Introduction

Aryl ethers are important structural components prevalent in numerous natural products and pharmaceuticals, including vancomycin, asenapine, and riccardin C (Fig. 1a)1. Moreover, aryl ether structures exhibit a wide range of bioactivities, such as antiviral, antimicrobial, and amyloidogenesis inhibition properties, and are also utilized as chiral receptors and chiral ligands2. Over the past few decades, significant effort has been devoted to developing mild and general methods for synthesizing aryl ethers in both industry and academia3. Many catalytic systems, particularly those employing transition-metal-catalysed C−O cross-coupling, have been developed to operate on an increasingly broad scope of aryl halides, alcohols, or phenols4,5,6. Among these, copper-mediated C−O cross-coupling has garnered substantial attention due to its low cost, low toxicity, and broad applicability7,8,9,10. Despite these advancements, enantioselective techniques for synthesizing chiral aryl ethers remain significantly underdeveloped, even though the optically active products produced are highly valuable. In 2013, Beaudry and co-workers reported Cu-catalysed enantioselective synthesis of planar chiral meta-cyclophanes related to diarylether heptanoid natural products via enantioselective Ullmann cross-coupling reactions11. However, the enantioselectivity achieved was only up to 44%. Subsequently, Cai and co-workers reported copper-catalysed, highly enantioselective intramolecular O-arylation coupling reaction, forming chiral 2,3-dihydrobenzofurans using an asymmetric desymmetrization strategy (Fig. 1b)12. In 2020, Gu and co-workers developed a borinic acid-activated site-selective O-arylation of diols in a copper-catalysed asymmetric ring-opening reaction of cyclic diaryliodonium salts13. In this reaction, borinic acids nucleophilically activate diols by forming borolanuide, although this also limits the range of alcohol nucleophiles. More recently, Miller developed atroposelective Cu-catalysed C−O cross-coupling of phenols utilizing a guanidiny-lated peptidic ligand based on remote symmetry-breaking processes in diarylmethine substrates14,15,16. However, the reaction is limited to arylation reagents with o-trifluoromethanesulfonamide substitution. Despite these achievements, asymmetric O-arylation reactions are currently limited to desymmetrization strategies, and the range of applicable substrates remains greatly restricted. Thus, there is a strong appeal for the development of more facile and direct O-arylation routes.

Fig. 1: Chiral pillar[n]arenes: background and proposal.
figure 1

a Representative natural products, bioactive molecules, ligands and functional molecules. b Previous copper-catalysed asymmetric O-arylation reaction. c Dynamic kinetic resolution of carbonyl, oximino, and hydroxyl compounds. d This design: Copper-catalysed dynamic kinetic resolution O-arylation with diaryliodonium salts.

Full size image

Dynamic kinetic resolution (DKR), which can transform racemic compounds into enantiomerically enriched products with a theoretically quantitative yield, has proven to be an elegant and powerful tool for enantioselective synthesis (Fig. 1c)17,18,19. While DKR is a potent approach for establishing absolute configuration, it generally requires a mechanistic pathway for racemization. The DKR of carbonyl compounds bearing two substituents at the α-carbon has garnered considerable attention over the past decades20,21,22,23,24,25,26,27, especially since Noyori’s groundbreaking work on the asymmetric hydrogenation of β-ketoesters28. The success of these reactions relies on the acidity of the α-carbon, which facilitates the interconversion of enantiomers through enol tautomerization. However, despite the widespread application of tautomeric dynamic kinetic resolution in asymmetric synthesis, extending this strategy to other stereochemical motifs has been challenging. Oximes are attractive building blocks in organic synthesis due to their ability to readily interconvert into a variety of functional groups, such as nitrones, nitriles, amides, isoxazolines, and oxime ethers29,30. Due to the acidity of the oxime hydroxyl group, oximes can interconvert with nitroso compounds through double bond migration under acidic or basic conditions, offering the potential for dynamic kinetic resolution of oximes. Despite significant advancements by Takemoto31, You32, Carreira33, Zhang34, Liu35,36, and Zhang37 in the field of enantioselective and chemoselective allylic substitution, hydroamination, heterodifunctionalization, and carboetherification reactions—catalysed by Ir, Cu, and Pd complexes to achieve O- or N-allylation and alkylation of oxime derivatives—all of these reactions involve the construction of carbon-centered chiral compounds using achiral oximes as nucleophiles. To date, dynamic kinetic resolution achieved by prochiral oximes remains unexplored. Besides the interconversion of two enantiomers through tautomerization, another strategy to achieve dynamic kinetic resolution involves the ring-opening/ring-closing equilibrium of biaryl acetals, as developed by Akiyama38,39,40,41,42,43,44,45. Importantly, this DKR strategy can reduce the electron density on the phenolic aromatic ring, thereby avoiding the competition of phenol carbon arylation. We hypothesized that constructing axially chiral styrene compounds via copper-catalysed O-arylation of prochiral phenol hydroxyl groups might be achievable. Herein, we report a copper-catalysed dynamic kinetic asymmetric O-arylation of oximes and phenols with diaryliodonium salts46,47,48,49,50,51,52,53,54,55,56,57,58, delivering inherently and axially chiral aryl ethers in good yields with exceptional regio- and enantioselectivities (Fig. 1d)59,60,61,62,63,64,65,66.

Results

Optimization studies

At the outset of our studies, we selected tribenzotropone oxime (1a), derived from tribenzotropone, as an appropriate substrate for the catalytic enantioselective arylation process. Tribenzotropone oxime exhibits the favorable property of existing in equilibrium with its nitroso form in solution under basic conditions, facilitating tautomeric dynamic kinetic transformation. This hypothesis, however, faces significant challenges in selectively regulating O-, N-, or C-arylation67,68,69,70. To explore this, we first studied the arylation of tribenzotropone oxime 1a with Ph₂IPF₆ 2a (Table 1). In the presence of 10 mol % CuI, 12 mol % chiral bisoxazoline ligand L1, and NaHCO₃ as the base in dichloromethane, the selectively O-arylated product 3a was obtained in 61% yield with 42% ee at 25 oC (entry 1), confirming that copper salts could indeed catalyse this DKR transformation. Encouraged by this initial result, we investigated several copper catalysts, including Cu(OAc)₂, Cu(OTf)₂, and CuBr (Table 1, entries 2–4). However, when the reaction was performed with Cu(OAc)₂ or Cu(OTf)₂ as catalysts, almost no product 3a was obtained, and the main by-product was tribenzotropone, resulting from the decomposition of oxime 1a (entries 2–3). CuBr, on the other hand, provided the best result, yielding 3a in 72% yield with 43% ee (entry 4). We then examined several chiral bisoxazolines with different substituents (entries 5–13). The use of tBu-BOX ligand L2 proved inefficient (entry 5). The desired product was obtained in 70% yield with 38% ee when the cis-diphenyl-substituted oxazoline ligand BOX L3 was employed (entry 6). Next, we explored substituents on the methylene linker between the two oxazoline rings in the BOX ligand. Enantioselectivity was significantly improved with the gem-dibenzyl-substituted (S,S)-diphenylbisoxazoline L5, yielding 3a in 71% yield and 92% ee (entry 8). Further modification of the benzyl substituents on the oxazoline ligands did not enhance enantioselectivity (entries 9–10). Interestingly, the enantioselectivity of 3a was greatly reduced when methyl-benzyl-substituted oxazoline L8 was used, compared to the gem-dibenzyl-substituted ligand L5 (entry 11), highlighting the importance of alkyl substituents for stereoselective control. The cyclopentane-substituted (S,S)-diphenylbisoxazoline ligand L9 was not effective, providing only moderate yield and enantioselectivity (entry 12). When the sterically encumbered indaBOX ligand L10 was employed, product 3a was afforded in 52% yield with only 9% ee (entry 13). Screening different reaction media identified acetone as the best solvent (entries 14–17). Using L5 as the ligand in acetone, we then examined the effect of the counteranion of the diaryliodonium salt on the reaction outcome (X=BF₄, OTf). A comparable result was obtained with Ph₂IBF₄ 2a’ as the arylation reagent (entry 18). However, when Ph₂IOTf 2a” was used, although the enantioselectivity was 92%, the yield significantly decreased (entry 19). Additionally, the use of molecular sieves was tested, but they did not have a significant impact on the reaction outcome (entry 20).

Table 1 Optimization of the reaction conditions.aView full size image
Full size table

Scope of the reaction

With the optimized conditions in hand, we then explored the substrate scope of this enantioselective O-arylation reaction (Fig. 2). First, we examined the scope of diaryliodonium salts. A variety of functional groups were well-tolerated as meta or para substitutions on the aryl ring of the diaryliodonium salts, including halogens (F, 3b, 3l; Cl, 3f, 3m; Br, 3c, 3n), methyl (3g, 3j), trifluoromethyl (3i, 3o), trifluoromethoxy (3d, 3k), ester (3e, 3p), and nitro (3h) groups. The inherently chiral oxime ethers 3 were obtained in 65%–85% yields and 81%–97% ee. The absolute configurations of 3a and 3f were determined by X-ray crystallographic analysis (CCDC: 3a 2372246; 3f 2372247). Notably, aryl substrates bearing C−Br bonds remained unaltered by the Cu(I)-catalysed arylation, providing a complementary platform for further elaboration through conventional metal-catalysed coupling reactions (3c, 3n). Diaryliodonium salts with either electron-donating or electron-withdrawing groups at the ortho-position of the aryl ring also proved to be suitable substrates, although the yields were slightly reduced when these groups were incorporated (3q-3t). Furthermore, a series of diaryliodonium salts containing multi-substituted aryl groups were successfully employed in this arylation reaction, affording the target oxime ethers in 62%–90% yields and 91%–99% ee (3u-3ab).

Fig. 2: Scope of diaryliodonium salts.
figure 2

Reaction conditions: tribenzotropone oxime 1a (0.12 mmol), diaryliodonium salts [Ar-I-Ar]PF6 2 (0.1 mmol), CuBr (10 mol%), L5 (12 mol%), NaHCO3 (1.0 equiv.), acetone (0.5 mL) in a sealed vial at 25 °C for 48 h.

Full size image

Encouraged by the previous results, we next investigated the substrate scope of tribenzotropone oximes 1. As shown in Fig. 3, a variety of electronically diverse tribenzotropone oximes, with substitutions on either the bottom or edge aromatic rings, performed well in this process, delivering the O-arylated products 3ac-3ao in 65%–87% yields with diastereoselectivities ranging from 1:1 to 1.4:1. The enantioselectivity of the diastereomers was nearly identical, with most achieving enantioselectivities above 90%. Notably, the reaction is not limited to polyphenyl-substituted oximes; it also proceeds smoothly with the introduction of nitrogen-containing heterocycles (3ap). However, the enantioselectivity of the product was reduced. We assessed the racemization of 3ap at various time intervals (see Supplementary Information Figure S5 for details). Experimental results indicated that the configurational stability of product 3ap was relatively low at 25 °C in iPrOH, and the measured rotational energy barrier was 24.1 kcal/mol for 3ap (at 25 °C in iPrOH). This suggests that the substrate 3ap has relatively low activation energy, and extended reaction times lead to slow racemization.

Fig. 3: Scope of tribenzotropone oximes.
figure 3

Reaction conditions: Oxime 1 (0.12 mmol), diaryliodonium salts [Ph-I-Ph]PF6 2a (0.1 mmol), CuBr (10 mol%), L5 (12 mol%), NaHCO3 (1.0 equiv.), acetone (0.5 mL) in a sealed vial at 25 °C for 48 h.

Full size image

Having examined the asymmetric C−O coupling of oximes with diaryliodonium salts, we next examined the dynamic kinetic asymmetric arylation of phenols (see Supplementary information Table S2 for detail optimization of the reaction conditions). Figure 4 shows that a diverse array of diaryliodonium salts and 1-enal substituted 2-naphthols were successfully converted into axially chiral styrene products with good yields and enantioselectivities71,72,73,74,75,76. First, we examined the reactivity of a simple diaryliodonium salt 2 with a series of analogs that included methyl, halogen groups, including F, Cl, Br and trifluoromethyl substituents, which successfully yielded the targeted products 5a–5f. An X-ray diffraction analysis of a single crystal of 5a confirmed that its absolute configuration was S. Substrates carrying an electron-donating or electron-withdrawing group at the meta-positions on phenyl rings (5g–5j) resulted in high yields and excellent enantioselectivities. It is worth noting that the product 5a (24.4 kcal/mol, iPrOH, 25 °C) underwent slightly racemization during column chromatography. Consequently, the aldehyde was converted into an alcohol with a higher energy barrier, allowing us to conduct further substrate scope tests. When an electron-donating methoxy group was used, the corresponding product 6k was obtained in 74% yield with 92% ee. Moreover, substrates featuring electron-withdrawing functional groups such as ester (6l), trifluoromethoxy (6m and 6o), and even nitro (6n) groups have emerged as appropriate choices, affording good-to-high yields while maintaining high levels of enantioselectivity. A significant exception was noted for the diaryliodonium salt 6p, which had a methyl group at its ortho position. This alteration resulted in a decrease in the enantioselectivity, emphasizing the pronounced influence of steric hindrance at this location on stereochemical control. Double-halogen-substituted diaryliodonium salts (6q–6t), demonstrating good substrate compatibility and enantioselectivities ranging from 90% to 93%, were also efficiently synthesized via this method. Subsequently, we investigated the broad spectrum of 1-enal substituted 2-naphthols 4 by introducing various substituents on aromatic rings R1 and R2. Various labile functionalities were efficiently incorporated into the R1 of naphthols, including bromine (6u and 6y), ester (6v), nitrile (6w), phenyl (6x), and methoxy (6z) groups. This indicates the high tolerance of different functional groups in the current catalytic approach. Notably, the nitrile-substituted substrate was suitable for the described reaction, although it resulted in slightly reduced enantioselectivity (6w). Subsequently, the aromatic ring R2 featuring various ortho- and meta-substituents, such as phenyl, halogen, nitro, and trifluoromethyl groups, were effectively utilized, producing the desired the product (6aa–6af) in yields of 76–94% and excellent enantioselectivities. Para-substituent substrates, whether with electron-donating or electron-withdrawing substituents, smoothly participated in the reaction, showing good enantioselectivity and resulting in the corresponding products 6ag–6ak in yields of 51–95%. Importantly, this method was effective for the transformation of electron-rich naphthyl aldehyde substrate (6al). Similarly, substitution with heteroaromatic furan and thiophene groups resulted in seamless reactions affording the products 6am and 6an, respectively, achieving yields of 52–56% and >82% ee.

Fig. 4: Scope of diaryliodonium salts and phenols.
figure 4

Reaction Conditions: a4 (0.12 mmol), 2 (0.10 mmol), CuBr (10 mol%), L7 (12 mol%), Na2CO3 (1.5 equiv.) in DCE (1.0 mL) at −20 °C for 48 h; isolated yields; ee values were determined by chiral HPLC. bAfter reaction, NaBH4 (1.5 equiv.) in 1.0 mL MeOH was added and the reaction was stirred at −20 °C for 4 h. cReacted for 5 days.

Full size image

The effectiveness of our method was further demonstrated through the modification of drug and natural product derivatives (Fig. 5a). Substrates derived from febuxostat, lithocholic acid, and linoleic acid were successfully transformed into the corresponding products 8a–8c with respectable yields and enantio- and diastereoselectivities. To further highlight the synthetic value of this method, a series of experiments was performed (Fig. 5b) using 4a and 2a as substrates. The synthesis was smoothly conducted on a 2.0 mmol scale under standard conditions. This process resulted in the production of product 6a with 75% yield while maintaining the enantioselectivity. Subsequently, allylic alcohol product 6a successfully reacted with acetic anhydride and iodomethane, resulting in the production of the corresponding esterification and alkylation 9a and 9b, respectively, with a slight decrease in the enantiomeric ratio. Additionally, the silylation of 6a was performed, leading to the synthesis of 9c with a high enantioselectivity.

Fig. 5: Scope of natural product derivatives and synthetic application.
figure 5

a Reaction Conditions: 7 (0.12 mmol), 2 (0.10 mmol), CuBr (10 mol%), L7 (12 mol%), Na2CO3 (1.5 equiv.) in DCE (1.0 mL) at −20 °C for 48 h; isolated yields; ee values were determined by chiral HPLC. b Scaled-up reaction and synthetic application.

Full size image

Stereochemical stability study and control experiments

To assess the stereochemical stability of these chiral aryl ether compounds, we conducted racemization experiments on oxime 1a, oxime ether 3a, and axially chiral styrene 6ak over time at various temperatures. As depicted in Fig. 6a, the ee value of 3a decreased slightly to 83% after 3.5 h in iPrOH at 70 °C. When the solution was maintained at 80 °C, the ee value of 3a gradually dropped to 75% over 3.5 h. Based on these experimental results, the rotational barrier for 3a was calculated to be 28.9 kcal/mol at 80 °C. Interestingly, oxime 1a exhibited similar stereochemical stability to oxime ether 3a, with a rotational barrier of 27.7 kcal/mol at 60 °C. Further experiments showed that axially chiral styrene 6ak remained relatively stable at 50 °C in iPrOH, with a measured rotational energy barrier of 28.1 kcal/mol at 70 °C. Given the relatively high inversion energy barrier of oximes, several controlled experiments were conducted to investigate their dynamic kinetic transformation at room temperature (Fig. 6b). We assessed the stability of enantioenriched oxime (+)-1a under base or base/copper conditions similar to those in the optimal process. In these cases, the remaining oxime 1a was partially or almost fully racemized. Additionally, we demonstrated that the reaction of enantioenriched (+)-1a under standard conditions, but without the chiral ligand L5, yielded 3a in 60% yield with 66% ee after 12 h. Furthermore, the reaction of (+)-1a with (S,S)-L5 as a ligand resulted in the productive, matched formation of the inherently chiral 3a with high ee. Conversely, the mismatched experiment using (R,R)-L5 as the ligand produced the opposite enantiomer of 3a in a similar yield but with 80% ee. These results confirmed that during the reaction, the optically active oxime 1a underwent racemization and then reacted with the matching ligand to achieve highly enantioselective arylation products. Collectively, these experiments elucidated the process of dynamic kinetic resolution in oximes 1.

Fig. 6: Mechanistic investigations.
figure 6

a Racemization experiments on oxime 1a, oxime ether 3a, and axially chiral styrene 6ak. b Control experiments for the dynamic kinetic resolution of 1a. c Kinetic experiment towards a dynamic kinetic resolution (DKR) pathway. d Non-linear experiments. e Hammett plot for selected diaryliodonium salts.

Full size image

To clarify the mechanism of the asymmetric C-O coupling reaction of phenols, a sequence of initial experiments was performed. Initially, a Cu-catalyzed O-arylation reaction between substrates 4a and 2a was performed under standard conditions to study the dynamic kinetic resolution (DKR) process (Fig. 6c). During the reaction, the ee value of 4a remained racemic, which indicated that the reaction underwent a transition state involving a five-membered aryl alkene lactol under the influence of sodium carbonate (base). This suggests that rapid dynamic kinetic resolution occurred under these reaction conditions. Initially, the ee value of 5a increased during the reaction, reaching a peak (91% ee) and stabilizing after 11 h. This pattern could be due to a strong background reaction that occurred during the early stages of the process. Additionally, experiments on nonlinear effects were carried out (Fig. 6d). The observed positive non-linear effect between 5a and L7 implied that the actual catalyst in the catalytic cycle might not be a mononuclear CuBr·L7 complex77,78,79. To better understand the enantiodetermining step of the O-arylation reaction, we examined the Hammett electronic parameters of different para- and meta-substituted diaryliodonium salts and enantioselectivities of the resulting products (Fig. 6e). Hammett plots exhibited a linear correlation (R2 = 0.9655) with a negative slope (ρ = −0.3500), suggested that copper intermediates generated from electron-rich diaryliodonium salts react more effectively with 1-enal substituted 2-naphthol 1a than those from electron-deficient diaryliodonium salts.

To further elucidate the origins of the regio- and enantioselectivity for the C−O cross-coupling reaction, density functional theory (DFT) calculations were performed with the B3LYP-D3 functional (see SI for computational details). As shown in Fig. 7a, starting from the CuI(Br)(L5) complex int1, the oxidative addition of the diaryliodonium salt80,81 2 has an activation free energy barrier of 24.2 kcal/mol via TS_OA and is slightly exergonic to lead to the CuIII(Br)(Ph)(L5) intermediate int2. Following that, the ligand exchange between oximes (+)-1, NaHCO3, and BF4− is thermodynamically favorable to provide int3, in which the O atom of the substrate binds to the Cu. Subsequently, the reductive elimination via transition state TS1 would afford the O-Ph coupling product. Alternatively, int3 would first isomerize to a less stable intermediate int4, in which the N atom binds to the Cu, and followed by the reductive elimination through TS2 to produce the N-Ph coupling product. By comparison, the pathway leading to the experimentally observed O-Ph coupling product (via TS1) is 3.2 kcal/mol more favorable than that for the N-Ph coupling product (via TS2). Thus, computations are consistent with experiment on the regioselectivity. It should be noted that we have also considered other possible pathways for the coupling step64,71 (e.g., the transmetalation-like pathway) but none of them were more favorable than the reductive elimination mechanism (see SI for details).

Fig. 7: DFT studies.
figure 7

a The free energy profiles for the oxidation, ligand exchange, and reductive elimination steps of the reaction. b The 3D geometries for the selectivity-determining transition states (bond distances are given in Å).

Full size image

Calculations were conducted to elucidate the enantioselectivity of the reaction. Starting from intermediate int2, the ligand exchange between oximes (-)-1, NaHCO3, and BF4− leads to the formation of int3’. However, this pathway is thermodynamically less favorable compared to the formation of int3. The subsequent reductive elimination occurs via transition state TS1’, which would generate the enantiomer opposite to the experimentally observed product. This minor pathway, progressing from int3’ to TS1’, has a higher energy barrier by 1.4 kcal/mol compared to the major pathway from int3 to TS1. These findings are consistent with the experimentally observed enantioselectivity.

A detailed analysis of the selectivity-determining transition states (Fig. 7b) reveals key factors influencing the reaction’s enantioselectivity. In the N-Ph coupling transition state TS2, significant steric repulsions occur between the ligand and the substrates, with the closest H-H distances measured at 2.06 and 2.12 Å, respectively. In contrast, no such steric repulsion is observed in the O-Ph coupling transition state TS1. Additionally, TS1 benefits from favorable C-H…π interactions between the phenyl groups of the ligand and oximes, as well as within the ligand itself, with distances of 2.64 and 2.63 Å, respectively. In addition, there is also a favorable C-H…Br interaction between the substrate and the Br atom (see SI for details of NCI analysis). In the minor enantioselective transition state TS1’, there is only one C-H…π interaction were found. Consequently, the observed enantioselectivity is primarily attributed to these non-covalent interactions, while both non-covalent interactions and steric effects contribute to the regioselectivity of the reaction.

Discussion

In conclusion, we reported a dynamic kinetic asymmetric C−O cross-coupling of oximes and phenols via copper/BOX-catalysed enantioselective O-arylation with diaryliodonium salts. This approach efficiently prepared a wide range of inherently and axially chiral oxime ethers and axially chiral styrenes in high yields with excellent regio- and enantioselectivities. Control experiments revealed the dynamic kinetic resolution process of oximes, while DFT studies provided insights into the origins of regio- and enantioselectivity. The reaction features mild conditions, good functional group tolerance, readily available starting materials, and commercial bisoxazoline ligands, indicating its potential for broad applications in organic synthesis.

Methods

General procedure for the synthesis of chiral aryl oxime ethers 3

Under air atmosphere, to a mixture of 1 (0.24 mmol, 1.2 equiv.), 2 (0.20 mmol, 1.0 equiv.), CuBr (2.9 mg, 0.02 mmol, 10 mol%), L5 (11.7 mg, 0.024 mmol, 12 mol%), NaHCO3 (16.8 mg, 0.20 mmol, 1.0 equiv.) in a sealed vial was added acetone (1.0 mL) at 25 °C. Upon completion of the reaction, the solvent was removed under vacuum, and the crude product was purified directly by column chromatography to yield the desired product 3.

General procedure for the synthesis of chiral aryl diaryl ethers 5

Under an air atmosphere, a mixture of compound 4 (0.12 mmol, 1.2 equiv.), compound 2 (0.10 mmol, 1.0 equiv.), CuBr (1.4 mg, 0.01 mmol, 10 mol%), L7 (7.2 mg, 0.012 mmol, 12 mol%), and Na₂CO₃ (15.9 mg, 0.15 mmol, 1.5 equiv.) was added in a sealed vial. This was followed by the addition of 1.0 mL of DCE. The reaction mixture was then stirred at −20 °C for 48 h. Upon completion, the solvent was removed under vacuum, and the crude product was purified directly by column chromatography to obtain the desired product 5.

Additional information

Supplementary information and chemical compound information are available in the online version of the paper. Reprints and permission information is available online at http://www.nature.com/reprints/.