Abstract

Background

Two fundamental features of living systems are heredity and metabolism, the latter being controlled by the former [1-3]. Although heredity is often considered to be exclusively dependent on template replication of nucleic acid polymers, it is not the only way of storing and transmitting information. Membranes [4] and epigenetic chromatin-markings [5], for example, are considered to be replicators providing a limited but important part of cellular inheritance. From the chemical point of view the essence of replication is autocatalysis [6], that is, when a compound catalyses its own formation (for example, DNA is needed for the synthesis of more DNA). One key model of minimal life [7] suggests that in addition to template replication and membrane growth, metabolism is also autocatalytic and, hence, results in replication.

In a trivial sense the cytoplasm is autocatalytic in that just the membrane and DNA alone are incapable of replication. DNA may code for the constituents of cytoplasm, but without these constituents there is no machinery to do anything. Enzymes would need to be added to the system. Enzymes catalyze the synthesis of more enzymes, so the enzymatic machinery can be regarded as autocatalytic. However, according to Gánti's theory [1,8,9], metabolism is autocatalytic at the level of small molecules (intermediates) as well, hence mere addition of enzymes and raw materials should be unable to kick-start the system. This being so we should be able to identify autocatalytic components of metabolism that are, in effect, replicators. It has been pointed out that the Calvin cycle [1] and the reductive citric acid cycle [10] are such autocatalytic networks.

As Gánti [7] pointed out, any member of an autocatalytic cycle is an autocatalyst. Thus, the reductive citric acid cycle can be launched with any of the intermediates, including, for example, fumarate, succinate, citrate, or oxaloacetate. The same is true of metabolic networks, but networks present additional complications because of their complicated stoichiometric structure. Consider the Calvin cycle as analyzed by Gánti [11]. If we provide the system with the necessary coenzymes (including ATP) and CO₂, one molecule of 3-phospho-glycerate is still not sufficient for autocatalytic growth: one needs three molecules of 3-phospho-glycerate to produce a fourth one, and, ultimately, three new molecules in addition to the three with which the system was successfully launched. But since we are dealing with a network, alternative starting compound sets are possible, such as (xylulose-5-P AND erythrose-4-P), OR (dioxyacetone-P AND 3-P-glyceraldehyde AND erythrose-4-P). The more complex the autocatalytic network, the more alternative sets we can expect to be identified, but this expectation is reduced by the also increasing number of interconversions due to alternative reaction pathways in large systems. But in general a few alternative obligatorily autocatalytic sets can be expected to be present.

The relevance of these cycles is not clear, however, as these are embedded in larger networks of reactions, through which the cycle intermediates could potentially be reconstructed. Thus, the existence of an autocatalytic sub-network does not guarantee that the whole network is autocatalytic (Figure ​(Figure1).1). Conversely, lack of an obvious smallish (easy-to-identify) autocatalytic cycle does not prevent the whole metabolism from being autocatalytic as several auto- and cross-catalytic (Figure ​(Figure1c)1c) molecules might be present in the network, which can be produced remote from where they are consumed in the first place (Figure ​(Figure1d).1d). We can reformulate the empirical question as follows: are the intermediates of a given metabolic network accessible just from the raw materials (members of a food set), or does one need to add some molecule(s) from the network itself? It is obvious that the answer will depend on the nature of the organism and the specified food set. A richer medium may allow the synthesis of compounds that would otherwise be inaccessible without some help 'from within' (that is, an autocatalytic component; Figure ​Figure1).1). From now on we call a metabolic network autocatalytic if at least one additional non-food metabolite must be added to the network of an organism to render it complete so that all its known reactants become accessible by a series of biochemical reactions. Importantly, a mere kinetic autocatalytic effect of, say, a cofactor, is not sufficient to include it in the set of autocatalytic compounds: if this compound can be synthesized by at least one alternative pathway just from the food set and other autocatalytic molecules, then we do not add it to the set of autocatalytic compounds. Thus, the autocatalytic metabolite set of an organism includes all compounds that can be synthesized by small molecule metabolism, have an autocatalytic nature, and that must be present within the cell because otherwise the network, or part of it, would halt. We refer to such cases as obligate autocatalysis.

Open in a separate window

Figure 1

Various metabolic organizations. (a) A protocell showing an indispensable autocatalytic metabolite. (b) A richer medium is able to kick-start metabolism because A can be formed from Z. (c) The set of autocatalytic molecules is composed of A and B, a pair of cross-catalytic molecules (that is, A is required to kick-start the biosynthetic route to B, and B is required to kick-start the biosynthetic route to A). Although the excess molecules are produced in a mirror fashion, it is easy to identify the autocatalytic compounds. (d) If reactions involving A and B are embedded in a large network, it may not be easy to identify the autocatalytic compounds: note that the excess A is produced remote from where it was consumed in the first place. Symmetrically, the same holds for B. Inner metabolite A, food X and Z, waste Y and W.

Another intellectual source behind the present study is an earlier nutrient-related analysis of Escherichia coli metabolism by Romero and Karp [12] aiming at identifying incomplete regions of a pathway database. Although the authors identified what they called 'bootstrapping molecules' (that is, those required to bootstrap the entire metabolism), these are typically not the same as those in our autocatalytic metabolite sets. For example, vitamin B12 cannot be synthesized by E. coli, thus it is a bootstrapping molecule sensu Romero and Karp, but is clearly not an autocatalytic metabolite by our definition. Moreover, the authors identified the set of bootstrap molecules with the preconception that certain compounds should necessarily be bootstrapping. Thus, it remains to be investigated in an unbiased way whether large metabolic networks contain autocatalytic components.

In the present study we test the idea that intermediary metabolisms of extant organisms are autocatalytic and we attempt to systematically identify autocatalytic sub-networks in different species using recently published high-quality metabolic network reconstructions. Our analysis not only shows that ATP is a universal obligatory autocatalytic compound, but also reveals species-specific differences in the set of autocatalytic molecules and variations in the structure of some autocatalytic sub-networks. These findings lend strong support to the view that replication in living systems is not restricted to macromolecules, but also involves small-molecule metabolism [1,8,9], albeit with limited heredity [13].

Results

Organism-specific autocatalytic compound sets

In three of the investigated networks ATP is not the only necessary autocatalytic compound. As expected, the largest number of autocatalytic metabolites is present in the photoautotrophic species Synechocystis sp., which requires only a limited set of inorganic food molecules for autotrophic growth. Analysis of the metabolic network of Synechocystis sp. strain PCC6803 reveals four additional autocatalytic sub-networks (Figure ​(Figure2).2). The Calvin cycle is clearly autocatalytic when the food set comprises only inorganic compounds: sugar is needed to fix CO₂ and produce more sugars. As different sugars are inter-convertible, any one of 138 different molecular species can fulfill this requirement. The Calvin cycle, however, does not remain autocatalytic upon inclusion of organic compounds in the food set (Additional data file 1). Furthermore, the biosynthesis of NAD⁺, coenzyme A (CoA) and tetrahydrofolate (THF) was also found to be autocatalytic in Synechocystis, irrespective of the food set (Table ​(Table1,1, Figure ​Figure22).

Open in a separate window

Figure 2

Autocatalytic synthesis of coenzymes in Synechocystis sp. (a) Coenzyme A. (b) NAD⁺. (c) Tetrahydrofolate. Autocatalytic metabolites involved in a reaction are indicated above the arrows. Dashed lines point to the reactions where the autocatalytic metabolites are involved in their own synthesis. See Additional data file 3 for the full names of metabolites.

Apparently, the reactions of the autocatalytic cycles identified in Synechocystis (with the exception of the Calvin cycle) are present in most of the studied organisms (Table ​(Table2):2): for instance, enzymes of CoA biosynthesis are found in all studied species. However, these metabolic routes do not necessarily operate as autocatalytic sub-networks in other organisms, either due to the possibility to uptake certain intermediates from the environment or due to the presence of enzymatic reactions leading to key intermediates. To further investigate this issue, we repeated the analysis of the E. coli network under a condition where only glucose and inorganic compounds were included in the food set (that is, a minimal medium). We found that in addition to ATP, NAD⁺, CoA and quinones also behave as autocatalytic compounds under this condition (Additional data file 1), demonstrating that uptake of certain intermediates from the environment can kick-start the corresponding autocatalytic sub-networks. For example, uptake of cysteine renders biosynthesis of CoA non-autocatalytic in E. coli. On the other hand, the fact that we did not identify THF as an autocatalytic metabolite in this species can best be explained by structural differences between the Synechocystis and E. coli networks: the possibility to convert AMP to GMP via IMP in the nucleotide salvage pathway of E. coli renders THF synthesis non-autocatalytic, even when folate is absent from the medium.

Table 2

The presence/absence of pathways involved in the biosyntheses of potentially autocatalytic cofactors

	NAD+	CoA	THF
Escherichia coli	+	+	+
Heliobacter pylori	+*	+	+
Staphylococcus aureus	P	+	+
Saccharomyces cerevisiae	P	+	+
Lactococcus lactis	P	+	+
Streptomyces coelicolor	+	+	+
Mycobacterium tuberculosis	+	+	+
Methanosarcina barkeri	P	+	P
Geobacter sulfurreducens	+*	+	+
Synechocystis	+	+	+
Minimal metabolism	P	P	P

Open in a separate window

A plus sign (+) indicates the biosynthetic route is fully present in the network; 'P' indicates the biosynthetic route is partially present in the network; asterisks indicate the biosynthetic route is slightly different from the one found in Synecocystis sp.

Equivalent compounds in autocatalytic coenzyme synthesis cycles

In line with the general considerations on autocatalytic cycles, we find that ADP and ATP are both autocatalysts since they are intermediates of the same autocatalytic cycle [11]. However, analysis of autocatalytic coenzyme synthesis in general is a challenge. Following the notation of Gánti [7], let the loaded form of a coenzyme be Q*, and the carrier molecule be Q (for example, the NADH:NAD⁺, acetyl-CoA:CoA, ATP:ADP pairs). Consider the following imaginary biosynthesis of this coenzyme (Figure ​(Figure3).3). Suppose external compounds A, X and F are provided to this cycle. This looks like an ordinary autocatalytic cycle with asymmetric branches leading to the copies of Q (similar to the topology of the reductive citric acid cycle). If so, B is also an autocatalyst. If B could be synthesized from other external materials, Q and B would cease being obligatory autocatalytic. Let, however, X be an intermediate of the whole network, thus an internal compound. In this case providing B does not settle the issue, because now we should look into the synthesis of X in other parts of the whole network. Consider, for example, the topology of the autocatalytic synthesis of NAD⁺ as shown in Figure S1 in Additional data file 1. It suggests that aspartate could replace NAD⁺, and seemingly the same applies to glutamate. However, launching the NAD⁺ synthesis cycle with glutamate also requires oxaloacetate, and the latter requires NAD⁺, hence glutamate cannot replace NAD⁺ in the obligatorily autocatalytic set, but aspartate can. Thus, identifying the sets of equivalent autocatalytic compounds embedded in large networks requires a systems-level approach. We provide a list of equivalent autocatalytic metabolite sets (that is, sets of autocatalytic compounds, usually forming reaction cycles, where the initial presence of any one compound is sufficient to render the others accessible) for different compound families and different organisms in Additional data file 1.

Open in a separate window

Figure 3

General scheme for the autocatalytic synthesis of a coenzyme. The coenzyme (carrier molecule) and its loaded form are denoted by Q and Q*, respectively. A, X and F are external compounds provided to the cycle. B is an intermediate in the cycle. G and Y are by-products of the cycle. B can be considered an autocatalyst if X is provided as an external compound. However, if X is an intermediate of the whole network (that is, an internal compound) then providing B does not necessarily launch the cycle because biosynthesis of X might require the presence of coenzyme Q*.

Discussion

We performed systems-level analysis of diverse metabolic networks to demonstrate that intermediary metabolisms contain obligatory autocatalytic biochemical cycles and, hence, qualify as replicators [6]. We found that intermediary metabolism is obligatorily autocatalytic for ATP (even if the system is able to uptake ATP via ATP-ADP exchange). Conceptually, our finding lends support to the view that a small but crucial part of inheritance is provided by the autocatalytic molecules of metabolism [1]. In sharp contrast to DNA-based replication, however, autocatalytic metabolic cycles are not modular replicators since replication is not based on the successive addition of modules, but rather proceeds progressively [13]. Moreover, although nucleic acids have practically unlimited potential to store information (the number of sequence types vastly exceeds the number of individuals in any realistic system), autocatalytic networks of metabolites can have very limited heredity only because the number of alternative types is likely to be small [13]. Our finding, that different forms of autocatalytic sub-networks are associated with NAD⁺ biosynthesis in two different organisms (Figures S3 and S6 in Additional data file 1) demonstrates that alternative forms of metabolic replicators is not a mere hypothetical possibility [33]. However, the evolutionary role of this variation remains questionable since rival variants should be present in the same population for competition and natural selection to take place. In contemporary systems the metabolic pathways are defined by highly specific catalytic activities provided by the genetically encoded enzymes. It is an open question whether alternative metabolic replicators can exist in their absence - today, the only known such replicator is the formose reaction [34,35] (producing sugars autocatalytically from formaldehyde). However, the formose reaction is non-informational [6]: no alternative cycles are known that would propagate themselves in a hereditary fashion.

The result that even heterotrophs contain metabolic replicators deserves special attention. For autotrophs the presence of clearly autocatalytic sub-networks as the Calvin cycle or the reductive citric acid cycle suggested that at least one of its intermediates, or some related compound, would be in the set of autocatalytic metabolites, given that there are no alternative synthetic routes, which is an empirical issue. We have settled this issue for Synechocystis in favor of sugar metabolism being truly autocatalytic in the autotrophic mode. In a similar vein, it will be interesting to examine in the future the autocatalytic compounds of autotrophic microbes running the reverse citric acid cycle (itself also being an autocatalytic cycle for CO₂ fixation). In contrast, the presence of metabolic replicators in heterotrophs may seem less obvious, since they consume organic compounds in the food set - yet we find at least ATP to be always such a replicator, and occasionally other coenzymes also.

Could the presence of coenzyme replicators be an ancestral feature of intermediary metabolism? As King [36] observed a while ago, the biosynthesis of coenzymes seems to be auto- and cross-catalytic and this may be partly due to ancient metabolic history. This also seems to hold partially in our analysis: for example, THF and NAD⁺ are needed in CoA synthesis (Figure ​(Figure2).2). Nucleotide coenzymes may well be molecular fossils [37] from an RNA world [38]. Considering the fact that they participate in so many reactions and that it would be very hard to replace them after the evolutionary build-up of the enzymatic system, their auto- and cross-catalytic nature indeed speaks for their primitive ancestry in metabolism [33,39]. The fact that comparative analysis of reduced endosymbiont genomes does not suggest coenzyme synthesis in top-down-derived minimal organisms [40] is no argument against such ancestry. If we accept that most of the coenzymatic biochemical reactions cannot be run at an acceptable speed for a primitive cell without the coenzymes, then the only remaining option is a heterotrophic uptake of the precursors of these coenzymes (compare [16]). But unless all the coenzyme precursors (vitamins) were abiogenically synthesized in an environment chemically different from the primitive protocells, the latter may have just been running the needed reactions inside. Early coenzyme synthesis, just as primitive metabolism in general [41], may have been closer to primordial chemistry rather than modern biochemistry. The fact that suggestions for reconstructed minimal cells thriving under nutrient-rich conditions do not contain coenzyme synthesis does not imply that the last universal common ancestor did not have coenzyme synthesis. Indeed, a recent estimate of the gene content of the last universal common ancestor reveals that it might have possessed a fairly complex genome similar to those of free-living prokaryotes, including genes encoding certain enzymatic steps of NAD⁺, CoA and THF biosynthesis [42]. With time, an originally autocatalytic metabolic compound may cease to remain such, as novel routes of synthesis, based on a reduced set of autocatalytic molecules, are discovered by genetic evolution. If this option is not available, the only solution is to evolve an alternative, still autocatalytic, synthetic pathway (analogous to the replacement of one enzyme, taking part in DNA replication, by another).

Our finding that even a minimal metabolism is autocatalytic at the level of small molecules has important implications for attempts to design a synthetic cell. Most efforts to build an artificial self-reproducing system from scratch have focused on constructing simple chemical supersystems capable of template replication and membrane growth, but lacking a metabolic subsystem (see [43] for a review). However, future aims to design a synthetic cell with complex intermediary metabolism should incorporate our findings on the existence of autocatalytic compounds. Moreover, future studies should address the question of whether gene regulatory and signaling networks contain autocatalytic components analogous to those found in metabolism (for example, the product of a positive feedback loop [5]). Thus, an extension of our network-based approach could be used to identify the minimal set of cellular network components that should possibly be provided to kick-start an artificial cell.

Conclusion

The current study constitutes, to our knowledge, the first systematic search for replicators embedded in large biochemical networks. Although parts of metabolism that are autocatalytic in isolation (for example, Calvin cycle, glycolysis) have been put forward previously, it remained unknown whether these cycles operate in an obligatorily autocatalytic manner when embedded in larger networks. Our analysis of the small molecule metabolism of 10 living organisms and an inferred minimal metabolism suggests that all metabolic networks have at least one universal autocatalytic molecule, ATP (or equivalent compounds). Conceptually, this finding supports the view that a small but important part of inheritance is provided by the set of autocatalytic compounds of intermediary metabolism. Although ATP appears to be the only universal autocatalytic metabolite, other, organism-specific autocatalytic molecules have been identified in the forms of nucleotide cofactors (such as CoA, NAD⁺ and THF) and sugars or sugar-containing compounds (in the autotrophic metabolism of a photosynthetic bacterium). Importantly, the metabolic pathways associated with these autocatalytic nucleotide cofactors are present in many organisms, but they do not necessarily operate in an autocatalytic manner, as the autocatalytic compounds can be synthesized from food molecules or with the help of alternative pathways. This finding clearly underlines the need for a systems-level approach to identify obligate replicators embedded in large metabolic networks. Our work also has relevance for attempts to create synthetic cells, as some of these autocatalytic molecules will presumably be needed to be added to the system as the system cannot synthesize them without their initial presence.

Materials and methods

Abbreviations

CoA, coenzyme A; THF, tetrahydrofolate.

Authors' contributions

ESz conceived the idea for the study. All authors contributed to the design and planning of the research. BP performed the Flux Balance Analyses, and ÁK performed the scope analyses and network curations. All authors were involved in writing the manuscript. All authors approved the final version of the manuscript.

Additional data files

The following additional data are available. Additional data file 1 includes details of the scope analysis for each organism and information on the reconstruction of the Synechocystis network. The figures (S1-S9) show the autocatalytic production of the identified metabolites. The table (S2) lists the relevant statistics for each metabolic network analyzed. Additional data file 2 is an Excel table presenting the list of producible metabolites for each metabolic network. If the network reconstruction used abbreviated names for the metabolites, then the abbreviations are also included for ease of comparisons with the reconstruction. Additional data file 3 is an Excel table describing the metabolic reconstruction of Synechocystis sp PCC 6803. The separate worksheets lists are: 1, the metabolites involved in the metabolic reconstruction, with their abbreviations and identifier used in the MetaCyc database; 2, the reactions with reaction ID, pathway, EC number and references; 3, the metabolites that cannot be produced in the network (dead end metabolites); 4, the reactions that were left out from the metabolic network, and the reason for the exclusion; and 5, the references and notes for the worksheets.

Supplementary Material

Acknowledgements

References