Effects of Genetic Perturbation on Seasonal Life History Plasticity

See allHide authors and affiliations

Science  13 Feb 2009:
Vol. 323, Issue 5916, pp. 930-934
DOI: 10.1126/science.1165826

This article has a correction. Please see:


Like many species, the model plant Arabidopsis thaliana exhibits multiple different life histories in natural environments. We grew mutants impaired in different signaling pathways in field experiments across the species' native European range in order to dissect the mechanisms underlying this variation. Unexpectedly, mutational loss at loci implicated in the cold requirement for flowering had little effect on life history except in late-summer cohorts. A genetically informed photothermal model of progression toward flowering explained most of the observed variation and predicted an abrupt transition from autumn flowering to spring flowering in late-summer germinants. Environmental signals control the timing of this transition, creating a critical window of acute sensitivity to genetic and climatic change that may be common for seasonally regulated life history traits.

The seasonal timing of critical life history events is often under strong natural selection, requiring plants to integrate and respond appropriately to multiple environmental signals. In the case of flowering time, these include day length, ambient temperature, and winter chilling (vernalization) cues. Genetic mechanisms of flowering response to environmental cues have been well characterized in the model plant Arabidopsis thaliana in the laboratory. However, the response of Arabidopsis to complex, natural environmental cues is unknown. The native distribution of Arabidopsis encompasses climates from the Mediterranean, where it is a winter annual that germinates in late autumn and flowers in late winter or early spring, to northern Scandinavia, where the species germinates in early autumn, overwinters, and flowers in late spring and early summer. In England and northwestern Europe, Arabidopsis either is a winter annual or undergoes rapid cycling by germinating in early autumn, spring, or summer, thus flowering without vernalization (1, 2). Thus, in nature the species experiences marked variation in photothermal environments across sites and seasons.

The timing and environmental sensitivity of flowering in Arabidopsis are regulated by a network of genes in several converging pathways (35) (fig. S1). Inductive long days are perceived by the photoperiod pathway, which accelerates flowering acting through GIGANTEA (GI) and CONSTANS (CO) (6, 7). The gibberellin pathway (8) and high temperatures (912) also promote flowering. These signals activate floral integrator genes such as FLOWERING LOCUS T (FT) (3). The ability of these integrators to respond is controlled by repressor genes, notably FLOWERING LOCUS C (FLC), which in turn is activated by genes such as FRIGIDA (FRI) and is repressed by vernalization and by genes in the autonomous pathway such as LUMINIDEPENDENS (LD) and FVE (13). Prolonged exposure to cold induces VERNALIZATION-INSENSITIVE-3 (VIN3), which initiates stable epigenetic repression of FLC (1315). Deficiencies in FLC activators such as FRI remove the vernalization requirement for flowering and may cause a rapid-cycling life history (16, 17).

To assess the relative contributions of different signaling pathways across environments, we grew mutants impaired in response to specific environmental signals and documented environmental conditions at five field sites spanning the native European climatic range: Oulu, Finland; Norwich, UK; Cologne and Halle, Germany; and Valencia, Spain (figs. S2 and S3) (18). Across these sites, we established a total of nine seasonal cohorts timed to coincide with the germination of local natural populations (fig. S2). To compare the relative contribution of the photoperiod pathway to flowering across sites and seasons, we examined the flowering delay in mutants of CO, which is activated by long-day signals, and in mutants of GI, which promotes flowering both by activating CO (19) and through a CO-independent derepression of FT (20). Both photoperiod pathway mutants (co-2 in Ler, gi-2 in Col) showed initiation of flowering significantly later than in the wild type in all plantings except for those planted at Halle in the fall (Fig. 1 and table S1). These mutants flowered especially late in the summer plantings and the Norwich fall cohort, planted under declining long days (Fig. 1).

Fig. 1.

Sensitivity of flowering-time pathways across sites and seasons. Bars show mean bolting times (±SE) of mutants in different flowering-time pathways along with their wild-type controls. (A) Norwich summer 2007 planting; (B) Oulu summer 2006; (C) Norwich spring 2007; (D) Cologne spring 2007; (E) Norwich fall 2006; (F) Cologne fall 2006; (G) Halle fall 2006; (H) Valencia fall 2006. Note the difference in scale among locations and seasons.

To examine the effects of different FLC repression levels, and thus vernalization requirements, we compared the fri null Col-0 wild-type to Col-FRI-Sf2, which contains an introgressed strong allele of the FLC activator FRI that naturally occurs in the Sf-2 ecotype (21). We also examined the effects of the autonomous pathway mutants ld-1, fve-3, and fve-4, which impair repression of FLC expression and delay flowering in the absence of vernalization, even in a null fri background (22). Col-FRI-Sf2 and the autonomous pathway mutants flowered at similar times within each of the plantings. Although FRI introgression and autonomous pathway mutation were expected to confer constitutive winter-annual life histories on the basis of greenhouse studies (6, 21), these lines displayed delays of only ∼10 days when planted in summer cohorts, which lacked vernalization (table S2). Thus, FLC-induced repression of flowering can be overridden, possibly as a result of natural temperature fluctuations (fig. S4) (23) or light conditions (21). Col-FRI-Sf2 and the autonomous pathway mutants both demonstrated small but significant delays in bolting in the spring cohorts and displayed an expected winter-annual life history in all autumn plantings (Fig. 1 and table S1). However, the Col wild type behaved as a rapid cycler (its laboratory phenotype) only in Norwich, where it flowered in fall more than 100 days earlier than did Col-FRI-Sf2 and Col plants containing autonomous pathway mutations. Unexpectedly, in the remaining autumn cohorts, Col flowered much later than in Norwich, displaying a clear winter-annual life history in Halle. Thus, loss of FRI function converts winter annuals to rapid cyclers only under a narrow range of natural environmental conditions.

The vin3-1 deletion, which is deficient in its ability to respond to vernalization cues, had no significant effect on flowering time in the summer cohorts relative to the vernalization-sensitive background wild type (Col-FRI-Sf2) (Fig. 1 and table S1), as expected. In contrast, vin3-1 mutants flowered later than the control in autumn plantings in a manner that was correlated with exposure to vernalizing temperatures (fig. S3C). Additionally, vin3-1 mutation significantly delayed flowering in spring cohorts (Fig. 1), where plants were exposed to vernalizing temperatures early in development (fig. S3C). Thus, vernalization signals predominated in the later autumn cohorts, photoperiod predominated in summer plantings, and both substantially contributed to the flowering response of spring-germinating plants.

We created a genetically informed photothermal model of Arabidopsis development in which plants transition from nonflowering rosettes to bolting (the first visible sign of floral initiation) once they have accumulated a threshold number of environmentally determined photothermal developmental units—temporal measures that incorporate climatic conditions and are interconvertible with real time [e.g., (24)]. Phenology models empirically scale photothermal units so that developmental transitions occur at a common threshold of accumulated units independent of particular environmental conditions. The accumulation of appropriately scaled photothermal units over time will therefore show the developmental progression of plants toward the initiation of reproductive growth (Fig. 2). We extended classical phenological modeling approaches by linking individual scaling factors to the activities of specific genes and their regulators. In our gene network–based model, each genotype accumulated photothermal units on the basis of day lengths and hourly temperature averages in each experimental planting, modified according to the environmental sensitivity of the line. For example, in our model, both photoperiod mutants and wild-type lines (Col versus gi-2, Ler versus co-2) were assumed to develop at similar rates under short days.

Fig. 2.

The accumulation of modified photothermal units from the time of sowing for genotypes in the Col background differing in environmental sensitivity. The model estimated the threshold for bolting at 2604 modified photothermal units (horizontal dashed line), a constant for all environments and genotypes. Each genotype is depicted by a single trace, which shows the accumulation of modified photothermal units in that particular mutant or wild-type background. Observed bolting times (days on the x axis, modified photothermal units on the y axis) with confidence intervals are shown on each of the accumulation traces. Note differences in scale of x axis among seasons. (A) Norwich summer 2007; (B) Norwich spring 2007; (C) Norwich fall 2006; (D) Halle fall 2006.

Photoperiod mutants did not respond to increasing day length, but the modeled developmental rate of wild-type plants accelerated above a critical short day length (Fig. 2) until a maximum rate was reached at a saturating long day length (25). Chamber experiments confirmed this piecewise linear response to day length in the Ler accession and identified a critical short day length of ∼10 hours. However, flowering in the co-2 mutant was insensitive to day length (fig. S5). Development was also assumed to accelerate during vernalization because of the repression of FLC, with the modeled degree of acceleration varying according to genetic background (Ler versus Col), the allelic state of FRI (Col fri versus Col FRI), and the presence or absence of functional alleles in the autonomous pathway (Col versus ld and fve); however, vin3-1 plants did not experience any FLC repression from exposure to cold temperatures (Fig. 2). A novel feature of the model, resulting from the scaling of developmental rate with pathway activities, was that all mutants within a given genetic background were predicted to flower at a common threshold of modified photothermal units. Parameter values for the critical day length window, vernalization effectiveness temperature function, time until vernalization saturation, and the degree-hour base temperature were drawn from empirical studies (table S3). Short-day developmental rate and the initial levels of FLC repression of development were optimized to the lowest coefficient of variation of modified photothermal units accumulated at observed bolting times across all plantings (18).

Our best-fit model explained more than 92% of the observed overall variance in days to bolting (Fig. 3A) and was in agreement with observed reversals in flowering-time order of different genotypes across plantings (e.g., gi-2 versus Col-FRI-Sf2, Cologne and Norwich spring versus fall; Fig. 2 and table S2). The model, with parameter values derived from our field experiment, also accurately predicted the behavior of plants not used in the model's construction and accounted for most of the flowering-time variation of Col, Ler, and co mutants under a wide range of controlled conditions [(11) and table S4; R2 = 0.85 for the data in both studies, Fig. 3B]. Thus, a modified photothermal model, informed by understanding of pathway function and empirical measurements of gene effects in controlled environments, can be used to quantify the shifting balance and sensitivity of flowering-time pathways under a broad range of climate conditions.

Fig. 3.

Observed versus predicted bolting times. (A) Field-observed bolting phenotypes of Ler-1, co-2, Col-0, gi-2, Col-FRI-Sf2, vin3-1 (in Col-FRI-Sf2), and fve-3 across the nine experimental plantings and the predicted phenotypes from the fitted model. Line means with confidence intervals are shown. Confidence intervals for the modified photothermal units (MPTU) were back-calculated to find the day at which the lower and upper 95% confidence limit for the bolting threshold was reached. (B) Verification against growth chamber data not used in model construction, showing results from Ler, co (both co-2 and co-3), and Col across nine treatments in two experiments.

We examined the relative sensitivity of flowering time to parameters representing vernalization sensing, day length sensing, short-day development rate, and baseline FLC levels (table S3) and found that flowering time was most sensitive to the initial FLC level in both Col and Ler. The short-day development rate, which represents the baseline at which plants proceed toward flowering in the absence of long-day or vernalization cues, had a large effect on flowering time. Indeed, development was slowed by 60 to 74% before vernalization depending on background, and short-day development proceeded only 63 to 69% as quickly as in long days. Similar genetically based sensitivity to photoperiod and/or vernalization cues has been observed in other species and traits (24, 26), in some cases with similar integration of converging pathways (2729). The linking of scaling factors to pathway activity depends on network architecture rather than gene identity, which suggests that similar models will prove useful in other species, even where the genes involved do not share a common evolutionary history.

Our model also was able to predict the effects of novel environments on flowering time across genotypes. For example, we used the model to generate predicted reaction norms of flowering date to germination timing in Norwich and Cologne, using actual daily on-site photothermal data to simulate the flowering dates of seeds germinating on successive days over a yearly cycle (Fig. 4). The model predicted a striking, abrupt transition from rapid-cycling to winter-annual life histories for all genotypes germinating in late summer. During this narrow temporal window of sensitivity, all genotypes switched from autumn to spring flowering; the high-FLC lines Col-FRI-Sf2 and vin3-1 transitioned between life histories earliest and showed the largest flowering-time response (Fig. 4). A series of repeated plantings into the field in Cologne of Col and Col-FRI-Sf2 seedlings throughout the autumn and spring confirmed the predicted abrupt life history transition (Fig. 4B, R2 = 0.94). This rapid rise in bolting time with later germination in autumn will be general across many climates, as it is driven by common patterns of seasonal periodicity of light and temperature cues (fig. S6). Thus, the impact of genetic perturbation on flowering time and life history expression under natural conditions is acutely sensitive to germination timing.

Fig. 4.

Timing of bolting as a function of time of germination. (A) In Norwich, a narrow window of extreme sensitivity to germination timing is predicted when sown in early autumn. Timing of experimental sowings is shown by arrows. (B) Model validation in Cologne with measured bolting data from repeated plantings of FRI functional and nonfunctional ecotypes into the field from September 2007 through March 2008. Our model predicted 94% of the variation in bolting time, confirming the expectation that FRI results in a life history conversion in early-autumn cohorts.

Our results suggest that A. thaliana ecotypes cannot simply be divided into two discrete classes of winter-annual and rapid-cycling genotypes. Rather, most ecotypes may be capable of both life histories but vary in the sensitivity and timing of the rapid transition between them. Consequently, natural variation in flowering pathways may have the greatest phenotypic expression and exposure to natural selection in climates that permit late-summer germination. Although the current model was developed for A. thaliana, the underlying principle of linking genetic architecture to environmental sensitivity is broadly applicable across plant species and seasonal traits. Our ability to predict the consequences of genetic variation under current and future climates has implications for optimizing plant breeding and cultivation strategies.

Supporting Online Material

Materials and Methods

Figs. S1 to S11

Tables S1 to S6


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article