Plant reproduction under elevated CO2 conditions: a meta-analysis of reports on 79 crop and wild species

Authors

  • Leanne M. Jablonski,

    1. Department of Evolution, Ecology and Organismal Biology, The Ohio State University, 1735 Neil Avenue, Columbus, OH 43210, USA;
    2. Marianist Environmental Education Center, 4435 East Patterson Road, Dayton, OH 45430, USA;
    Search for more papers by this author
  • Xianzhong Wang,

    1. Department of Biology, Indiana University-Purdue University Indianapolis, Indianapolis, IN 46202, USA
    Search for more papers by this author
  • Peter S. Curtis

    Corresponding author
    1. Department of Evolution, Ecology and Organismal Biology, The Ohio State University, 1735 Neil Avenue, Columbus, OH 43210, USA;
      Author for correspondence: Peter S. Curtis Tel: +1 614 292 0835 Fax: +1 614 292 2030 Email: curtis.7@osu.edu
    Search for more papers by this author

Author for correspondence: Peter S. Curtis Tel: +1 614 292 0835 Fax: +1 614 292 2030 Email: curtis.7@osu.edu

Summary

  • • Reproductive traits are key characteristics for predicting the response of communities and ecosystems to global change.
  • • We used meta-analysis to integrate data on eight reproductive traits from 159 CO 2 enrichment papers that provided information on 79 species.
  • • Across all species, CO 2 enrichment (500–800 µl l −1 ) resulted in more flowers (+19%), more fruits (+18%), more seeds (+16%), greater individual seed mass (+4%), greater total seed mass (+25%), and lower seed nitrogen concentration, (N) (−14%). Crops and undomesticated (wild) species did not differ in total mass response to elevated CO 2 (+31%), but crops allocated more mass to reproduction and produced more fruits (+28% vs +4%) and seeds (+21% vs +4%) than did wild species when grown at high CO 2 . Seed [N] was not affected by high CO 2 concentrations in legumes, but declined significantly in most nonlegumes.
  • • Our results provide robust estimates of average plant reproductive responses to CO 2 enrichment and demonstrate important differences among individual taxa and among functional groups. In particular, crops were more responsive to elevated CO 2 than were wild species. These differences and the substantial decline in seed [N] in many species have broad implications for the functioning of future natural and agro-ecosystems.

Introduction

How plant reproduction responds to rising CO2 has important implications for the performance of both natural (Körner et al., 1996) and agro-ecosystems (Wittwer, 1995; Bazzaz & Sombroek, 1996; Rosenzweig & Hillel, 1998). For crops, changes in the mass, number, and nutrient concentration of fruits and seeds with elevated CO2 could affect agricultural production practices and food quality (Conroy et al., 1994; Murray, 1995; Idso & Idso, 2001). Changes in reproductive success of undomesticated species could alter the composition and hence the functioning of unmanaged plant communities (Bazzaz, 1996; Grunzweig & Körner, 2001). Since vegetative phase responses to elevated CO2 are not always well correlated with those of reproductive traits (e.g., Ackerly & Bazzaz, 1995; Farnsworth & Bazzaz, 1995; Jablonski, 1997), predicting the response of plants to future atmospheric CO2 conditions requires investigation of the effects of CO2 enrichment throughout a plant's life cycle (Norby et al., 2001).

Domesticated crops and undomesticated species (henceforth referred to as wild species) often differ markedly in their patterns of carbon and nitrogen allocation, and might also be expected to differ in their reproductive responses to CO2 enrichment. Crop species have been bred to maximize yield and produce harvestable products of consistent size and quality. Indeed, intense selection by plant breeders for increased harvest index (proportion of biomass invested in the harvested organ) over the past century may have provided a mechanism for some crops to take advantage of rising atmospheric CO2 in boosting fruit and grain yield (Hall & Ziska, 2000). Most modern crop varieties also have been bred to perform best under relatively predictable and benign environmental conditions, or to show a broad range of adaptability to stress (Evans, 1993). However, wild species are subject to natural selection from pollinators and dispersal agents and are subject to more diverse allocational trade-offs to sustain growth, resource acquisition, reproduction and defense. They are therefore likely to be more plastic than crop plants in a greater number of vegetative and reproductive traits (Waller, 1988) and less likely to allocate as great a fraction of available resources to reproductive structures.

Other functional groupings of plants that have shown consistent and distinctive vegetative responses to elevated CO2 are based on photosynthetic physiology or ability to fix atmospheric nitrogen. Among herbaceous species, N2-fixing legumes are typically the most responsive to CO2, followed by nonlegume C3 species, with C4 species being the least responsive (Poorter, 1993; Wand et al., 1999). Since vegetative- and reproductive-phase physiology are often well coupled (Egli, 1998), this rank order in vegetative responsiveness to CO2 may generally hold true for other life history traits. However, physiological changes that can occur under high CO2 such as reductions in leaf nitrogen concentration ([N]), accumulation of leaf starch, and downregulation of photosynthesis (Stitt & Krapp, 1999; Körner, 2000), may affect the carbon and nitrogen supply available to reproductive organs (Lawlor, 2002) and in turn constrain the magnitude of potential responses to high CO2 during this phase of the life cycle.

Kimball (1983 ) conducted the first quantitative review of crop yield responses to elevated CO 2 and reported an average increase of 33% across 37 agricultural species, which included fiber, root/tuber, and leaf crops. Increases at high CO 2 were comparable, or in some cases less, for crops whose yield was as reproductive organs per se : +23% for fruit crops, +31% for C 3 grains, +31% for legume seed, and +12% for flower crops ( Kimball, 1986 ). Cure (1985 ) and Cure and Acock (1986 ) assembled yield data for 10 major crops (grain, leaf, tuber and fiber) and found an average increase with CO 2 doubling of 41%. Soybean (28 studies) and wheat (17 studies) were the most common species examined, and yield responses were similar to those reported for mass accumulation, with little change in harvest index. Amthor (2001 ), in a semiquantitative review of wheat yield from 50 publications, found that CO 2 enrichment increased grain mass 31%. Ainsworth et al. (2002 ) used meta-analysis on soybean experiments published from 1980 to 2000 and found that yield (pod number) was stimulated 24%, total biomass increased 37%, and harvest index declined 11% in elevated compared to ambient CO 2 -grown plants.

Reviews of elevated CO2 responses by wild species have focused primarily on the vegetative phase (Poorter, 1993, 1998; Poorter & Nagel, 2000; Poorter & Perez-Soba, 2001). Wild species appear to show less stimulation by high CO2 than do crops, and fast-growing wild species respond more to CO2 than do slow-growing wild species (Poorter, 1993, 1998). For plants grown to reproductive maturity, reviews have been largely qualitative (Murray, 1995) or have targeted particular plant groups. For example, Ackerly and Bazzaz (1995) reported a 30% enhancement in reproductive mass across six old field annuals, with the magnitude of the response depending on life-cycle stage and planting density. They found that the C3 species were stimulated more by elevated CO2 (+50%) than were the C4 species (+10%). Newton (1991) observed a trend towards higher seed weight per plant and higher individual seed weight, but a variable response in seed numbers, among nine pasture and old-field species at elevated compared with ambient CO2. Elevated CO2 effects on seed nutrients have been the least studied of seed quality parameters, and no effect was seen in initial studies of wheat, maize, soybean or Scirpus (reviewed in Newton, 1991), although Conroy et al. (1994) and Murray (1995) reported a decline in wheat ear and grain [N] under high CO2.

Our objectives in this meta-analysis were twofold: to provide a comprehensive, quantitative synthesis of published reports on plant reproductive responses to elevated CO2, and to test for differences among plant functional groups or among growth conditions in affecting the magnitude of these responses. We considered CO2 effects on the number and mass of reproductive parts, which are key measures of reproductive effort, and on individual seed mass and seed [N], which reflect seed quality. We hypothesized that the effects of elevated CO2 on reproductive effort would be quantitatively similar to CO2 effects on vegetative mass, with a similar rank order in response among legumes, nonlegume C3, and C4 plants. However, because of artificial selection for increased carbon allocation to fruits and seeds, we expected a greater response to CO2 enrichment by crops than by wild species. We also hypothesized that because of decreased leaf [N] under high CO2 (Cotrufo et al., 1998; Stitt & Krapp, 1999), which is a substantial contributor to seed protein (Murray, 1995), seed [N] would be similarly reduced. Our results provide the first comprehensive and statistically robust summary of plant reproduction under CO2 enrichment and demonstrate important differences between the likely behavior of natural vs agro-ecosystems in a higher CO2 environment.

Methods

Database development

Bibliographic resources used in developing the database were Strain and Cure (1986, 1994), Curtis et al. (1999), Jones et al. (1999) and the Institute for Scientific Information's Science Citation Index. Our literature survey was comprehensive for publications appearing between 1983 and 2000 that reported data on reproductive responses to elevated CO2. Here, we report our analysis of eight specific reproductive traits: flower number, fruit number, fruit mass, seed number, total seed mass, individual seed mass, seed [N] and reproductive allocation (reproductive tissue mass/total plant mass). From the set of papers reporting data on these reproductive characteristics we also analysed data on plant mass (above- plus below-ground including both vegetative and reproductive tissues).

The following data reporting requirements were necessary for a publication to be included in the analysis: response means of elevated (x̄e) and ambient (x̄e) CO2-grown plants reported as numerical or graphical data; ambient CO2 treatments < 400 µl l−1 CO2 and elevated CO2 treatments between 500 µl l−1 and 800 µl l−1 CO2; and the entire plant exposed to the CO2 treatment for the majority of its life cycle. Our statistical methods require independence among individual observations (response means) so only one measurement point per variable per treatment per experiment was used. These data were taken either following the longest period of CO2 exposure or from the peak value prior to plant senescence. ‘Treatment’ included species identity as well as crossed experimental factors interacting with CO2, such as nutrient level or competition with other plants. Thus, a multifactorial experiment examining a number of species or using multiple interacting treatments could contribute several response means for each variable measured (e.g., Chaudhuri et al., 1990; Navas et al., 1997).

We analyzed results from 159 publications that yielded 1391 separate pairs of means distributed across the nine response variables. It is important to note that sample sizes, k, in the meta-analysis refer to pairs of means, not separate publications. Of these pairs of means, or studies, 75% were from 24 domesticated crop species and 25% from 55 wild species (Appendix 1). In our analysis, wild species were those that had not experienced significant artificial selection for harvestable yield. In most cases, wild species were characteristic of unmanaged ecosystems, but they also included some economically important, but undomesticated species such as Avena barbata, Poa annua, and Arabidopsis thaliana. Among studies of crop species, 31% involved wheat (Triticum aestivum), 23% soybean (Glycine max), 19% rice (Oryza sativa), and < 5% each from barley (Hordeum vulgare), cotton (Gossypium hirsutum), maize (Zea mays) and other crop species. Studies were approximately evenly distributed between those conducted in outdoors growth chambers, glasshouses, and indoor growth facilities, respectively, while only 3% were conducted under free-air CO2 enrichment (FACE).

We were specifically interested in how categorical variables describing biological characteristics of the study species or aspects of the experimental design influenced the magnitude of elevated CO2 responses. The biological categories considered were taxonomic identity at the generic level (Taxon), whether the plant was a crop or a wild species (Crop/wild), and whether it was a legume, a nonlegume C3 species or a C4 species (Leg/C3/C4). In most cases only one species per taxon was studied, but occasionally data from several wild species in a single genus were reported (e.g., Bromus, Ipomoea and Polygonum) and in two cases genera (Raphanus and Hordeum) contained both a crop and a wild species. Experimental design categories were: elevated CO2 treatment level ([CO2]), treated as a class variable with 50 µl l−1 divisions between 500 µl l−1 and 800 µl l−1 CO2; the rooting volume, or pot size, of the growth facility (Pot), treated as a class variable with divisions of < 1 l, 1–5 l, 5–10 l, 10–30 l and > 30 l including in-ground planting; and the nature of any experimental treatments interacting with elevated CO2 (X-trt) such as low soil nitrogen, low or high temperature, low light availability, high ozone, high UV-B radiation, water deficits, etc. The basis of coding decisions for the X-trt category is described in more detail by Curtis (1996) and Curtis and Wang (1998). We also tested for effects of several other categorical variables such as annual or perennial life history, whether a response variable was originally expressed on a per plant or per unit land area basis (as was often done for yield parameters in agronomic studies), the duration of the CO2 treatment, and type of CO2 exposure facility (growth chamber, greenhouse, field chamber or FACE ring); however, these are not discussed here as they showed little or no between-group heterogeneity (i.e. no statistical effect of category).

Statistical analysis

Meta-analytical studies depend on some estimate of treatment effect size, commonly the magnitude of an experimental treatment mean (in this case, x̄e) relative to the control treatment mean (in this case, x̄a) (Gurevitch et al., 2001). A common effect size metric in elevated CO2 studies, and that used here, is the response ratio, r = x̄e/x̄a. Hedges et al. (1999) described statistical methods for integrating the log-transformation of r, lr(= ln r) across independent studies and these methods have been incorporated into statistical software for performing meta-analyses (metawin v. 2.0; Rosenberg et al., 2000). In summarizing results from k independent studies (pairs of means), effect sizes are normally weighted by the reciprocal of their variance, w, giving greater weight to experiments whose estimates have greater precision and hence increasing the precision of the combined estimate. For example, the weighted mean log ratio (inline image) is

image(Eqn 1)

While superior from a statistical standpoint, such weighting requires knowledge of both experimental sample size and the variance of treatment means from each of the k studies. Unfortunately, these basic measures often have not been reported in elevated CO2 studies. Using weighted effect sizes thus can result in the exclusion of many studies owing to inadequate reporting in the primary literature.

More recently, nonparametric resampling techniques (randomization tests and bootstrapping) have been developed for meta-analysis that allow the determination of significance levels and the generation of confidence limits for test statistics that do not meet the assumptions of the parametric tests developed for weighted effect sizes (Adams et al., 1997). In order to include the broadest range of studies in our database, we have used the resampling methods in metawin for all of our statistical evaluations, setting wi = 1 for all i in Eqn 1. Thus, we do not consider differences in precision among individual studies.

An important goal of our meta-analysis was to understand the source of variation in CO2 effect size among studies and to determine whether particular biological or experimental design categories elicited quantitatively different responses. In a procedure analogous to the partitioning of variance in an analysis of variance, the total heterogeneity for a group of comparisons (QT) is partitioned into within-group heterogeneity (Qw) and between-group heterogeneity (Qb), such that QT = Qw + Qb (Hedges & Olkin, 1985). Partitioning of variance proceeded in two steps. First, between-group heterogeneity (Qb) for each categorical variable was examined across all data for a given response variable. Second, the data set was subdivided according to levels of those categorical variables revealing significant Qb and the first step repeated. Mean log ratios were calculated when the number of categorical variables exhibiting significant Qb had been reduced to one or zero, suggesting no further partitioning of the dataset was justified. Categorical subdivisions could only be compared if they were represented by at least two separate studies (i.e. if k≥ 2). Thus, the results of a single study of a particular genus (e.g. seed number in Stachys; Navas et al. 1997) would be included in a comparison of the responses of wild vs crop species, but not in a comparison of the responses of individual taxa. Significance was established at the P < 0.05 level unless otherwise noted.

Results

Reproductive effort

Growth at elevated CO2 resulted in the production of significantly more flowers compared with growth at ambient CO2 (+19%, Fig. 1), averaged across 110 different studies. This result was consistent across categories, with no significant between-group heterogeneity (Qb) among biological categories and only one modest difference due to an experimental design factor, [CO2] (Table 1). The significant [CO2] Qb resulted from five studies, all using strawberry (Fragaria ananassa, see Appendix 1), that reported large responses to CO2 in the 550–599 µl l−1 range. Stimulation of fruit number by CO2 enrichment (+18%, Fig. 1) was very similar to that for flower number but the magnitude of this response differed between crop and wild species (Table 1). Crop species showed a much stronger response to CO2 enrichment in fruit production (+28%) than did wild species (+4%, ns) (Fig. 2a). The significant effect of interacting treatment variables (significant X-trt Qb, Table 1) in crops was due to a reduced CO2 response in plants grown under low nutrient stress (k = 4).

Figure 1.

Reproductive and mass accumulation responses to experimental CO 2 enrichment in 79 crop and wild species. Mean ± 95% confidence interval. The number of studies, k , for each response variable is shown in parentheses.

Table 1.  Within-group heterogeneity (Q b ) for plant reproductive and growth responses to CO 2 enrichment among the k studies contributing data to each response variable. Categorical variables distinguishing biological characteristics of each study were the species identity at the level of genus (Taxon), whether the plant was a crop or undomesticated species (Crop/wild) or whether it was a legume, a nonlegume C 3 plant, or a C 4 plant (Leg/C 3 /C 4 ). Categorical variables distinguishing experimental design features of each study were the elevated CO 2 treatment concentration ([CO 2 ]), the pot volume (Pot), and the presence of any interacting experimental treatments (X-trts). Significance of Q b denoted by * , P < 0.05; ** , P < 0.01; *** , P < 0.001
Response variableBiological categoriesExperiment design categories
kTaxonCrop/wildLeg/C3/C4[CO2]PotX-trts
  • 1

    Does not include studies with interacting experimental treatments in addition to CO 2 .

  • 2

    Does not include studies with [CO 2 ] < 550 p.p.m.

Flower number110   1.63   0.090.26   1.09*0.44   0.26
Fruit number100   3.20   1.02**0.53   0.060.71   0.96
Crop 61   1.050.06   0.130.39   1.20*
Wild 39   1.200.02   0.060.52   0.11
Fruit mass117   8.71   1.94*2.94*   0.252.13   6.93*
Legume 45   0.91   0.21   0.150.81   1.92
C3 471   2.26   0.11   0.470.10
Seed number230   6.41***   0.59 *0.01   1.66**1.16*   1.37
Crop1402   0.90*0.07   0.170.54   0.35
Wild 822   4.29*0.27   1.050.09   0.69
Total seed mass315   3.09   1.73***0.42   0.421.38*   1.34
Crop292   0.890.33   0.190.57   1.15
Wild 23   0.690.09   0.120.06
Individual seed mass184   0.27   0.010.09*< 0.010.19**   0.22
Legume 42< 0.01   0.02   0.030.07   0.09
C3140   0.18   0.01   0.020.18*   0.16
C4  2  –
Seed [N] 33   0.32**< 0.010.10*   0.070.25**   0.09
Legume  4  –
C3 29   0.21**< 0.01   0.060.16*< 0.01
Plant mass226   5.69***   0.272.06***   1.12*0.92   2.44**
Legume 48   0.19< 0.01   1.35**0.80   2.28***
C3168   3.56**   0.04   0.141.00*   1.43
C4 10   0.11< 0.01   0.020.02
Reproductive allocation 82   4.99   1.11*0.13   0.711.12   1.52
Crop 34   1.750.02   0.231.37   1.32
Wild 48   2.270.13   0.301.34   0.89
Figure 2.

Comparative responses of crop and wild species (a), and legume, nonlegume C 3 , and C 4 species (b), to experimental CO 2 enrichment. Only those categories showing significant between-group heterogeneity are illustrated. Mean ± 95% confidence interval. The number of studies, k , for each response variable is shown in parentheses.

Fruit mass increased less under high CO2 (+12%, Fig. 1) than did fruit number and was not significantly different from zero because of substantial among-study variation. Partitioning the fruit mass dataset within the Leg/C3/C4 categorical group (Table 1) revealed that legumes had a strong positive response (+37%, Fig. 2b) while the magnitude of the response by nonlegume C3 plants was further affected by interacting experimental treatments (X-trts). However, nonlegume C3 plants grown under optimal conditions (i.e. no X-trts) also showed a significant stimulation in fruit mass by elevated CO2 (+15%, Fig. 2b). There were no reports of C4 fruit mass at high CO2 in our dataset.

Seed numbers were greater in elevated compared with ambient CO2-grown plants (+16%, Fig. 1) and there were several significant sources of variation within categorical groups (Table 1). Eight studies with elevated CO2 treatments between 500 and 549 µl l−1 showed very high average CO2 effect size, causing a significant [CO2] Qb. Among all other studies (k = 222), crop species produced significantly more seeds at high CO2 (+21%, Fig. 2a), while wild species as a group did not (+4%, ns). There was significant heterogeneity among wild taxa in their response to CO2 enrichment, however, with approximately half showing positive and others negative or no significant response to the CO2 treatment (Table 1, Fig. 3a). Crops were consistently positive in their seed number response to high CO2, but with significantly less stimulation in the C4 maize (Zea) than in some C3 species such as rice (Oryza) (Fig. 3a).

Figure 3.

Taxon-specific responses in seed number (a) and seed [N] (b) to experimental CO 2 enrichment among wild (closed symbols) and crop (open symbols) species. Starred genera are represented by more than one species. Species shown illustrate the range of responses among taxa in each functional group, but do not include all taxa in the database. Mean ± 95% confidence interval. The number of studies, k , for each taxon is shown in parentheses.

Consistent with these effects on seed number, total seed mass increased substantially under CO2 enrichment across all studies (+25%, Fig. 1), but with significant differences between the performance of crops (+28%, Table 1, Fig. 2a) and wild species (−4%, ns). It is noteworthy that the crop/wild categorical group showed the greatest Qb in this large dataset (k = 315), and that within these subgroups there was no other significant Qb for any biological or experimental design category. However, there were over 10 times fewer studies reporting total seed mass data from wild species than from crop species, and no studies with common X-trts involving wild species, making it impossible to test for a significant X-trt Qb in this group (Table 1).

Seed quality

We found a small, but significant increase in individual seed mass with CO2 enrichment across all 184 studies (+4%, Fig. 1). This response was greatest in legumes (+8%, Fig. 2b), less for nonlegume C3 plants (+3%) and absent in C4 plants (−2%, ns), although sample size for the latter was very small (k = 2). The significant [CO2] Qb we observed was caused by a group of 10 studies, conducted at 600–649 µml l−1 CO2, that showed an unusually high treatment response.

Seed [N] was reduced in elevated compared with ambient CO2-grown plants across all studies (−14%, Fig. 1), but there also was significant variation in the magnitude of this response among taxa (Table 1). Among legumes, seed [N] was unaffected by growth at high CO2, while in nonlegume C3 plants it was reduced 15% (Fig. 2b). There was significant variation among taxa in the latter group, however. Rice, like the legume soybean, showed no CO2 effect, wheat seed [N] was reduced more than 20%; cotton and barley were intermediate in their responses (Fig. 3b). Wild species as a group did not differ from crop species in seed [N] responses to elevated CO2 (Table 1) and showed similar levels of variation among individual taxa (Fig. 3b). Sample sizes for this group were very small, however, limiting the strength of among-taxa comparisons.

Total mass and reproductive allocation

Mass response to CO2 enrichment was highly significant, with total plant mass increasing 31% relative to controls across the 226 studies in our database (Fig. 1). There was considerable heterogeneity among studies, however, with significant differences within both biological and experimental design categories (Table 1). Partitioning the database according to the Leg/C3/C4 category reduced the number of heterogeneous categorical groups and revealed a much higher biomass response to CO2 by legumes (+56%) than either nonlegume C3 (+26%) or C4 (+8%) plants (Fig. 2b). Remaining significant Qb among studies conducted with legumes was caused largely by an unusually high response to CO2 among three CO2 × low temperature studies (significant X-trt Qb) and secondarily by two studies showing very low CO2 responses when exposure CO2 concentrations were less than 549 µl l−1 (significant [CO2] Qb). Among the large number of studies conducted with nonlegume C3 plants, there were two heterogeneous groupings, the most significant being among different taxa. For example, rice (+38%, k = 58) and cotton (+44%, k = 13) responded more in mass accumulation to CO2 enrichment than did strawberry (+11%, k = 3) or wild radish (Raphanus raphanistrum, +5%, k = 8). We also found a significantly smaller mass response in nonlegume C3 plants grown in pots < 5 l compared with those grown in larger pots or in the ground (significant Pot Qb).

Overall, reproductive allocation was unaffected by CO2 enrichment (−5%, ns, Fig. 1). The only categorical group to show significant heterogeneity in this regard was among crop and wild species (Table 1). Crop species showed considerable variability in their response to elevated CO2, but, on average, were unaffected (+10%, ns, Fig. 2b). Wild species, however, had significantly lower reproductive allocation (−14%) in elevated compared with ambient CO2 conditions. We also examined CO2 effects on harvest index (yield mass/above-ground vegetative mass), which was only reported for crops, and found no difference from the effects on reproductive allocation (data not shown).

Discussion

Reproductive effort

Our meta-analysis showed that for crop plants, growth under elevated CO2 conditions resulted in a significant increase in all measures of reproductive effort. Our calculation of a 28% increase in total seed mass with CO2 enrichment is consistent with previous estimates of CO2 effects on crop seed yield ranging from 28% to 35% (Kimball, 1983, 1986; Cure, 1985; Cure & Acock, 1986; Amthor, 2001). Underlying this average response, however, was significant variation among individual crop species. For example, rice (+42% in seed number) was much more responsive to CO2 enrichment than was soybean (+20%), wheat (+15%) or maize (+5%). Species-specific responses to elevated CO2 in crop yield are well known but the mechanism(s) accounting for these differences are not always clear. We found that generally, the functional groupings of legumes, nonlegume C3 and C4 plants showed a similar rank order of response in reproductive traits as that found for many vegetative traits under high CO2 (e.g., Poorter, 1993; Wand et al., 1999). That is, legumes were the most responsive, followed by nonlegume C3 species and lastly C4 species. This supports our hypothesis of a linkage between vegetative physiology and reproductive effort in crop plants at high CO2. The one notable exception to this general pattern was in rice. The high degree of responsiveness to elevated CO2 by rice could be related to the widespread use of very high-yielding, semi-dwarf cultivars with exceptionally high photosynthetic rates coupled with high spikelet sink strength (Horie et al., 2000), or perhaps also to the typical growth of rice under flooded, high nutrient conditions. Variation among species in their phenological or ontogenetic responses to CO2, which can affect total seed production (e.g., Ward et al., 2000; LaDeau & Clark, 2001), could also help to explain these interspecific differences.

In contrast to the behavior of crops under high CO2 conditions, wild species, on average, showed no significant change in reproductive effort with CO2 enrichment. This supports our original hypothesis, but as with crop plants we found substantial interspecific variation in the absolute magnitude of the CO2 response. However, while crop plants all showed some degree of stimulation in reproductive effort by high CO2, wild species were equally distributed among those that were stimulated and those that were inhibited by the CO2 treatment. The broad range of responses to CO2 among the 55 wild species in our database (e.g. from −25% to +30% in seed number) at one level simply reflects the substantial response variability reported even for closely related wild species (e.g., Farnsworth & Bazzaz, 1995) or for those occurring in the same communities (e.g., Ackerly & Bazzaz, 1995; Smith et al., 2000). We found, however, that this overall difference in reproductive effort was not reflected in differences in mass accumulation, where crop and wild species were statistically indistinguishable, both showing a strong positive response to high CO2. Rather, there was a difference in CO2 effects on reproductive allocation, the ratio of reproductive mass to total plant mass. In crops, elevated CO2 had no significant effect on reproductive allocation, which is consistent with the generally very weak effect of CO2 on allometry in the vegetative phase (Poorter & Nagel, 2000). In wild species, allocation to reproduction declined 14% at high CO2, suggesting more carbon allocation to structural, defensive, or other nonreproductive tissues under these conditions.

Systematic differences in experimental growth conditions could have contributed to the differences we found in the responses to CO2 by crop and wild species. Experiments on crops typically duplicated agricultural conditions while wild species were often studied in situ or in more complex experimental arrays (e.g., Schappi, 1996; Navas et al., 1997; Grunzweig & Körner, 2000; Smith et al., 2000). Greater variability in environmental conditions in native ecosystems compared with agro-ecosystems would probably contribute to greater variability in wild species responsiveness to elevated CO2. Approximately 30% of the studies using wild species in our database were conducted under field conditions where plants experienced natural levels of resource heterogeneity and competition. However, unless an interacting variable such as nutrient level or competition was explicitly identified in the original study, these field conditions were considered equivalent to those of unstressed plants growing in pots or in an agricultural setting. Thus, we cannot attribute crop/wild differences in CO2 responsiveness to differing levels of environmental stress per se, as we found no clear statistical evidence for effects of interacting environmental factors on the magnitude of these reproductive responses to CO2 (i.e. few cases of significant X-trt Qb). This contrasts with the well-documented effects of, for example, soil fertility and drought on the magnitude of vegetative phase responses to elevated CO2 (Poorter & Perez-Soba, 2001). It is possible that results from experimental treatments of sufficient severity to markedly inhibit reproduction would not have been well represented in our dataset, thereby reducing our ability to detect these environmental effects.

Seed quality

Ours is the first broad synthesis of seed quality responses to elevated CO2. We found that individual seed mass responded only weakly to CO2 enrichment (+4%); by comparison, mineral nutrient additions can increase individual seed mass from 5% to 25% (Fenner, 1992). Plant nutrient status affects seed traits at several phases of reproduction, with the size and number of seeds largely determined at the time of flower bud initiation (Willson, 1983; Roach & Wulff, 1987). However, seed-filling is influenced by subsequent environmental conditions, particularly nutrient status and the level of carbon assimilation (Fenner, 1992; Egli, 1998; Bazzaz et al., 2000), suggesting a capacity for greater seed-filling under high CO2. Overall responsiveness to CO2 may be limited, though, by breeding for uniform seed size in crops (Almekinders & Louwaars, 1999) and selection pressures for optimal seed size in wild species (Leishman et al., 2000).

We found a 14% reduction in seed [N] averaged across all species. Seed [N] is often coupled tightly to leaf [N] (Murray, 1995), which in turn is often reduced under elevated CO2 (Cotrufo et al., 1998). Although we did not examine CO2 effects on leaf [N], the 16% reduction in nonlegume C3 seed [N] we documented is identical to the percentage reduction in C3 vegetative tissue [N] reported by Cotrufo et al. (1998) from a quantitative synthesis of 278 studies. As with CO2 effects on leaf [N], reductions in seed [N] could be caused both by a carbon dilution effect (reflected in increased individual seed mass) and by diversion of internal nitrogen resources to increased seed production (reflected in increased seed numbers). These observations suggest that the strong linkage between vegetative and reproductive nitrogen levels observed under ambient CO2 (Fenner, 1986; Bazzaz et al., 1987; Evans, 1993; Lawlor, 2002) is maintained under elevated CO2.

Two notable exceptions to the pattern of decline in seed [N] at high CO2 were in legumes and rice. Legumes are often able to use increased carbon gain under elevated CO2 for increased N2-fixation (Allen & Boote, 2000) and thus may be able to increase seed number and mass without a loss in seed [N]. Legumes also show less decline in leaf [N] at high CO2 compared with nonlegume C3 species (Cotrufo et al., 1998). The response of rice is more difficult to explain. Horie et al. (2000) noted that under high CO2, rice has increased spikelet production efficiency per unit plant nitrogen, and typically shows complete grain-filling if nitrogen fertilization is adequate, so rice seeds may draw more on soil nitrogen and hence rely less on remobilization of leaf nitrogen than do seeds in other crop species. We found no evidence for a difference in seed [N] responses to CO2 between crops and wild species, but sample sizes were very small for the latter (k = 5) and included no cases where a given wild species was examined in more than one study.

These reproductive effort and seed quality responses could have important agronomic and ecological implications. For example, while wheat and barley both showed strong positive responses to elevated CO2 in total seed number (approx. +15%), their seed [N] was even more strongly reduced (approx. −20%), indicating a potential decline in total nitrogen allocation to reproduction. This trade-off should be considered when calculating the perceived benefits of rising atmospheric CO2 on world food production (e.g., Wittwer, 1995; Bazzaz & Sombroek, 1996; Rosenzweig & Hillel, 1998). Ecologically, changes in the number of flowers, fruits, and seeds, and their nutritional quality, could have far-reaching consequences. For example, seed [N] can affect establishment dynamics by affecting germinability, growth rate, and competition (Fenner, 1991, 1992; Charest & Potvin, 1993; Wulff, 1995; Gutterman, 2000). Interspecific differences in responses of seed number and seed [N] to elevated CO2 also could affect competitive interactions among co-occurring species (Smith et al., 2000). Indeed, changes in seed nutrient composition could underlie the altered germination success and seedling size observed from seeds of plants grown under elevated CO2 (reviewed in Bazzaz & McConnaughay, 1992). These effects could also be magnified trophically through impacts on pollinators, frugivores, and granivores (Coviella & Trumble, 1999; Whittaker, 1999, but see Bezemer & Jones, 1998). Together, these community-level interactions could alter biodiversity, plant community dynamics and ecosystem functioning (Körner, 2000).

In conclusion, our meta-analysis of reproductive responses to elevated CO2 for the first time encompasses a suite of traits relating to both reproductive effort and seed quality, and directly compares the responses of diverse crop and wild species. As a group, crops were much more responsive to CO2 enrichment than were wild species, and showed less interspecific variation in the magnitude of their response to elevated CO2 than did wild species. These differences were not reflected in CO2 effects on vegetative growth, where crop and wild species responded identically, but rather in a significantly lower reproductive allocation at high CO2 in wild species. Overall, both crop and wild species had reduced seed [N] at high CO2, pointing to an important trade-off between CO2 effects on reproductive effort and effects on reproductive quality. Reproductive traits are necessary additions to the suite of morphological, physiological and developmental characteristics used in predicting the response of individuals, communities and ecosystems to global change (Körner et al., 1996). Variation in the magnitude of CO2 effects on reproductive effort and seed quality across species and functional groups have broad implications for the functioning of future natural and agro-ecosystems.

Acknowledgements

We thank Michael Jones for help with database development and Quirine Ketterings for data entry assistance. Support for this work was provided by the National Science Foundation, grant number IBN-97-27159 to P. S. C.

Appendices

Appendix 1

Table 2. The species, response variables, and publications used in the meta-analysis. For each entry it is noted whether the species was considered a crop (C) or wild (W) species, whether the species was a legume (L), nonlegume C3, or C4 species, whether data on number of flowers (NFLW), number of fruits (NFRT), mass of fruits (MFRT), number of seeds (NSD), total mass of seeds (MSDT), mass of individual seeds (MSDI), seed nitrogen concentration ([N]SD), reproductive allocation (AR), or total plant mass (MTOT) were included in the database (indicated by an asterisk) and the publication from which these data were extracted, which are listed after the Table.
SpeciesC/WL/C3/C4NFLWNFRTMFRTNSDMSDT[N]SDMSDIARMTOTReferences
Abutilon theophrastiWC3       ** 19
Abutilon theophrastiWC3   ** *   20
Abutilon theophrastiWC3  *       21
Abutilon theophrastiWC3       ** 35
Abutilon theophrastiWC3** *  * * 54
Aegilops kotschyiWC3   ****   58
Aegilops peregrinaWC3   ****   58
Agrostis caninaWC3      * * 52
Agrostis capillarisWC3       ** 18
Alstroemeria aurantiacaCC3*        143
Amaranthus retroflexusWC4       ** 19
Amaranthus retroflexusWC4       ** 35
Arabidopsis thalianaWC3     *     7
Arabidopsis thalianaWC3 * *  *  144
Arabidopsis thalianaWC3 ***    *146
Arachis hypogaeaCL  *    ** 33
Arachis hypogaeaCL  *       34
Arachis hypogaeaCL ****    138
Avena barbataWC3   ****   68
Avena barbataWC3   *      69
Avena sativaCC3    * *  129
Betonica officinalisWC3*        128
Brassica junceaWC3 ***  * *112
Brassica kaberWC3 *       147
Bromus erectusWC3      *  139
Bromus lanceolatusWC3   *   * 105
Bromus madritensisWC3   *  *** 65
Bromus madritensisWC3   *  *  136
Bromus mollisWC3   ** * * 81
Bromus rubensWC3    *   * 64
Calluna vulgarisWC3*        152
Capsicum annuumCC3***     *109
Capsicum annuumCC3  *     *110
Cardamine hirsutaWC3   *  *   83
Carduus pycnocephalusWC3   *     105
Carex curvulaWC3*  ** *  130
Cassia obtusifoliaWL **** *** 47
Centaurea jaceaWC3*        128
Cucumis sativusCC3  *     * 61
Cucurbita pepoCC3  *     * 61
Datura stramoniumWC3  *       54
Dimorphotheca pluvialisWC3 *     * 145
Echinochloa crus-galliWC4       **111
Echinochloa glabrescensWC4    *   *  3
Eichhornia crassipesWC3*       *137
Eleusine indicaWC4       **111
Festuca viviparaWC3       ** 18
Fragaria ananassaCC3***    ** 39
Fragaria ananassaCC3 **  * *  31
Gentianella germanicaWC3*  ** **  50
Glycine maxCL  * *** *  4
Glycine maxCL **** * *  5
Glycine maxCL  * *   *  6
Glycine maxCL   ** *    9
Glycine maxCL  *       25
Glycine maxCL   **** * 36
Glycine maxCL   ** *** 37
Glycine maxCL * ** * * 48
Glycine maxCL    *   * 51
Glycine maxCL **** *   62
Glycine maxCL **** *   63
Glycine maxCL    *     66
Glycine maxCL      * * 82
Glycine maxCL **** *   93
Glycine maxCL **    ** 94
Glycine maxCL *  * *  100
Glycine maxCL **** * *113
Glycine maxCL    *    114
Glycine maxCL    *    123
Glycine maxCL **      124
Glycine maxCL * ** *  134
Glycine maxCL    *   *140
Glycine maxCL  *     *155
Glycine maxCL ** * * *158
Glycine maxCL ** * *  159
Gossypium barbadenseCC3  *     *118
Gossypium hirsutumCC3    *     75
Gossypium hirsutumCC3    ** **115
Gossypium hirsutumCC3*        119
Gossypium hirsutumCC3  *     *120
Gossypium hirsutumCC3*      **121
Gossypium hirsutumCC3*       *122
Hordeum spontaneumWC3   ****   58
Hordeum vulgareCC3    * *   43
Hordeum vulgareCC3    ***   46
Hordeum vulgareCC3*  **** * 78
Hordeum vulgareCC3    *     86
Hordeum vulgareCC3     *    87
Hordeum vulgareCC3*  ** *  116
Hordeum vulgareCC3    * *  129
Hordeum vulgareCC3   ****  141
Hordeum vulgareCC3   ** *  148
Ipomoea purpureaWC3 **** *** 47
Leontodon helveticusWC3*  *     130
Lepidium lasiocarpumWC3   *  *  136
Lolium perenneWC3*     *  108
Lotus corniculatusWL*         26
Lotus corniculatusWL*        128
Lycopersicon esculentumCC3  *       61
Lycopersicon esculentumCC3 *        80
Lycopersicon esculentumCC3*       * 92
Lycopersicon esculentumCC3  *      135
Medicago orbicularisWL   *   * 105
Oryza sativaCC3    *   *  3
Oryza sativaCC3   ** *   10
Oryza sativaCC3   ** *   11
Oryza sativaCC3   ** *   12
Oryza sativaCC3   ** *   13
Oryza sativaCC3    * *   14
Oryza sativaCC3   ** * * 24
Oryza sativaCC3      *   30
Oryza sativaCC3    *     70
Oryza sativaCC3*  ** * * 73
Oryza sativaCC3    * *   74
Oryza sativaCC3       *  85
Oryza sativaCC3    *   * 99
Oryza sativaCC3   ** *  131
Oryza sativaCC3   ****  132
Oryza sativaCC3    *   *140
Oryza sativaCC3    *   *156
Oryza sativaCC3    * * *157
Panicum antidotaleWC4   *    * 55
Phalaenopsis x yukimaiCC3*         40
Phaseolus vulgarisCL*         71
Phaseolus vulgarisCL  *       86
Phlox drummondiiWC3*       * 54
Pisum sativumCL*** *   *104
Plantago lanceolataWC3   *  * * 41
Plantago lanceolataWC3        * 42
Poa alpinaWC3       ** 18
Poa annuaWC3   *  *   83
Polygonum persicariaWC3 **** *** 47
Raphanus raphanistrumWC3**        27
Raphanus raphanistrumWC3*  *      38
Raphanus raphanistrumWC3 ***   ** 67
Raphanus sativusCC3 **    ** 67
Rosa hybridaCC3*         32
Rosa hybridaCC3*        154
Rosa hybridaCC3*        153
Scabiosa columbariaWC3*        128
Senecio vulgarsWC3   *  *   83
Setaria faberiiWC4*         89
Sherardia arvensisWC3       * 105
Solanum melogenaCC3 **      106
Sorghum bicolorCC4       *   2
Sorghum bicolorCC4    *   *  6
Sorghum bicolorCC4    *     28
Sorghum bicolorCC4    *    123
Spergula arvensisWC3   *  *   83
Stachys arvensisWC3   *     105
Trifolium pratenseWL*        128
Trifolium repensWL      *  108
Triticum aestivumCC3   **      8
Triticum aestivumCC3*         15
Triticum aestivumCC3    *     16
Triticum aestivumCC3    * * * 17
Triticum aestivumCC3    * *   23
Triticum aestivumCC3   ** *   29
Triticum aestivumCC3    ***   43
Triticum aestivumCC3   ** *   44
Triticum aestivumCC3     *    45
Triticum aestivumCC3   *    * 53
Triticum aestivumCC3    *   * 56
Triticum aestivumCC3    *     57
Triticum aestivumCC3   ****   60
Triticum aestivumCC3    *     72
Triticum aestivumCC3    *     76
Triticum aestivumCC3*         84
Triticum aestivumCC3    *     86
Triticum aestivumCC3     *    87
Triticum aestivumCC3   ** *   88
Triticum aestivumCC3    *   * 90
Triticum aestivumCC3   ** * * 91
Triticum aestivumCC3   **  ** 95
Triticum aestivumCC3   ** *** 96
Triticum aestivumCC3    * * * 97
Triticum aestivumCC3*         98
Triticum aestivumCC3    * * *100
Triticum aestivumCC3   ** *  101
Triticum aestivumCC3   ** *  102
Triticum aestivumCC3   *  *  103
Triticum aestivumCC3   ** * *117
Triticum aestivumCC3    *    125
Triticum aestivumCC3   ** * *126
Triticum aestivumCC3    * * *129
Triticum aestivumCC3   ** * *133
Triticum aestivumCC3    *   *140
Triticum aestivumCC3   ****  141
Triticum aestivumCC3    *  * 142
Triticum aestivumCC3   ** *  148
Triticum aestivumCC3   *  *  149
Triticum aestivumCC3   **    150
Triticum aestivumCC3    *    151
Tropaeolum majusCC3*       * 79
Vaccinium myrtillusWC3 *        59
Vicia fabaCL    *     58
Vicia fabaCL*        107
Vigna unguiculataCL**         1
Vigna unguiculataCL * ** *   22
Vulpia ciliataWC3   *    * 49
Vulpia octofloraWC3   *  *  136
Zea maysCC4    *   * 77
Zea maysCC4    * * *100
Zea maysCC4    *    125
Zea maysCC4*  **  * 127
Unidentified wild speciesWC3 * *   * 105
 

Ancillary