1. Climate change is causing the growing season to expand and many plants are flowering earlier. However we know less about whether other components of reproductive phenology are altered or whether these changes in phenology are adaptive.
2. We evaluated reproductive phenology and fitness components for populations of Campanulastrum americanum sampled across an elevation gradient and reciprocally transplanted into common gardens at high and low elevations.
3. The low-elevation planting site had an expanded growing season that induced the advance of bolting, flowering, average flower date, and time to fruit maturity relative to the high-elevation site for transplants. With the exception of flowering initiation, each successive stage of reproduction was advanced more than the previous one, resulting in a compressed phenology in the warmer environment.
4. In contrast, populations from low elevation had a longer reproductive cycle when grown at both sites, with each phenological component extended relative to populations from high elevation. Fruit production indicated populations were locally adapted to elevation, suggesting these differences in phenology are adaptive.
5. Selection on phenological characters was stronger on transplants in the expanded low-elevation growing season, favouring delayed bolting and advanced flowering. Plastic response to the longer growing season was adaptive for flowering time but maladaptive for bolt initiation.
6.Synthesis. The compressed reproductive phenology favoured in the expanded growing season expected under climate change will largely be achieved with adaptive plasticity of individual phenological traits. Traits under selection in the longer growing season were genetically differentiated between populations that currently differ in growing season length, suggesting evolutionary malleability and likely modification of reproductive phenology in response to climate change.
In addition, although there is a wealth of data on earlier flowering in warming climates, we know less about how accelerating the initiation of reproduction may change subsequent reproductive events (Post et al. 2008). Earlier flowering plants have been shown to mature fruit earlier (Peñuelas, Filella & Comas 2002; Post et al. 2008), especially in species that bloom early in the growing season (Sherry et al. 2007). However, it is not known whether this change reflects patterns of maturation of individual fruits or simply an earlier initiation of reproduction. For example, if reproduction is tightly integrated across the life cycle (cf. Pigliucci 2003), the earlier reproductive development induced by warmer spring temperatures would also advance flowering time, flower deployment and fruit maturation, shifting the entire reproductive schedule. However, if later reproductive stages, e.g. flower production and fruit maturation, can respond independently to favourable conditions in the longer growing season, reproductive phenology might expand in warmer climates. Alternatively, later phenological stages may be accelerated in novel, warmer conditions, resulting in a compressed reproductive phenology. Finally, other aspects of the environment, such as water availability, may change with warmer temperatures and also alter reproductive phenology (Giménez-Benavides, Escudero & Iriondo 2007; Jentsch et al. 2009). A focus on individual elements of reproductive phenology is needed to understand the duration of reproduction as well as potential fitness consequences of an altered reproductive schedule.
Studies across elevations provide particularly useful insight into the phenological changes expected in response to an expanded growing season. At high elevations the growing season starts late and ends early, whereas at low elevations it begins earlier and ends later. Initiation of reproduction tracks elevation-associated changes in spring warming in a number of species (reviewed in Stinson 2004). With warming climates, populations at higher elevations will be faced with longer growing seasons similar to those of current lower elevations. Studies across elevations permit a comparison of response to growing season length over short geographic distances allowing regional-scale weather patterns and photoperiod to be held constant.
We took advantage of the differences in growing season length associated with elevation to determine (i) the effect of growing season length on individual components of reproductive phenology and (ii) whether plastic responses to a longer growing season were adaptive. Plants from high elevation were transplanted to low elevation to create an expanded growing season. By also transplanting in the complementary direction, it was possible to determine the extent of phenological differentiation between populations from different seasonal environments. Such differentiation provides insight into any potential long-term evolutionary response to changes in growing season length. We used Campanulastrum americanum, a monocarpic herb for which previous work found elevation-associated differences in flowering time in a uniform environment (Galloway, Etterson & Hamrick 2003), reproductive traits that were tightly integrated when flowering time was manipulated in a natural population (Galloway & Burgess 2009), and for which artificial selection rapidly altered flowering time (Burgess, Etterson & Galloway 2007). Using a reciprocal transplant design, we addressed the following specific questions. (i) What are the effects of an expanded growing season on the suite of phenological traits that span the initiation and progression of reproduction? (ii) Do plastic responses to growing season length involve a shift in all components of reproductive phenology, or do individual phenological components respond separately? (iii) Do populations from different growing season lengths differ in phenology and fitness components, and if so, is there evidence of local adaptation? (iv) Does phenotypic selection on phenological characters differ between low- and high-elevation environments, and are phenological responses to an expanded growing season adaptive?
Materials and methods
Campanulastrum americanum Small (=Campanula americana L; Campanulaceae) is an insect-pollinated outcrossing herb common to open deciduous woods, moist borders and steep slopes throughout eastern and central North America (Shetler 1962; Galloway, Cirigliano & Gremski 2002; Galloway, Etterson & Hamrick 2003). The present study was conducted in southwest Virginia, where natural populations of C. americanum are found across a range of elevations from c.500 m to 1400 m a.s.l. Campanulastrum americanum is monocarpic; rosettes must be vernalized prior to initiating bolting in the spring. Therefore seeds that germinate in the autumn, such as in the present study, have a winter annual life-history. Plants flower in mid- to late-summer with compact inflorescences at reproductive nodes on the main stem and lateral branches. Flowers typically last for 2 days (Evanhoe & Galloway 2002). Fruits contain 10–40 seeds, ripen after c.6 weeks (Galloway 2002), and are persistent, allowing fruit production to be assessed at the end of the season.
Planting site environments
Reproductive phenology and fitness components were assessed for plants grown in low- and high-elevation planting sites. The high-elevation planting site was located in the understorey of a mixed deciduous forest (1194 m) and the low-elevation site in a riparian mixed deciduous forest (514 m, see Table S1 in Supporting Information). Seasonal patterns of temperature and precipitation during the experiment were described with data from the Mountain Lake Biological Station (MLBS) and Kentland Farm meteorological data bases (3 and 0.5 km from and similar in elevation to the high- and low-elevation planting sites respectively). Temperature during the experiment was compared to a long-term mean (1972–1997) at each site. For the high-elevation planting site, data were obtained from the National Climatic Data Center for MLBS (NCDC Co-Op ID 445828); the nearest NCDC station to the low-elevation planting site was 10 km away and similar in elevation (NCDC Co-Op ID 440766). Photosynthetically Active Radiation (PAR) was measured at each planting site in cloud-free conditions using a portable PAR-meter on August 12 and 13. Measurements were taken at 1.5 m heights (slightly taller than most plants) at the four corners and middle of the eight planting blocks per site.
Fruits were collected by maternal plant from five natural populations located on an elevation gradient (see Table S1). Two populations, L1 and L2, were sampled from low-elevation sites on the flood plain of the New River; an intermediate elevation population was sampled at the base of Salt Pond Mountain; and two populations were sampled from high elevation, H1 and H2 (near the high-elevation planting site), near the tops of Beanfield and Salt Pond Mountains, respectively. Tests of the effect of planting site and population origin on phenology and fitness only included populations from the high and low elevations as only those populations were transplanted reciprocally. The phenotypic selection analysis also included the intermediate population to create a broader phenotypic distribution. Fruits were collected from approximately 60 haphazardly selected maternal plants per population.
Seeds were germinated in autumn 2004 under controlled conditions. Three seeds were sown per plug in plug trays filled with MetroMix 200® (Sun Gro Horticulture, Vancouver, British Columbia, Canada) for 50 maternal families in each population (30 for L2). This was replicated six times (eight for L2). The seeds were germinated under near-optimal conditions in a growth chamber (23 °C day/14 °C night; 12:12 L:D). The first seedling to germinate in each plug was retained and subsequent seedlings were removed. Seeds from the low-elevation populations germinated on average 1.5 days earlier than those from the high-elevation populations. A month after planting, seedlings were placed outdoors for 1 week to acclimate to ambient temperatures.
Seedlings were reciprocally transplanted into the two sites in the late autumn in a randomized block design. Eight blocks were created per site, and fenced to exclude mammalian herbivores. All blocks were cleared of existing vegetation before transplanting and weeded regularly throughout the experiment. In total, 1440 plants were transplanted with approximately 150 from each source population in each planting site distributed across families. Individuals were spaced 0.33 m from each other. Season extenders (plastic cups with bottoms cut-off, staked upside down) were affixed around each seedling to help reduce transplant shock in cool fall weather. Transplant mortality was assessed when season extenders were removed 1 week later; individuals that did not survive transplant were dropped from the data set (2.1%). Fall rosette size, measured as the number of leaves multiplied by the length of the longest leaf, was also assessed at this time. Survival (presence/absence) was determined prior to bolting in the spring and again when plants began flowering.
Reproductive phenology was followed from bolting through seed maturation. Bolting phenology was assessed by measuring plant height every 2 weeks until the opening of the first flower. Bolt initiation was recorded as the date of stem height > 1.5 cm. Plants were censused every 3 days for flowering initiation, defined as the first open flower. Bolting duration was the number of days from bolt initiation to flowering initiation. Floral display, scored as the number of open flowers, was censused every 6 days throughout the flowering season. Flowering duration was the number of days between an individual’s first and last flower. Seasonal patterns of flower production for each plant were summarized with ‘average flower date,’ the mean date that an individual’s flowers were produced. Average flower date was calculated by weighting each census date by the proportion of an individual’s total flower production that was open on that date and summing over all census dates (cf. Nuismer & Cunningham 2005). A small average flower date indicates most of an individual’s flowers are produced early in the season, whereas large values indicate greater late-season floral production. Finally, one open flower was tagged on each plant at weekly intervals. Because flowers are typically open for 2 days, the date flowers were tagged approximates the date of fruit initiation. Tagged fruits were checked approximately every 4 days for the opening of lateral pores which indicated maturity. Fruit maturation time was the number of days from when the flower was tagged until the fruit matured.
An index of reproductive duration was estimated from these sequential phenological traits as the sum of bolting duration, the time from flowering initiation to the average flower date and the maturation time of the first fruit. Fruit maturation time was only available for 41% of the plants due to an error in data collection. Therefore mean fruit maturation time, estimated for each population in each planting site, was used when data for an individual was missing. Calculating reproductive duration using average flower day, rather than flowering duration, results in estimates that are more conservative for demonstrating differences between planting sites and elevation of population origin.
Plants were harvested when almost all fruits had dehisced. They were dried, weighed for above-ground biomass and fruits counted. Cumulative fitness was estimated by number of fruits; individuals that did not survive were assigned a fitness of zero. Therefore cumulative fitness combines both survival and reproduction.
Planting site environments
Meteorological variables and light availability were compared between planting sites. A difference between sites for PAR was determined in an anova with planting site as a fixed effect and block nested within planting site as a random effect. Temperature and daily-accumulated precipitation were compared between planting sites by performing a repeated measures anova with planting site as a fixed effect.
Comparison of sites and populations
ancova was used to evaluate differences between planting sites and populations from low and high elevation for phenological characters and fitness components. Planting site, elevation of population origin (only populations from low and high elevation), population nested within elevation, the planting site × elevation interaction, and the planting site × population interaction were treated as fixed effects, and block nested within planting site was included as a random effect. Differences between sites correspond to environmental effects and those between elevations or populations represent genetic effects. A planting site × elevation interaction indicates that populations from different elevations differ in their plastic response to the environments of the planting sites, i.e. genetic differences in plasticity. Fall rosette size, a measure of variation among individuals at planting, was included as a covariate for all traits. Any differences in size due to the slightly earlier germination of the low-elevation populations will be accounted for by the covariate, rather than contributing to the results of adult traits. Survival between the fall rosette, spring rosette and flowering initiation stages was analyzed with a loglinear analysis using a comparable model to the above ancova, assuming a binomial distribution and a logit link function. Fruit maturation time was analyzed with repeated measures anova using the above analysis with the inclusion of fruit initiation date. Unfortunately the maturation date of a number of the later fruits is missing; therefore an additional analysis was conducted using the first tagged fruit to mature on each plant (fruits initiated in the first 2 weeks of flowering).
To evaluate whether phenological traits respond to local environments independently, they were reanalyzed using the previous phenological stage as an additional covariate. Path analysis is an alternate approach to addressing the interrelated phenological traits (see Table S2). However, path analysis is only effective when a branched diagram can be constructed; the linear nature of the sequential phenological traits measured here precluded this analytical approach.
To meet assumptions of normality, biomass and cumulative fitness+1 were ln-transformed while fruit production was square-root transformed. The significance testing is approximate for cumulative fitness due to the large number of zeros in the data set. However, analyses of fitness means across groups of individuals yielded qualitatively similar results.
Phenotypic selection analysis
Phenotypic selection acting on phenological traits was estimated for each planting site to evaluate whether plastic changes in trait expression between environments were adaptive (Lande & Arnold 1983; Mitchell-Olds & Shaw 1987). Individuals from all five populations were included to expand the phenotypic distribution and enhance the ability to detect selection (Wade & Kalisz 1990; Conner & Hartl 2004). The phenological traits, date of bolt initiation, date of flowering initiation, average flower date and flowering duration were included in the analysis. Bolting duration was not included because it is not independent of bolt initiation and flowering initiation. Analyses initially included fruit maturation time, but that variable was dropped due to lack of significant selection combined with missing data for a number of individuals. Biomass was included in the model to control for any covariance between phenology and plant size, and population was included to account for the lack of independence among plants from the same population. Traits were first standardized to a mean of zero and variance of unity. Pearson correlation coefficients were then calculated to determine correlations between characters (see Table S2). To ensure that multicollinearity among these traits did not influence results, we estimated variance inflation factors and found all were less than five, implying limited multicollinearity (Kutner, Nachtsheim & Neter 2004).
Relative fitness was calculated by dividing individual reproductive success (fruit production) for plants that survived to flowering by the site-specific mean (Lande & Arnold 1983). Selection differentials (Si) were calculated as the covariance between relative fitness and each standardized character (i), and are measures of direct linear selection on character i as well as any indirect effects from selection on correlated characters. Standardized linear selection gradients (βi) were calculated as partial regression coefficients from the multiple regression of relative fitness on all traits. Thus, βi is a measure of the effect of each trait i on relative fitness, holding all other traits fixed. Standardized nonlinear selection gradients (γi) were obtained as double the parameter estimates from a multiple regression analysis of relative fitness on all traits and their squares (Stinchcombe et al. 2008). Negative values of γi indicate a decelerating relationship between trait values and fitness, with stabilizing selection present when intermediate trait values have the greatest fitness. Positive values of γi indicate accelerating selection where a unit change in the trait is associated with a greater fitness increase for more extreme trait values.
We determined the significance of the selection gradients using 95% confidence intervals estimated by creating 10 000 replicate data sets in a bootstrap of the original data. The patterns of significance for both linear and quadratic selection gradients in each of the two planting sites were consistent with those of parametric tests, so the results of the original regressions are reported. ancovas were performed to determine whether the magnitude of linear or quadratic selection differed between sites with phenological traits as covariates and planting site a fixed effect. Significant interactions between planting site and the phenological traits indicate that the pattern of selection differs between environments.
Planting site environments
The planting sites at high and low elevation differed in temperature but not in accumulated precipitation (Site F1,686 = 2.97, P = 0.09; Site × Month F11,686 = 0.12, P = 0.99) or light (PAR Site: F1,64 = 2.37; P = 0.15). Repeated measures anova revealed that mean temperatures during the experiment were warmer at the low-elevation site (F1,686 = 78.4; P < 0.001) and fluctuated synchronously at both sites throughout the experiment (Site × Month: F11,686 = 0.21; P = 0.99). As a consequence, warm spring temperatures occurred earlier and fall frosts later at the low-elevation site. Temperatures at the planting sites during summer growth and reproduction were representative of long-term patterns at those locations (April–August 2005, elevation mean: high 14.99 °C, low 18.02 °C; long-term elevation mean ± SE: high 14.73 ± 1.83 °C, low 17.58 ± 2.00 °C).
In both planting sites, bolting phenology differed between populations from high and low elevations. Populations from high elevation initiated bolting late and rapidly produced a fully developed flowering stem in both sites, whereas those from low elevation initiated bolting early and required more time to produce a flowering stem (Table 1, Fig. 1a and d). On average, the bolt initiation date in the low-elevation site was advanced by about 7 days relative to the high-elevation site (Fig. 1a). However, planting site had no average effect on bolting duration because populations had opposite plastic responses to changes in elevation. Populations from low elevation had shorter bolting duration in the low-elevation site whereas populations from high elevation had shorter bolting duration high-elevation site (Table 1, Fig. 1d). As a result of these combined variables, there is less difference between populations from high and low elevations for flowering initiation date than for bolting traits. Similar to bolt initiation, flowering initiation was advanced by about 7 days at the low-elevation site (Fig. 1b). There were no differences between populations from low and high elevations for flowering initiation in the low-elevation site. However at the high-elevation site, populations from low elevation flowered significantly later than those from high elevation (Table 1, Fig. 1b).
Table 1. ancova of reproductive phenology for populations of Campanulastrum americanum sampled from high and low elevations and reciprocally transplanted into common gardens at those elevations. F-values are reported for fixed effects and Z-values for random effects. (*)P < 0.10, *P < 0.05, **P < 0.01, ***P < 0.001
Seasonal flowering patterns differed between the planting sites (Fig. 2). The low-elevation environment induced an advance in average flower date by c.14 days. In both sites, populations from high elevation had an earlier average flower date than those from low elevation (Table 1, Fig. 1c), indicating that populations from high elevation produce the majority of their flowers early in the season, whereas those from low elevation produced more flowers late in the season (Fig. 2). Flowering duration was almost twice as long at the high-elevation site than the low-elevation site (Table 1, Figs. 1e and 2). The flowering duration of populations from low and high elevations was similar when plants were grown at high elevation. However in the low-elevation site, flowering duration of populations from high elevation was less than those from low elevation. On average, populations from high elevation had shorter fruit maturation time than those from low elevation (Tables 1 and 2; Figs. 1f and 3). Fruit maturation time was also more rapid at the low-elevation site with the first fruit taking an average of 20 days less time to ripen than at the high-elevation site (Table 1; Fig. 1f). Over the season, time to fruit maturation of high- and low-elevation populations was similar when grown at low elevation (F1,178 = 0.59, P = 0.44) but in the high elevation site was shorter for high-elevation than low-elevation populations (F1,91 = 5.59, P = 0.02; site × elevation Table 2). Fruit maturation time shortened over the season at both planting sites, with final fruit maturation time over a week shorter than initial maturation time (Table 2, Fig. 3).
Table 2. Repeated measures anova to evaluate effects of planting site and elevation of origin on the maturation time of Campanulastrum americanum fruit initiated at weekly intervals over the reproductive season. F-values are reported for fixed effects and Z-values for random effects. Denominator degrees of freedom = 273 (Planting site denominator d.f. = 14)
Elevation of origin
Site × Elevation
Site × Population(Elev)
Date fruit initiated
The differences in phenology between plants grown in high- and low-elevation sites and between populations from high and low elevations were found for each successive reproductive stage. Patterns of differentiation between sites and between populations from different elevations changed little when the previous phenological stage was included as a covariate (Table 3). The exception was average flower date. After accounting for differences in initiation of flowering, greater divergence in average flower date was found between populations from low and high elevation in the low-elevation site than the high-elevation site (Table 3). However, for most traits populations from low- and high-elevation were more similar when grown in the low elevation site. Least-square means from the other analyses were very similar to Fig. 1, indicating that successive phenological components each responded to the local environment, but the pattern of response was similar across the reproductive phenology. As a consequence of the different responses among traits, reproductive duration was 21% shorter in the low-elevation site than the high-elevation site and 11 days shorter in plants that originated from high elevation than those from low elevation (Table 1, Fig. 4).
Table 3. ancova to evaluate stage-specific effects of planting site and elevation of origin on reproductive phenology of Campanulastrum americanum sampled from high- and low-elevation populations and reciprocally transplanted into common gardens at those elevations. For each trait, rosette size at transplant is the upper covariate and the previous phenological stage is the lower covariate. F-values are reported for fixed effects and Z-values for random effects. (*)P < 0.10, *P < 0.05, **P < 0.01, ***P < 0.001
The probability of an individual surviving to flower was a function of the planting site and the elevation from which it originated, but not their interaction (Table 4). Survival to flowering can be divided into two life cycle stages: the overwinter interval and the spring growth interval. Although statistically similar during the winter interval, survival was significantly greater in the low-elevation site during the spring interval and over the entire fall to flowering period (Table 4, Fig. 5a). Across both planting sites, populations from low elevation had a greater proportion of individuals survive to flowering than those from high elevation.
Table 4. Log-linear analysis of survival and analysis of covariance of fruit production and estimated fitness for populations of Campanulastrum americanum sampled from high and low elevations and reciprocally transplanted into common gardens at those elevations. Chi-square statistics reported for survival; for other traits F-values are reported for fixed effects and Z-values for random effects. **P < 0.01, ***P < 0.001
Although overall fruit production was greater in the high-elevation site, fruit production within each planting site was greater for plants grown at their native elevation (Table 4, Fig. 5b). Cumulative fitness was more similar between sites than its components due to opposite responses to the sites for survival and fruit production (Fig. 5c). At the high-elevation site cumulative fitness did not depend on a population’s origin, while populations from low elevation had greater cumulative fitness at the low-elevation site (Table 4, Fig. 5c). As a result, overall cumulative fitness was greater for populations from low elevation.
Phenotypic selection analysis
Phenotypic selection analyses revealed stronger total and direct selection on phenological traits in the expanded growing season of the low-elevation site than in the high-elevation site (Table 5). Total selection (S) favoured earlier bolt and flowering initiation, and a longer flowering duration at both sites, along with a later average flower date at the low-elevation site (Table 5). Direct linear selection (β) in the low-elevation site favoured later bolt initiation and earlier flowering (Table 5). The magnitude of selection on flowering initiation was c.2.5 times greater than on bolt initiation. Direct linear selection was not detected for any phenological traits in the high-elevation site.
Table 5. Standardized selection differentials (S), linear gradients (β), and quadratic gradients (γ) for populations of Campanulastrum americanum sampled along an elevation gradient and planted in common gardens at high and low elevations. Source population was also included in the models as a blocking term. Selection is compared between the planting sites with F-values in the Site × S, Site × β and Site × γ columns. High-elevation site N = 294, low-elevation site N = 245. (*)P < 0.10, *P < 0.05, **P < 0.01, ***P < 0.001
Site × S
Site × β
Site × γ
Ave flower date
Nonlinear selection (γ) was only detected in the low-elevation site (Table 5). A positive quadratic selection gradient for flowering initiation indicated that relative fitness was an accelerating function of earlier flowering such that earlier flowering resulted in a larger increase in fitness for early-flowering plants than for later-flowering ones. A negative quadratic selection gradient was found for average flower date. Graphical inspection revealed a peak within the phenotypic distribution, indicating stabilizing selection on patterns of flower deployment.
The expanded growing season resulted in a nonlinear shift in reproductive phenology. Bolting and flowering were each about 7 days earlier under the warmer low-elevation conditions indicating the advance of reproductive initiation by a week. Average flower date was 14 days earlier in the longer growing season, a week beyond that due to initiation of flowering, in part because flowering duration was reduced by about 12 days. Finally, fruits matured on average 20 days faster under the warmer conditions. In total, combining advances in initiation of flowering and average flower date with the reduced time to fruit maturity yielded a reproductive phenology in which fruit initiated on the average flower date matured on average 34 days earlier at the low-elevation site than the high-elevation site (Fig. 6).
Therefore in C. americanum, earlier initiation of bolting does not simply shift the reproductive phenology earlier. Nor is there an expansion of the phenological schedule to match the longer growing season as seen in some woody plants (Aerts et al. 2004; Chmielewski, Muller & Kuchler 2005). Instead, the reproductive phenology is compressed such that progression through the later phenological stages is accelerated relative to the earlier stages. Because advances in early phenological traits may move subsequent phenological events into cooler temperatures, reducing environmental differences between the sites, we estimated the growing degree days (GDD) accumulated in each site for each phenological event. We found greater GDD accumulations in the low-elevation site for each phenological stage (GDDlow site−GDDhigh site: bolting initiation 6.2, flowering initiation 153.9, average flower date 129.8, estimated maturation date of a fruit initiated on the average flower date 159.6). This indicates that individual phenological events occurred at warmer temperatures in the low-elevation sites which likely resulted in the non-linear shift in phenology, with reproduction occurring earlier, more rapidly and occupying less of the growing season. When flowering initiation date was manipulated in a natural C. americanum population, reproductive phenology was highly integrated and shifted to follow flowering time (Galloway & Burgess 2009). The exception was plants that flowered a month earlier than the natural population which had a compressed reproductive phenology similar to that seen in this study. By comparison, an investigation of in situ warming in 12 prairie species found the total reproductive period was shortened in four species, expanded in three, and unchanged in five (Sherry et al. 2007). Also, duration of the initial stages of reproductive phenology, examined under control and warmed conditions in a low-arctic site, revealed a compression of individual reproductive stages in two of three species (Post et al. 2008). These studies support the possibility that the nonlinear shift reported here may be found in other systems.
Such nonlinear dynamics among phenological traits can affect reproductive success and ecological interactions. Plants grown in the low-elevation site had reduced fecundity relative to those with longer reproductive schedules at the high-elevation site. This may be because plants initiated reproduction earlier at the low-elevation site. Timing of reproduction often determines final size, therefore earlier flowering may reduce fecundity (Roff 2002; in C. americanum, Burgess, Etterson & Galloway 2007). Alternatively, the compressed reproductive phenology, with reduced flowering duration, may have resulted in the production of fewer flowers if the rate of flower production was unchanged. The compressed reproductive schedule may also alter biotic interactions such as with pollinators or herbivores (Elzinga et al. 2007; Hegland et al. 2009) or other abiotic factors (Giménez-Benavides, Escudero & Iriondo 2007; Inouye 2008; Jentsch et al. 2009), thereby reducing fecundity. Finally, the reduced fecundity at the low-elevation site could be partially an experimental artefact because plantings at the two sites were initiated at the same time whereas in nature the timing of germination may differ between the sites. However, it is not known whether germination would be earlier or later at the low-elevation site making it difficult to predict its effect on fecundity. Further study to identify the specific cause of reduced fecundity would aid in predicting demographic responses to warmer conditions.
Evolutionary response to growing season length
Reproductive phenology was differentiated between C. americanum populations from low and high elevations. Regardless of planting site, populations from low elevation initiated bolting earlier and reached reproductive maturity more slowly, whereas populations from high elevation initiated bolting later and developed more quickly. As a result of rapid development, the late-bolting high-elevation populations initiated flowering at the same time as earlier-bolting low-elevation populations in the low-elevation site, and even earlier in the high-elevation site. Populations from high elevation produced more of their flowers toward the beginning of the reproductive season and flowered for a shorter duration on average than those from low elevation. Finally, fruits matured more rapidly in populations from high elevation than those from low elevation. In total, reproductive phenology was shorter for populations from high elevation with more rapid development, more condensed and earlier deployment of flowers, and faster fruit development.
Differences between low- and high-elevation populations likely reflect genetic divergence and suggest that local growing season length has selected for different patterns of reproductive phenology. Maternal environmental effects may also contribute to the differentiation of populations from low and high elevation (Roach & Wulff 1987). However, in other taxa we also find that the short growing seasons of alpine or high-latitude environments are mitigated by rapid development and reproduction (Arroyo, Armesto & Villagran 1981; Ratchke & Lacey 1985; Arft et al. 1999; Blionis, Halley & Vokou 2001; Olsson & Ågren 2002; Stinson 2004). The generality of these patterns reveals a common evolutionary response of reproductive phenology to growing season length and suggests that we may expect widespread evolution of a slower reproductive phenology in response to warmer climates.
Because high-elevation populations initiated bolting up to a week later than low-elevation populations, it is possible that the differences in bolting duration are due to environmental factors rather than genetic effects. Specifically, the later-bolting high-elevation populations experienced warmer temperatures when bolting which may have accelerated development and reduced bolting duration. This possibility is supported by a negative association between bolting duration and accumulated growing degree days, R = −0.78; P < 0.001 (see also Inouye, Morales & Dodge 2002). However, populations from high elevation still have a shorter bolting duration than those from low elevation after statistically accounting for these environmental effects, indicating a genetic basis to differences in bolting phenology.
Differentiation of a suite of traits across a physical gradient, such as elevational differences in reproductive phenology, is termed ‘ecotypic differentiation’ (Turesson 1922; Clausen, Keck & Hiesey 1948) and is typically the product of local adaptation. Fruit production in C. americanum was locally adapted to the environmental differences associated with elevation. Populations from high elevation had greater fruit production than those from low elevation when grown in the high-elevation site, and populations from low elevation produced more fruit than those from high elevation in the low-elevation site. However, cumulative fitness did not show local adaptation because survival of low-elevation populations was greater at both planting sites. Divergence in reproductive phenology between low- and high-elevation populations is expected to influence reproductive fitness components more than survival. Therefore ecotypic differentiation in reproductive phenology between high- and low-elevation populations is likely adaptive. Because common gardens were not replicated within an elevation, and unmeasured environmental factors may contribute to differences between high- and low-elevation sites, temperature cannot be isolated as the sole factor underlying local adaptation or phenological differences. However, previous work on a forest herb identified temperature and light availability as the major environmental factors influencing reproductive phenology (Dahlgren, von Zeipel & Ehrlen 2007). Therefore shifts in mean temperature and growing season length across elevations, but similar light and water availability, suggest that differences in temperature are likely to have led to adaptive differentiation of reproductive phenology between high and low elevation C. americanum.
Reproductive phenology of populations from high and low elevation was more similar when plants were grown under the expanded growing season of the low-elevation site. For all phenological traits except average flower date, there were differences in the response to growing season length between populations from low and high elevation. In most traits including overall reproductive duration, this difference in plasticity resulted in less phenotypic divergence in the low-elevation site than the high-elevation site. As a result, there was a phenotypic resemblance between the populations from high elevation and the local low-elevation populations in the low-elevation site. Whether this phenotypic similarity reveals historic selection of high-elevation populations in a longer growing season (reviewed in Ghalambor et al. 2007) or simply a common physiological response of accelerated growth and development under warmer conditions is not known. Temperatures during the year of study were representative of the twenty-five year average at each site. Regardless, convergence on a similar phenotype suggests adaptive plasticity of populations from high elevation to the warmer low-elevation climate.
Selection on phenology in an expanded growing season
Selection on phenological characters was stronger in the expanded growing season. We might imagine, all else being equal, that a longer growing season would favour later bolting and flowering, because that would allow plants to grow larger prior to flowering and therefore reproduce more per unit time (Roff 2002; Metcalf, Rose & Rees 2003). Following this expectation, in uniform growth conditions plants from lower latitudes, where growing seasons are longer, often flower later than those from higher latitudes (e.g. Weber & Schmid 1998; Olsson & Ågren 2002; Etterson 2004; Griffith & Watson 2005). A longer growing season might also be expected to favour individuals with a longer reproductive period and later average flower dates because these individuals would use the longer season to produce more reproductive structures. Following these expectations, in the longer growing season direct linear selection favoured later bolting (larger size). However, in contrast to expectations, earlier flowering individuals had greater fitness. Flowering duration did not affect fruit production, but plants that deployed their flowers with a similar timing to those currently growing naturally in the low-elevation site, i.e. with the same average flower date, had the greatest fitness. Thus, in an expanded growing season there was direct fitness advantages associated with attributes of populations from high elevation: delaying bolting and advancing flowering initiation.
Phenotypic plasticity modified flowering initiation and average flower date in the direction favoured by selection in the low-elevation site. Therefore adaptive plasticity for these traits provided populations beneficial short-term responses to the selective pressures imposed by the longer growing season (see also Etterson 2004). However, the plastic response of earlier-bolting in the low-elevation site was maladaptive. Maladaptive plasticity to a longer growing season suggests that evolutionary change will be required for this trait to enhance fitness under the projected warmer conditions. The relatively later initiation and shorter duration of bolting in the high-elevation populations suggest that such evolution is possible, despite the positive phenotypic correlation between bolting and flowering initiation (cf. Etterson & Shaw 2001; Hellmann & Pineda-Krch 2007).
In total, the results of this reciprocal transplant study suggest that individual components of reproductive phenology will play a central role in plant reproductive success as the global climate continues to warm. The combination of adaptive plasticity, expressed as nonlinear responses of reproductive phenology to warmer temperatures, and selection for changes in the timing and relative duration of phenological traits, suggests that we must consider the ecological and potential evolutionary changes of individual phenological components to forecast response to climate change.
We thank F. Kilkenny and K. Burgess for field assistance; D. Roach, H. Wilbur, P. Klinkhamer, several referees, and the Galloway lab for comments on previous versions; an MLBS fellowship to B.P.H. and NSF DEB-0316298 to L.F.G. for financial support; and MLBS and Kentland Farm of Virginia Tech for logistical support.