-
(i)
Study species
Lobelia inflata (Campanulaceae) is a monocarpic plant native to Eastern North America. It has multiple flowering schedules in the wild (both annual and biennial patterns have been observed), but reproduction is always semelparous in that the plant senesces after completion of flowering. Upon germination, L. inflata seeds form rosettes capable of overwintering. Reproduction is initiated as a reproductive stalk forms from a mature rosette; this event is termed “bolting” and occurs predictably if size, photoperiod and light quality thresholds are exceeded [43]. L. inflata has perfect flowers, reproduces sexually, and is obligately self-fertilizing. Outcrossing is prevented by a stamen tube, a structure which permits the release of pollen directly onto the stigma, but does not permit the release of pollen into the air, since it is sealed [37]. Analyses of polymorphic microsatellite loci [44] have revealed no evidence that outcrossing occurs in nature.
Bolting, which marks the beginning of a transition from a vegetative to a reproductive phase, is irreversible for L. inflata, and thus the timing of this “decision” has important fitness consequences. Inflorescences show an acropetal flowering pattern, where flowers are produced in series from the base to the tip of the stalk (and along each branch). Each flower progresses through easily observable stages: from bud, to flower formation, to anthesis, to “inflation” (where fruits resemble small balloons–hence the name Lobelia cinflata”) and finally to fruit maturation. Reproduction occurs as seeds are formed inside inflated ovules; the number of seeds in a fruit has been observed to depend on environmental unpredictability and reproductive timing [39, 45]. During reproduction, one or more shoots may branch off from the main stalk. Seeds disperse passively upon fruit maturation; once all fruits have reached this stage, a plant senesces.
-
(ii)
Experimental design
Of central importance to our design was the ecological significance of the timing of bolting, marking the (irreversible and terminal) initiation of reproduction. By manipulating the date of initiation of reproduction, we were able to control the length of time that plants had to reproduce. Because reproduction is terminated relatively consistently among individuals each year (around October 15th) with the onset of hard frosts–a phenotypically plastic reproductive response to a range of manipulated bolting dates could be effected.
We collected seeds from dead plants at the Petawawa Research Forest (Lat. 45°99’N, Long. 77°30’W) in eastern Ontario, Canada in October 2007. To maximize the potential inclusion of a variety of genetic lineages (and preclude the possible influence of atypical genotypes), we collected seeds from 21 parent plants in the field (each at least 50 m from each other). Each seed sample was used to found an experimental population of genotypically identical replicate plants, yielding 21 (potentially distinct) genetic lineages. To obtain offspring plants from each lineage we first placed groups of 100-200 seeds on moistened filter paper in petri dishes, then germinated seeds for 10-14 days in a BioChambers SG-30 seed germinator using a 12 h /12 h day/night light regimen (at 20°C with 85% humidity).
Seedlings were then moved to individual cells (dimensions: 7.6 X 7.6 cm) within trays of autoclaved soil, and were then transferred to a Biochambers AC-40 growth chamber for prebolting rosette growth under a 16 h/8 h day/night schedule (at 24°C/18°C day/night). Trays were watered twice weekly, and a 15-5-15 liquid fertilizer mixture (200 ppm N) was added once every two weeks. Seedlings were grown for approximately 8-9 weeks, forming small rosettes. Rosettes grew undisturbed until bolting; the emergence of a reproductive stalk upon bolting may be reliably detected [37]; a stalk taller than 4 cm is diagnostic of the onset of bolting. Seed germination and seedling translocation was planned so that plants would initiate bolting at four regular intervals, targeting the 15th of each month (±2 days) from June through September. Plants that bolted before the 13th or after the 17th of the month were excluded. The distribution of plants included in the study is shown in Table 1.
Bolted rosettes within each group were randomly assigned to one of two environments: a field site (at 45°23’N, 75°41’W) or to a growth chamber, which was programmed to mimic the photoperiod and light intensity of the field site. Lab plants were simply moved into the new chamber in their planting trays, and to minimize the difference in soil conditions between lab and field environments, field rosettes were planted along with the soil from their planting cell. Translocated plants were left to grow, reproduce and eventually senesce. Reproducing plants were monitored every two days until death.
Measurements of longest living leaf—a surrogate for rosette biomass [43], stalk height, stage of flower formation (visible bud, anthesis, mature flower, etc…), fruit maturation and branch initiation were performed on growing plants once every 4-6 days for all plants. Once they had senesced, plants were taken to the lab, measured, and harvested. At harvest, fruits were measured and removed to storage vials.
-
(iii)
Traits measured
Seven traits were assessed in this manipulation: three phenological traits, two fruiting and two seeding traits. The three phenological traits were days to first flower, size at first flowering and flowering duration. ‘Days to first flower’ is simply the number of days between bolting and the formation of the first flower. ‘Size at first flowering’ is the distance between the base of the stalk and the pedicel of the emerging terminal flower bud (measured by Vernier caliper–error: +/- 0.01 mm); this is the plant’s height as the first flower forms. Flowering duration is the number of days between the formation of the first flower and the maturation of the last flower (as it becomes a fruit).
The two fruiting traits included branches per plant and the total number of fruits produced. “Branches per plant” was the simple count of rami protruding from the main stalk. Upon death of a plant, fruits were counted and fruit location (on branch or reproductive stalk) recorded. The “total number of fruit” produced by a plant included all fruits at all positions, both on branches and the main reproductive stalk.
The two seed traits included seed size and the total number of seeds produced. To obtain a sufficient sample size while ensuring adequate replication, seed traits were obtained from a subsample of individuals from the June and September (early and late) bolting groups. Seed size was measured by: i) sampling the ten fruits at the 100th/90th/80th/&c… percentile position along the raceme–all fruits were used if there were fewer than 10 in total; ii) imbibing seeds on a moistened filter paper-lined petri dish for 72 hours; and iii) measuring seed dimensions using NIHimage 1.62b7. Seed number per fruit was determined by manual count under a light microscope. Estimates of total reproductive output (the number of seeds per plant) were calculated as the product of the mean number of seeds per fruit and the number of fruits per plant.
-
(iv)
Statistical analyses
Our aim was to assess the effect of bolting month as a predictor in a multivariate response; however, we first tested whether differences in any reproductive traits were explained by the genotype random effect. We used likelihood ratio tests to compare the proportion of total variability in response accounted for by two models [46]. The first model was a generalized linear model (GLM) that included only fixed effects and their interaction terms; and the second was a generalized linear mixed model (GLMM) that included the fixed effects and interaction effects from the first model, as well as genotypic lineage—which we considered a random effect—along with two interaction terms that included genotype. We then compared the restricted log likelihood values for each of these models to assess whether the inclusion of genotype significantly increases the predictive power of the model. This process was repeated for each of the seven reproductive traits measured. Where the GLMM did not significantly differ from the GLM in terms of the proportion of total variation explained by the model, we dropped the random effects, opting instead for the more parsimonious model. Below (Model 1 and 2) are the specifications for the two models we used.
For each of the seven reproductive traits, we constructed a GLM that included six predictors, and used a Poisson distribution and a log link. We included four fixed effects: year (categorical), bolting month (categorical), environment (i.e. lab or field–categorical), plant size (continuous, based on prebolting rosette leaf length), and two crossed effects: bolting month X year and environment X year. Effect sizes were measured using partial η2. Our full GLM model is below (Model 1).
Model 1. GLM
REPRODUCTIVE TRAIT =
BOLTING MONTH + SIZE + ENVIRONMENT + YEAR + BOLTING MONTH*YEAR + ENVIRONMENT*YEAR
We also constructed a GLMM for each variable, which included: (1) all six factors from the GLM; as well as (2) “genotypic lineage”, a random effect; and (3) “genotypic lineage X environment”, “genotypic lineage X bolting month”, and “bolting month X year and environment X year”, three interaction effects. Our full GLMM model is specified below (Model 2).
Model 2. GLMM
REPRODUCTIVE TRAIT =
BOLTING MONTH + GENOTYPE + SIZE + ENVIRONMENT + YEAR + GENOTYPE*ENVIRONMENT + GENOTYPE*BOLTING MONTH + BOLTING MONTH*YEAR + ENVIRONMENT*YEAR
This analysis contains biologically relevant random crossed effects (i.e. GENOTYPE*BOLTING MONTH), and we constructed the GLMM so that parameters were fitted using REML-based estimation, to avoid underestimation of the standard deviation of random effects [47, 48].