Sources and consequences of seed size variation in Lupinus perennis (Fabaceae): adaptive and non-adaptive hypotheses¹

Natural history and site description

Lupinus perennis L. (Fabaceae) occurs on well-drained sandy soils across the northern United States often in oak savannas and pine barrens (Dirig, 1994; Celebrezze, 1996). An indeterminate herbaceous perennial, L. perennis reproduces both vegetatively and by seed. Several stems may emerge from a single rhizome (Dirig, 1994), resprouting shortly after snowmelt in April. Flowering begins and peaks in late May, although some individuals continue to bloom through June, especially on secondary flower stalks. The self-compatible flowers occur on a raceme that matures from the base up. Bees (Bombus spp.) are the most common floral visitors (Bernhardt, 2000; S. Halpern, personal observation), but others include ants, butterflies, beetles, and thrips (S. Halpern, personal observation). Fruits and seeds begin to mature in June and most plants enter seasonal dormancy by early August. Fruits contain 0–7 filled seeds, which disperse short distances via explosive dehiscence. Seeds may germinate the summer they mature or in subsequent springs (S. Halpern, personal observation), and can persist at least 3 yr in the seed bank (Zaremba, 1992, as cited by Celebrezze, 1996).

I established an experimental population in a common garden at the Aldo Leopold Memorial Reserve (LMR), a 567-ha privately owned and cooperatively managed reserve located in Sauk County, Wisconsin, USA (43°33′ N, 89°40′ W). The LMR includes restored prairies, forests, marshes, and a former old-field common garden site used in this study. The garden site slopes with an eastern exposure, and is fenced to reduce deer browse; otherwise, damage to flowering stalks is extensive (S. Halpern, unpublished data). Regionally, the soil is sandy-gravely till deposited during the Wisconsin glaciation (Clayton and Attig, 1990); the garden site lies on Gotham loamy sand soils (Grundlach, 1980) that consist of 87% sand, 9% silt, and 4% clay (Hunt, 1987; Zolidis, 1987). The plots used in these experiments were treated with the herbicide Round-Up and tilled before initial plantings to reduce competition from weeds.

The climate at the LMR is typical temperate continental. From 1971 to 2000, mean temperatures ranged from −10.6°C in January to 20.6°C in July, and annual precipitation averaged 85.8 cm, with 70% arriving from April to September and about a third arriving during the peak growing season of L. perennis, from mid-April to mid-July (Wisconsin State Climatology Office, 2003).

Sources of variation in seed size

I examined sources of variation in seed size using 2-yr-old maternal plants grown in the LMR common garden. Maternal plants had grown from seeds that originated from several small populations within 30 km of the LMR site (S. Weber, Bluestem Farm, personal communication). I planted 10 blocks of three 1.5 × 1.5 m plots in April– May 1999. Each plot contained 36 individuals planted in 6 rows spaced 20 cm apart, resulting in 16 “center” plants (surrounded by conspecifics on all sides) and 20 “edge” plants per plot. Paths approximately 80 cm wide separated plots. During the 1999 growing season, I periodically weeded and watered plots to facilitate establishment of the 1080 plants.

I manipulated several aspects of the 2-yr-old maternal plants'growing environment during the 2000 growing season to test the effects of these conditions on the mean and variance in seed size. I manipulated resources indirectly by weeding within and around plots and mowing paths between plots, creating differences in inter- and intraspecific competition for plants on the edge of plots compared to those in the center. Specifically, center plants grew at higher conspecific densities and likely experienced greater intraspecific competition. I simulated herbivory and directly manipulated resources by clipping 50% of each new leaf weekly from five edge plants in each plot. Plants thus grew in one of three resource treatments: edge clipped, edge unclipped, or center unclipped. Plots also received an ambient or dry water treatment. Ambient plots received approximately 46.5 cm of natural rainfall over the three-mo growing season; from 1971 to 2000, mean precipitation during the same period was 29 cm (Wisconsin State Climatology Office, 2003). In dry plots, I withheld rainfall using temporary plastic rain-out shelters, deployed only during rainfall events. These shelters drained into gutters directed away from plots. Shelters intercepted 43 cm (92%) of the ambient rainfall and reduced soil moisture content in the top 70 cm by 37% (Halpern, 2003). However, physiological symptoms of water stress were not evident in adult plants in this experiment (Halpern, 2003), probably because their long tap roots accessed deep water resources. Although it appears that adults did not experience water stress, I include water treatment as a factor in analyses because it was a design element.

From 16 June to 7 July 2000, I collected all mature fruits as they ripened from all adults. Fruits from each maternal plant were stored separately and sorted into three harvest date categories: early (16–23 June), mid (23–30 June), and late (July). Seeds were stored in an air conditioned laboratory until March 2001, when they were cold-stratified at 4°C for at least 13 d before weighing. Using stratified random sampling, I selected 59 maternal plants distributed among the different water and resource treatment combinations. For each maternal plant, I counted the number of seeds produced and weighed up to 180 haphazardly selected seeds on a microbalance (Mettler AT261 DeltaRange, Columbus, Ohio, USA) at an accuracy of 0.01 mg; seeds were equally distributed among all harvests represented in that plant. I weighed all seeds for plants that produced ≤30 seeds; I weighed >30 seeds for the 40 plants included in the following experiment, which examined the effects of the size of a seed on the fitness of the individual that grew from it.

Consequences of variation in seed size

To determine the effects of seed size on seedling performance and fitness, I planted individually weighed and hand scarified seeds from 40 randomly selected maternal plants into plots in the same common garden site. The maternal plants included an equal number from dry and ambient plots, and from edge and center positions in the plots. To obtain additional data on germination, I planted 2 seeds/position (spaced 1 cm apart) from fecund families; if both seeds germinated, I randomly pulled one and measured all subsequent plant traits on the remaining individual. I planted seeds 8 cm apart in a hexagonal array on 18–19 April 2001. Each 1.7 × 1.7 m plot included six replicates per maternal family, in a completely randomized design. The experiment included 40 families × 6 replicates · family⁻¹ · plot⁻¹ × 3 plots/block × 5 blocks, for a total of 3600 plants; double planting led to an additional 1600 seeds from which germination data were collected and analyzed.

Plots received one of three water treatments: watered, ambient, or dry. Ambient plots received approximately 28 cm of natural rainfall over the 3-mo growing season. Watered plots received supplemental watering by hand 7 times, for a total of approximately 7 cm additional water. To avoid run-off, I watered at least 3 d after a natural rainfall event. Watered and ambient plots did not differ in soil moisture or in plant traits (Halpern, 2003), so I combined these treatments in all analyses. In dry plots, I again used temporary rain-out shelters, which intercepted 22 cm (79%) of precipitation and reduced soil moisture in the top 30 cm by 25%.

I surveyed individuals from April 2001 to June 2002 to estimate fitness effects of seed size in different moisture environments during a plant's first two growing seasons. I measured several components of fitness: germination (yes/no) and its timing, plant size in years 1 and 2, plant survival through years 1 and 2, and reproductive success. I surveyed for germination every 2– 3 d from 22 April to 24 May, and then approximately once per week through mid-July. I measured germination rate as 1/number of days to germinate; a germination rate of 0 indicates the seed never germinated. After thinning double planted seeds, I surveyed for survival and counted the number of true leaves present on each plant on 17–20 May, 6–15 June, 29 June–5 July, and 12–15 August (a total of 3388 individuals, after excluding those damaged by animals or damaged during handling). I recorded presence of plants in May 2002, and separately harvested and dried reproductive and vegetative aboveground biomass in June 2002. I weighed dry vegetative and reproductive biomass and counted all fruits and seeds.

Data analysis

To describe variation in seed size in this population, I plotted histograms of seed size for the entire data set and separately for each maternal plant. I also tested these distributions for normality, kurtosis, and skew. I calculated the mean seed mass and the coefficient of variation in seed mass for each maternal plant. In separate analyses, I used linear regression to assess the relationship between seed number and either mean seed mass or the coefficient of variation in seed mass, weighted for the number of seeds measured for an individual plant.

Variation in seed size

Seed mass was highly variable in the 5839 seeds weighed from 59 maternal plants. Seed mass was normally distributed (Kolmogorov-Smirnoff-Lillifors D = 0.008, P > 0.15) with a fivefold range in magnitude, from 8 mg to 41 mg (Fig. 1). The distribution is mesokurtic (kurtosis = 0.27) and is not significantly skewed (skew = −0.08). Within individual plants, seed mass varied 1.2-fold to 4.5-fold, with the range of masses spanning 4–27 mg and the coefficient of variation ranging from 0.06–0.26. Seed size was normally distributed in 46 of the 59 individuals (78%). In 9 of the 13 individuals with non-normal distributions of seed mass, the distribution was skewed strongly to the left, while 11 had strongly leptokurtic distributions (i.e., heavy tails).

Sources of variation in seed size

Within plants, this study detected no relationship between maternal parents'seed number and seed size (weighted linear regression, β = 3.8 × 10⁻⁶, t = 1.04, P = 0.304) or between seed number and the coefficient of variation in seed size (weighted linear regression, β = −3.8 × 10⁻⁵, t = 1.06, P = 0.295; Fig. 2). Therefore, this study did not detect a trade-off between seed size and seed number, or a change in seed size variation with increasing seed number.

Water availability, resources (competition and clipping), and harvest date did not significantly affect mean seed size (Table 1). Because not all individuals matured seeds during all three harvest periods, I also tested for the effects of water availability and resources on mean seed mass within harvest periods. These analyses also detected no statistically significant effects of maternal growing environment on seed mass (results not shown). Maternal plants varied substantially in mean seed mass (Table 1), and the effect of harvest period on mean seed mass varied among families; in other words, timing of seed maturation had different effects on seed mass among maternal plants. In addition, seed number was not significantly affected by water availability (F_1,55 = 1.58, P = 0.228), competition (F_1,55 = 1.94, P = 0.169), or clipping (F_1,55 = 0.98, P = 0.327).

In contrast, maternal growing environment did affect variation in seed size. Weighted ANOVA detected an effect of resource availability on the coefficient of variation (CV) in seed size (F_2,55 = 8.17, P = 0.0008). Edge plants (with reduced intraspecific competition) had lower CV in seed size than center plants (CV_edge = 0.096, CV_center = 0.123, planned contrast F_1,55 = 12.13, P = 0.0009). Although seeds from clipped plants also were more variable than seeds from unclipped plants (CV_clip = 0.099, CV_control = 0.096) this difference was not statistically significant (planned contrast, F_1,55 = 0.13, P = 0.78). These results suggest that intraspecific competition increases variation in seed sizes produced by an individual plant. It is possible that clipping also increases variation in seed size; the power to detect such differences was low in this study. The effect of water treatment on seed size variation also was not statistically significant (ANOVA, F_1,55 = 0.12, P = 0.73); plants in ambient conditions had slightly greater CV than plants in dry conditions (CV_ambient = 0.107, CV_dry = 0.105). The correlation between mean seed mass and the CV in seed mass was not significantly different than zero (r = −0.084, P = 0.528, N = 59)

Fitness consequences of variation in seed size

Seed mass affected all fitness components measured up to reproductive success (Tables 2, 3). For germination probability and germination rate, the effect of seed mass differed in the dry and ambient treatments (seed mass × treatment interaction; Tables 2A, 3). In the ambient environment, a 10-mg increase in seed mass tripled the odds of germination, while a 10-mg increase in seed mass had negligible effects in the dry treatment (Table 4A). A similar pattern occurred in germination rate: larger seeds germinated earlier in the ambient treatment, while the positive effects of seed mass on germination timing were smaller in the dry treatment (Fig. 3). Significance tests and effect sizes were similar when I reran germination analyses excluding double planted seeds (results not shown). Overall, germination was higher in the ambient treatment (92%) than in the dry treatment (61%). Larger seeds also had a higher probability of survival: a 10-mg increase in seed mass approximately doubled an individual's probability of survival (Table 4A). After accounting for deaths in the previous stage, a 10 mg increase in seed mass still doubled an individual's odds of surviving to July, although it had no effect on the probability of surviving the winter (Table 4B). The probability of flowering also doubled when considering all seeds planted. For just plants that survived the winter, increasing seed mass by 10 mg nearly doubled the probability of flowering in the ambient treatment but had a negligible effect in the dry treatment (Table 4B). Finally, increasing seed mass was associated with increasing aboveground plant size in both years 1 and 2 after excluding plants that had died. In contrast, seed mass had no detectable effect on reproductive biomass for those individuals that flowered (Fig. 3).

When detected, whole-plot effects (block and treatment) were strongest in the first year (Tables 3, 5). Except for maximum leaf number, all first-year traits were affected by treatment. A smaller proportion of seeds in the dry treatment germinated and survived when all individuals were included in the analysis (Table 5A). Seeds in the dry treatment also germinated later. When only survivors were included in the analysis, treatment effects were not detectable after July survival (Table 3, 5B). Therefore, the strong treatment effects arise primarily from impacts on early fitness components, germination, and establishment. If juveniles survived the first growing season, effects of water limitation dissipated and were no longer detectable in terms of biomass or flowering. The only fitness component that varied among blocks was germination probability (Table 5A). However, treatment effects varied significantly among blocks for germination probability, germination rate, and maximum leaf number. This interaction suggests that the effects of water environment on these traits differed over the relatively small spatial scale of this experiment (7 × 11 m in total).

Selection on seed size

This experiment detected strong selection for increasing seed size (β = 0.30, SE = 0.14, P < 0.0001, N = 2996). Standardized linear selection gradients predict a shift of 0.30 standard deviations in mean seed mass per generation of selection. This analysis detected neither a treatment (F_1,4 = 0.10, P = 0.77) nor a treatment-by-trait interaction (F_1,2284 = 0.12, P = 0.73). Quadratic selection gradients were not significantly different from zero (β = 0.05, SE = 0.05, P = 0.29, N = 2996). They provide no evidence for curvature in the fitness function, suggesting that selection on seed size is directional rather than nonlinear. Again, no treatment effect (F_1,4 = 0.08, P = 0.79) or trait-by-treatment interaction (F_1,2983 = 0.10, P = 0.76) were detected.

Sources of variation in seed size

The fivefold variation in seed size observed in this study was not associated with environmental conditions. Mean seed size differed substantially among individuals but not among maternal growing environments or seed maturation times. Many other studies similarly have found individual differences in seed size (e.g., Dolan, 1984; Thompson, 1984; Fenster, 1991; Winn, 1991; Krannitz, 1997; Mendez, 1997; Vaughton and Ramsey, 1997, 1998; Castro, 1999; Simons and Johnston, 2000), which may arise from environmental maternal effects (Platenkamp and Shaw, 1993; Gutterman, 2000), genetic maternal effects (Byers et al., 1997), or both (Hereford and Moriuchi, in press). In contrast to this study, differences in seed size among environments are common, likely due in part to low heritabilities and strong environmental maternal effects in this trait (Platenkamp and Shaw, 1993; Byers et al., 1997). However, the effect of a particular environment is not clearly predictable across systems. For example, similar resource manipulations can affect mean seed size in different directions (Platenkamp and Shaw, 1993; Vaughton and Ramsey, 1998; Galloway, 2001): artificial defoliation reduces mean seed size in some species (Vaughton and Ramsey, 1997, 1998) but not in others (this study, Ågren, 1989; Krannitz, 1997); maternal plant size positively affects seed size in some species (Dolan, 1984) but not others (Krannitz, 1997; Vaughton and Ramsey, 1997); and seasonal effects on mean seed size vary from substantial (Fenster, 1991; Winn, 1991; Byers et al., 1997; Vaughton and Ramsey, 1998) to undetectable (Krannitz, 1997; Vaughton and Ramsey, 1997).

In contrast to mean seed size, maternal growing environment affected variability in the seed sizes produced by an individual L. perennis plant. Seed size variation was greater for plants growing at higher compared to lower densities of intraspecific competitors; a similar trend related to clipping was not statistically significant. These results are consistent with a scenario in which environmental conditions affect control over seed size. An increase in trait variation under stress usually is interpreted as maladaptive developmental instability (reviewed in Simons and Johnston, 1997). Therefore, variation in seed size in L. perennis may occur in part due to constraints on equitable seed provisioning when resources become more limiting. It is also possible that developmental instability might be a mechanism for producing adaptive variation (Simons and Johnston, 1997), and differences in variation between treatments occur because its expression changes across environments. Determining whether the observed variation is adaptive would require information about the relationship between fitness and seed size variation as a parental trait, which this study did not test.

In fact, variation in offspring size as a parental trait has received much less empirical attention than identifying sources of within-plant variation in seed size, and few studies have explicitly considered the effects of maternal environment on seed size variation within a plant. In plants, Krannitz (1997) found that a stressor, proportion of twigs browsed, was associated with variability in seed size in Purshia tridentate (Pursh) DC. Recent studies in trematodes (Poulin and Hamilton, 2000) and fish (Koops et al., 2003) also have explicitly studied variation in offspring size among environments. By correlating this variation with estimates of environmental predictability, Koops et al. (2003) argued that egg size variation within female trout arises as part of a bet-hedging strategy in unpredictable environments. Similar empirical work with plants could help clarify whether and when an individual that produces more variable seeds is favored.

Consequences of differences in seed size

For offspring, fitness consequences of seed size are dramatic and persistent in Lupinus perennis. Most fitness components measured in the first two seasons were affected by seed size, although some of those effects weakened over time or after accounting for mortality in earlier stages. Larger seed size may allow for faster seedling growth, giving individuals a head start that carries over into greater survival and reproduction in subsequent years. For L. perennis, a head start may be important in terms of root establishment. Adult L. perennis avoid water stress despite growing in sandy, drought-prone soils, apparently because their long tap-roots reach deep water sources (Dirig, 1994; Halpern, 2003). Plants that reach deep water faster, having grown from larger seeds, may photosynthesize and grow continuously, while smaller plants from smaller seeds may be more dependent on rain and associated surface water.

Seed size effects are similarly persistent in some other plants, including Lobelia inflata, which has very small seeds (Simons and Johnston, 2000). However, seed size also is seen commonly as an important component of maternal effects whose influence tends to dissipate during the first few weeks of seedling establishment (Harper et al., 1970; Roach and Wulff, 1987). Evidence to support this conclusion comes from empirical examples where positive effects of seed size on germination and plant size only persist during early establishment (Dolan, 1984; Tremayne and Richards, 2000; but see Stanton, 1984; Simons and Johnston, 2000). The duration of seed size effects may vary in association with life history characteristics or habitat conditions, or it may be idiosyncratic. Comparative studies that control for similarities due to shared evolutionary history (Felsenstein, 1985; Martins and Hansen, 1996) could best test the importance of these potential explanations for differences in seed size effects.

In this study, it is somewhat surprising that seed size had little effect on germination probability in the dry treatment. Larger seeds often contain additional resources, which could be allocated to root growth in seedlings and thus could increase access to soil water. Indeed, L. perennis seeds that germinate in drier soils in the greenhouse have longer and heavier roots (Halpern, 2003). Alternatively, smaller seeds could be favored if they affected root:shoot ratios, and hence water losses to transpiration under drought conditions (Hendrix et al., 1991). Because seed size can affect dormancy (Murdoch and Ellis, 2000), it is possible that smaller seeds were likely to remain dormant in the dry treatment. In this experiment, however, it seems unlikely that seeds in the dry treatment entered the seed bank and remained viable: all seeds were hand-scarified, and I observed no new germinants in spring 2002, when treatments were no longer applied. Seed size can also affect probability of seed predation (Reader, 1993; Moegenburg, 1996; Alexander et al., 2001; Sousa et al., 2003). However, L. perennis seeds are not attacked by pre- or post-dispersal predators at this site (S. Halpern, unpublished data), so it is also unlikely that greater seed predation influenced germination rates differentially among treatments.

The large fitness benefits of increasing seed size did not differ between environments. The impact of seed size on fitness did not vary among ambient and dry environments for most fitness components, and selection on seed size did not differ among ambient and dry environments. Thus, this study provides little support for conditions underlying one common adaptive explanation for seed size variability, that temporal or spatial variation in establishment conditions creates a mosaic of optimal seed sizes on the landscape (Janzen, 1977; Capinera, 1979; reviewed in McGinley et al., 1987). These results contrast with several studies that manipulate environmental characteristics in other systems; for example, Eriksson (1999) found some evidence that the positive benefits of larger seed size increased in Convallaria majalis L. when seedlings grew at higher densities of a competitor, while Wulff (1986b) reported differences in the magnitude of effects of seed size on seedling survival among habitat types for Desmodium paniculatum (L.) DC.

This study only measured the fitness consequences of seed size for offspring. Only one empirical study has simultaneously estimated selection on seed size for both parent and offspring (Mojonnier, 1998). She found high heritability but no selection on seed size for maternal plants of Ipomoea purpurea (L.) Roth, and low heritability but strong positive directional selection for seed size for offspring plants. If the same pattern holds in L. perennis, variation in seed size could persist because it is a neutral trait in terms of the maternal plant's fitness.

Seed size vs. number trade-off?

Life history theory predicts that resource limitations induce a fundamental genetic trade-off between number of offspring and offspring size (Smith and Fretwell, 1974), but evidence of trade-offs between seed size and number in plants is inconsistent. A genetic or phenotypic trade-off has been reported in numerous studies (e.g., Ågren, 1989; Mehlman, 1993; Byers et al., 1997; Vaughton and Ramsey, 1997, 1998; Eriksson, 1999; Simons and Johnston, 2000) but not in others (e.g., Schaal, 1980; Mendez, 1997; Vaughton and Ramsey, 1997; Galloway, 2001; Karrenberg and Suter, 2003; reviewed in Venable, 1992), including this one (Fig. 2a). It is possible that differences in microenvironment or plant size affect resources available for seed provisioning, thus obscuring actual trade-offs among plants (van Noordwijk and de Jong, 1986; Venable, 1992). In this study seed number increased with plant size (measured as leaf number) (r = 0.22, P = 0.006, N = 151), although mean seed mass was not associated with plant size (r = 0.09, P = 0.65, N = 24). Therefore, this study provides no evidence for a trade-off but cannot rule one out. Similarly, plant modules such as racemes or fruits may experience different resource environments and therefore provision seeds differently. For example, Mehlman (1993) observed a size-number trade-off among fruits within individuals but not among individuals, and Simons and Johnston (2000) only detected a size-number trade-off after controlling for fruit size. Although the effects of position and plant modularity on seed size are well recognized (reviewed in McGinley et al., 1987), the implications for the evolutionary ecology of seed size and life-history trade-offs have not received as much attention.

Conclusion

The fivefold variation in seed sizes observed in this study with Lupinus perennis contributes to important and long-lasting differences in fitness among offspring. In addition, stressful conditions increase the variability in the size of seeds an individual produces. This pattern could arise from a lack of parental control over seed size that is magnified when resource limitation increases. Few studies have explicitly examined the causes of seed or egg size variation within an individual (but see Poulin and Hamilton, 2000; Koops et al., 2003), despite its potential importance in life history and seed evolutionary ecology. Although they present challenges, such studies could contribute to new insights about the evolution of seed size variation. Given the large differences among plants in their seed size variation, it also might be possible to examine selection on seed size variation per se in various environmental contexts, thereby explicitly testing the premise that increased variation is favored under some conditions.