1Seed mass is widely regarded as a key plant trait, yet is notoriously variable within an individual species. This study reports the first field experiments to disentangle the relative importance of four major correlates of seed mass variation: resource limitation; plant height; seed number; and seasonality.
2Variation in reproductive traits in Impatiens glandulifera Royle was partitioned among six populations established across an elevation gradient, among plants at each elevation and within individual plants.
3Seed mass was the least variable trait both at the individual and population levels, while the number of pods per plant varied most. Seed mass variation was greatest among individual plants (42%) and was least among populations (28%).
4Taller plants produced proportionally more seed pods and more seeds per pod. Seed mass was not related to the number of pods per plant, or the number of seeds per pod. Thus with increasing resources, I. glandulifera allocated resources to produce more seeds rather than larger seeds.
5Seed mass was negatively correlated with the heat sum over the period of seed maturation, explaining increased seed mass at higher elevations and increasing seed mass over time at most elevations. This may indicate that environmental condition pre- and post-fertilization determines the relationship between seed mass and elevation, thereby limiting the potential for evolutionary selection for seed-size optima.
6The study warns against correlative studies of seed mass variation that select arbitrary populations or ignore the temporal dynamics in seed mass, as these may generate spurious correlations among life-history traits.
7The success of I. glandulifera as an invasive species may reflect the extended period of seed release, considerable seed mass variation and large seed size under conditions of environmental severity. These traits will facilitate the exploitation of spatially and temporally heterogeneous natural environments commonly found in riparian ecosystems.
1Seed mass should decrease (and its variation increase) with increasing resource limitation due to increased sibling competition.
2Seed mass should increase (and its variance decrease) with increasing plant height as a result of an allometric relationship between these variables (Thompson & Rabinowitz 1989).
3Seed mass should decrease (and its variance increase) with total number of seeds produced because, for a given amount of resources, increases in seed mass may come at the expense of producing fewer seeds.
4Seed mass should decrease (and its variance increase) through the season (due to an ever shorter period available for seed maturation and/or asymmetric competition for resources from seeds formed earlier).
The considerable literature on seed mass variation provides support both for and against these hypotheses, and reflects the intricate interplay between resource competition and pollination rates, physiological responses and development constraints (Leishman et al. 2000). Current insights into seed mass variation rely heavily on correlative rather than experimental approaches, and no comprehensive field tests of all four hypotheses exist. Rather, present knowledge is drawn from a catalogue of disparate studies each tackling different hypotheses, often on markedly different plant species. This has led to conflicting results and has limited the opportunities for synthesis. This study aims to address this heterogeneous background by testing all four hypotheses simultaneously in order to provide a foundation for an integrated, functional understanding of seed mass variation.
To test these predictions requires variation in seed mass to be partitioned among populations spanning a marked environmental gradient, among plants within these populations, and within the individual plants themselves. Although several studies have attempted to partition seed mass variation, most have examined only a single population (Winn 1991; Mehlman 1993; Obeso 1993; Eriksson 1999). Where more than one population has been examined, only occasionally has a study covered an explicit environmental gradient (Ågren 1989; Stocklin & Favre 1994; Susko & Lovett-Doust 2000). Even these studies suffer from a correlative approach that may confound effects due to local adaptation (Susko & Lovett-Doust 2000), or fail to tease apart the impacts on seed mass of multiple covarying environmental gradients such as shade (Wulff 1986; Ågren 1989); plant density (Winn & Miller 1995); soil nutrient status (Wulff 1986); soil pH (Susko & Lovett-Doust 2000). In the past these limitations have been addressed by collecting seeds from the field and examining allocation patterns of the resulting plants under laboratory conditions (Wulff 1986; Winn & Gross 1993; Susko & Lovett-Doust 2000). However, extrapolation of results back to the field is often problematic (Wulff 1986). The foregoing highlights the piecemeal state of the current understanding of seed mass variation and the need for a different approach.
The present study reports the results of a novel approach in which patterns of seed mass variation were examined in plants arising from the experimental sowing of Impatiens glandulifera Royle (Balsaminaceae) along a marked climate gradient. Our aims were to:
1test the hypotheses regarding seed mass variation within and between individuals as well as among populations distributed across a climatic gradient;
2assess the extent of compensation in seed allocation in response to resource limitation;
3elucidate the environmental drivers of variability in seed mass.
Impatiens glandulifera (hereafter Impatiens) is suitable for such analyses because of its discrete packaging of seeds within pods, its annual life history, and the important role climate plays on plant performance (particularly reproductive traits) and distribution (Collingham et al. 2000; Willis & Hulme 2002).
Materials and methods
Impatiens glandulifera is the tallest (up to 3 m) annual species in the British flora (Grime et al. 1988), having become naturalized in the late 19th century following its introduction from Asia (Beerling & Perrins 1993). Detailed information on the floral biology can be found in Beerling & Perrins (1993). Spurred, helmet-like flowers are produced in few-flowered racemes in leaf axils. Flowers produce substantial volumes of nectar and are highly attractive to pollinators, often to the detriment of neighbouring species (Chittka & Schurkens 2001). Although flowers are hermaphrodite and self-compatible (Valentine 1978), widespread Bombus spp. readily visit the flowers and deposit sufficient pollen to result in often 100% seed set (Titze 2000). Flowers and resulting multiseeded capsules (pods) are produced sequentially over a period extending from July–October When ripe the capsules explode at the slightest touch, dispersing seeds ballistically over a radius of 5 m. Seeds are among the largest for annual plants in the British flora, and weigh on average 22 mg (range 2–35 mg). Pods may contain between four and 16 seeds with, on average, between five and six ripe seeds per pod (Beerling & Perrins 1993). Plants produce a terminal inflorescence at the start of the flowering period and then, given adequate resources and space, can subsequently produce lateral inflorescences arising from branches off lower nodes of the main stem. Flowers and seed pods develop sequentially on florets within an inflorescence, with the distal flowers developing first and proximal flowers developing last, opposite to the development pattern of many plant species (reviewed by Lee 1988). In dense stands Impatiens produces almost all its seed pods on the terminal inflorescence, with very little production of pods from lateral inflorescences. However, when conditions allow, the lower inflorescences may produce the majority of seed pods.
site location and experimental design
Ripe seeds of Impatiens were collected from a single population in Durham City and sown along an elevation transect in north-east England (Table 1). Six stations were established in the River Wear catchment, including sites beyond the species’ current range (Rookhope and Dun Fell). Details of the station locations and design are described more fully by Willis & Hulme (2002). Twenty seeds were planted in 10 replicate pots containing a similar soil/compost mixture at each station, reducing much of the variation associated with natural stands. Seeds were planted in October 1997, and the resulting plants followed during the 1998 growing season. Mortality of seeds and seedlings led to similar flowering plant densities at all stations (20–80 plants m−2), which were comparable to natural densities (Willis & Hulme 2002).
Table 1. Summary statistics for each of six elevation stations with regard to the number of pods per plant, seed content per pod and seed mass for experimentally sown Impatiens glandulifera
Number of pods per plant
Population CV (%)
Seed content per pod
Population CV (%)
Mean individual CV (%)
Individual seed mass
Population CV (%)
Mean individual CV (%)
Plant performance was assessed fortnightly by recording plant height as well as seed-pod and flower production. To enable cross-pollination but prevent seed dispersal, flowers were left open until the initiation of pod formation, at which point seed pods were enclosed in muslin bags. Seeds were collected from plants on a regular basis through the summer in order to record seed numbers from all pods produced. A subsample of the seeds collected was weighed on the day of collection. Seeds and pods were recorded as originating from either the terminal inflorescence or a lateral inflorescence. At the Durham City station, the sequence of ripe pod production along terminal and lateral inflorescences was recorded, along with corresponding values for seed number and mass.
Climate data over the study period (October 1997–October 1998) were obtained from a standard meteorological station at Durham City and an automated station on Great Dun Fell. From these it was possible to interpolate daily temperature at the remaining stations and derive bioclimate variables (for details see Willis & Hulme 2002). Daily heat sums, measured as growing degree-days >5 °C (GDD5), were calculated. From these data heat sums were estimated for the period up to seedling emergence (pre-emergent heat sum, PRHS, Jan–Mar); for the growth season (post-emergent heat sum, POHS, April–September); for the entire year from sowing (ALLHS); and for periods from seedling emergence up to the day of seed set (SEEDHS). Finally, as data collected from the study stations indicated that the usual time to produce a flower, ripen the pod and set seed was ≈4 weeks, an additional variable was created summing GDD5 for the periods of flowering to seed set (4WKHS). Estimates of wind speed and precipitation would also have been desirable, but these variables could not be as reliably interpolated between stations (Hulme et al. 1995). However, at the two meteorological stations there was a strong negative correlation between GDD5 and precipitation.
Data were examined using nested anova and regression analyses (Norusis 2000). Regressions used mean data from individual plants to examine relationships (e.g. plant height and seed traits). Nested anovas were undertaken on measurements of individual seeds and pods, unless stated otherwise. To produce balanced models, the analysis of seed number per pod among stations and among plants eliminated the highest elevation station and used five randomly selected pods per plant. Analysis of seed mass between plants and stations similarly used 10 random seeds per plant with the number of stations reduced to four, eliminating the two high elevation stations. Analyses were undertaken using each of the bioclimate variables separately as explanatory variables.
partitioning variation in reproductive traits between stations and plants
Over the entire 1998 flowering period, individual plants differed across two orders of magnitude in the number of pods they produced, ranging from one to 270 (Table 1). The considerable interplant differences within individual elevation stations accounted for 72% of the variation in pod production in the experiments, although significant differences among elevation stations were still found (F4,102 = 7·63, P < 0·001, anova on log-transformed number of pods per plant). There was no significant trend with elevation in either the mean or variance in pod production, and no correlation between mean pod production at a station and interplant variation (rs = 0·021, df = 4, P > 0·1; Table 1). The number of seeds per pod ranged from one to 19 (Table 1) and varied significantly both within and among stations (nested anova: within station, F20,75 = 4·45, P < 0·001; among station, F4,20 = 11·28, P < 0·001), with 41·9 and 27·7% of variation explained within and among stations, respectively). There was no significant trend with elevation in either the mean or variance in seed number per pod, and no correlation between mean pod production at a station and interplant variation (rs = 0·009, df = 4, P > 0·1). Seed mass varied significantly within and between stations (nested anova: within station, F36,399 = 21·99, P < 0·001; among station, F3,399= 56·22, P < 0·001), with 60·0 and 13·0% of variation explained within and among stations, respectively. Seed mass distributions were all positively skewed, which was attributable to one or a few large-seeded individuals in each population (Fig. 1). A significant trend of heavier seeds at higher elevations was found (rs = 0·955, df = 4, P < 0·01; Table 1; Fig. 1). Although the coefficients of variance (CV) did not differ significantly among stations (Kruskall–Wallis df = 4, χ2 = 5·051, P > 0·1), the range in seed mass (as described by the ratio of largest to smallest seed) declined progressively from almost 12 at Durham City to less than 2 at Dun Fell (Table 1). Overall, the highest population variation in reproductive traits was found in the number of pods per plant, and the least in individual seed mass.
seasonal variation in reproductive traits
Following individual plants throughout the 2 month fruiting period highlighted changes in seed output and quality over time. Pod production was initiated and subsequently peaked progressively later with increasing elevation (Fig. 2). Although facing a shorter season at higher elevations, plants at all but the highest elevation compensated by producing, on average, more pods per unit time once pod production began. Thus while, in late August, a negative relationship existed between elevation and numbers of pods per plant (rs = −0·964, df = 4, P < 0·01), the trend had disappeared by late September (rs = 0·297, df = 4, P > 0·05). Number of seeds per pod varied during the fruiting period (Fig. 3), being highest at the start of the fruiting period, declining as fruiting progressed, and increasing towards the end of the period. This trend may reflect within-plant allocation patterns, as revealed by more detailed sampling at the Durham City station where, within individual plants, pods produced earlier had more seeds than later pods on the same inflorescence (t39 = 6·94, P < 0·001). Although, on average, low-elevation stations produced smaller seeds than stations at higher elevations (Table 1), seed mass varied seasonally (Fig. 4). Over the fruiting period, the mean seed mass of the lower elevation stations increased towards a maximum of ≈25 mg, such that at the end of the season mean seed mass was similar at all stations.
correlates among reproductive traits
Not surprisingly, taller plants produced proportionally more seed pods (rs = 0·85, df = 99, P < 0·001; Fig. 5a) as a result of an increased number of lateral inflorescences. Plants that produced more seed pods, however, also produced more seeds per pod (rs = 0·50, df = 68, P < 0·001; Fig. 5b). This trend arises partly as a result of pods on lateral inflorescences containing more seeds per pod than those on the terminal inflorescence (lateral mean = 8·49 ± 0·41 seeds per pod; terminal mean = 7·04 ± 0·37 seeds per pod; paired t-test, t34 = 4·87, P < 0·001). Individual seed mass was not related to either the number of pods per plant (rs = 0·17, df = 66, P > 0·1) or the number of seeds per pod (rs = 0·18, df = 61, P > 0·1). There was no difference in seed mass between seeds originating from terminal vs lateral inflorescences of the same plant at any given time (t-tests of plants with 30 or more seeds from both terminal and lateral inflorescences were all non-significant). Thus with increasing resources, Impatiens allocates resources to produce more, rather than larger, seeds.
impact of climate on reproductive output
None of the measures of heat sum (PRHS, POHS, ALLHS) explained significant variation in the number of pods produced per plant during the 1998 growing season. The heat sum over the period of flowering and seed development (4WKHS) was by far the best predictor of seed mass, and revealed a strong negative relationship between heat sum and mass of seeds produced (r2 = 0·717, F1,22 = 55·74, P < 0·001; Fig. 6).
The reproductive traits in Impatiens, as in numerous plant species (Michaels et al. 1988; Leishman et al. 2000), show marked intraindividual and intrapopulation variation. Of the three reproductive traits, seed mass was the least variable at both individual and population levels, while the number of pods per plant varied most. This pattern is consistent with the general finding that seed mass is the least variable component of yield (Harper et al. 1970; Haig & Westoby 1988). However, although seed mass was the least variable trait, it still exhibited significant variation. The mean population CV in seed mass was over 39%, and is high compared to an average of 28% for 39 plant species in eastern-central Illinois (Michaels et al. 1988). The detailed experimental approach in this study presents a unique opportunity to assess the correlates of seed mass variation and their adaptive significance.
Partitioning the variation in seed mass revealed the greatest source to be that arising among individual plants (42%) within a station, while the least was found among elevation stations (28%). Within an elevation station, over 53% of the variation in seed mass occurred among plants and, compared to an average of 38% for the species examined by Michaels et al. (1988), represents a relatively high percentage. This finding is striking for at least two reasons. First, it is contrary to what might be seen as the norm for herbaceous plants; second, it is maintained even in the face of a marked elevational trend that led to a doubling in mean seed mass and a tenfold reduction in its range (Table 1). Although no information was available on genotypic variation, marked variation in seed mass among individual plants may reflect an adaptive response to changes in resources.
At all elevations, self-thinning led to significant plant-size hierarchies. The interplant variation in seed mass could simply reflect that larger plants produce larger seeds, and it is evident that one or two plants in each population had markedly larger seeds than the rest (Fig. 1). Both the number of pods per plant and the seeds produced per pod were a function of plant height, and resulted in seed production varying by between two and three orders of magnitude among plants at the same elevation. Yet even with considerable variation in seed number, there was no evidence of a trade-off with seed mass and it appears that, with increasing resources, Impatiens tends to produce more rather than larger seeds. Although there is empirical support for a trade-off in seed mass and number across different species (Shaanker et al. 1988; Henery & Westoby 2001; Leishman 2001), within species there is evidence for (Ågren 1989; Vaughton & Ramsey 1998; Eriksson 1999; Parciak 2002); and against (Wulff 1986; Winn & Gross 1993; Mendez 1997; Simons & Johnston 2000) such a trade-off. The Impatiens data argue against such a trade-off, as plant size strongly determined seed number but not seed mass. By simultaneously assessing the covariance among reproductive and somatic traits and controlling for resource limitation, this study provides the first experimental support for models of optimum seed mass that predict plants responding to changes in resource supply by a change in seed number rather than mass (Haig & Westoby 1988).
While seed mass did not vary with either plant height or seed number, a significant trend of increasing seed mass with higher elevations was found. Could such variation reflect selection on individual plant traits? Seed mass has previously been found to increase (Hurka & Benneweg 1979; Oyama 1993; Lord 1994; Murray et al. 2003); decline (Baker 1972; Totland & Birks 1996; Kollmann & Pflugshaupt 2001; Murray et al. 2003); or be invariant (Greimler & Dobes 2000; Buckley et al. 2003; Murray et al. 2003) with elevation. This indicates that the selective advantages of a large seed mass in the face of environmental severity at higher elevations (Hurka & Benneweg 1979; Lord 1994) are finely balanced by a shorter growing and seed-maturation period (Baker 1972). However, previous studies of seed mass variation along elevation gradients have not assessed the role of climate on seed mass. This study highlights that lower heat sums at higher elevations explain most of the variation in seed mass along the elevation gradient, and the trend is consistent with laboratory studies that have revealed low temperatures to increase seed mass (Wulff 1986; Lacey 1996). The timing of changes in resource supply may be crucial and, if occurring after seed number is determined, a response in seed mass might be expected (Harper et al. 1970; Haig & Westoby 1988; Eriksson 1999). Post-zygotic parental temperatures strongly determine seed mass (Lacey 1996), and this may explain why, in Impatiens, the heat sum over the 4 weeks preceding seed maturation rather than over the entire growing season was most strongly correlated with seed mass. This effect of temperature was seen both in comparisons of mean seed mass across the elevation gradient and also over time within a population. Thus any adaptive explanations regarding improved performance at higher elevations are not convincing arguments for seasonal variation within an elevation. Why, then, does seed mass sometimes decline with elevation? A tentative hypothesis regarding declines in seed mass with elevation is that, in these species, allocation patterns are determined pre- rather than post-fertilization (McKee & Richards 1996). This may explain why declines in seed mass with elevation can coincide with a reduction in seed number (Totland & Birks 1996; Kollmann & Pflugshaupt 2001). However, it should also be borne in mind that simple differences in the timing of seed collection in correlative studies may produce artefactual relationships between seed mass and elevation. So, while adaptive explanations may be sought for such trends in seed mass, these must be viewed in the context of physiological and developmental constraints relating to direct temperature effects. Whether the variations in seed traits driven by such constraints actually result in increased, decreased or invariant consequences for success requires further work.
It appears that Impatiens does not fit the expectation as to how resources might influence seed mass variation (Lee 1988; Michaels et al. 1988; Leishman et al. 2000). Seed mass increased, rather than decreased, with increasing resource limitation at higher elevations; was not related to plant height or fecundity; and seeds produced late in the season were often larger than the first seeds to mature. In addition, while intraindividual seed mass variation was high, there were marked differences in seed mass among individual plants. However, rather than obscure the search for general trends, the experimental approach and detailed monitoring have significantly advanced understanding of the interplay between allocation patterns and the environment. Of particular importance is the identification of climate as a crucial influence on both mean and variance in seed mass. The study also warns against correlative studies of seed mass variation that fail to account for environmental gradients by selecting arbitrary populations, or that ignore the temporal dynamics in seed mass by sampling populations on a single occasion. Had this study followed a correlative approach using natural populations, it is likely that trends in increased seed mass at higher elevations and significant individual variation in seed mass would probably have been detected. However, this combination of trends could lead to adaptive explanations of increased selection for larger seed mass at higher elevations. The experimental approach adopted here argues against an adaptive explanation. Thus, while individual plants may vary in mean seed mass, this does not appear to be related to increased fitness (plant size and fecundity), nor to increased survival in the face of environmental severity (Moles & Westoby 2004). Rather, the results reiterate increasing evidence that plasticity in seed mass may be as important a life-history characteristic as mean seed mass, and comparative analyses based solely around means and not variances may miss fundamental aspects of plant ecology (Geritz 1995). For example, the success of Impatiens as an invasive annual species may reflect intraindividual variation in seed mass that potentially facilitates the exploitation of heterogeneous environments (Venable & Brown 1988; Rees & Westoby 1997), especially in wetland ecosystems (Yanful & Maun 1996). Thus the extended period of seed release, considerable variation in seed mass, and increased seed mass in response to environmental severity are all likely to maximize the opportunities for colonization and establishment in this invasive species.
We would like to thank Northumbrian Water and English Nature for their permission to site study plots in enclosures on their land. Thanks also to Andy Joyce for his assistance in interpolating the climatic data. This research was funded by the Natural Environment Research Council through an award to P.E.H. under the Large Scale Processes in Ecology and Hydrology Thematic Programme.