Metabolite profiles reveal interspecific variation in operation of the Calvin–Benson cycle in both C4 and C3 plants

Interspecific divergence in metabolite profiles in C3 and C4 species points to differing evolutionary trajectories of the Calvin–Benson cycle in different seed plant lineages


Introduction
The Calvin-Benson cycle (CBC) evolved ~2 billion years ago (Rasmussen et al., 2008), is the most abundant biochemical pathway on Earth in terms of nitrogen investment (Ellis, 1979;Raven, 2013), and plays a dominant role in the global carbon (C) and O 2 cycles. The CBC can be divided into three partial processes; fixation of CO 2 (ribulose-1,5-bisphosphate carboxylase-oxygenase) RuBisCO into a 3-C compound, 3-phosphoglycerate (3PGA), reduction of 3PGA to triose phosphate (triose-P) using ATP and NADPH from the light reactions, and a series of reactions that use triose-P to regenerate ribulose 1,5-bisphosphate (RuBP) (von Caemmerer and Farquhar, 1981;Heldt, 2005;Stitt et al., 2010;Adam, 2017). The net gain in C exits the CBC and is converted into endproducts. Despite its evolutionary age, the pathway's structure is essentially unchanged from cyanobacteria to angiosperms.
This conservation of the CBC pathway structure is remarkable. The CBC evolved in a world in which CO 2 concentrations were very high and O 2 concentrations were very low. Over geological time, there has been a dramatic rise in atmospheric O 2 and decline in atmospheric CO 2 . This uncovered a side reaction with O 2 , which competes with CO 2 as a substrate for RuBisCO, leading to the formation of 2-phosphoglycolate (2PG) (Lorimer and Andrews, 1973;Lorimer, 1981;Tcherkez et al., 2006). 2PG is recycled via an energetically wasteful process termed photorespiration that results in the loss of 0.5 CO 2 per scavenged molecule of 2PG (Somerville, 2001;Heldt, 2005). In the current atmosphere with 0.04% CO 2 and 21% O 2 , in C 3 plants about every fourth reaction is with O 2 instead of CO 2 , leading to a 20-30% decrease in the net rate of photosynthesis (Osmond, 1981;Sharkey, 1988;Long et al., 2006;Betti et al., 2016). This side reaction decreases nitrogen use efficiency, because higher amounts of protein must be invested in the photosynthetic apparatus. This includes an especially large investment in RuBisCO, which has a relatively low catalytic rate and represents up to half of leaf protein (Ellis, 1979;Betti et al., 2016). It negatively impacts water use efficiency because a higher internal CO 2 concentration is required to support a given net rate of photosynthesis, which in turn requires higher stomatal conductance and higher evaporative water loss (Ort et al., 2015;Betti et al., 2016).
Cyanobacteria and eukaryotic algae possess C-concentrating mechanisms (CCMs) that accumulate CO 2 in RuBisCOcontaining microstructures, the carboxysome in cyanobacteria and the pyrenoid in eukaryotic algae (Badger et al., 1998;Giordano et al., 2005;Kerfeld and Melnicki, 2016;Raven et al., 2017). These microstructures were lost in plant lineages that colonized the land. A second type of CCM evolved in terrestrial plants in the last 30 million years (Sage et al., 2012;Raven et al., 2017), coinciding with the decline of CO 2 from ~1000 ppm to <300 ppm during the Oligocene (Christin et al., 2008;Zachos et al., 2008;Edwards et al., 2010). These CCMs are in essence biochemical CO 2 pumps, in which bicarbonate is fixed into 4-C acids that are subsequently decarboxylated to generate a high internal CO 2 concentration. In C 4 plants, bicarbonate is typically captured by phosphenolpyruvate (PEP) carboxylase in mesophyll cells, and 4-C acids diffuse to bundle sheath cells, which are located internally within the leaf and contain RuBisCO and the rest of the CBC (Hatch, 2002;von Caemmerer and Furbank, 2003;Sage et al., 2012;Sage, 2017). There is substantial diversity in the pathway of C 4 photosynthesis; for example, which 4-C and 3-C metabolites are involved in the CCM, how the 4-C acid is decarboxylated, and to what extent PSII activity is lost in the bundle sheath chloroplasts. C 4 photosynthesis evolved independently >65 times in separate lineages among the angiosperms, and C 4 species currently represent ~3% of terrestrial plant species and account for 23% of total terrestrial C gain (Still et al., 2003;Sage et al., 2011;Sage, 2017). An analogous biochemical CO 2 pump evolved in plants with Crassulacean acid metabolism (CAM); bicarbonate is assimilated in the dark into 4-C acids, which are decarboxylated in the light to provide CO 2 for the CBC (Shameer et al., 2018). CAM evolved in at least 35 independent lineages and is found in ~6% of current terrestrial plant species (Silvera et al., 2010). Parallel evolution of C 4 and CAM in many lineages underlines the strong selective pressure exerted by low CO 2 in the recent geological past.
CCMs are complex traits. For example, C 4 photosynthesis requires major changes in leaf development and anatomy, gene expression patterns, and the location, levels, and properties of hundreds of enzymes and transporters (Sage et al., 2012;Heckmann et al., 2013;Sage, 2017). It is likely that its evolution involved successive steps, including the development of denser venation, modification of the size and functionality of bundle sheath cells, and stepwise specialization of metabolism in the bundle sheath and mesophyll cells (McKown and Dengler, 2007;Kocacinar et al., 2008;Nelson, 2011;Sage et al., 2013;Mallmann et al., 2014). This multistep evolutionary trajectory may explain why CCMs evolved in only a relatively small fraction of terrestrial plant lineages (Heckmann, 2016).
Low CO 2 will have exerted massive selective pressure on the CBC in species that did not evolve a CCM, representing ~90% of existing terrestrial plant species (Silvera et al., 2010;Sage, 2017). Pressure will also have been exerted by other environmental factors such as water availability, temperature, and nutrient availability (Raven et al., 2017). Indeed, terrestrial C 3 plants exhibit substantial variation in photosynthetic rate, with large differences between annuals and perennials, and considerable differences within these groups (Evans, 1989;Wullschleger, 1993). This includes variation in photosynthetic rate between phylogenetically related species (Galmés et al., 2014b) and within species (Driever et al., 2014). Factors contributing to variation in photosynthetic rate include differences in the rate of electron transport and carboxylation (Wullschleger, 1993), leaf nitrogen content and photosynthetic nitrogen use efficiency (Field and Mooney, 1986;Evans, 1989;Hikosaka, 2010), and differing investment strategies in short-lived (deciduous) and long-lived (evergreen) leaves (Wright et al., 2004;Donovan et al., 2011).
We know relatively little about whether there is interspecific variation in the CBC in C 3 plants (Lawson et al., 2012). It is well established that RuBisCO kinetics have evolved over a long geological time scale, with selectivity for CO 2 rising and catalytic rate declining between cyanobacteria and higher plants (Jordan and Ogren, 1981;Badger et al., 1998;Tcherkez et al., 2006;Savir et al., 2010;Sharwood et al., 2016a, b). Intriguingly, there is also variance over shorter evolutionary time scales. RuBisCO kinetics vary between quite closely related C 3 species (Yeoh et al., 1980;Galmés et al., 2014a;Prins et al., 2016). In perennial oak, ecological adaptations have been linked to specific amino acid polymorphisms in RuBisCO (Hermida-Carrera et al., 2017). RuBisCO is inhibited by RuBP and low molecular weight inhibitors that derive from catalytic infidelities of RuBisCO or, like 2-carboxyarabinitol 1-phosphate, are synthesized by other enzymes (Yeoh et al., 1980;Parry et al., 2008). There is surprising diversity in the levels and dynamics of these low molecular weight inhibitors in different C 3 species (Servaites et al., 1986;Moore et al., 1993;Charlet et al., 1997;Parry et al., 2008) and, incidentally, different C 4 species (Carmo-Silva et al., 2010). CP12 is a small regulatory protein that interacts with NADP-glyceraldehyde-3-phosphate dehydrogenase (NADP-GAPDH) and phosphoribulokinase (PRK) (Gontero and Maberly, 2012;López-Calcagno et al., 2014). The action of CP12 varies between C 3 species (Howard et al., 2011;López-Calcagno et al., 2014), again pointing to interspecies variation in CBC regulation.
Some of the strongest evidence that the CBC can adapt to selection or relaxation of selection in a relatively short evolutionary time comes from studies of C 4 species. Compared with C 3 species, C 4 species contain forms of RuBisCO with a lower affinity for CO 2 and faster catalytic turnover (Yeoh et al., 1980;Sage and Seemann, 1993;Kapralov et al., 2011;Galmés et al., 2014b;Sharwood et al., 2016a, b), allowing a substantial decrease in RuBisCO abundance (Long, 1999;Ghannoum et al., 2005;Sharwood et al., 2016a, b, c). Such changes are found even within the tribe Paniceae in which C 4 photosynthesis evolved recently (Sharwood et al., 2016a).
The operation of a pathway depends on many factors, including the abundance of the participating enzymes, their kinetic properties, and the action of regulatory mechanisms on individual enzymes and sets of enzymes. It is laborious to characterize variation in all these potential factors. Analyses of steady-state metabolite levels provide a top-down strategy to search for variation in pathway operation. This is because changes in enzyme abundance, properties, or regulation will all lead to changes in the relative levels of the metabolic intermediates in a pathway.
Information about CBC intermediate levels in different C 3 species is rather sparse. Most previous studies in C 3 plants focused on RuBP (e.g. Sage and Seemann, 1993) or a handful of metabolites such as 3PGA, triose-P, and fructose 1,6-bisphosphate (FBP), and were restricted to single species (see Stitt et al., 2010 for references). A similar picture holds for C 4 plants (Stitt and Heldt, 1985;Usuda, 1987;Leegood andvon Caemmerer, 1988, 1989). The reason was partly conceptual, reflecting the idea that photosynthesis is usually limited by the light reactions or RuBisCO (Farquhar et al., 1980). Subsequent work has highlighted that photosynthesis can also be limited by reactions in the remainder of the CBC (see Stitt et al., 2010 for a review), especially sedoheptulose-1,7-bisphosphatase (SBPase) (Raines et al., 2000;Lefebvre et al., 2005;Zhu et al., 2007;Ding et al., 2016;Driever et al., 2017;Simkin et al., 2017). There were also technical reasons; until ~10 years ago it was impossible to quantify many CBC intermediates routinely. This is now possible using HPLC-MS/MS (Cruz et al., 2008;Arrivault et al., 2009;Hasunuma et al., 2010;Ma et al., 2014).
In this study, we have profiled CBC intermediates in four C 4 species and five C 3 species, representing diverse plant lineages including eudicots and monocots. We used these data to address two questions. The first is whether CBC intermediates display different profiles in C 3 and C 4 species, as would be expected if the presence of a CCM allows a different mode of CBC operation. This question provides a check that expected differences in CBC operation can be detected as changes in CBC metabolite profiles. In particular, we might expect that the lower abundance of RuBisCO (see above) results in lower levels of RuBP. Furthermore, C 4 species with dimorphic chloroplasts might have enhanced levels of 3PGA and triose-P to support an intercellular shuttle that transfers energy from the mesophyll to the bundle sheath cells. The second and major question is whether there are interspecific differences between C 3 species. This would have important implications for the evolution of the CBC and the need for a better understanding of the pathway in a broader range of C 3 species, including many of our major crops.

Plant growth and harvest
Nine species (of which eight were phylogenetically diverse; Supplementary  Fig. S1 at JXB online) were grown as described in Supplementary Table  S1. Material was harvested by cutting leaves and quenching them immediately in a bath of liquid N 2 under growth irradiance, avoiding shading.

Metabolite analyses
Plant material was ground to a fine powder by hand in a mortar precooled with liquid N 2 or in a cryo-robot (Stitt et al., 2007) and stored at -80 °C. Metabolites were extracted and quantified by LC-MS/MS (Arrivault et al., 2009). All samples were spiked with stable isotopelabelled internal standards for correction of ion suppression and other matrix effects (Arrivault et al., 2015). 3PGA gives a broad, poorly defined peak in LC-MS/MS and was therefore quantified enzymatically (Merlo et al., 1993).

Gas exchange
CO 2 assimilation was measured using the fourth fully expanded Zea mays leaf or 5-week-old Arabidopsis thaliana rosettes using an open-flow infrared gas exchange analyser system (LI-COR Inc., Lincoln, NE, USA; www.licor.com) equipped with an integrated fluorescence chamber head (LI-6400-40, 2 cm 2 leaf chamber for Z. mays; LI-6400-17 whole-plant Arabidopsis chamber for A. thaliana; LI-COR Inc.). CO 2 was kept at 400 µmol mol -1 , leaf temperature at 29 °C for Z. mays and at 20 °C for A. thaliana, and relative humidity at 65%. 13 CO 2 labelling with M. esculenta The fifth or sixth fully expanded leaf from the top of a 9-week-old plant was labelled ( Supplementary Fig. S2A), starting 2 h into the light period. The leaf was placed in the labelling chamber ( Supplementary Fig. S2B, C; see Arrivault et al., 2017). Gases were supplied from individual bottles and controlled by gas-flow controllers (Brooks instruments; www. brooksinstrument.com). The labelling chamber was initially supplied with 79% N 2 , 21% O 2 , and 420 ppm 12 CO 2 . After 1 min, 12 CO 2 was replaced by 13 CO 2 . Samples were collected after 10, 20, 40, or 60 s, or 2, 5, 10, 30, or 60 min, in random order. Gas flow was 10 l min −1 for pulses of up to 1 min, and 5 l min −1 for longer pulses. Unlabelled samples (t=0) were collected after 1 min in unlabelled gas mixture. The chamber was maintained at growth cabinet temperature (28 °C) by circulating water from a water bath. Gases were passed through a humidifier in the water bath after mixing and before entering the measuring chamber. Light intensity at the leaf surface was kept as in the growth cabinet (250 µmol m −2 s −1 ) by supplying additional light (FL-460 Lighting Unit, Walz, Effeltrich, Germany). Material was quenched by dropping a copper rod, pre-cooled in liquid N 2 , down a hollow tube incorporated in the chamber lid, thereby freeze-clamping a 1.9 cm diameter (~40 mg FW) leaf disc ( Supplementary Fig. S2C, D). 13 CO 2 -labelled samples were analysed by LC-MS/MS and GC-MS, and isotopomer distribution (%) and enrichment (%) were calculated as in Arrivault et al. (2017).

Statistical analyses
Statistical analysis was performed in R Studio Version 0.99.896 (www. rstudio.com) with R version 3.3.0 (https://cran.r-project.org/) using either Student's t-test (R default package stats) or an ANOVA (Sums of Squares Type II) followed by the Tukey's Honest Significant Differences (HSD) post-test (R package agricolae). Details are provided in the figure legends.

Metabolite levels at growth irradiance
We profiled CBC metabolites in four C 4 species from the NADP-malic enzyme subtype including two monocots (Zea mays and Setaria viridis) and two eudicots (Flaveria bidentis and F. trinervia), and five C 3 species including two monocots (Oryza sativa, Triticium aestivum) and three eudicots (Arabidopsis thaliana, Nicotiana tabacum, and Manihot esculenta). Each species was grown with non-saturating irradiance (range of 60-133% of that required for half-maximal rates of photosynthesis) and appropriate temperature for rapid, healthy growth, and harvested under growth irradiance at least 2 h after the beginning of the light period (for details, see Supplementary Table S1). CBC intermediates and 2PG levels were determined by LC-MS/MS, using isotope-labelled internal standards to obtain reliable quantification, or enzymatically (3PGA). The signals for ribulose-5-phosphate (Ru5P) and xylulose-5-phosphate (Xu5P) overlapped, so they were combined ('Ru5P+Xu5P'). Otherwise, we were able to quantify all CBC intermediates except 1,3-bisphosphoglycerate, glyceraldehyde 3-phosphate, and erythrose 4-phosphate. Metabolites were initially normalized on FW.
CBC metabolite levels varied greatly between species ( Fig.  1; Supplementary Dataset S1). This involved differences in the absolute and the relative levels of metabolites. Some of the observed changes were expected, for example the low levels of 2PG in C 4 compared with C 3 species, reflecting the lower rate of photorespiration in the C 4 plants (note, 2PG amounts are multiplied by 10 for better visualization in Fig. 1). RuBP levels were lower in C 4 compared with C 3 species, probably reflecting lower abundance of RuBisCO in C 4 plants. However, other interspecies differences were unexpected, in particular the rather diverse profiles in the five C 3 species. Features that varied between the C 3 species included the absolute levels of individual metabolites such as 3PGA, triose-P, Ru5P+Xu5P, the level of RuBP compared with metabolites involved in RuBP regeneration, and the relative levels of metabolite pairs, for example FBP and fructose 6-phosphate (F6P) or sedoheptulose 1,7-bisphosphate (SBP) and sedoheptulose 7-phosphate (S7P).

Metabolite levels in Z. mays and A. thaliana at different irradiances
One potential complication of a cross-species comparison is that each species has a different light saturation response, making it difficult to standardize growth and harvest conditions across species. We grew and harvested all species at moderate and limiting irradiance, using lower irradiance for species whose photosynthesis saturates at lower light intensities (Supplementary Table S1). In addition, for Z. mays and A. thaliana, we asked whether short-term changes in irradiance lead to major changes in the metabolite profile, using an additional lower irradiance for Z. mays ( Fig. 2A, covering the range from 40% to 133% of that required for half-maximal rates of photosynthesis), and a lower and a higher near-saturating irradiance for A. thaliana (Fig. 2B, covering the range from 67% to 233% of that required for half-maximal rates of photosynthesis). The metabolite profiles were not greatly altered for either species (Fig. 2C), except that higher irradiance tended to lead to a general increase in metabolite levels. Metabolite levels in a given species were strongly correlated irrespective of irradiance (r>0.98), whereas metabolite levels were poorly correlated between species (Fig. 2D).

Participation of pools in photosynthesis
Our approach assumes that the investigated metabolites are predominantly involved in the CBC. If they are also involved in another pathway, the total content will not provide reliable information about the size of the CBC pool. Published 13 C labelling kinetics validate this assumption for N. tabacum, A. thaliana, and Z. mays (Hasunuma et al., 2010;Szecowka et al., 2013;Arrivault et al., 2017); after pulsing with 13 CO 2 , all of the CBC metabolites showed a rapid rise in 13 C enrichment to reach a final value of ≥80%. One exception was SBP in maize, where 13 C enrichment plateaued at ~14%. We performed analogous 13 CO 2 labelling experiments for M. esculenta which, like Z. mays, is a subtropical species adapted to highlight conditions. We also chose M. esculenta because it has been suggested to be a C 4 or C 3 -C 4 intermediate species El-Sharkawy and Cock, 1987). A subsequent study showed that M. esculenta performs C 3 photosynthesis (Edwards et al., 1990; see also De Souza et al., 2017;De Souza and Long, 2018). Time-resolved 13 CO 2 labelling would provide a further test that M. esculenta is a C 3 species In M. esculenta, CBC intermediates rose rapidly to high (>75%) 13 C enrichment (Supplementary Fig. S3A; Supplementary Dataset S2) except for SBP where enrichment plateaued at ~40% and about half of the SBP remained in the unlabelled form after 60 min ( Supplementary Fig. S3B). Otherwise, the labelling time series in M. esculenta resembled published time series for the C 3 plants A. thaliana (Szecowka et al., 2013) and N. tabacum (Hasunuma et al., 2010). In particular, labelling of 4-C acids was very slow ( Supplementary Fig. S3C).

Chlorophyll and protein
Leaf composition varies between species (see the Introduction). This could contribute to interspecific differences in absolute . CO 2 assimilation rate in Z. mays and A. thaliana, and CBC metabolite profiles in Z. mays and A. thaliana at different short-term irradiances. Zea mays and A. thaliana were grown at 550 µmol m −2 s −1 and 120 µmol m −2 s −1 irradiance, respectively. CO 2 assimilation rate in (A) Z. mays (n=10) and (B) A. thaliana (n=9). The results are shown as mean (µmol CO 2 m −2 s −1 and µmol CO 2 g FW −1 h −1 , for Z. mays and A. thaliana, respectively) ±SD. Arrows indicate the irradiances at which leaves were sampled for metabolite analysis. (C) Zea mays was harvested at growth irradiance (Zm, medium irradiance) or after being subjected for 4 h to 160 µmol m −2 s −1 (ZmL, low irradiance) from the beginning of the light period. Arabidopsis thaliana was harvested at growth irradiance (At, medium irradiance) or subjected for 15 min to 80 µmol m −2 s −1 or 280 µmol m −2 s −1 (AtL, low and AtH, high irradiances, respectively). Quenching of metabolism and harvest of leaf tissue were performed at least 4 h after the beginning of the light period. 2PG amounts are multiplied by 10 for better visibility. Asterisks indicate graphs already presented in Fig. 1. The results are shown as mean (nmol g FW −1 ) ±SD. (D) Correlation analysis. The metabolite data shown in (B) and (C) were used to perform Pearson's correlation analysis between data sets from the same species at different irradiances, and correlations between different species. Before performing the correlation analysis, each data set was normalized by calculating the amount of carbon in a given metabolite, and dividing it by the total carbon in all metabolites in that data set. This was done to avoid secondary correlation due to any interspecies differences in leaf composition. The results are given as r and the higher correlations are indicated in bold. All correlations were positive. The original data are presented in Supplementary Dataset S1. metabolite levels; in particular, differences in leaf composition could lead to systematically higher or lower levels of all metabolites. We therefore determined total chlorophyll and protein contents in the leaf material used for metabolite analyses. Total chlorophyll content (Fig. 3A) was similar on a FW basis in all species except for O. sativa and M. esculenta, which had considerably higher values. Protein content on a FW basis (Fig. 3B) was similar in all species except for lower values in N. tabacum, and higher values in O. sativa and, especially, M. esculenta. These results partly explain why CBC metabolite levels on a FW basis tended to be low in N. tabacum and high in O. sativa and M. esculenta (Fig. 1).

Principal component analysis
We performed principal component (PC) analyses to provide an integrated overview of the CBC metabolite profiles in the nine species. PC analysis gives information about which samples (here, different species) are closely related or separated, and which variables (here, metabolites) contribute to this relationship. The analysis was performed with z-scored data (i.e. normalizing the individual values of a given variable on the mean value for that variable) to ensure that each metabolite adopted an equally important role in the analysis, independent of its absolute abundance. Each individual sample was included separately in the analysis to provide an overview of the quality of within-species replication. We included the low light maize and the low and high light Arabidopsis samples to further test the impact of prevailing irradiance. In the analyses shown in Fig. 4, we omitted 2PG to focus solely on the CBC and exclude effects due to lower photorespiration in C 4 plants. We also omitted SBP because of the labelling data (Supplementary Fig. S3; Arrivault et al., 2017) indicating that in some species part of the SBP pool is not involved in the CBC. For comparison, analyses including 2PG and SBP are provided in Supplementary Figs S4-S7.
As previously mentioned, some cross-species variation in metabolite levels may be driven by changes in leaf composition. We therefore performed PC analyses on data sets in which the metabolites were normalized on FW ( Fig. 4A; Supplementary  Fig. S4), total chlorophyll content ( Fig. 4B; Supplementary  Fig. S5), or protein content ( Fig. 4C; Supplementary Fig. S6). We also performed PC analysis on a dimensionless data set in which, for a given species, the amount of C in a given metabolite was divided by the total amount of C in all CBC intermediates plus 2PG ( Fig. 4D; Supplementary Fig. S7). In total, we performed 16 PC analyses with different metabolite data sets and normalizations. In interpreting the plots, we focused on features that were seen in all or the vast majority of these analyses.
In analyses with the FW-, chlorophyll-, and protein-normalized data sets and the dimensionless data set, PC1 accounted for 44-46, 33-35, 34-36, and 39-40%, respectively, of the total variance, while PC2 accounted for 17-20, 20-23, 21-22, and 19-22%, respectively ( Fig. 4A-C; Supplementary Figs S4-S6). In all cases, replicates for a given species grouped together, showing that within-species variance was smaller than interspecies differences. This included samples harvested at low and ambient light intensities for Z. mays and for A. thaliana. The A. thaliana samples collected at high light grouped separately from the other A. thaliana samples, but well removed from the other species in PC analyses with the FW-, chlorophyll-, and protein-normalized data sets. In PC analyses with the dimensionless data set, A. thaliana samples from all three light intensities grouped together ( Fig. 4D; Supplementary Fig. S7), showing that increasing light intensity led mainly to a general increase in metabolite levels rather than to changes in their relative levels.
Inspection of the species distribution in the PC plots leads to three main conclusions. First, the PC analyses almost always separated C 4 species from C 3 species; this holds irrespective of how the metabolite data are normalized, and whether 2PG and SBP were excluded (Fig. 4) or included ( Supplementary  Figs S4-S7). Manihot esculenta showed a slight overlap with the Flaveria spp. in the analyses using metabolites minus SBP and 2PG, when the data set was normalized on protein (Fig. 4C), but was fully separated from all of the C 4 species in the 15 other PC analyses. A. thaliana in low light showed a slight overlap with Z. mays or S. viridis in two (all metabolites normalized on FW, metabolites minus SBP normalized on FW; Supplementary    The analyses were performed on the metabolite data set excluding 2PG and SBP (2PG was omitted to avoid systematic bias between C 3 and C 4 species due the differing rates of RuBP oxygenation, and SBP was omitted because in some species part of the pool may not be involved in the CBC; see text for details). Metabolite amounts were normalized on (A) FW, (B) total chlorophyll content, or (C) protein content, or (D) were transformed into a dimensionless data set. For dimensionless data set determination, in a given sample, the level of each metabolite was first transformed to C equivalent values by multiplying the amount (nmol g FW −1 ) by the number of C atoms in the metabolite. The C equivalent amounts of all CBC intermediates plus 2PG were then summed. In the last step, the C equivalent value of a given metabolite was divided by the summed C equivalent values. The transformed values and calculation steps are provided in Supplementary Dataset 1. This transformation generates a dimensionless data set (provided in Supplementary Dataset S1) in which each metabolite receives a value equal to its fractional contribution to all the C in CBC metabolites plus 2PG. As this data set is dimensionless, there is no systematic bias due to differences in leaf composition. The distribution of C 4 species (green) and C 3 species ( Supplementary Fig. S4 (amounts normalized on FW), Supplementary Fig. S5 (amounts normalized on total chlorophyll content), Supplementary Fig. S6 (amounts normalized on protein content), and Supplementary Fig. S7 (dimensionless). The original data are presented in Supplementary Dataset S1. and M. esculenta separated strongly from other C 3 species in PC1 ( Fig. 4A; Supplementary Fig. S4). This was less marked in the PC analysis based on chlorophyll-or protein-normalized data (Fig. 4B, C; Supplementary Figs S5, S6), indicating that the strong separation in the analysis with FW-normalized data is partly driven by secondary effects due to leaf composition. Similarly, N. tabacum was less strongly separated from the other four C 3 species in the PC analysis with protein-normalized data than with FW-or chlorophyll-normalized data. Despite these small shifts in the relationships, the five C 3 species still separated from each other in the PC analyses with the chlorophyll-and protein-normalized data sets, as well as with the dimensionless data set ( Fig. 4D; Supplementary Fig. S7). The metabolite loadings ( Fig. 4; Supplementary Figs S4-S7) reveal that the separation of C 4 from C 3 species was driven not only by lower levels of RuBP and (when included) 2PG, but also by other CBC metabolites. 3PGA and triose-P contributed to the separation of the C 4 species Z. mays and S. viridis from F. trinervia and F. bidentis (see the Discussion). Almost every metabolite contributed to the separation between the five C 3 species, with large contributions from RuBP, FBP, F6P, S7P, ribose 5-phosphate (R5P), triose-P, and 3PGA.
We repeated the PC analysis on a data set including only C 3 species and with metabolites normalized on total chlorophyll content or protein content, and with a dimensionless data set ( Fig. 5; Supplementary Figs S8-S10). Replicate samples from a given species grouped closely together. A. thaliana, N. tabacum, and M. esculenta were clearly separated from T. aestivum and O. sativa, which were only weakly separated. The high irradiance A. thaliana samples grouped separately from the low and medium light A. thaliana samples, but in the same tangent, and were clearly separated from the other four C 3 species. Metabolite loadings revealed strong contributions from 3PGA, triose-P, RuBP, FBP, F6P, and S7P to the separation.

Coefficient of variance
We calculated the coefficient of variance (CV) to determine which metabolites showed the greatest interspecies variance for all nine species together, for the four C 4 species, and for the five C 3 species (Fig. 6). To avoid influence due to leaf composition, this analysis was performed on the dimensionless data set. Across all species (Fig. 6A), the highest CV was for 2PG, followed by FBP, RuBP, triose-P, SBP, and R5P. When only C 4 species are considered (Fig. 6B), the highest CV was for SBP, followed by 2PG, RuBP, R5P, FBP, and triose-P. When only C 3 species are considered (Fig. 6C), the highest CV was for FBP, followed by Ru5P+Xu5P, 2PG, R5P, and 3PGA.

Correlation analysis
When metabolite profiles are compared across different genotypes, they typically generate a correlation network (Meyer et al., 2007;Sulpice et al., 2009Sulpice et al., , 2013Zhang et al., 2015;Wu et al., 2016). This reflects features of the underlying metabolic pathways that are maintained across genotypes and generate conserved relationships between metabolites. Our data set allowed us to apply this approach to interspecies variation in the CBC. The analyses were performed on the metabolite data set excluding SBP. Metabolite data were normalized on (A) total chlorophyll content and (B) total protein, and (C) using a dimensionless data set (see legend of Fig. 4). The distribution of C 3 species is shown on PC1 and PC2 (O. sativa, Os; T. aestivum, Ta; A. thaliana, AtL, At, and AtH; N. tabacum, Nt; M. esculenta, Me). The loadings of CBC intermediates in PC1 and PC2 are shown in red. PC analyses with the full metabolite data set and with all metabolites except either 2PG or SBP are shown in Supplementary Fig. S8 (amounts normalized on total chlorophyll content), Supplementary Fig. S9 (amounts normalized on total protein), and Supplementary Fig. S10 (dimensionless). The original data are presented in Supplementary Dataset S1.
We performed pairwise PC analysis and clustering on CBC metabolites (Fig. 7) using the dimensionless data set to avoid bias from changes in leaf composition. We also searched for relationships between 2PG and the CBC metabolites. Metabolite pairs that are linked by irreversible reactions (Bassham and Krause, 1969;Mettler et al., 2014) are indicated by black boxes in the figure. Correlation analysis and clustering were performed for all nine species (Fig. 7A), for the four C 4 species (Fig. 7B), and for the five C 3 species (Fig. 7C). To aid visual comparison across the species sets, correlation coefficients are also shown in Supplementary Fig. S11 with the metabolites in a fixed order corresponding to CBC topology.
The CBC correlation network for all nine species (Fig. 7A) contained six positive correlations (e.g. F6P versus S7P; all pairwise comparisons between RuBP, FBP, and SBP), many non-significant relationships [e.g. FBP versus F6P; SBP versus S7P; RuBP versus R5P and Ru5P+Xu5P (here collectively called pentose-P)], and 13 negative correlations (e.g. 3PGA or triose-P versus most other CBC metabolites). In some cases, the correlations were driven by differences between C 4 and C 3 species; for example, the positive correlation between 2PG and RuBP is driven by the lower levels of both metabolites in C 4 compared with C 3 species (see Fig. 1). However, in many cases, the correlations were also seen within the subset of C 4 and within the subset of C 3 species (see Supplementary Fig.  S11 and below).
The correlation network for CBC metabolites in C 4 species (Fig. 7B) contained nine positive (e.g. FBP versus SBP; FBP versus RuBP; and triose-P versus FBP, SBP, and RuBP) and 13 negative (e.g. 3PGA versus triose-P, FBP, and SBP; triose-P versus S7P and pentose-P; RuBP versus F6P, S7P, and pentose-P; and SBP versus S7P) relationships. There was no significant relationship between FBP and F6P. The correlation network for CBC metabolites in C 3 species (Fig. 7C) contained six positive (e.g. all pairwise comparisons between RuBP, FBP, and SBP; and F6P versus S7P) and 11 negative (e.g. 3PGA versus triose-P, FBP, SBP, and RuBP; RuBP versus F6P and S7P; and SBP versus S7P) relationships. There was no significant relationship between FBP and F6P, or between RuBP and pentose-P. 2PG correlated positively with S7P and negatively with RuBP in C 4 and C 3 species, respectively, and positively with 3PGA and negatively with triose-P and FBP in C 3 species.
The correlation networks can be interpreted by relating them to CBC topology ( Fig. 7D; see also Supplementary Fig.  S11). Figure 7D focuses on correlations seen within the subset of C 4 species and within the subset of C 3 species. Triose-Ps are used to synthesize FBP and SBP in reversible reactions catalysed by aldolase. This may explain the positive correlations between triose-P and FBP or SBP (except for SBP in C 3 plants). FBPase and SBPase catalyse irreversible reactions. The non-significant or negative correlations between FBP and F6P and between SBP and S7P point to interspecies variance in the regulation of FBPase and SBPase. This may also explain the absence of a positive correlation between triose-P and pentose-P that otherwise might have been expected because pentose-Ps are formed from triose-P and F6P or S7P in reversible reactions catalysed by transketolase (TK). The negative relationship between pentose-P and RuBP points to interspecies variation in the regulation of PRK. Further, the positive correlations of FBP and SBP with RuBP (see Supplementary Fig.  S11) indicate that FBPase and SBPase activity vary reciprocally to PRK activity and/or co-ordinately with binding or use of RuBP by RuBisCO.

Discussion
The CBC is an ancient pathway that has been under selective pressure due to the long-term increase of the O 2 :CO 2 ratio in the atmosphere and particularly over the last 30 million years  Fig. 6. Coefficient of variance (CV) of CBC metabolites between species. The CV (SD/mean×100) is a standardized quantity describing the dispersion of a population distribution (Simpson and Roe, 1939). The analysis was performed on a dimensionless data set that was generated as described for Fig. 4D. The transformed data (presented in Supplementary Dataset S1) were used to calculate the bootstrapped CV for each metabolite (30 bootstrap iterations). The 95% confidence interval was estimated using the basic bootstrap method. Statistically significant differences between metabolites are indicated by letters (ANOVA on the bootstrap results followed by the Tukey's HSD post-test). (A) All species, (B) only C 4 species, and (C) only C 3 species. R u B P T r io s e -P F B P F 6 P S B P S 7 P R 5 P R u 5 P + X u 5 P 2 P G 3 P G A R5P Triose-P SBP F6P S7P 3PGA Ru5P+Xu5P FBP RuBP 2PG R u B P T r io s e -P F B P F 6 P S B P S 7 P R 5 P R u 5 P + X u 5 P 2 P G 3 P G A C species The metabolites taken for each enzyme to depict the substrate-product relationship as a heat map (scale panels B-C) are:  Fig. 7. Correlation between levels of CBC metabolites. The correlations were performed with all individual samples from a dimensionless data set, generated as described for Fig. 4D. The transformed data were used to calculate the Pearson's correlation matrix on every pair of metabolites. Correlation values are given in the figure panels and indicated by a heat map. The adjacent dendrograms show clusters defined using the complete linkage method (Sørensen, 1948). Non-significant correlations (P≥0.05; two-sided Student's t-test) are set as zero. Metabolite pairs that are linked by irreversible reactions are indicated by a black box. (A) All species, (B) only C 4 species, and (C) only C 3 species. An alternative display is provided in Supplementary Fig. S11, with the same fixed order of metabolites in each panel, corresponding to the reaction sequence in the CBC. The same heat map scale is used for (A-C). (D) Schematic representation of interspecies variance in the ratio of substrate abundance:product abundance for different CBC enzymes. Enzymes that catalyse irreversible reactions are highlighted in bold. For each enzyme reaction, the substrate and product that were compared are indicated in the list below the display. This display is schematic because some metabolites were not measured (erythrose 4-phosphate, E4P; and glyceraldehyde 3-phosphate, GAP) or were not separated (Ru5P and Xu5P). For reactions using GAP, it is assumed that GAP and dihydroxyacetone phosphate (DHAP) are in equilibrium. For transketolase (TK), two reactions were separated (termed TK a and TK b ). For TK a , the reactant E4P was missing, and only the relationships between triose-P and F6P and Ru5P+Xu5P are shown. For Tk b , the plot focuses on the relationship between S7P and R5P or Ru5P+Xu5P. The display shows the alternative pairs of metabolites compared, with the upper and lower symbols in the display corresponding to the upper and lower pair in the list. A similar display mode is used for the carboxylation and oxygenation reactions of Rubisco. The correlation coefficients are taken from (B) and (C), using the same heat map scale. Results are shown separated for correlations between the four C 4 species (squares) and the five C 3 species (circles). The analysis is not shown for the combined C 4 plus C 3 species set because, in this case, some relationships are driven by differences between C 4 and C 3 species. Additional abbreviations: fructose 1,6-bisphosphate aldolase (FBP ald), phosphoglycerate kinase (PGK), ribose 5-phosphate isomerase (R5P isom), sedoheptulose 1,7-bisphosphate aldolase (SBP ald). due to falling CO 2 concentrations, which led to independent evolution of a CCM in >100 terrestrial plant lineages. However, the vast majority of terrestrial species did not evolve a CCM, probably because they were unable to follow the multistep evolutionary trajectory that was required to acquire this complex trait (Sage et al., 2012;Christin and Osborne, 2013;Heckmann et al., 2013). Present-day C 3 plants nevertheless will have been subject to similar selective pressures to those that drove the evolution of C 4 or CAM photosynthesis. Indeed, in the absence of a CCM, the selective pressures on the CBC may have been even greater than in plants that did evolve a CCM. In addition to low CO 2 , it is likely that environmental factors such as irradiance, temperature, and nutrient and water availability exerted more or less selective pressure, depending on the local environment, and leading to different evolutionary trajectories in different populations. While it is well documented that there is large variation in photosynthetic rate between terrestrial species (Evans, 1989;Wullschleger, 1993;Lawson et al., 2012), previous studies of the underlying causes have focused on leaf morphology and composition (Field and Mooney, 1986;Evans, 1989;Hikosaka, 2010;Poorter et al., 2015;Díaz et al., 2016), stomatal conductance (Lawson et al., 2012), and the kinetic characteristics of RuBisCO (Yeoh et al., 1980;Jordan and Ogren, 1981;Badger et al., 1998;Tcherkez et al., 2006;Galmés et al., 2014b;Prins et al., 2016;Sharwood et al., 2016a, b). Little is known about whether the CBC operates in a highly conserved manner or in different modes in different C 3 species.
We have used metabolite profiling as an unbiased strategy to search for interspecific variance in CBC operation. The underlying assumption is that changes in the balance between different enzymatic steps will lead to changes in the relative levels of pathway intermediates. This approach is top down, in the sense that it does not make assumptions about whether the observed variance is due to changes in gene expression and protein abundance, enzyme kinetics, or regulatory networks that act on the enzymes. We applied it to search for differences in CBC operation between C 4 and C 3 plants, and within C 3 species. As our aim was to compare CBC operation across species, we focused exclusively on the metabolites that are involved in the CBC plus 2PG, the immediate product of the RuBisCO oxygenation reaction. We excluded metabolites involved further downstream in photorespiration and metabolites involved in the CO 2 -concentrating shuttle in C 4 plants, which have nonphotosynthetic functions in C 3 plants.
Our interspecies comparison required important control experiments and cross-checks during data analysis. First, plant species differ in their photosynthetic rate and its dependence on light, temperature, and the availability of water, nutrients, and CO 2 (see the Introduction). We grew and harvested plants in a light regime that was limiting for that species, rather than using identical conditions for all species. In these conditions, RuBP regeneration is likely to be limiting, and effects of light stress are avoided. Importantly, we showed for one C 4 species (Z. mays) and one C 3 species (A. thaliana) that although increased harvest irradiance led to higher levels of metabolites, it did not strongly alter their relative levels (Fig. 2). Secondly, it is important that the CBC pool accounts for most or all of the total content of a given metabolite. Analysis of published data for two C 3 (N. tabacum and A. thaliana), one C 4 (Z. mays) species (Hasunuma et al., 2010;Szecowka et al., 2013;Arrivault et al., 2017), and a new data set for the C 3 species M. esculenta ( Supplementary Fig. S3) showed that CBC intermediates exhibit a rapid rise in 13 C enrichment to a high level after supplying 13 CO 2 . This provides evidence that most of the total pool is indeed involved in the CBC. This conclusion is supported by published subcellular fractionation studies, in which most CBC intermediates are exclusively or largely confined to the plastid (Gerhardt et al., 1987;Szecowka et al., 2013). The only exception was SBP, which was only partially labelled in Z. mays and M. esculenta. We do not know whether there is a separate pool of SBP that is not involved in CO 2 fixation, or if these plant species contain an unknown metabolite with an identical chromatographic behaviour, mass, and fragmentation pattern to SBP. In our interpretation of the metabolite profiles, we took care that our conclusions did not depend on inclusion of SBP. A third set of controls addressed the issue that leaf composition varies between species, with the result that absolute values for metabolite content will depend on the unit in which they are given. We analysed metabolite data normalized on FW, chlorophyll, or protein content, and also used a dimensionless data set in which metabolite levels were expressed relative to each other. Our interpretation focused on results that were independent of how the data were normalized. Importantly, inclusion of the dimensionless data set eliminated secondary correlations due to differences in leaf composition, and placed the emphasis on relative rather than absolute levels of metabolites. It minimizes contributions from differing light regimes, which had less effect on relative than on absolute metabolite levels (see above).
We included four C 4 species in our panel to test if CBC profiles could distinguish between species in which it is known that the CBC operates in a different context from that of C 3 plants. The CBC operates at a much higher intercellular CO 2 concentration in C 4 than in C 3 plants, and RuBisCO has a higher affinity for CO 2 , and an increased catalytic rate in C 4 compared with C 3 species (see the Introduction). PC analysis confirmed that CBC metabolite profiles allow C 4 and C 3 species to be distinguished ( Fig. 4; Supplementary Figs S4-S7).
As expected, C 4 species had lower 2PG and RuBP than C 3 species (Fig. 1). However, the separation in the PC analysis was also seen when 2PG was excluded, and was driven by several other CBC intermediates, pointing to broader changes in CBC operation between C 4 and C 3 species.
The four C 4 species belong to the NADP-malic enzyme C 4 subtype. Interestingly, PC analysis separated Z. mays and S. viridis from the two Flaveria spp. Whilst this might reflect a difference between monocots and eudicots, the PC vectors indicated that this separation reflected higher levels of 3PGA and, in particular, triose-P in Z. mays and S. viridis ( Fig. 4; Supplementary Figs S4-S7; see also Fig. 1). Most NADPmalic enzyme C 4 subtypes, including Z. mays, have dimorphous chloroplasts with little or no PSII activity in the bundle sheath cells (Munekage, 2016). They operate an intercellular shuttle in which 3PGA moves from the bundle sheath to the mesophyll cells and is reduced to triose-P, which returns to the bundle sheath. Intercellular movement is thought to occur by diffusion (Hatch and Osmond, 1976), driven by concentration gradients that require the build-up of large pools of 3PGA and triose-P in the bundle sheath and mesophyll cells, respectively (Leegood, 1985;Stitt and Heldt, 1985;Arrivault et al., 2017). Flaveria bidentis and F. trinervia can have PSII activity in the bundle sheath chloroplasts, although to a varying extent depending on conditions (Laetsch and Price, 1969;Höfer et al., 1992;Meister et al., 1996;Nakamura et al., 2013). Their separation in the PC analysis from Z. mays and S. viridis might reflect decreased reliance on this intercellular shuttle.
Our panel included five C 3 species, two monocots (O. sativa and T. aestivum) and three eudicots (A. thaliana, N. tabacum, and M. esculenta), with the individual species representing different phylogenetic lineages (Supplementary Fig. S1) and originating in differing climatic zones. The three C 3 eudicot species represent two of the major lineages within the eudicots, namely the asterids (N. tabacum) and rosids (A. thaliana and M. esculenta), that contain 41% and 24% of all angiosperms, respectively. There was considerable interspecies variation in CBC metabolite profiles. This was evident from visual inspection of the metabolite levels ( Fig. 1) and was confirmed by PC (Figs 4, 5; Supplementary Figs S4-S10) and variance (Fig. 6) analyses.
When metabolites were expressed on a FW basis, some of the variation was due to differences in leaf composition, with a strong trend to higher absolute levels in O. sativa and M. esculenta, reflecting their high chlorophyll and protein content. The high protein content in O. sativa may be linked to changes in leaf anatomy that enhance mesophyll transfer conductance, including small deeply lobed cells and densely arranged chloroplasts and stromules at the cell surface (Sage and Sage, 2009;Busch et al., 2013). This high mesophyll transfer conductance may prevent internal CO 2 from being drawn down by the high CBC activity that results from the high protein and metabolite content per unit FW in O. sativa. The high protein content in M. esculenta resembles the findings of previous reports (Awoyinka et al., 1995;Nassar and Marques, 2006), and could explain the high rates of photosynthesis in this species.
However, the five C 3 species still showed differing CBC metabolite profiles when metabolites were expressed on a chlorophyll or protein basis, or when the analyses were performed with a dimensionless data set. Variation was driven by many metabolites including RuBP, 3PGA, triose-P, FBP, F6P, S7P, and Ru5P+Xu5P. This variation points to different operating modes of the CBC in different C 3 species. There were also differences in 2PG content; this might be related to the rate of RuBisCO oxygenation or removal of 2PG by 2-phosphoglycolate phosphatase.
Cross-species correlation analysis ( Fig. 7; Supplementary  Fig. S11) revealed that in both C 4 and C 3 species, the interspecies variance often included parallel changes of FBP, SBP, and RuBP, and unrelated or even reciprocal changes of these metabolites to F6P, S7P, and pentose-P. This is consistent with interspecies variation in the balance between FBPase, SBPase, PRK, and RuBisCO activity. It could reflect differences in the abundance or the regulation of these enzymes, both within C 3 species and within C 4 species, and between C 3 and C 4 species. Little is known about the expression, characteristics, and regulation of CBC enzymes in different species, with (see the Introduction) the exception of RuBisCO.
Our results do not reveal when and under what circumstances the variation in CBC function in C 3 species appeared. It is tempting to link it with the selection pressure that led to the appearance of C 4 and CAM photosynthesis, but it is likely to have started even earlier. Further, as pointed out by Zhu et al. (2007), it is possible that different C 3 species are following different trajectories during the increase in CO 2 levels in very recent evolutionary time. Our results also indicate that there is no strong connection between phylogeny and the diversity in CBC metabolite profiles in C 3 species. In the PC analyses (Figs 4, 5; Supplementary Figs S4-S10), the two monocot species are often closely related, but the three eudicot species are highly diverse, and a given eudicot is often more closely related to the monocot species than to the other eudicot species. Unlike changes in genome sequence, complex emergent phenotypes may not accrue in a linear manner, and phylogenetically distinct species may undergo convergent evolution whilst phylogenetically related species may undergo divergent evolution, depending on the selective pressure they experience. Better understanding of the relationship between diversity in CBC profile, phylogeny, and evolution will require studies both with more phylogenetically diverse species and with more dense sampling in short evolutionary space.
In conclusion, marked differences in CBC metabolite profiles between five C 3 species, including the major crop plants O. sativum, T. aestivum, and M. esculenta, and the important model plants A. thaliana and N. tabacum, reveal interspecies variation in the operating mode of the CBC in C 3 plants. This probably reflects independent evolution of CBC regulation in different plant lineages, in analogy to the independent evolution of a CCM in different plant lineages. These findings, together with emerging evidence for interspecies variation in the properties of specific CBC enzymes (see the Introduction) and the growing realization that efficient photosynthesis requires integrated operation of the CBC (Stitt et al., 2010;Raines, 2011;Simkin et al., 2017), highlight the need for a mechanistic understanding of CBC regulation in a wider range of species. This will be an important step towards improving C 3 photosynthesis and crop productivity.

Supplementary data
Supplementary data are available at JXB online. Fig. S1. Phylogenetic distribution based on APGIII of the tested plant species. Fig. S2. Experimental set-up for 13 CO 2 labelling of M. esculenta. Fig. S3. 13 C enrichment (%) of CBC metabolites, relative abundance (%) of SBP isotopomers, and 13 C enrichment (%) of malate, aspartate, pyruvate, and alanine in M. esculenta. Fig. S4. PC analyses of all species using metabolite data normalized on FW (supplementary analyses to Fig. 4A). Fig. S5. PC analyses on all species using metabolite data normalized on total chlorophyll content (supplementary analyses to Fig. 4B). Fig. S6. PC analyses on all species using metabolite data normalized on protein content (supplementary analyses to Fig.  4C). Fig. S7. PC analyses on all species using a dimensionless data set (supplementary analyses to Fig. 4D). Fig. S8. PC analyses on C 3 species only, using metabolite data normalized on total chlorophyll content (supplementary analyses to Fig. 5A). Fig. S9. PC analyses on C 3 species only, using metabolite data normalized on protein content (supplementary analyses to Fig. 5B). Fig. S10. PC analyses on C 3 species only, using a dimensionless data set (supplementary analyses to Fig. 5C). Fig. S11. Correlation between levels of CBC metabolites, with metabolites shown in a fixed order reflecting the reaction sequence in the CBC. Table S1. Growth conditions and photosynthetic rates. Dataset S1. Metabolite levels, total chlorophyll, and protein contents in different species.
Dataset S2. Labelling kinetics of CBC and other intermediates after exposing M. esculenta to 13 CO 2 (supplementary data to Supplementary Fig. S3).