Photorespiration pathways in a chemolithoautotroph

Carbon fixation via the Calvin cycle is constrained by the side activity of Rubisco with dioxygen, generating 2-phosphoglycolate. The metabolic recycling of 2-phosphoglycolate, an essential process termed photorespiration, was extensively studied in photoautotrophic organisms, including plants, algae, and cyanobacteria, but remains uncharacterized in chemolithoautotrophic bacteria. Here, we study photorespiration in the model chemolithoautotroph Cupriavidus necator (Ralstonia eutropha) by characterizing the proxy-process of glycolate metabolism, performing comparative transcriptomics of autotrophic growth under low and high CO2 concentrations, and testing autotrophic growth phenotypes of gene deletion strains at ambient CO2. We find that the canonical plant-like C2 cycle does not operate in this bacterium and instead the bacterial-like glycerate pathway is the main photorespiratory pathway. Upon disruption of the glycerate pathway, we find that an oxidative pathway, which we term the malate cycle, supports photorespiration. In this cycle, glyoxylate is condensed with acetyl-CoA to give malate, which undergoes two oxidative decarboxylation steps to regenerate acetyl-CoA. When both pathways are disrupted, autotrophic growth is abolished at ambient CO2. We present bioinformatic data suggesting that the malate cycle may support photorespiration in diverse chemolithoautotrophic bacteria. This study thus demonstrates a so-far unknown photorespiration pathway, highlighting important diversity in microbial carbon fixation metabolism.


Introduction
The Calvin cycle is responsible for the vast majority of carbon fixation in the biosphere. However, its activity is constrained by the low rate and the limited substrate specificity of its carboxylating enzyme Rubisco. The oxygenase side-activity of Rubisco converts ribulose 1,5-bisphosphate (RuBP) into 3-phosphoglycerate (3PG) and 2-phosphoglycolate (2PG), a dead-end metabolite which can inhibit the Calvin cycle (1,2). The metabolic recycling of 2PG, termed photorespiration, is an essential process for most organisms that grow autotrophically via the Calvin cycle (3). Photorespiration has been extensively studied in photosynthetic organisms, including plants, algae, and cyanobacteria (4)(5)(6)(7). The only identified photorespiration pathway in plants is the so-called C 2 cycle (Fig. 1), in which 2PG is first dephosphorylated to glycolate, then oxidized to glyoxylate, and subsequently aminated to glycine. One glycine molecule is decarboxylated to give 5,10methylene-THF which reacts with another glycine to yield serine. Serine is then deaminated to hydroxypyruvate, further reduced to glycerate, and finally phosphorylated to generate the Calvin cycle intermediate 3PG.
Using gene deletion studies, the cyanobacterium Synechocystis sp. PCC6803 was demonstrated to harbor two photorespiratory routes in addition to the C 2 cycle (5,8). In the glycerate pathway, two glyoxylate molecules are condensed to tartronate semialdehyde, which is subsequently reduced to glycerate and phosphorylated to 3PG (Fig. 1). Alternatively, in the oxalate decarboxylation pathway, glyoxylate is oxidized to oxalate and decarboxylated to formate, which is finally oxidized to CO 2 (Fig. 1). Growth at ambient CO 2 was abolished only when all three pathways were deleted, indicating that each of these routes can participate in photorespiration (5).
Photorespiration was initially named to describe light-dependent CO 2 release in plants (9). Even though the name implies otherwise, photorespiration is not restricted to phototrophs but must occur in all organisms that use the Calvin cycle for autotrophic growth in the presence of oxygen. In fact, a wide range of chemolithoautotrophic microorganisms that employ the Calvin cycle under oxic conditions -including bacteria that oxidize hydrogen, ferrous iron, sulfur, or ammonia (10,11) -must also cope with the oxygenase side-activity of Rubisco by recycling or removing 2PG. Yet, despite the obvious physiological significance of photorespiration to chemolithoautotrophs, it has so far received only scarce attention.
In this study, we explore metabolic routes involved in photorespiration of Cupriavidus necator H16 (formerly known as Ralstonia eutropha or Alcaligenes eutrophus), the best-studied chemolithoautotrophic microorganism that fixes CO 2 fixation via the Calvin cycle (12)(13)(14). Unlike cyanobacteria, C. necator does not harbor a CO 2 concentrating mechanism (i.e., a carboxysome with appropriate inorganic carbon transporters), as evident from the relatively high CO 2 specificity of its Rubisco, which falls within the range reported for plants but is much higher than that found in cyanobacteria (15)(16)(17). Very little is known about photorespiration in C. necator. Previous studies found evidence only for the first steps of photorespiration -2PG dephosphorylation and glycolate oxidation (18)(19)(20). However, the specific routes that metabolize glyoxylate remain elusive. glycolate and then oxidized, by the glycolate dehydrogenase complex, to give glyoxylate. Glyoxylate can be further metabolized via four routes: the C2 cycle (i.e., the plant photorespiration pathway), the glycerate pathway, the oxalyl-CoA decarboxylation pathway, and malate cycle. The latter two routes completely oxidize glycoxylate to CO2. Note that that malate cycle may also proceed via oxidation of malate to oxaloacetate, which can be converted to pyruvate either directly or via phosphoenolpyruvate. Abbreviated Here, we identify the native photorespiration pathways of C. necator by performing comparative transcriptomic analysis and conducting growth experiments with gene deletion strains. We show that the C 2 cycle and decarboxylation via oxalate do not support photorespiration in C. necator. We find that the 5 glycerate pathway is the primary photorespiration route in this bacterium. A second pathway, which we term the malate cycle, carries all photorespiratory flux when the glycerate pathway is disrupted. This route was previously unknown to operate in nature and can completely oxidize glyoxylate to CO 2 . In this cycle, glyoxylate is condensed with acetyl-CoA to generate malate, which then undergoes oxidative decarboxylation twice to regenerate acetyl-CoA (Fig. 1). Only when both the glycerate pathway and the malate cycle are disrupted is autotrophic growth at ambient CO 2 abolished. This study therefore fills an important gap in our understanding of chemolithoautotrophic metabolism.

Multiple pathways in C. necator support growth on glycolate
As glycolate metabolism is at the core of photorespiration (Fig. 1), we decided to start by exploring the metabolic pathways that can support the growth of C. necator on this C 2 carbon source. First, we focused on the initial oxidation of glycolate to glyoxylate, which is the first step in all glycolate-metabolizing routes. The operon glcDEF is annotated to encode different subunits of the glycolate dehydrogenase complex (21). The gene kch, which is also part of the operon, is annotated as an ion transport channel, but as it shows homology to FAD-binding proteins, including subunit D of glycolate dehydrogenase from Ralstonia syzygii R24, it is more likely to encode another subunit of the complex. While a wild-type C. necator could efficiently grow on glycolate (doubling time of 3.2 ± 0.1 h, 'WT' in Fig. 2a), a strain deleted in the glcDEF operon failed to grow on this carbon source ('∆GDH' in Fig. 2a). This confirms that GlcDEF plays an essential role in oxidizing glycolate to glyoxylate.
necator also harbors the key components of the C 2 cycle, that is, the glycine cleavage system (encoded by gcvTHP) and serine hydroxymethyltransferase (glyA). While no dedicated glycine-dependent or serinedependent transaminases are annotated in the genome of C. necator (Fig. 1), it is possible that glyoxylate amination and serine deamination are supported by one or more of the endogenous transaminase enzymes.
Finally, the Synechocystis-like oxalate decarboxylation pathway seems not to be present in C. necator, as no oxalate decarboxylase gene could be found. However, a similar route might be used. Specifically, oxalyl-CoA decarboxylase and formyl-CoA transferase -the genes for which, oxc and frc, are annotated and found in a single operon -could enable a glyoxylate oxidation route. In this putative route, oxalate is activated to oxalyl-CoA, which is then decarboxylated to formyl-CoA, converted to formate, and finally oxidized to CO 2 ( Fig. 1). However, no CoA-acylating glyoxylate dehydrogenase (catalyzing the first reaction of the pathway) is annotated in the genome of C. necator. Hence, we performed a BLAST search of the bacterium's genome using a CoA-acylating glyoxylate dehydrogenase from Methylobacterium extorquens (panE2) (22) as a query. We identified the gene apbA2, located in close proximity to the frc-oxc operon and showing high sequence homology to the M. extorquens gene (66% similarity, 51% identity), as a probable candidate glyoxylate dehydrogenase.
To test whether ApbA2 can indeed catalyze the reversible CoA-acylating glyoxylate dehydrogenase reaction in vivo, we tested the growth of a ∆apbA2 strain on oxalate. Growth on oxalate can proceed via two routes (Supplementary Fig. S1): (i) assimilation to central metabolism via the glycerate pathway, which depends on the activity of a CoA-acylating glyoxylate dehydrogenase (in the reverse direction of that required for growth on glycolate); and (ii) complete oxidation of oxalate which can be followed by carbon fixation via the Calvin cycle. Hence, if ApbA2 serves as a CoA-acylating glyoxylate dehydrogenase, its deletion should impair growth on oxalate. Indeed, while a wild-type C. necator grew efficiently on oxalate (doubling time 6.5 ± 0.1, Supplementary Fig. S1) growth of the ∆apbA2 strain on this carbon source was impaired (doubling time of 10 ± 1.1 h, Supplementary Fig. S1). This deletion strain showed a similar growth phenotype to that observed upon deletion of the enzymes of the glycerate pathway (doubling time of 13 ± 0.7 h, Supplementary Fig.   S1). It therefore seems that ApbA2 can indeed act as a CoA-acylating glyoxylate dehydrogenase.
Interestingly, deletion of the frc-oxc operon completely abolished growth on oxalate ( Supplementary Fig.   S1), presumably as the supply of reducing power via oxalate oxidation is necessary to support the assimilation of this highly oxidized substrate.
By generating three distinct mutant strains, carrying deletions in the gcl-hyi2-tsr operon, the gcvTHP operon, or the frc-oxc operon, we explored the contribution of each candidate route to growth on glycolate. While the deletion of the latter two operons did not affect growth on glycolate (doubling time of 3.4 ± 0.1 h, '∆C2 ∆OX ∆MC' in Fig. 2a), deletion of the gcl-hyi2-tsr operon resulted in a substantially lower growth rate (doubling time of 7.9 ± 0.1 h, '∆GP' in Fig. 2a), indicating that the glycerate pathway is the main route for growth on glycolate (Fig. 2b). Still, the ability of the strain lacking the gcl-hyi2-tsr operon to grow on glycolate implies that other routes can support glyoxylate metabolism.
Next, we deleted the gcvTHP operon or the frc-oxc operon in the strain already lacking the gcl-hyi2-tsr operon. We found that further disruption of the oxalyl-CoA decarboxylation pathway substantially reduced the growth rate Yet, only few other enzymes can react with glyoxylate and even fewer exist in C. necator. For example, the β-hydroxyaspartate cycle was recently shown to enable growth on glycolate via glyoxylate assimilation (23), but the genome of C. necator does not encode its key glyoxylate assimilating enzyme, β-hydroxyaspartate aldolase. We were able to identify only one other enzyme in this bacterium that can react with glyoxylate: malate synthase, a key component of the glyoxylate shunt that condenses glyoxylate with acetyl-CoA to generate malate (24,25). Indeed, when the gene encoding malate synthase (aceB) was deleted in the strain deleted in the gcl-hyi2-tsr operon, growth on glycolate was completely abolished ('∆GP ∆MC' in Fig. 2a). On the other hand, deletion of aceB in a strain in which the glycerate pathway is still active did not affect growth on glycolate ('∆C2 ∆OX ∆MC' in Fig. 2a). The main route glycolate assimilation in C. necator growth is glycolate dehydrogenase followed by the glycerate pathway. (c) In the absence of the glycerate pathway, glyoxylate is decarboxylated via the oxalyl-CoA decarboxylation pathway and the malate cycle, the latter route being the dominant one. Generated CO2 and reducing equivalents are utilized by the Calvin cycle to support biomass formation. Strain labels: ∆C2, C2 cycle knockout (∆gcvTHP); ∆GDH, glycolate dehydrogenase knockout (∆glcD-kch-glcE-glcF); ∆GP, glycerate pathway knockout (∆gclhyi-tsr); ∆MC, malate cycle knockout (∆aceB); ∆OX, oxalate decarboxylation knockout (frc-oxc); ∆Rub, Rubisco knockout (∆cbbS2-cbbL2 ∆cbbSp-∆cbbLp); WT, wild-type.
For glyoxylate metabolism via malate synthase to proceed, the co-substrate acetyl-CoA must be regenerated. Such regeneration requires malate to undergo oxidative decarboxylation twice, first to pyruvate and then to acetyl-CoA. The combined activity of the malic enzyme and pyruvate dehydrogenase can support this double oxidation. The result is a 'malate cycle', composed of malate synthase, malic enzyme, and pyruvate dehydrogenase, which together completely oxidize glyoxylate to CO 2 while generating two NAD(P)H molecules (Fig. 1). (We note that this cycle could alternatively proceed via malate oxidation to oxaloacetate, which is then converted to phosphoenolpyruvate via phosphoenolpyruvate carboxykinase, 8 and further metabolized to pyruvate and acetyl-CoA; this alternative malate cycle would result in the same net decarboxylation reaction).
C. necator is expected to grow on glycolate via the malate cycle only by using the generated reducing power to fix CO 2 via the Calvin cycle (Fig. 2c). To test if this is indeed the case, we deleted all genes encoding for Rubisco (cbbS2, cbbL2, cbbSp, and cbbLp) in the strain disrupted in the glycerate pathway (i.e., deleted in the gcl-hyi2-tsr operon). As anticipated, this strain was not able to grow on glycolate ('∆GP ∆Rub' in Fig.   2a); note that the deletion of Rubisco in a wild-type strain did not affect growth on glycolate as the glycerate pathway is still active (doubling time of 3.3 h, '∆Rub' in Fig. 2a). These results confirm that growth on glycolate via the malate cycle strictly depends on the carbon fixation via the Calvin cycle.
To summarize, growth on glycolate mainly depends on the glycerate pathway in a wild-type C. necator (Fig.   2a). When this route is disrupted, glyoxylate is metabolized via a combination of the malate cycle and the oxalyl-CoA decarboxylase pathway, both of which depend on CO 2 fixation for growth (Fig. 2c). Under these conditions, the malate cycle carries most of the flux and is essential for growth, while the oxalyl-CoA decarboxylase pathway has a secondary role. Yet, it is still not clear whether the pathways that support growth on glycolate also participate in photorespiration during autotrophic growth at ambient CO 2 .

Photorespiration in C. necator is supported by the glycerate pathway and the malate cycle
To study which of the pathways of glycolate metabolism also participates in photorespiration, we compared the transcript levels of a wild-type C. necator growing autotrophically on hydrogen at ambient CO 2 concentrations (≈0.04%) versus elevated CO 2 concentrations (10%); the latter condition suppresses the oxygenation reaction. We found that the genes encoding for the Calvin cycle enzymes were overexpressed between 4-and 12-fold under ambient CO 2 concentration (Supplementary Data 1). This result is expected as higher activity of the cycle is needed to compensate for the decreased rate of Rubisco and the carbon loss from photorespiration. Furthermore, the genes encoding the first steps of photorespiration, that is, 2PG phosphatase (cbbZ2 and cbbZp) and the glyoxylate dehydrogenase complex, were overexpressed between 4-and 8-fold ( Fig. 3 and Supplementary Data 1).
Genes encoding the enzymes of the glycerate pathway were among the most highly upregulated at ambient CO 2 ( Fig. 3 and Supplementary Data 1): gcl and hyi2 were more than 300-fold overexpressed, tsr was ≈50fold upregulated, and ttuD1 was ≈25-fold upregulated. On the other hand, the genes related to other potential glyoxylate metabolism routes -that is, the C 2 pathway, the oxalyl-CoA decarboxylation pathway, and the malate cycle -were not overexpressed at ambient CO 2 (Supplementary Data 1). This suggests that the glycerate pathway is the main photorespiration route. However, the fact that the genes of the other pathways were not overexpressed does not necessarily mean that they do not participate in photorespiration. Specifically, it could be that their basal expression levels are sufficient to support the required activity. For example, the genes encoding for the components of the malate cycle -aceB, maeA, maeB, pdhA1, pdhB, and pdhL -are highly expressed both under ambient and high CO 2 concentrations (all among the 10% most highly expressed in both conditions; see Supplementary Data 1); hence, the malate cycle could play a role in photorespiration. Only genes that are significantly upregulated and related to photorespiration pathways are labelled in the plot (red dots To determine the relative importance of each candidate pathway in photorespiration, we tested wild-type C. necator and several of the gene deletion strains described above for their ability to grow autotrophically at ambient CO 2 (Fig. 4). Autotrophic growth of wild-type C. necator under these conditions resulted in a much lower growth rate than observed at 10% CO 2 (doubling time of 21 ± 0.7 h vs. ~3 h, solid vs. dashed 'WT' lines in Fig. 4, respectively). This difference is expected from the lower carboxylation rate of Rubisco at low CO 2 concentrations and the relatively high rate of the oxygenation reaction which leads to CO 2 release, thus directly counteracting carbon fixation.
A strain lacking all routes of glyoxylate metabolism besides the glycerate pathway did not show reduced growth at ambient CO 2 (doubling time of 20 ± 2 h, '∆C2 ∆OX ∆MC' in Fig. 4). On the other hand, a strain in which the glycerate pathway was disrupted displayed a substantially lower growth rate (doubling time of 35 ± 1 h, '∆GP' in Fig. 4). This suggests that the glycerate pathway is the major route of photorespiration, but also that it can be replaced by other pathways. Further deletion of the malate cycle in the strain lacking the glycerate pathway completely abolished autotrophic growth at ambient CO 2 ('∆GP ∆MC' in Fig. 4). On the other hand, disruption of the C 2 pathway and the oxalyl-CoA decarboxylation pathway in the strain lacking the glycerate pathway did not alter its growth phenotype (doubling time of 31 ± 1 h, '∆GP ∆C2 ∆OX' in Fig.   4). These results clearly show that the malate cycle can participate in photorespiration while the other two pathways contribute little, if any, to this process. Notably, deletion of both the glycerate pathway and the malate cycle did not affect autotrophic growth at high CO 2 concentrations (dashed '∆GP ∆MC' line in Fig. 4), as photorespiration is expected to be negligible under these conditions due to competitive inhibition of oxygenation at high CO 2 . We were also interested to explore the outcome of disrupting photorespiratory metabolism upstream of glyoxylate. We found that a strain deleted in glycolate dehydrogenase could grow autotrophically under ambient CO 2 concentration, albeit at substantially lower growth rate and yield (doubling time of 38 ± 4 h, '∆GDH' in Fig. 4). Supporting previous studies (18,19), we found that glycolate accumulates in the medium during autotrophic growth of this strain at ambient CO 2 (Fig. 5). Since glycolate is secreted by this strain and is not further oxidized, we hypothesized that further deletion of all four glyoxylate metabolism routes would not affect the growth phenotype; we indeed found this to be the case (doubling time of 41 h, '∆GDH ∆GP ∆C2 ∆OX ∆MC' in Fig. 4). Moreover, as expected, no glycolate was detected during the autotrophic growth of a wild-type strain at ambient CO 2 . Growth experiments were conducted in 700 mL bioreactor cultures on minimal medium (JMM) with a continuous sparging of gas (6.25 L/minute) with ambient air + 4% hydrogen. Biological duplicates were measured and SD for the glycolate concentrations are shown, replicates showed identical growth curves (±5%) and a representative curve is shown. Glycolate was measured by ion chromatography as explained in materials and methods.
It therefore seems that C. necator can excrete glycolate if necessary, thus enabling the autotrophic growth of a strain lacking glycolate dehydrogenase. In contrast, C. necator seems to be incapable of secreting glyoxylate, which prevents the growth of a strain lacking the glycerate pathway and the malate cycle.
Indeed, we could not detect glyoxylate in the growth medium of any tested strains.

The malate cycle may be prevalent in chemolithoautotrophs using the Calvin cycle
To explore the potential distribution of the malate cycle, we searched for the occurrence of malate synthase (PFAM 01274 (26)) in bacteria that harbor a bona-fide Rubisco, and are therefore likely to operate the Calvin cycle and rely on a photorespiration to metabolize 2PG (Methods). Only 2% of cyanobacterial species harbor malate synthase, indicating that the malate cycle is likely uncommon among oxygenic photoautotrophic bacteria (Table 1). However, of the remaining ≈2000 non-cyanobacterial genomes found to encode Rubisco, ≈60% also encode a malate synthase. These include bacteria that grow chemolithoautotrophically by oxidizing either inorganic compounds (e.g., hydrogen, ammonia, and sulfur compounds) or organic one-carbon compounds (e.g., formate or methanol), as well as non-oxygenic phototrophs that use the Calvin cycle (such as purple non-sulfur bacteria). In several key orders of aerobic autotrophic bacteria, most genomes that harbor a Rubisco also encode a malate synthase (Table 1)  The other enzymes of the malate cycle -the malic enzyme and pyruvate dehydrogenase or their alternatives (e.g., malate dehydrogenase, phosphoenolpyruvate carboxykinase) -are quite ubiquitous.
However, these enzymes are not strictly necessary. While growth on glycolate via the malate cycle must be 13 accompanied by the regeneration of acetyl-CoA and thus complete oxidation of glyoxylate, in photorespiration, acetyl-CoA does not need to be regenerated. Rather, the glyoxylate produced in photorespiration can be condensed with acetyl-CoA generated from the carbon fixation process and the resulting malate assimilated into biomass. In this case, the malate 'cycle' is not a real cycle but rather represents a linear photorespiration route to reassimilate glyoxylate into central metabolism. We suspect that some chemolithoautotrophs using malate synthase as part of their photorespiratory metabolism actually operate the linear than rather than cyclic version of this pathway.

Discussion
This study aimed to fill gaps in our knowledge of photorespiratory metabolism in chemolithoautotrophic microorganisms. In the model chemolithoautotroph C. necator, we confirmed the role of the glycolate dehydrogenase complex in photorespiration and further revealed that two metabolic routes can sustain photorespiratory flux and thus support autotrophic growth at ambient CO 2 .
The glycolate dehydrogenase complex of C. necator is homologous to other bacterial glycolate dehydrogenases. The cofactor used as an electron acceptor for glycolate oxidation has not yet been elucidated in any bacteria. It is sometimes proposed, especially in cyanobacterial photorespiration, that NAD + serves as the electron acceptor. However, this is highly doubtful as the change in Gibbs energy for the reaction glycolate + NAD + = glyoxylate + NADH is very high (∆ r G' m > 40 kJ/mol, pH 7.5, ionic strength of 0.25 mM; ∆ r G' m corresponds to metabolite concentration of 1 mM (27)). Instead, it is more likely that glycolate transfers its electrons, via a flavin adenine dinucleotide cofactor, to a quinone, the reduction potential of which is substantially higher than that of NAD (E' m ~ 0 mV rather than ~ -300 mV, respectively).
Unlike most phototrophic organisms, which mostly cannot grow heterotrophically, C. necator can be grown on various organic compounds. We used this metabolic versatility to explore photorespiration via a proxyprocess of glycolate metabolism. Indeed, we found that photorespiration and growth on glycolate generally rely on the same routes, that is, the glycerate pathway and the malate cycle. In contrast, the oxalyl-CoA decarboxylation pathway seems to participate, albeit marginally, only in glycolate metabolism but not in photorespiration. This might be attributed to differences in regulation or to the higher concentrations of glyoxylate available when supplying glycolate as a sole carbon source.
Autotrophic growth phenotypes of our C. necator gene deletion strains show that the C 2 pathway contributes negligibly to photorespiration and glycolate metabolism. While C. necator harbors the main components of this pathway, transaminase enzymes that accept glycine and serine could not be identified. It might be the case that these enzymatic activities are completely missing in the bacterium. Alternatively, it could be that endogenous transaminases can aminate glyoxylate and deaminate serine but the C 2 pathway shows only low activity due to inadequate regulation.
A key finding of our work is the existence of the malate cycle as a supporting route for photorespiration and growth on glycolate. It is difficult to determine whether this route plays a role in the wild-type strain, as its deletion does not seem to hamper growth when the glycerate pathway is present. It could be that the malate cycle carries non-negligible flux only when the glycerate pathway is disrupted and glyoxylate begins to accumulate. Alternatively, considering the relative high expression of malate synthase, it is possible that the pathway always supports a substantial fraction of glyoxylate metabolism, but not enough to affect growth once deleted.
The glycerate pathway is the most efficient, naturally occurring photorespiration route in terms of consumption of ATP and reducing power (28). On the other hand, complete decarboxylation of glyoxylate to CO 2 -as supported by the cyanobacterial oxalate decarboxylation pathway as well as the oxalyl-CoA decarboxylation pathway and the malate cycle described here -is arguably the least efficient photorespiration mode, as it requires higher activity of the Calvin cycle to compensate for the lost carbon.
This might explain why the deletion of the glycerate pathway in C. necator, such that the malate cycle carries the entire photorespiratory flux, resulted in lower growth rate and yield (Fig. 4). Similarly, the superiority of the glycerate pathway might explain why it serves as the major photorespiration route both in cyanobacteria and in C. necator.
Despite the relative inefficiency of the malate cycle, its implementation in plants was suggested to boost carbon fixation and photosynthesis (28)(29)(30)(31). Recently, the heterologous expression of malate synthase and glycolate dehydrogenase within the chloroplast of the agricultural crop Nicotiana tabacum led to a substantial increase in photosynthetic rate and yield (29). As the malate cycle is less efficient than the natural C 2 pathway, this growth enhancement is not easy to explain and was suggested to be related to the release of CO 2 in the chloroplast, rather than the mitochondria, thus suppressing the oxygenase sidereaction and promoting Rubisco's carboxylation.
As malate synthase is present in many chemolithoautotrophs that use the Calvin cycle (Table 1), it is tempting to suggest that it contributes to photorespiration in many of these bacteria. However, the occurrence of malate synthase in cyanobacteria is much lower (Table 1). Synechocystis, the only bacterial photoautotroph for which photorespiration has been physiologically characterized (5,8), probably lacks malate synthase and hence cannot operate the malate cycle (32,33). However, the presence of malate synthase has been confirmed in some cyanobacteria (34)(35)(36)(37), leading to the suggestion that they may use the malate cycle. We leave it for future studies to explore the occurrence of the malate cycle and other photorespiration routes in cyanobacteria and chemolithoautotrophs. Such investigations will generate further insights into the evolutionary history of photorespiratory metabolism.

Strains, conjugations and gene deletions
C. necator H16 (DSMZ 428) was used for transcriptome studies. Growth experiments were performed for a C. necator H16 strain knocked out for polyhydroxybutyrate biosynthesis (∆phaC1) (38), in which other gene deletions were performed. The ∆phaC1 strain grows in nutrient non-limiting similar to the wild-type and does not result in PHB granules that could disturb optical density measurements.
Cloning of plasmids was performed in E. coli DH5α, whereas E. coli S17-1 was used for conjugation of mobilizable plasmids to C. necator by biparental overnight spot mating. A complete overview of strain genotypes used in this study can be found in Table 2.

Table 2. Strains used in this study
Gene deletions were performed with the pLO3 suicide vector, as previously described (41,42 (40) selected on agar plates with tetracycline (10 µg/mL) and 10 µg/mL gentamycin tot counter-select for E. coli.
Next, some transconjugants were grown in an overnight liquid culture (without tetracycline) to support a second homologous recombination event. The overnight cultures were plated on LB with 10% sucrose to allow for SacB counter selection. Resulting colonies were screened by colony PCR (OneTaq, Thermo Scientific) to identify gene-deleted strains (primers in Supplementary Table S1), which were further verified for having lost tetracycline resistance. Biological duplicates of each knockout strain were constructed.

Growth medium and conditions
C. necator and E. coli were cultivated for routine cultivation and genetic modifications on Lysogeny Broth (LB) (1% NaCl, 0.5% yeast extract and 1% tryptone). Routine cultivation was performed in 3 mL medium in 12 mL glass culture tubes in a Kuhner shaker incubator (240 rpm) at 30ºC for C. necator and 37ºC for E.
coli. Growth characterization and transcriptomic experiments of C. necator were performed in J Minimal Medium (JMM) as reported previously (43).

RNA isolation and sequencing
For transcriptome analysis, cells were grown in 10 mL JMM (without carbon source) in 100 mL erlenmeyers within a 10 L desiccator filled with 4% hydrogen + 96% ambient air, or 4% hydrogen + 10% CO 2 and 86% air. To maintain hydrogen and CO 2 , the gas phase in the desiccator was exchanged at least twice a day.
Biological replicate cultures were harvested in log phase (2 mL culture for OD 600 ~0.2 for ambient CO 2 , 1 mL for OD 600 ~0.4 for 10% CO 2 ) and stabilized by RNA Protect Bacteria Kit (Qiagen). Next, cells were lysed using lysozyme and a bead-beating step with glass beads (Retschmill, MM200), for 5 minutes at 30 hertz.
RNA was purified using the RNeasy Mini kit (Qiagen) according to manufacturer's instructions and oncolumn DNAase digestion (DNase kit Qiagen). rRNA depletion (RiboZero kit), cDNA library preparation, and paired-end 150 bp read sequencing (Illumina HiSeq 3000) was performed by the Max Planck Genome Centre Cologne, Germany.

Transcriptome data analysis
Sequence data of all samples were mapped with STAR v2.5.4b using default parameters (44). Ensembl version 38 genome reference in FASTA format and Ensembl version 38 cDNA Annotation in GTF format were used for genome indexing with adapted parameters for genome size (--genomeSAindexNbases 10) and read length (--sjdbOverhang 150). Anti-strand reads out of the ReadsPerGene files, which are automatically generated by STAR, were used in two different ways: calculating reads per kilo base of exon per million mapped reads (RPKM) for sample-wise transcript abundances as well as merging in order to perform a differential expression analysis with DESeq2 (45) as guided by the rnaseqGene Bioconductor workflow (https://bioconductor.org/packages/release/workflows/vignettes/rnaseqGene/inst/doc/rnaseqGene.html).
Briefly, samples were grouped by single parameter condition (ambient CO 2 and 10% CO 2 ), read count data were then loaded with DESeqDataSetFromMatrix to create a DeSeqDataSet object to subsequently run the standard analysis consisting of the functions DESeq and Results. Then, log2-transformed fold changes (log2fc) for ambient CO 2 compared to 10% CO 2 and absolute log10 of adjusted p-values were determined and were visualized in a Volcano plot.

Supernatant analysis for glycolate and glyoxylate
Glycolate concentrations in culture supernatant were determined by ion chromatography (IC) analysis.
Supernatant samples were diluted 1:100 in ultrapure water (Milli-Q). The samples were analyzed in an ICS 3000 (Dionex) ion chromatography system, which was combined with a AS50 auto sampler. The samples were run through a Dionex™ IonPac™ AS11 IC column (4 mm diameter, 250 mm length (044076)) and a guard column (4mm diameter, 50mm length (044078)). Samples were run following KOH eluent gradient: 1 mM from 0 to 5 minutes, 1 mM to 15 mM from 5 to 14 minutes,15 mM to 30 mM from 14 to 23 minutes,30 mM to 60 mM from 23 to 31 min at a flow rate of 0.015 mL/minute. The experimental data were analyzed using Chromeleon 6.8. Concentrations were calculated based on a standard curve generated for sodium glycolate (Sigma-Aldrich) in JMM medium. Glyoxylate concentrations were determined by a colorimetric assay based on a reported protocol (46). Specifically, 216 µL from the supernatant sample were mixed with 24 µL of 1% w/v phenylhydrazine in 0.1 M HCl and incubated for 10 minutes at 60ºC and cooled down. To 100 µL of this mixture, 50 µL concentrated HCl and 20 µL 1.6% w/v potassium ferricyanide were added, while background control samples were prepared with 100 µL of reacted sample mixture with 50 µL concentrated HCl and 20 µL MQ water. These samples were incubated for exactly 12 minutes and then absorbance of 1,5-diphenylformazan at 520 nm was recorded by in a BioTek Epoch 2. Differences in absorbance were calculated for each sample by subtracting absorbance from background controls and glyoxylate concentration could be determined based on a standard curve. All supernatant samples of all autotrophic cultures in this work resulted in negligible levels of <0.1 mM glyoxylate.

Genomic prevalence analysis of Rubisco and malate synthase
Lists of bacterial genomes containing the large subunit of Rubisco (PF00016) or malate synthase (PF01274) were downloaded from the AnnoTree website (47) on April 27 th h 2020 by searching for the appropriate protein families. As of writing, AnnoTree uses version 89 of the GTDB taxonomy (48) , which was retrieved from the GTDB website on the same date. Rubisco sequences were filtered by using usearch (49) to remove any amino acid sequences with >30% identity to known Rubisco-like proteins (RLPs, type IV Rubiscos). The list of Rubisco-like proteins was drawn from (50) and Rubisco-like proteins were removed because they do not catalyze the carboxylation reaction (51). Organisms encoding R and malate synthase were identified by using their GTDB IDs to merge the two tables along with taxonomic information. This permitted calculation of the fraction of Rubisco-containing genomes that also contain malate synthase for each order in the GTDB taxonomy. Analyses were performed in a custom Python script that is available here: https://github.com/flamholz/malate_synthase/blob/master/pipeline/01_plot_co_occurrence.ipynb.