Natural quantitative genetic variance in plant growth differs in response to ecologically relevant temperature heterogeneity

www.ecolevol.org Ecology and Evolution 2016; 6: 7574–75857574 | © 2016 The Authors. Ecology and Evolution published by John Wiley & Sons Ltd. Received: 27 April 2016 | Revised: 22 July 2016 | Accepted: 28 August 2016 DOI: 10.1002/ece3.2482 Abstract Adaptation to largescale spatial heterogeneity in the environment accounts for a major proportion of genetic diversity within species. Theory predicts the erosion of adaptive genetic variation on a withinpopulation level, but considerable genetic diversity is often found locally. Genetic diversity could be expected to be maintained within populations in temporally or spatially variable conditions if genotypic rank orders vary across contrasting microenvironmental settings. Taking advantage of fineresolution environmental data, we tested the hypothesis that temperature heterogeneity among years could be one factor maintaining quantitative genetic diversity within a natural and genetically diverse plant population. We sampled maternal families of Boechera stricta, an Arabidopsis thaliana relative, at one location in the central Rocky Mountains and grew them in three treatments that, based on records from an adjacent weather station, simulated hourly temperature changes at the native site during three summers with differing mean temperatures. Treatment had a significant effect on all traits, with 2–3fold increase in aboveand belowground biomass and the highest allocation to roots observed in the treatment simulating the warmest summer on record at the site. Treatment affected bivariate associations between traits, with the weakest correlation between aboveand belowground biomass in the warmest treatment. The magnitude of quantitative genetic variation for all traits differed across treatments: Genetic variance of biomass was 0 in the warmest treatment, while highly significant diversity was found in average conditions, resulting in broadsense heritability of 0.31. Significant genotype × environment interactions across all treatments were found only in roottoshoot ratio. Therefore, temperature variation among summers appears unlikely to account for the observed levels of local genetic variation in size in this perennial species, but may influence family rank order in growth allocation. Our results indicate that natural environmental fluctuations can have a large impact on the magnitude of withinpopulation quantitative genetic variance. KEYWORDS environmental heterogeneity, genotype × environment interaction, maintenance of genetic diversity, temperature, temporal heterogeneity 1Department of Botany, University of Wyoming, Laramie, WY, USA 2Program in Ecology, University of Wyoming, Laramie, WY, USA 3Department of Molecular Biology, University of Wyoming, Laramie, WY, USA Correspondence Matti J. Salmela, Department of Botany, University of Wyoming, Laramie, Wyoming, USA. Email: [email protected]om Funding information National Science Foundation, Grant/Award Number: IOS-1025965 Current address Matti J. Salmela, Natural Resources Institute Finland, Vantaa, Finland ORIGINAL RESEARCH Natural quantitative genetic variance in plant growth differs in response to ecologically relevant temperature heterogeneity Matti J. Salmela1 | Brent E. Ewers1,2 | Cynthia Weinig1,2,3 This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. | 7575 Salmela et al. 1 | INTRODUCTION In the wild, native populations often exhibit a fitness advantage in their home environments. This phenomenon, local adaptation, is frequently documented among populations found across heterogeneous environments and is known to maintain genetic diversity in fitness and many other quantitative traits among populations (Hereford, 2009; Leimu & Fischer, 2008). On a more local withinpopulation level, however, natural selection is expected to reduce genetic variation as nonoptimal phenotypes are removed from populations (Kawecki & Ebert, 2004). In contrast to these predictions, substantial quantitative genetic variation is frequently found within populations, even in traits closely related to fitness that are assumed to experience strong selection (Barton & Keightley, 2002; Hill, 2010; Houle, 1992; Johnson & Barton, 2005). For example, flowering time exhibits genetic diversity not just among but also within populations in the mainly selffertilizing model species Arabidopsis thaliana (L.) Heynh. in Fennoscandia and Spain (Kuittinen, Mattila, & Savolainen, 1997; MéndezVigo, Gomaa, AlonsoBlanco, & Pico, 2013; Stenøien, Fenster, Tonteri, & Savolainen, 2005), while variation in timing of growth is often distributed in a similar fashion in highly outcrossing forest trees (Alberto et al., 2011; Mimura & Aitken, 2007; Savolainen, Bokma, GarcíaGil, Komulainen, & Repo, 2004). In spite of extensive theoretical work on the potential of different evolutionary and ecological forces to maintain genetic variation within populations, the causes of such local diversity remain for the most part poorly understood (Johnson & Barton, 2005). Mutationselection balance on its own is considered unlikely to account for the observed high levels of genetic diversity, while various forms of balancing selection have been hypothesized to play a causal role (Barton & Keightley, 2002; Charlesworth, 2015; Hill, 2010; Johnson & Barton, 2005). Further, gene flow among genetically distinct populations from divergent environments may be a source of variation within populations (Barton, 1999). Because a population’s evolutionary potential in a changing environment is determined largely by the magnitude of heritable genetic diversity in adaptive traits (Falconer & Mackay, 1996), the maintenance of genetic variation is relevant not only to basic research on evolutionary processes but also to applied conservation biology and rare species’ management (Hoffmann & Sgrò, 2011). Indeed, empirical evidence that at least some populations in the wild have been able to adapt to environmental changes indicates that adaptive genetic diversity existed in previous generations (Franks, Weber, & Aitken, 2014; Merilä & Hendry, 2014; see also Geerts et al., 2015; Irwin, Finkel, MüllerKarger, & Ghinaglia, 2015). In addition to largescale spatial variation in the environment that often leads to genetically differentiated populations, environmental conditions may also vary at much finer spatial and temporal scales (Linhart & Grant, 1996). Consequently, a population may encounter variable conditions at its home site, which could select for the potential to express a wider range of phenotypes across distinct microenvironments. Such phenotypic variation could be achieved for instance by genetic diversity such that different genotypes have the highest fitness in differing local environmental settings, that is, by genotype × environment interactions (Ellner & Hairston, 1994; Gillespie & Turelli, 1989; Hedrick, 1995), or via phenotypic plasticity of a generalist genotype (Kawecki & Ebert, 2004). Empirical field studies have shown that performance ranks of genotypes may change as a result of variation in interand intraspecific competition (Baron, Richirt, Villoutreix, Amsellem, & Roux, 2015; Shaw, Platenkamp, Shaw, & Podolsky, 1995) or disturbance (McLeod, Scascitelli, & Vellend, 2012); genotype × environment interactions in fitness have even been described on a scale of just 10 cm within a single old field (Stratton, 1994). In Betula pendula Roth in Finland, however, forest ground heterogeneity on a local level affected overall growth but was not sufficient to shift genotypic ranks (Mikola et al., 2014). Genetic mapping studies on recombinant inbred lines and experimental hybrids of model systems such as Arabidopsis, Boechera, Drosophila, and Mimulus in the field and under controlled experimental conditions have provided insights into the genetic basis of genotype × environment interactions across highly divergent environmental conditions. These studies commonly show environmentspecific QTL effects for numerous quantitative traits from morphology to reproductive fitness (reviewed in Mackay, 2009; Savolainen, Lascoux, & Merilä, 2013; Weinig, Ewers, & Welch, 2014). Genetically variable populations of predominantly selffertilizing plant species like the mustard Boechera stricta (Graham) AlShehbaz (Song, Clauss, Pepper, & MitchellOlds, 2006) are suitable study systems for exploring the maintenance of local genetic diversity in varying environments because longdistance gene flow may be reasonably excluded as the source of high levels of withinpopulation adaptive genetic diversity (cf. Yeaman & Jarvis, 2006). Previously, Salmela et al. (2016) observed considerable quantitative genetic diversity in the circadian clock, the endogenous timekeeper regulating daily oscillations in numerous traits, and various growth traits among maternal families of B. stricta sampled at a highelevation site in the central Rocky Mountains in the USA. In this population, the range of family means in circadian period, one characteristic of the circadian clock, accounted for over 50% of the range that has previously been reported in a global set of 150 A. thaliana genotypes by Michael et al. (2003). Quantitative variation in the clock was associated with growth so that families with longer circadian periods grew more rapidly and to a larger size, but had a lower roottoshoot ratio. Because the circadian clock and its variability are generally thought to reflect adaptation to variable light–dark cycles (e.g., Michael et al., 2003), Salmela et al. (2016) hypothesized that local genetic diversity in the clock could be maintained by performance tradeoffs of families across seasonal environments with differing photoperiods and temperatures. Yet, growth chamber environments simulating different months within a growing season resulted in genotype × environment interactions that were larger than the overall genetic effect only for roottoshoot allocation. Higher roottoshoot ratio may enhance survival in stressful and lowresource environments (Poorter et al., 2012; Wilson, 1988), an adaptive hypothesis also suggested by the larger roottoshoot ratios found in populations from higherelevation environments in this region. Still, the exact factors enabling the maintenance of significant finescale genetic diversity in this population remain unknown. Fluctuating selection driven by temporal variation in environmental factors has been proposed to be common in the wild (Siepielski, 7576 | Salmela et al. DiBattista, & Carlson, 2009). Temporally variable environmental conditions have also been hypothesized to shape the patterns of adaptive genetic diversity within populations (Salmela, 2014), but in general, theoretical studies suggest that conditions under which temporal variation in the environment can maintain genetic diversity are more restrictive than those for spatial heterogeneity (Hedrick, 2006). Some evidence supports the role of temporal environmental heterogeneity in maintaining genetic variation within populations: Different growth seasons within a year have been reported to favor divergent genotypes in Taraxacum officinale (L.) Weber ex F.H. Wigg (Vavrek, McGraw, & Yang, 1996), while annually varying moisture conditions in a desert may conserve flower color polymorphism in the annual Linanthus parryae (A.Gray) Greene (Schemske & Bierzychudek, 2001). In the fish Xiphophorus variatus, the level of temperature variability within days and years has been found to be positively associated with withinpopulation diversity in tail spot richness (Culumber & Tobler, 2016), with the fitness rank of different spot types depending for instance on the thermal environment (Culumber, Schumer, Monks, & Tobler, 2015). In this study, we were interested in further exploring the capacity of temporal abiotic variability to sustain local quantitative genetic variation in the wild using the aforementioned B. stricta population from the central Rocky Mountains as our study system. Two features in particular make this population an interesting model for assessing the possible environmental causes of local genetic diversity: It has already been shown to be genetically diverse in multiple quantitative traits (Salmela et al., 2016), and its home site is located close to a weather station with longterm and fineresolution temperature records. The population experiences for instance variable summer temperature conditions; between 1995 and 2014, mean June and July temperatures at the site varied between 8.3–15°C and 13–17.2°C, respectively. Hourly recorded temperature data from multiple years allowed us to investigate the effects of heterogeneity in a single abiotic factor, providing the unique opportunity to simulate temporally fluctuating and ecologically relevant natural environments in controlled experimental settings. We grew replicates of the same set of naturally occurring maternal families across treatments with differing average temperatures, hypothesizing that if variable summer temperature regimes were capable of preserving quantitative genetic diversity in this population, not only would the treatments affect overall means but also induce genotype × environment interactions in growth and its allocation. Moreover, we expected the magnitude of these interactions to be large compared to the average effect of family and a large proportion of the interaction variance to be due to rank shifts of families among simulated growing season temperature cycles. 2 | MATERIALS AND METHODS 2.1 | Sampling We collected seed material for the study by maternal family at a distance of about 300–500 m from the South Brush Creek SNOTEL weather station (41.333°N, 106.500°W; elevation 2,572 m) in southeastern Wyoming on 26 July 2012. The inbreeding coefficient in populations of B. stricta is high (0.74–0.98 in Song et al., 2006), suggesting that maternal progeny consist mostly of fullsibs. All except one of the sampled plants were located along a transect of approximately 300 m. Because we were interested in natural patterns of variation present in the population at the time of sampling rather than characterizing the genetic background of quantitative trait variation, we used wildcollected seeds in the experiment. Of the 21 families investigated in this study, 19 were included in the study by Salmela et al. (2016) that documented significant withinpopulation genetic variation in the circadian clock and growth. 2.2 | Treatments Based on records from the weather station (http://www.wcc. nrcs.usda.gov/nwcc/site?sitenum=772), we programmed growth chamber compartments (PGC9/2 with Percival Advanced Intellus Environmental Controller, Percival Scientific, Perry, IN, USA) to track hourly temperature changes at the site during an 8week period on June 1–July 25 in 2009, 2010, and 2012. Thus, each treatment consisted of over 1,300 temperature steps. In the following sections of the article, we will refer to these treatments by the corresponding year. We chose to focus on June and July because an earlier study indicated the most vigorous growth in a treatment simulating early growing season conditions (Salmela et al., 2016) and because in May the site regularly experiences freezing temperatures as low as −15°C that cannot be achieved in the chambers used. Daily mean temperatures in each treatment are shown in Figure 1. These treatments represented a range of mean June temperatures at the site (10.2°C in 2009, 12.0°C in 2010, and 15.0°C in 2012) during a period from which hourly recorded temperature data were available; the summer of 2012 is the hottest on record at the site and in the rest of Wyoming (NOAA National Overview for Annual 2012). Due to lack of hourly data, we did not simulate the coolest June on record (1998, with average temperature of 8.3°C). Average daily maximum and minimum temperatures in June were 17.4°C and 3.6°C in 2009, 19.6°C and 4.2°C in 2010, and 23.4°C and 5.1°C in 2012. Under chamber conditions, mean July temperatures were 14.7°C in both 2009 and 2010, and 16.3°C in 2012. This resulted in overall treatment means of approximately 12.5°C in 2009, 13.4°C in 2010, and 15.7°C in 2012. Due to chamber restrictions on minimum temperature (4°C in the dark, 10°C with all lights on), overall experimental conditions were slightly (0.09–0.23°C) higher than those in the field. Photoperiod was approximately 15 hr, with gradual changes in a similar fashion in all treatments according to local sunrise and sunset times over the course of the 8week period. To simulate dawn and dusk, 50% of the lights were on during the first and last hour of the photoperiod. Maximum photosynthetically active irradiance, measured with the light meter LI250 (LICOR Biosciences, Lincoln, NE, USA) in all chambers before the experiment, was approximately 250 μmol photons m−2 s−1 at the plant level. We did not replicate treatments; thus, family × treatment effects cannot be unequivocally attributed to temperature. However, variation in other controllable factors is likely to have been small compared to temperature due to very similar light levels in all chamber compartments. Daily monitoring of temperature | 7577 Salmela et al. by thermometers also indicated that chamber conditions adhered to programmed settings. We germinated and planted seeds from the 21 families, with ten replicates per family in each treatment, following the protocol in Salmela et al. (2016). We did not score timing of germination due to rapid germination on moist paper in all the families within 4 days following cold stratification. We divided the seedlings into the three treatments, with ten blocks within each and one replicate per family randomized within a block. We watered pots to field capacity every 2 days and rotated flats within the chamber compartments twice every week. The species requires vernalization in order to flower; due to difficulties in simulating ecologically relevant highelevation autumn, winter, and spring conditions in our experimental settings (air temperature at the site can drop below −30°C), we measured the equivalent of firstyear growth only. After 8 weeks, we scored the number of leaves and the length of the longest leaf in all replicates and sampled whole plants for aboveand belowground biomass and roottoshoot ratio. We dried plants in an oven at 65°C for 3 days before biomass measurements. Leaf measures were strongly correlated with aboveground biomass (the number of leaves: r ≥ .807, p < .0001 in all treatments; the longest leaf length vs. √aboveground biomass: r ≥ .923, p < .0001 in all treatments). Because qualitatively similar results were obtained with all correlated measures, we will use biomass only as a measure of aboveground plant growth. We sampled plants at 8 weeks because the oldest leaves often start to senesce at this point in chamber conditions (M.J. Salmela, personal observation) and because an 8week period is likely to account for a large proportion of the annual growing season at the home site of the population. 2.3 | Statistical analyses We included families with at least four replicates in two treatments in the analysis. Prior to formal statistical tests, we examined whether our data met assumptions of analysis of variance. We detected variance heterogeneity of residuals among treatments in all traits, with variances varying 6.6fold for aboveground biomass, 9.5fold for belowground biomass, and 4.8fold for roottoshoot ratio. These ranges of variation resulted from the 2012 treatment that had the most phenotypic variation in all traits; variance differences between 2009 and 2010 were less than twofold. To determine the influence of unequal variances, we ran all statistical analyses with the complete dataset using both original and transformed measurements. While transformations reduced variance heterogeneity, they did not change the outcome of the analyses or interpretation of the results. This outcome probably reflects the robustness of analysis of variance to unequal variances in the case of similar sample sizes in all treatments. Therefore, we used the raw values in the analyses presented here. We first tested for the significant effects of family and block within each treatment using general linear models. We consider variation among maternal families to represent genetic diversity within the population, and thus, proportion of total variation explained by amongfamily diversity in each treatment is an estimate of broadsense heritability (H2). We estimated variance components for family, block, and residual variation using the REML approach. We examined associations between the traits within treatments using replicate values and Pearson’s correlation. Scatterplots suggested the slope between aboveand belowground biomass varies among treatments. To test for treatment differences in slope (ß1), we performed a regression analysis of aboveground biomass on belowground biomass in which we included treatment as a categorical factor and the aboveground biomass × treatment interaction term. To investigate the contributions of different sources of variation across all three treatments and between pairs of treatments, we used the model aboveor belowground biomass or roottoshoot ratio = treatment + family + family × treatment + block(treatment). We considered treatment a fixed factor. When nonsignificant at the FIGURE1 Variation in daily mean temperatures in growth chamber treatments simulating three summers with differing average temperatures at South Brush Creek in southeastern Wyoming. Overall treatment means were 12.5°C in 2009, 13.4°C in 2010, and 15.7°C in 2012. Due to the hourly tracking of temperature heterogeneity at the site, each treatment consisted of over 1,300 different temperature steps 7578 | Salmela et al. 0.10 level, we removed the interaction term from the analysis. Given that amongfamily variation measures genetic diversity within the population, the family × treatment term defines genotype × environment interactions. We estimated variance components for the random factors using the REML approach. Genotype × environment interactions can maintain diversity if they are caused by rank shifts of genotypes rather than by differences in amonggenotype variance among treatments (MitchellOlds, 1992). In order to separate the effects of shifts in family rank order vs. amongfamily variance on a statistically significant interaction variance component, we used the equation (Cockerham, 1963): where VFamily × treatment is the variance component due to family × treatment interaction, σi and σj are the square roots of amongfamily variances in treatments i and j, rGE is the genetic correlation between treatments i and j, and t is the number of treatments. The first part of the equation accounts for variance due to changes in rank order among treatments, while the second part accounts for treatment differences in amongfamily variance. We estimated genetic correlations using family variance components within each treatment and across pairs of treatments (Windig, 1997): where VFamily is the variance component due to family across the two environments, and VFamily, treatment i and VFamily, treatment j are the variance components due to family in treatments i and j. We carried out all statistical analyses with IBM SPSS Statistics Version 23. 3 | RESULTS Table 1 shows general linear model results for the three traits within each treatment, along with treatment means and measures V Family ×treatment = (∑ i<j [2σiσj(1−rGE)+(σi−σj)2] ) ∕(t(t− 1), r GE =VFamily∕ √( VFamily, treatment i×VFamily, treatment j ), TABLE1 General linear model results for aboveand belowground biomass and roottoshoot ratio in a population of Boechera stricta within each treatment simulating differing growing season temperature conditions at South Brush Creek in southeastern Wyoming Factor df Aboveground biomass Belowground biomass Roottoshoot ratio MS Fratio pvalue % MS Fratio pvalue % MS Fratio pvalue % 2009 Family 18 0.00124 1.84 * 9.96a0.0000557 1.58 .0784 6.53a0.00734 2.18 ** 11.7a Block 9 0.00119 1.76 .0840 4.92 0.0000607 1.72 .0931 4.89 0.0111 3.30 ** 11.9 Residual 105 0.000674 85.1 0.0000352 88.6 0.00336 76.3 Mean (±SE) 0.0557 (±0.00243) g 0.0111 (±0.000547) g 0.202 (±0.00570) VP7.99 × 10−4 4.02 × 10−5 4.43 × 10−3 VF7.96 × 10−5 2.63 × 10−6 5.21 × 10−4 2010 Family 20 0.00460 5.35 **** 31.0a0.000179 4.52 **** 26.2a0.00590 3.24 **** 20.1a Block 9 0.00311 3.61 *** 8.84 0.000163 4.10 *** 11.0 0.00659 3.62 *** 11.4 Residual 147 0.000860 60.2 0.0000396 62.8 0.00182 68.5 Mean (±SE) 0.0573 (±0.00282) g 0.0105 (±0.000593) g 0.174 (±0.00384) VP1.42 × 10−3 6.28 × 10−5 2.66 × 10−3 VF4.41 × 10−4 1.64 × 10−5 5.35 × 10−4 2012 Family 19 0.00469 0.947 ns 0a0.000377 0.957 ns 0a0.0275 3.04 *** 19.5a Block 9 0.000855 0.173 ns 0 0.000200 0.508 ns 00.0298 3.29 ** 11.2 Residual 120 0.00495 100 0.000394 100 0.00906 69.3 Mean (±SE) 0.115 (±0.00561) g 0.0311 (±0.00160) g 0.268 (±0.00934) VP4.69 × 10−3 3.83 × 10−4 1.30 × 10−2 VF0 0 2.52 × 10−3 The percentage shows the proportion of total variation explained by each factor. VP = variance component for phenotypic variation; VF = variance component for amongfamily (genetic) variation. ns = p > .10, *p < .05, **p < .01, ***p < .001, ****p < .0001. aEstimate of broadsense heritability (H2). | 7579 Salmela et al. of phenotypic and genetic diversity. In 2009, the family effect was significant for aboveground biomass and roottoshoot ratio, an indication of genetic variation in the population. For belowground biomass, the family effect was close to significance at the 0.05 level (p = .078). In 2010, the family effect was highly significant for all traits. Although 2012 was the most phenotypically variable treatment, we found no significant genetic variation for aboveor belowground biomass in this year; variation was entirely residual. For roottoshoot ratio, however, family differences in 2012 were highly significant. For aboveground biomass, phenotypic and genetic variation in 2009 were 56.3% and 18.0%, respectively, of those in 2010, resulting in higher broadsense heritability in 2010. The pattern was similar for belowground biomass: Phenotypic and genetic variation in 2009 were 64.0% and 16.0%, respectively, of those in 2010. These differences were not due to the two families in 2010 for which we lacked data in 2009; when they were excluded, we obtained very similar estimates of withintreatment diversity. Similar levels of genetic diversity were expressed in roottoshoot ratio in 2009 and 2010, but broadsense heritability in 2009 was lower due to 66.5% more phenotypic variation in this treatment. We observed the highest phenotypic and genetic variation in roottoshoot ratio in 2012. However, due to increases of similar magnitude in all variance components, broadsense heritability in 2012 was similar to that in 2010. Aboveand belowground biomass were positively correlated in all treatments (Figure 2). This association was the strongest in 2010 (Figure 2A) and the weakest in 2012 (Figure 2B). A significant aboveground biomass × treatment interaction in the regression analysis indicated treatment differences in the slope between the two variables (F2,453 = 4.13, p < .05), with very similar estimates in 2009 and 2010 (2009: ß1 = 0.192, 95% CI: 0.171–0.212; 2010: ß1 = 0.191, 95% CI: 0.178–0.204; Figure 2A) and a steeper one in 2012 (ß1 = 0.233, 95% CI: 0.206–0.260; Figure 2B). Belowground biomass was positively and moderately correlated with roottoshoot ratio in all treatments (2009: r = .470, p < .0001, 95% CI: 0.330–0.605; 2010: r = .592, p < .0001, 95% CI: 0.482–0.689; 2012: r = .569, p < .0001, 95% CI: 0.464–0.664). In 2010, aboveground biomass showed a weak positive correlation with roottoshoot ratio (r = .288, p < .001, 95% CI: 0.153–0.422). In 2009 and 2012, this correlation was not significantly different from zero. Treatment had a highly significant effect on all three traits (Table 2). Aboveground biomass was twice and belowground biomass about three times larger in 2012 than in 2009 or 2010, and thus, also roottoshoot ratio displayed its highest treatment mean in 2012 (Table 1). For aboveand belowground biomass, the only significant random factor across all treatments was family. We found a significant family × treatment interaction term for roottoshoot ratio, revealing variation in phenotypic plasticity among maternal families in the population. The variance component due to the interaction was about 27% bigger (6.49 × 10−4) than that of family (5.10 × 10−4). Approximately 63% of the interaction variance was due to rank changes of families across the treatments (Figure 3A), resulting in very similar magnitude of variance components due to family vs. rank order changes. In accordance with the moderate overall family effect, genetic correlations for roottoshoot ratio between treatments were positive: 0.720 between 2009 and 2010, 0.527 between 2009 and 2012, and 0.521 between 2010 and 2012. Table 3 shows results for pairwise comparisons of treatments. We found highly significant treatment differences in biomass between 2009 and 2012, and 2010 and 2012. Although the overall effect of treatment was not significant in the comparison of 2009 and 2010, we observed a significant family × treatment interaction for aboveground biomass. However, the variance component due to family was 1.73 times bigger than that due to the interaction, signaling that the significant interaction arises mainly from the treatment differences in amongfamily variance (Table 1). For belowground biomass in the same treatment comparison, the pattern was comparable: The family × treatment interaction FIGURE2 Variable associations between aboveand belowground biomass within a population of Boechera stricta sampled at South Brush Creek in southeastern Wyoming and grown in three treatments simulating differing summer temperature conditions at its native site. 2012 differed from the two other treatments in its overall trait means and variances. For clarity, we show data for 2012 in a separate figure. (A) In treatments simulating the summers of 2009 and 2010, the correlations and slopes between the traits were not significantly different from each other. (B) In the treatment simulating the summer of 2012, the correlation between the traits was significantly weaker than in 2010. Also, the slope was significantly steeper (ß1 = 0.233) than the one in 2010 (ß1 = 0.191) 7580 | Salmela et al. was close to significance at the 0.05 level (p = .0880). For roottoshoot ratio, pairwise comparisons of treatments indicated moderate differences in mean between 2009 and 2010, and highly significant differences between 2009 and 2012, and 2010 and 2012. The difference between 2009 and 2010, despite the lack of significant treatment effects in its two components, was likely caused by small but opposite changes in the two biomass measures: Overall aboveground biomass was slightly larger in 2010 than in 2009, while the opposite pattern was found for belowground biomass (Table 1). In the comparison of 2009 and 2010 to 2012, we detected some support for a family × treatment interaction (p = .0941). In the pairwise comparisons of 2009 and 2010 to 2012, we observed more statistical support for family × treatment interactions than for family. Together with variation in the strength of genetic correlations, the results indicate that differing family responses to 2012 in particular account for the significant interaction term in roottoshoot ratio across all treatments. 4 | DISCUSSION In this study, we combined environmental data with quantitative genetic approaches to examine whether temporal environmental heterogeneity could act as one mechanism maintaining genetic variation in growth and its allocation within a plant population experiencing variable conditions at its home site. In order to carry out the experiment in as ecologically relevant chamber conditions as possible, we took advantage of fineresolution temperature data from a weather station that was located in the vicinity of the sampled population. This enabled us to quantify the potential of natural amongyear heterogeneity in growing season temperatures to induce genotype × environment interactions in firstyear growth traits. Selffertilization is expected to reduce genetic diversity and consequently limit longterm survival prospects of populations (reviewed in Wright, Kalisz, & Slotte, 2013). Yet, populations of predominantly selffertilizing plant species like A. thaliana and B. stricta may harbor considerable levels of local quantitative and molecular genetic variation (Baron et al., 2015; Kuittinen et al., 1997; MéndezVigo et al., 2013; Salmela et al., 2016; Siemens, Haugen, Matzner, & Vanasma, 2009; Song et al., 2006; Stenøien et al., 2005). Because the rate at which the population mean for a quantitative trait can change is positively correlated with the amount of genetic diversity in the trait (Falconer & Mackay, 1996; Houle, 1992), understanding the maintenance of genetic variation in the wild has become one of the key questions in modern evolutionary and conservation biology (Franks et al., 2014; TABLE2 General linear model results for aboveand belowground biomass and roottoshoot ratio across all treatments Factor df Aboveground biomass Belowground biomass df Roottoshoot ratio MS Fratio pvalue MS Fratio pvalue MS Fratio pvalue Treatment 2 0.156 87.1 **** 0.0193 128 **** 2 0.344 17.7 **** Family 20 0.00582 2.71 *** 0.000256 1.65 * 20 0.0215 2.29 * Family × treatment 37 0.00950 2.07 *** Block(treatment) 27 0.00177 0.822 ns 0.000151 0.972 ns 27 0.0158 3.45 **** Residual 409 0.00215 0.000155 372 0.00459 ns = p > .10, * p < .05, ** p < .01, *** p < .001, **** p < .0001. Interaction terms were included in the models when significant at the p < .10 level. FIGURE3 (A) Reaction norms of roottoshoot ratio for naturally occurring maternal families of Boechera stricta that were sampled at South Brush Creek in southeastern Wyoming and grown in three treatments simulating summers with differing average temperatures at the population’s home site. (B) The negative association between mean June temperature and precipitation during the month at South Brush Creek between 1995 and 2014 (r = −.737, p < .001) | 7581 Salmela et al. Merilä & Hendry, 2014). Our results revealed not only that there was quantitative genetic variation in the highelevation B. stricta population from the Rocky Mountains but also that the magnitude of such diversity varied greatly among treatments mimicking natural growing season temperature heterogeneity. Genetic diversity in aboveground biomass was the highest under the summer temperatures of 2010, while in the treatment simulating the summer of 2009 that was on average only 0.9°C cooler than 2010, genetic variance had reduced by over 80%. For belowground biomass, we found a qualitatively similar pattern, although in this case the family effect in 2009 was a little weaker. A divergent pattern of genetic diversity emerged in the treatment simulating the warmest summer recorded at the home site of the population: Variation in both aboveand belowground biomass was completely residual in 2012. Due to the divergence in variances, broadsense heritability of biomass varied between 0 and 0.31 across two treatments with a mean temperature difference of only 2.3°C. Heritabilities and genetic variances commonly vary across environments, which complicates predicting responses to selection (Charmantier & Garant, 2005; Falconer & Mackay, 1996). Lower levels of quantitative genetic diversity may be expressed in unfavorable conditions (Charmantier & Garant, 2005; Hoffmann & Merilä, 1999). Thus, relative to the intermediate temperatures in 2010, the cooler growth environment during the month of June 2009 may have contributed to the reduced genetic variability in biomass in this year. However, the substantial biomass increase observed in response to the simulated 2012 conditions indicates that the unusually warm temperatures on their own were not detrimental to our study population. Instead, the lack of quantitative genetic diversity in biomass in this treatment may be due to the atypical high temperatures, or the novel combination of high temperature and moisture that this particular population would be unlikely to encounter in nature. Indeed, one limitation of our study is that although we programmed our chamber treatments to track closely natural temperature fluctuations that had occurred at the site, we investigated the effects of a single environmental variable only. We kept plants wellwatered throughout the experiment, but records from the weather station show that no precipitation occurred at the site in June 2012. The region experienced severe drought during the summer of 2012 (NOAA National Overview for Annual 2012), which would be expected to increase allocation to roots but also limit biomass accumulation in the shoot (Mokany, Raison, & Prokushkin, 2006; Poorter et al., 2012). Therefore, a differing response might have been observed in our experiment had we simulated natural variation in multiple environmental factors simultaneously. In the wild, natural selection operates on total phenotypic variance shaped by both genetic and nongenetic factors (Falconer & Mackay, 1996). Because we were interested in the response of a natural population to the environmental heterogeneity of its native environment, we used wildcollected seeds in the experiment. As a consequence, it TABLE3 General linear model results for pairwise treatment comparisons Factor 2009–2010 2009–2012 2010–2012 df MS Fratio pvalue df MS Fratio pvalue df MS Fratio pvalue Aboveground biomass Treatment 1 0.000417 0.159 ns 1 0.208 164 **** 1 0.244 121 **** Family 20 0.00382 2.62 * 20 0.00386 1.34 ns 20 0.00582 2.12 ** Family × treatment 18 0.00146 1.86 * Block(treatment) 18 0.00215 2.74 *** 18 0.00105 0.362 ns 18 0.00196 0.715 ns Residual 252 0.000783 242 0.00289 286 0.00275 Belowground biomass Treatment 1 0.0000261 0.217 ns 1 0.0244 165 **** 1 0.0317 166 **** Family 20 0.000158 2.77 * 20 0.000224 0.992 ns 20 0.000287 1.42 ns Family × treatment 18 0.0000569 1.51 .0880 Block(treatment) 18 0.000112 2.95 **** 18 0.000137 0.608 ns 18 0.000190 0.935 ns Residual 252 0.0000378 242 0.000226 286 0.000203 Roottoshoot ratio Treatment 1 0.0488 5.31 * 1 0.287 11.5 ** 1 0.687 29.3 **** Family 20 0.00935 2.55 * 20 0.0213 1.62 ns 20 0.0212 1.83 .0949 Family × treatment 18 0.00367 1.49 .0941 17 0.0131 2.05 ** 19 0.0117 2.30 ** Block(treatment) 18 0.00883 3.59 **** 18 0.0204 3.19 **** 18 0.0182 3.59 **** Residual 252 0.00246 225 0.00640 267 0.00507 ns = p > .10, *p < .05, **p < .01, ***p < .001, ****p < .0001. Interaction terms were included in the models when significant at the p < .10 level. 7582 | Salmela et al. is possible that our estimates of genetic variance also include a component due to maternal effects. Variation in the environment may result in differential maternal provisioning during seed development, the effects of which could resemble those caused by segregating genetic diversity (e.g., Bischoff & MüllerSchärer, 2010). Further, maternal effects may be adaptive so that they enhance offspring fitness in conditions similar to those experienced by the parents (Galloway & Etterson, 2007). Drought experienced by parents has been found to enhance drought tolerance of offspring in B. stricta (Alsdurf, Ripley, Matzner, & Siemens, 2013), but the general importance of maternal effects to trait variation in this species is not known. While we cannot rule out potential transgenerational effects on the observed phenotypic variation, it is likely that the seeds used in the experiment matured under similar temperature and moisture conditions because all seed collections took place on the same day and within a range of a few hundred meters at a single location. Regardless of the precise underlying causes, it is noteworthy that in our experiment, the greatest potential for evolutionary responses to selection in biomass existed under intermediate temperatures and that patterns of variation under such conditions were not indicative of those in novel environmental settings caused by higher temperature. One factor that may play a role in the maintenance of multiple phenotypically distinct lineages in a population is environmental heterogeneity on a local scale (Gillespie & Turelli, 1989; Linhart & Grant, 1996; Salmela, 2014). We hypothesized that the existence of finescale genetic variation in growth in our study system could be related to amongyear variation in growing season temperature conditions in the home environment. Size is a frequent fitness proxy in plants, and shoot size has been reported to correlate moderately with reproductive effort in B. stricta (Siemens & Haugen, 2013); thus, changes in rank order for biomass could be expected to influence fitness ranks, too. In comparison with 2009 and 2010, we found that average aboveground biomass doubled and belowground biomass tripled under the unusually warm conditions of 2012. Biomass increase in response to warm temperatures is common in plants, especially when growing in nonlimiting conditions (Lin, Xia, & Wan, 2010; Pregitzer, King, Burton, & Brown, 2000). However, despite the significant treatment and family effects, we observed no significant family × treatment interactions in biomass across the three temperature environments. In the pairwise comparison of 2009 and 2010, the family × treatment interaction terms were significant or close to significant, but the variance components due to family were of greater magnitude than those for the interactions. This mirrors the observed treatment differences in genetic variances with little change in family rank order across temperature regimes. The lack of family × treatment interactions in biomass suggests that temperature variation among growing seasons alone is unlikely to account for the genetic diversity in growth in our study system (MitchellOlds, 1992). This is in accordance with a previous experiment that explored the possibility that photoperiodic and temperature variation within a potential growing season could maintain genetic diversity in growth traits in the same B. stricta population (Salmela et al., 2016). In the earlier study, month within a growing season (June, August, or September) had a large effect on overall growth, but more variation was due to family than family × treatment interactions across the three different months. In these growing season treatments, both photoperiod and temperature varied and the range of variation in temperature among treatments was more extensive than in the current study. Nevertheless, although we did not detect significant family × treatment interactions for biomass, conditional neutrality at the level of maternal family may slow the erosion of genetic variation (cf. Fry, Heinsohn, & Mackay, 1996; Schnee & Thompson, 1984). Specifically, in years similar to our 2009 and 2010 treatments, genetic variation in size would be expressed; but in years similar to 2012, genetic diversity in the population could be masked, and selection might not differentiate among distinct maternal lineages (e.g., Merilä, 1997). It is also possible that finescale genetic diversity in growth in this perennial species is maintained by extensive temporal heterogeneity in low winter, spring, and autumn temperatures that could not be simulated in the chambers used, or by other abiotic or biotic factors, such as fine spatial heterogeneity in soil or competition (Baron et al., 2015; Shaw et al., 1995; Stratton, 1994). Patterns of variation expressed for roottoshoot ratio differed from the two other traits. The trait exhibited significant genetic diversity in all three treatments, and in contrast to biomass, the most genetic variation was expressed in 2012. The two other treatments had very similar levels of genetic variation, but differences in the amount of residual variation resulted in broadsense heritability in 2010 being almost twice as high as in 2009. The significant genetic diversity in this trait in 2012 despite the lack of such variation in its two components may be due a differing association between aboveand belowground biomass in this treatment: The correlation between the traits was weaker and belowground biomass gain per an increase in aboveground biomass larger in 2012 than in 2010. Similar to biomass, we detected the highest overall mean for roottoshoot ratio in 2012. The trait is known to be sensitive to different environmental cues such as light and moisture, and limitations in aboveor belowground resources often result in increased allocation to the corresponding parts of the plant (Poorter & Nagel, 2000). In our experiment, plants grew in wellwatered conditions but variable temperatures. Reduced water uptake by roots in cooler temperatures may underlie slightly increased allocation to roots in 2009 (Poorter et al., 2012). Based on amongspecies patterns of variation, roottoshoot ratio is expected to be negatively correlated with shoot size in herbaceous plants (Poorter et al., 2012; Wilson, 1988; see also Mokany et al., 2006); in our study focusing on a much finer level of withinspecies diversity found in the wild, increases in aboveand belowground biomass in the 2012 treatment also led to higher roottoshoot ratios. Larger allocation to roots in this treatment may be related to the patterns of covariation between temperature and precipitation in the population’s native environment: During a 20year period (1995– 2014), average June temperature and precipitation increment during the month at the site were negatively correlated (r = −.737, p < .001, Figure 3B). During Junes of 2012 and 2013, the 2 years with the highest mean June temperatures, no precipitation was recorded at the site, conditions under which increased allocation to roots would be