Water can be transformed to valuable hydrogen gas (H2) through serial chemical processes, which allow it to be applied as a potential transfer hydrogenation reagent in the reduction of unsaturated compounds1,2. However, the main challenges for water activation originate from two features (Fig. 1a). (1) Whereas deprotonation of water (pKa = 15.74) can be readily achieved with an appropriate base, homolytic cleavage is challenging due to the high thermodynamic stability of H2O as a result of the high bond dissociation energy (BDE = 118 kcal mol−1) of the O–H bonds3,4. (2) Further, the OH radical (hydroxyl radical), if generated, cannot provide the second hydrogen atom, as this would lead to a high-energy oxygen atom5. As a result, catalysts or mediators are required, and current approaches to water activation follow three different strategies: oxidative addition activation6,7,8, metal–ligand cooperation activation9 and coordination-induced bond weakening10,11,12 (Fig. 1b). The first two approaches mainly focus on the use of d-block transition metals, and transformations dealing with main group elements remain very rare. Several examples capitalizing on the concept of coordination-induced bond weakening through generation of the corresponding H2O adducts have been disclosed using transition metal complexes or main group elements with empty d or p orbitals10. However, as H2O adduct formation also increases the acidity, the majority of such H2O–complexes react as proton donors. Considering the homolytic O–H bond cleavage, a few examples are known, with samarium(II)13,14,15,16,17,18, titanium(III)19,20,21, borane22,23,24 and bismuth(II)25 compounds, among others, enabling homolytic O–H bond cleavage processes. These activated water complexes have been shown to reduce reactive radicals, leading to closed-shell products, in which hydrogen atom transfer (HAT) is thermodynamically favoured (Fig. 1c)10,19,20,21,22,23,24,25,26. By contrast, examples of intermolecular HAT from H2O–intermediates to closed-shell π systems leading to reactive H adduct C radicals, especially considering metal-free H2O adduct intermediates, remain desirable. Notably, in situ-generated Fe, Mn and Co hydride complexes have been intensively investigated as HAT reagents in the reaction with alkenes to give the corresponding H adduct C radicals, in so-called metal hydride hydrogen atom transfers (MHAT)27,28,29. In these MHAT processes, the H atom to be delivered is generally derived from a hydride source, such as a silane. Intermolecular HAT to π systems from activated water are known for SmX2–H2O adducts, albeit with limited scope, and metal-free variants need to be uncovered30,31,32,33.

Fig. 1: Background and concept.
figure 1

a, Water as a potential transfer hydrogenation reagent; BDE and pKa. b, Overview of different water activation modes and examples for activated H2O adducts. c, Radical reductions through intermolecular HAT from water adducts and HAT to a closed-shell π system. Challenges associated with the latter reactivity mode. d, This work: photocatalytic phosphine-mediated water activation to provide a ‘free’ H atom through sequential heterolytic and homolytic cleavage of two O–H bonds in the activated water adduct. MG, main group; TM, transition metal.

Considering transition metal-free variants, there are several challenges to be overcome. The currently known water adducts that are able to act as HAT reagents are relatively stable, so that hydrogen transfer to π systems leading to reactive C radicals is endergonic. As a consequence, except in the examples with Sm(II)30,31,32,33,34,35,36, only HAT to reactive radicals is established with such species. Furthermore, the H2O-based H donors should be readily generated, be sufficiently reactive to engage in thermodynamically favoured HAT to π systems, but at the same time must have a sufficient lifetime, which renders reagent design highly challenging. It is remarkable that a few examples of metal-free radical H atom transfers to realize hydrogenation of closed-shell π systems have been reported. Thus, 9,10-dihydroanthracene, xanthene or tetralin were shown by Rüchardt and co-workers to transfer an H atom to closed-shell π systems37,38,39. However, as these H donors themselves are stable closed-shell compounds, intermolecular HAT in a retrodisproportionation reaction is strongly endothermic and kinetically not favoured. Therefore, these processes are very narrow in scope and only occur at temperatures above 300 °C.

Considering that phosphine radical cations can act as tunable mediators in photoredox catalysis for the generation of carbon- and heteroatom-centred radicals through deoxygenation (Fig. 1d)40,41, we designed a related approach for water activation by interaction of a phosphine radical cation through its singly occupied molecular orbital with water, using photoredox catalysis. The PR3–H2O radical cation can be deprotonated, generating a PR3–OH radical intermediate in a heterolytic O–H bond cleavage, as previously suggested by Pandey and co-workers42. The strong BDE of the P=O bond in phosphine oxides ensures a high thermodynamic driving force for the homolytic cleavage of the second O–H bond43, enabling the PR3–OH intermediate to behave like a ‘free’ H atom, providing the opportunity to explore H atom radical chemistry.

To experimentally validate the reaction design and H atom reactivity of the proposed PR3–OH intermediate, triphenylphosphine (PPh3)-mediated hydrogen evolution was investigated first. Pleasingly, with [Ir(dF(CF3)ppy)2(dtbbpy)](PF6)] (dF(CF3)ppy, 2-(2,4-difluorophenyl)-5-trifluoromethylpyridine; dtbbpy, 4,4′-di-tert-butyl-2,2′-bipyridine, PC1) as the photocatalyst and H2O as the hydrogen source in acetonitrile under irradiation with a 5 W blue light-emitting diode (LED), generation of H2 was experimentally verified in both solution and gas phase, with concomitant formation of triphenylphosphine oxide as the by-product (Fig. 2a). This important experiment indicated that the proposed radical PPh3–OH intermediate can be generated, and further that it has a sufficient lifetime so that intermolecular reactions with such an H donor become feasible, as shown by the successful generation of H2. Encouraged by this finding, transfer hydrogenation of unactivated alkenes with water as the H donor was attempted (Fig. 2b). According to the proposal, the first H atom of the activated H2O is released as a proton (Fig. 1d), and therefore we decided to use a thiol co-catalyst that is able to transform the proton donor (PPh3–OH2+) into a radical H donor through thiolate protonation. The thiol in turn is capable of efficiently reducing an alkyl radical. The key species for the designed process is the suggested PR3–OH intermediate, which should show sufficient reactivity to transfer its H atom intermolecularly to an unactivated alkene.

Fig. 2: Reaction design and mechanism analysis.
figure 2

a, Hydrogen evolution through photocatalytic triphenylphosphine-mediated water activation. b, General mechanism for HAT of PR3–OH intermediates to closed-shell unactivated alkenes. c, Reaction optimization for transfer hydrogenation of 1a and BDE of the O–H bond in various PR3–OH intermediates, obtained by density functional theory (DFT) calculations. Yields were determined by gas chromatography–mass spectrometry (GC–MS) with tetradecane as the internal standard. Calculated activation energy for the intermolecular HAT from the P3–OH intermediate to styrene and propene as model reaction. The structure displays the spin density distribution of the Ph3P–OH radical indicating delocalization of the radical spin from the phosphorus atom into the adjacent phenyl ring. P6, bis(4-methoxyphenyl)(methyl)phosphine. ΔG, reaction free energy (298K); ΔG, free energy barrier (298K); ΔH298(vac), enthalpy of bond dissociation.

4-Allyltoluene (1a) was selected as the model substrate to evaluate the feasibility of the targeted radical alkene transfer hydrogenation. Pleasingly, using tris(4-methoxyphenyl)phosphine P1 as mediator, PC1 as photocatalyst, 2,4,6-triisopropylbenzenethiol (HAT1) as HAT catalyst and H2O as hydrogen source in acetonitrile under blue LED irradiation, the hydrogenation product 2a could be obtained in 86% isolated yield at 20 °C (Fig. 2c, entry 1). Other triaryl phosphines P2P6 provided worse results (Fig. 2c, entries 2–6), and hydrogenation failed with the thiol co-catalyst HAT2 (Fig. 2c, entry 7). Other frequently used Ir-based photocatalysts (PC2, PC3) provided lower yields (Fig. 2c, entries 8 and 9) and all reagents are required for successful hydrogenation of 1a (Supplementary Information).

To get an idea regarding the structure and the H atom affinity of the H donor, density functional theory (DFT) calculations were performed (Fig. 2c and Supplementary Information). First, the O–H bond dissociation energy of different PAryl3–OH intermediates was calculated to be very low for all compounds considered, ranging from 5.0 to 13.4 kcal mol−1. We note that the lowest bond energies are obtained for the H atom donors formed with P1 and P4, which show the highest yields in the hydrogenation of 1a. Interestingly, we found that transfer of the H atom to the ortho-position of the aryl substituent leads to a more stable structure (shown for P3). However, due to the non-linearity of that intramolecular HAT, we did not find any transition state for such a process and therefore this cyclohexadienyl radical was not further considered as the H atom donor in these reactions. Intermolecular HAT of the P3-derived PAryl3–OH radical intermediate was calculated for styrene and propene as the H atom acceptors. In agreement with the experimental findings, HAT to propene is feasible at room temperature with a free energy barrier of 19.3 kcal mol−1. As expected, due to the radical stabilization of the styryl radical, an even lower barrier was found for HAT to styrene (10.6 kcal mol−1).

With the optimized conditions in hand, various unactivated alkenes bearing different functionalities could be efficiently hydrogenated. Allylarenes carrying electron-donating or -withdrawing substituents at the arene moiety provided the corresponding propylarenes 2a2f with high yields (Fig. 3a). The radical hydrogenation was compatible with commonly used functional groups, including boronic acid pinacol ester (2g), pyridine (2h), ether (2i), triphenylsilyl (2j), oxazolidinone (2k), amide (2l and 2m), sulfonamide (2n), free alcohols (2o and 2r), aldehyde (2p) and thiophene (2q) moieties. Disubstituted alkenes could also be reduced, albeit with moderate yields (2s and 2t). Our mild protocol also permits late-stage hydrogenation of more complex compounds bearing a terminal unactivated double bond, as documented by the reduction of sclareol (2u), allylestrenol (2v), β-estradiol (2w) and an estrone derivative (2x). Notably, the reduction of allylestrenol 1v furnished the product 2v with complete chemoselectivity and the trisubstituted alkene moiety in the ring remained unreacted. The more reactive styrene derivatives could be readily hydrogenated using PPh3 (P3) as the H atom mediator and reaction time could be shortened (48 h). Various styrene derivatives, including α- and β-substituted congeners, as well as trisubstituted systems, could be reduced in moderate to excellent yields (Fig. 3b, 4a4l). The steroidal compound 3m was hydrogenated with complete chemo- and diastereoselectivity.

Fig. 3: Transfer hydrogenation substrate scope.
figure 3

a, The reduction of unactivated alkenes. b, The reduction of styrene derivatives. Yields given correspond to isolated yields. See Supplementary Information for specific reduction conditions that deviate from the standard conditions. Diastereoselectivity was determined by 1H NMR spectroscopy. aTriaryl phosphine P1 as the mediator in combination with 0.2 equivalent of HAT1. bTriaryl phosphine P3 as the mediator in combination with 0.15 equivalent of HAT1. cYield obtained for gram-scale experiment (10.0 mmol scale). dGC yield with tetradecane as internal standard. eReaction conducted at 40 °C. fYield for gram-scale experiment (5.0 mmol scale). Bpin, boronic acid pinacol ester; d.r., diastereomeric ratio; Ts, tosyl.

Fig. 4: The reduction of naphthalenes.
figure 4

Yields given correspond to isolated yields. See Supplementary Information for specific reduction conditions that deviate from the standard conditions. Diastereoselectivity was determined by 1H NMR spectroscopy. aTriaryl phosphine P1 as the mediator in combination with 0.2 equivalent of HAT1. bYield obtained for gram-scale experiment (10.0 mmol scale).

Hydrogenation of arenes provides an atom-economic, straightforward and efficient route for the construction of three-dimensional structures through easily accessed planar aromatic compounds. However, arene reduction through H atom transfer is more difficult than alkene hydrogenation, as the initial HAT leads to de-aromatization. As is known, radical MHAT by Mn/Co/Fe hydrides to arenes is very challenging44,45. Motivated by our calculations, which showed that intramolecular H transfer to the arene moiety in the PAryl3–OH radical intermediate is thermodynamically feasible (Fig. 2c), we next attempted the challenging hydrogenation of arenes (Fig. 4). We were very pleased to find that naphthalene (5a) and its derivatives with ester (5b), cyano (5c) and hydroxy (5d) groups could be hydrogenated to afford the corresponding tetrahydronaphthalenes with high yields. Hydrogenation of the second arene moiety was not observed in these cases, which can be understood by the larger resonance energy of the arene moiety in tetrahydronaphthalene compared to naphthalene. Naphthalenes with electron-withdrawing substituents were regioselectively de-aromatized at the substituted arene ring. Naphthalene derivatives bearing electron-donating methyl or methoxy groups (see compounds 5f and 5g) provided mixtures of the two possible tetrahydronaphthalenes. The regioselectivity issue was further investigated by DFT calculations (Supplementary Information). For 1-methylnaphthalene (5f) the activation barrier for the intermolecular HAT from the P1–OH radical to the 2-, 3-, 4-, 5-, 6-, 7- and 8-position was calculated. All HATs are exothermic (−17.9 to −23.5 kcal mol−1) and occur with low barriers (6.3 to 9.7 kcal mol−1). The lowest barrier was calculated for the HAT to the C4-position (6.3 kcal mol−1), but H transfer to C8 and C5 leading to the other regioisomers showed similar barriers (6.5 and 6.9 kcal mol−1, respectively). This explains why no selectivity was obtained for the hydrogenation of 5f. The barrier for the HAT correlates well with the thermodynamic stability of the H adduct (exothermicity, C4-position: −23.5 kcal mol−1, C8-position: −23.5 kcal mol−1, C5-position: −23.0 kcal mol−1). With this knowledge in hand, we also calculated the exothermicity for the HAT to 1-(methoxycarbonyl)naphthalene (5b) and 2-cyanonaphthalene (5c), for which perfect regiocontrol was achieved in the hydrogenation. Again, seven regioisomeric H adducts were considered for both cases. For the ester 5b, HAT to the C4-position was thermodynamically most favoured (−26.2 kcal mol−1) leading to the experimentally observed regioisomer 6b. The thermodynamically most stable H adduct in the reaction with 5c resulted from C1 addition (−27.0 kcal mol−1), which eventually leads to the observed isomer 6c, revealing that the regioselectivity in these hydrogenations is mainly controlled by the stability of the de-aromatized H adduct radical.

Naphthols deserve special attention: 1-naphthol gave the phenol 6d as a single reduction product with complete regiocontrol, whereas 2-naphthol provided three different products 6e, 6e′ and 6e′′. Formation of the ketone 6e′′ indicates that the enol formed by 1,4-reduction of the naphthalene core is a likely intermediate in the hydrogenation of 5e. The more complex naphthalene 5h was regioselectively hydrogenated to give 6h as a 1:1 mixture of diastereoisomers. Hydrogenation also worked well for multisubstituted naphthalenes (5i5k). Regioselective hydrogenation occurred at the more electron poor arene moiety (6j and 6k). Disubstituted naphthalenes with both substituents at the same ring afforded the targeted tetrahydronaphthalenes as diastereoisomeric mixtures (6l6n). As expected for 1,3-dimethylnaphthalene, 6l was formed as a major regioisomer along with 6l′, which was isolated as a 2:1 (cis:trans) diastereoisomeric mixture. Complete regiocontrol was noted for the hydrogenation of the naphthalenes 5m and 5n. Very good 1,2-stereocontrol was achieved for 6m (cis:trans = 10:1), whereas the 1,4-stereoinduction was moderate (6n, 2:1). Anthracene 5o and its cyanated derivative 5p were reduced with complete regiocontrol at the most activated central ring. Comparable yields were obtained for the gram-scale preparation of 2b, 4d and 6c, documenting the practicality of this hydrogenation process. The generated phosphine oxides could be almost quantitatively recovered (Supplementary Information) and are easily reduced to give the corresponding starting phosphines46.

The unique H atom reducing ability of this photocatalytic phosphine-mediated water activation system, in which hydrogenation proceeds through radical intermediates, was further verified by realizing direct skeletal editing of quinolines (Fig. 5a)47. Such a hydrogenative rearrangement is unknown for any other radical reducing system and cannot be achieved through transition metal-mediated hydrogenation processes. Several indole derivatives could be directly constructed by a reductive ring contraction of the corresponding quinoline derivatives. The rearrangement of 2-substituted quinolines (5q–s and 5u) afforded the corresponding 2,3-disubstituted indoles 6q–s and 6u in moderate yields. A 2,4-disubstituted quinoline (5t) also engaged in this transformation to give the indole 6t, and a possible mechanism is presented in Fig. 5a. Intermediate Int1 is generated through an initial H atom transfer of the P1–OH intermediate to the quinoline C(4)-position. Subsequent neophyl-type rearrangement through Int2 leads to Int3, which is finally reduced by the thiol co-catalyst to provide the product indole. The suggested mechanism was supported by a deuteration experiment: replacing H2O with D2O provided the bis-deuterated indole 6u-D2, showing that the initial HAT by the P1–OH intermediate and also the final HAT from the thiol co-catalyst occur at the same carbon atom.

Fig. 5: Applications and mechanistic studies.
figure 5

a, Skeletal editing from quinolines to indoles by photocatalytic phosphine-mediated H atom transfer. b, Radical cyclization and ring-opening experiments. c, Transfer hydrogenation with deuterated water to prepare deuterated reduced products. Kinetic isotope effect (KIE) measurement.

We next explored the relative reactivity of two different π systems, either by installing them into the same substrate or by running competition experiments of two π systems at low conversion (Supplementary Information). The alkene moiety in 2-vinylnaphthalene (3n) is more reactive than the naphthalene core, and 2-ethylnaphthalene was obtained with complete chemoselectivity using PPh3 (P3). The same outcome was noted for the more reactive phosphine P1 (Supplementary Information). Comparing an activated alkene with an unactivated terminal alkene in 4-allyloxystyrene, we found excellent chemoselectivity for the hydrogenation of the styrenic double bond with the milder P3 as mediator, and allyl (4-ethylphenyl) ether was obtained in 82% yield. However, both double bonds in 4-allyloxystyrene were reduced with the more reactive phosphine P1. Thus, the chemoselectivity of the hydrogenation of 4-allyloxystyrene can be controlled by tuning of H atom reactivity through variation of the PAryl3 component. This is reminiscent of the reactivity tuning in organometallic hydrides through ligand variation. Considering styrene pairs that slightly differ in their electronic structure by variation of their para-substituents, low selectivity was noted using the milder PPh3-derived H donor (Me versus CN = 1:1.4; Me versus MeO = 1:2). Steric effects play a role as β-methyl- (trans) and also α-methyl-substituted styrenes are less reactive than their corresponding unsubstituted styrenes (by around 2:1).

As an additional advantage of the radical-type hydrogenation over well-established Pd/Rh/Ir-catalysed processes, which generally operate with H2 gas48,49,50, hydrogenations with concomitant cyclization or ring opening were studied. At the same time, these investigations provided further experimental support for the radical nature of these transformations (Fig. 5b). Four different dienes 7, 9, 11 and 13 were subjected to the standard conditions and the targeted reductive cyclization products 8, 10, 12 and 14 were obtained. In all cases, hydrogenated non-cyclized derivatives were not observed, showing that the cyclization is faster than direct reduction under the applied conditions. For the Si compound 7, 6-endo-trig cyclization occurred exclusively, whereas the other three systems cyclized through the 5-exo mode and the five-membered rings were formed with good to excellent diastereoselectivities. Vinylcyclopropane 15 reacted to produce the reduced ring-opening product 16. As a side product, alkene 17 was isolated, probably resulting from double bond isomerization of 16. Control experiments revealed that H2O is the exclusive hydrogen source in the hydrogenation and the solvent (acetonitrile) is not involved (Fig. 5c), indicating the potential of this radical hydrogenation for the preparation of deuterated compounds using cheap D2O, as realized for the preparation of 2d-D2 and 6j-D4.

Kinetic experiments, from running the reaction of 3a in either H2O or in D2O at low conversion, indicated that the HAT process between the PR3–OH intermediate and the alkene is the rate-determining step, with a kinetic isotope effect (KIE) value of 3.2 (Fig. 5c). This is further supported by an additional experiment in which hydrogenation was conducted using a 1:1 mixture of H2O:D2O with a lower deuterium incorporation observed for the β-position (conducted with 3c). Moreover, we found first-order kinetics with respect to the alkene, and zeroth-order kinetics with respect to the thiol co-catalyst, further supporting that the initial HAT to the alkene is the rate-determining step (conducted with 1a and 1j; Supplementary Information). Stern–Volmer quenching studies confirmed the initial oxidation of the phosphine by the excited Ir photocatalyst, and quantum yield measurements revealed that a radical chain reaction is not in operation. Overall, all these experiments and the calculations support the suggested mechanism depicted in Fig. 2b.

In summary, in situ-generated PR3–OH radicals have been shown to react as HAT reagents with various alkenes and naphthalenes to give the corresponding H atom addition adduct radicals at room temperature. The key PR3–OH radicals, which can be considered as formal ‘free’ H atoms, are readily generated through photocatalytic phosphine-mediated water activation. Thiols are used as catalytic co-reductants for the radical hydrogenation of π systems, transforming the initially generated PR3–OH2 radical cation intermediate, which is an efficient proton donor, into a radical hydrogen atom donor. This co-catalysis approach ensures that both H atoms of water can be used as H atom donors in the reduction of π systems. Considering hydrogenation of alkenes, the introduced metal-free PR3–OH intermediate has comparable reactivity to the established transition metal-based systems, which engage in MHAT reactions. Intra- or intermolecular hydrogen atom transfer to reactive carbon- or heteroatom-centred radicals is a highly useful tool in modern synthesis for radical translocation. However, HAT to closed-shell systems has not been well developed and we are confident that the strategy presented here will open the door to unexplored ‘H atom radical chemistry’.