1. How Natural Selection could Evolve Metabolism. Chrisantha Fernando School of Computer Science University of Birmingham, UK San Sebastian, September 2006 Chrisantha Fernando School of Computer Science University of Birmingham, UK San Sebastian, September 2006

2. The Problem Defined:  The multiple source hypothesis proposes that a great variety of organic synthesis routes could have produced organic molecular variety.  Eg…..  The multiple source hypothesis proposes that a great variety of organic synthesis routes could have produced organic molecular variety.  Eg…..

3.  A.I. Oparin & J.B.S. Haldane (1924,1929). UV light energy, Fox and Dose, Folsome, etc.. Amino acid polym.  G. Wachtershauser. FeS/H 2 S reducing power produces COO -, -S -, -COS -, neg charged metabolism on surface.  S. Miller. Electrical discharges make Aas, aldehydes, etc.. in reducing conditions  C. de Duve. Thioester metabolism on surfaces.  H. Morowitz. Reverse Citric acid cycle on mineral surfaces, (autocatalytic step so far a problem).  Decker, Ganti. Formose cycle feeding on Formaldehyde and other aldehydes formed from perhaps a Miller like reaction or CO and H 2 O + light at pressure (Hazen et al).  Chyba and Astrobiology. Organics from space, meteoric origin of organics.  Tommy Gold. Abiotic synthesis of hydrocarbons! (Controversial).  A.I. Oparin & J.B.S. Haldane (1924,1929). UV light energy, Fox and Dose, Folsome, etc.. Amino acid polym.  G. Wachtershauser. FeS/H 2 S reducing power produces COO -, -S -, -COS -, neg charged metabolism on surface.  S. Miller. Electrical discharges make Aas, aldehydes, etc.. in reducing conditions  C. de Duve. Thioester metabolism on surfaces.  H. Morowitz. Reverse Citric acid cycle on mineral surfaces, (autocatalytic step so far a problem).  Decker, Ganti. Formose cycle feeding on Formaldehyde and other aldehydes formed from perhaps a Miller like reaction or CO and H 2 O + light at pressure (Hazen et al).  Chyba and Astrobiology. Organics from space, meteoric origin of organics.  Tommy Gold. Abiotic synthesis of hydrocarbons! (Controversial).

4. Ingredients but no Recipe?  Given a wide range of ingredients to choose from, what process resulted in the origin of metabolism?  We seek  I. The minimal autopoetic chemical organization capable of the “recursive generation of functional constraints”, initially in the absence of template replication.  II. The properties of the ingredients, and the recipe, required to produce a metabolism, ultimately capable of evolving template replication.  Given a wide range of ingredients to choose from, what process resulted in the origin of metabolism?  We seek  I. The minimal autopoetic chemical organization capable of the “recursive generation of functional constraints”, initially in the absence of template replication.  II. The properties of the ingredients, and the recipe, required to produce a metabolism, ultimately capable of evolving template replication.

5. Selection and Drift in Phase Separated Spots  Operin suggested natural selection between coacervates and Dyson modeled the effect of drift in such systems and the probability of transition from simpler to more ‘complex’ metabolisms.  Thought experiments and models of natural selection between coacervate-like entities.  Replication rate is correlated with steady-state light absorption due to: direct growth enhancement and an energy constraint on information transmission.  Novelties arise through chemical ‘avalanches’.  Operin suggested natural selection between coacervates and Dyson modeled the effect of drift in such systems and the probability of transition from simpler to more ‘complex’ metabolisms.  Thought experiments and models of natural selection between coacervate-like entities.  Replication rate is correlated with steady-state light absorption due to: direct growth enhancement and an energy constraint on information transmission.  Novelties arise through chemical ‘avalanches’.

6. Assumptions  Assumption 1: Phase separated spots form on a liquid surface. They proliferate due to supply of oily ‘food’ material that perhaps falls in rain. They divide by agitation and are subject to random loss, with geological recycling of their ‘food’ back to the atmosphere.

8. The First Autotrophic Unit is at the Level of a Geophysical Cycle Chemical recycling Physical recycling

9.  Assumption 2: Within the oil phase, there exists a potential light absorbing reaction that can re-form abbb and a high energy molecule ba. But to start with there is no abb so this reaction cannot actually take place.  Assumption 2.1. In addition, light absorbing products may be much longer, and contribute more to spot growth than non-driven reactions. Long molecules stay in spots better than short ones.  Assumption 2: Within the oil phase, there exists a potential light absorbing reaction that can re-form abbb and a high energy molecule ba. But to start with there is no abb so this reaction cannot actually take place.  Assumption 2.1. In addition, light absorbing products may be much longer, and contribute more to spot growth than non-driven reactions. Long molecules stay in spots better than short ones. abb + abb + light ----> ba + abbb

10. 1.A light absorbing reaction may produce molecules that make the spot grow faster directly. or longer molecule

11.  Assumption 3: Spots able to absorb abbb more quickly obviously proliferate faster.  If a spot contains high energy species that react with abbb molecules in chemical reactions that produce spot localized contents, then abbb absorption rate can be increased greatly (irreversibility of this reaction is ++ favoured).  Assumption 3: Spots able to absorb abbb more quickly obviously proliferate faster.  If a spot contains high energy species that react with abbb molecules in chemical reactions that produce spot localized contents, then abbb absorption rate can be increased greatly (irreversibility of this reaction is ++ favoured).

12. 2. High energy reactants can shift the equilibrium of ‘food’ promoting growth.

13.  Assumption 4: Novel species are produced by rare bimolecular rearrangement reactions between existing species, e.g. abbb and other food species, e.g.  abbb + abbb -(v. low flux ~ 0)-> ba + bbbbba  Both ba and bbbbba are novel species, present at initially very low concentration e.g. 0.0000001M, in actual fact, much less than this, i.e. 1 molecule!  Assumption 4: Novel species are produced by rare bimolecular rearrangement reactions between existing species, e.g. abbb and other food species, e.g.  abbb + abbb -(v. low flux ~ 0)-> ba + bbbbba  Both ba and bbbbba are novel species, present at initially very low concentration e.g. 0.0000001M, in actual fact, much less than this, i.e. 1 molecule!

14.  Assumption 5: Once a novel species is produced, it may react in spontaneous high flux reactions with a proportion of other species,e.g. 0.01%, 1%, 5% etc…  However, by definition such novel reactions must have insignificant reverse rates if their product is any existing species, otherwise, reverse flux through this reaction would have been observed before the novel rare species arose, and the novel species would therefore not have been novel!  Assumption 5: Once a novel species is produced, it may react in spontaneous high flux reactions with a proportion of other species,e.g. 0.01%, 1%, 5% etc…  However, by definition such novel reactions must have insignificant reverse rates if their product is any existing species, otherwise, reverse flux through this reaction would have been observed before the novel rare species arose, and the novel species would therefore not have been novel!

16.  Assumption 6: For each novel species produced in high flux reactions, the high flux reactions it takes part in are again calculated, resulting in a potential ‘avalanche’ of novel reactions. Avalanche size distribution will depend on the proportion P each species reacts with.

17. aaa + b bbbbbaa + a 1. 2. 3. Determining the Avalanche.

18. Energy for Information

19. 3. Energy required to replicate babb, (high k 2 ). 4. Energy required for babb to have influence (high k 3 ).

20.  Assumption 7 : Division occurs due to absorption of abbb at regular intervals.  Replication rate is increased if abbb undergoes chemical reactions with the material in the spot to produce spot phase material, (longer polymers being more likely to be spot-phase). Division occurs when the volume reaches some noisy threshold.  One definition of fitness is time to reach this threshold. In the simulation, fitness could be defined as the total mass of material absorbed into the spot in fixed time T.  Assumption 7 : Division occurs due to absorption of abbb at regular intervals.  Replication rate is increased if abbb undergoes chemical reactions with the material in the spot to produce spot phase material, (longer polymers being more likely to be spot-phase). Division occurs when the volume reaches some noisy threshold.  One definition of fitness is time to reach this threshold. In the simulation, fitness could be defined as the total mass of material absorbed into the spot in fixed time T.

21. Low energy vs. high energy spots.  Spots growing by incorporation of low energy material are disadvantaged because,  Absorption of abbb by chemical reactions is thermodynamically unfavorable.  Unable to create the thermodynamically unfavorable longer polymers that greatly benefit spot growth.  Unable to sustain autocatalysis required for the ‘channeling’ of the reaction network.  High energy autocatalytic particles can more effectively channel the reaction network than low energy autocatalytic particles.  Spots growing by incorporation of low energy material are disadvantaged because,  Absorption of abbb by chemical reactions is thermodynamically unfavorable.  Unable to create the thermodynamically unfavorable longer polymers that greatly benefit spot growth.  Unable to sustain autocatalysis required for the ‘channeling’ of the reaction network.  High energy autocatalytic particles can more effectively channel the reaction network than low energy autocatalytic particles.

22. Selection at the Level of the Geophysical Unit?  Low energy spot material requires more energy for recycling back to abbb by the autotrophic geophysical metabolism.  If [abbb] in rain is limited by recycling rate, then geophysical units with greater abbb production mediate greater energy flux, more replication, and hence more natural selection.  A spot that produces abbb not only benefits itself, but benefits all other spots in the geophysical unit.  Low energy spot material requires more energy for recycling back to abbb by the autotrophic geophysical metabolism.  If [abbb] in rain is limited by recycling rate, then geophysical units with greater abbb production mediate greater energy flux, more replication, and hence more natural selection.  A spot that produces abbb not only benefits itself, but benefits all other spots in the geophysical unit.

23. Preliminary Models.  In the models so far I did not have this elaborate geophysical unit conception, but attempted to directly select for increased steady-state light absorption in spots.  The artificial selection experiment is intended to capture some of the dynamics just presented.  In the models so far I did not have this elaborate geophysical unit conception, but attempted to directly select for increased steady-state light absorption in spots.  The artificial selection experiment is intended to capture some of the dynamics just presented.

24. Aim of the Models  I use a 1:1 GA to artificially seelct ‘spots’ satisfying a fitness function.  What chemical organizations are produced, e.g. autocatalytic cycles etc..?  What is the ratio of harmful reactions to beneficial reactions as a function of the properties of chemical variation, the food set and the fitness function?  I use a 1:1 GA to artificially seelct ‘spots’ satisfying a fitness function.  What chemical organizations are produced, e.g. autocatalytic cycles etc..?  What is the ratio of harmful reactions to beneficial reactions as a function of the properties of chemical variation, the food set and the fitness function?

25. Main Findings  Experiments selecting for high steady-state light absorption show adaptive avalanches v. rare because.  Probability of growth of novel low concentration product is small.  Probability that novel product has beneficial spot-level effect is small.  Bimolecular rearrangement reactions tend to produce divergent networks (not closed), not such a problem if autocatalytic growth of novel product is occurring.  Increasing food set size increases probability of adaptations.  Early experiments selecting for abbb uptake rate: Light absorption reaction not always utilized.  Adaptive frequency decreases over evolution?  Experiments selecting for high steady-state light absorption show adaptive avalanches v. rare because.  Probability of growth of novel low concentration product is small.  Probability that novel product has beneficial spot-level effect is small.  Bimolecular rearrangement reactions tend to produce divergent networks (not closed), not such a problem if autocatalytic growth of novel product is occurring.  Increasing food set size increases probability of adaptations.  Early experiments selecting for abbb uptake rate: Light absorption reaction not always utilized.  Adaptive frequency decreases over evolution?

26. The Chemistry  Bimolecular rearrangement reactions of binary atoms, e.g.  abbb + ba abb + abb  Each molecule has free energy of formation, G.  Two types of reaction.  Irreversible (and reversible) exogonic reactions (heat producing)  Irreversible endogonic reaction (light absorbing)  When generating novel reactions ensure they are always spontaneous.  Bimolecular rearrangement reactions of binary atoms, e.g.  abbb + ba abb + abb  Each molecule has free energy of formation, G.  Two types of reaction.  Irreversible (and reversible) exogonic reactions (heat producing)  Irreversible endogonic reaction (light absorbing)  When generating novel reactions ensure they are always spontaneous.

27. k f = 0.01*e -dG/RT k b = 0.01, dG = (G products - G reactants ) R = gas constant, T = 300K The Initial Network

28. Artificial Selection

29. Control Experiment  Information (i.e. modification of reaction structure) implicit. Now reactions can just be formed and removed randomly! No need for the evolution of “engram autocatalysis”.  Autocatalytic reaction network evolved.  High energy hub molecules drive reactions, e.g. babb.  Information (i.e. modification of reaction structure) implicit. Now reactions can just be formed and removed randomly! No need for the evolution of “engram autocatalysis”.  Autocatalytic reaction network evolved.  High energy hub molecules drive reactions, e.g. babb.

32.  Fitness is largely unaffected if the network is initialized with 100mM abbb plus any one of the following species at 0.1mM; ba, ab, abb, babb, babbb, bbbbbba.  However, if the network is initialized with 100mM abbb alone, or with 100mM abbb + 0.1mM bbab, bbabab, bbbba, bab, bbabb, or bbbaabab, etc… then fitness = 0.  I.e. the network reacts abbb (food) with inherited high energy molecules (self) to produce abb.  Fitness is largely unaffected if the network is initialized with 100mM abbb plus any one of the following species at 0.1mM; ba, ab, abb, babb, babbb, bbbbbba.  However, if the network is initialized with 100mM abbb alone, or with 100mM abbb + 0.1mM bbab, bbabab, bbbba, bab, bbabb, or bbbaabab, etc… then fitness = 0.  I.e. the network reacts abbb (food) with inherited high energy molecules (self) to produce abb.

34. Full Experiment  Avalanche type variation introduced.  No adaptations evolved with just abbb as food, within 80,000 generations, and after several runs! To solve this problem….  More food molecule types introduced.  ab,aab, aaab, aabb, abbb, aaaab, aaabb, aabbb, abbbb  Only one initial reaction defined at outset.  abb + abb ---> ba + abbb  Therefore initial fitness = 0.  More possible small molecule reactions undefined.  Avalanche type variation introduced.  No adaptations evolved with just abbb as food, within 80,000 generations, and after several runs! To solve this problem….  More food molecule types introduced.  ab,aab, aaab, aabb, abbb, aaaab, aaabb, aabbb, abbbb  Only one initial reaction defined at outset.  abb + abb ---> ba + abbb  Therefore initial fitness = 0.  More possible small molecule reactions undefined.

35. Fitness 3Sept_10_0_05 Generation Fitness

36. 3Sept_10_0_05 Time Conc abbb,aaab,aabb, (aaaab)* Not used. ba abbb aab abb

37. 3Sept_10_0_05 Rare novel molecule 0.68kJ 1kJ 5kJ 1kJ 0.15kJ 0.27kJ 0.4kJ - High energy food used to drive new reactions. -ba from light not used? 1.68 13 91 1

38. 5Sept15Sept2 5Sept3 5Sept4 No preference for particles of the same length to react. No b related catalysis probability effect. 1/L not 1/L 2 1.Preference for same length species to react. 2.Probability of a particle being catalytic is related to the proportion of p. 3.Prob reaction proportional to 1/L 2. 4.Initial [rare species] = 0.00001 Generation Fitness

39. 5Sept5. No length dependence5Sept6 Initial [rare] = 0.0000001 Not 0.00001

40. 5Sept1 The Evolutionary History of Experiment 5Sept1 661661 662662 12211221 2200,2201,2205 2329 2427 1.07852 91.9089 2200 75.852 20.6873 2201 2205 23.8513 2329 2427

41. The Evolutionary History of Experiment 5Sept2 [abb] 296.704 [abb] 5.27797 [abb] 221.193

42. Properties of Evolved Networks  Many different food molecules are typically utilized, e.g. aabb, abbb, aaab, abbb.  Novel products (X) often catalyse a reaction with a food molecule, e.g.  X + abbb ---> X + babb The novel product typically then reacts with another food molecule to re-form the X, e.g. babb + aaab ----> X + Waste See next diagram…  Many different food molecules are typically utilized, e.g. aabb, abbb, aaab, abbb.  Novel products (X) often catalyse a reaction with a food molecule, e.g.  X + abbb ---> X + babb The novel product typically then reacts with another food molecule to re-form the X, e.g. babb + aaab ----> X + Waste See next diagram…

43. X F1F1 a F2F2 W X This is two step autocatalysis, i.Step one, catalysis. ii.Step two, formation of another catalyst. The complex networks exhibit many such two step autocatalysis reactions. Engram Autocatalysis was evolved.

44. Conclusions  Often the child networks (even without mutation) are less fit that the parent network, due to a non- heritable adaptation in the parent.  Sometimes there are children of greater fitness than parents produced in quick succession, as [abb] increases over several generations due to the same adaptation in the parent. This can mask harmful reactions in the offspring.  The use of the 1:1 GA is restrictive, not allowing selection for robustness to variation, as would be expected with a larger population size, and a high probability of variation.  Often the child networks (even without mutation) are less fit that the parent network, due to a non- heritable adaptation in the parent.  Sometimes there are children of greater fitness than parents produced in quick succession, as [abb] increases over several generations due to the same adaptation in the parent. This can mask harmful reactions in the offspring.  The use of the 1:1 GA is restrictive, not allowing selection for robustness to variation, as would be expected with a larger population size, and a high probability of variation.

45.  Ecological dynamics between spots having different metabolic roles are not considered.  Novel reactions are sustained by the evolution of autocatalytic species, “engram autocatalysts”.  The light absorbing molecule itself shows autocatalytic growth, since this maximizes fitness of the spot, “growth autocatalysis”.  No micro-mutation is necessary if sufficient variety of random catalytic avalanches exists, and if harmful avalanches can be prevented spreading.  Ecological dynamics between spots having different metabolic roles are not considered.  Novel reactions are sustained by the evolution of autocatalytic species, “engram autocatalysts”.  The light absorbing molecule itself shows autocatalytic growth, since this maximizes fitness of the spot, “growth autocatalysis”.  No micro-mutation is necessary if sufficient variety of random catalytic avalanches exists, and if harmful avalanches can be prevented spreading.

46. Fitness = Maximization of polymer mass ~ Spot Growth Rate  Fitness = Sum over all i species of  Species i Length x [Species i ] at end of trial.  Excluding all food molecules.  Initial conditions as before,I.e. all food [molecule] = 100mM, except [abbb] = 0mM  Will the light absorbing reaction of abb be utilized to produce high energy species, as a side-effect of selection for growth rate?  Fitness = Sum over all i species of  Species i Length x [Species i ] at end of trial.  Excluding all food molecules.  Initial conditions as before,I.e. all food [molecule] = 100mM, except [abbb] = 0mM  Will the light absorbing reaction of abb be utilized to produce high energy species, as a side-effect of selection for growth rate?

47. 5Mass1 Gen Fit 5Mass2 5Mass3 5Mass4 Gen No further adaptation in the whole trial!?!?!? No further adaptation in the whole trial!?!?!? No further adaptation in the whole trial!?!?!? Why are avalanches So unlikely to be adaptive?

48. 5Mass55Mass6

49. 831.259 593 The Evolutionary History of Experiment 5Mass1 528 Light absorbing reaction lost.

50. The Evolutionary History of Experiment 5Mass2 139 abb not produced.

51. The Evolutionary History of Experiment 5Mass3 323 abb produced.

52. The Evolutionary History of Experiment 5Mass4 abb produced but light absorbing reaction lost

53. The Evolutionary History of Experiment 5Mass6 abb produced.

54. Conclusion  In some runs abb is produced, but some reactions loose the light absorbing reaction.  Further analysis is required to know how much heat dissipation is based on energy obtained from food molecules vs. from light.  Why are there such long periods without an adaptive avalanche, when networks are larger?  In some runs abb is produced, but some reactions loose the light absorbing reaction.  Further analysis is required to know how much heat dissipation is based on energy obtained from food molecules vs. from light.  Why are there such long periods without an adaptive avalanche, when networks are larger?

55. Thanks to…  Kepa Ruiz-Mirazo  Jon Rowe  Eors Szathmary  Graham Cairns-Smith  Guenter Wachtershauser  Kepa Ruiz-Mirazo  Jon Rowe  Eors Szathmary  Graham Cairns-Smith  Guenter Wachtershauser