Introduction

Atropisomers1,2,3,4,5,6,7 represent an important class of fundamental chiral compounds due to their various applications in pharmaceuticals8,9, materials10,11 and organic synthesis12. Therefore, catalytic enantioselective synthesis of atropisomers has become a frontier area of research13,14,15,16,17. In this area, fast developments have been achieved in catalytic atroposelective construction of open-chain axially chiral aryl-based scaffolds (Fig. 1a)18,19,20,21. However, in stark contrast, catalytic atroposelective construction of axially chiral aryl-fused medium-sized rings is rather underdeveloped22,23, which might be ascribed to the inherent difficulties in overcoming the unfavorable entropy and enthalpy effect during the construction of medium-sized rings as well as restraining the high tendency towards racemization of the axial chirality. Nevertheless, given that axially chiral aryl-fused medium-sized rings constitute the structural motifs of numerous natural products, bioactive molecules, and chiral catalysts (Fig. 1b)24,25,26,27, catalytic atroposelective construction of such axially chiral medium-sized rings is of great importance yet very challenging.

Fig. 1: Current status of catalytic atroposelective synthesis of axially chiral aryl-fused medium-sized rings and the remaining challenges.
figure 1

a Profile of catalytic atroposelective synthesis of axially chiral aryl-based scaffolds. b Representative axially chiral aryl-fused medium-sized rings. c Catalytic atroposelective synthesis of axially chiral aryl-fused medium-sized rings.

Full size image

Currently, the constructed axially chiral aryl-fused medium-sized rings are confined to seven-membered, eight-membered and nine-membered rings (Fig. 1c). Among them, most of the reports focused on catalytic atroposelective construction of axially chiral aryl-fused seven-membered rings28,29,30,31,32,33,34,35,36, and only several reports described catalytic atroposelective construction of axially chiral aryl-fused eight-membered rings37,38,39,40,41, which are more challenging to construct than seven-membered rings. Evidently, with the increase of the ring size, the construction of axially chiral aryl-fused medium-sized rings become more and more challenging. Due to the structural characteristics of unfavorable transannular interaction and less stable configuration of axially chiral aryl-fused nine-membered rings, catalytic atroposelective construction of such rings is far more challenging, and there is only one report by Yan’s group in 2022, who utilized the strategy of catalytic asymmetric intramolecular cyclization to construct axially chiral diaryl-fused nine-membered rings in the presence of a cinchona-derived squaramide catalyst42. In spite of this elegant work, there are still some remaining challenges in this field, mainly include: 1) designing new classes of axially chiral aryl-fused nine-membered rings; 2) developing new strategies to construct aryl-fused nine-membered rings; 3) studying the properties and applications of such axially chiral scaffolds.

To address the above challenges, we consider to design a class of axially chiral arylalkene-fused nine-membered rings because axially chiral arylalkenes have recently been recognized as a type of intriguing atropisomers and their catalytic atroposelective synthesis has absorbed intensive attention from the synthetic chemists (Fig. 2a)43,44,45. Nevertheless, the majority of the synthesized axially chiral arylalkenes are six-membered ones such as styrenes, N-vinylquinolinones, N-vinylquinazolinones and so on46,47,48,49,50,51,52,53. By contrast, catalytic atroposelective synthesis of axially chiral five-membered arylalkenes with weak configurational stability is less explored and more challenging, which is limited to axially chiral alkenylindoles and alkenylfurans54,55,56. So, it is also highly valuable to develop efficient strategies toward catalytic atroposelective synthesis of axially chiral nine-membered arylalkenes. In this context and considering the importance of axially chiral alkenylindoles as chiral catalysts and bioactive molecules, we conceive to merge axially chiral alkenylindole scaffold with nine-membered ring, thus designing axially chiral alkenylindole-fused nine-membered rings as a class of axially chiral aryl-fused nine-membered rings (Fig. 2b). Notably, indole-fused nine-membered rings are also the core structures of some bioactive molecules (Fig. 2c)42,57,58,59. So, this design will not only add a member to the family of axially chiral nine-membered arylalkenes, but also provide an opportunity for discovering bioactive molecules.

Fig. 2: Design of axially chiral alkenylindole-fused nine-membered rings and synthetic strategy.
figure 2

a Profile of catalytic atroposelective synthesis of axially chiral arylalkenes. b Design of axially chiral alkenylindole-fused nine-membered rings. c Bioactive molecules containing indole-fused nine-membered rings. d Profile of catalytic asymmetric (4 + 5) cycloadditions. e Our strategy for constructing axially chiral alkenylindole-fused nine-membered rings.

Full size image

To design a synthetic strategy for this class of axially chiral alkenylindole-fused nine-membered rings, we envision whether catalytic asymmetric (4 + 5) cycloaddition could be an efficient strategy toward this goal. In recent years, catalytic asymmetric (4 + 5) cycloaddition has become a powerful tool for constructing chiral nine-membered rings in an enantioselective manner (Fig. 2d)60,61,62,63,64,65,66. However, only metal-catalyzed asymmetric (4 + 5) cycloadditions have been established so far62,63,64,65,66, while organocatalytic asymmetric (4 + 5) cycloadditions have never been documented in the literature, which require exquisite design of the reaction. To develop organocatalytic asymmetric (4 + 5) cycloadditions for the synthesis of axially chiral alkenylindole-fused nine-membered rings, based on our interests in organocatalytic asymmetric construction of indole-based chiral scaffolds67,68, we design a chiral phosphoric acid (CPA)-catalyzed asymmetric formal (4 + 5) cycloaddition of 3-alkynyl-2-indolylmethanols with 2-indolylethanols (Fig. 2e). In this design, 3-alkynyl-2-indolylmethanols are utilized as 1,4-dielectrophilic platform molecules which were designed by our group69,70, and 2-indolylethanols are conceived to serve as 1,5-dinucleophiles for realizing formal (4 + 5) cycloadditions with 3-alkynyl-2-indolylmethanols. Moreover, CPAs are supposed as suitable organocatalysts because CPAs are capable of activating the reactants or intermediates by forming hydrogen-bonding and ion-pairing interactions, therefore facilitating the designed formal (4 + 5) cycloaddition and controlling the enantioselectivity of the generated axially chiral alkenylindole-fused nine-membered rings. The significance of this study lies in that it will not only realize the organocatalytic asymmetric formal (4 + 5) cycloaddition, but also accomplish the catalytic atroposelective synthesis of axially chiral arylalkene-fused nine-membered rings and provide an efficient strategy to construct axially chiral aryl-fused nine-membered rings. Besides, we will perform investigations on the reaction pathway, activation mode and product applications, which will give an in-depth understanding of the organocatalytic asymmetric formal (4 + 5) cycloaddition and this class of axially chiral scaffolds.

Results

Optimization of reaction conditions

To verify our design, 3-alkynyl-2-indolylmethanol 1a and 2-indolylethanol 2a were selected as model substrates to conduct the designed asymmetric formal (4 + 5) cycloaddition in the presence of CPA (R)−4a (Table 1, entry 1). As anticipated, this (4 + 5) cycloaddition smoothly occurred to afford the desired axially chiral alkenylindole-fused nine-membered ring 3aa albeit with a low yield and a poor enantioselectivity (12% yield, 12% ee). To improve the yield and enantioselectivity of product 3aa, we evaluated an array of CPAs 4 (entries 2–11). Among them, CPA (R)−4j with large steric aryl groups at the 3,3′-positions was the optimal choice to deliver product 3aa in the highest enantioselectivity of 85% ee although with a low yield of 13% (entry 10). Additionally, several H8-BINOL and SPINOL-derived CPAs were also investigated, but none of them was better than (R)−4j regarding to control the enantioselectivity (Supplementary Table 6). Then, changing different solvents showed that product 3aa could hardly be generated in other reaction media (entries 12–16). Subsequently, the molar ratio of the substrates was modulated (entries 17-18), and the yield of 3aa could be slightly enhanced to 20% when increasing the amount of substrate 2a (entry 17). In addition, some other reaction parameters such as reaction temperature, additives, concentration, co-catalysts, catalyst loading, halogen-containing solvents and reaction time were investigated (Supplementary Table 7). Finally, it was discovered that increasing the catalyst loading could lead to a greatly increased yield of 55% with an improved enantioselectivity of 94% ee (entry 19). So, such reaction conditions were chosen as the optimal conditions for this formal (4 + 5) cycloaddition.

Table 1 Optimization of reaction conditionsa
Full size table

Investigation on the substrate scopes

With the optimal conditions identified, the substrate scope for this catalytic asymmetric formal (4 + 5) cycloaddition was investigated (Fig. 3). A wide range of 3-alkynyl-2-indolylmethanols 1 and 2-indolylethanols 2 worked efficiently, which afforded the desired axially chiral alkenylindole-fused nine-membered rings 3 in moderate to good yields (up to 74%) with overall high enantioselectivities (up to 98% ee). In detail, substrates 2b–2h with electronically different R1 groups at C5-C7 positions were fully compatible with this (4 + 5) cycloaddition, affording the structurally varied products 3ab–3ah in moderate yields (40%–66%) with good enantioselectivities (83%–94% ee). Moreover, 3-alkynyl-2-indolylmethanols 1b–1d bearing meta- or para-substituted phenyl groups on the Ar moiety participated well in the formal (4 + 5) cycloaddition, delivering axially chiral products 3ba–3dh in generally high enantioselectivities (86%–98% ee).

Fig. 3: Substrate scope of catalytic asymmetric formal (4 + 5) cycloaddition.
figure 3

Reaction conditions: 1 (0.1 mmol), 2 (0.2 mmol), (R)−4j (30 mol%), DCE (1 mL), 3 Å MS (100 mg), and 30 °C for 5 h. Isolated yields were provided, and ee values were determined using HPLC. a35 mol% (R)−4j.

Full size image

Besides, the formal (4 + 5) cycloaddition was also applicable to substrates 1e–1g with different R groups, generating products 3ea-3gh in moderate yields (41%-74%) with good enantioselectivities (81%-98% ee). In addition, the absolute configuration of product 3aa was determined to be (Ra) by single-crystal X-ray diffraction analysis. Moreover, we compared the results of using 20 mol% and 30 mol% catalyst loading (Supplementary Fig. 3). It was discovered some products 3 were obtained in moderate yields and high enantioselectivities with 20 mol% catalyst loading, which were only slightly inferior to those achieved with 30 mol% catalyst loading. So, these results demonstrated that 20 mol% catalyst loading is also feasible to the (4 + 5) cycloaddition.

Investigation on configurational stability and potential applications

Next, we investigated the configurational stability and potential applications of this class of axially chiral alkenylindole-fused nine-membered rings (Fig. 4). The rotational barrier of product 3aa was experimentally calculated as 28.86 kcal·mol−1 (Fig. 4a), which was only slightly higher than the energy barrier (24.0 kcal·mol−1) required for isolating the atropisomers. This result further demonstrated the enormous challenges in constructing alkenylindole-fused nine-membered rings due to the low configurational stability of such skeletons. Moreover, we focused our attention on the potential applications of such axially chiral alkenylindole-fused nine-membered rings (Fig. 4b–d). First, scale-up reaction of 1a with 2h was performed on one-mmol-scale (Fig. 4b), and a maintained yield with a good enantioselectivity was observed (68%, 91% ee). Subsequently, the synthetic transformations were also investigated (Fig. 4c). Compound 5 was successfully synthesized via the diamination of product 3aa. Reaction of product 3aa with Ph2PCl gave chiral phosphane 6 in a moderate yield, and compound 6 could be oxidated with H2O2 to afford phosphamide derivative 7 in a good yield with a maintained enantioselectivity. To our delight, chiral phosphane 6 could act as a chiral ligand in palladium-catalyzed asymmetric allylic alkylation reaction, which generated product 10 in 21% yield with 55% ee (Fig. 4d)71,72. This result demonstrated the potential application of axially chiral alkenylindole-fused nine-membered rings in the development of chiral ligands.

Fig. 4: Investigation on configurational stability and potential applications.
figure 4

a Investigation on the configurational stability. b One-mmol-scale reaction. c Synthetic transformations. d Application in asymmetric catalysis. e Some products 3 with antitumor activity to PC-3 cells. NMP = N-Methylpyrrolidone. BSA = N,O-Bis(trimethylsilyl)acetamide.

Full size image

Furthermore, the potential bioactivity of axially chiral alkenylindole-fused nine-membered rings 3 was investigated via evaluating their cytotoxicity against the human prostate PC-3 cancer cell line (Supplementary Tables 1-3). As shown in Fig. 4e, several products 3 with (Ra)-configuration exhibited inhibition effects against PC-3 cancer cells. Among them, products (Ra)-3ah and (Ra)-3ac showed better cytotoxicity (6.8 and 17.9 μM). To further investigate the difference of two enantiomers on the cytotoxicity against cancer cells, products (Sa)-3ah and (Sa)-3ac were also evaluated. It was discovered that axially chiral alkenylindole-fused nine-membered rings with (Sa)-configuration exhibited higher cytotoxicity than their enantiomers, and (Sa)-3ac exhibited the highest cytotoxicity (4.2 μM). These results implied that axially chiral alkenylindole-fused nine-membered rings have promising applications in medicinal chemistry and it is important to synthesize the two enantiomers of chiral compounds to clarify the difference in their bioactivity.

Control experiments

To provide an in-depth understanding of the mechanism for the catalytic asymmetric formal (4 + 5) cycloaddition, some control experiments were conducted (Fig. 5). By monitoring the reaction of substrates 1a and 2a, we found that the two substrates were converted into an intermediate at first, then the intermediate was further converted into product 3aa. Specifically, when the reaction of 1a with 2a was performed for only 20 min, only trace of product 3aa was detected with the generation of ether derivant 11 in 27% isolated yield (Fig. 5a). When compound 11 was subjected to the standard reaction conditions, product 3aa could be obtained in a moderate yield of 49% with a good enantioselectivity of 91% ee (Fig. 5b), which was similar to the result of the model reaction. These results showed that the ether derivant 11 generated from the dehydration of substrates 1a and 2a was indeed an important intermediate product of this catalytic asymmetric formal (4 + 5) cycloaddition. In fact, the conversion of intermediate product 11 to final product 3aa was sluggish, and the similar outcomes were observed in other cases, which resulted in the observed moderate yields of some products 3. Besides, to disclose the possible activation mode of CPA to the substrates, N-methyl-protected substrates 1h and 2j were respectively examined in this (4 + 5) cycloaddition under the standard conditions, and the yields and enantioselectivities of the corresponding products 3ha and 3aj were dramatically decreased (Fig. 5c–d). These outcomes indicated that the hydrogen-bonding interactions between CPA (R)-4j and the NH groups of substrates 1–2 were crucial to control the reactivity and the enantioselectivity.

Fig. 5: Control experiments.
figure 5

a, Control experiment to detect the intermediate. b, Control experiment to transform the intermediate into final product. c, Control experiment to study the role of the NH group in substrates 1. d, Control experiment to study the role of the NH group in substrates 2.

Full size image

Density functional theory (DFT) calculations on the reaction mechanism

To further elucidate the possible reaction mechanism and activation mode, theoretical calculations were carried out for the catalytic asymmetric formal (4 + 5) cycloaddition between 1a and 2a (Fig. 6). All the calculations were carried out with B3LYP-GD3BJ73,74 functional in Gaussian 16 Rev. A.0375. Additionally, we have performed a conformational search on the key transition states and intermediates using CREST (version 2.11.1)76. As illustrated in Fig. 6a, in the presence of CPA (R)-4j, 3-alkynyl-2-indolylmethanol 1a underwent a dehydration process via transition state TS1, which generated carbocation intermediate INT2’ with an energy barrier of 16.1 kcal·mol−1. Notably, the NH group of 3-alkynyl-2-indolylmethanol 1a formed a hydrogen bond (b1 = 1.889 Å) with the catalyst in TS1, which demonstrated that the NH group of 3-alkynyl-2-indolylmethanol 1a is important in facilitating this dehydration process. After releasing the water molecule from INT2’ to INT2, in INT3, CPA (R)-4j formed three hydrogen bonds with the carbocation intermediate and 2-indolylethanol 2a, which promoted the nucleophilic addition of 2a to the carbocation intermediate via TS2 to afford ether derivant 11 in INT4. This result is consistent with the control experiments (Fig. 5a–b), confirming that compound 11 is indeed the key intermediate product of this reaction. Then, INT4 transformed into INT5 by shifting the position of the catalyst, which facilitated the subsequent isomerization process of INT5 to allene-iminium intermediate INT6 via TS3. Afterwards, the intramolecular addition of indole C3-position to the allene moiety was the key step to construct the nine-membered ring and generate the axial chirality, which occurred via enantioselectivity-determining transition state TS4 to deliver axially chiral alkenylindole-based iminium intermediate INT7. Finally, the rearomatization of indole ring occurred via TS5, resulting in the regeneration of catalyst (R)-4j and the production of alkenylindole-fused nine-membered ring (Ra)-3aa. In the reaction pathway, the rate-determining transition state is TS4 with a maximum reaction energy barrier of 23.6 kcal·mol−1, which agrees with the optimal reaction temperature. Moreover, the NH group of the two substrates formed hydrogen bonds with CPA (R)-4j in almost the whole reaction pathway, which played an important role in controlling the reactivity and enantioselectivity of the catalytic asymmetric formal (4 + 5) cycloaddition and explained the results of control experiments (Fig. 5c–d).

Fig. 6: Theoretical calculations on the reaction pathway of the catalytic asymmetric formal (4 + 5) cycloaddition (kcal·mol−1).
figure 6

a Calculated Gibbs free energy profile of the reaction. b Key transition states leading to the two enantiomers.

Full size image

To explain the absolute configuration of product (Ra)-3aa observed in this (4 + 5) cycloaddition, we compared the enantioselectivity-determining transition states TS4 and TS4’, which corresponded to the formation of products (Ra)-3aa and (Sa)-3aa (Fig. 6b). The difference of Gibbs energy barrier in transition states TS4 (11.6 kcal·mol−1) and TS4’(14.1 kcal·mol−1) was 2.5 kcal·mol−1, which provided a reasonable explanation for the high enantioselectivity (94% ee) and the (Ra)-configuration of product 3aa. The discrepancy of energy barrier between the two transition states TS4 and TS4’ may be caused by the difference of the interactions between the substrates and catalyst (R)-4j. Structural analysis showed that one hydrogen-bonding interaction (b1 = 1.646 Å), two CH····O interactions (b2 = 2.150 Å, b3 = 2.360 Å) and one CH····π interaction (b4 = 2.444 Å) were observed in TS4, whereas only one hydrogen-bonding interaction (b1 = 1.728 Å) and one CH····O interaction (b2 = 2.047 Å) were observed in TS4’, resulting in the lower energy barrier of TS4 than TS4’. Additionally, all these interactions could be visualized by a plot of the noncovalent interactions (NCI) (Supplementary Fig. 1).

Investigation on constructing alkenylindole-fused eight-membered rings

To expand the application of the cycloaddition strategy in the construction of arylalkene-fused medium-sized rings, we further tried the organocatalytic formal (4 + 4) cycloaddition of 3-alkynyl-2-indolylmethanols 1 with 2-indolylmethanols 12 for constructing alkenylindole-fused eight-membered rings 13 (Fig. 7). As shown in Fig. 7a, initially, the reaction of 3-alkynyl-2-indolylmethanol 1a with 2-indolylmethanol 12a was carried out under the catalysis of CPA (R)-4e in DCE at 0 oC, which delivered the desired axially chiral alkenylindole-fused eight-membered ring 13aa but with an unsatisfying result (38% yield, 10% ee). To obtain more information on this class of alkenylindole-fused eight-membered rings, the configurational stability and the rotational barrier of product 13aa were studied. Unfortunately, the racemization process of product 13aa was fast at 25 oC, and the theoretically calculated rotational barrier of product 13aa (21.0 kcal·mol−1) was lower than the rotational barrier (24.0 kcal·mol−1) required for atropisomers with relatively stable configuration. This result implied that this class of alkenylindole-fused eight-membered rings have very unstable configurations, which can hardly be synthesized in an atroposelective manner. So, we focused our attention to develop the strategy of organocatalytic formal (4 + 4) cycloaddition for the synthesis of achiral alkenylindole-fused eight-membered rings and investigate the application of such products.

Fig. 7: Organocatalytic formal (4 + 4) cycloaddition for constructing alkenylindole-fused eight-membered rings.
figure 7

a Investigation on configurational stability and rotational barrier. b Substrate scope of organocatalytic formal (4 + 4) cycloaddition. c One-mmol-scale reaction. d Investigation on the antitumor activity.

Full size image

After establishing the optimal reaction conditions (Supplementary Tables 8–9), the substrate scope of this formal (4 + 4) cycloaddition was examined in the presence of phosphoric acid 4l (Fig. 7b). A number of 3-alkynyl-2-indolylmethanols 1a–1l containing different Ar and R groups were all well-tolerated, which afforded the corresponding alkenylindole-fused eight-membered rings 13aa–13la in moderate to high yields (57%–98%). In details, 3-alkynyl-2-indolylmethanols 1b–1j bearing substituents at either meta- or para-positions of the phenyl ring performed well in the (4 + 4) cycloaddition to give the desired products 13ba–13ja in good yields. Besides, C5- and C6-substituted 3-alkynyl-2-indolylmethanols 1e–1l were accommodated as well to furnish products 13ea–13la in acceptable yields. Moreover, a series of 2-indolylmethanols 12b–12i with substituents at C4-C6 positions of the indole moiety could serve as suitable substrates for the formal (4 + 4) cycloaddition, resulting in the production of alkenylindole-fused eight-membered rings 13ab–13ai in generally high yields (up to 95%). Besides, the structure of product 13aa was confirmed by single-crystal X-ray diffraction analysis.

To demonstrate the practicality of this formal (4 + 4) cycloaddition, one-mmol-scale reaction of substrates 1a and 12a was performed, which afforded product 13aa in a good yield of 79% (Fig. 7c). To find the possible bioactivity of alkenylindole-fused eight-membered rings, a preliminary evaluation on the cytotoxicity of selected products 13 was carried out (Supplementary Table 4). As shown in Fig. 7d, a series of products 13 exhibited some extent of cytotoxicity against PC-3 cancer cells, and the IC50 values were ranging from 7.1 to 19.8 μM. Among them, product 13ka exhibited potent cytotoxicity against PC-3 cancer cells with a low IC50 value of 7.1 μM. These results indicated that this class of alkenylindole-fused eight-membered rings might find potential applications in medicinal chemistry.

Discussion

In conclusion, we have established the catalytic atroposelective construction of axially chiral arylalkene-fused nine-membered rings, which makes use of organocatalytic asymmetric formal (4 + 5) cycloadditions of 3-alkynyl-2-indolylmethanols with 2-indolylethanols. By this strategy, various axially chiral alkenylindole-fused nine-membered rings were constructed in high enantioselectivities with moderate to good yields (up to 74% yield, 98% ee). Theoretical calculations provide an in-depth understanding on the reaction pathway and activation mode of this organocatalytic asymmetric formal (4 + 5) cycloaddition, and application explorations demonstrate the great potential of these axially chiral arylalkene-fused nine-membered rings in asymmetric catalysis and medicinal chemistry. Besides, this cycloaddition strategy could be extended to the construction of alkenylindole-fused eight-membered rings via organocatalytic formal (4 + 4) cycloadditions of 3-alkynyl-2-indolylmethanols with 2-indolylmethanols. This work has not only realized the organocatalytic asymmetric formal (4 + 5) cycloaddition, but also provided an efficient strategy for the construction of axially chiral aryl-fused nine-membered rings, which will add some contents to cycloaddition reactions, axially chiral chemistry and chiral indole chemistry.

Methods

General procedure for the synthesis of products 3

3-alkynyl-2-indolylmethanols 1 (0.1 mmol), 2-indolylethanols 2 (0.2 mmol), chiral phosphoric acid (R)−4j (0.03 mmol) and 3 Å MS (100 mg) were added to a reaction tube. Then, DCE (1.0 mL) was added to the reaction mixture, which was stirred at 30 oC for 5 h. After the completion of the reaction which was indicated by TLC, the reaction mixture was purified through flash column chromatography (petroleum ether/ethyl acetate = 4/1) on silica gel to afford pure products 3.

General procedure for the synthesis of products 13

3-alkynyl-2-indolylmethanols 1 (0.2 mmol), 2-indolylmethanols 12 (0.4 mmol), 4 l (0.04 mmol) and 3 Å MS (200 mg) were added to a reaction tube. Then, DCE (8.0 mL) was added to the reaction mixture, which was stirred at 0 oC for 5 h. After the completion of the reaction which was indicated by TLC, the reaction mixture was purified through flash column chromatography (petroleum ether/ethyl acetate = 4/1) on silica gel to afford pure products 13.

Cytotoxicity test

Human prostate cancer cell line (PC-3) was provided by Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.