A. thaliana wildtype Col-0 and the mutant hpr1-1 (SALK067724) were grown in hydroponic culture for 50 days in a growth chamber with 8 h/16 h light/dark regime (100 μmol m⁻² s⁻¹; 22°C/16°C). For the first 17 days plants were grown at ambient CO₂ (450 ± 20 ppm). Afterwards half of the plants were transferred to eCO₂ (1,000 ± 20 ppm). The hydroponic medium was as described in Brauner et al. (2014) with the difference that no ammonium was supplied and the nitrate concentration reduced to 2.175 mM. This assured sufficient, but not excess supply of N as identified in experiments with different levels of nitrate supply. For measurements of metabolites and enzyme activities, full rosettes of 50 day old plants were harvested every 2 h over a full diurnal cycle at fivefold replication under ambient as well as eCO2, resulting in a total of 120 samples. Photosynthesis measurements were conducted 1 week before harvesting as described by Küstner et al. (2019). Briefly, an infrared gas analysis system (Uras 3G; Hartmann and Braun AG, Frankfurt am Main, Germany) equipped with five custom made whole-rosette cuvettes and a sixth channel for measurement of CO₂ in surrounding atmosphere was operated at a flow rate of 40 l/h, and each of the channels was measured sequentially for 6 min over a full diurnal cycle at a rate of 0.1 Hz. From the measured time points, splines were generated to yield a complete course of photosynthetic activity. All measurements were conducted at growth light and the respective CO₂ concentration. Using the result of the photosynthesis measurements the photorespiratory activity was calculated according to the method developed by Sharkey (1988). For Γ* the values identified by Kraemer et al. (2021a) were used. As a first step, the ratio between oxigenation and carboxylation- Φ-was calculated according to equation (1).

Using this ratio the rate of oxygenation was calculated according to equation (2).

The photorespiratory activity used in the model was calculated as the oxygenation rate divided by 2.

Gly, Ser, α-KG, glucose, fructose and sucrose were measured by quantitative GC–MS/MS. Samples were extracted using 750 μl methanol with 25 nmol ribitol as internal standard. After 15 min at 70°C followed by shaking for 10 min at RT samples were centrifuged (5 min 17,000 g). The supernatant was transferred to a new vessel, and 400 μl of H₂O were added. After incubation for 10 min at 95°C samples were agitated for 10 min at RT. Following centrifugation (5 min, 17,000 g) the supernatants were pooled. Subsequently, 300 μl H₂O and 200 μl chloroform were added. After centrifugation (2 min, 17,000 g) the two phases were separated and the polar phase was dried in a speedvac and used for analysis. Dried samples were derivatized using 20 μl of methoxamine dissolved in pyridine (40 mg/ml) by incubation for 90 min at 30°C. Next, 80 μl N-methyl-N-(trimethylsilyl)trifluoracetamide (MSTFA) were added and the solution was incubated for 30 min at 50°C. Metabolites were measured by gas-chromatography coupled to mass-spectrometry (GC–MS/MS). For injection, 1 μl of the derivatized sample was used. The GC–MS/MS device was a GCMS-TQ8040 (Shimadzu, Kyoto, Japan) using helium as carrier gas at a flow of 1.12 ml/min. The stationary phase was a 30 m Optima 5MS-0.25 μm fused silica capillary column. Injection temperature was 230°C. The transfer line and ion source were set to 250°C and 200°C, respectively. The initial temperature of the column oven was 80°C and this was increased by 15°C/min until the final temperature of 330°C was reached and held for 6 min. After a solvent delay of 4.6 min, spectra of the MS device were recorded in the multiple reaction mode (MRM) with specific target-ions for each metabolite. External standards were used for quantification.

Starch, hexose-phosphates and the total amino-acid pool were determined photometrically and carbonic acids (fumaric acid, malic acid and citric acid) by HPLC as described by Küstner et al. (2019). Ammonium was quantified according to Vega-Mas et al. (2015). Glu and Gln were measured according to Graham and Aprison (1966) and Pérez-de la Mora et al. (1989).

Activity of nitrate reductase was measured as described by Scholl et al. (1974). Activity of hydroxy-pyruvate-reductase was determined according to Bauwe (2017). GS activity was measured according to Berteli et al. (1995) and Silveira et al. (2003). Briefly, protein was extracted into a 100 mM Tris–HCl (pH 7.6) buffer containing 2.5 mM dithiothreitol and 10 mM MgCl₂. The assay buffer contained 125 mM Tris–HCl (pH 7.6), 5 mM ATP, 80 mM MgSO4, 125 mM hydroxylamine-NaOH (pH 7) and 100 mM glutamate (pH 7.2). To an assay buffer volume of 80 μl, 120 μl of protein extract were added and incubated for different time-points (0, 20, 25 and 30 min). To determine the background, an assay buffer was used without hydroxylamine-NaOH (pH 7). The reaction was stopped using 60 μl of a solution consisting of 1.5 ml 10% w/v FeCl₃*6 H2O in 0.2 N HCl, 1.5 ml 24% w/v trichloroacetic acid and 1.5 ml 20% v/v HCl. Afterwards, the absorption was determined at 540 nm. For calibration curves L-glutamic acid γ-monohydroxamate was used.

All enzyme activities were determined at the beginning, middle and end of the light phase as well as in the middle of the night. Values for the remaining time-points were calculated by spline interpolation. For the calculation of reaction velocity, substrate concentration was calculated from the measured metabolite contents, assuming that 1 g of plant biomass had a volume of 0.77 ml.

Data evaluation, normalization, visualization and statistics were performed in Microsoft Excel (RRID:SCR_016137) and the R software (R Project for Statistical Computing, RRID:SCR_001905). The experimental design was a randomized complete block design with CO₂ level as block and genotype as treatment. If not otherwise stated, two-way ANOVA for genotype and CO₂ treatment effects were used for statistical analysis. Parameter optimization was performed in a way that in silico time courses best matched the measured time courses. For optimization and simulations the R-package paropt was used (Kraemer et al., 2021b).

Net photosynthetic rates (PS) increased by about 50% in the wildtype, when CO₂ concentration was raised to 1,000 ppm. The wildtype had significantly higher PS than the mutant (p < 2e-16) under both conditions (Col-0 at ambient CO₂ = 85.4 ± 12.8 μmol/g * h, Col-0 at eCO₂ = 128.1 ± 44.9 μmol/g * h, hpr1-1 at ambient CO₂ = 76.8 ± 19.9 μmol/g * h, hpr1-1 at eCO₂ = 98.6 ± 35.8 μmol/g * h). An increase in net photosynthesis by about 50% at eCO₂ in the wildtype is in accordance with previous observations (Jauregui et al., 2016). However, a stronger reduction of photosynthesis in hpr1-1 was reported for higher light intensities by Timm et al. (2008).

Despite the lower PS rate and no obvious deviations in hexose-phosphate levels (Supplementary Image 1), the combined pool of malate and fumarate (MF) was larger in the hpr1-1 mutant than in wildtype (p = 4.1e-10). Citrate (Cit) was elevated in hpr1-1 only at ambient CO₂ (Supplementary Image 1; p = 2.47e-7) and showed a different response to rising CO₂ levels: while it increased in the wildtype (p = 7.1e-4, t-test for wildtype), it declined in the mutant (p = 5.08e-6, t-test for hpr1-1).

The PR intermediates Gly and Ser accumulated in the mutant under both conditions (Figures 1A,B). As expected, the contents were higher at ambient CO₂. For NH₄⁺, which is also produced during PR, significant genotype and treatment effects were observed with higher levels in hpr1-1 and a general decrease under eCO₂ (p = 5.67e-5 and p = 4.38e-6, respectively).

Similar to the NH₄⁺ content, Glu levels (Figure 2A) were elevated in the mutant as compared to wildtype (p = 7.71e-4) and significantly increased in mutant plants at ambient CO₂ as compared to eCO₂ (p = 2.03e-10). This effect could not be observed for Col-0. Neither genotype nor treatment effects were observed for Gln (Figure 2B). However, when considering the mutant alone, an elevated level at ambient CO₂ became obvious (p = 0.0163). The α-KG content (Figure 2C) was strongly increased for hpr1-1 at ambient CO_2, Considering that Glu is converted to α-KG when used as substrate for transamination of glyoxylate in the peroxisome, elevated α-KG may point to an increased Glu turnover for transamination of glyoxylate.

Furthermore, the amino acid (AA) pool (Figure 2D) was elevated in the mutant compared to wildtype (p < 2.0e-16). Considering only mutant data, an increase at eCO₂ was significant (p < 2.0e-16). Thus, no indications for N-starvation in the mutant were found.

Maximum activities were determined for three enzymes central to de novo N assimilation, amino acid metabolism and PR. For NR, the Col-0 wildtype showed the highest activity at eCO₂ (Supplementary Image 2A). This constituted a significant genotype effect (p = 0.0346) that was unexpected considering reports of decreased N assimilation at eCO₂ (Bloom et al., 2014). In addition, a time effect with highest activity at midday was observed independent of the condition (p = 0.0017). GS activity (Supplementary Image 2B) was clearly affected by the CO₂ level with higher activity at ambient CO₂ than at eCO₂ (p = 4.14e-13). Notably, hpr1-1 had even higher GS activity than Col-0 when grown at ambient CO₂ (p = 0.0021), indicating a higher potential for turnover in the GS/GOGAT cycle. Not surprisingly, a genotype effect was found for HPR activity (Supplementary Image 2C; p < 2e-16). In addition, a treatment effect was detected, which was dependent on the hpr1-1 data (p = 0.0461).

With the aim of mathematically simulating C/N interactions at eCO₂, a dynamic model based on ordinary differential equations (ODE) was constructed that covered carbon as well as nitrogen acquisition. The model structure is depicted in Figure 3. Based on measurements of photosynthesis and respiration as well as metabolite content at 8 time points over a diurnal cycle, kinetic parameters were identified that allowed simulation of metabolite dynamics with only the metabolite levels at the beginning of the light period (t = 0) given. The identified parameters (see Supplementary Table 1) describe turnover rates for C- and N-compounds. Besides concentrations of O₂ and CO₂, the plant biomass (BM), starch level, amount of assimilates exported from the leaves (EXP) and nitrate level are regarded as beyond the system boundaries, which means that, e.g., the rate of starch synthesis is included in simulations, but not the actual amount that has accumulated at a certain time point. In contrast, levels of Gly, Ser, HP, MF, Cit, α-KG, Glu, Gln, AA and NH₄⁺ are included in the simulations. For the carbon fluxes to biomass and sink organs (hp2BM/EXP; note: all fluxes in italics), to or among carboxylic acids (hp2MF, Cit2MF, MF2Cit, Cit2KG), as well as the C/N fluxes from serine to amino acids (Ser2AA) and amino acids to biomass and export (AA2BM/EXP), mass balance kinetics were applied. The remaining reactions are represented as Michaelis–Menten kinetics. Kinetic equations for enzyme reactions are given in file Supplementary Data Sheet 1. The full ODE system is given in file Supplementary Data Sheet 2.

Initial steps of the PR pathway are represented by a reaction yielding Gly, which means that all reactions from oxygenation of ribulose-bisphosphate to glyoxylate formation were not resolved, because absolute quantification of the respective intermediates is not reliable. Input into the pathway was calculated on the basis of PS according to Sharkey (1988). Gly is converted to Ser, which was considered precursor for the HPR reaction that feeds back into the pool of sugar-phosphates. Alternatively, Ser can flow into the pool of the other AA, representing the reaction of Ser-pyruvate aminotransferase.

All short-lived intermediates of the Calvin-Benson-Cycle, triose-phosphates and hexose-phosphates were represented by the HP pool, which is quantitatively dominated by glucose-6-phosphate and fructose-6-phosphate. The latter were quantified enzymatically (see Material and Methods). Photosynthesis, measured as CO₂ uptake per time, constituted the carbon input into this pool. Measured values were corrected for mitochondrial respiration and PR input. Mitochondrial respiration was measured during the dark phase and set constant over the entire diurnal cycle as earlier suggested by Küstner et al. (2019) and Kraemer et al. (2021a). The HP pool was set as the precursor for cell wall synthesis and the supply of carbon to sinks. It should be mentioned that the transport metabolite, sucrose, attains much higher levels than measured for HP, which is substrate for cell wall biosynthesis. While this would affect the ratio of BM formation and transport rates, when kinetic parameters are constrained, this was not the case in the current model, which left wide ranges for the BM/EXP parameter, because assimilate export was not in focus. In the model, assimilate export and biomass formation were represented by the combined flux hp2BM/EXP. In addition, HP is in exchange with starch that serves as carbon source during the night. Starch production during the day and degradation at night was assumed to be linear as outlined by Stitt and Zeeman (2012). As demonstrated by Gauthier et al. (2010) and Tcherkez et al. (2009), Cit is not, or only to a very low extent, produced during the day, but accumulates during the night and is used for α-KG production during the next day (Cheung et al., 2014). Therefore, the substrate for accumulation of MF during the light phase was considered to be HP, thus referring to anaplerotic reactions. However, the possibility of a flux from Cit to MF was not excluded. MF is used for mitochondrial respiration.

During the reaction from Gly to Ser NH₄⁺ is produced and, together with Glu, can be used by GS to produce Gln (Supplementary Image 2). Gln can transaminate α-KG, catalyzed by GOGAT, to produce two molecules of Glu. Two reactions can yield α-KG that is required for AA production. The first reaction is from Cit (Cheung et al., 2014) representing the TCA cycle and cytosolic isocitrate dehydrogenase activity (Hodges, 2002). The second is from Glu and includes the reaction of Glu-glyoxylate-aminotransferase, which is part of PR, as well as mitochondrial Glu dehydrogenase and transamination reactions involving Glu. For Gly production, a source of N is required. This can derive from Ser, which, in the model, is included in the combined HPR flux. Alternatively, Glu can serve as N donor for Gly production. The PR flux into the system, the HPR flux, and the reaction from Glu to α-KG are all N₁-fluxes. Thus, the three fluxes can be balanced, and Glu to α-KG can be expressed as PR minus HPR.

Another source of NH₄⁺ is the reduction of nitrate by NR (Supplementary Image 2). Modeling this de novo N fixation is complicated by the fact that a large proportion of nitrate is stored in the vacuole and not accessible by NR. To account for this compartmentation, the content was adjusted based on the cellular proportion of cytosol, which is about 5% of total cell volume in mature leaves (Koffler et al., 2013). NR is the only flux providing de novo assimilated N. Thus, the step towards AA production was linked to NR. According to Gauthier et al. (2010) about 50% of Ser consist of newly assimilated N. Accordingly, we modeled the flux from Glu to AA as NR multiplied by the correction factor, correct, which was set in the interval [0.5, 1]. Because the PR intermediates Gly and Ser are the only other sinks for de novo assimilated N in our model, correct likewise determines the deposition of new N in these compounds.

Because PR flux, as defined by Sharkey (1988), is directly coupled to photosynthetic activity, the pathway would not operate in the dark, which would, in our model, exclude turnover of Gly and Ser after light-off. However, especially in the case of hpr1-1 at ambient CO₂, Ser turnover was substantially extended into the night. This was accounted for by adding the summand α to the formula for PR, which allows operation of pathway reactions described by Walton and Butt (1981) independent of photosynthetic activity.

The step from Gly to Ser is catalyzed by the two enzymes GDC and SHMT. For both reactions Michaelis–Menten kinetics were employed. However, turnover of SHMT cannot exceed that of GDC, and thus SHMT was limited to the GDC value. In contrast, it is known from the literature that the GDC flux may well be substantially higher than the flux of SHMT (Rebeille et al., 1994).

Because the model structure presented in Figure 3 does not allow the amino acids Gly and Ser to contribute to biomass via the AA2BM/EXP route, a separate pathway was established for these compounds. Based on proteome information for Arabidopsis (Berardini et al., 2015) a proportion of about 10% of the amino acids in protein should be Ser. However, Rubisco, which makes up about one third of total protein (Atkinson et al., 2017), has 14% Ser, and thus, the flux from Ser to BM was set to 14% of the flux from AA to BM/EXP.

According to the above described model structure several rounds of parameter optimization were conducted that yielded results for most states that were covered by the measured standard deviation with an averaged error between 0.05 and 0.1% per state and time point. However, Glu levels were consistently underestimated, particularly during the night as shown in Figure 4A. Choi et al. (1999) reported that the GS enzyme of Canavalina lineata is activated by a reduction at two cysteine residues, which are conserved among all known plastidial GS sequences. The authors demonstrated that reductants like dithiothreitol increase the activity of the plastidial isoform. Considering that dithiothreitol was added during protein extraction (see “Materials and Methods”), it is very likely that GS activity was overestimated, especially during the night. It is known that, following light-off, the redox milieu of the chloroplasts changes rapidly (Dietz and Hell, 2015). However, the extent of change in GS activity is unknown. Therefore, a numerical experiment was set up to mathematically assess GS activity during the night (see Figure 4B). Nocturnal GS activity was intentionally reduced in steps of 10%, and 10 simulations of model states were conducted for each step. Simulations were accepted, when results for all states lay within the measured standard deviation. As can be seen in Figure 4B, the cumulative error of simulations was minimized for nocturnal GS activity between 20 and 50% of the in vitro activity. This agrees with the data of Choi et al. (1999), who showed that dithiothreitol increased the activity by about two-fold. Thus, we added a factor in the model by which GS can be down-regulated by the optimizer during the night.

As can be seen from the model structure in Figure 3, fluxes such as HPR summarize multiple enzymatic steps, for which kinetic parameters may deviate from those determined for individual enzymes that are rate-limiting under most conditions. Thus, for identification of the k_m of the HPR reaction a broad interval was set in order to integrate activity of several enzymes and transporters, for which no parameter boundaries are known. This includes transport of Ser out of the mitochondria and into the peroxisomes, its de-amination by Ser-glyoxylate transaminase, the actual reduction by HPR and finally the phosphorylation of glycerate before its re-integration into the Calvin-Benson cycle. Most importantly, it is unclear whether in Col-0 HPR constitutes the rate-limiting step in the PR pathway.

Setting a broad interval for the HPR parameters created a problem for simulations in hpr1-1, because it allowed for a purely mathematical compensation of the low v_max in the mutant by choosing an adequately low k_m. To prevent this, the identified minimum of the k_m value for HPR in Col-0 was set as lower boundary for simulations of the hpr1-1 mutant with the underlying assumption that the k_m, being a feature of a protein, cannot easily change. However, using this approach the simulated Ser pool was slightly higher then measured for the mutant at eCO₂ (Figure 5) albeit not at ambient. Opening of the lower boundary to allow a further 30% reduction of the k_m could correct this, as demonstrated by the simulations in Figure 1B, and was thus used as setting for all simulations.

Using the parameter sets obtained from the simulations, flux rates at time points of harvest were calculated. Figure 6 shows calculated flux rates for PR-relevant reactions. As expected, HPR flux was higher at ambient than eCO₂ (p < 2e-16) due to the larger PR input. Similar, GDC and SHMT (Figures 6A,B) were increased at ambient CO₂ as compared to eCO₂ (p < 2e-16). Surprisingly, HPR (Figure 6C) showed higher flux rates in the mutant as compared to wildtype (p < 2e-16). This resulted from extremely high levels of Ser, which is the substrate in our compiled HPR reaction, and indicated a metabolic state clearly different from wildtype. For instance, the Ser levels already at the beginning of the day are substantially elevated in the mutant at ambient CO₂. Moreover, in comparison to wildtype the hpr1-1 mutant showed increased levels for GDC (p = 1.16e-10). In contrast, the SHMT reaction tended to be increased in the wildtype, though this was not significant.

Flux within the GS/GOGAT cycle (Figure 7) was tightly linked to PR. For GS and GOGAT genotype and treatment effects were observed (treatment effect on GS and GOGAT: p < 2e-16; genotype effect on GS: p < 2e-16; genotype effect on GOGAT: p = 5.63e-14). The highest flux was obtained at ambient CO₂ for hpr1-1, followed by Col-0 at ambient. For plants grown at eCO₂, a slightly higher flux was observed for hpr1-1 as compared to wildtype, supporting an earlier finding that PR takes place even at a CO₂ concentration of 1,000 ppm (Kraemer et al., 2021a). In summary, the mutant showed a higher turnover in the GS/GOGAT cycle compared to the wildtype, and this turnover was decreased at eCO₂. However, flux of NR (Figure 7C) behaved differently. NR flux was higher in Col-0 and rose at eCO₂, revealing significant genotype and treatment effects (p = 5.66e-11 and p = 4.79e-07, respectively).

Several studies have pointed out that eCO₂ decreases the N content of plant biomass (Bloom et al., 2014; Andrews et al., 2020). However, the underlying mechanism is unclear. Low stomatal conductance at high internal CO₂ concentrations could reduce nitrate availability, but in contrast to observations for wheat under eCO₂ (Del Pozo et al., 2007), we found no indications for reduced mineral content in Arabidopsis wildtype plants (Supplementary Image 3A). Although no consistent change of foliar nitrate was observed in the wildtype, it was significantly lower in the hpr1-1 mutant especially under ambient CO₂. This points to a restricted uptake capability in hpr1-1, probably because of energetic constraints (see below).

The alternative possibility that nitrate reduction was inhibited at eCO₂ (Bloom et al., 2014; Zhao et al., 2021) was also not supported in the current study. We found that the v_max for NR increased at eCO₂ (Figure 7C), and the calculated NR flux was higher at eCO₂ in wildtype plants. The observation that NR was consistently higher in Col-0 as compared to the hpr1-1 mutant, which had high levels of free AA, further argues against insufficient N supply. Thus, our results are in support of reports favoring a dilution of total N to accompany increased carbon fixation and biomass formation under eCO₂ (Andrews et al., 2019). For eCO₂, we found a higher ratio of NR to GDC (Figure 8A). Both fluxes feed into the pool of ammonium used for Gln synthesis. A low GDC contribution would reduce the load on the GS/GOGAT cycle, allowing a higher proportion of de novo N assimilation. Indeed, we calculated a lower flux for GS/GOGAT at eCO₂ (Figures 7A,B), demonstrating a tight link between PR and GS/GOGAT turnover. This link has also been shown by Häusler et al. (1994) and Wallsgrove et al. (1987), who reported that barley mutants lacking GS activity suffered under photorespiratory conditions. In addition, it is known that even short incubations at eCO₂ result in a reduction of GS and GOGAT activity (Guo et al., 2013; Wu et al., 2020). Our finding that the ratio of NR to GDC activity was elevated at eCO₂ in the wildtype as well as the mutant indicates that the GS/GOGAT cycle contained more newly assimilated N at eCO₂, and this further argues against N starvation at eCO₂.

In a previous study comparing plants exposed to eCO₂ either short or long term, we revealed that PR allows deposition of already assimilated N in low carbon-containing AA like Gly and Ser, thus providing carbon skeletons for de novo N assimilation (Kraemer et al., 2021a). While this can explain photosynthetic acclimation to long-term eCO₂, which results in reduced PS as well as NR rate, it does not explain an imbalance in the ratio of both fluxes. Whether sufficient N for the production of biomass was assimilated at eCO₂ should be reflected in the ratio of the carbon flux HP2BMEXP and the de novo N fixation by NR. Figure 8B shows a significant but not dramatic increase of this ratio at eCO₂. Considering the increased NR flux, this indicates that carbon assimilation was even more stimulated than N fixation at eCO₂, which would support the concept of N dilution at eCO₂ (Wong, 1990; Kuehny et al., 1991; Gifford et al., 2000; Taub and Wang, 2008). This is also substantiated by the strongly increased starch levels (Supplementary Image 3B). In the mutant, however, we and others (Timm et al., 2008) found a large pool of free amino acids (Figures 1, 2), which apparently contradicts an increased C/N ratio.

As mentioned above, an energetic constraint could be responsible for the metabolic disturbance in the mutant. As shown in Figure 7, the fluxes for GS and GOGAT are increased in hpr1-1 as compared to wildtype especially at ambient CO₂. Thus, large amounts of ATP and reduced ferredoxin are required to sustain synthesis of Glu, which is needed for the removal of glyoxylate. Figure 8A shows that the NR-to-GDC ratio was lowest for hpr1-1 at ambient, thus indicating futile cycling of ammonium without net gain of biomass.

While a loss of cellular energy could explain the differences in biomass formation of wildtype and mutant at different CO₂ concentrations, which have already been described (Timm et al., 2008), futile cycling of already assimilated ammonium could be regarded as conflicting with high levels of amino acids in the mutant. Accumulation of starch, carboxylates and AA in the mutant could simply result from a slow growth rate as suggested by Génard et al. (2014), but than a reason different from resource limitation must underly stunted growth and chlorotic phenotype of hpr1-1.

A possible explanation could be poisoning by photorespiratory intermediates such as phosphoglycolate, glycolate or glyoxylate (Anderson, 1971; Dellero et al., 2016). As can be seen in Figure 6 the GDC and SHMT fluxes are strongly increased in hpr1-1 at ambient CO₂. But still the HPR flux is substantially extended into the night. This shows that even after light-off photorespiratory intermediates had to be recycled, which were not metabolized during the day in spite of the increased GDC and SHMT activity. It can thus be assumed that phosphoglycolate accumulates during the day, and this would strongly inhibit triose-phosphate isomerase (Anderson, 1971), which would in turn block carbon assimilation in the Calvin-Benson cycle. But how should a bottleneck in the last step of the PR pathway cause accumulation of the early metabolites? As we have already described (Kraemer et al., 2021a), simulations of a metabolic model for the hpr1-1 mutant pointed to an additional source of glycolate, which is independent from oxygenation of ribulose-bisphosphate. A very likely candidate is the non-enzymatic oxidation of hydroxypyruvate by H₂O₂ in the peroxisome, which yields glycolate (Walton and Butt, 1981). This reaction is promoted by the large amount of Ser in the mutant at ambient CO₂. Bao et al. (2021) showed that already a fourfold increase of Ser levels in the catalase mutant cat2 caused a significant increase in hydroxypyruvate decarboxylation. Considering that Ser levels in hpr1-1 were about tenfold higher than wildtype at the end of the day, it is highly likely that non-enzymatic decarboxylation of hydroxypyruvate takes place in hpr1-1. Thus, in the hpr1-1 mutant more glycolate is produced in relation to photosynthetic carbon acquisition as compared to the wildtype. This would not only increase the probability of a toxic effect, but also cause additional loss of assimilated carbon in the form of CO₂.

Besides toxication by photorespiratory intermediates, it would also be possible that an increased level of ammonium (Figure 1C) could interfere with ATP production, because it is in equilibrium with ammonia that acts as an uncoupling agent of ATP synthesis.

Finally, the metabolic bottleneck created by the hpr1-1 mutation caused large amounts of C and N being bound in the form of Gly and Ser. As a consequence, the equilibrium of free amino-acids, brought about by transamination reactions, might be severely disturbed, and this could interfere with protein synthesis in the shoot as well as the supply of the roots and other sinks with AA. Not only N but also C compounds show altered distributions in hpr1-1 (Supplementary Image 1; Figure 2). Especially the carboxylates Cit, malate and fumarate, were significantly enriched, while glucose was reduced in hpr1-1. We did not detect a drop in foliar sucrose content, but the lowered nitrate content of hpr1-1 shoots in ambient CO₂ might indicate low carbon supply to the root system. The low glucose level in leaves could result from enhanced use in the pentose-phosphate pathway as suggested by Li et al. (2019), who reported that high pentose-phosphate pathway activity could provide additional CO₂ that would alleviate the PR syndrome, but restrict biomass formation. A combination of the above described effects could contributes to the mutant phenotype of hpr1-1.