Article | Published:

# Sulfate radicals enable a non-enzymatic Krebs cycle precursor

Nature Ecology & Evolution volume 1, Article number: 0083 (2017) | Download Citation

## Abstract

The evolutionary origins of the Krebs cycle (tricarboxylic acid cycle) are not currently clear. Despite the existence of a simple non-enzymatic Krebs cycle catalyst being dismissed only a few years ago as ‘an appeal to magic’, citrate and other intermediates have since been discovered on a carbonaceous meteorite and do interconvert non-enzymatically. To identify a metabolism-like non-enzymatic Krebs cycle catalyst, we used combinatorial, quantitative high-throughput metabolomics to systematically screen iron and sulfate compounds in a reaction mixture that orients on the typical components of Archaean sediment. Krebs cycle intermediates were found to be stable in water and in the presence of most molecule species, including simple iron sulfate minerals. However, in the presence of sulfate radicals generated from peroxydisulfate, the intermediates underwent 24 interconversion reactions. These non-enzymatic reactions covered the critical topology of the oxidative Krebs cycle, the glyoxylate shunt and the succinic-semialdehyde pathway. Assembled in a chemical network, the reactions achieved over 90% carbon recovery. Our results show that a non-enzymatic precursor of the Krebs cycle is biologically sensible, efficient, and forms spontaneously in the presence of sulfate radicals.

The tricarboxylic acid (TCA) cycle, or Krebs cycle, is a central metabolic pathway, typically of oxidative function. This metabolic pathway provides precursors for the biosynthesis of amino acids, and plays an essential role in fatty-acid breakdown, cellular respiration, and energy and redox metabolism1. The widespread occurrence of at least its oxidative reactions2,3 indicates that at least parts of the Krebs cycle originated at a very early stage in evolution; perhaps dating back to the origin of life4,​5,​6. A frequently discussed hypothesis proposes that the Krebs cycle obtained its structural topology by Darwinian selection principles that were enabled by the ‘ribonucleic acid (RNA) world’. A post-genetic origin implies that pathway topologies are subject to progressive change, implying that the modern Krebs cycle could differ substantially from its early precursors7. However, a post-genetic origin of metabolism struggles to explain how multiple, structurally complex enzymes came into being initially; enzymes themselves are made-up from the metabolic products of the Krebs cycle. Second, a Darwinian origin for the metabolic network topology has difficulties in explaining the high number of reactions that recur between kingdoms despite a lack of enzyme sequence conservation8. An alternative hypothesis thus proposes that at least the key metabolic reactions originated from environmental chemistry. In this scenario, inorganic catalysis determines the basic structure of metabolism9,​10,​11,​12.

As enzymatic mechanisms of the Krebs cycle have limited resemblance to inorganic catalysis, the idea of a non-enzymatic origin was received sceptically by many5,13. For instance, Leslie Orgel, a leading scientist in shaping the RNA world hypothesis, dismissed the possibility that a simple inorganic catalyst that could replace a series set of TCA-like reactions as ‘an appeal to magic’5. However, despite the fact that a simple catalyst was indeed missing, others have argued that several Krebs cycle metabolites form in organic chemical reactions9,10. Meanwhile, the presence of a series of TCA intermediates has been confirmed on a carbonaceous meteorite14. Furthermore, citrate and other TCA intermediates undergo highly efficient reductive interconversion reactions on semiconductor particles when exposed to strong ultraviolet light15. A non-enzymatic precursor to the Krebs cycle is therefore catalytically possible. Moreover, the existence of a unifying, simple catalyst has become plausible. Non-enzymatic reactions that replicate two other metabolic pathways, glycolysis and the pentose phosphate pathway, are united in their common dependence on ferrous iron as the catalyst and co-substrate16,17. Fe(ii) is abundant in typical Archaean sediments18,19, implying that general chemical environments, rather than niche conditions, may have been key to shaping the structure of metabolic pathways.

We chose a systematic screening strategy whereby ~4,850 absolute quantitative experiments were performed to examine the reactivity of TCA intermediates in the presence of typical Archaean sediment constituents, as well as related iron and sulfur species. We found that the TCA intermediates were unreactive in the presence of the majority of iron and sulfate combinations, including the simple mineral, ferrous sulfide (FeS). However, in the presence of the radical donor peroxydisulfate, we detected 24 non-enzymatic interconversion reactions. These reactions resemble the isomerization and oxidative reactions of the enzymatic Krebs cycle, the glyoxylate shunt and the succinic semialdehyde pathway, so that their critical topologies are covered. A chemical network assembled from these reactions achieves more than 90% carbon recovery, forming a plausible non-enzymatic precursor for the origin of the early Krebs cycle.

## Results

We optimized the throughput of a liquid-chromatography multiple reaction monitoring (LC-MRM) method to allow for absolute quantification of TCA intermediates in the 6,100 samples eventually generated by this study. We next assessed the reactivity of citrate, cis-aconitate, isocitrate, α-ketoglutarate, succinate, fumarate, malate and oxaloacetate when heated in water in an artificial nitrogen atmosphere (to replicate the low oxygen concentrations of the pre-metabolic world20). TCA intermediates were stable and did not spontaneously react to form other TCA intermediates or other products (Fig. 1a,b and Supplementary Fig. 1). The only exception was oxaloacetate, which formed pyruvate via decarboxylation (Fig. 1b and Supplementary Fig. 2). The Krebs cycle thus contrasts with the glycolysis and pentose phosphate pathways, in which several intermediates spontaneously interconvert in water16,17. Non-enzymatic TCA-like reactivity was therefore highly dependent on enabling chemistry. Furthermore, these experiments served as important controls because they confirmed the absence of enzymatic contamination.

Transition metals were prebiotically abundant and remain common metabolic catalysts today21. We tested a transition metal mixture that reflects the iron, cobalt, nickel and molybdenum concentrations of typical Archaean sediments18,19,22 and found that isocitrate reacted to form α-ketoglutarate, succinate and pyruvate, and α-ketoglutarate reacted to form succinate (Fig. 1c and Supplementary Table 1). These reactions were dependent on Fe(ii) (Supplementary Fig. 3 and Supplementary Table 2), and therefore probably dependent on the α-hydroxyl or α-keto moieties that are specifically present in isocitrate and α-ketoglutarate—since these moieties form iron binding sites (Fig. 1d)23. Furthermore, these reactions followed a clear pH profile (Supplementary Fig. 4), similar to Fe(ii)-mediated glycolysis-like reactions17.

The low number of reactions ruled out the possibility that transition-metal catalysis is sufficient to enable a non-enzymatic Krebs cycle (Fig. 1d). A major bottleneck occurs at the conversions of citrate to isocitrate, and succinate to fumarate, which are both typically catalysed by iron–sulfur-cluster enzymes24. Iron–sulfur minerals such as pyrite could have acted as catalysts in a prebiotic metabolism4, but the surface chemistry of these minerals bears little resemblance to the chemistry of metabolic catalysis. We tested ferrous sulfide as a simple mimetic of an iron–sulfur mineral, but detected no TCA-like reactions (Supplementary Tables 3 and 4, and Supplementary Figs 5 and 6). We continued with a systematic screen, testing ten sulfur-containing compounds, seven iron sources and five TCA substrates in all possible combinations (Fig. 2a). Again, most reaction mixtures had no effect on TCA intermediates (Fig. 2b). However, a comprehensive spectrum of reactivity was enabled on addition of peroxydisulfate (Fig. 2b and Supplementary Table 3). In the two-timepoint screening format, we detected up to 13 TCA-metabolite forming reactions (Fig. 2b and Supplementary Figs 5 and 6).

We noticed that when combined with iron sources, in particular ferrous sulfide, the reaction spectrum enabled by peroxydisulfate was substantially altered. Most reactions were accelerated, while some intermediates were no longer observed at the expense of additional TCA metabolites that were formed (Supplementary Table 1). We therefore continued with a detailed and quantitative (kinetic) characterization of the reactions starting from citrate, cis-aconitate, isocitrate, α-ketoglutarate, succinate, succinic semialdehyde, fumarate and malate in the presence of peroxydisulfate and ferrous sulfide. In the quantitative time-series analyses, we detected 24 reactions that interconvert TCA metabolites (Fig. 3a). A network graph assembled from these reactions showed that a high proportion of the TCA-cycle (its oxidative and isomerization reactions), the glyoxylate shunt, and the succinic semialdehyde pathway were covered by the topology of this non-enzymatic system (Fig. 3b,c).

Considering that our conditions were optimized to obtain a maximum number of TCA-like reactions, rather than maximum yield, the non-enzymatic network was found to be highly efficient. After two hours, 91.9% of the carbon was recovered in TCA intermediates, of which 42.3% were newly formed TCA metabolites. Only 8.1% were non-TCA metabolites, including carbon dioxide (Fig. 3c).

The 24 non-enzymatic reactions occurred at an average rate of 0.280 μmol min−1, for 100 μM substrates in the presence of peroxydisulfate and ferrous sulfide; however, that rate decreased to just 0.0433 μmol min−1 in the absence of ferrous sulfide. The conversion of succinic semialdehyde to succinate was the fastest reaction, while that of citrate to succinate was the slowest reaction (Fig. 3a and Supplementary Fig. 7). In addition, the two reactions were enabled by Fe(ii) (Fig. 1), and were substantially faster in the peroxydisulfate/ferrous sulfide milieu, and respectively occurred at 33.0 and 34.7% rate in the Fe(ii) milieu (Supplementary Fig. 7 and Supplementary Table 1). Furthermore, the intermediates that were exclusively detected in the presence of peroxydisulfate are probably exclusive to the simple milieu, as they are converted to the downstream product at a faster rate than they are formed, once the iron sources are combined with peroxydisulfate (Fig. 3a). Peroxydisulfate can act via a one (radical)- or two-electron mechanism25, and it possesses a high oxidative potential26,27 owing to the donation of sulfate radicals that form on activation by iron and iron-containing sediments such as pyrite28, other trace metals29,30, as well as photolysis and thermolysis31,32. Hypothesizing that the TCA-like reactions were dependent on the formation of these radicals, we investigated the effect of hydrogen peroxide (H2O2), which in the presence of Fe(ii) produces hydroxyl radicals via the Fenton reaction. Of the 24 reactions, 10 were also enabled by H2O2 (Fig. 4a and Supplementary Table 5); however, the specificity was 83% lower and the cumulative reaction rate was 91.1% lower (Fig. 4a). Next, we exploited the differential scavenging properties of 2-propanol and tert-butanol. While 2-propanol efficiently scavenged both hydroxyl and sulfate radicals ($k(HO•)$ = 1.9 × 109 M−1 s−1, $k(SO4•−)$ = 8.2 × 107 M−1 s−1), tert-butanol was less efficient at scavenging the sulfate radical ($k(SO4•−)$ = 8.5 × 105 M−1 s−1), than the hydroxyl radical ($k(HO•)$ = 5.2 × 108 M−1 s−1) (Fig. 4b, left panel)33. TCA-like reactions were not prevented by tert-butanol, but scavenging of sulfate radicals by 2-propanol abolished most of reactivity (Fig. 4b,c and Supplementary Table 5). Radical quenching had a lesser impact on some reactions, which could be explained by incomplete scavenging or by the non-radicalic activity of peroxydisulfate and its product (Fig. 4d). However, for the majority of reactions, the presence of sulfate radicals was essential.

## Discussion

All living cells possess a metabolic network whose topological structure is conserved across kingdoms; this includes pathways like glycolysis and its variants, where the enzymes are not sequence conserved. Sometimes confused with another important question about whether life was of heterotrophic or autotrophic origin34, the origin of this metabolic network remains a largely unsolved problem35,​36,​37. An increasing number of experiments reveal that its core structural parts resemble spontaneous chemical reactions and non-enzymatic catalysis16,21,38. A theory that explains the metabolic origin by an environmental inorganic chemistry is indeed attractive, as it facilitates a stepwise scenario for the origin of enzymes39,40. In the presence of a chemical network, not all enzymes that form a pathway need to come into being at the same time to achieve a functional unit41. Moreover, a chemical system can be improved in steps starting from its most limiting reaction.

Fe(ii) dependent non-enzymatic reactions that resemble glycolysis and the pentose phosphate pathway (PPP)16,17 reveal that metabolism-like reaction networks can form on the basis of a simple and prebiotically abundant catalyst. Seen from another perspective, chemical reactions that are driven by the most abundant Archaean transition metal are part of the metabolic network. However, two key questions remain unanswered. First, there is currently no prebiotically plausible scenario that forms glycolytic and PPP precursors; for example, glucose-6-phosphate and 6-phosphogluconate, respective. It is thus not known how a non-enzymatic glycolysis could escape equilibrium. Second, glycolysis- and PPP-like reactions cannot explain the origin of other pathways that depend on a different chemistry, one without phosphorylated intermediates. The Krebs cycle is an example of such a pathway.

What is attractive about the Krebs cycle is that the non-enzymatic formation of its intermediates has been repeatedly described, including their presence in high concentrations on a carbonaceous meteorite13,​14,​15. In our systematic search for a plausible Krebs cycle catalyst, we excluded chemical conditions that cannot persist within the boundaries of a cell (in particular, ultraviolet light, high pressure, temperatures above 100 °C), and we avoided niche conditions and metals that play virtually no role in metabolism (titanium, borate). We only screened molecules based on elements important for metabolism that are frequent components of Archaean sediment, used low-oxygen conditions and compromised on a temperature (70 °C) that was high enough to facilitate the measurement of slow reactions, yet low enough to be compatible with life.

The fact that almost all the reactions were dependent on the presence of sulfate radicals could be indicative of the prebiotic environment in which the Krebs cycle emerged. Although the reactive radicals themselves have not been preserved over billions of years, there remain indirect traces of an early sulfur redox chemistry. Microfossils in 3.4-billion-year-old rocks from Western Australia are some of the earliest traces of life on Earth, and are characterized by their formation in pyrite and their biotic origin44. This contrasts with glycolysis and the PPP, which depend on Fe(ii) but not on sulfur species16. This observation might shed new light on the endosymbiont theory. While glycolysis and the PPP are typically cytoplasmic, in most eukaryotic species, both the TCA cycle and the assembly of iron–sulfate clusters occur in the mitochondira45.

How could a series of non-enzymatic reactions enable the evolution of the Krebs cycle, which, as well as interconverting and oxidizing metabolites, plays a role in anabolism and includes key co-factor-coupled reactions? To complete the Krebs cycle, a C–C bond forming reaction to yield citrate (catalysed by citrate synthase) is required but this was missing from the non-enzymatic series. However, a full or reductive cycle is present in only a subset of biological species, and thus is probably not the evolutionary starting point for the Krebs cycle2,3. Indeed, an oxidative directionality is the expected thermodynamic outcome for non-enzymatic reactions. Cofactor-dependent reactions reverse the directionality in living system and prevent the system from attaining equilibrium, these are considered a speciality of enzymatic catalysis. The closure of the TCA cycle via a citrate-synthase-like reactivity means that a significant advantage could be achieved by adding just a single enzymatic step. This provides an interesting case for the evolution of an enzyme-catalysed metabolism. Another possibility is that the first versions of the Krebs cycle could have been heterotrophic in nature, with precursors coming from environmental chemical reactions that did not necessarily bear a resemblance to metabolic pathways14,15. Alternatively, early closure of the Krebs cycle might not have been required if its oxidative reactions were fed by other metabolism-like reactions, such as a pyruvate carboxylation that enabled coupling to an early glycolysis. The existence of a non-enzymatic pyruvate carboxylase reaction could unify key issues in explaining the origins of glycolysis, the PPP and the Krebs cycle.

In summary, we describe a mutually compatible, efficient, non-enzymatic catalysis that interconverts TCA cycle intermediates. The reactions were enabled by sulfate radicals that form on activation of peroxydisulfate. The resulting chemical network covers the topology of the conserved (oxidative) part of the Krebs cycle, and achieves more than 90% carbon efficiency. The simplicity of these ‘environmental’ conditions, which were based around a unifying sulfate catalyst, shows that the Krebs cycle could have emerged from a non-enzymatic precursor that forms spontaneously in the presence of sulfate radicals.

## Methods

### General study design

#### Sample size

Experiments exploring the available chemical reaction space are constrained by combinatorics: the number of samples multiplies with each additional condition to be tested. To address this issue, we simplified and optimized sample preparation, liquid-chromatography selective reaction monitoring (LC-SRM) and statistical analyses for screening purposes, at the expense of exact reaction rate measurements. In subsequent verification experiments, we used more resource-intensive methods for optimal chromatographic separation and absolute quantification of reaction rates. This allowed us to measure more than 4,850 samples along with 1,250 controls, external standards and blanks, and to efficiently analyse and manually curate the corresponding ~65,000 LC-SRM chromatograms.

#### Replicates and time series

Each calculated reaction rate is based on a time-series experiment with six collection points, which were performed in at least three replicates. Rates were only calculated for those reactions that showed significant product accumulation in all three replicates. Consequently, a rate calculation for one reaction under a given condition comprised 18 measurement points.

#### Sample collection/endpoints

Sampling points in time-course experiments were typically set to 0, 10, 30, 60, 120 and 300 min to cover both relatively fast and slow reactions. The reaction of oxaloacetate to pyruvate was the only one that was too fast for these intervals and we designed an alternative experimental setup with narrower intervals to follow this reaction.

#### Quality control and outliers

Peaks were identified by matching retention time and fragmentation properties to externally measured chemically pure standards. Automated peak picking and integration of specific transitions by MassHunter (Agilent) was backed up by additional manual inspection of all peaks (see LC-SRM method section). We controlled for cross-contaminations and carry-over by repeated measurement of blanks and substrate-free controls.

### Materials

Metabolite standards were obtained at high purity. If not otherwise indicated, the product number refers to Sigma-Aldrich: sodium citrate (71635), sodium isocitrate (I1252), cis-aconitic acid (A3412), sodium α-ketoglutarate (K2010), sodium succinate (14160), succinic semialdehyde (Santa Cruz Biotechnology, F1114), sodium fumarate (F1506), l-malic acid (M6413), sodium pyruvate (P2256), oxaloacetic acid (O4126), iron acetate (339199), FeCl2 (372870), FeCl3 (157740), Fe(ClO4)2 (334081), Fe(ClO4)3 (309281), ferrocene (F408), FeS(268704), H3PO4 (P5811), 2-mercaptoethanol (Merck Millipore, 805740), cysteine (30095), dl-ethionine (E5139), dimethylsulfoxide (D8418), homocysteic acid (69453), NaHSO3 (Acros Organics, 41944), methionine (M9375), ammonium peroxydisulfate (Fischer Scientific, 10219790), sodium sulfite Na2SO3 (Fischer Scientific, 10070400), sodium sulfate Na2SO4 (Fischer Scientific, 10493372). All water was obtained commercially at a purity suitable for ultra-performance liquid chromatography (UPLC) and mass spectrometry (Biosolve Chemicals, Cat no. 23214102).

### LC-SRM method for TCA metabolite quantification

For quantification of TCA intermediates, an Agilent 1290 Infinity Binary LC system with an online coupled Agilent 6460 triple quadrupole mass spectrometer was used. Sample separation was achieved on a Zorbax Eclipse Plus C18 Rapid Resolution column (1.8 μm, 2.1 mm × 50 mm; column temperature, 30 °C; Agilent). Solvent A contained 5% methanol, 0.2% acetic acid and 10 mM tributylamine in UPLC-grade water (Greyhound) and Solvent B was 100% methanol (Greyhound). Injection of 1.5–2.5 μl of sample onto the column at 0.5 ml min−1 was followed by gradient elution according to the UPLC gradient conditions given in Supplementary Tables 7–11. For each injection the needle was washed with water:acetonitrile (2:1) containing FlushPort to prevent sample carryover. Including washing and re-equilibration, this resulted in cycle times of between 4.4 and 16 min, depending on the gradient conditions (Supplementary Tables 7–11).

Mass spectrometric quantification of specific products was performed in multiple or selective reaction monitoring (MRM or SRM) mode. Instrument parameters and metabolite transitions for each feature were optimized using pure commercially available standards (Supplementary Tables 12 and 13). Tandem mass spectrometry data were analysed using a Masshunter Workstation (Agilent) via its QQQ analysis software. All automatically integrated peaks were manually curated to ensure high data quality and consistency with regularly measured quality-control standards. Repeatedly measured external standard dilution series were used for determination of absolute metabolite concentrations. Further analysis and fitting of reaction rates were performed in R (www.R-project.org)46

### Determination of reaction rate

For each replicate, the recorded time series was used to determine product formation rates (time-dependent accumulation of reaction products). Different fitting algorithms were used to account for varying modes of product formation, caused by the dissimilar chemical properties of substrates, products and co-products, as well as other effects such as intermediate stabilities, reaction orders or reaction-induced changes to the chemical environment. The models used were: (1) a least-squares linear model for continuous product formation; (2) a nonlinear growth model, $y=ae−b⋅cx$, which is part of the SSgompertz module in R; (3) the maximum initial growth rate within the linear range using the least squares model, to be used for products with dynamic formation/degradation behaviour; and (4) and an exponential decay model, y = a 2bx + c. The best fitting model was chosen according to the respective coefficient of determination (R2) values and was confirmed by manual review of the plotted curves.

### Statistical analysis of iron/sulfur reactivity screen

Differential raw peak area data (ΔXi = Xt=300 − Xt=0) between t = 300 min and t = 0 min were calculated from the individual LC-SRM peak areas extracted using the Masshunter software (Agilent) and were indicative the accumulation of products (positive values) or the removal of substrates (negative values). Assuming a normal distribution, we then calculated z-scores (zi) (for a graphical representation, see Supplementary Fig. 5): $zi=Xi−median(X)σ(X)$ with σ(X) being approximated by the median absolute deviation (MAD) according to: σ = MAD(X) × 1.4826. Two-tailed P-values were then calculated from the z-scores using a normal distribution function, f(x) where the mean of the population, μ, was 0 and σ was 1: $(1)p=2f(x)=22π×σ×e(|zi|22)$

All calculations were performed in R using standard inbuilt functions, including mad() and pnorm(). A graphical representation of the results can be found in Fig. 2b and Supplementary Fig. 6.

### Iron-rich Archaean sediment simulation experiments

The TCA substrates (100 μM) were combined with a freshly prepared metal-rich mixture (200 μM FeCl2, 10 nM CoCl2, 400 nM NiCl2, 10 nM MoO4 and 100 μM H3PO4) in a low-oxygen N2 atmosphere generated by three repeated vacuum/N2 gas cycles in an anoxic chamber (Coy Lab Products). The mixtures were then sealed in glass vials designed for high-performance liquid chromatography (HPLC) (Agilent, 5182-0717 and 5182-0716), and incubated at 70 °C in a water bath. For all substrates, other than oxaloacetate, samples were collected after 0, 10, 30, 60, 120 and 300 min by rapid cooling on ice. This resulted in measurement of 486 samples plus controls. After incubation, the reaction mixtures were transferred to 384-well plates (Greiner Bio-One, 781186) under normoxic conditions and stored at −80 °C for subsequent LC-SRM measurement using the UPLC gradient conditions specified in Supplementary Table 7.

### Individual metal effects

To determine the contribution of individual metals towards facilitating specific TCA metabolite interconversions we tested each of the constituents of the Archaean sediment simulation for their catalytic potential. The same experimental setup as above was applied, using isocitrate and α-ketoglutarate as substrates and mixing them with each metal individually (200 μM FeCl2, 200 μM FeCl3, 10 nM CoCl2, 400 nM NiCl2, 10 nM MoO4, 100 μM H3PO4) under low-oxygen conditions; normoxic conditions were used for Fe(iii). Substrates and products in 324 samples plus controls were measured via the LC-SRM method using the UPLC gradient conditions given in Supplementary Table 8.

Reaction mixtures containing oxaloacetate were generated by combining oxaloacetate with FeCl2 at the same concentrations as stated above but under normoxic conditions, directly followed by incubation at 40 °C in the autosampler of the Agilent 1290 HPLC system. Instead of stopping the reaction on ice, the HPLC system was programmed to automatically inject 1.5 μl of the sample onto the column every 5 min for up to 2.5 h. The three replicates for the FeCl2 condition and the iron-free control were measured in alternating order. Substrates and products in 113 samples plus controls were measured via the LC-SRM method using the UPLC gradient conditions in Supplementary Table 8.

### Dependence on pH

To determine the effects of pH on ferrous-iron-induced non-enzymatic reactivity, the interconversion experiments were carried out in the presence of 5 mM sodium phosphate buffering at pH 3, 5, 6, 7, 8 and 9, in a manner analogous to the experiments described in ref. 17. Furthermore, the reaction mixture consisted of 100 μM isocitrate (or water as control) and 200 μM iron(ii) chloride (Sigma-Aldrich, 372870). The correct starting pH was controlled for by initial measurement, and kept constant throughout the incubations via the buffer capacity of the respective buffer. Substrates and products in 252 samples, plus 56 controls, were measured by the LC-SRM method using the UPLC gradient conditions in Supplementary Table 9.

### Screening of iron/sulfur compounds for their catalytic impact on TCA cycle intermediates

For screening of iron and/or sulfur species that actively promote non-enzymatic reactivity among TCA metabolites, the experimental setup was adapted to a 96-well format. Samples were prepared in glass-coated 96 well plates (Thermo Scientific, 60180-P300), sealed under low-oxygen conditions and additionally vacuum-packed to prevent any possible well-to-well cross-contamination. The substrates (100 μM), citrate, cis-aconitate, succinate, malate and fumarate were used individually and mixed with all possible combinations of the iron sources (FeC2H3O2, FeCl2, FeCl3, Fe(ClO4)2, Fe(ClO4)3, ferrocene, FeS and iron-free controls) and inorganic or organic sulfur species (2-mercaptoethanol, cysteine, DL-ethionine, dimethylsulfoxide, homocysteic acid, NaHSO3, methionine, (NH4)2S2O8, Na2SO3, Na2SO4 and sulfur-free controls); all these iron and sulfur species were used at 200 μM. The samples were heated to 70 °C in a water bath for 0 and 300 min. This resulted in a total of 1,320 samples plus controls. Samples were then transferred to 384-well plates (Greiner Bio-One, 781186) under normoxic conditions and stored at −80 °C for subsequent LC-SRM measurement using the UPLC gradient conditions in Supplementary Table 10.

### Non-enzymatic TCA qualitative and quantitative (kinetic) validation experiments

On the basis of the statistical analysis of the iron/sulfur screening results, we selected the most potent non-enzymatic interconversion conditions and verified them in a time series experiment with three replicates using a comprehensive targeted mass spectrometric quantification method. The substrates citrate, isocitrate, cis-aconitate, α-ketoglutarate, succinate, succinic semialdehyde, fumarate and malate were diluted to 100 μM in UPLC grade water and mixed with 200 μM peroxydisulfate in the presence of absence of 200 μM ferrous sulfide (FeS). In analogy to the experiments above, these mixtures were sealed under anoxic conditions and were incubated at 70 °C in a water bath. Samples were collected after 0, 10, 30, 60, 120 and 300 min by rapid cooling on ice, resulting in 432 samples plus controls, which were transferred to 384 well plates and stored at −80 °C for subsequent LC-SRM analysis using the UPLC gradient conditions given in Supplementary Table 11.

### Recovery and specificity calculations

Recovery rates were quantified via the quotient of the respective TCA metabolites at timepoint t = 0 and the total sum of concentrations measured after 2 h of incubation at 70 °C. The amount of products at timepoint t = 2 h divided by the total sum of TCA metabolites (substrate + products) was calculated to determine the percentage of specific TCA products. Note that the absolute quantification of citrate in the presence of peroxydisulfate (and with it the calculated recovery rate) was hampered by an unusually high variance (coefficient of variance, CV = 25.6%) for this metabolite due to an unknown mechanism/chemical interference (the average variance for all other features, metabolites, and recovery rates remained low: CV = 3.48%). For this technical reason, the recovery rate calculated for citrate and the efficiency of the reactions that started from citrate, had a higher uncertainty according to this CV value.

### Comparison of hydrogen peroxide and peroxydisulfate as radical donors for non-enzymatic TCA-like reactions

To investigate the potential contribution of sulfur radical species to the non-enzymatic reactivity facilitated by peroxydisulfate, we compared them with hydroxyl radicals generated from hydrogen peroxide. In three replicates, we incubated 100 μM citrate, isocitrate, cis-aconitate, alpha-ketoglutarate, succinate, fumarate and malate with 200 μM hydrogen peroxide or 200 μM peroxydisulfate at 70 °C under anoxic conditions and collected samples after 0, 10, 30, 60, 120 and 300 min (378 samples plus controls) and measured them via LC-SRM using the UPLC gradient conditions in Supplementary Table 11.

### Sulfate and hydroxyl radical scavenging experiments

The differential radical scavenging properties of 2-propanol and tert-butanol were exploited to test for the importance of sulfate over hydroxyl radicals. In these experiments, 200 μM peroxydisulfate was mixed with 100 μM citrate, isocitrate, cis-aconitate, α-ketoglutarate, succinate, fumarate or malate individually and each combination was treated with 500 μM 2-propanol, 500 μM tert-butanol or water as a control. Samples were sealed under anoxic conditions, heated to 70 °C in a water bath and incubated for 0, 10, 30, 60, 120 and 300 min (504 samples plus controls) and then measured by LC-SRM analysis using the UPLC gradient conditions specified in Supplementary Table 11.

### Data availability

The reaction rate data are provided in Supplementary Tables 1, 2, 3, 4, 5 and the raw quantification data has been deposited in the Mendeley Data repository (http://dx.doi.org/10.17632/vgpmnzdz55.1).

How to cite this article: Keller, M. A., Kampjut, D., Harrison, S. A. & Ralser, M. Sulfate radicals enable a non-enzymatic Krebs cycle precursor. Nat. Ecol. Evol. 1, 0083 (2017).

## References

1. 1.

& The role of citric acid in intermediate metabolism in animal tissues. FEBS Lett. 117(suppl.), K1–K10 (1980).

2. 2.

, & Variation and evolution of the citric-acid cycle: a genomic perspective. Trends Microbiol. 7, 281–291 (1999).

3. 3.

, & The puzzle of the Krebs citric acid cycle: assembling the pieces of chemically feasible reactions, and opportunism in the design of metabolic pathways during evolution. J. Mol. Evol. 43, 293–303 (1996).

4. 4.

Evolution of the first metabolic cycles. Proc. Natl Acad. Sci. USA 87, 200–204 (1990).

5. 5.

The implausibility of metabolic cycles on the prebiotic Earth. PLoS Biol. 6, e18 (2008).

6. 6.

, , , & The origin of life in alkaline hydrothermal vents. Astrobiology 16, 181–197 (2016).

7. 7.

, , & The origin of RNA and ‘My Grandfather’s Axe’. Chem. Biol. 20, 466–474 (2013).

8. 8.

, , & Carbohydrate metabolism in Archaea: current insights into unusual enzymes and pathways and their regulation. Microbiol. Mol. Biol. Rev. 78, 89–175 (2014).

9. 9.

, , & The origin of intermediary metabolism. Proc. Natl Acad. Sci. USA 97, 7704–7708 (2000).

10. 10.

& Universality in intermediary metabolism. Proc. Natl Acad. Sci. USA 101, 13168–13173 (2004).

11. 11.

& in Cellular Origin and Life in Extreme Habitats and Astrobiology Vol. 7 (eds Seckbach, J., Chela-Flores, J., Owen, T. & Raulin, F.) 129–132 (Springer, 2004).

12. 12.

A replicator was not involved in the origin of life. IUBMB Life 49, 173–176 (2000).

13. 13.

, & Uncertainty of prebiotic scenarios: the case of the non-enzymatic reverse tricarboxylic acid cycle. Sci. Rep. 5, 8009 (2015).

14. 14.

, , , & Detection and formation scenario of citric acid, pyruvic acid, and other possible metabolism precursors in carbonaceous meteorites. Proc. Natl Acad. Sci. USA. 108, 14015–14020 (2011).

15. 15.

& Driving parts of Krebs cycle in reverse through mineral photochemistry. J. Am. Chem. Soc. 128, 16032–16033 (2006).

16. 16.

, & Non-enzymatic glycolysis and pentose phosphate pathway-like reactions in a plausible Archaean ocean. Mol. Syst. Biol. 10, 725 (2014).

17. 17.

et al. Conditional iron and pH-dependent activity of a non-enzymatic glycolysis and pentose phosphate pathway. Sci. Adv. 2, e1501235 (2016).

18. 18.

, & Iron isotope constraints on the Archaean and Paleoproterozoic ocean redox state. Science 307, 1088–1091 (2005).

19. 19.

, & The bioinorganic chemistry of the ancient ocean: the co-evolution of cyanobacterial metal requirements and biogeochemical cycles at the Archaean–Proterozoic boundary? Inorganica Chim. Acta 356, 308–318 (2003).

20. 20.

, & The Archaean sulfur cycle and the early history of atmospheric oxygen. Science 288, 658–661 (2000).

21. 21.

& Metal catalysts and the origin of life. Elements 12, 413–418 (2016).

22. 22.

Biogeochemical signatures through time as inferred from whole microbial genomes. Am. J. Sci. 305, 467–502 (2005).

23. 23.

& Iron-catalyzed oxidative decarboxylation of benzoylformic acid. J. Am. Chem. Soc. 101, 2221–2222 (1979).

24. 24.

& Metabolic regulation of citrate and iron by aconitases: role of iron-sulfur cluster biogenesis. Biometals 20, 549–564 (2007).

25. 25.

& The chemistry of persulfate. I. The kinetics and mechanism of the decomposition of the persulfate ion in aqueous medium 1. J. Am. Chem. Soc. 73, 3055–3059 (1951).

26. 26.

, & Persistence of persulfate in uncontaminated aquifer materials. Environ. Sci. Technol. 44, 3098–3104 (2010).

27. 27.

The free-radical chemistry of persulfate-based total organic carbon analyzers. Mar. Chem. 41, 91–103 (1993).

28. 28.

, , , & Oxidative degradation of MTBE by pyrite-activated persulfate: proposed reaction pathways. Ind. Eng. Chem. Res. 49, 8858–8864 (2010).

29. 29.

, , & Persulfate oxidation for in situ remediation of TCE. I. Activated by ferrous ion with and without a persulfate–thiosulfate redox couple. Chemosphere 55, 1213–1223 (2004).

30. 30.

, & Persulfate activation by naturally occurring trace minerals. J. Hazard. Mater. 196, 153–159 (2011).

31. 31.

On the photolysis of simple anions and neutral molecules as sources of O/OH, SOx and Cl in aqueous solution. Phys. Chem. Chem. Phys. 9, 3935–3964 (2007).

32. 32.

, & Thermally activated persulfate oxidation of trichloroethylene: experimental investigation of reaction orders. Ind. Eng. Chem. Res. 47, 2912–2918 (2008).

33. 33.

et al. Oxidative and reductive pathways in iron-ethylenediaminetetraacetic acid–activated persulfate systems. J. Environ. Eng. 138, 411–418 (2012).

34. 34.

, & On the origin of heterotrophy. Trends Microbiol. 24, 12–25 (2016).

35. 35.

et al. The moderately efficient enzyme: evolutionary and physicochemical trends shaping enzyme parameters. Biochemistry 50, 4402–4410 (2011).

36. 36.

Accuracy-rate tradeoffs: How do enzymes meet demands of selectivity and catalytic efficiency? Curr. Opin. Chem. Biol. 21, 73–80 (2014).

37. 37.

An open question on the origin of life: the first forms of metabolism. Chem. Biodivers. 9, 2635–2647 (2012).

38. 38.

& Spontaneous emergence of S-adenosylmethionine and the evolution of methylation. Angew. Chem. Int. Ed. 56, 343–345 (2017).

39. 39.

The RNA world and the origin of metabolic enzymes. Biochem. Soc. Trans. 42, 985–988 (2014).

40. 40.

, & The widespread role of non-enzymatic reactions in cellular metabolism. Curr. Opin. Biotechnol. 34, 153–161 (2015).

41. 41.

On the evolution of biochemical syntheses. Proc. Natl Acad. Sci. USA 31, 153–157 (1945).

42. 42.

et al. Application of capillary electrophoresis-mass spectrometry to synthetic in vitro glycolysis studies. Electrophoresis 25, 1996–2002 (2004).

43. 43.

Strategies of antioxidant defense. Eur. J. Biochem. 215, 213–219 (1993).

44. 44.

, , , & Microfossils of sulphur-metabolizing cells in 3.4-billion-year-old rocks of Western Australia. Nat. Geosci. 4, 698–702 (2011).

45. 45.

& Maturation of cellular Fe–S proteins: an essential function of mitochondria. Trends Biochem. Sci. 25, 352–356 (2000).

46. 46.

R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2015).

## Acknowledgements

We thank G. Averill and T. Littmann for helping with experiments. This work was supported by the Francis Crick Institute, which receives its core funding from Cancer Research UK (FC001134), the UK Medical Research Council (FC001134) and the Wellcome Trust (FC001134). M.R. is supported by a Wellcome Trust grant, RG 093735/Z/10/Z, and a European Research Council Starting Grant, 260809. M.A.K. is supported by an Erwin Schrödinger postdoctoral fellowship (FWF, Austria, J3341). D.K. is supported by an Ad Futura studentship (Slovene Scholarship Fund).

## Affiliations

1. ### Department of Biochemistry and Cambridge Systems Biology Centre, University of Cambridge, 80 Tennis Court Road, Cambridge CB2 1GA, UK

• Markus A. Keller
• , Domen Kampjut
• , Stuart A. Harrison
•  & Markus Ralser
2. ### Division of Biological Chemistry, Biocenter, Medical University of Innsbruck, Innrain 80-82, 6020 Innsbruck, Austria

• Markus A. Keller
3. ### Division of Human Genetics, Medical University of Innsbruck, Peter-Mayr-Straße 1, 6020 Innsbruck, Austria

• Markus A. Keller
4. ### The Molecular Biology of Metabolism Laboratory, The Francis Crick Institute, 1 Midland Road, London NW1 1AT, UK

• Markus Ralser

## Authors

### Contributions

M.A.K. and M.R. designed the research. M.A.K., D.K. and S.A.H. performed the research. M.A.K. and M.R. wrote the first draft of the paper, and all authors contributed to finalizing the manuscript.

### Competing interests

The authors declare no competing financial interests.

## Corresponding author

Correspondence to Markus Ralser.

## PDF files

1. 1.

### Supplementary Information

Supplementary Tables 6–13, Supplementary Figures 1–7.

## Excel files

1. 1.

### Supplementary Table 1

Reaction rate data for controls, Fe(II), peroxydisulfate and peroxydisulfate/ferrous sulfide.

2. 2.

### Supplementary Table 2

Metal dependency rate data.

3. 3.

### Supplementary Table 3

Z-score data.

4. 4.

### Supplementary Table 4

Complete reaction list.

5. 5.

### Supplementary Table 5

Scavenger experiment reaction rate data.