A prebiotic basis for ATP as the universal energy currency

ATP is universally conserved as the principal energy currency in cells, driving metabolism through phosphorylation and condensation reactions. Such deep conservation suggests that ATP arose at an early stage of biochemical evolution. Yet purine synthesis requires six phosphorylation steps linked to ATP hydrolysis. This autocatalytic requirement for ATP to synthesize ATP implies the need for an earlier prebiotic ATP-equivalent, which could drive protometabolism before purine synthesis. Why this early phosphorylating agent was replaced, and specifically with ATP rather than other nucleotide triphosphates, remains a mystery. Here we show that the deep conservation of ATP reflects its prebiotic chemistry in relation to another universally conserved intermediate, acetyl phosphate, which bridges between thioester and phosphate metabolism by linking acetyl CoA to the substrate-level phosphorylation of ADP. We confirm earlier results showing that acetyl phosphate can phosphorylate ADP to ATP at nearly 20 % yield in water in the presence of Fe3+ ions. We then show that Fe3+ and acetyl phosphate are surprisingly favoured: a panel of other prebiotically relevant ions and minerals did not catalyze ADP phosphorylation; nor did a number of other potentially prebiotic phosphorylating agents. Only carbamoyl phosphate showed some modest phosphorylating activity. Critically, we show that acetyl phosphate does not phosphorylate other nucleotide diphosphates or free pyrophosphate in water. The phosphorylation of ADP monomers seems to be favoured by the interaction between the N6 amino group on the adenine ring with Fe3+ coupled to acetyl phosphate. Our findings suggest that the reason ATP is universally conserved across life is that its formation is chemically favoured in aqueous solution under mild prebiotic conditions.


ADP phosphorylation occurs in a range of aqueous prebiotic environments 145
We next explored the conditions under which Fe 3+ catalyses the phosphorylation of ADP by acetyl 146 phosphate, specifically pH, temperature, water activity and pressure. We found that the reaction is 147 strongly sensitive to pH, and occurs most readily under mildly acidic conditions, with an optimum pH 148 of ~5.5-6, the uncorrected default pH of the reaction (Fig. 2a). Slightly more acidic conditions (pH 4) 149 suppressed the yield a little, but more alkaline conditions had a much stronger suppressive effect. 150 ATP yield fell by around three quarters at pH 7, and collapsed to nearly zero at pH 9. While this sharp 151 sensitivity to pH might seem at first sight limiting, in the Discussion we show that, on the contrary, it 152 could be valuable in generating disequilibria, enabling ATP hydrolysis to power work. 153 ATP yield was less acutely sensitive to temperature, at least between 20 and 50 °C. Over 24 154 hours, the overall ATP yield reflects both synthesis and hydrolysis. We found that 30 °C optimised 155 yield across 24 hours, by promoting synthesis within the first 4 hours while limiting hydrolysis over 156 the subsequent 20 hours (Fig. 2b). The rate of synthesis was a little lower at 20 °C, but this was 157 offset by slightly less hydrolysis over 24 hours. ATP synthesis was markedly faster at 50 °C, but so too 158 was hydrolysis, which already lowered yields within the first 2 hours and cut them to less than a 159 quarter of those at 30 °C after 24 hours. If ATP is to power work, as in modern cells, then hydrolysis 160 in itself is not an issue, but rather needs to be coupled to other reactions such as the 161 phosphorylation or condensation of substrates. Such processes also tend to take place over minutes 162 to hours [10], meaning that temperature has a relatively trivial effect, with the yield after 2-3 hours 163 being similar at all three temperatures studied, at around 10-15 % (Fig. 2b). This implies that 164 temperature would not be a strong limiting factor on many possible prebiotic environments. 165 More surprisingly, ATP yield was greatest at high water activity, either in HPLC-grade water 166 or in suspended silica (Fig. 2c). Adding NaCl lowered ATP yield, albeit not dramatically. Moderate 167 NaCl concentration (300 mM, giving a total reaction ionic strength of 303.75 mM) lowered ATP yield 168 by around a fifth. Modern ocean salinity (600 mM NaCl, reaction ionic strength 603.75 mM) and 169 higher salinity (1 M NaCl, reaction ionic strength 1.004 M) both roughly halved the yield. This 170 suggests that the effect of solutes does not only reflect ionic strength, which was confirmed by the 171 addition of other solutes. Dissolved silicate (10 mM SiO2) also halved ATP yield, even though the 172 ionic strength in this case was only 123.75 mM (Fig. 2c). Likewise, higher Mg 2+ and Ca 2+ 173 concentrations (50 mM and 10 mM, respectively) as part of a modern ocean mix collapsed ATP 174 yields to nearly zero (Fig. 2c), presumably because Ca 2+ and Mg 2+ promote ATP hydrolysis [42,43]. 175 While this might suggest that ATP synthesis could not occur in modern oceans, Mg 2+ and Ca 2+ 176 concentrations can in fact vary considerably in ocean environments (see Discussion). We show later 177 that lower Mg 2+ and Ca 2+ concentrations (~2 mM) actually promote ATP synthesis. 178 High pressure (80 bar) had very little effect on ATP synthesis (Fig. 2d). This is consistent with 179 the work of Leibrock, Bayer, and Lüdemann (1995), who showed that high pressure promotes ATP 180 hydrolysis, but only at pressures ≥ 300 bar. The slightly greater ATP yield at ambient pressure in our 181 experiment may be attributable to greater evaporation in the open (non-pressurized) system. This 182 was clearly the case in the absence of Fe 3+ , where most of the ATP detected was not produced by 183 phosphorylation of ADP, but contamination of the ADP commercial standard via the manufacturing 184 process, then concentrated by evaporation at ambient pressure (SI Fig. 3 Given the diverse reaction kinetics anticipated with these different phosphorylating agents, 193 we carried out experiments at both at 30 °C (the optimal temperature for AcP) and 50 °C (as most 194 phosphate donors are less labile than AcP and so might be more effective at higher temperatures), 195 as well as pH 5.5-6, 7 and 9. As shown in Fig. 3, no other phosphorylating agent was as effective as 196 AcP at synthesising ATP in the presence of Fe 3+ . The only other phosphorylating agent to show any 197 notable efficacy was carbamoyl phosphate (CP), which is similar in structure to AcP; it has a 198 carbamate (-CO-NH2) rather than acetate (-CO-CH3) bound to phosphate. CP produced about half the 199 ATP yield of AcP at 20 °C and pH 5.5-6 ( Fig. 3a), but barely a quarter of the yield at pH 7 ( Fig. 3b). At 200 pH 9, only cyclic trimetaphosphate (cTMP) produced any ATP at all, albeit after a delay of more than 201 20 hours (Fig. 3c). 202 At 50 °C, CP generated ATP continuously over 24 hours at pH 5.5-6, despite producing only 203 half the yield in the first 2 hours. The fact that ATP yield declined over time with AcP indicates that 204 ATP was hydrolysed over hours at 50 °C; it was not replenished because AcP also hydrolysed at that 205 temperature [10]. While CP has a similarly low thermal stability, the primary decomposition product 206 is cyanate [57], which is itself a proficient condensing agent [58]. This likely contributed to a balance 207 between the synthesis and hydrolysis of ATP over 24 hours. Only AcP formed any ATP at 50 °C and 208 pH 7 (Fig. 3e), consistent with the pH sensitivity of CP seen at 30 °C. CP did form ATP at low yield at 209 50 °C and pH 9 (Fig. 3f), and we can infer again that it is due to the decomposition product cyanate; . 210 The main conclusion here is that from a panel of eight plausibly prebiotic phosphorylating agents, 211 only AcP was capable of generating an ATP yield of > 10% in water at both 30 and 50 °C. The only 212 other agent to show remotely comparable efficacy at mildly acidic pH was CP, but its maximal yield 213 was half that of AcP. The fact that CP was capable of synthesising ATP at low yield under warm 214 alkaline condition (50 °C, pH 9) in fact lowers its phosphorylating potential as it is less capable of 215 sustaining a disequilibrium of ATP/ADP ratio in a dynamic pH environment (see Discussion). 216 217

Phosphorylation of ADP to ATP is unique among nucleotide diphosphates 218
We next explored the propensity of AcP to phosphorylate other canonical nucleotide diphosphates 219 (NDPs), specifically cytidine diphosphate (CDP), guanosine diphosphate (GDP), uridine diphosphate 220 (UDP) and inosine diphosphate (IDP). While not a canonical base, inosine is the precursor to both 221 adenosine and guanosine in purine synthesis. Importantly, from a mechanistic point of view, inosine 222 lacks the amino group incorporated at different positions onto the purine rings of adenosine and 223 guanosine, but like GDP, IDP has an oxygen in place of the N6 amino group of adenosine. The results 224 clearly show that AcP will phosphorylate ADP but not other NDPs (Fig. 4a-e), demonstrating a strong 225 dependence on the structure of the nucleobase. For all NDPs, a peak for the corresponding 226 triphosphate was present at the start of the reaction, but this did not change over 3 hours for any 227 NDP except ADP. As noted above for ADP, the presence of the NTP at 0 h can be ascribed to minor 228 contamination of the commercial standard during the manufacturing process. 229 To explore the dependence of phosphorylation on the nucleobase, and to establish whether 230 Fe 3+ interacts directly with the base as well as its diphosphate tail, ADP was substituted by potassium 231 pyrophosphate (PPi) in the reaction mixture with AcP and Fe 3+ . No triphosphate was detected by 31 P-232 NMR (Fig. 4f), which suggests that the adenine ring does indeed need to interact directly with Fe 3+ . 233 We note that Fe 3+ heavily interferes with 31 P-NMR spectroscopy due to its paramagnetism. To 234 minimize the presence of Fe 3+ in the sample, we therefore performed solid-phase extraction twice 235 before NMR. Despite this precaution, the experimental samples still showed some deformation, 236 suggesting that Fe 3+ continued to interact with the phosphate groups (SI Fig. 4) [59]. Nonetheless, 237 this small deformation is cosmetic and does not conceal the absence of triphosphate in the reaction 238 mixture. We also considered whether Fe 3+ could interact with the adenine ring but not the 239 diphosphate tail, analysing the phosphorylation of AMP to ADP. AcP did indeed phosphorylate AMP 240 to ADP in the presence of Fe 3+ (SI Fig. 5) but at considerably lower yield than ADP to ATP. Thus, Fe 3+ 241 interacts preferentially with the purine ring coupled to the diphosphate tail. 242 The fact that neither pyrimidine NDP could be phosphorylated suggests that the purine ring 243 (or at least adenosine) is essential for positioning the interactions between Fe 3+ and AcP. ADP has an 244 amino group at N6, whereas GDP has a carbonyl at C6 and an amino group at N2; inosine has a 245 carbonyl group at C6; and both GDP and IDP have a protonated N at N1. We infer that the critical 246 moiety in the adenosine ring for phosphorylation by AcP with Fe 3+ as catalyst must be the N6-amino 247 group of adenosine, as the IDP and GDP ring structures are equivalent elsewhere. In particular, from 248 a mechanistic point of view, we note that the N7 is equivalent in all three purine rings, so although 249 this might also interact with Fe 3+ , as suggested by others [60][61][62][63], it cannot be the critical moiety. 250 251

Catalysis of ADP phosphorylation does not involve nucleotide stacking 252
To understand how Fe 3+ catalyses the phosphorylation of ADP to ATP, we tested the effect of varying 253 the Fe 3+ ion concentration. Holding the ADP and AcP concentrations constant at 1 mM and 4 mM, 254 respectively, we varied the Fe 3+ concentration from 0.05 to 2 mM. We found that the maximal ATP 255 yield was produced by 1 mM Fe 3+ , indicating that the optimal ADP:Fe 3+ stoichiometry of the reaction 256 was 1:1 (Fig. 5a). Following Kitani et al. [33]we confirmed that low concentrations of either Mg 2+ or 257 Ca 2+ (up to 2 mM) slightly increased the ATP yield in the presence of 1 mM Fe 3+ . This suggests that 258 either of these divalent cations can stabilise the newly formed ATP and liberate Fe 3+ to catalyse the 259 next phosphorylation of ADP (Fig. 5a). 260 We next conducted a kinetic study of the phosphorylation reaction, specifically varying the 261 ADP concentration and monitoring the reaction rate. The resulting curve resembled a characteristic 262 Michaelis-Menten mechanism for an enzyme, indicating that Fe 3+ does indeed act as a catalyst (Fig.  263 5b). The question remained whether a single Fe 3+ was interacting directly with a single ADP and AcP, 264 or whether larger units such as stacked ADP rings were involved. Stacking can alter the geometry of 265 which group interacts with Fe 3+ (SI Fig. 6) and has previously been suggested as a possible 266 mechanism[64]. However, MALDI-ToF analysis, which can sensitively detect stacked nucleotides, 267 showed no difference between the ADP control and the reaction sample; the main visible peaks 268 appeared to be dimers of ADP/AMP present in the commercial ADP standard, possibly due to freeze-269 drying during production of ADP [65] (Fig. 5c). This demonstrates that stacking of ADP to coordinate 270 the Fe 3+ ion does not occur as a mechanistic step in the reaction. That in turn constrains more tightly 271 which groups in the base could potentially interact with metal ions such as Fe 3+ . 272 Altogether, our results suggest that the high charge density of Fe 3+ allows it to interact 273 directly with the N6 amino group on the adenine ring, while anchoring AcP in position for its 274 phosphate group to interact with the diphosphate tail of ADP, giving a taut conformation of ADP 275 ( Fig. 6a). The interaction with the dianion has been proposed before [66,67] and is key because at 276 the optimal pH of 5.5-6, the first two hydroxyl groups of ADP (pKa 0.9 and 2.8) are deprotonated, 277 while the external OH group (pKa 6.8) remains protonated, and is therefore not available for 278 nucleophilic attack [68]. The interaction of the two deprotonated OH groups with Fe 3+ has the effect 279 of lowering the pKa of the outermost OH group, thus deprotonating it and enhancing its 280 nucleophilicity (Fig. 6b). The phosphate group of AcP is readily positioned for nucleophilic attack by 281 the newly deprotonated Oof ADP, forming ATP (Fig. 6c). This mechanism also explains why Ca 2+ and 282 Mg 2+ slightly increase the rate of reaction; these ions are able to displace Fe 3+ from the ATP product 283 (as they interact better with the triphosphate tail; Fig. 6d), freeing the Fe 3+ to catalyse further 284 reactions (Fig. 6e). Our results support the following conclusions: (i) acetyl phosphate (AcP) efficiently phosphorylates 288 ADP to ATP, but only in the presence of Fe 3+ ions as catalyst (Fig. 1); (ii) the reaction takes place in 289 water and can occur in a wide range of aqueous environments (Fig. 2); (iii) no other phosphorylating 290 agent tested was as effective as AcP (Fig. 3); and (iv) adenine is unique among canonical nucleobases 291 in facilitating the phosphorylation of its nucleotide diphosphate to the triphosphate (Fig. 4). Taken 292 together, these findings suggest that the pre-eminence of ATP in biology has its roots in aqueous 293 prebiotic chemistry. The substrate-level phosphorylation of ADP to ATP by AcP is uniquely facilitated 294 in water under prebiotic conditions and remains the fulcrum between thioester and phosphate 295 metabolism in bacteria and archaea today [2]. This implies that ATP became the universal energy 296 currency of life not as the endpoint of genetic selection or some frozen accident, but for 297 fundamental chemical reasons, and probably in a monomer world before the polymerization of RNA, 298 DNA and proteins. 299 The work presented here provides a compelling basis for each of these statements, but also 300 raises a number of questions. Why ferric iron? Unlike AcP or ATP itself there is no clear link with 301 biology in this case; we had expected other ions more commonly associated with nucleotides, 302 notably Mg 2+ or Ca 2+ [39,40], to play a more clear-cut role. In fact, their catalytic effect was only 303 noticeable in the presence of Fe 3+ , as has been reported before, whereas higher concentrations, 304 equivalent to modern ocean conditions, precluded ATP synthesis. We infer that the reason Fe 3+ plays 305 a unique role relates in part to its high charge density and small ionic radius. The fact that only ADP 306 could be phosphorylated among canonical nucleobases suggests that Fe 3+ interacts directly with the 307 N6 amino group on the adenine ring as well as the N7 previously noted by others [60][61][62][63]. But the 308 interactions between Fe 3+ and the N7 moiety alone cannot explain our results, as no triphosphate 309 was formed in the absence of the N6-amino group, for example in the case of GDP. The fact that ADP 310 is phosphorylated more readily than AMP (SI Fig. 5) indicates that Fe 3+ also interacts with the 311 diphosphate tail of ADP. And the fact that the optimal stoichiometry of Fe 3+ to ADP is 1:1, coupled 312 with the absence of evidence for stacking of bases by MALDI-ToF (Fig. 5), indicates that a single Fe 3+ 313 ion interacts with a single ADP, and necessarily also with a single AcP. 314 As shown in Fig. 6, these stipulations require an unusually taut molecular configuration of 315 ADP, far from the loose conformation usually depicted, if only for ease of presentation. The 316 orientation of the adenine ring in relation to metal ions has long been disputed, with some arguing 317 that it should face the opposite way in apposition to the phosphate tail [69]. Others have suggested 318 an equivalent orientation to that proposed here [67,70], some specifically with Fe 3+ [60,61]. In any 319 case, this taut conformation almost certainly requires the interacting ion to have a high charge 320 density and small ionic radius, to draw each of these groups into close enough proximity to react. 321 Among the cations tested here, Fe 3+ has the highest charge density and the smallest ionic radius 322 [71]. Nonetheless, some of the other ions studied, notably Cr 3+ and Co 3+ , have a similar ionic radius 323 and charge density, yet do not have a remotely comparable catalytic effect, so the size and charge 324 density cannot be the only explanation for our results. The electronic configuration of Fe 3+ may also 325 play a role: unlike Cr 3+ and Co 3+ , Fe 3+ has the electronic configuration [Ar]3d 5 , having all 5 d orbitals 326 half occupied. However, Mn 2+ , which can substitute Mg 2+ in the catalytic centre of acetate kinase, 327 has an equivalent 3d orbital, yet yielded negative results in our experiments. If so, then size, charge 328 density and electronic configuration might all play a role. These possibilities need to be explored in 329 future work. 330 Why acetyl phosphate? The idea that this small (2-carbon) molecule might have acted as a 331 phosphoryl donor at the origin of life has a long history, going back to Lipmann himself 332 [10,13,17,19,[24][25][26][27][28][29], as indeed does its confounding potential as an acetyl donor. Acetyl phosphate 333 still plays a global signalling and energy transduction role in bacteria [72], in part because its free 334 energy of hydrolysis (and therefore its phosphorylating potential) is greater than that of ATP (ΔG o´ = 335 -43 kJ mol -1 versus -31 kJ mol -1 , respectively). When complexed in a 1:1 ratio with ADP, therefore, 336 AcP has the potential to transfer its phosphate to form ATP, and so serves as a labile energy source 337 in cells, linked to the excretion of acetate as waste. But the actual change in ΔG depends on how far 338 from equilibrium the ratio of AcP/Ac + Pi or ATP/ADP + Pi has been pushed, and hence varies 339 depending on conditions. In our experiments, all phosphoryl donors were added at equivalent 340 excess. The fact that the ΔG o´ for hydrolysis of PEP (-62 kJ mol -1 ) and CP (-51 kJ mol -1 ) are markedly 341 greater than that for AcP means that free-energy change is only part of the explanation for the 342 efficacy of AcP. The fact that ATP was primarily formed by AcP in the presence of Fe 3+ ions instead 343 implies that the critical factors were (i) the position of the two phosphoester oxygen atoms in 344 relation to the Fe 3+ , and (ii) the phosphate group in relation to the diphosphate tail of ADP, as shown 345 in Fig. 6. In other words, both AcP and ADP are favoured not for selective or thermodynamic 346 reasons, but kinetic -because their chemistry is facilitated by molecular geometry in aqueous 347 prebiotic environments. 348 The only other molecule with equivalent geometry in this regard is carbamoyl phosphate 349 (CP), which our model would therefore predict should have some phosphorylating efficacy. CP was 350 indeed the only other species to show significant phosphorylating activity in our system (Fig. 3) own results, these findings suggest that both AcP and CP are molecular 'living fossils' of prebiotic 357 chemistry, retaining a role in modern metabolism due to their felicitous chemistry. But despite these 358 similarities, CP was less effective than AcP at generating ATP under mildly acidic to neutral 359 conditions (Fig. 3). This difference holds important connotations for its ability to power work. 360 A major question for prebiotic chemistry is how can an energy currency power work? As 361 noted in the Introduction, there is nothing special about the bonds in ATP; rather, the ATP synthase 362 powers a disequilibrium in the ratio of ADP to ATP, which amounts to 10 orders of magnitude from 363 equilibrium in the cytosol of modern cells. Only that disequilibrium powers work; no equilibrium 364 mixture of ATP and ADP can power anything. But molecular engines such as the ATP synthase use 365 ratchet-like mechanical mechanisms to convert environmental redox disequilibria into the highly 366 skewed ratio of ADP to ATP [81]. How could a simple prebiotic system composed of monomers 367 sustain a disequilibrium in the ratio of ATP to ADP that powers work? One possibility is that the 368 environment itself could sustain critical disequilibria across short distances, such as membranes. The 369 fundamental disequilibrium that drives work in essentially all cells is the proton-motive force -at its 370 simplest, the difference in proton concentration, or pH, across membranes. This mechanism is highly 371 relevant to ATP, given the strong dependence of ATP synthesis versus hydrolysis on pH, specifically 372 because the phosphorylation potential of ATP depends on its free energy of hydrolysis, which 373 increases with pH [82,83]. Far from being an environmental limitation, the narrow pH range 374 facilitating ATP synthesis reported here may therefore help to drive work in a monomer prebiotic 375 world. 376 Dynamic environments such as alkaline hydrothermal systems can sustain steep pH 377 gradients across thin inorganic barriers, as mildly acidic Hadean ocean waters (pH 5-6) continually 378 mix with strongly alkaline hydrothermal fluids (pH 9-11) in microporous labyrinths that operate as locally acidic conditions close to the barriers, followed by hydrolysis linked to phosphorylation under 389 more alkaline conditions in the cytosol of protocells. At face value, the ATP yield reported here at pH 390 5-5-6 after 10 hours was 17.4 % (corresponding to 156.5 µM) while the yield at pH 9 was 0.043 %, 391 corresponding to 0.4 µM, a difference of 400-fold. Thus, a geologically sustained difference in pH 392 across membranes could drive a local disequilibrium in the ATP/ADP ratio of 2-3 orders of 393 magnitude, enough to power work even in the absence of other possible factors such as 394 temperature. Higher temperatures (50 °C) promote both the rapid synthesis and hydrolysis of ATP 395 (Fig. 2b), which should amplify this driving force. We stress that these considerations require further 396 elucidation, but in principle steep pH gradients can drive a disequilibrium in the ATP/ADP ratio that 397 powers work. 398 Are these far-from-equilibrium conditions consistent with the high water-activity and low 399 ion requirements for optimal ATP synthesis in our experiments? High concentrations of Mg 2+ (50 400 mM) and Ca 2+ (10 mM) precluded ATP synthesis, implying that this chemistry would not be favoured 401 in modern oceans, but would be feasible in freshwater systems. Likewise, ferrous iron could be 402 oxidized to ferric iron by photochemical reactions or oxidants such as NO derived from volcanic 403 emissions, meteorite impacts or lightning strikes, which also points to terrestrial geothermal systems 404

Reaction setup 459
Depending on the solubility of the analytes, reactions were carried out in either a stationary (SciQuip 460 HP120-S) or a shaking (ThermoMix HM100-Pro) dry block heater. 461 For the reaction, stock solutions of di-nucleotides (sodium salts, ≥96%, Sigma-Aldrich), 462 phosphorylating agents and metal catalyst were freshly prepared as to avoid freeze-thawing (10 mM 463 for reactions to be analysed via HPLC, 1 M for reactions to be analysed via NMR). Except where 464 indicated, the ratios of analytes in a solution were 1(ADP):4(AcP) and 1(Fe 3+ ):2(ADP). When needed 465 the pH was adjusted using aqueous HCL and NaOH (1 M or 3 M) 466 After checking the pH (Fisher Scientific accumet AE150 meter with VWR semi-micro pH 467 electrode), samples were taken at time-points (0, 10 and 30 min, 1 to 5, 10 and 24h) and, unless 468 otherwise specified, immediately frozen at −80 °C for next-day analysis. Samples were prepared at collection by spinning at 4,000rpm for 2 minutes and diluting 200 µL in 495 800 µL of EDTA solution (500 µL in 100 mM PO4 buffer at pH 7.1) prior to freezing, in order to chelate 496 the Fe 3+ ions in solution that would otherwise block the HPLC column. 497 Thawed samples were filtered using syringe filters (ANP1322, 0.22 μm PTFE Syringe filter, Gilson 498 Scientific Ltd.) attached to a 1 mL sterile syringe (BD Plastipak Syringes) in 2 mL headspace vials and 499 analysed on an HPLC instrument (Agilent Technologies, 1260 Infinity II); peaks were identified using 500 pure standards. The wavelengths for UV detection were usually set at 254 nm and 260 nm (most 501 suitable for cyclic rings such as adenosine), while the column tray temperature was maintained at 502 room temperature. Two different columns were used depending on the pH of the sample being 503 analysed: Poroshell 120 EC-C18 for pH 2-8 and Poroshell HPH-C18 for pH 9-11. 504 Mobile phase A consisted of 80 mM phosphate buffer (made by mixing equal parts of 505 potassium phosphate dibasic (40 mM) and potassium phosphate monobasic (40 mM) salts dissolved 506 in water) adjusted to pH 5.8 using 3 M HCl and filtered with 0.2 μm nylon membrane filters 507 (GNWP04700, 0.2 μm pore size, Merck Millipore Ltd.), while mobile phase B consisted of 100% 508 methanol. The injection volume was 1 μL, with a flow rate of 1 mL/min, and the run was an isocratic 509 gradient that consisted of 95% B for 5 minutes. 510 For experiments using nucleotide diphosphates with different bases, analyses were carried 511 out on a Polaris C18-A column, with mobile phase A consisting of 10 mM potassium phosphate 512 monobasic buffer with 10 mM Tetrabutylammonium hydroxide (TBAH) adjusted to pH 8 using 3 M 513 HCl and filtered with 0.2 μm nylon membrane filters (GNWP04700, 0.2 μm pore size, Merck Millipore 514 Ltd.), while mobile phase B consisted of 100% methanol (method described in Table 2). The 515 wavelengths for UV detection were set at 254, 260, and 271 nm for guanosine, uridine and inosine, 516 and cytidine, respectively. 517 518  Table 3) were employed to preserve the column: Flush 1 was used every 12-15 samples, then three 523 rounds of Flush 1 followed by one run of Flush 2 were run prior to switching off the machine. 524 525 Flow rate 1 mL/min 1 mL/min 528 computational analysis was done using Agilent OpenLAB software (ChemStation Edition). Each peak 529 was manually integrated using the calibration curves as reference and the raw file was exported for 530 data manipulation. As residual ATP is present in the ADP commercial standard, the yield of the 531 reaction is calculated by subtracting the reading for ATP at timepoint 0 from all subsequent timepoint 532 readings. 533 534 31 P-NMR 535 As iron is paramagnetic and thus tampers with NMR spectra, samples prepared for 31 P-NMR were 536 purified using solid phase extraction (SPE) after thawing. The SPE cartridge (InertSep ME-1, 537 300mg/3mL) was equilibrated with 3mL of 100% methanol and then washed with 3mL H2O, after 538 which the sample was passed through and collected. The procedure was tested on control samples to 539 ensure appropriate recovery 540 A volume of 0.9 mL of purified sample was added to 0.1 mL of D2O and dispensed in an NMR 541 tube (Norell Standard Series 5mm Precision NMR Sampling Tubes) for analysis ( 1 H decoupling, Bruker 542 Avance 400 MHz, 52 scans). The data was processed using the Bruker TopSpin 4.0.7 software and 543 peaks were identified using pure standards. 544

ESI MS 546
Electrospray Ionisation Mass Spectrometry was used to confirm the identity of ATP through MS/MS. 547 After purification through SPE (see previous section) the sample was loaded into a 0.5 mL glass syringe 548 (Gastight Syringe Model 1750 RN, Hamilton) and directly infused into the mass spectrometer (Finnigan 549 LTQ Linear Ion Trap mass spectrometer) at a flow rate of 10 μL/min. To avoid contaminations, the 550 syringe and line were flushed with 100% methanol before and after sample infusion, and the spectra 551 recorded. 552 The mass spectrometer was operated in negative ion mode and the capillary voltage was set 553 at -16 V. Data were collected from 100 to 2000 m/z with an acquisition rate of 5 spectra per second. 554 For the MS/MS Ar was used as the collision gas and the collision energy was adjusted to 30 eV. The 555 software Xcalibur (Thermo Scientific) was used for method setup and data processing. 556 557

MALDI-ToF MS 558
Samples were thawed and desalted using a protocol adapted from Burcar et al.[101]. Two solvents 559 were prepared: an ACN solution consisting of 50% acetonitrile in water and a 0.1 M TEAA solution in 560 water. 561 Using a Millipore C18 zip tip (Sigma), 10 µL of ACN solution were aspirated and discarded 3 562 times. The three rinses were repeated with 10 µL of the TEAA solution. To allow for the retention of 563 the analyte by the zip tip matrix, 10 µL of sample were aspirated up and down eight times and then 564 discarded. A volume of 10 µL of water were aspirated and discarded, followed by 10 µL of the TEAA 565 solution and once again 10 µL of water. A volume of 4 µL of ACN were slowly aspirated up and down 566 three times and deposited into a small Eppendorf microcentrifuge tube. 567 The MALDI-ToF protocol used was designed by Whicher et al. [10]. The matrix consisted of 568 2,4,6-trihydroxyacetophenone monohydrate (THAP) and ammonium citrate dibasic, and was freshly 569 prepared before the analysis using equal volumes of stocks that were maintained at 4°C for a 570 maximum of a week. 571 A volume of 2 μL of matrix solution was mixed with 2 μL of sample, deposited onto a clean 572 steel MALDI-ToF plate and allowed to evaporate for 30 minutes before the introduction of the steel 573 plate into the instrument (Waters micro MX mass spectrometer). The analytical conditions were: 574 reflectron and negative ion mode, 400 au of laser power, 2000 V of pulse, 2500 V of the detector, 575 12,000 V of flight tube, 5200 V of reflector, 3738 V of negative anode, and 500-5000 amu of scan 576 range. The mass spectrometer was calibrated using a low-molecular-weight oligonucleotide standard 577 (comprising of a DNA 4-mer, 5-mer, 7-mer, 9-mer, and 11-mer (Bruker Daltonics)). Each 578 oligonucleotide standard was initially dissolved in 100 μL water, divided in aliquots and frozen at −80 579 °C. A fresh aliquot was used at each analytical calibration. 580 AcP catalysed by Fe 3+ at 30°C and pH ~5.5-6 at the beginning of the reaction (0 h, broken line, blue) 891 and after 3 hours (solid line, red). The molecular structure of each base forming the nucleotides is 892 shown. (f) 31 P-NMR spectrum of PPi (bottom, blue), PPPi (middle, red) and the reaction PPi (1 mM) + 893 AcP (4 mM) + Fe 3+ (500 μM) at 30°C and pH ~5.5-6 after 3h (top, green).