A formal model of autocatalytic sets emerging in an RNA replicator system
 Wim Hordijk^{1}Email author and
 Mike Steel^{2}
DOI: 10.1186/1759220843
© Hordijk and Steel; licensee Chemistry Central Ltd. 2013
Received: 15 November 2012
Accepted: 29 January 2013
Published: 26 February 2013
Abstract
Background
The idea that autocatalytic sets played an important role in the origin of life is not new. However, the likelihood of autocatalytic sets emerging spontaneously has long been debated. Recently, progress has been made along two different lines. Experimental results have shown that autocatalytic sets can indeed emerge in real chemical systems, and theoretical work has shown that the existence of such selfsustaining sets is highly likely in formal models of chemical systems. Here, we take a first step towards merging these two lines of work by constructing and investigating a formal model of a real chemical system of RNA replicators exhibiting autocatalytic sets.
Results
We show that the formal model accurately reproduces recent experimental results on an RNA replicator system, in particular how the system goes through a sequence of larger and larger autocatalytic sets, and how a cooperative (autocatalytic) system can outcompete an equivalent selfish system. Moreover, the model provides additional insights that could not be obtained from experiments alone, and it suggests several experimentally testable hypotheses.
Conclusions
Given these additional insights and predictions, the modeling framework provides a better and more detailed understanding of the nature of chemical systems in general and the emergence of autocatalytic sets in particular. This provides an important first step in combining experimental and theoretical work on autocatalytic sets in the context of the orgin of life.
Background
Recently, significant new experimental results on spontaneous network formation among cooperative RNA replicators were reported [1]. These results continue and strengthen a line of ongoing work on creating autocatalytic sets in real chemical systems [2–5]. Moreover, they show the plausibility and viability of the idea of autocatalytic sets, especially in the context of the origin of life, as already developed in various forms several decades ago [6–12].
However, such chemical experiments, important as they are, are difficult, costly, and timeconsuming to perform. In contrast, in our own work we have developed a theoretical framework of autocatalytic sets, which we consider a necessary condition for the origin of life, that can be studied computationally and mathematically [13–20]. This framework has provided many important insights into the emergence and structure of autocatalytic sets in its own right, and is considered to provide theoretical support for experimental observations [1, 21].
The formal framework has, so far, mostly been applied to an abstract model of a chemical reaction system known as the binary polymer model. Even though this model already has a fair amount of chemical realism (for example, some recent experiments are an almost literal chemical implementation of the binary polymer model [5]), a direct application to a real chemical system was still lacking. Here, we construct and analyze a formal model version of the recent experimental RNA replicator system [1], and show how this results in:

an accurate reproduction of experimental results,

the correction of a misinterpretation of some of the original results,

additional results and insights that could not be obtained from the experiments alone, and

testable predictions about the behavior of the chemical system.
With this, we claim that the formal framework can be applied directly and meaningfully to real chemical systems, generating additional insights and predictions that are hard to obtain from experiments alone, thus providing a better understanding of real (bio)chemical systems. This represents an important first step towards merging experimental and theoretical work on autocatalytic sets.
Chemical reaction systems and autocatalytic sets
 1.
reflexively autocatalytic (RA): each reaction $r\in {\mathcal{R}}^{\prime}$ is catalyzed by at least one molecule type involved in ${\mathcal{R}}^{\prime}$, and
 2.
foodgenerated (F): all reactants in ${\mathcal{R}}^{\prime}$ can be created from the food set F by using a series of reactions only from ${\mathcal{R}}^{\prime}$ itself.
A more formal (mathematical) definition of RAF sets is provided in [14, 17], including an efficient algorithm for finding RAF sets in a general CRS. It was shown (using the binary polymer model) that RAF sets are highly likely to exist, even for very moderate levels of catalysis (between one and two reactions catalyzed per molecule, on average) [14–16], and that this result still holds when a more realistic “templatebased” form of catalysis is used [17, 18].
The RAF sets that are found by the RAF algorithm are called maximal RAF sets (maxRAFs). However, it was shown that a maxRAF can often be decomposed into several smaller subsets which themselves are RAF sets (subRAFs) [19]. If such a subRAF cannot be reduced any further without losing the RAF property, it is referred to as an irreducible RAF (irrRAF). The existence of multiple autocatalytic subsets can actually give rise to an evolutionary process [22], and the emergence of larger and larger autocatalytic sets over time [19].
A model of the RNA replicator system
The existence of an RNA world is considered a crucial step in the origin of life [23, 24]. However, the transition from prebiotic chemistry to this stage of life is still poorly understood. An autocatalytic set of cooperative molecules is likely to have been required for this transition to happen. In recent experiments [1], it was shown that particular mixtures of RNA fragments that selfassemble into selfreplicating ribozymes spontaneously form cooperative catalytic networks. Furthermore, when such cooperative networks compete directly with selfish cycles, the former grow faster, indicating an intrinsic ability of collections of RNA molecules to evolve greater complexity through cooperation. These RNA replictor experiments thus highlight the advantages of cooperative behavior (as autocatalytic sets), even at the early molecular stages of life [1].
The main idea behind this RNA replicator system is the assembly (ligation) of ribozymes (catalytic RNA molecules) from two smaller RNA fragments. These ribozymes can then catalyze the assembly of other ribozymes (or in some cases their own assembly). Which ribozymes catalyze which assembly reactions is determined by one specific nucleotide in the “guide sequence” of the (potential) catalyst and one other specific nucleotide in the “target sequence” of a reactant. If these two nucleotides are each other’s base pair complement, then the given ribozyme can catalyze the given reaction. Starting with RNA fragments with different nucleotides in these guide and target sequences, a mixture of auto and crosscatalytic RNA replicators evolves over time (by means of the assembly reactions taking place), in which cooperative networks (autocatalytic sets) form spontaneously [1].
The ribozymes in this system are labeled Mj N, where M denotes the specific nucleotide (A, C, G, or U) in the guide sequence of an RNA molecule, and N denotes the specific nucleotide in its target sequence [1]. The value of j denotes the specific location where the ribozyme was assembled from two smaller RNA fragments (there are three possible locations in the original experiments). However, as in some of the experimental results [1], for the purposes of looking for autocatalytic sets, this value can be ignored. So, in total there are 4×4=16 possible ribozymes Mj N. For example, Gj A and Aj C are two such ribozymes.
The reaction set X in the formal model of this RNA replicator system contains these 16 Mj N ribozymes plus the smaller RNA fragments from which they are assembled. In fact, these RNA fragments are the subset of molecules that forms the food set F. In the model, we simply lump these fragments together into just two food molecule types (i.e., each ribozyme is the assembly of two “generic” fragments). For most of the results shown here this suffices, although a refinement will be made later on in one of the computational experiments (below).
The reaction set $\mathcal{R}$ in the model consists of the 16 assembly reactions that create the ribozymes from food molecules. For simplicity, the notation Mj N will denote both a molecule type (ribozyme) as well as the reaction that created it. Finally, the catalysis set C consists of all molecule/reaction pairs where the nucleotide M in the guide sequence of a molecule is the basepair complement of the nucleotide N in the target sequence of a reaction. For example, molecule Gj A can catalyze the reaction that produces molecule Aj C, since G and C are complementary. The catalysis set C thus consists of all such matching pairs (16×4=64 in total).
The full mathematical definition of the CRS $Q=\{X,\mathcal{R},C\}$ of the RNA replicator system is thus as follows:

F = { f_{1}, f_{2}}

X = F ∪ {Mj N  M,N ∈ {A,C,G,U}}

= {Mj N  M,N ∈ {A,C,G,U}}$\mathcal{R}$

C = {(M_{1}j N_{1}, M_{2}j N_{2})  M _{ i },N _{ i } ∈ {A,C,G,U}, i∈{1,2}, and (M_{1},N_{2}) ∈ {(A,U),(C,G),(G,C),(U,A)}}
Note that since the reactants in all 16 reactions in $\mathcal{R}$ are food molecules, every subset ${\mathcal{R}}^{\prime}\subseteq \mathcal{R}$ is automatically foodgenerated (F). Therefore, identifying possible RAF sets in this system only requires checking the (RA) part of the definition. We now analyze this model in detail and compare the results with those from the original experiments [1].
Results and discussion
The existence of RAF sets
Applying the RAF algorithm to the reaction set $\mathcal{R}$ (as defined above), returns the set $\mathcal{R}$ itself. In other words, the system as a whole (all 16 reactions) forms a maximal RAF set. Moreover, this would be expected, given the extent of catalysis, even if the actual assignment of catalysis was randomized. In fact, under such a (random) model the probability that $\mathcal{R}$ forms an RAF is approximately (1−e^{−4})^{16} = 74%. This can be derived as follows.
which, for C=64 and $\left\mathcal{R}\right=16$ (as in the RNA replicator model) evaluates to (1 − e^{−4})^{16}, as claimed. Notice that Eqn. (4) is essentially independent of X (the number of molecules) provided this is not too small.
The structure of RAF sets
This Hasse diagram contains 68 nodes, i.e., there are 68 possible subRAFs in the given 7reaction RAF set. Note that there are 2^{7}=128 possible subsets of a set of seven elements. So, more than half of these actually form RAF sets themselves, which shows how “rich” and diverse the RNA replicator system really is. In fact, if the same ratio (about half) holds for the full 16reaction maximal RAF set, then we can expect more than 2^{16}/2≈32,000 nodes (subRAFs) in the Hasse diagram of the maxRAF, i.e., far too many to visualize in a meaningful way.
The edges in the Hasse diagram of an RAF set show the many possible ways in which the full RAF set can be built up from smaller subsets, or, in other words, how RAF sets can emerge and evolve [19, 22]. Which of these trajectories is actually followed depends on, for example, initial conditions and stochastic events such as “spontaneous” (uncatalyzed) reactions. As shown above, in the original experiment the system went through a stage of a particular 7reaction RAF set, which then grew to an 11reaction RAF set. However, what the Hasse diagram suggests, given the many possible subRAFs and ways of combining them into larger RAFs, is that when the experiment is repeated, most likely a different trajectory will be followed. This is a testable prediction that follows directly from the formal model and its results.
The emergence of RAF sets
Performing the actual chemical experiments in a laboratory is costly and timeconsuming. However, we can use the formal model to simulate molecular flow on the reaction network [20]. Using the wellknown Gillespie algorithm [25, 26], we performed such simulations on the full 16reaction model, starting with an initial supply of food molecules (RNA fragments) only. The Gillespie algorithm is a stochastic (Monte Carlo) simulation procedure for modeling the transient behavior of molecular systems. It is an alternative method to solving the reaction rate equations numerically, by instead using the reaction rates as “reaction probabilities”. Based on these probabilities and the current molecular concentrations, at each step in the simulation, random variables are generated that determine the time and type of the next reaction event.
With these results we do not intend to claim that the simulation is an exact reproduction of the original experiment, where at regular intervals 10% of the current solution was transferred to a new solution of fresh RNA fragments [1]. Although we can include such “transfer” steps in our simulations as well, we mainly wish to show that the overall process of going through larger and larger subRAFs is accurately reproduced by the model, and to confirm that in each repetition of the experiment (simulation), a different trajectory is indeed followed, as suggested by the Hasse diagram.
The advantage of RAF sets
There is a simple explanation for the difference between cooperation and selfishness in this model. To get the cooperative system (RAF set) going, only one of the three (green) reactions has to happen “spontaneously”, i.e., uncatalyzed; this is always possible, but at a lower rate (by a factor c) than a ribozymecatalyzed reaction. This happens around time point 0.14 in the simulation shown in Figure 5. However, in the selfish system, all three (red) reactions have to happen spontaneously at least once to get the full system going. This results in an almost three times longer waiting time on average. In the simulation shown in Figure 5, the third spontaneous (uncatalyzed) red reaction happens around time point 0.35. By that time, however, the (green) RAF set has already built up enough “momentum” to outcompete the selfish (red) system, i.e., enough molecules of types E1, E2, and E3 are already around to increasingly catalyze each other’s assembly. However, due to the stochastic nature of the system (having to wait for uncatalyzed reactions to happen), occasionally the selfish (red) system gets a headstart (with low probability) and outcompetes the cooperative (green) system.
Finally, there is also an advantage for the RAF set in terms of robustness. Suppose that at some point all molecules of type E3 and S3 are removed from the system, and a new supply of their food molecules (f_{5} and f_{6}) is provided. This happens around time point 2 in Figure 5, at the sudden dip in the concentrations. Since there are still E1 and E2 molecules present, the (green) RAF set quickly recovers without any delay. However, the (red) selfish system again has to wait until reaction S3 happens uncatalyzed at least once (which still has not happened by time point 3). Clearly, this suggests that RAF sets are more robust than selfish systems against perturbations, another prediction that can be tested with real chemical experiments.
Conclusions
We have taken the chemical RNA replicator system described recently [1] and formalized and investigated it within our mathematical RAF framework. As the experimental results showed, autocatalytic sets seem to form spontaneously in this chemical system [1]. The existence of RAF sets in this system is verified by the formal model. In fact, many of the experimental results are accurately reproduced by the model, such as the emergence of larger and larger RAF sets over time, and the advantage of cooperative systems over selfish systems when they compete for the same resources. Of course there are many refinements and additional details that can be added to the current model, but even in its most basic form it already captures the main structural and dynamical properties of the real chemical system.
Moreover, the modeling approach provides several additional results and insights. First, it allows for a correction in the misinterpretation of some of the original experimental results. Second, it shows the “richness” of the RNA system in terms of the many subRAFs and ways they can be combined into larger subRAFs (as visualized by the Hasse diagram). Third, this “richness” suggests the testable prediction that each repetition of the experiment will, most likely, follow a different trajectory towards a realization of the maximal RAF set, which is confirmed computationally by the molecular flow simulations presented here. Lastly, it provides a simple explanation for the advantage of RAF sets over selfish systems, together with another testable prediction that RAF sets are more robust than selfish systems against perturbations.
In conclusion, we have shown that the formal RAF framework can be directly and meaningfully applied to real chemical systems, generating additional insights that are hard or even impossible to obtain from experiments alone, and thus provides a more detailed understanding of the nature of chemical systems in general and the emergence of autocatalytic sets in particular. This forms an important and much needed first step towards merging experimental and theoretical lines of work on autocatalytic sets in the context of the origin of life.
Declarations
Authors’ Affiliations
References
 Vaidya N, Manapat ML, Chen IA, XulviBrunet R, Hayden EJ, Lehman N: Spontaneous network formation among cooperative RNA replicators. Nature 2012, 491: 72–77. 10.1038/nature11549View ArticleGoogle Scholar
 Sievers D, von Kiedrowski G: Selfreplication of complementary nucleotidebased oligomers. Nature 1994, 369: 221–224. 10.1038/369221a0View ArticleGoogle Scholar
 Ashkenasy G, Jegasia R, Yadav M, Ghadiri MR: Design of a directed molecular network. PNAS 2004,101(30):10872–10877. 10.1073/pnas.0402674101View ArticleGoogle Scholar
 Hayden EJ, von Kiedrowski G, Lehman N: Systems chemistry on ribozyme selfconstruction: Evidence for anabolic autocatalysis in a recombination network. Angewandte Chemie International Edition 2008, 120: 8552–8556.View ArticleGoogle Scholar
 Taran O, Thoennessen O, Achilles K, von Kiedrowski G: Synthesis of informationcarrying polymers of mixed sequences from double stranded short deoxynucleotides. J Syst Chem 2010, 1: 9. 10.1186/1759220819View ArticleGoogle Scholar
 Kauffman SA: Cellular homeostasis, epigenesis and replication in randomly aggregated macromolecular systems. J Cybern 1971, 1: 71–96. 10.1080/01969727108545830View ArticleGoogle Scholar
 Kauffman SA: Autocatalytic sets of proteins. J Theor Biol 1986, 119: 1–24. 10.1016/S00225193(86)800479View ArticleGoogle Scholar
 Kauffman SA: The Origins of Order. New York: Oxford University Press; 1993.Google Scholar
 Eigen M, Schuster P: The hypercycle: a principle of natural selforganization. Part A: Emergence of the hypercycle. Naturwissenschaften 1977, 64: 541–565. 10.1007/BF00450633View ArticleGoogle Scholar
 Dyson FJ: A model for the origin of life. J Mol Evol 1982, 18: 344–350. 10.1007/BF01733901View ArticleGoogle Scholar
 Rosen R: Life Itself. New York: Columbia University Press; 1991.Google Scholar
 Gánti T: Biogenesis itself. J Theor Biol 1997, 187: 583–593. 10.1006/jtbi.1996.0391View ArticleGoogle Scholar
 Steel M: The emergence of a selfcatalysing structure in abstract originoflife models. Appl Math Lett 2000, 3: 91–95.View ArticleGoogle Scholar
 Hordijk W, Steel M: Detecting autocatalytic, selfsustaining sets in chemical reaction systems. J Theor Biol 2004,227(4):451–461. 10.1016/j.jtbi.2003.11.020View ArticleGoogle Scholar
 Mossel E, Steel M: Random biochemical networks: The probability of selfsustaining autocatalysis. J Theor Biol 2005,233(3):327–336. 10.1016/j.jtbi.2004.10.011View ArticleGoogle Scholar
 Hordijk W, Hein J, Steel M: Autocatalytic sets and the origin of life. Entropy 2010,12(7):1733–1742. 10.3390/e12071733View ArticleGoogle Scholar
 Hordijk W, Kauffman SA, Steel M: Required levels of catalysis for emergence of autocatalytic sets in models of chemical reaction systems. Int J Mol Sci 2011,12(5):3085–3101.View ArticleGoogle Scholar
 Hordijk W, Steel M: Predicting templatebased catalysis rates in a simple catalytic reaction model. J Theor Biol 2012, 295: 132–138.View ArticleGoogle Scholar
 Hordijk W, Steel M, Kauffman S: The structure of autocatalytic sets: Evolvability, enablement, and emergence. Acta Biotheoretica 2012,60(4):379–392. 10.1007/s1044101291651View ArticleGoogle Scholar
 Hordijk W, Steel M: Autocatalytic sets extended: Dynamics, inhibition, and a generalization. J Syst Chem 2012, 3: 5. 10.1186/1759220835View ArticleGoogle Scholar
 Martin W, Russel MJ: On the origin of biochemistry at an alkaline hydrothermal vent. Philos Trans R Soc B 2007, 362: 1887–1925. 10.1098/rstb.2006.1881View ArticleGoogle Scholar
 Vasas V, Fernando C, Santos M, Kauffman S, Sathmáry E: Evolution before genes. Biol Direct 2012, 7: 1. 10.1186/1745615071View ArticleGoogle Scholar
 Gilbert W: The RNA world. Nature 1986, 319: 618.View ArticleGoogle Scholar
 Joyce GF: RNA evolution and the origins of life. Nature 1989, 338: 217–224. 10.1038/338217a0View ArticleGoogle Scholar
 Gillespie DT: A general method for numerically simulating the stochastic time evolution of coupled chemical reactions. J Comput Phys 1976, 22: 403–434. 10.1016/00219991(76)900413View ArticleGoogle Scholar
 Gillespie DT: Exact stochastic simulation of coupled chemical reactions. J Phys Chem 1977,81(25):2340–2361. 10.1021/j100540a008View ArticleGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.