Plant growth and competition at elevated CO₂: on winners, losers and functional groups

1. Seed mass and germination

What will be the effect of elevated CO₂ on the size of the seeds produced? With an improved carbohydrate availability of the mother plant at elevated CO₂, one may expect that one of the constraints on seed mass would be lifted and that seed size would increase. On the other hand, seed mass per se is generally far less variable than the total number of seeds produced (Harper, 1977). The effects reported in the literature are variable, even within a species. Most of the studies have focused on the crop species Triticum aestivum and Glycine max. For both species decreases as well as increases in seed mass have been observed (Fig. 1a), the variation most likely depending on cultivar and/or environmental conditions. On average, both Triticum and Glycine seeds increase significantly in mass with CO₂ concentration, with a larger stimulation for the latter (2% vs 9%, P < 0.05).

Will all species be stimulated in seed mass? Not to give excessive weight to a frequently measured species we averaged all observations per species before the analysis, under the assumption that environmental effects are leveled out across groups of species. The results for a total of 50 species are shown in Fig. 1(b). In a number of cases, seeds of high CO₂ plants were found to be smaller, with differences ranging from almost nil to 25% (e.g. Bromus rubens, Huxman et al., 1998). Others have reported increases in seed mass, up to 45% (e.g. Cassia fasciculata, Farnsworth & Bazzaz, 1995). Averaged over all herbaceous species, the effect of maternal CO₂ on seed mass was nil, and there were no significant differences (P > 0.5) between monocots, herbaceous legumes and other herbaceous dicots. Thus, we conclude that across these functional groups, seed mass will not contribute to a significant extent to the observed biomass stimulation. In a number of species, the difference may affect plant size of the next generation, though. A spectacular example is the seed mass of the only woody species investigated so far, Pinus taeda (Hussain et al., 2001). High CO₂ plants produced 90% heavier seeds, with a much higher lipid concentration. Lipids do have a high C-concentration and it requires relatively large amounts of glucose to be synthesised (Penning de Vries et al., 1974). If plants are C-limited, one could expect that species with a high lipid concentration in their seeds would show the largest increase in seed mass when mother plants were grown at high CO₂. The fact that a species like Glycine max, which also has a high lipid concentration, shows a strong response as well (Fig. 1a), does fit in with this idea. However, a third species with a high oil concentration in the seeds, Brassica juncea, is among the ones with the most negative response (–20%; Tousignant & Potvin, 1996). Clearly, more systematic experiments in this field are required before a proper answer can be provided.

Germination per se is generally not affected directly by a higher CO₂ concentration (Morse & Bazzaz, 1994; Andalo et al., 1996; Huxman et al., 1998). Given the overall high CO₂ concentration in the soil, it is questionable whether a doubling in the atmospheric CO₂ concentration would have any direct effect (Ward & Strain, 1999). However, maternal effects have been reported, with a much lower germination rate for seeds from high CO₂ parents of two Ipomoea species (Farnsworth & Bazzaz, 1995) up to an increased germination rate in Pinus taeda (Hussain et al., 2001). The consequence of such maternal effects are as yet only scarcely studied (Tousignant & Potvin, 1996; Bezemer et al., 1998), although they may have a profound impact on the growth and population dynamics of different species through changes in size hierarchy (Morse & Bazzaz, 1994) or reproductive behaviour (Bazzaz et al., 1992). In most growth experiments discussed in this paper, however, seeds from the same batch were used for both control and high-CO₂ grown plants. Thus, in the analysis of biomass responses (section V), differences in seed mass or germination effects will not play a role.

2. RGR, ULR and LAR

To analyse the growth response of plants to a given environmental factor, the concept of growth analysis can be used, a top-down approach tightly connected to the carbon budget of the plant. Growth then is analysed in terms of ‘Relative Growth Rate’ (RGR, the increase in biomass per unit time and per unit biomass present). This parameter can be factorised into the ‘Unit Leaf Rate’ (ULR, the increase in biomass per unit leaf area and per unit time), a factor closely related to the carbon gain and losses of the plant per unit leaf area, and the ‘Leaf Area Ratio’ (LAR), which indicates the amount of leaf area per unit plant mass (Evans, 1972). A problem is that the effect of CO₂ on RGR is relatively small and often time-dependent, occurring only at the early stage of plant growth (references in Poorter, 1993; Centritto et al., 1999). In those experiments where seedlings were pregrown at control levels of CO₂ and then transferred to high CO₂, a transient stimulation of RGR is apparent (Fonseca et al., 1996; Gibeaut et al., 2001). In experiments where the seeds are germinated at different CO₂ levels, the CO₂-induced RGR stimulation is already partly over by the time of the first harvest (Wong, 1993; Roumet et al., 1996; Bunce, 1997). This has to be kept in mind when the distribution of RGR data is considered. Averaged over a wide range of experiments with C3 species, RGR increases by 8% (Fig. 2a; 130 observations in 50 experiments), but the stimulation does not statistically differ from zero, due to quite a number of cases where RGR is not affected at all.

Although changes in RGR are small and only transient in time, the growth components underlying RGR do shift more substantially and over a longer time. The largest change is in ULR, with an average increase of 24% (Fig. 2a; P < 0.001). The increase in ULR is balanced by a decrease in LAR of on average 13% (P < 0.001). There are some differences between groups of species in this respect, but they are generally small and only marginally significant. The exception is formed by the herbaceous monocots and dicots: although there is no difference in RGR stimulation between the two groups, the dicots were showing a stronger increase in ULR at elevated CO₂ (28% vs 11%, P < 0.001) and a stronger decease in LAR (−17% vs −7%, P < 0.001).

3. Components of ULR

A simple way to break down ULR is to consider it as composed of three parameters (Poorter, 2002). The first is the leaf-area based carbon gain in photosynthesis, not determined as a momentary rate at light saturation as in many physiological analyses, but under growth conditions and integrated over the day (PS_A). In this way, diurnal variation in light intensity and light distribution are taken into account. Moreover, different leaves on a plant may be in a different physiological stage, which necessitates measurements on whole plants rather than on one specific leaf. The second parameter is the fraction of the daily fixed C that is not spent in respiration, exuded or ‘lost’ in another way, such as volitalization, but retained in the plant to form part of its ‘structural’ biomass (FCI, fraction of fixed C incorporated). The third parameter is the carbon concentration of the plant material ([C]), which indicates how much C has to be invested in C-skeletons to build a given unit of biomass. In formula

(eqn 1)

Although there is a wide range of information available at the level of the whole plant in terms of changes in growth, and even more at the individual leaf level in terms of C-fixation, we know very little about the whole-plant carbon fluxes and concentrations that basically link the two. In general, we may expect the daily rate of photosynthesis to increase with elevated CO₂, simply because it happens in most or all individual leaves. This indeed is confirmed by measurements in which the C-gain of whole plants is analyzed (Fig. 2b; P < 0.001), although the stimulation can be transient when measurements are made over longer time (Mousseau, 1993; Roumet et al., 2000). Interestingly, although the dataset is small (n = 35) and does not adequately cover the various functional groups, herbaceous dicots showed a stronger increase in photosynthesis than the monocots (31% vs 12%, P < 0.05), which is in line with the stronger increase in ULR observed above.

Information on the fraction of daily fixed C that is incorporated into the plant's biomass is very scarce. With the rate of photosynthesis increased and the rate of respiration hardly affected by CO₂ (Bruhn et al., 2002), one would expect the C-losses relative to the C-gain to decrease and – assuming exudation and volatilization to be constant – FCI would go up. This does indeed occur, although the average increase (6%; Fig. 2b) does not significantly differ from zero and is small compared to the average change in whole-plant photosynthesis (28%).

Carbon concentrations of whole plants are seldomly reported. The C-concentration of leaves generally decreases somewhat for plants that have a high C-concentration by nature, such as woody species, and increases somewhat for plants that inherently have a low C-concentration, such as fast-growing herbs, but the changes are mostly small and do not exceed the 3% (Poorter et al., 1992; Poorter et al., 1997). As changes in the C-concentration of stems and roots are generally less pronounced (Den Hertog et al., 1993), and shifts in allocation between leaves, stems and roots are minimal (Poorter & Nagel, 2000), we expect changes in the C-concentration of the whole plant to be marginal. This is confirmed by the few data in the literature, showing an average increase of 1% only (Fig. 2b; P < 0.05). Thus, as far as information is present, it seems that the observed increase in ULR is mainly due to an increased rate of whole plant photosynthesis per unit leaf area, whereas changes in FCI or whole plant C concentration are minor factors (Fig. 2a,b). Similar conclusions were drawn by Wong (1990) and Evans et al. (2000) in comparisons of ULRs with photosynthetic rates measured on individual leaves.

4. Components of LAR

LAR is the product of the Leaf Mass Fraction (LMF, the fraction of total plant biomass that is allocated to leaves) and the Specific Leaf Area (SLA, amount of leaf area per unit biomass). The change in LMF at elevated CO₂ is generally small (Stulen & Den Hertog, 1993; Poorter & Nagel, 2000; Fig. 2a), although a CO₂-induced shift in allocation towards the roots may occur occasionally (Stulen & Den Hertog, 1993; Sigurdsson et al., 2001), though not invariably (Maroco et al., 2002) when plants are grown at a low nutrient availability. There was no difference in LMF change when various functional groups were compared.

The change in SLA is much stronger (Fig. 2a) and a decrease occurs in almost all C3 plants under a wide range of environmental conditions. SLA depends on differences in leaf anatomy, and should be reflected in the chemical composition on a leaf area basis. Although a meta-analysis is lacking, accumulation of nonstructural carbohydrates is likely to be the main factor for the decrease in SLA (Wong, 1990; Roumet et al., 1996) resulting in an increase in leaf density (Roumet et al., 1999). However, additional effects have been reported, such as an increase in the leaf thickness due to more cell layers or a larger cell size (Thomas & Harvey, 1983; Mousseau & Enoch, 1989; Sims et al., 1998; Lin et al., 2001). There is a substantial difference between the herbaceous monocots and dicots, with dicots decreasing 19% in SLA at elevated CO₂, and monocots decreasing only 7% (P < 0.001). We have no idea to what extent this can be a consequence of the observed larger stimulation of photosynthesis in dicots. Monocots more often accumulate soluble carbohydrates such as fructans, whereas most dicots accumulate more starch (Lambers et al., 1998). If starch accumulation would have less of a feedback on photosynthesis than the accumulation of soluble sugars, this could perhaps form an explanation for the differential decrease in SLA. However, this is at the moment merely a speculative hypothesis.

5. Growth response along a CO₂ gradient

This paper focuses mainly on the effect of a twofold increase in the atmospheric CO₂ concentration, from c. 350 µl l⁻¹−700 µl l⁻¹. However, CO₂ concentrations have covered a much wider range throughout geological time scales, with values estimated as high as 6000 µl l⁻¹ during the Paleozoïcum (500 million years ago) and as low 200 µl l⁻¹ during the late Pleistocene (15 thousand years ago; Berner, 1997). How do growth parameters respond to a wider range of CO₂ levels? One of the most extended growth experiments is that of Neales & Nicholls (1978) on wheat. As can be seen in Fig. 3, strongest changes in parameters like ULR and SLA are in the low concentration range (200–400 µl l⁻¹). Similar to what happens with whole-plant photosynthesis, a clear saturation is shown for these growth parameters at higher CO₂ levels. Another point that should be stressed is that growth analyses generally deal with plants in the vegetative phase. Responses of plants in the generative phase are not necessarily similar to those of vegetative plants (Thomas et al., 1999), a conclusion that probably holds strongest for annual plant species.

1. BER distribution

Our literature compilation on the growth response of plants to elevated CO₂ comprised approx. 350 experiments. Some of these were on one species, others on more, and in a number of cases the same species was studied in different experiments, yielding a total of 800 BER observations for approx. 350 species. The overall distribution of the observed BER values is shown in Fig. 4. Clearly, there is wide variation between observations, with some results showing a BER higher than four, implying an over 300% increase in plant mass due to elevated CO₂, and others a BER lower than 0.7, implying a more than 30% decrease in mass for high-CO₂ plants. The most simple and straightforward explanation would be that these differences are due to variation in response between species, with some species showing consistently high BER values across experiments and others showing low values. We tested this hypothesis by selecting only those data in the compilation for which at least three independent observations per species in different experiments had been made. We arrived at a total of 495 observations for 70 species, which were subjected to an analysis of variance with log_e-transformed BER values as the dependent and species as the independent variable. The factor species explained 31% of the total sum of squares. If we would consider species as a random factor, assuming that these 65 were just a sample out of the total range of species that could be considered, then only 18% of the total variance is due to species. This leaves an uncomfortable 82% as ‘unexplained’.

What may be the reason for such a high variability independent of species? A first possibility is that the biomass stimulation depends on developmental stage. Thus, starting with equally sized plants at the beginning of the CO₂ enrichment, one would expect BER to increase from one to a certain higher value, with possibly only small changes (mostly decreases) thereafter. That may apply to herbs that enter the generative phase (Thomas et al., 1999) and has also been observed for tree seedlings (Norby et al., 1995; Hättenschwiler et al., 1997; Idso, 1999). However, there is as yet insufficient insight into the nature and reasons for these ontogenetic trends, especially in perennial plants. In most herbaceous species it seems that the main part of the growth stimulation occurs within a relatively short period of 2 wk after the start of the CO₂ treatment. Availability of extra sinks to which a high-CO₂ plant can allocate their extra fixed C may play an important role (Reekie et al., 1998). We tried to filter out part of the effect of developmental stage by focusing on vegetative plants. However, even then BER may change nonpredictably during the growth period.

A second possibility for the high intraspecific variability in BER is that the CO₂-induced growth stimulation is strongly dependent on the environmental differences under which experiments are conducted. We expect controlled experiments, which form the main topic of this review, to differ most in the integrated daily quantum flux and possibly in water supply (cf. Garnier & Freijsen, 1994). These two factors have, on average, only moderate effects on BER (see section VI), which are by far too small to explain the range in BER values of Fig. 4. Stronger effects are to be expected if nutrients become limited, in which case we expect BER to be ‘underestimated’, or in laboratories in high-ozone areas, where BER may be ‘overestimated’ due to a strong positive CO₂ × ozone interaction. Especially water and nutrient availability are strongly dependent on the relative size of plant and pot, whereas an environmental factor such as O₃ concentration is hardly ever determined. Therefore, there is not an easy way to statistically control for these environmental differences. Lloyd & Farquhar (2000) suggested an iterative restricted maximum likelihood approach to analyse the effect of species and experiment concurrently. Although less sensitive to unbalanced designs than most statistical tests, such an analysis is only fruitful when larger scale experiments are compared that have a range of species in common. This severely restricts the amount of information that can be used. The approach we will follow is to minimise the between-experiment effects by encompassing information from as many experiments as possible. However, we fully agree with Lloyd & Farquhar (2000) and Gurevitch et al. (2001) that factorial experiments with a range of species as well as environmental conditions are needed to further test our insights at this point.

A third reason for large variability within species may be that different researchers used different accessions or genotypes, which vary in their response to elevated CO₂. Such variation has been ascribed for a number of species (Wulff & Alexander, 1985; Curtis et al., 1996; Schmid et al., 1996; Klus et al., 2001). In most cases, however, there are no independent experiments confirming that a specific genotype that responds strongly in experiment A is also the one with the strongest response in experiment B. As long as such information is not present, it may be questionable to what extent these genotypic differences really play a quantitative role. The reason for this will be discussed in the next paragraphs.

2. Plant-to-plant variability

The three causes given above are biological as well as deterministic by nature, and have been discussed on various occasions. A fourth reason for variation in BER, which has hardly received attention, is that individual plants within an experimental population vary in biomass (within group variation), sometimes to a large extent. Although all experiments in the literature are based on a number of replicates to avoid biased conclusions, plant-to-plant variability per se has hardly ever been the focus of attention of biologists, with the notable exception of size inequality in dense, highly competitive stands (Weiner et al., 1990; Wyszomirski et al., 1999). In the next three paragraphs we assess the importance of plant-to-plant variability. To this end we investigated the variation present in experimental plant populations, the effect of elevated CO₂ thereupon, and the consequences for the reliability of the BER estimate.

One way to characterise plant-to-plant variability is to calculate the standard deviation of the log_e-transformed biomass data (S_lnM). In order to obtain an impression about the size of S_lnM in experimental populations used for growth analysis, Poorter & Garnier (1996) compiled such values for a variety of published experiments. We extended this compilation with data from a range of other studies, separating experiments with herbaceous species from those with tree seedlings. The distribution graphs of this compilation are shown in Fig. 5(a) and (b), respectively. In some cases investigators have selected for homogeneity of their plants before the onset of the experiment, others have taken a random sample from all seedlings available. The data on S_lnM therefore will underestimate the biological variability of the species, but gives a good indication about the homogeneity of the plants used in growth experiments. Taken over all species, the median value of S_lnM is 0.31, with 20% of the values below 0.21, and 20% above 0.51. Variability in populations of herbaceous species (Fig. 5a) is generally lower than that in populations of tree seedlings (Fig. 5b; P < 0.001). Possible reasons could be that a number of experiments on herbaceous species were carried out with genetically homogeneous crop plants and that in a number of tree species seeds are difficult to germinate, causing considerable variation in germination times and therefore large differences in plant size.

Having obtained an impression of plant variability in growth experiments, we turn to the next question to be answered: does the elevated CO₂ treatment affect plant-to-plant variability? This is a relevant question, not only because it could alter the population dynamics of various species, but also because it may affect the strength of the conclusions based on a given sampling scheme. Increases in population variability have been reported, for example, for plants fumigated with SO₂ (Coleman et al., 1990). To test whether this is also the case for plants grown at elevated CO₂, we calculated the S_lnM observed per harvest in a range of CO₂-enrichment studies, with data kindly provided by a number of authors. Thereafter we calculated the ratio of the standard deviations in log_e-transformed dry mass of the elevated-CO₂ and the control plants. The distribution of this ratio for 150 harvests on 60 species in 14 experiments is given in Fig. 5(c). The range is wide, most likely because a proper estimate of population variance is difficult to achieve with just 5 or 10 plants harvested. The average ratio, however, is 1.01 and does not deviate significantly from unity. Therefore, we conclude that plant populations do not become more variable at elevated CO₂. There is no indication of any difference between herbaceous and woody species in this respect (P > 0.8).

To what extent can plant-to-plant variability affect the outcome of CO₂ experiments? Let us assume an extreme case, i.e. the true BER for all 800 observations compiled in Fig. 4 was actually the same, with a value equal to the median of all observations (1.41). For most experiments we do not know the population variability, but we can arrive at an educated guess. Let us assume that the experimental populations can be categorised into three groups, with either a low, an intermediate or a high S_lnM, as derived from the 20th, 50th and 80th percentile of the distribution given in Fig. 5(a) and (b). The third part of information that has to be known is the number of plants harvested per treatment. For the 350 experiments compiled here, a low value is 5 (P20), the median is 8 and a high value is 12 (P80). Using all this information, we postulated a species with a high CO₂ and a control population that differed in mass by 41%, and with equal variability in plant mass. We then simulated experiments with three different population variabilities in plant mass (σ_lnM= 0.21, 0.31 and 0.51) and, for each variability, three different sample sizes (n = 5, 8 and 12). For each of the nine combinations we computer-generated 10 000 experiments, and for every experiment we calculated an average BER. The resulting distribution of 90 000 BER values, which we consider to be representative of what could be expected from a compilation of all of the actually observed BER results, is shown as the continuous line in Fig. 4. Compared to the actually observed distribution, the simulated distribution is surprisingly similar and no significant difference between the two could be found with a χ² test ( df = 17, P > 0.15). Therefore, we have to conclude that variability in the plant population plays an unfortunate, but large role in determining the outcome of experiments. Clearly, the number of plants harvested is often insufficient given the range in plant mass within an experimental group of plants. This is especially problematic as variability in BER is the sum of the variability in both the numerator and denominator of the BER calculation (cf. Jasienski & Bazzaz, 1999).

3. Consequences for data interpretation

What are the take-home-messages of this analysis? The first conclusion is that in these type of experiments attention should be given to a sufficiently homogeneous experimental population. Clearly, the outcome of single experiments may vary to a large extent if highly variable populations are investigated. This is the more critical as the growth stimulating effect of CO₂ is relatively small compared to the effect of other environmental variables such as nutrient availability or light intensity, where plant masses may vary fivefold or more across treatments. Consequently, it requires more precision to separate the CO₂ effects from error variation than in the case of light or nutrient effects. Second, credibility for one or another hypothesis explaining interspecific variation in growth responses of plants cannot come from results of one or two specific experiments, because repeatability is not very high. Rather, we think that a wider range of experiments should be performed before we can conclude that a (group of) species behaves differently with respect to CO₂ than another (group of) species. Such an approach bears the risk that experiments with different experimental conditions are compared. As discussed above, it is impossible to quantify these effects for each of the experiments described in the literature. This may imply that differences between species are confounded to a certain extent with growth conditions. However, given the large number of species and experiments, the chances that this may drastically affect the outcome of the analysis seem small. The other side of the coin is that the wide variety of experimental conditions allows for more general statements (Poorter, 1993). Thirdly, no matter what groups of species are compared, or how similar experimental conditions were, there will always be a lot of scatter, so the amount of variation explained by any contrast between groups of species will at best be modest. For the analysis of BER in the next section of this paper, we choose to consider variation at the level of species. Therefore, we averaged all data per species, assuming that most of the variation at the within-species level was random.

There is one other aspect that deserves attention. The timescale for the compiled experiments varies from 10 d to 50 months, with a median value of 4, 7 and 16 wk for crop, wild herbaceous and woody species, respectively. As discussed in section III, most of the CO₂-induced growth stimulation is generally found at the beginning of the treatment. However, for most species time-dependent changes have not been investigated and the phenomenon is in general ill-understood. It can be shown that it is the absolute difference in RGR that translates into a relative biomass stimulation after a given period of time. From the BER at the end of the experiment and the duration of the CO₂ fumigation we can back-calculate the average absolute stimulation in RGR over the whole experimental period, assuming equal seed or seedling size at the onset of CO₂ enrichment (for a mathematical derivation see Poorter et al., 1996). In the analyses in section V, we decided to consider only those differences between groups relevant where both the average BER change and the RGR stimulation show similar direction and statistical significance.

1. C3, C4 and CAM

In this section we analyze to what extent functional groups of species differ in their growth response to elevated CO₂. The most obvious categorization that could be envisaged is based on the type of photosynthesis (Bazzaz & Catovsky, 2002). The commonest group of species, the C3 species, has a limiting CO₂ concentration at the level of the chloroplast and therefore can be expected to respond with an increased C-gain upon a rise in the CO₂ concentration around the leaves. Plants with a C4 type of photosynthesis posses a CO₂ concentrating mechanism that increases the CO₂ concentration at the site of Rubisco to c. 2000 µl l⁻¹ (Sage, 2001). At this concentration the oxygenating function of Rubisco is repressed, and the carboxylating function is almost saturated. Purely on that basis, C4 plants are expected to respond not or at best marginally to a rise in atmospheric CO₂.

Analysis of the BER values complies with these notions. C3 plants show, on average, the strongest response (+45%; 300 species) and C4 plants the smallest (+12%; 40 species; Fig. 6). Both groups have BER values significantly higher than one, implying that at least a number of C4 species are stimulated in their growth as well. Several explanations are possible for this intriguing response (Ghannoum et al., 2000). First, photosynthesis may not be completely saturated at current CO₂ concentrations in some or all C4 species. Second, CO₂-induced decreases in stomatal conductance may reduce transpiration, thereby conserving soil water that can be deployed for extra photosynthesis later in time. Thirdly, the decrease in transpiration could increase leaf temperature. As C4 photosynthesis is strongly dependent on temperature, this could be a likely explanation as well. A fourth alternative that has been mentioned is that cotyledons of C4 species may be C3 like, or that young developing leaves of C4 species are leaky and therefore may increase photosynthesis at elevated CO₂ (Dai et al., 1995). However, this last option is not considered likely (Ghannoum et al., 2000). The increase in daily photosynthesis required to explain the improved growth is in the range of 2% only (Poorter, 1993), a value that may easily be reached via any of these alternatives.

CAM species are also stimulated, with an average response in between that of C3 and C4 species (23%; P < 0.001). One of the physiological reasons for the positive growth response is that a number of these species are CAM-facultative, they can switch to a C3 type of photosynthesis when water is available. Others show direct CO₂ fixation by Rubisco early in the morning or late in the afternoon. Under these circumstances an increased C-fixation is to be expected. However, CO₂ fixation by PEP-carboxylase during the night is also stimulated by elevated CO₂ (Li et al., 2002). This is an ill-understood phenomenon, as it is generally thought that this enzyme is saturated at current CO₂ levels (Drennan & Nobel, 2000). Unfortunately the total number of CAM species measured is very low (nine), so any generalization for this group of species is premature. One interesting test would be to compare the response of species that are CAM facultative and those that are obligatory CAM in their CO₂ fixation. A point of attention is that in quite some cases CO₂ treatment did not start with seeds or recently germinated seedlings, but with larger plants or cuttings. Calculating a BER value on the basis of total biomass may then underestimate the real growth response. It would require a time-course of the growth stimulation to analyse whether the time-dependency of the stimulation, as observed for C3 species, also occurs in this group of plants.

Testing with a one-way ANOVA we found the differences between plants of the photosynthetic groups highly significant (P < 0.001). However, again the proportion of the sum of squares explained by the model is low, being 10% only. Notwithstanding this variation, the absolute differences between the functional groups of species are quite consistent and do hardly differ from earlier analyses, based on 1/3 or 2/3 of the now available data (Poorter, 1993; Poorter et al., 1996). We checked the effect of data base size by randomly dividing the 800 observations into three groups, recalculating mean responses of C3, C4 and CAM species for each of the groups. Although small differences between the three data sets were present, they did not affect the main conclusion to any extent. Therefore, we conclude that the sample size of this database is sufficient for robust conclusions with regard to the investigated C3 and C4 species.

2. Within C4 species

There are indications that not all C4 species respond to the same extent. Ziska & Bunce (1997), for example, found Amaranthus retroflexus and five other C4 weed species to respond more strongly than Zea mays and some other C4 crop species. Differences in BER between Amaranthus retroflexus and Zea mays have been quite systematic, indicating a species–specific response (Poorter et al., 1996). These species belong to different C4-subtypes, which vary in leaf morphology as well as in the enzymes used for decarboxylation of the C4 product formed. The NAD-ME type is considered to have vascular bundle sheath cells that are the most leaky for CO₂, and therefore may respond more to an increase in CO₂. LeCain & Morgan (1998), however, found NAD-ME species to have lower BER values than NADP-ME species and a similar trend was observed by Wand et al. (2001). Ziska & Bunce (1997) could not relate differences in BER to the subtype. Averaged over all observations available, there was no difference between species of the two groups (Table 1). There are only few data for C4 species of the PCK subtype, so we did not include them in the tests, but contrary to the other subtypes the species investigated so far showed a negative growth response. A firm conclusion awaits more evidence.

Alternatively, a difference between Amaranthus and Zea might be caused by the very different genetic background, with one species belonging to the dicots, the other to the monocots. The number of dicots investigated is small and although they seem to respond more than monocots, the difference is not significant. Simply analyzing C4-subtype and lineage (monocots vs dicots) separately may ignore the fact that these factors could be confounded to a certain extent. This would be especially true if, by chance, the investigated monocots were almost all from the NADP-ME subtype and the dicots from the NAD-ME subtype, or vice versa. Therefore we analysed the BER and RGR response of all NAD-ME and NADP-ME species with a multiple regression, in which we entered C4-subtype and lineage as dummy variables. As long as two traits are not completely correlated, such a regression can separate the effect of both factors. Finally, we also tested the significance of the difference between the groups without aggregating data per species. Table 1 shows the result of these analyses: there is no systematic effect of subtype or lineage in the compiled data set, independently of the way we tested the data and whether the growth stimulation was expressed as BER value or as increase in RGR. For the time being, with so few data on PCK species, we conclude that any systematic difference between C4 species has to be explained by other factors than subtype or lineage. Wand et al. (2001) suggested that C4 species with a large growth potential responded more than slower-growing species. This is a factor worth investigating, although it seems at variance with the meta-analysis of Wand et al. (1999). They found wild C4 species to respond more than what is generally found for weedy or crop C4 species, where we would presume these last two groups to be faster-growing.

3. Within C3 herbs

Plants within the group of C3 species share the same type of photosynthesis, but will that imply that they all behave the same in response to elevated CO₂? We considered three different contrasts: the one between monocots with dicots, the possibility to fix atmospheric nitrogen in one symbiotic relationship or another, and the effect of inherent differences in growth rate. Again we carried out a multiple regression with these factors as dummy variables to exclude to some extent the fact that these factors are not distributed randomly over our database. Table 2 shows that there was a significant increase in the RGR stimulation of those plants that are capable of symbiotic N₂ fixation. However, as the actual N₂-fixation strongly depends on nitrate availability in the root environment (Lambers et al., 1998), it is not known to what extent these species really were relying on atmospherically fixed N. Moreover, there was no significant difference in BER. There was a tendency for monocots to respond more strongly than dicots, but again the statistical significance strongly depended on the way the data were treated, and whether the stimulation was expressed as BER difference or RGR increase. Therefore, we conclude for the moment that monocots and dicots show similar growth responses to elevated CO₂.

The largest and most consistent difference was the one between inherently fast- and slow-growing species. No matter how the data were analysed, fast-growing species showed a much greater response than slow-growing species, the difference being of the same magnitude as that for C3 and C4 plants in general. A simple explanation is that fast-growing species operate with a higher LAR than slow-growing species (Poorter & Van der Werf, 1998). If elevated CO₂ stimulates photosynthesis and also ULR to the same extent in slow- and fast-growing plants, then the proportional increase in RGR is the same for species of both groups, but the absolute increase in RGR is larger for the fast-growing species. And as it is the absolute increase in RGR that determines the relative response after a given period of time, this could explain the results. However, other differences between the species could play a role as well. Source–sink interaction is an important factor in determining the growth response, with the highest BER values observed for species with large sinks (Reekie et al., 1998). We categorised all crop species as being fast-growing. They generally have growth rates that come close to those of the inherently fast-growing wild species. It may well be that such species are better able to lay down new meristems and invest the extra-fixed C in structural biomass than slow-growing species.

There have been other efforts in the literature to classify groups of species with contrasting characteristics. Indeterminate as opposed to determinate growth has been mentioned as a factor that may increase the growth response to CO₂ (Oechel & Strain, 1985; Ziska & Bunce, 2000), again because it will be easier to deploy extra-accumulated sugars. Mycorrhizal species may easily metabolise sugars in the fungal network, with possible beneficial effects (Díaz et al., 1993), but in another experiment a nonmycorrhizal species like Carex flacca was responding better than all other species (Leadley & Körner, 1996). Another contrast that has been studied is between plants that load their sugars differently into the phloem. These differences between symplastic and apoplastic phloem loaders are correlated with life form and ecological niche (Van Bel, 1999). Körner et al. (1995) investigated possible differences in starch accumulation between the groups, but were unable to find any. Differences in growth response have not been studied.

4. Within woody C3 species

In terms of number of species investigated, more work has been done on woody species than on herbaceous ones. With regard to the BER, no difference at all could be found between these two groups of species (Fig. 6). However, most experiments with woody species last longer than those for herbaceous species, implying that the RGR stimulation is much smaller in the case of woody species. It is obvious that most of the work on woody species is focused on young tree seedlings, with a median duration of CO₂ enrichment of 16 wk only. Therefore, it is difficult to forecast longer-term responses of larger trees on this basis. Longer-term enrichment studies on a few species indicate that an average stimulation of 30–50% is achieved in the field as well as in open-top chambers (Curtis & Wang, 1998; Idso, 1999). This is not different from the average stimulation for younger trees (Fig. 6).

In this analysis we considered possible functional groups on the basis of three contrasts: evergreen vs deciduous species, N₂-fixing plants compared to those without the possibility of N₂-fixation, and Gymnosperms vs Angiosperms. Only the last group showed a significant difference, with Angiosperm seedlings showing a somewhat stronger response than Gymnosperms (Table 3). This is in agreement with what was found by Ceulemans & Mousseau (1994), but Curtis & Wang (1998) did not find significant differences, and Saxe et al. (1998) even claimed the opposite. Another point of uncertainty is that evergreen species are generally slower-growing than deciduous species. Therefore, we had expected to see similar differences as for fast- and slow-growing herbaceous species.

Another attempt for classification has been made by Kerstiens (2001). In a meta-analysis of a more specific data set of woody species, he found BER to be higher for shade-tolerant species than for intolerant ones, especially at high light. This was partially confirmed in an experiment included in Bazzaz & Catovsky (2002) where coniferous seedlings complied with these expectations, but not the deciduous species. On the contrary, Winter & Lovelock (1999) found the shade-intolerant pioneer species to show larger responses than the shade-tolerant climax species. Combined with the fact that most experiments cover such a limited part of the life cycle of these plants, we have to conclude that classification of woody species into functional groups is still complicated.

5. Fast-growing versus slow-growing

In the above analysis of the woody species, we did not classify these species with respect to their growth rate, as we felt that we did not have a good overview over this parameter. For herbaceous C3 plants we were able to categorise species as slow-, intermediate and fast-growing. These categories are to some extent arbitrary, and attention has been drawn to the fact that these differences between fast- and slow-growing species may not show up in each experiment (Lloyd & Farquhar, 2000). Apart from the variability discussed in section IV, such ‘inconsistencies’ may also be due to misclassification of the species, or to specific conditions used in a particular experiment. We therefore extended the analysis, by selecting those experiments for which RGR of these species was specifically measured. Different experiments have different time-frames, and RGR is not always measured from the start of the CO₂ enrichment. Therefore, we derived the increase in RGR due to elevated CO₂ from the difference in the BER values and the duration of the CO₂ enrichment, as discussed in section IV. If the absolute increase in RGR is plotted against the measured growth rate of control plants, a strongly positive correlation is found: the higher the RGR at 350 µl l⁻¹ CO₂, the stronger is the absolute RGR response (Fig. 7; r² = 0.52; P < 0.001).

The relationship shown in Fig. 7 comprises both herbaceous and woody species. The good fit may partly be fortuitous: by considering the absolute increase in RGR we standardise time in the sense that the average growth stimulation per day over the whole experimental period is calculated. As most of the response is in the beginning of the growth period, and as experiments with tree seedlings are generally lasting for a longer time than those with herbaceous plants, the correlation may in part be due to the fact that tree seedlings grow more slowly than herbaceous plants, and are measured over a much longer time span. The positive relationship, however, also holds when herbaceous and woody species are considered separately, with some indication that woody species respond a bit stronger at a given RGR than herbaceous species (P < 0.05). An alternative test is to analyse the relationship within experiments, thereby minimizing the possibility that slow-growing species were experiencing less favourable conditions than the fast-growing species in the compilation. In almost all experiments where a range of species with different growth rates is compared (either tree seedlings or herbaceous plants), the faster-growing species respond more strongly than slow-growing species. In a number of cases the response is positive, but statistically above the 0.05 level (Campbell et al., 1991; Roumet et al., 1996; Cornelissen et al., 1999), whereas in others the response is significant (Poorter, 1993; Bunce, 1997; Atkin et al., 1999; Winter & Lovelock, 1999). For 20 out of the 24 experiments that included species comparisons and determined RGR values, slopes of the regression lines were positive, and the average slope (0.088, very similar to the overall slope of 0.095) was significantly higher than zero (P < 0.001). Therefore, we consider the broad picture emerging from the literature confirmed by most individual papers, and we conclude that under favourable conditions high-RGR species will respond more strongly to elevated CO₂ than low-RGR species. The mean difference in BER is of similar magnitude as the one observed between C3 and C4 species. The fact that fast-growing species respond more strongly to elevated CO₂ has a clear parallel in the growth responses observed at different light or nutrient availabilities. Also in those cases the inherently fast-growing species will respond strongest.

1. Limiting conditions

One of the goals of the quantitative analysis presented in section V is to find a classification that might be helpful as a basis for prediction of changes in natural vegetations. Will C3 species expand relative to C4 species, and will fast-growing species thrive at the cost of slow-growing ones? There are a number of complications to which we briefly would like to draw the attention. First, the analysis in section V is carried out with species that were grown under more or less ‘optimal’ conditions. This will allow most plants to show their maximal response to CO₂, without environmental constraints or stresses. In a natural environment conditions will generally be less favourable. Poorter & Pérez-Soba (2001) reviewed the CO₂ response of isolated plants when grown under a variety of stresses. Table 4 gives the average BER of C3 species in the close-to-optimal situation as well as in the case that a given environmental factor causes biomass to be reduced by 50% at control levels of CO₂. For most of these factors (irradiance, water, salinity, UV-B) changes in BER are small. Interactions are more substantial in ozone-stressed plants, for which BER is strongly promoted, and for cold- or nutrient stressed plants, where the BER is clearly lower than for plants grown at close-to-optimal conditions. In most natural environments nutrient availability will be low, so purely on that basis we expect the response under those conditions to be small.

2. Competition versus isolated plants

The second factor that makes a difference between most laboratory experiments and the field is that plants in the lab often grow without any mutual interference at the leaf or root level. This implies that an extra investment in leaves or roots can immediately pay off in the form of extra carbon and nutrient capture, which will, in the absence of sink limitation, result in an extra stimulation in growth. The situation is different when plants are grown together. At low density, total biomass of a monoculture will increase linearly with density, but as crowding becomes stronger the biomass of the stand saturates to a maximum level, with only very limited space for each individual. Under crowded conditions, extra leaf area will not necessarily lead to extra carbon gain. Since both the threshold density and the slope vary among species, the simplest comparison is the biomass of isolated plants with those in crowded monocultures. Under those conditions, woody and herbaceous species are generally responding less to elevated CO₂ than individually grown plants (Du Cloux et al., 1987; Wayne & Bazzaz, 1995; Retuerto et al., 1996; Navas et al., 1999). Taken over a range of experiments, no correlation was found between the response of a species grown in isolation and in monoculture (Fig. 8a, r² = 0.06).

The next level of complexity comes in when mixed stands are analysed. The response to elevated CO₂ of a particular species will then not only depend on its own physiological and morphological characteristics, but is also determined by the secondary interactions that arise with the other species that are competing for the same resources (Firbank & Watkinson, 1990). Therefore, the correlation between the BER of a given species grown in isolation and in competition with other species can be expected to be even lower as the one between isolated plants and monocultures, and this happens to be the case (Fig. 8b, r² = 0.00). The most unpredictable step appears to be the transition from isolated plants to monocultures. The second step, from monostands to mixed stands, shows a much better correlation (Fig. 8c, r² = 0.33), confirming a previous study by Navas et al. (1999) on artificial herbaceous communities. Therefore, we conclude that any prediction of species responses in a vegetation would be better off with growth analyses at the stand level than at the level of the individual.

3. Functional groups

Alternatively, we could use published competition experiments, to test whether specific groups of plants profit more than others. A good example is shown in Fig. 9, where results of Winter & Lovelock (1999) and Lovelock et al. (1998) are combined. They grew isolated seedlings of nine tropical tree species in open top chambers at ambient and elevated CO₂, and found a stronger response for the fast-growing pioneer species as compared to the slow-growing climax species. This is in agreement with the conclusions of section V. However, when almost the same set of plant species was grown in competition, BER values for all species ranged around 1, with no difference between species that responded strongly or weakly in isolation. Another example of a very poor correlation between the prediction for isolated plants and those in competition are experiments in a calcareous grassland vegetation. The species that showed the strongest growth response, both in the field as in the lab was Carex flacca (Leadley & Körner, 1996; Stöcklin & Körner, 1999), a species with a very low potential growth rate (Van der Werf et al., 1993).

Is it possible to discriminate between groups of species that form ‘winners’ and ‘losers’ in competitive situations? An extended review is given by Reynolds (1996). Similarly as for isolated plants we analysed a number of competition experiments retrospectively for differences in response between functional groups of species. To this end, we used the BER of the whole artificial or natural vegetation as a calibration point. For each species of the mixture we calculated the BER of that species, and divided it by the BER value of the whole vegetation. If this ratio is higher than 1.0 the species is profiting disproportionately and is designated as a ‘winner’. If the ratio is lower than 1.0 the plant would lose out compared to the whole vegetation. On the one hand, it may be naive to test for such a general response for a given group of species, as competition will strongly depend on the competing species that are present, as well as the specific environmental conditions. On the other hand, small differences between species that are hardly of relevance for plants grown in isolation can be of crucial importance in a competitive situation and may magnify differences in response between species. We felt it appropriate to formally test for these winners and losers anyway. We restricted our analysis to competition experiments carried out with herbs, as most of the work in this field has concentrated on this group, but excluded a-priori those species from the analysis that represented less than 2% of the total biomass of the vegetation, as the behaviour of these plants may be erratic if only a few individuals are present. Finally, given the strong difference in response of nutrient-rich and nutrient-stressed plants (Table 4), we classified experiments as carried out under either high or low nutrient conditions. A classical problem in this case is that the observations on different species within a competition experiment are definitely not independent of each other. A very conservative solution is to use only one species per experiment. This would have resulted in a serious loss of information, an aspect we considered more problematic than statistical independence (Gurevitch et al., 2001).

The results are shown in Table 5. As in section V, we analysed the differences both as simple contrasts and in a multiple regression. For experiments with high nutrient levels, the only significant difference found was between C3 and C4 species, with C4 species being the losers at elevated CO₂. However, the number of observations for C4 species is rather low (< 15). The difference remains significant in the multiple regression analysis and is in line with the difference we have seen between C3 and C4 species at the individual plant level (Fig. 6). By contrast with the observations at the individual plant level, fast- and slow-growing species respond exactly similar to elevated CO₂ under competition. Again, the number of species in one of the categories is low, but as it is in accordance with the idea that there is little scope for fast-growing plants in a vegetation to profit from the extra investments they made, we have as yet no reason to doubt these conclusions. No differences were found between N₂-fixing species and other dicots, or between monocots and dicots in general.

For competition at low nutrient levels, the situation is different. Here no differences between C3 and C4 species are observed, although we stress again that the number of C4 species investigated is low. The fact that C4 species are not negatively affected under these circumstances might well be due to the fact that at a low nutrient level a large CO₂ response of C3 species is precluded (Table 4). The exception to this rule is the group of species capable of symbiotic N₂ fixation, they are clearly the winners under these conditions. There is some indication of a difference between dicots and monocots, but closer analysis showed that the response only came from the nitrogen-fixing dicots. Decreases in grasses and increases in dicots, mostly due to an enhanced biomass of legumes, were found in most field studies (Schäppi, 1996; Clark et al., 1997, Navas et al. 1997; Lüscher et al., 1998; Warwick et al., 1998; Reich et al., 2001). The increase in leguminous species is particularly evident under conditions of low N and high P availability (Stöcklin & Körner, 1999; Körner, 2001).

Competition per se can be thought of as consisting of two components: the competitive effect, which is the ability of a plant to suppress neighbours, and the competitive response, the ability of a plant to tolerate its neighbours (Goldberg, 1990). An estimate of the first component is the dominance of a species in a community. In some field studies, not so much the dominant but some subordinate species were found to be highly responsive to CO₂ (Leadley & Körner, 1996; Clark et al., 1997, Navas et al. 1997; Berntson et al., 1998; Stöcklin & Körner, 1999). It has therefore been suggested that elevated CO₂ may reduce the overall size difference between dominant and subordinate plants (Catovsky & Bazzaz, 2002). Does that imply that we can consider subordinate species as a special ‘response group’, whose inherently low competitive effect is compensated for by a high responsiveness to CO₂? As mentioned above, just by the nature of the fact that a species forms a minority in a vegetation, it may show larger proportional fluctuations than dominant species. It could well be that large proportional increases in a species strongly draw the attention of the researchers. Considered over all competition experiments compiled, we tested whether subordinate species are more often winners than dominant species, calculating the percentage of the total stand biomass taken up by a given species as an estimate for dominance. Using this parameter as the independent variable and the winner scale as the dependent variable, we did not find any indication of a difference between subordinate and dominant species (Fig. 10a; r² = 0.00, P > 0.4), although the former show larger variability in their response to CO₂ than the latter.

The second factor that may play a role in the response to CO₂ is the reaction of a species to competition from neighbouring vegetation. It can be estimated by the Relative Competition Intensity (RCI; Wilson & Keddy, 1986; Keddy et al., 1998), which is defined as the absolute decrease in the biomass of species because of competition, normalised against the biomass of isolated individuals. An RCI value of zero means that competition has no effect on plant performance, whereas an RCI value of one corresponds to complete competitive exclusion. Catovsky & Bazzaz (2002) suggested that tree species with a high response to elevated CO₂ were those that suffered less from neighbouring plants when grown in competition. However, tested for herbaceous plants, we were not able to find a correlation between our winner scale and RCI (Fig. 10b; r² = 0.07, P > 0.1).

In conclusion, there is no scope for using the response of isolated plants as a predictor of changes in the vegetation. Furthermore, there is no relationship between the competitive ability of a species and its responsiveness to CO₂. As far as differences can be generalised over a larger group of experiments, C3 species may win from C4 species in vegetations with a high nutrient availability, and nitrogen-fixing dicots may profit at low nutrient availability.