QUANTITATIVE GENETICS AND THE PERSISTENCE OF ENVIRONMENTAL EFFECTS IN CLONALLY PROPAGATED ORGANISMS

Transcription

1 Evolution, 54(2), 2000, pp QUANTITATIVE GENETICS AND THE PERSISTENCE OF ENVIRONMENTAL EFFECTS IN CLONALLY PROPAGATED ORGANISMS KENT E. SCHWAEGERLE, 1 HEATHER MCINTYRE, AND CHRISTOPHER SWINGLEY Institute of Arctic Biology and Department of Biology and Wildlife, University of Alaska Fairbanks, Fairbanks, Alaska ffkes@aurora.alaska.edu Abstract. Phenotype is often viewed as a product of genes and the environment in which these genes are expressed. However, numerous studies have shown that environment can cause lasting changes in phenotype that can be passed from one generation to the next, much as genes are transmitted. In clonally propagated organisms, persistence of environmental effects has been observed in a range of plant and animal species, but has rarely been the object of study. We measured the persistence and magnitude of environmental effects on phenotype over three clonal generations in the arctic sedge Eriophorum vaginatum. We found that the environment in which tillers developed had large effects on their later performance (parental effects) and that these effects were in part independent of the size of tillers. The magnitude and persistence of environmental effects did not differ between environmental treatments or among genotypes. However, after 52 weeks of growth and two rounds of clonal propagation, grandparental treatment effects were not significant. We describe methods that can be used in quantitative genetics studies of clonal organisms to reduce bias in estimates of genotypic and environmental variance and argue that the persistence of environmental effects in clonal plant material has ecological and evolutionary consequences similar to those described for maternal environmental effects in sexual organisms. Key words. C effects, carry-over effects, clonal propagation, Eriophorum, intergenerational environmental effects, maternal effects, quantitative genetics. The goal of quantitative genetics is to understand how genes and environment combine to determine phenotypic variation in populations. The simplest quantitative genetics models are built on the notion that an individual s phenotype is determined by genes inherited from parents and by environment experienced during development. However, many studies have shown that environment can cause changes in phenotype that can be passed from one generation to the next, thus influencing phenotypic resemblance among relatives much as genes do. In sexually reproducing organisms, parental environmental effects on offspring phenotype have been demonstrated in a wide range of plant and animal species (reviews in Roach and Wulff 1987; Mazer and Gorchov 1996; Mousseau and Fox 1998) and have been shown to have ecological and evolutionary consequences (Kirkpatrick and Lande 1989; Crone 1997; Ginzburg 1998). Intergenerational environmental effects in clonal organisms have received less attention. In clonal organisms, environment may bring about changes that are transmitted from cell to cell, so that phenotype of mitotically derived tissues is influenced by environments experienced by ancestral tissues. The magnitude and persistence of this somatic memory (W. S. Armbruster, pers. comm. 1998) is largely unknown and may be an important source of variation in clonal organisms. Although quantitative genetic models have been developed that account for maternal environmental effects in sexually reproducing organisms (e.g., Kirkpatrick and Lande 1989; Roff 1998; Wade 1998), clonal organisms present some unique possibilities and challenges. Ideally, asexual propagation permits estimation of genotypic variance (V G ) of traits, that portion of the total phenotypic variance (V P ) of traits attributable to genetic differences among individuals. Genotypes sampled from a wild population are asexually propagated, and phenotypic resemblance of genetically identical 2000 The Society for the Study of Evolution. All rights reserved. Received June 3, Accepted September 14, offspring is used to estimate genotypic variance (V G ) and broad-sense heritability (V G /V P ) of metric traits. But, clonally propagated genotypes provide estimates of genetic variance of traits that may be biased by intergenerational environmental effects (Reznick et al. 1986). If environmental differences among parents cause differences in phenotype of clonal offspring, intergenerational environmental effects will upwardly bias estimates of genotypic variance. Unbiased estimates of broad-sense heritability depend on experimental designs that provide independent estimates of genotypic and environmental effects. We argue that the use of clonal propagation to estimate broad-sense heritabilities is a valuable and underutilized tool in quantitative genetics. First, most quantitative genetic studies have focused on narrow-sense heritability because in sexually reproducing populations response to selection is directly proportional to the additive genetic portion of the total phenotypic variance. However, in long-lived organisms faced with a period of rapid environmental change (e.g., change associated with global warming), the persistence of populations may depend on genotypic differences among extant genotypes. Change may occur so rapidly relative to the generation time of the organism that sexual recombination is less important to population survival than the physiological tolerances of individuals. Second, some researchers have argued that broad-sense heritability may be a useful measure of a population s long-term capacity for evolutionary change because: (1) additive genetic variance may often be a major component of total genetic variance; and (2) estimates of genotypic variance are usually much more precise than estimates of additive genetic variance. Third, unbiased estimates of environmental variation (V E ), which are useful in studies of phenotypic plasticity and environmental heterogeneity (e.g., Antonovics et al. 1987; Schmid 1992), depend

2 PERSISTENCE OF ENVIRONMENTAL EFFECTS 453 on clean estimates of V G. Fourth, estimation of narrow-sense heritability requires breeding designs that may be difficult or impossible in some long-lived clonal species. And finally, extant genotypes in long-lived perennials may be particularly useful in understanding genetic trade-offs and adaptation to microenvironmental heterogeneity (Vrijenhoek 1979; Stettler and Bradshaw 1994; Vrijenhoek and Pfeiler 1997). This is because differential mortality of genotypes can lead to populations of individuals with specific combinations of genetic traits that may represent unique solutions to the problem of survival (e.g., Lynch 1984; Ayre 1985; Hartnett et al. 1987; McFadden 1997). The purpose of this study was to measure the magnitude and persistence of environmental effects in clonally propagated genotypes of Eriophorum vaginatum. To accomplish this, plants were raised in contrasting environments (treatments) for a period of time (C0 generation); tillers from these plants ( clonal offspring ) were raised in two contrasting environments (C1 generation); and finally tillers from the C1 generation (C2 generation) were grown in a common environment and examined for grandparental (C0) and parental (C1) environmental effects. The experimental design follows the protocol of many earlier studies of broad-sense heritability (e.g., Platenkamp 1990) where genotypes collected from the field (C0 generation) are brought into a uniform controlled environment (C1 generation) where they are held for a period of time to reduce or eliminate environmental effects from the field. Finally, propagules from these plants (C2 generation) are used in an experiment to estimate broad-sense heritability of traits. Our study differs from these studies in that we use experimental treatments to induce environmental effects in our C0 generation. We tested the following hypotheses. (1) Parental and grandparental environments influence phenotype of clonally propagated individuals. These effects could arise through differences in size of clonal propagules, through differences in storage of growth resources, and through changes in gene regulation that persist across clonal generations. (2) Environmental effects will be more apparent and persist longer in low-resource environments than in high-resource environments. We reasoned that tillers taken from plants in environments contrasting in resource availability may differ in amount of stored resources. When raised in low-resource environments, differences in stored resources may result in greater differences in growth than in environments where resources are less limiting. (3) Environmental effects will be greater and more persistent in some genotypes than others. Slower-growing genotypes, which may be more adapted to low-resource environments (Grime 1977; Chapin et al. 1993), may be less sensitive to resource availability; consequently growth of tillers from these plants may be less sensitive to previous environments. (4) Environmental effects in clonal offspring can arise from environmental variation in uniform controlled environments, so that estimates of broadsense heritability based on organisms propagated in the laboratory or greenhouse may be biased by environmental effects induced during this period of propagation. FIG. 1. Two episodes of clonal propagation were used in this experiment. During the C0 generation replicates of 30 genotypes were treated with two light and two nutrient levels in a factorial design (HH, HL, LH, LL). Each of these plants was divided, replanted, and raised at two nutrient levels (H and L) during the C1 generation. Finally, each of these plants was harvested after nine weeks and replanted into two nutrient treatments (H and L) or after 37 weeks and replanted into a uniform environment (not depicted here). MATERIALS AND METHODS Eriophorum vaginatum is a common, arctic/alpine sedge with a circumpolar distribution. The plant grows as a tussock, a dense clump of tillers connected by short woody rhizomes, and may stand 50 cm above the soil surface. Tussocks presumably represent a single genetic individual and may live 100 yr or more (Polozova 1970). In the field, individual tillers often survive 10 or more years and give rising to zero to 15 daughter tillers, with tiller production peaking at three to five years (Fetcher and Shaver 1983). Tussocks flower in early spring and produce numerous small seeds in a cottony mass, giving rise to a common name, cotton grass. Seedling recruitment is common following disturbance in tundra plant communities, but rare in established populations (Gartner et al. 1986). C0 Generation Tussocks of E. vaginatum were collected from three sites in Alaska: Smith Lake near Fairbanks in interior Alaska (15 plants), Toolik Research Station on the north slope of the Brooks Range (11 plants), and Barrow on the cold, wet coast of the Arctic Ocean (four plants). In August 1994 we weighed and planted 16 tillers from each tussock into 236-ml styrofoam cups (480 total) filled with coarse sand, and randomly assigned one cup from each tussock to 16 plastic flats (one replicate of each genotype in each flat; 480 plants total; Fig. 1). Flats were randomly assigned treatments: two levels of light and two levels of nutrients in a factorial design. The eight flats assigned to low light were each covered with a wooden frame supporting cheesecloth that reduced light by 50%. Throughout this study we fertilized plants twice per week with differing concentrations of a complete nutrient

3 454 KENT E. SCHWAEGERLE ET AL. solution (11 mm N; 0.5 mm P; 6.0 mm K) similar to Evans solution (Salisbury and Ross 1992, p. 119). During the C0 generation, high-nutrient treatment cups received solution diluted to 50%; low-nutrient cups received solution diluted to 5%. All 16 flats were assigned positions on a greenhouse bench at random with regard to treatment. Flats and frames were rotated at weekly intervals. After 18 weeks we judged that plants receiving the high-nutrient treatment were showing reduced growth rate due to self competition. These plants were removed from cups, washed, and weighed. One tiller from each cup was weighed and replanted to ensure a more consistent environmental treatment throughout the C0 generation. Treatments continued another 23 weeks (41-week total), at which time plants were removed from cups, washed, and weighed. Treatment effects were assessed by comparing mean final fresh mass of plants and mean relative growth rates (RGR). For low-nutrient plants RGR was calculated as: the difference in the logs (initial fresh mass and final fresh mass) divided by the number of days from planting to harvest. Because high-nutrient plants were replanted as single tillers after 18 weeks, RGR was calculated for the first 18 weeks of the experiment and again for the last 23 weeks. The weighted mean of these two was used to summarize RGR of highnutrient plants. C1 Generation After weighing, plants from the C0 generation were divided into one to four tillers which were planted into coarse sand in styrofoam cups (1478 cups total, Fig. 1). Cups were assigned to high- and low-fertilization regimes of 30% and 15% concentrations of the nutrient solution described above. PROC PLAN (SAS Institute 1990) was used to randomize spatial position of cups. To assess environmental effects during the course of the C1 generation, we conducted a nondestructive measurement of plant size. Two weeks after planting we measured the length of each leaf on all plants. Plant size was described as the summed length of leaves on each plant. One-third of the cups were selected using the RANUNI function and PROC SORT (SAS Institute 1990) and harvested nine weeks after planting. The remaining two-thirds of the plants were harvested 37 weeks after planting. Plants were removed from cups, roots were washed, and total fresh mass was determined. C2 Generation: First and Second Planting Two separate harvests of the C1 generation provided an opportunity to evaluate the magnitude of C0 and C1 treatment effects after nine weeks and 37 weeks of the C1 treatments. C1 plants harvested at nine weeks were used to establish the first planting of the C2 generation (Fig. 1). Two tillers were taken from each C1 plant. We weighed each tiller, counted number of leaves, measured the length of each leaf, and planted tillers in 236-ml styrofoam cups filled with coarse sand (719 cups total). One tiller from each C1 plant was assigned to a C2 high-nutrient treatment and one to a C2 low-nutrient treatment. Plants assigned to high- and low-nutrient treatments received weekly additions of 35 ml of 30% and 15% concentration of nutrient solution, respectively. PROC PLAN (SAS Institute 1990) was used to assign random positions with respect to all statistical effects on a greenhouse bench. Plants were raised for 12 weeks and harvested. We determined fresh and dry mass of leaves, infloresences, rhizomes, and roots. C1 plants harvested at 37 weeks were used to establish the second planting of the C2 generation. From each C1 plant we took two tillers, weighed, counted number of leaves, measured length of leaves, and planted each tiller in a 236-ml styrofoam cup filled with coarse sand (1683 cups total). PROC PLAN (SAS Institute 1990) was used to assign random positions to cups on two adjacent greenhouse benches (blocks) with one tiller from each C1 plant assigned to each block. Plants were fertilized twice per week with 35 ml of 20% concentration of the nutrient solution described above. Plants were harvested after 14 weeks. Fresh and dry masses of leaves, inflorescences, rhizome, and roots were recorded. During the growth period, plant size was measured biweekly using a digital imaging system. Each individual plant was set before a video camera interfaced with a microcomputer using commercially available hardware and software. A short computer program was written to count numbers of black and nonblack pixels and to translate these data into image area. Statistical Analyses All dependent variables were examined for heteroscedasticity, skewness, and kurtosis and transformed appropriately before statistical analysis. Mixed-model analysis of variance was used to analyze data. Because this study focuses on environmental sources of variance rather that genetic variation within and among populations and because genotypes do not represent samples from some normally distributed population, we treated plant genotype as a fixed effect and ignored population effects throughout our analyses. The design called for a five-way fixed- effects model with the following main effects: light treatment during the C0 generation, nutrient treatment during the C0 generation, nutrient treatment during the C1 generation, nutrient treatment during the C2 generation, and genotype. Random effects included: replication nested within the interaction between light, nutrient, and genotype during the C0 generation and replication during the C1 generation nested within C0 replications. Interactions between fixed effects and between fixed and random effects were included in the model, but when higher-order interactions were not significant (P 0.15), we also used a reduced model that was limited to first-, second-, and third-order interactions. Although most statistical analyses followed this model, some hypotheses required variations on this basic model; these are discussed in the Results. Prior to each analysis PROC FREQ (SAS Institute 1990) was used to locate missing treatment combinations (empty cells) in the dataset. In situations with empty cells, we allowed PROC GLM (SAS Institute 1990) to construct Type IV hypotheses for model effects. As a check on Type IV results, we dropped those genotypes from the analysis that did not experience all treatment combinations and proceeded

4 PERSISTENCE OF ENVIRONMENTAL EFFECTS 455 TABLE 1. Fresh mass (g) of plants at planting (initial) and harvest (final) of each clonal generation summarized as least-square means. The C1 generation was divided into two groups and harvested at two times (H1 and H2). Week measured indicates when in the course of the experiment the measurements were taken. Dashes indicate means that do not exist within the experimental design. To interpret C0 treatment means in the C1 generation consider the 1478 plants raised in the C1 generation. About half of these plants were propagated from parents grown in ambient light and displayed a mean mass of 0.62 g at planting. The other half were taken from parents grown in 50% ambient light and displayed an initial mean mass of 0.60 g for a grand mean of 0.61 g. Similarly, about half of the 1478 plants had parents in the 50% nutrient treatment (mean 0.77 g) and half had parents in the 5% nutrient treatment (mean 0.44 g; grand mean 0.61 g). Finally, of these same 1478 plants, half were randomly assigned in generation C1 to the 30% nutrient treatment, and these had a mean initial size of 0.62 g. The other half were assigned to the 15% nutrient treatment, and these had a mean of 0.59 g (grand mean 0.61 g). C0 Treatments Light ambient Light 50% ambient Nutrients 50% Nutrients 5% C1 Treatments Nutrients 30% Nutrients 15% C2 First planting Nutrients 30% Nutrients 15% First planting Initial C0 generation Final Second planting Initial Final C1 generation Initial H1 H First planting Initial Final C2 generation Second planting C2 Second planting Nutrients 20% Grand mean Sample size Week measured Initial Final with Type III hypothesis testing. We saw strong congruence between Type III (reduced dataset) and Type IV (full dataset) results. When several dependent variables occurred, the MANOVA statement (SAS Institute 1990) was used to test multivariate hypotheses about model effects, thus providing an experimentwide test of statistical effects. Analyses described above assess probability of Type I errors, but Type II errors are also of concern in this study. A Type II error occurs when a study lacks the statistical power to detect a real difference between two sampled populations, and consequently a null hypothesis is not rejected when, in fact, it is false. Power analysis has often been used to describe the probability of making Type II errors, but several authors have argued that a posteriori estimates of power can be misleading and that confidence intervals on population differences provide more meaningful information about study results (review in Gerard et al. 1998). To address the problem of Type II errors in this study, we used the ESTIMATE option of PROC GLM (SAS Institute 1990) to calculate confidence intervals for estimated differences between treatment means whenever treatment effects were not significant (P 0.05). Estimated repeatabilities of clonal replicates were calculated as the among-replicate variance divided by the sum of the among- and within-replicate variances. Standard errors of repeatability and variance components are given by Becker (1984, pp ). RESULTS C0 Generation Plant growth was greatly influenced by treatments. Plants raised for 23 weeks with 50% nutrient solution were five times the mass of plants raised for 41 weeks with 5% nutrient solution (least-squares means with initial mass as covariate; P ; Table 1). Mean RGR throughout the 41-week treatment period indicated nearly fourfold difference in mean RGR between nutrient treatments (P ). Least-square mean RGR in high nutients was day 1, corresponding to a fresh mass doubling time of 36 days; least-square mean RGR in low nutrients was day 1, corresponding to a doubling time of 133 days. Nearly threefold differences in RGR were observed among genotypes (P ). Other significant effects included light (P 0.02), light nutrient (P 0.002), light genotype (P 0.02), nutrient genotype (P ), and tray within light nutrient (P 0.006). C1 Generation Size of tillers at planting showed strong C0 nutrient effects (P ; Table 1). Tillers from the C0 low-nutrient treatment had on average 43% less mass than plants from the high-nutrient treatment. C0 light treatments had little or no effect on mass of tillers (P 0.8), but significant differences in tiller mass were observed among genotypes (P ), trays within light nutrient (P ), and among clonal replicates nested within all the above effects (P ). Three measurements of plant size during the C1 generation revealed a decline in C0 treatment effects and an increase in C1 treatment effects. (1) Two weeks after planting, the summed leaf length of C1 plants carrying C0 low-nutrient effects was 38% less than that of C1 plants carrying C0 highnutrient effects (P ; Fig. 2A) and, counterintuitively,

5 456 KENT E. SCHWAEGERLE ET AL. FIG. 2. Plant size during the C1 generation measured (A) two weeks; (B) nine weeks; and (C) 37 weeks after planting and summarized by simple means. Differences between CO nutrient treatments represent carry-over environmental effects from the C0 generation, which diminished during the C1 generation. Differences between C1 nutrient treatments represent effects of immediate environment, which increased during the C1 generation. the summed leaf length of C1 plants carrying C0 low-light effects was 19% greater than that of C1 plants carrying C0 high-light effects (P ; not shown in Fig. 2). This later difference may reflect allocation differences between tillers from high- and low-light treatments during the C0 generation, rather than size differences. (2) Nine weeks after the C1 planting, at time of first harvest, C0 and C1 treatments showed combined effects on plant size (Fig. 2B). Mass of plants in the C1 low-nutrient treatment was 47% less than that of C1 plants in the high-nutrient treatment (P ). Fresh mass of plants carrying C0 low-nutrient effects was 30% less than that of plants carrying C0 high-nutrient effects (P ). C0 light treatments had no significant effects on fresh mass (P 0.13). (3) Thirty-seven weeks after planting, at the time of the second C1 harvest, the pattern had changed (Fig. 2C). Plant mass was determined largely by nutrient treatments during the C1 generation: Plants from the C1 low-nutrient treatment had 82% less fresh mass than plants from the high-nutrient treatment (P ). C0 nutrient treatments still exhibited marginally significant effects on plant size (P 0.05), but the difference was small and not in the expected direction; fresh mass of C1 plants with C0 low-nutrient effects was 4% greater than that of plants carrying C0 high-nutrient effects. C0 light treatments had no detectable effect on fresh mass (P 0.7). We found little or no evidence for genotypic variation in the persistence of environmental effects during the C1 generation. Summed leaf length measured two weeks after planting showed small but statistically significant interaction between genotype and C0 nutrient treatment (P 0.02) and between genotype and C0 light treatment (P 0.03). Interactions between genotype and C0 treatments for measures of plant size at nine weeks and 37 weeks were not significant. We found little or no evidence that the magnitude and persistence of C0 treatment effects differed between C1 nutrient treatments. Measures of plant size taken after two weeks showed no interactions between C1 nutrient treatments and C0 treatments. However, a small but significant interaction (P 0.01) was observed at 37 weeks. Significant variation was found among C0 replicates nested within C0 treatment-genotype combinations, indicating that plants derived from the same clonal replicate displayed differences from plants taken from other replicates with the same genotype and C0 treatment (P ). For final fresh mass, repeatability of replicates was 0.36, indicating that after removing treatment and genotypic effects, 36% of the remaining phenotypic variance in fresh mass was attributable to variation among C0 replicates. Interaction between C1 nutrient treatments and C0 replicates was not significant (P 0.10). C2 Generation: First Planting Tillers used to start the C2 generation (first planting) showed significant C1 nutrient effects, but small or no C0 treatment effects. Tillers from the C1 low-nutrient treatment had 14% less fresh mass (P 0.01) and 22% less summed length of leaves (P ) than high-nutrient plants. Tiller size traits also differed among genotypes (P ). C0 light and C0 nutrient treatments and all interactions were not significant (P 0.05). The 95% confidence intervals for the estimated differences in initial tiller mass between C0 light treatments and between C0 nutrient treatments were mg and mg, respectively, where the mean tiller mass was 830 mg. This result indicates that real differences as great 20% of mean tiller mass may have gone undetected in the analysis. Size of tillers at planting was closely associated with plant size at harvest, indicating ample opportunity for C1 environmental effects to influence C2 plants through effect of initial tiller size. Specifically, initial tiller fresh mass, summed leaf length, number of leaves, and interaction between summed length and number of leaves all had significant effects on final dry mass of plants (P 0.05), together accounting for 26% of the variance in final dry mass. We found no evidence that C0 light and C0 nutrient treatments influenced growth or allocation in the C2 generation. MANOVA of fresh and dry mass of leaves, rhizomes, roots, and inflorescences revealed no significant C0 light effects (P 0.44), no significant C0 nutrient effects (P 0.81), and

6 PERSISTENCE OF ENVIRONMENTAL EFFECTS 457 no significant interaction (P 0.14). The 95% confidence interval for the estimated differences in final mass between C0 light treatments and between C0 nutrient treatments were g and g, respectively, where the mean plant mass was 7.8 g. This 6% interval about the mean indicates that any grandparental effects undetected by this study were very small. C1 nutrient treatments had significant effects on growth and allocation of C2 plants that were independent of size at planting. MANOVA of fresh and dry mass of plant parts revealed significant differences among genotypes (P ), C1 nutrient treatments (P 0.002), and C2 nutrient treatments (P ). Final total mass of plants carrying C1 low-nutrient effects was 12% less than the mass of plants carrying high-nutrient effects (P ). Mean percent of total dry mass allocated to inflorescences was 15% and 12% in plants carrying C1 high- and low-nutrient effects, respectively (P ). Mean percent of leaf fresh mass attributable to water was 85% and 84% in plants carrying C1 highand low-nutrient effects, respectively (P 0.01). We found little or no evidence that the magnitude and persistence of environmental effects depends on environment. Interaction between C1 nutrient and C2 nutrient effects from MANOVA was not significant (P 0.10), indicating little or no difference in C1 effects between C2 high- and low-nutrient treatments. Estimated differences in final mass between plants carrying C1 high- and low-nutrient effects when raised in the C2 high-nutrient treatment was 1.2 g 0.3 (SE) and 1.1 g 0.2 (SE) when raised in the C2 lownutrient treatment, indicating a large opportunity for Type II error associated with the interaction between C1 and C2 nutrient treatments. Persistence and magnitude of C1 nutrient effects was independent of genotype despite large treatment effects and large differences in growth rates among genotypes. Leastsquare mean final mass was 5.6 g in the slowest growing genotype and 10.6 g in the fastest growing genotype. MAN- OVA of masses of plant parts showed no interaction between genotype and C0 treatment effects (P 0.3) or between genotype and C1 nutrient effects (P 0.6), indicating that differences in plant size and allocation caused by C0 and C1 treatments did not vary significantly among genotypes. C2 Generation: Second Planting Tillers used to start the C2 generation (second planting) showed strong C1 nutrient effects and some evidence of C0 treatment effects. Tillers from the C1 low-nutrient treatment had 13% less fresh mass, 7% fewer leaves, and 23% less summed length of leaves (P ) than tillers from the C1 high-nutrient treatment. Substantial differences were observed among clones (P ), and smaller but statistically significant effects included C0 nutrient treatments (P 0.04); interaction between C0 light and C0 nutrient effects (P 0.04); interaction between C0 nutrient and C1 nutrient effects (P 0.01); interaction between C0 light, C0 nutrient, and C1 nutrient effects (P ); interaction between C1 nutrient and genotype effects (P 0.006); and interaction between C0 light, C0 nutrient, C1 nutrient, and genotype effects (P 0.04). These last two interactions provide some FIG. 3. Growth of plants during the second planting of the C2 generation was measured using a digital imaging system. Plants carrying C1 high-nutrient effects were consistently larger than plants carrying C1 low-nutrient effects, but the proportional difference declined. Graph shows means 2 SE. evidence that C0 and C1 environmental effects varied among genotypes. Size of tillers at planting displayed significant effects on plant size throughout the C2 generation. Tiller mass, summed leaf length, number of leaves, interaction between mass and summed leaf length, and interaction between summed leaf length and number of leaves at planting were all significant predictors of final plant mass and image area measurements (P 0.05). For example, these five effects alone accounted for 61% of the variance in image area in the second week of the study and for 27% of the variance in total dry mass at harvest, indicating a major C1 nutrient effect on C2 plants through the effect of initial tiller size. Measures of tiller size at time of planting were used as covariates in subsequent analyses. We found little evidence that C0 treatments influenced plant growth during the C2 generation (second planting) when the effect of initial tiller size was removed. MANOVA of fresh and dry mass of leaves, rhizomes, roots, and inflorescences indicated that final plant mass and allocation was not affected by C0 light treatment (P 0.4), C0 nutrient treatment (P 0.1), or their interaction (P 0.4). Power of the experimental design would have allowed resolution of 3 4% differences in C2 plants due to C0 effects. No effect of C0 replicates within genotype-treatment combinations was observed (P 0.33). C1 nutrient treatments had significant effects on C2 plants, independent of tiller size at time of planting. Digital imaging during the course of the experiment showed that as plants grew C1 treatment effects became proportionally smaller (Fig. 3); C2 plants carrying C1 low-nutrient effects were 23% smaller in the second week, but only 10% smaller at harvest. The coefficient of variation for image area declined from 48 to 41 and from 41 to 36 in plants taken from C1 low- and high-nutrient treatments, respectively. Plant mass and allocation at harvest were also significantly influenced by C1 nutrient treatments. Plants carrying C1 low-nutrient effects

7 458 KENT E. SCHWAEGERLE ET AL. had 17% less total dry mass than plants carrying C1 highnutrient effects and displayed differences in allocation to inflorescences and roots. Mean percent of the total dry mass allocated to inflorescences was 16% and 19% in plants carrying C1 low- and high-nutrient effects, respectively. Mean percent allocation to roots was 22% and 20% in plants carrying C1 low- and high-nutrient effects, respectively. Large variance among C1 replicates nested within C1 treatment-genotype combinations was observed. For final dry mass, the repeatability of replicates was 0.43, indicating that 43% of the variance in final dry mass can be attributed to environmental differences among replicates within C1 treatment-genotype combinations. Causes of this within-clone, environmental effect could be intergenerational environmental effects or time of planting effects, as discussed above (see C1 generation). Because we sequentially planted two tillers from each C1 replicate (e.g., AABBCCDD), we can distinguish between these alternatives by partitioning residual sums of squares from the fixed-effects model in two ways: (1) among pairs of plants taken from the same pot and planted in sequence (e.g., AA BB CC DD); and (2) among pairs of plants that were planted in sequence, but were not taken from the same pot (AB BC CD). Estimated variance among C1 replicates was (SE). Estimated variance among unrelated plants planted in sequence was (SE). Based on the standard errors associated with these estimates, the difference (4.2) is significantly greater than zero (P 0.05) and represents an estimate of phenotypic variance caused by intergenerational (C1) environmental effects. The persistence and magnitude of C1 nutrient effects did not vary significantly among genotypes. MANOVAs of image data and mass of plant parts at harvest showed no interactions between genotype and C1 nutrient effects (P 0.4 and P 0.6, respectively). DISCUSSION This study shows that environment can cause changes in phenotype that can be passed from parents to clonal offspring. Similar phenomena have been observed in field-grown potatoes (Went 1959), Mimulus guttatus (Libby and Jund 1962), eastern cottonwood (Wilcox and Farmer 1968), western hemlock (Foster et al. 1984), larch (Radosta et al. 1994), and Daphnia species (Lynch and Ennis 1983; Tessier et al. 1983; Goulden and Henry 1984). Hume and Cavers (1981) used the term carry-over effects to describe preconditioning effects passed between generations, and the term C effects (Lerner 1958) has been used in the forestry literature to describe within-clone environmental effects. Plants grown in high-nutrient environments consistently produced tillers that were larger, grew faster, and allocated biomass differently than plants from low-nutrient environments. Initial size of propagules (tillers) explained much of the variation in later size. This relationship declined over time, but still accounted for as much as 27% of the variance in biomass after 15 weeks of growth. Similarly, in sexually reproducing plants, maternal effects on postjuvenile traits in a variety of plant species have been attributed to differences in propagule (seed) size (Roach and Wulff 1987; Galloway 1995; Donohue and Schmitt 1998; Thiede 1998). We also found intergenerational effects that were independent of size. After adjusting for initial size, tillers taken from plants from high-nutrient treatments grew faster than tillers taken from low-nutrient plants. Three mechanisms may contribute to this result: (1) Tillers from high-nutrient treatments may carry more or qualitatively better growth resources; (2) high-resource environments may stimulate the initiation of a greater number of active meristems (M. Weih, pers. comm. 1998); and (3) high-nutrient treatments may cause persistent changes in gene expression (e.g., Vöchting 1904; Doorenbos 1965; Bussey and Fields 1974; Banks 1979; Alexander and Wulff 1985; Karban and Myers 1989) that promoted growth. This result follows Went (1959), who reported that intergenerational environmental effects in potato were independent of propagule size. Environment of clonal grandparents had little effect on plant size and allocation. At the start of the C2 generation, size of tillers displayed small and marginally significant C0 effects, indicating that differences in tiller size and vigor established in the C0 generation caused differences during the C1 generation, which in turn influenced size of tillers used to initiate the C2 generation. However, we found no statistically significant effect of C0 treatments on final plant size and allocation in the C2 generation. If C0 effects on the C2 generation exist but went undetected in this study due to limited sample size, then we estimate final size differences in the C2 generation caused by C0 treatments were less than 4% of final plant mass. In contrast, grandparental environmental effects have been observed in sexually reproducing plant species (Miao et al. 1991; Case et al. 1996). The decline in C0 effects through the C1 generation and through the C2 generation could have been caused by a swamping of treatment effects by random environmental effects introduced during the study. Environmental heterogeneity in the greenhouse causing variation in plant growth rates may have gradually blurred earlier treatment effects. Furthermore, random changes in plant status introduced during propagation (e.g., Schmid and Bazzaz 1990) may obscure earlier treatment effects. The observed decline in intergenerational effects during the first few weeks of plant growth points to these mechanisms. Alternatively, decline of carry-over effects could have arisen by a different mechanism. Plants grown in small containers with finite resources may converge in phenotype due to self competition, so that earlier environmental effects gradually decline. Supporting this later hypothesis, imaging data throughout the second planting of the C2 generation showed that plant growth followed a logistic form (Fig. 3). Plants reached a certain size then growth rate declined. Furthermore, variance in image area declined relative to the mean during the course of the experiment, adding more evidence that the size of plants from different genetic and treatment backgrounds converged in phenotype. The persistence of intergenerational effects was correlated with the magnitude of direct treatment effects. We found that C0 nutrient treatments had much greater immediate effects on growth rates than C0 light or C1 nutrient treatments and subsequently had greater effects on the next clonal generation. This result is not surprising considering the tenfold difference in nutrient availability between C0 nutrient treat-

8 PERSISTENCE OF ENVIRONMENTAL EFFECTS 459 ments and the only twofold difference in resource availability between C0 light and C1 nutrient treatments. From this result we argue that environmental factors that have strong direct effects on plant growth will also cause greater and/or more persistent intergenerational effects. We found little or no evidence that the magnitude and persistence of environmental effects differed among environments or among genotypes. Interactions between treatments applied in successive clonal generations were generally not significant, indicating that tillers taken from a particular nutrient treatment respond similarly to nutrient availability. Thus, our prediction that intergenerational effects would be greater in low-resource environments was not supported. This result suggests that storage of growth resources may not be the primary cause of intergenerational effects in this species. Similarly, interactions between genotypes and treatments during the previous clonal generation were generally not significant, providing little evidence that slower-growing genotypes had smaller intergenerational environmental effects. Incidental heterogeneity in the greenhouse environment within experimental treatments also caused differences among plants that were passed from one generation to the next. In the C1 and C2 generations, we observed significant clonal replicate effects within treatment-genotype combinations, indicating that tillers taken from different plants having the same genotype-treatment background often differed in growth characteristics. This within-clone, environmental effect may represent an intergenerational environmental effect associated with spatial variation in the greenhouse. Alternatively, variation among replicates may represent an environmental effect associated with time of planting: Tillers from each clonal replicate were planted at the same time. Therefore, any inconsistencies in planting technique or potting medium would inflate this variance estimate. The second planting of the C2 generation provided an opportunity to distinguish between these two effects and revealed that about half of the among-replicate variance was caused by intergenerational environmental effects caused by variation in the greenhouse and about half was caused by time of planting effects. Despite the evidence for intergenerational effects provided by this study, plants showed a great ability to respond to change in their immediate environment. After only a few weeks of growth, direct treatment effects were generally greater than intergenerational environmental effects once the effects of initial size were removed, indicating that natural selection has favored rapid response to the availability of resources. In contrast, highly persistent environmental effects would be adaptive in situations where morphological and physiological adjustments in the present environment are likely to preacclimate propagules to future environments (Fox et al. 1999). Such adjustments are likely when environmental changes occur predictably (e.g., seasonal change). The rapid response to nutrients exhibited by plants in this study suggests that E. vaginatum has adapted to unpredictable changes in nutrient availability. Somatic mutation is a source of genetic and phenotypic variance in studies using clonally propagated organisms. Mutations at loci influencing phenotype can be passed mitotically so that asexual descendants from a single genetic individual may have somewhat different genetic composition and therefore may display different phenotypes (Whitham and Slobodchikoff 1981). For example, in collecting material for our study, we sampled tussocks that may represent genetic mosaics. Also, during the course of this experiment somatic mutation may have further contributed mutational variance among replicates from a tussock. Genetic differences among tillers from the same tussock would contribute to phenotypic variance among replicates just as intergenerational environmental effects would. These two very different causes of variance among clonal replicates can be distinguished by examining within-clone variance over time and randomizing across generations for environmental effects (e.g., Lynch 1985; Lynch et al. 1998). However, the absence of significant differences among the C2 clonal descendants of C0 replicates of genotypes from the field indicates that somatic mutation did not have a major effect on the results of this study. The persistence of environmental effects in clonal organisms has ecological and evolutionary consequences. The growth and proliferation of clonal organisms may depend, in part, on past environmental conditions, thus introducing a time lag in growth and/or population response to a changing environment. Indeed, Molau and Shaver (1997) have suggested that flower production in tundra populations of E. vaginatum may be determined by weather conditions several years earlier. Furthermore, Crone (1997) and Ginzburg (1998) have demostrated that intergenerational environmental effects can influence demography in sexually reproducing species. Evolutionary consequences of clonal environmental effects may parallel the consequences of maternal effects in sexual organisms, as outlined by Kirkpatrick and Lande (1989). Response to selection in a clonal population will depend on the direction and magnitude of selection on parental and offspring generations as well as the phenotypic correlation between parent and clonal offspring, caused by intergenerational environmental effects. For example, if a change in environment caused natural selection for high growth rates in E. vaginatum, then the positive correlation between parental and offspring growth rates caused by carryover effects as observed in this study would result in a time lag in evolutionary response. This is because plants with below-average genotypic values (Falconer and Mackay 1996) but high environmental values for growth rate would produce offspring with above-average growth rates, and thus these plants would make a greater contribution to future generations than their genotypic value merits. If selection for growth rate was constant across clonal generations, evolutionary response would increase asymptotically as the correlation between genotypic value and phenotypic value increases. Then, if selection were relaxed, evolutionary response would continue for several generations, declining geometrically as random environmental effects erode the correlation between genotypic and phenotypic values. This evolutionary momentum could have consequences for the persistence of populations of long-lived clonal organisms in changing environments. The persistence of environmental effects has methodological consequences for a variety of studies using clonal organisms. Carry-over effects must be considered in studies of genotypic variation and covariation of traits, studies using asexual organisms to estimate the contribution of mutation

9 460 KENT E. SCHWAEGERLE ET AL. to genetic variation in quantitative traits (e.g., Lynch et al. 1998), and studies using clonally propagated plants and animals as phytometers (Antonovics et al. 1987) or biomonitors of environmental toxins (e.g., Lovett Doust et al. 1993; Biernacki and Lovett Doust 1997). Quantitative genetic studies of natural clonal populations have acknowledged the problem of environmental effects biasing estimates of genotypic variance, and researchers have gone to varying lengths to avoid the problem. Most of these studies use a holding period in a uniform environment prior to the experiment to purge clonal material of environmental effects from the field and use measures of initial size of asexual propagules to discount effects of previous history (e.g., McKee and Ebert 1996). However, our data show that persistent environmental effects can arise even within uniform controlled environments and that asexual propagules can carry intergenerational effects that are independent of size. Future experiments to estimate genotypic variance and covariance of metric traits can incorporate additional procedures to reduce environmental bias in estimates of broadsense heritability. Because genotypes cannot be separated from their history, no experimental design can ensure that estimates of genotypic variance are free of intergenerational environmental effects. Instead, researchers should seek methods that reduce and/or provide some quantitative information about this source of error. Libby and Jund (1962) outline two approaches: (1) a statistical approach in which clones provide an unbiased estimate of V E, sexually produced individuals provide an estimate of total phenotypic variance, and genotypic variance is estimated by subtraction; and (2) an experimental approach that involves replication of genotypes prior to a holding period. Our experiments with E. vaginatum paralleled this second approach. Genotypes sampled from the field are clonally propagated. Multiple replicates of each genotype are raised for an extended holding period with care that replicates are randomly distributed with respect to all potential environmental variables (e.g., initial size, potting media, spatial position, fertilization). This ensures that environmental effects introduced during the holding period are randomly distributed among genotypes. Finally, replicates are propagated and these offspring are used in an experiment to estimate genotypic variance. Total phenotypic variance can be partitioned among genotypes, among replicates of genotypes during the holding period, and within replicates. Variance among genotypes estimates V G. Variance among replicates within genotypes quantifies environmental effects persisting from the holding period (plus mutational variance), and variance within replicates estimates environmental variance (V E ) during the final experiment. If a controlled uniform environment is used during the holding period, environmental effects from the field will gradually be lost and introduction of new environmental effects during the holding period will be minimized. Alternatively, the holding period can be designed around a series of microenvironments (e.g., in the field), so that when clones are put into the experimental setting, all clonal replicates carry the naturally occurring range of parental environmental effects. Such a design would provide more accurate estimates of the actual genotypic variance in the field (Libby and Jund 1962). Quantitative genetic analysis of natural clonal populations can also incorporate measurements of propagule size, which may often be a useful indicator of previous environments (e.g., Watson and Cook 1987). However, our results showed that treatments affected not only the size of propagules, but also the allocation of biomass within propagules, which, in turn, was an important predictor of later performance. Furthermore, we also found carry-over effects that were independent of both size and allocation differences among propagules. This indicates that biases arising from intergenerational environmental effects can be reduced, but not eliminated, in future studies by incorporating a more complete description of the size and allocation in asexual propagules. ACKNOWLEDGMENTS We thank W. S. Armbruster, J. P. Bryant, R. A. Stafford, E. D. Parker, Jr., and M. Weih for comments on earlier versions of the manuscript and U. Molau for specimens from Barrow. This research was support by National Science Foundation grant OPP to KES. LITERATURE CITED Alexander, H. M., and R. D. Wulff Experimental ecological genetics in Plantago. X. The effects of maternal temperature on seed and seedling characters in P. lanceolata. J. Ecol. 73: Antonovics, J., K. Clay, and J. Schmitt The measurement of small-scale environmental heterogeneity using clonal transplants of Anthoxanthum odoratum and Danthonia spicata. Oecologia 71: Ayre, D. J Localized adaptation of clones of the sea anemone Actinia tenebrosa. Evolution 39: Banks, M. S Plant regeneration from callus from two growth phases of English ivy, Hedera helix L. Z. Pflanzenphysiol. 92: Becker, W. A Manual of quantitative genetics. Academic Enterprises, Pullman, WA. Biernacki, M., and J. Lovett Doust Vallisneria americana (Hydrocharitaceae) as a biomonitor of aquatic ecosystems: comparison of cloned genotypes. Am. J. Bot. 84: Bussey, H., and M. A. Fields A model for stably inherited environmentally induced changes in plants. Nature 251: Case, A. L., E. P. Lacey, and R. G. Hopkins Parental effects in Plantago lanceolata L. II. Manipulation of grandparental temperature and parental flowering time. Heredity 76: Chapin, F. S., III, K. Autumn, and F. Pugnaire Evolution of suites of traits in response to environmental stress. Am. Nat. 142:S78 S92. Crone, E. E Parental environmental effects and cyclical dynamics in plant populations. Am. Nat. 150: Donohue, K., and J. Schmitt Maternal environmental effects in plants: adaptive plasiticity? Pp in T. A. Mousseau and C. W. Fox, eds. Maternal effects as adaptations. Oxford Univ. Press, New York. Doorenbos, J Juvenile and adult phases in woody plants. Encyc. Plant Physiol. 15: Falconer, D. S., and T. F. C. Mackay Introduction to quantitative genetics. 4th ed. Longman, Harlow, U.K. Fetcher, N., and G. R. Shaver Life histories of tillers of Eriophorum vaginatum in relation to tundra disturbance. J. Ecol. 71: Foster, G. S., R. K. Campbell, and W. T. Adams Heritability, gain, and C effects in rooting of western hemlock cuttings. Can. J. For. Res. 14: Fox, C. W., M. E. Czesak, T. A. Mousseau, and D. A. Roff