and are adapted, partly by differences in flowering time, to native ranges differing in precipitation.

Methods

In these subspecies and in recombinant inbred lines (RILs) from a cross between them, we scored traits related to drought adaptation (timing of seed germination and of flowering, succulence, pressure–volume curve variables) in common environments.

Results

The subspecies native to more arid environments (parviflora) exhibited slower seed germination in saturated conditions, earlier flowering, and greater succulence, likely indicating superior drought avoidance, drought escape, and dehydration resistance via water storage. The other subspecies (xantiana) had lower osmotic potential at full turgor and lower water potential at turgor loss, implying superior dehydration tolerance. Genetic correlations among RILs suggest facilitated evolution of some trait combinations and independence of others. Where genetic correlations exist, subspecies differences fell along them, with the exception of differences in succulence and turgor loss point. In that case, subspecies difference overcame genetic correlations, possibly reflecting strong selection and/or antagonistic genetic correlations with other traits.

Conclusions

Clarkia xantiana subspecies’ differ in multiple mechanisms of drought adaptation. Genetic architecture generally does not seem to have constrained the evolution of these mechanisms, and it may have facilitated the evolution of some of trait combinations.

Plant adaptation to the soil water deficits and dehydration risks associated with climatic drought can involve diverse processes (Passioura, 1996; Tardieu, 2012; Gilbert and Medina, 2016; Volaire, 2018). For example, plants may resist dehydration due to drying soil by reducing stomatal conductance (Martin‐StPaul et al., 2017) or storing water in succulent tissues (Ogburn and Edwards, 2012), and they may tolerate dehydration by resisting turgor loss (Maréchaux et al., 2015) or resisting hydraulic cavitation (Lens et al., 2016; Zhang et al., 2016). Where precipitation is ephemeral (e.g., some deserts) or seasonal (e.g., Mediterranean climates), several other strategies can provide drought adaptation (Kooyers, 2015). Such plants can escape approaching drought by reproducing early (Volis et al., 2002; Sherrard and Maherali, 2006) or avoid drought altogether by producing seeds that do not germinate without prolonged exposure to saturated soil, a cue to sustained water availability (Kos and Poschlod, 2010; Zeng et al., 2010; Duncan et al., 2019a, 2019b, 2019a, 2019b). Seed dormancy can create persistent seed banks that spread the risk of encountering severe drought (and reproductive failure) among years that vary in precipitation (Cohen, 1966; Venable and Lawlor, 1980; Adondakis and Venable, 2004; Donohue et al., 2010; Saatkamp et al., 2011; Gremer and Venable, 2014). Seeking to understand how adaptation to drought evolves by these and other mechanisms is timely, given the increasing frequency and severity of climatic drought worldwide (Sheffield and Wood, 2008; Taylor et al., 2012; Keellings and Engström, 2019; Koutroulis et al., 2019).

Adaptation to drought by escaping it—maturing before drought becomes severe—appears common in annuals and herbaceous perennials. Considerable evidence suggests that when and where drought arrives early, earlier flowering evolves. For example, earlier flowering is associated with xeric distributions among populations or species in Arabidopsis (Brassicaceae) (e.g., McKay et al., 2003; Paccard et al., 2014; Monroe et al., 2018), Clarkia (Vasek, 1968; Runions and Geber, 2000; Mazer et al., 2004), Erodium (Geraniaceae) (Latimer et al., 2019), Lupinus (Fabaceae) (Berger and Ludwig, 2014), Mimulus (Erythranthe) (Phyrmaceae) (e.g., Hall and Willis, 2006; Wu et al., 2010; Ferris and Willis, 2018; Mantel and Sweigart, 2019), Oryza (Poaceae) (Groen et al., 2020), and Streptanthus s.l. (Brassicaceae) (Pearse et al., 2019). In populations of Brassica rapa (Brassicaceae) (Franks, 2011; Hamann et al., 2018), multi‐year droughts led to the evolution of earlier flowering (but see Dickman et al. [2019] for a contrasting outcome in Mimulus laciniatus).

When early flowering enables drought escape, how do other mechanisms of drought adaptation evolve? One pattern is a trade‐off between drought escape achieved by early maturity versus dehydration avoidance achieved by stomatal regulation of gas exchange (Geber and Dawson, 1990; Heschel and Riginos, 2005; Emms et al., 2018). This trade‐off suggests not only that selection for other drought adaptations might be relaxed if early maturity enables drought escape, but also that to achieve early maturity requires sacrificing some other mechanisms of drought adaptation.

Less is known about how drought‐avoiding seed‐germination behavior evolves with other traits. The relationships may be complex because germination timing can alter the selective environment experienced later in life (Donohue, 2002). Some pertinent evidence comes from long‐term studies of winter annuals in North America’s Sonoran Desert. Correlations exist in that community between the water relations of seed germination and long‐term water‐use efficiency (Kimball et al., 2011; Huang et al., 2016; Liu et al., 2020). A strategy of slow germination restricted to high soil water potentials (with high fractions of seeds remaining dormant) correlates with high growth rate and low water‐use efficiency (estimated by stable isotopes); a contrasting alternative combines rapid germination (with high germination fractions) at lower soil water potentials with high water‐use efficiency. Thus, species with the first set of trait values are more likely to avoid drought, sacrificing their ability to postpone dehydration, while those with the second set are more likely to encounter drought but maintain performance for a time, as it intensifies. These sets of traits also correlate with demography; drought‐avoiding and ‐escaping species having particularly variable demography among years (Huang et al., 2016). In contrast, a resurrection study of Mimulus laciniatus of seeds sampled from populations before and after California’s multi‐year megadrought in the 2010s (see Kogan and Guo, 2015 for examination of this megadrought and its severity), Dickman et al. (2019) reported not only the evolution of slightly later flowering noted above but also more rapid—not slower—seed germination. As studies of tissue water‐relations traits such as succulence, turgor loss point, and cavitation resistance are rare in annuals and other herbaceous plants (but see Lens et al., 2016; Doria et al., 2019), it is not clear how these features contribute to adaptation and/or evolve with other drought adaptations in annuals.

How traits that confer drought avoidance, drought escape, dehydration resistance, and dehydration tolerance evolve in the short term will depend on the underlying genetic structure of the traits involved (Juenger, 2013). In annual plants to date, there has been extensive research on genetic variation and possible constraints on model plants such as Arabidopsis thaliana (e.g., McKay et al., 2003, 2008; Ågren et al., 2013; Lovell et al., 2013; Mojica et al., 2016; Taylor et al., 2017; Auge et al., 2019) but much less work on other wild species. Investigations of genetic variation and covariation in drought‐related traits besides phenology and gas exchange appear to be very rare in wild annuals, leaving unaddressed whether combinations of these traits are facilitated, independent, or constrained by genetic architecture.

We address some of these knowledge gaps by investigating recent evolutionary divergence in water relations in the California annual plant Clarkia xantiana. Two recognized subspecies of C. xantiana are distinct lineages that separated approximately 65,000 years ago (Pettengill and Moeller, 2012b). They are differently adapted to geographic regions with contrasting winter precipitation, partly because the subspecies from more arid areas flowers earlier (Geber and Eckhart, 2005; Benning et al., 2019). In a common environment, we scored phenological and physiological traits expected to affect adaptation to drought of populations of each subspecies and of recombinant inbred lines (RILs) from a cross between them. We asked whether (1) C. xantiana subspecies differ in germination timing and tissue water relations as well in flowering phenology; (2) there is substantial genetic variance and covariance in these traits among RILs; (3) genetic correlations indicate facilitated, constrained, or independent evolution of trait combinations; and (4) subspecies differences fall along genetic correlations or suggest evolution away from them. We consider what patterns of subspecies differences and genetic correlations suggest about the evolutionary history and physiology mechanisms of drought adaptation in this species and other wild annuals.

MATERIALS AND METHODS

Study system

Clarkia xantiana A.Gray occurs in mountainous, sparsely populated areas of inland southern and central California, usually between 500 and 1600 m a.s.l., on steep, sandy slopes in grassland, pine–oak savanna, and openings in chaparral (Lewis and Lewis, 1955; Eckhart and Geber, 1999). The majority of the distribution is on U.S. public land that is grazed by cattle, and it occurs mainly within the pre‐contact territories of the Kawaiisu, Kitanemuk, Tabtulabl, Western Shoshone, and Yokuts peoples (Conservation Biology Institute, 2020). Its two recognized subspecies, C. xantiana subsp. xantiana (hereafter “xantiana”) and C. xantiana subsp. parviflora (Eastw.) F.H. Lewis & P.H. Raven (“parviflora”) have mostly allopatric distributions, with parviflora occupying more arid areas leeward of mountain ranges that cast rain shadows (Eckhart and Geber, 1999; Eckhart et al., 2011). A narrow sympatric zone, in the Kern River drainage of the southern Sierra Nevada, represents secondary contact (Pettengill and Moeller, 2012a; Briscoe Runquist and Moeller, 2014). The subspecies differ in mating system, with xantiana strongly outcrossing and parviflora highly self‐pollinating (Eckhart and Geber, 1999; Fausto et al., 2001).

The early maturity of parviflora contributes to its ability to occupy more arid regions than xantiana. In common environments in nature and in cultivation, parviflora begins to flower 1–4 weeks earlier than xantiana (Moore and Lewis, 1965; Eckhart and Geber, 1999; Runions and Geber, 2000; Eckhart et al., 2004). In reciprocal transplant experiments, each subspecies has higher lifetime fitness (also higher than that of the other subspecies) in its native geographic range; later flowering by xantiana in parviflora’s range leads to higher mortality before setting seed, partly because of dehydration (Geber and Eckhart, 2005). In parviflora’s exclusive range, early flowering also increases survival by reducing the risk of being eaten by small mammals (Geber and Eckhart, 2005; Benning et al., 2019). In its exclusive range xantiana benefits from its larger stature (correlated with its later flowering; Eckhart et al., 2004), which appears to reduce susceptibility to fungal attack and also may increase competitive ability in the more productive plant communities xantiana inhabits (Geber and Eckhart, 2005).

The RILs studied here were produced by four generations of single‐seed descent from individuals in the F₂ generation of a cross between a xantiana population near the upper end of the lower Kern River Canyon, below Isabella Dam, and a parviflora population from the Pacific Crest Trail near Walker Pass, 45 km east. These are sites “8” and “47,” respectively, from previous publications (e.g., Pettengill and Moeller, 2012a). Including the selfed F₁ generation that created the F₂, the RILS had experienced five generations of self‐fertilization. Because phenotypic variation within RILs is almost exclusively environmental and variation among them almost exclusively genetic, among‐RIL variance estimates broad‐sense genetic variance in the cross, and correlations among RIL means estimate broad‐sense genetic correlations (Bulmer, 1980; Hegmann and Possidente, 1981; Falconer and Mckay, 1996). Heritabilities and genetic correlations in the RIL “population” do not represent the genetic architecture of any single wild population but rather describe the quantitative genetics of differences between subspecies. To make it feasible to estimate tissue water‐relations traits on several replicate individuals per RIL, from time‐ and labor‐intensive pressure–volume curves (see below), we selected at random a modest number of RILs (14) from those with sufficient seeds to carry out the study. For the sake of feasibility, this feature traded off power to detect significant genetic correlations. At α < 0.05, with 12 degrees of freedom, r must exceed 0.532.

Experimental design and data collection

We cultivated parent populations and RILs in two temporal blocks, each block lasting approximately 10 weeks. Seeds from the RILs came from their F₅ parents cultivated in a greenhouse in 2017, stored dry at room temperature for one month and at 4°C thereafter. The seeds from parental populations came from bulk collections of approximately 50 maternal families collected in natural populations in June 2018, with seeds stored dry at room temperature for 1 month and then dry at 4°C. The seeds from the parent populations that founded the RILs were collected in the early 2010s. For this study, we wanted to use seeds of similar age that had experienced similar storage conditions. With these materials, we scored a series of traits based on their demonstrated and hypothesized contributions to drought adaptation (see above; Table 1).

TABLE 1.

Measured traits and their interpretations in terms of drought adaptation.

Trait	Description and interpretation
T GERM	Time from sowing to germination. Seeds that require prolonged periods at high water potentials to germinate increase the likelihood of drought avoidance (Bradford, 1990; Bradford and Still, 2004; Kos and Poschlod, 2010).
T FLOWER	Time from germination to flowering. Plants that flower earlier minimize exposure to terminal water‐deficit, providing drought escape (Franks, 2011; Kooyers, 2015; Monroe et al., 2018).
SWC	Saturated water content (succulence). Plants may store water, relying on it during soil water deficit, delaying dehydration (MacDougal et al., 1913; Ogburn and Edwards, 2012). SWCLEAF and SWCSTEM are mass‐based leaf succulence and stem succulence, respectively. SWCPV, estimated from a pressure–volume curve, is the succulence of shoot tips with stem, leaves, and flower buds.
RWCTLP	Water content (relative to maximum) at turgor loss. Lower values indicate greater ability to maintain turgor, thus tolerating dehydration.
π O	Tissue osmotic potential at full turgor, lowered by dissolved solutes. Lower values (farther below 0) index the ability to tolerate dehydration (Farrell et al., 2017).
Ѱ TLP	Shoot water potential at the turgor loss point. Below this point, plants wilt. Lower values (farther below 0) confer greater tolerance of dehydration (Bartlett et al., 2012; Farrell et al., 2017).

Each of the two complete blocks of the experiment began when we sowed 10 seeds per RIL and per subspecies, placing seeds in individual positions on indented seed‐germination blotting paper (Seedburo, Des Plaines, IL, USA) inside 15‐cm Petri dishes. We distributed seeds of different origin systematically (“checkerboard fashion”) into grids to avoid any genotype–environment correlation. Cotton balls placed underneath the paper, wetted with 10 mL of de‐ionized water, kept the paper consistently moist, and sealing with Parafilm (Bemis, Inc., Neenah, WI, USA) reduced moisture loss. After sowing we moved dishes into a growth chamber programmed for 14 h of dark (at 5°C) and 10 h of light (at 15°C), simulating soil‐surface conditions during natural seed germination (V. M. Eckhart, unpublished data). Sampling daily, we scored the time to germination for each seed as days from sowing (T _GERM) as radicles emerged. Terminating the seed‐germination period at 14 d in each block, we found that germination was virtually complete. Only 19 of 320 seeds (6%) failed to germinate in 14 d. We conducted squeeze tests (Baskin and Baskin, 2014) using a dissecting microscope to determine viability of ungerminated seeds. Seeds that, when squeezed open, were found to contain intact, white embryos we considered viable (15 of 19).

Upon seed germination, we transplanted each seedling into a 13 cm wide by 8 cm deep round plastic pot filled with equal volumes of potting soil (FLX, Hummert, Earth City, MO, USA) and fritted clay (Profile Products, Buffalo Grove, IL, USA). At transplanting, we applied a blind labeling protocol intended to reduce bias in data collection. Each plant received a label with a unique code with no other identity cue, and each day’s sample of newly transplanted seedlings we distributed widely and haphazardly on two 6.4‐m² benches in a 22.5‐m² greenhouse room. Only after we completed data collection for each block of the experiment did we return to the codebook to identify plants. Thus, in all subsequent data collection, we were blind to the plant’s identity until termination of its block of the experiment. Metal‐halide, high‐intensity discharge lamps supplemented light for 14 h per day, during which set points were 21°C for heating and 30°C for cooling. While lights were off, set points were 13°C to heat and 18°C to cool. We applied 100 mL of fertilizer (Jack’s Classic 20‐20‐20 with Micronutrients, Allentown, PA, USA) per plant twice before flowering, and irrigation kept plants well watered throughout. Of 160 transplanted seedlings per block, 97 in block 1 and 88 in block 2 survived to flowering to be scored for adult traits, an average of 11.5 individuals per RIL or parent population (8–16 individuals) from 20 seeds sown across the two blocks.

After recording an individual’s time from germination to flowering (T _FLOWER), we scored the plant for tissue water relations that day or on the first day it became feasible, subject to the limitation that it was possible to process only 5–6 plants per workday. To hydrate plants, we irrigated to saturation 1–4 h before collecting data. On the top 5 cm of the central shoot (Vilagrosa et al., 2014), we measured water potentials and pressure‐volume curves (Turner, 1988; Sack et al., 2011) by the bench method, using a pressure chamber (Model 1000, PMS Instrument Co., Albany, OR, USA). To accelerate desiccation after shoot tips lost turgor, we placed a fan on low speed 10 cm away from the excised shoots. We analyzed data only from those pressure‐volume curves that met the following conditions: (1) initial water potential exceeded −1.00 MPa; (2) water potential could be estimated at seven or more points during dehydration; and (3) the curves possessed a single, obvious inflection point, indicating turgor loss. A few other anomalies (e.g., curves for which shoot water potential increased during dry‐down, presumably because of measurement error) compelled us to eliminate two other curves. In all, we scored pressure–volume curves for 151 plants, retaining 136 curves for data analysis. We used the spreadsheet tool of Sack et al. (2011) to calculate four tissue water relations variables: (1) saturated water content at full turgor (SWC_PV; PV refers to pressure–volume for this measure of succulence, which applies to shoot tips), (2) water potential at the turgor loss point (Ψ _TLP), (3) osmotic potential at full turgor (π _o), and (4) relative water content at the turgor loss point (RWC_TLP).

For each plant, on the same day we obtained its pressure–volume curve, we estimated the mass‐based succulence of its stems and leaves: water mass/dry mass (Ogburn and Edwards, 2012). To estimate stem succulence (SWC_STEM), we cut a 2‐cm length of stem from each individual immediately below the shoot tip used for pressure–volume analysis, weighing it immediately and then again after 48 h of convection drying at 65°C. To estimate an individual’s leaf succulence (SWC_LEAF), we removed all leaves except those on the shoot tip that had been clipped off for pressure–volume analysis, weighing them immediately and again after 48 h of drying at 65°C.

Statistical analyses

We completed statistical analyses using R version 3.6.2 (R Core Team, 2019). We assessed normality for each variable with a Shapiro–Wilk normality test (base stats version 3.6.2; R Core Team, 2019). For normally distributed variables we compared parent subspecies differences through a Wald χ ² test with subspecies (fixed) and block (random) effects using lme4 (version 1.1.21; Bates et al., 2015) and car packages (version 3.0.5; Fox and Weisberg, 2019). We used a restricted maximum‐likelihood (REML) approach and derivative‐free bound‐constrained optimization with bobyqa. To estimate the significance of the block effect, we used the ranova function from the lmerTest package (version 3.1.0; Kuznetsova et al., 2017). Only for T _FLOWER was there a block effect, slightly later flowering in block 2 (P = 0.028, LRT = 4.849). For T _GERM, which was not normally distributed, we compared subspecies with the Scheirer–Ray–Hare extension of the Kruskal–Wallis test in the rcompanion package (version 2.3.7; Mangiafico, 2019).

For the data on RILs, we estimated variance components using the lme4 (version 1.1.21; Bates et al., 2015) and insight packages (version 0.7.1; Lüdecke et al., 2019), where both RIL identity (reflecting genetic variation) and block (reflecting between‐block environmental variation) were treated as random. For SWC_STEM and Ψ _TLP the variance due to block approached 0. Following the guidance of Barr et al. (2013), we kept the block effect in all models. We calculated 95% confidence intervals of broad‐sense heritability (among‐RIL variance over total phenotypic variance) based on the χ ² distribution.

To characterize the genetic correlation structure among traits, we used two approaches. First, we calculated pairwise correlations among RIL means using the corrplot package (version 0.84; Wei and Simko, 2017), interpreting these as broad‐sense genetic correlations. For significant correlations, we examined bivariate relationships, plotting 95% confidence ellipses using ggplot2 (version 3.3.2; Wickham, 2016). To these plots, we added subspecies means to visualize subspecies’ positions in trait space alongside RIL means. Second, we looked for larger structural patterns with a principal component analysis (PCA) of RIL means using the factoextra (version 1.0.6; Kassambara and Mundt, 2019) and Hmisc packages (version 4.3.0; Harrell and Dupont, 2019). We used singular value decomposition (SVD) to perform the PCA with the prcomp() function. We included six traits in the analysis: T _GERM, T _FLOWER, Ψ _TLP, π _o, SWC_STEM, and SWC_LEAF, excluding SWC_PV because it correlated strongly with SWC_STEM and RWC_TLP because it did not differ between subspecies. To visualize how parent subspecies compare to RILs, we placed parents in PC1–PC2 space by multiplying z‐scores for each trait and subspecies by principal component loadings. Z‐scores were trait‐ and subspecies‐specific, calculated as the difference between each subspecies’ mean trait value and the mean of that trait among the RILs, divided by standard deviation of the RIL values.

RESULTS

Parent populations of the two subspecies differed distinctly in almost every trait measured (Fig. 1). Seed germination time differed marginally (H _{1, 15} = 3.10; P = 0.078), the trend being that xantiana seeds germinated in two‐thirds the time of parviflora seeds, approximately 2 days earlier (Fig. 1A). After germination, parviflora flowered on average at 40 days (±1.2 SE), sooner by a third (13 days) than xantiana (Fig. 1B; ${\chi }_1^2$ = 45.18; P = 1.8 × 10^‐11). In all measures of succulence parviflora exceeded xantiana, by 25% in shoot‐tip succulence (SWC_PV) (Fig. 1C; ${\chi }_1^2$ = 7.14; P = 0.0075), 25% in leaf succulence (Fig. 1D; ${\chi }_1^2$ = 6.44; P = 0.0112), and 50% in stem succulence (Fig. 1E; ${\chi }_1^2$ = 14.85; P = 0.00012). Water‐potential traits also differed, with xantiana exhibiting lower averages by 0.2 MPa in π _o (Fig. 1G; ${\chi }_1^2$ = 5.38; P = 0.0203) and 0.3 MPa in Ψ _TLP (Fig. 1H; ${\chi }_1^2$ = 5.09; P = 0.024). RWC_TLP was the only variable with no evidence of a subspecies difference (Fig. 1F; ${\chi }_1^2$ = 0.53; P = 0.47), both subspecies averaging close to 80% (±1.9 SE for xantiana and ±2.2 SE for parviflora). Therefore, we did not analyze among‐RIL variation in RWC_TLP.

FIGURE 1.

Subspecies (i.e., parent population) comparisons for phenological and physiological traits. Symbols are means ± 1 SE. *P < 0.05, **P < 0.01, ***P < 0.001.

For every trait, the range among RILs exceeded that of the parental populations, and the among‐RIL component of variance was conventionally statistically significant, indicating the presence of substantial genetic variance (Table 2; Appendix S1). Broad‐sense heritabilities ranged from 0.10 (95% CI 0.06–0.22) for SWC_STEM to 0.37 (95% CI 0.26–0.61) for T _FLOWER (Table 2). Succulence measures varied nearly 4‐fold in heritability, suggesting somewhat independent inheritance.

TABLE 2.

Analysis of genetic variance among 14 recombinant inbred lines for 7 water‐relations traits. Chi‐square and associated P values are the results for the random RIL effect in these models.

Trait	${\mathit{\chi }}_{13}^2$ for RIL effect	P	Broad‐sense heritability (V RIL/V P) (95% CIs)
T GERM	43.23	4 × 10‐5	0.16 (0.51, 0.33)
T FLOWER	112.33	2 × 10‐16	0.37 (0.26, 0.61)
SWCPV	41.66	7 × 10‐5	0.18 (0.11, 0.36)
SWCSTEM	27.34	0.011	0.09 (0.06, 0.22)
SWCLEAF	91.970	2 × 10‐13	0.35 (0.23, 0.59)
Ѱ TLP	27.50	0.01063	0.12 (0.07, 0.26)
π o	37.59	0.00033	0.18 (0.11, 0.36)

Six of 21 pairwise genetic correlations were strong enough to be conventionally statistically significant (Figs. 2, 3). Strong negative correlations included those between T _FLOWER and both SWC_LEAF and SWC_PV, between Ψ _TLP and both SWC_STEM and SWC_PV, and between T _GERM and π _o (Figs. 2, 3). Succulence variables correlated positively with each other, significantly so for SWC_STEM and SWC_PV. In contrast, the two water potential variables (Ψ _TLP and π _o) were not correlated (r = 0.19, P = 0.518). For the significant correlations between succulence variables and between the succulence variables and flowering time, parent means fell more or less along the genetic correlations among RIL means (Fig. 3). For the two succulence–Ψ _TLP relationships, however, the subspecies differences were orthogonal to the negative genetic correlations; xantiana had an exceptionally low Ψ _TLP for its succulence (Fig. 3C, E). A less‐dramatic exception is the negative genetic correlation between T _GERM and π _o; parviflora (but also one RIL) had unusually late germination for its osmotic potential at full turgor (Fig. 3).

FIGURE 2.

Matrix of pairwise correlation coefficients (Pearson’s r) among recombinant inbred line means. Coefficients for which P < 0.05 (i.e., for df = 12,│r│ > 0.532) include color‐ and shape‐coded ellipses.

FIGURE 3.

Scatterplots of significant genetic correlations (P < 0.05) among recombinant inbred line means, with 95% confidence ellipses. Locations of parent means appear as “x” and “p” inside boxes.

The PCA of RIL means revealed two components that explained approximately 69% of the variation (Table 3; Fig. 4). The first component (PC1, accounting for 42.1%) mainly indexes T _FLOWER from late to early, SWC_LEAF and SWC_STEM from low to high, and Ψ _TLP from high to low (Table 3; Fig. 4). The second component (PC2, accounting for 27%) mainly indexes T _GERM from slow to fast and π _o from low to high (Table 3; Fig. 4). In terms of drought adaptation mechanisms, high values of PC1 reflect drought escape (early flowering) plus components of dehydration resistance (high SWC) and dehydration tolerance (Ψ _TLP). Low values of PC2 represent genotypes that combine slow T _GERM (drought avoidance) and low π _o (dehydration tolerance). Plotting parent populations in the RIL‐mean PC1–PC2 space shows parviflora as having a high value along the PC1 axis and near zero on PC2, reflecting high succulence and early flowering (and, to some extent, slow germination), despite parviflora having higher Ψ _TLP than xantiana (Fig. 4). Meanwhile, xantiana positions near zero on PC1 but low on PC2, despite xantiana having somewhat more rapid germination than parviflora (Fig. 4).

TABLE 3.

Trait loadings for the principal component analysis of 14 recombinant inbred lines.

Trait	Loading on PC1	Loading on PC2
T GERM	0.173	–0.662
T FLOWER	–0.489	–0.227
SWCSTEM	0.514	0.107
SWCLEAF	0.445	0.161
Ѱ TLP	–0.506	0.090
π o	–0.114	0.682

FIGURE 4.

Principal components of recombinant inbred line means of six phenological and physiological variables, with vectors indicating trait loadings. Points represent recombinant inbred lines. Subspecies positions, plotted in this PC1–PC2 space, appear as “x” and “p” inside boxes.

DISCUSSION

Evolutionary plant physiology can provide an integrative lens to examine mechanisms, possible trade‐offs, and potential plant responses to climate change. Focusing this lens on evolutionary divergence of possible drought adaptations in Clarkia xantiana subspecies, we discovered that the subspecies differ in several traits that may represent adaptations to their contrasting environments and that genetic correlations among traits may influence evolutionary trajectories. Our study is one of the first to estimate pressure–volume traits in wild herbaceous plants, and the first, to our knowledge, to estimate their genetic architecture. In this system, genetic architecture generally does not seem to have constrained the evolution of subspecies differences in these traits. In some cases, genetic correlations likely facilitated the evolution of the combinations of water‐relations traits the subspecies possess. With respect to tissue succulence and turgor loss point, subspecies’ divergence seems to have overcome genetic correlations that lie in the opposite direction as subspecies differences. Below we discuss the implications of the subspecies differences in these and other water‐relations traits, and the evolutionary implications of genetic variation and covariation that underlies these differences, noting some future studies of the physiology and evolution of drought adaptation likely to be informative.

Subspecies differences

This research adds depth and breadth to the evidence that C. xantiana subspecies differ distinctly in phenology and physiology. Previous research (Vasek, 1968; Runions and Geber, 2000; Mazer et al., 2004) documented the subspecies difference in flowering time also reported here. There also is some evidence that, like other drought escapers (Geber and Dawson, 1990; Heschel and Riginos, 2005; Emms et al., 2018), parviflora sacrifices long‐term water‐use efficiency for higher rates of gas exchange, perhaps an inevitable trade‐off (Mazer et al., 2010). Here we found that parviflora also has higher tissue osmotic potential at full turgor (π ₀) and higher turgor loss point (Ψ _TLP) than xantiana, indicating that xantiana has superior ability to tolerate dehydration. By flowering later, xantiana may experience stronger selection than parviflora for conservative water‐use and dehydration tolerance. Subspecies parviflora does not appear, however, to sacrifice other potential drought adaptations, as it has slower seed germination than xantiana on a saturated substrate, and it is more succulent than xantiana at the flowering stage. Longer time requirements for germination may confer drought avoidance, reducing the likelihood of germination in places and times where soil water is insufficient to complete a life cycle (Duncan et al., 2019a) and increasing the likelihood of establishing persistent seed banks that spread the risk of reproductive failure due to terminal drought (Saatkamp et al., 2011). Succulence may allow parviflora to delay, for a time, the effects of soil water deficit.

It is premature to conclude that these subspecies differences all represent adaptive individual traits or sets of traits. The questions of whether germination timing and succulence have the above hypothesized benefits in parviflora’s native environments and whether xantiana’s lower turgor loss point is adaptive because it flowers late, call for additional studies, such as phenotypic selection analysis (cf. Kimball et al., 2013). It also would be valuable to assess whether parviflora and xantiana differ in bet‐hedging seed dormancy and the demographic correlates of seed germination found across phylogenetically diverse sets of winter annuals (Huang et al., 2016).

Genetic variation and evolution

The RILs of a cross between the subspecies contained substantial genetic variance and covariance in water relations and phenology traits. We acknowledge some caveats in interpreting quantitative genetic variables in this study. First, as noted above, the number of RILs was small, for the sake of feasibility. The small sample reduced statistical power, and though the sample should not have been biased, it likely underestimated the range of possible genotypes and phenotypes in a hybrid population. Nevertheless, the RILs did display transgressive segregation: wider ranges in trait values of hybrids than parents (Rieseberg et al., 1999; Devi et al., 2009; Shekoofa et al., 2013; Polania et al., 2017). Second, broad‐sense heritabilities and genetic correlations estimated from RILs of a subspecies cross cannot (and are not intended to) predict the precise evolutionary potentials or trajectories of these traits in any particular contemporary population. As our estimates are of broad‐sense genetic variation, if we were to apply them directly to natural populations, they would be more appropriate for populations of the self‐fertilizing parviflora than the outcrossing xantiana.

The quantitative genetic data here do, however, provide a view of the genetic variation that underlies subspecies differentiation. We interpret this variation, cautiously, as representing potential directions of drought adaptation that might be evolutionarily independent, facilitated, or constrained. If selection acts on a pair of traits in the same direction as their genetic correlation, there will be both direct and correlated evolutionary responses, facilitating adaptation; in contrast, evolutionary responses may be constrained when selection acts orthogonally to a genetic correlation (Antonovics, 1976). Sufficiently strong selection, which can happen with large changes in drought timing (Franks et al., 2007; Groen et al., 2020), can overcome constraints (Conner et al., 2011), and such responses can be facilitated by restricted recombination that preserves favorable trait combinations (Lowry and Willis, 2010; Lowry et al., 2019).

Because genetic correlations among RILs represent mostly genetic covariance due to pleiotropy (Hegmann and Possidente, 1981), they cannot detect genetic covariance due to linkage disequilibrium, which may occur in natural populations due to correlated selection and/or to inbreeding (Lynch and Walsh, 1998). If the subspecies have diverged in ways that do not fall along genetic correlations in the RIL population, one or more of these evolutionary and genetic mechanisms (strong selection that overcame constraints; linkage disequilibrium that modified genetic correlations or preserved favorable, possibly epistatic combinations of alleles among loci) may be responsible. For example, Lowry and Willis (2010) demonstrated the importance of chromosomal inversions in Mimulus guttatus, finding that an inversion contributes to local adaptation, life‐history differences, and ecological reproductive isolation. Inversion orientation affects gibberellic acid biosynthesis and activity, with numerous effects on phenotypes and adaptation (Lowry et al., 2019).

Interpreting the genetic correlations in this way and interpreting drought adaptations as in Table 1, we did not find evidence of strong genetic constraints in this trait set. Instead, the evolution of several drought adaptation mechanisms appears to be independent and able to evolve freely (T _GERM and T _FLOWER; T _GERM and succulence; T _FLOWER, Ψ _TLP, and π _o). Other combinations of drought adaptations might be facilitated by genetic correlations. For example, mutually reinforcing responses would be expected if there were selection for early flowering (drought escape) plus high succulence (dehydration resistance)—a combination that appears in parviflora—or if there were selection for low π ₀ (dehydration tolerance) and slow germination (drought avoidance). Early flowering is often accompanied by low water‐use efficiency (low drought tolerance) (Geber and Dawson, 1990; McKay et al., 2003; Heschel and Riginos, 2005; Mazer et al., 2010), although there is some conflicting evidence (Sherrard and Maherali, 2006; Donovan et al., 2007; Wu et al., 2010).

By summarizing genetic covariance structure among multiple traits, the PCA enabled a broader view, though still a tentative one. PC1 suggests the potential for facilitated evolution of drought escape by early flowering, plus drought tolerance by succulence and low turgor loss point. PC2 seems to capture components of drought avoidance (delayed germination) and drought tolerance (low osmotic potential at full turgor). In this way, joint responses to selection for drought avoidance by slow germination plus dehydration tolerance by low osmotic potential at full turgor seem likely to be facilitated. Though genetic trade‐offs among drought‐adaptation mechanisms clearly occur (e.g., McKay et al., 2003; Franks and Weis, 2008), the present study shows that evolutionary independence and even facilitation also are possible.

When genetic correlations do occur, insight into their causes can be informed by further genetic analysis. McKay et al. (2003) found a strong positive genetic correlation between δ¹³C (a proxy of long‐term water‐use efficiency) and time to flowering among 39 naturally occurring genotypes from across the climatic range of Arabidopsis thaliana. The positive genetic correlation arose in part from allelic variation at two loci, FLC and FRI, known to affect flowering time. Subsequent work confirmed the importance of those alleles and their pleiotropic effects in drought adaptation (Lovell et al., 2013).

One finding in the present study presents a particular puzzle. Subspecies have diverged in a way that suggests a trade‐off between high SWC and low Ψ _TLP, but the underlying genetic variation suggests facilitated evolution of that trait combination. It is plausible that in annual plant populations experiencing drying soil as rainfall declines and evaporative demand increases, selection would favor both dehydration delay by high SWC and dehydration tolerance by low Ψ _TLP. These traits correlate genetically, in this direction, yet xantiana deviates substantially, having a combination of exceptionally low Ψ _TLP and low SWC. On the other hand, xantiana’s SWC has evolved to a position that fits along the genetic correlation between that trait and T _FLOWER. This outcome might reflect strong selection and/or antagonistic genetic correlations with other traits. We speculate that strong selection for low Ψ _TLP in xantiana led to the evolution of genotypes that are dehydration tolerant but still follow the genetic correlation between late flowering and low SWC. Perhaps that correlation was harder to overcome, developmentally or genetically, and/or perhaps SWC is less important to fitness than flowering time and turgor loss point.

The environments in our study also suggest that we be cautious in our interpretations. The shared, well‐watered environments in which we did this research sacrificed realism for simplicity. Estimating succulence and pressure–volume traits during soil drydown may reveal constraints that we did not detect. Common, more or less stable environments also minimized the potential for important phenology–environment interactions that can happen in nature. In seasonal environments, changes in the timing of life‐cycle events also change the environmental conditions experienced during those life stages, in turn affecting the expression of those traits (Donohue et al., 2010). For example, when it is transplanted to parviflora’s exclusive, more arid range, xantiana’s flowers take longer to develop from visible buds to anthesis than they do in their native range, apparently because plants experience more severe water stress; consequently flowering occurs even later than would be expected in a constant environment (Eckhart et al., 2004). Because phenology, measured here as times to germination and flowering, generally alters environmental experiences in this way, phenology shifts can lead to pleiotropic changes in other traits, including drought adaptations (Monroe et al., 2018; Auge et al., 2019; Takou et al., 2019).

Finally, we point out another finding about succulence that deserves further investigation. Although SWC_LEAF and SWC_STEM loaded similarly along PC1 in the RIL‐mean analysis, genetic correlations between these traits were not particularly strong, and within xantiana succulence varied considerably among organs. It is possible that the succulence of different organs can evolve more or less independently (Eggli and Nyffeler, 2009; Feng et al., 2019). Feng et al. (2019) demonstrated in a pair of sympatric sister species of Primulina that floral and vegetative traits differ in their underlying genetic structure, suggesting that suites of traits related to these organs are unlinked. Moreover, it is then possible for different selective pressures to act on different organs. Other recent evidence examines floral and vegetative trait differences across 22 species, showing that flowers exhibit different and more diverse hydraulic traits than vegetative tissues (Roddy et al., 2019). Specifically, Roddy et al. (2019) notes that succulence is higher for flowers than vegetative organs, proposing that flowers do not rely as heavily on stomatal transpiration to maintain water balance as do vegetative organs. Functionally, this variation in succulence between plant organs represents a form of hydraulic segmentation, where organs are differentially important to the water continuum (Tyree et al., 1993; Pivovaroff et al., 2014; Johnson et al., 2016). Stem succulence may play an important role in drought tolerance (Hasanuzzaman et al., 2017). The fact that C. xantiana stem xylem is unusual in containing gelatinous fibers suggests anatomical specialization for water storage for terminal drought (Carlquist, 1975, 2014).

CONCLUSIONS

Since two lineages of C. xantiana separated approximately 65,000 years ago, the subspecies have evolved to differ not only in their potential for drought escape by early flowering but also in other water‐relations traits, including germination timing and tissue water relations, physiological traits that have received less attention than in most similar studies. The genetic architecture underlying subspecies differences suggests that much of this divergence occurred independent of or facilitated by genetic correlations. Further work should address the quantitative contributions to drought adaptation of traits besides flowering time, possible mechanisms of pleiotropy with flowering‐time change, the reasons for contrasting directions between subspecies differences and negative genetic correlations between succulence and turgor loss point, and the causes and consequences of organ‐specific succulence.

AUTHOR CONTRIBUTIONS

T.E.B and V.M.E. conceived the study and designed the experiment. T.E.B. set up the experiment and collected the data. T.E.B. analyzed the data, with V.M.E.’s assistance. T.E.B. completed the first draft of the manuscript, and V.M.E. and T.E.B. edited and revised subsequent versions.

Supporting information

APPENDIX S1. Figures of trait variation among recombinant inbred lines (RILs).

Burnette, T. E. and Eckhart V. M.. 2021. Evolutionary divergence of potential drought adaptations between two subspecies of an annual plant: Are trait combinations facilitated, independent, or constrained?. American Journal of Botany 108(2): 309–319.

Data Availability

Data available from the Dryad Digital Repository: doi.org/10.5061/dryad.k3j9kd55k (Burnette and Eckhart, 2020).