The synthesis of the complete set of canonical Watson–Crick base-pairing nucleosides [adenosine (A), cytidine (C), guanosine (G) and uridine (U)] under conditions that do not violate the accepted plausible geochemical environments on early Earth is an essential step towards elucidating the origins of life on Earth1,2,3. However, while many 'plausible' nucleoside candidates have been suggested to have played a role at the origins of life4,5,6, the concurrent prebiotic synthesis of a complete set of nucleoside monomers remains an unresolved challenge for any of the proposed genetic polymers [e.g., ribonucleic acid (RNA), arabinonucleic acid (ANA), threonucleic acid (TNA) and pyranosyl-ribonucleic acid (pRNA)]3,5,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24. Accordingly, we set out to elucidate chemical reactions that could address this problem. We have previously reported a prebiotic synthesis of pyrimidine ribonucleotides 1C and 1U16. More recently, we reported the divergent synthesis of 1C, 1U and 8-oxo-purine ribonucleotides 2A and 2I (Fig. 1, red arrows)22. However, no divergent prebiotic synthesis of pyrimidine and purine nucleoside monomers bearing the canonical Watson–Crick base-pairing nucleobases has yet been elucidated7,2123.

Fig. 1
figure 1

Divergent prebiotic nucleotides synthesis. Red arrows: Previous work; a prebiotic pathway to cytidine-2′,3′-cyclic phosphate (1C), uridine-2′,3′-cyclic phosphate (1U), 8-oxo-adenosine-2′,3′-cyclic phosphate (2A) and 8-oxo-inosine-2′,3′-cyclic phosphate (2I)16,22. Dashed arrow: Hydrolysis of 8,2′-anhydropurines (3A, 3I and 3G) is not observed, which provides chemical differentiation from 2,2′-anhydropyrimidine (3C) that readily hydrolyses to β-arabino-adenosine (ara-4C). Blue arrows: This work; a prebiotic pathway to β-arabino-cytidine (ara-4C), β-arabino-uridine (ara-4U), β-arabino-adenosine (ara-4A), β-arabino-inosine (ara-4I) and β-arabino-guanosine (ara-4G)

ANA displays many properties that make it an attractive candidate for the first genetic polymer of life. ANAs can equilibrate between helix and stem-loop structures, which mimic DNA and RNA, respectively25. ANA can form a complementary Watson–Crick base-paired duplex with RNA26,27, and can be readily transcribed (from DNA) and reverse transcribed (to DNA)28. Additionally, Holliger and co-workers recently evolved catalytic ANA-zymes that can achieve RNA phosphodiester cleavage6. Notably, the ANA phosphodiester backbone is also far more resistant to hydrolysis than its RNA analogue26.

Anhydrocytidine (3C), a key intermediate in our previously reported prebiotic pyrimidine synthesis16, undergoes hydrolysis at near neutral pH (≥6.5)22 to quantitatively afford arabino-cytidine (ara-4C; Fig. 1, blue arrow). This facile hydrolysis suggests that a simple prebiotic synthesis of arabino-nucleotides may be achievable. Importantly, the synthesis of a complete set of arabinosides requires differential reactivity between the purine and pyrimidine precursors 3A/3G and 3C (Fig. 1), respectively16,22. Pyrimidine ara-4C can be accessed by direct hydrolysis of 3C, whereas hydrolysis of 3A and 3G would furnish 8-oxo-purines (8-oxo-4A and 8-oxo-4G; Fig. 1, dashed arrow)22,29 rather than the desired purine nucleosides ara-4A and ara-4G. Therefore, it is of note that purine precursors 3A22, 3I and 3G are highly resistant to alkaline hydrolysis, even in extremely alkaline (pH >12) solutions. Accordingly, we viewed this subtle, yet pronounced, difference in reactivity as an ideal source for chemical differentiation that could be exploited while building the canonical nucleobases on a preformed furanosyl-sugar scaffold. We envisaged sulfur—a critical element in the development of divergent ribonucleoside syntheses22, with widespread use in prebiotic chemistry22,30,31,32,33,34,35—would hold the key to site-selective purine reduction.

Here, we demonstrate a divergent route to synthesise a complete set of canonical (A, G, C and U) nucleosides from one plausibly prebiotic reaction sequence. Interestingly, photochemical reduction of (intermediate) 8-mercaptopurines is observed to be highly efficient for the desired canonical purines (A and G), but not the non-canonical purine inosine (I). The facile prebiotic synthesis of ANA indicates that it may have played an important role during the origins of life, and the selective photochemical reduction of 8-mercaptopurines provides a physical mechanism for prebiotic nucleobase selection en route to the Watson–Crick base-pairing nucleosides.


8-Mercaptopurine synthesis

We suspected that addition of hydrogen sulfide (H2S) to 3A, 3G and 3I in water would selectively introduce sulfur at the C8-carbon atom, and consequently direct regiospecific reduction of the canonical purine nucleobases on the preformed sugar scaffold (Fig. 1, blue arrows). Previously, Ikehara and Ogiso established that 3A reacts with liquid H2S in pyridine at 100 °C (sealed in a steel tube) to afford 8-mercapto-arabino-adenosine (ara-5A)36. However, 3A, 3G and 3I were remarkably stable to alkaline hydrolysis (Supplementary Figs. 1 and 2). Therefore, we began our investigation by exploring the mild, plausibly prebiotic, aqueous thiolysis of the prebiotic purine precursor 3A22,33,35. Incubation of H2S (670 mM) with 3A (35.7 mM) in water (pH 7, 60 °C, 7 d) afforded remarkably clean conversion of 3A to ara-5A (73%; Fig. 2, Table 1 and Supplementary Fig. 3).

Fig. 2
figure 2

Prebiotic synthesis of arabino-adenosine (ara-4A). 1H NMR spectra (600 MHz, 9:1 H2O/D2O, 25 °C, δ = 5.5–9.0 ppm) showing: a anhydroadenosine (3A; red). b Crude 8-mercapto-arabino-adenosine (ara-5A; blue) observed upon incubation of anhydroadenosine (3A; 35.7 mM) and H2S (670 mM, 60 °C, 7 d, pH 7). c Crude arabino-adenosine (ara-4A; purple) after irradiation (λ = 300 nm) of 8-mercapto-arabino-adenosine (ara-5A; 2 mM, room temperature, 16 h, pH 7)

Table 1 Percentage conversions for nucleoside reactions

Intrigued by the remarkably clean, high-conversion thiolysis of 3A, we next investigated the thiolysis (pH 7, 60 °C, 7 d) of 3G and 3I. We observed ara-5G (83%) and ara-5I (78%; 71% isolated yield after 5 d) (Table 1 and Supplementary Figs. 4 and 5). Next, 1:1 3A/3I was subjected to thiolysis and gave a high conversion to both ara-5A (66%) and ara-5I (75%) (Supplementary Fig. 6). Even under these mild, aqueous conditions, highly efficient thiolysis of 3A, 3G and 3I was observed, demonstrating that sulfur can be regioselectively introduced to the C8-carbon atom of purines under plausibly prebiotic conditions.

Photochemical purine reduction

For purine reduction, we initially investigated the effect of UV light on 8-mercaptopurines 5, because the atmosphere of the early Earth lacked an ozone layer37, allowing UV light (λ > 204 nm) to irradiate Earth’s surface38,39. We envisaged that UV irradiation of ara-5A would lead to π–π* excitation, followed by C–S bond fragmentation to afford an N-heterocyclic carbene tautomer of ara-4A. Pleasingly, when ara-5A (2.00 mM, pH 7) was irradiated (λ = 300 or 254 nm) in water we observed extremely clean conversion to ara-4A in 66% and 70%, respectively, after 16 h (Fig. 2 and Supplementary Figs. 79). Notably, ara-4A was the only nucleoside product observed after irradiation. Irradiation (λ = 300 nm) of the inosine analogue ara-5I gave arabino-inosine ara-4I (Supplementary Fig. 10). However, upon complete consumption of ara-5I, the observed conversion to ara-4I (15%) was significantly lower than for ara-4A (66%) (Table 1). Next, the two-step thiolysis/irradiation sequence was investigated for 3G. Anhydronucleoside 3G (5.85 mM) was thiolysed (H2S, pH 7, 60 °C, 7 d) and then irradiated (λ = 300 nm). The reaction was monitored until complete consumption of the intermediate (ara-5G) was observed (Supplementary Fig. 11). Once again, we observed a good conversion (59% after 22 h; Table 1) to the desired product arabino-guanosine (ara-4G). Irradiation of 1:1 ara-5A/ara-5I confirmed the disparity between the conversions observed for the reduction to yield the canonical nucleobases A and G and wobble base-pairing I: 62% and 13% conversion to ara-4A and ara-4I, respectively, was observed (Supplementary Fig. 12). Notably, the irradiation (λ = 300 or 254 nm) of 1:1 ara-4A/ara-4I (Supplementary Fig. 13) demonstrated equal product stability with respect to UV irradiation (e.g., 73% each after 16 h at λ = 300 nm), and therefore did not account for the intriguing differential percentage conversion observed between the canonical (A and G) and non-canonical (I) 8-mercapto-nucleosides. Subsequently, we irradiated the ribo-mercaptopurines ribo-5A, ribo-5G and ribo-5I, and observed highly efficient photo-reduction of the A and G ribo-nucleosides, but poor yielding reduction of the I ribo-nucleoside (Table 1 and Supplementary Figs. 1416).

Quantum chemical studies

The observed difference in the photo-reduction of 5A and 5G compared to 5I cannot be directly connected to any remarkable differences in their UV-absorption features (Supplementary Figs. 117119). However, quantum chemical studies [ADC(2)]40,41 and femtosecond transient absorption spectroscopy (FTAS) suggest that after initial UV excitation these purines undergo different singlet-to-triplet decay pathways.

UV excitation of mercaptopurines (5) was calculated to result in the population of low-lying excited singlet states (ππCS*, ππRING* and nπCS*), which all exhibit significant spin–orbit coupling (SOC) with the triplet manifold (Supplementary Table 2 and Supplementary Discussion). Therefore, these singlet states are expected to undergo efficient intersystem crossing (ISC) to populate the triplet excited states of 5, as noted for other thionucleosides and thionucleobases42,43,44,45. Two important triplet minima were identified in our calculations (Fig. 3). First, the T1(3ππCS*) triplet minimum (Fig. 3a), which leads to C8-thiocarbonyl elongation (≈0.1 Å) and localisation of unpaired electrons on the sp3-hybridised C8-carbon and sulfur atoms. Second, the T1(3ππRING*) triplet minimum (Fig. 3a), which leads to pyrimidine ring puckering and localisation of unpaired electrons on the partially sp3-hybridised C2- and C5-carbon atoms. The ara-5A 3ππring* triplet is significantly higher in energy than the ara-5A 3ππCS* triplet; consequently, population of the 3ππCS* triplet is predicted to be the dominant pathway during ISC following UV excitation of ara-5A (Fig. 3d). Accordingly, we expect that C8-photo-reduction is triggered in the 3ππCS* triplet state of 8-mercaptopurines 5. In contrast to ara-5A, which is expected to only populate the 3ππCS* triplet state, ara-5I and ara-5G are predicted to populate both triplet states (3ππCS* and 3ππring*). Optimisation of the mercaptopurine excited state geometries indicated that the T1 hypersurface of ara-5I (Fig. 3c) and ara-5G (Supplementary Fig. 51) are similar in topography to those reported for thiopyrimidine nucleobases46. Consequently, we expected ara-5I and ara-5G to exhibit similar UV excitation behaviour to thiopyrimidines, which are observed to undergo two competing excited state decay pathways46.

Fig. 3
figure 3

Photochemical properties of ara-5I and ara-5A. a Molecular orbitals for T1 states (3ππCS* and 3ππring*) of ara-5I. b Calculated potential energy profile of ara-5A T1 3ππCS* state C–S bond homolysis furnishing triplet sulfur atom and singlet C8-carbene tautomer of ara-4A. Blue line = T1-state energy, black line = ground-state energy. The minimum-energy path along the C–S distance (Å) was obtained by calculation at the ADC(2)/cc-pVTZ level of theory. c Calculated ara-5I triplet excited state T1 topography. Parabolas fitted to calculated T1 minima energies and optimised T2/T1 minimum-energy state crossing, which represents the transition state between these minima. d Calculated ara-5A triplet excited state T1 topography. No ring-puckered minimum was found on the T1 hypersurface due to the higher 3ππring* state energy. ef FTAS recorded between λ = 340 and 530 nm with λ = 290 nm excitation (pump) pulses. e FTAS of ara-5I recorded between λ = 340 and 530 nm with λ = 290 nm excitation (pump) pulses. Inset: Calculated excited state absorption spectra for the two possible configurations of ara-5I. f FTAS of ara-5A recorded between λ = 340 and 530 nm with λ = 290 nm excitation (pump) pulses. Inset: Calculated excited state absorption spectra for the singlet and triplet states of ara-5A

Short excited state lifetimes are expected for the redox active T1(3ππCS*) triplet states, due to their very large SOC values (ara-5A: 99.5 cm−1; ara-5I: 81.4 cm−1) and low (0.23 eV) state-crossing barrier to the electronic ground state (S0). Conversely, the ara-5I T1(3ππRING*) triplet state exhibits a very low SOC value (1.44 cm−1) and is predicted to be much longer lived. Moreover, the ara-5I T1(3ππRING*) triplet minimum is estimated to be 0.2 eV lower in energy than the redox active ara-5I T1(3ππCS*) triplet minimum, and predominant population of the non-redox active T1(3ππRING*) triplet minimum is expected following UV excitation of ara-5I because these two states are only separated by a moderate (0.55 eV) energy barrier.

Upon first inspection it might seem surprising that, like ara-5I, both triplet states can be populated in ara-5G given its observed efficient photo-reduction. However, the ara-5G T1(3ππRING*) and T1(3ππCS*) triplet minima are calculated to be nearly isoenergetic, and therefore these minima are expected to easily interconvert. Accordingly, efficient photo-reduction of ara-5G is thought to occur by continual repopulation of the redox active ara-5G T1(3ππCS*) triplet minimum.

Femtosecond transient absorption spectroscopy

We next sought to verify our theoretical predictions by FTAS for ara-5I (Fig. 3e) and ara-5A (Fig. 3f). The absorbance (λ = 380–450 nm) observed in the first picosecond following ara-5A excitation matches the position and structure of the excited state absorption (ESA) spectrum simulated from the S1(1ππCS*) singlet minimum of ara-5A (Fig. 3f, inset). Initial singlet (1ππCS*) state population following photo-excitation is also consistent with our calculated vertical excitation energies (Supplementary Table 1), which indicate that λ = 290–300 nm excitation would primarily populate this state (Fig. 3e). At probe time delays >2 ps after incident excitation of ara-5A, the absorbance evolved to cover a broader range of probe wavelengths, and the FTAS spectrum exhibits two characteristic features near λ = 360 and 480 nm. Although these features are partly covered by the bleaching bands, they match the simulated ara-5A T1(3ππCS*) triplet state ESA spectrum (Fig. 3f, inset), which indicates population of this triplet state. The observed excited state lifetime of ara-5A (≈70 ns at λ = 510 nm) is also consistent with the predicted efficient photo-relaxation of this triplet state, which is characterised by high SOC with the electronic ground state.

Although excitation at λ = 290 nm could potentially populate both ara-5I singlet states (i.e., 1ππCS* and 1ππring*), we did not observe any features in the FTAS spectrum that would correspond to the 1ππCS* singlet state. No significant change in absorbance was recorded during the first 0.4 ps (Fig. 3e). We were unable to simulate the ESA spectrum for the ara-5I S1(1ππring*) singlet state (its minimum-energy geometry coincided with the S1/S0 conical intersection), but we anticipate that population of this singlet state dominates the 0.4 ps immediately after ara-5I excitation. The emergence of a sharp band at λ = 350 nm between 0.6 and 1.0 ps was assigned to population of the redox active T1(3ππCS*) triplet state, which could be efficiently accessed from the initial S1(1ππring*) singlet state because of the molecular orbital change associated with an S1 → T1 transition47. The redox active ara-5I T1(3ππCS*) triplet minimum is observed to be efficiently depopulated over the next 6 ps. The excited state population transfers to the lower-energy ara-5I T1(3ππring*) triplet minimum (Fig. 3e), which is confirmed by the long excited state lifetime (≈12 μs at λ = 440 nm) and the two characteristic bands (λmax = 350 and 490 nm) in the ara-5I FTAS measurements, which are both consistent with the simulated ESA spectrum for the redox inactive triplet state (Fig. 3e, inset). The estimated excited state lifetime associated with the T1(3ππring*) triplet state is thought to be sufficient to enable bimolecular reactions and, consequently, the photochemical degradation pathways in ara-5I, which are not observed for ara-5A or ara-5G. Photo-reduction appears to correlate directly with population of the T1(3ππCS*) triplet state for ara-5A, and we anticipate that the same triplet state would be responsible for photo-reduction in ara-5I and ara-5G. In particular, the observed short-lived population of the ara-5I T1(3ππCS*) triplet state (0.6–7.0 ps) likely explains the poor photo-reduction observed for this nucleoside.

It appears probable that C–S bond homolysis in ara-5A, initiated by population of the T1(3ππCS*) triplet minimum, would furnish the singlet C8-carbene tautomer of purine ara-4A (Fig. 3b). Our calculations indicate that C–S bond homolysis requires 0.9 eV to liberate the triplet sulfur atom. Excitation at λ = 300 nm would excite ara-5A ~1.2 eV above the T1(3ππCS*) triplet minimum, which would be sufficient to promote efficient photo-reduction. In principle, C–S bond homolysis (Fig. 3b) would be reversible, however tautomerisation of the intermediate C8-carbene to ara-4 would prevent reformation of ara-5.

8-Mercaptopurine oxidation

Encouraged by the photochemical reduction of ara-5A, ara-5G and ara-5I, we investigated how to increase the efficiency of purine reduction. Simplicity and prebiotic plausibility led us to consider hydrogen peroxide (H2O2) in nucleobase reduction48,49,50. In our hands, the reported reaction of H2O2 and ara-5A in acidic methanol36 gave ara-4A (50% after 16 h, Supplementary Fig. 17). Pleasingly, oxidative disproportionation of ara-5A (50 mM) and ara-5I (50 mM) under plausibly prebiotic conditions (H2O2 (3 equiv), pH 7, water, room temperature, 3 h) afforded excellent conversion to ara-4A (85%) and ara-4I (90%), respectively (Table 1 and Supplementary Figs. 18 and 19). Reaction of 1:1 ara-5A/ara-5I afforded ara-4A (88%) and ara-4I (91%) (Supplementary Fig. 20). Next, we investigated the one-pot thiolysis/oxidation of 3G. Incubation of 3G (35 mM) with H2S (714 mM) in water (pH 7, 60 °C, 7 d) then addition of H2O2 (375 µmol; pH 7, room temperature, 2 h) furnished ara-4G (89%) (Table 1 and Supplementary Fig.  21).

Pyrimidine thiolysis

Having demonstrated the efficient conversion of 3A, 3G and 3I to arabino-purines ara-4A, ara-4G and ara-4I, respectively, we next investigated the reactivity of pyrimidine precursor 3C under comparable conditions. Sutherland and co-workers reported that the reaction of α-ribo-anhydrocytidine (α-ribo-3C) with sodium hydrogen sulfide (NaHS) in formamide at 50 °C yields 2-thio-α-cytidine (α-ribo-6C; 84%; Fig. 4b.iv)35. Similarly, upon submitting 3C to these conditions (NaHS, formamide, 50 °C), we observed conversion to ara-4C (24%) and 2-thio-arabino-cytidine (ara-6C; 31%) (Fig. 4a.iv; Supplementary Fig. 24). Conversely, incubation of 3C (28 mM) with H2S (0.14 M) in water (pH 7, 60 °C) yielded mostly ara-4C (68%) alongside some conversion of ara-4C to 4-thio-arabino-uridine (ara-7U; 22%) after 2 d (Fig. 4a.i–ii; Supplementary Fig. 25); <5% ara-6C was observed. Further conversion of ara-4C to ara-7U was observed upon prolonged incubation (57:38 ara-4C/ara-7U after 7 d; Fig. 4a.ii). Although in formamide the addition of NaHS to 3C furnishes ara-6C, in water rapid hydrolysis of 3C to ara-4C occurs, which subsequently undergoes slow nucleophilic substitution at the C4-carbon atom to furnish ara-7U.

Fig. 4
figure 4

Thiolysis and oxidation of pyrimidine nucleosides. a i. H2S thiolysis of arabino-anhydrocytidine (3C) in water (pH 7) furnished β-arabino-cytidine (ara-4C; quant.) rapidly; ii. ara-4C then undergoes slow addition of H2S to afford 4-thio-β-arabino-uridine (ara-7U). iii. H2O2 oxidation of ara-7U in water (pH 7) furnished arabino-uridine (ara-4U, 78%). iv. Thiolysis of 3C in formamide yielded 2-thio-β-arabino-cytosine (ara-6C, 31%). v. H2O2 oxidation of ara-6C in water (pH 7) furnished 3C (quant.). b i–ii. H2S thiolysis of ribo-anhydrocytidine (α-ribo-3C) in water (pH 7) furnished α-ribo-cytidine (α-ribo-4C), which then undergoes addition of H2S to afford 4-thio-α-ribo-uridine (α-ribo-7U) in 91% combined conversion. iii. H2O2 oxidation of α-ribo-7U in water (pH 7) furnished α-ribo-uridine (α-ribo-4U, 93%). iv–v. Sutherland and co-workers reported the synthesis of 2-thio-β-ribo-cytidine (β-ribo-6C) by thiolysis of ribo-anhydrocytidine (α-ribo-3C) in formamide, followed by aqueous irradiation (λ = 254 nm)35. vi. H2O2 oxidation of 2-thio-β-ribo-cytosine (β-ribo-6C) furnished canonical ribonucleoside β-cytidine (β-ribo-4C, quant.) between pH 7 and 9. vii. Conversely, H2O2 oxidation of β-ribo-6C predominately furnished pyridimide 4-amino-pyrimidine-riboside (β-ribo-9, 76%) at pH 3. viii. Due to the proximity of the nucleobase and C2′-hydroxyl moieties, H2O2 oxidation of α-ribo-6C furnished α-ribo-3C (quant.). ix–x. The sequential reaction of H2S and H2O2 with β-ribo-4C in water (pH 7) furnished β-ribo-4U

The observed C4-thiolysis of ara-4C to ara-7U is not limited to the β-arabino-stereochemistry shown in Fig. 4a.ii. Submitting α-ribo-3C (28 mM) to the same aqueous thiolysis conditions afforded α-ribo-cytidine (α-ribo-4C, 52%; Fig. 4b.ii) and 4-thio-α-ribo-uridine (α-ribo-7U, 39%; Fig. 4b.ix). Incubation of ara-4C, β-ribo-4C and α-ribo-4C with H2S (pH 7) led to partial (unoptimised) conversion to the 4-thiouridines ara-7U (7 d, 40%), β-ribo-7U (7 d, 16%) and α-ribo-7U (8 d, 63%), respectively (Table 1), alongside recovered cytidine starting material [ara-4C (52%), β-ribo-4C (84%) and α-ribo-4C (37%)] (Supplementary Figs. 2830). We were surprised to observe slower C4-thiolysis for the canonical trans-1′,2′-isomer (β-ribo-4C) than for the cis-1′,2′-isomers (α-ribo-4C and ara-4C). Simultaneous thiolysis of 1:1 ara-4C/β-ribo-4C verified this rate difference; after 7 d more ara-7U (52%) than β-ribo-7U (18%) was observed (Supplementary Fig. 31), indicating the relative nucleobase/C2′-OH orientation affects the rate of nucleophilic substitution at the distal C4-position of cytidine nucleotides. Finally, we incubated pyrimidine 3C and purine 3A (1:1) with H2S (pH 7, 60 °C, 7 d). Gratifyingly, ara-4C (70%), ara-7U (30%) and ara-5A (71%) were cleanly furnished (Supplementary Fig. 27).

Pyrimidine oxidation

An efficient protocol to convert ara-7U to ara-4U would indicate that nucleosides ara-4C and ara-4U could be readily generated in comparable yields from thiolysis of 3C alongside conversion of 3A/3G to ara-4A/ara-4G, respectively. We envisioned that the thiocarbonyl moiety of ara-7U would be readily oxidised, however we suspected that oxidation would activate the pyrimidine nucleobase to hydrolysis (rather than reduction as observed for the purines). To test our hypothesis, we incubated ara-7U (50 mM), β-ribo-7U (50 mM) and α-ribo-7U (50 mM) with H2O2 (0.15 m) in water (pH 7, room temperature, 3.5 h). Each reaction proceeded smoothly to give the respective uridine (ara-4U (78%); β-ribo-4U (78%); α-ribo-4U (93%); Table 1, Fig. 4 and Supplementary Figs. 38, 4041), which validated our prediction. We next investigated thiolysis and selective hydrolysis in a one-pot reaction: 3C (35.7 mM) was incubated with H2S (0.71 M) in water (pH 7, 60 °C, 7 d) and then H2O2 (375 µmol) was added. We observed concomitant formation of ara-4C (62%) and ara-4U (25%) (Supplementary Fig. 39).

The reactions of 2-thiocytidines ara-6C and α-ribo-6C with H2O2 (150 mM) in water (pH 7, room temperature, 2 h) cleanly regenerated anhydrocytidines 3C (82%; Fig. 4a.v) and α-ribo-3C (80%; Fig. 4b.viii), respectively, demonstrating a clear switch in reactivity relative to ara-7U due to the proximity of the C2′-hydroxyl and thiocarbonyl moieties. Therefore, it is of note that canonical cytidine β-ribo-6C has an anti-1′,2′-disposition, and we expected to observe hydrolysis during H2O2 oxidation of β-ribo-6C. As expected, incubation of β-ribo-6C (38.4 mM) with H2O2 (232 mM) in phosphate buffer (pH 7–9, room temperature, 7 h) afforded a quantitative conversion to β-ribo-4C (Fig.; Supplementary Fig. 44). Interestingly, buffering the oxidation at pH 3 with glycine afforded β-ribo-4C (24%) alongside 4-amino-pyrimidine-riboside (β-ribo-9; 76%; Fig. 4b.vii; Supplementary Fig. 44). It is likely that cytidine protonation switches on C2-reduction by promoting access to the carbene intermediate required for reduction (Fig. 4b.vii)51. This hypothesis is supported by the simultaneous oxidation of ara-7U and β-ribo-6C with H2O2 (pH 3, glycine buffer, room temperature), which affords β-ribo-4C (20%) and β-ribo-9 (72%) from β-ribo-6C, but exclusively ara-4U (92%) from ara-7U (Supplementary Fig. 45).

It is particularly interesting, with respect to the origins of life, that purine C8-reduction is facile and quantitative at neutral pH, whereas pyrimidine C2-reduction is only observed at low pH. Consequently, the reaction of β-ribo-6C with H2O2 at neutral pH furnishes canonical β-ribo-4C cleanly, but protects against, and even reverses, the formation of non-canonical ara-6C and α-ribo-6C. This provides a facile and selective method to convert β-ribo-6C35 to canonical nucleoside β-ribo-4C.

Concomitant purine and pyrimidine synthesis

To form information-rich ANA nucleic acid oligomers, all four Watson–Crick base-pairing nucleosides (ara-4A, ara-4C, ara-4G and ara-4U) need to accrue at the same time in the same environment, ideally from the same set of chemical reactions. We reacted 1:1:1 3C, 3A and 3G with H2S (pH 7, 60 °C; Supplementary Fig. 46), and after 7 d we observed ara-5A (65%), ara-5G (62%), ara-4C (55%) and ara-7U (35%). Subsequent incubation with H2O2 (pH 7, room temperature, 2 h) gave the desired Watson–Crick base-pairing products ara-4A (53%), ara-4G (62%), ara-4C (47%) and ara-4U (35%) without purification or isolation of intermediate products (Fig. 5). Thus, we achieved a plausibly prebiotic divergent synthesis of Watson–Crick base-pairing nucleosides, and have marked ANA as a likely candidate for the nucleic acid of early evolution.

Fig. 5
figure 5

Concomitant synthesis of four Watson–Crick base-pairing nucleosides. a One-pot reaction of an equimolar mixture of anhydroadenosine (3A), anhydroguanosine (3G) and anhydrocytidine (3C) with H2S (20 equiv., pH 7) in water, then H2O2 (20 equiv., pH 7) in water affords adenosine ara-4A (53%), guanosine ara-4G (62%), cytidine ara-4C (47%) and uridine ara-4U (35%). The isourea moieties (blue) of the anhydropurines (3A/G) and anhydropyrimidine (3C) were differentiated in situ to afford the canonical Watson–Crick base-pairing nucleobase moieties regiospecifically on a furanosyl-sugar scaffold, by reduction (orange) and hydrolysis (magenta), respectively. b 1H NMR spectrum (600 MHz, 9:1 H2O/D2O, 25 °C, δ = 5.5–7.0 ppm) showing the prebiotic synthesis of cytidine ara-4C, uridine ara-4U, adenosine ara-4A and guanosine ara-4G following the sequential addition of H2S and H2O2 to anhydrocytidine 3C, anhydroadenosine 3A and anhydroguanosine 3G in water. c 1H NMR spectrum (600 MHz, 9:1 H2O/D2O, 25 °C, δ = 5.5–7.0 ppm) of substrates anhydrocytidine 3C, anhydroadenosine 3A and anhydroguanosine 3G in water


Through the reaction of pyrimidine and purine precursors (3C and 3A, respectively), which can be accessed divergently from a single prebiotic substrate16,22, we have elucidated an efficient plausibly prebiotic method to simultaneously form arabino-pyrimidine and arabino-purine nucleosides. Both the purine and pyrimidine nucleosides are delivered with regiospecific glycosylation on a furanose-specific arabinose sugar moiety. The reaction of H2S in water unlocks a purine-specific C8-reduction, by either UV irradiation or H2O2 oxidation; H2O2 oxidation is especially high yielding, however, UV irradiation caused selective destruction of ara-5I/β-ribo-5I over ara-5A/β-ribo-5A and ara-5G/β-ribo-5G. The remarkable purity of the arabino-purines delivered by either route is highly encouraging, but the direct mechanism for canonical purine nucleobases selection by UV light provides a mechanism (based in the physical properties of the purine bases) for nucleobase selection prior to their incorporation into nucleic acid biopolymers at the origins of life.

In water H2S also undergoes selective, but slow, addition to the cytidine C4-position, and—when coupled with oxidative hydrolysis—this provides a selective mechanism to convert cytidines to uridines. Importantly, H2S did not convert A/G to I/xanthine (X) demonstrating another remarkable, but essential, reactivity difference between the canonical nucleosides.

Although our focus has been on developing new chemistry, not on assessing geochemical boundary conditions, we note that the availability of H2S and H2O2 are both plausibly prebiotic, and although their sequential reaction implicates different redox states this is not geochemically implausible. The entire planet is at redox disequilibrium, and layered redox gradients are geochemically common. Outgassing volatile compounds (from Earth’s interior) are likely to have played a critical role in determining prebiotic chemistry, and local redox environments (for example, due to volcanic outgassing, meteorite impacts, photo-oxidation and atmospheric water dissociation and hydrogen escape) are expected to provide significant variation from the global average52. H2S is produced through a distinct mechanism from H2O2, thus discrete redox zones could be readily established. The oxidation state of early Earth (Hadean) magmas are not well constrained, however, zircons that pre-date the known rock record suggest average oxygen fugacities may have been similar to the present-day conditions53. Accordingly, sulfide-leaching and H2S-outgassing would have occurred on the early Earth, with sulfur outgassing rates as high as 1011.5 cm−2 s−1 possible during major volcanogenic emplacement of basaltic plains54. Equally, photo-dissociation generates H2O2 from water, and H2O2 is a likely key environmental oxidant prior to global oxidation37,50,55. It seems reasonable to suppose that distinct (redox) reactivity could be controlled by geochemical localisation and the different physicochemical processes that yield and accumulate feedstock molecules. Geochemical H2S outgassing and atmospheric H2O2 production are well suited to this specific localisation. Moreover, it has recently been proposed that H2O2 can accumulate within ices in an anoxic atmosphere56. Melting H2O2-rich ices could augment H2O2 delivery into aqueous environments and, in principle, could help to provide a mechanism for the sequential delivery of H2S and H2O2 into, for example, a flowing stream system or pool. Although the reactions of H2S and H2O2 must occur in sequence to achieve purine reduction by the described oxidative mechanism, it is of note that the purine anhydronucleotides (3A and 3G) were observed to be stable to H2O2 oxidation and the reduced purines (ara-4A and ara-4G) were observed to be stable to H2S addition. Therefore, cycling material between these redox conditions (or across this redox gradient) is not considered to be problematic for this chemistry; indeed, a one-pot two-step H2S/H2O2 reduction has been demonstrated. Importantly, photochemical mercaptopurine reduction does not require two redox states, only H2S and UV light (at λ = 250–300 nm), both of which are expected to be in adequate (simultaneous) supply on the early Earth when sulfur outgassing rates are less than 1011.5 cm−2 s−1. The localised enhancement in H2S can be achieved in surface hydrothermal systems with shallow water reservoirs (to allow UV penetration)54. In terms of the UV-light dose environment, we note that in the laboratory only six of the available sixteen lamps in the Rayonet irradiation chamber were employed, which corresponds to 39,000 erg s−1 cm−2 and 30,900 erg s−1 cm−2 at λ = 254 nm and 300 nm, respectively. The integrated surface flux (λ = 200–300 nm) delivered to the Earth by the early Sun is estimated to be about 2700 erg s−1 cm−2 (and within roughly an order of magnitude of the experimental apparatus), which provides ample flux to achieve the desired transformations and preserve prebiotic plausibility39. Crucially, the two distinct mercaptopurine 5 reductions demonstrate, not only the specific value of sulfur in prebiotic nucleoside synthesis, but also, more generally, the value of chemical redundancy. Two mechanisms for mercaptopurine reduction that operate under different conditions but both furnish the same purine products renders the overall transformation more robust, which may be especially important to consider in the context of the origins of life. Chemical redundancy is highly likely to improve pathway or network robustness, which may be an essential feature of sustained proto-metabolism in a (geo)chemically fluctuating environment3,18.

The different reactions observed between 3C and 3A/G with H2S is ideally suited to concomitant synthesis of the canonical nucleobases on preformed sugar scaffolds. Although the arabino-stereochemistry is not found in extant genetics, it is highly plausible that ANA could have been a precursor to RNA in early life or that early co-evolution of mixed RNA/ANA systems could have been superseded by RNA/DNA systems. The simplicity and efficiency of the synthesis of A, C, G and U arabinosides indicates that further investigations into the synthesis of 3G and the potential for ANA and RNA co-evolution are both warranted.


General procedure A. Thiolysis

Sodium hydrosulfide or disodium sulfide nonahydrate (20 equiv.) was dissolved in H2O/D2O (9:1) at pH 7. Nucleoside(s) (1 equiv.) was added and incubated at pH 7 and 60 °C. The reaction progress was monitored periodically by NMR spectroscopy. Upon completion, the reaction was cooled to room temperature and sparged of hydrogen sulfide with nitrogen or argon gas. The solution was adjusted to pH 6.5 and analysed by NMR spectroscopy.

General procedure B. UV irradiation

A degassed aqueous solution of nucleoside(s) (2.00 mM, pH 6.5) was irradiated in a Rayonet reactor (SNE Ultraviolet Co.) housing six RPR-2537A or RPR-3000A lamps (with principal emission at λ = 254 nm and λ = 300 nm, respectively) at 38 °C under an argon atmosphere. After irradiation, the reaction was allowed to cool to room temperature, and then lyophilised. The lyophilisate was dissolved in D2O (500 µL) and analysed by NMR spectroscopy. A solution of potassium hydrogen phthalate (0.100 M, 50.0 µL, 5.00 µmol in D2O) was added as an internal NMR standard and NMR spectra were reacquired.

General procedure C. Nucleoside oxidation

Nucleoside(s) (1.0 equiv., 50 mM) and potassium hydrogen phthalate (0.2 equiv.) were dissolved in H2O/D2O (9:1) or buffer solution. The pH of the solution was adjusted to pH 7 and NMR spectra were acquired. Hydrogen peroxide (30% w/w solution in H2O) was added and NMR spectra were periodically acquired at pH 7.

Computational methods

The vertical excitation energies, excited state geometries and excited state harmonic vibrational frequencies of ara-5I, ara-5A and ara-5G were computed by using the algebraic diagrammatic construction to the second-order method [ADC(2)]40,41,57, and the cc-pVTZ basis set. The MP2/cc-pVTZ method was used to optimise the corresponding ground-state geometries. Spin–orbit coupling matrix elements were calculated at the CASPT2/SA-CASSCF(10,9)/cc-pVTZ-DK level58, including the second-order Douglas–Kroll–Hess transformation to account for scalar relativistic effects. Calculated ESA spectra were obtained by convolution of vertical excitation energies and oscillator strengths with normalised Gaussian functions (0.20 eV half-width). Vertical excitation energies necessary for the ESA spectra were calculated from the corresponding excited state minima and 11 excitations were taken into account in each case. ESA cross-sections were generated using the GaussSum programme59. Minimum-energy crossing point (MECPs) geometries were optimised with the in-house implementation of the method proposed by Levine, Coe and Martinez60. The energies and gradients in electronically excited states were computed at the ADC(2) level, whereas the corresponding properties for electronic ground states were obtained at the MP2 level in the MECP geometry optimisations41. The MECP optimisation steps were performed with the Broyden–Fletcher–Goldfarb–Shanno quasi-Newton scheme available in the internal optimiser of Turbomole 7.161. All ADC(2) and MP2 calculations were performed with Turbomole 7.1, and Molcas 8.062 was employed for all the CASPT2/SA-CASSCF calculations. Graphical representations of the molecular geometries and orbitals were generated with IboView63.

Femtosecond transient absorption spectroscopy

FTAS experiments were performed with a Solstice Ace pulsed laser system (Spectra-Physics, Newport Co.), which produces 97 fs pulses at λ = 800 nm. To generate the white light continuum probe pulses (λ = 320–700 nm) in the Helios-Fire spectrometer (Ultrafast Systems, LLC), a fraction of the fundamental beam was focused on a thin CaF2 crystal. The λ = 290 nm excitation (pump) pulses were generated by passing the remainder of the fundamental beam through an optical parametric amplifier (TOPAS, Light Conversion, Ltd.). FTAS experiments of aqueous samples of ara-5A (4 mM, pH 7.4) and ara-5I (4 mM, pH 7.4) were performed in 2 mm optical path length quartz cuvettes (Starna Cells, Inc.). LabView Surface Xplorer software (Ultrafast Systems, LLC) was used to reduce the FTAS data, apply corrections for group velocity dispersion of the white light probe, and extract and analyse spectra and their time dependence.

X-ray diffraction

All diffraction data were collected by using a four-circle Agilent SuperNova (Dual Source) single crystal X-ray diffractometer with a micro-focus CuKα X-ray beam (λ = 1.54184 Å) and an Atlas CCD detector. The crystal temperature was controlled by using an Oxford Instruments Cryojet5. Unit cell determination, data reduction and analytical numeric absorption correction using a multifaceted crystal were carried out using the CrysAlisPro programme64. The crystal structures were solved with the ShelXS programme and refined by least squares on the basis of F2 with the ShelXL programme65. All non-hydrogen atoms were refined anisotropically by the full-matrix least-squares method. Hydrogen atoms affiliated with oxygen and nitrogen atoms were refined isotropically in positions identified by the difference Fourier map, or in geometrically constrained positions. Hydrogen atoms associated with carbon atoms were refined isotropically in geometrically constrained positions.