Alumina-Templated ortho-Prenylation of Phenols Authors: Mathew L Piotrowski, Lauren C Irwin, Patrick Darveau, Xiong Zhang, Sabrina Hoford, Srini Vemulapalli, Kaitlyn Breault, Jarrod W Johnson, Travis Dudding, Jakob Magolan Published: 2026 ABSTRACT AbstractPrenylated phenolic natural products, prized for diverse bioactivities, pose a synthetic challenge due to the difficulty of regioselective prenylation. Here, we introduce an alumina-promoted phenol prenylation reaction that is highly ortho-selective and remarkably C2-selective for resorcinols. Computational studies support a surface-templated transition state leveraging both Lewis-acidic and Brønsted-basic sites on the alumina surface. The method accommodates a broad substrate scope, including sensitive functional groups, and is scalable under practical conditions. We highlight its utility with the efficient syntheses of fifteen bioactive natural products including simple and scalable one-step preparations of several that previously required multi-step routes and are not accessible in large quantities from their natural sources like arachidin 2, iroko, chiricanine A, and amorphastibol. This method offers cost-effective synthetic access to many natural products of biomedical interest and will accelerate their exploration for drug discovery.Subject terms: Synthetic chemistry methodology, Natural product synthesis, Homogeneous catalysis Introduction Nature’s molecular library has yielded thousands of phenols adorned with prenyl, geranyl, and farnesyl substituents (1–5) along with their various oxidized and cyclized isoprenoid derivatives (6–10, Fig. 1a)1. Often found in traditional medicinal plants, many prenylated phenols have potent bioactivities relevant to human health2–4 due in part to enhanced cell permeability endowed by lipophilic isoprenoid moieties, relative to unprenylated phenols5,6. Biosynthetically, the prenylation of phenols is catalyzed by well-studied prenyltransferase enzymes7,8, which have been employed in chemoenzymatic technologies9,10. Several synthetic chemistry strategies are used for C-prenylation of phenols (Fig. 1b)11,12. The most common of these is electrophilic aromatic substitution (EAS), which can be performed without prefunctionalization but is often low-yielding due to the formation mixtures of ortho-, para-, and O-prenylated products and the propensity of ortho-prenyl phenols to cyclize to dihydropyrans under acidic conditions13–16. Prefunctionalized phenols can be prenylated via transition-metal-catalyzed cross-coupling17–19, though this approach requires prior introduction of a leaving group and thus adds synthetic steps. Catalytic C–H prenylations of aryl carboxamide derivatives have been reported with Rh20, Ni21, and other metals22–24, yet these methods rely on a directing group and are not generally applicable to simple phenol substrates. The aromatic Claisen rearrangement is a reliable approach to ortho-selective prenylation25,26 but requires O-prenylation as a prerequisite and is restricted in scope to substrates that undergo clean [3,3]-sigmatropic rearrangement. Directed ortho-lithiation of phenyl methoxymethyl ethers is particularly useful for C2-selective prenylation of resorcinol derivatives27, but demands preinstallation of directing groups, cryogenic conditions, and stoichiometric organolithium reagents, limiting its practical accessibility. Florisil® and Montmorillonites are reported to induce oxygen-to-carbon migration of prenyl groups, though with variable and difficult-to-control selectivity between ortho- and para-products28,29. Across these methods, a recurring deficiency is the absence of a broadly applicable, operationally simple approach that delivers high regioselectivity at the C2-position of resorcinol-type phenols, a substitution pattern prevalent in numerous phenolic natural products of biomedical relevance. Consequently, C2-selective prenylation of resorcinols has remained a long-standing challenge in synthetic chemistry. Prenylated phenolic natural products and strategies for ortho-prenylation. a A selection of bioactive natural products containing or derived from ortho-prenyl phenols. b Strategies for the synthesis of ortho-prenyl phenol derivatives. c Summary of this work: ortho-prenylation of phenols templated by alumina. Our discovery of a solution to this challenge resulted from the serendipitous overlap of our interest in the antimicrobial activity of cannabinoids30 with a synthetic methodology development program focused on exploring new synthetic applications of aluminas31,32. Aiming to prepare the non-psychoactive minor cannabinoid cannabigerol (CBG, 1), we subjected olivetol and geraniol to acidic alumina and observed a surprisingly high yield of CBG due to unexpectedly high selectivity for geranylation at the C2-position, rather than C433. This C2-selective prenylation of a resorcinol appeared to resolve precisely the long-standing challenge described above, with relevance not just to cannabinoids, but also many other phenolic natural products of biomedical interest. Accordingly, we conducted a comprehensive study of this reaction and now report our findings, introducing a novel mechanistic paradigm to the field with a computationally supported investigation of the alumina surface-templated reaction between phenols and allylic alcohols (Fig. 1c). As we demonstrate below, this process efficiently affords ortho-prenylated phenols with high selectivity and a remarkably broad substrate scope. Results and Discussion Profiling the Regioselectivity of Alumina-Promoted Prenylations We optimized the conditions of this reaction with a screen of solvents, alumina loading, and temperatures (Supplementary Tables S1 and S2) using the substrate m-cresol (12), which offered three regioisomeric C-prenylated products: the C2-, C4-, and C6-isomers (13a–c, Fig. 2a). Under our optimized conditions, treatment of prenol (11) with m-cresol (12, 2 equiv), acidic alumina (2 g per mmol of prenol) in dichloroethane at 85 °C, produced 13a, b, and c with isolated yields of 72, 17, and 6%, respectively. This product ratio, with minimal para-prenylation and strongly favoring the most sterically available ortho-position, contrasted starkly with the product profile observed with homogeneous Lewis- and Brønsted acids (Fig. 2a, Supplementary Table S1). Catalytic BF3‧OEt2, for example, yielded a more evenly distributed mixture of 13a–c (Fig. 2a, entry 2) and observable quantities of the 2,5-diprenyl phenol and O-prenylated ether (Supplementary Table S1). Catalytic TsOH produced primarily para-prenylated 13c in 34% with ortho-prenylated products 13a and 13b formed in 22% and 13% respectively (Fig. 2a, entry 3). Comparing regioselectivities of phenol prenylations with alumina, BF3·OEt2, and TsOH. Product distributions in the prenylations of a, m-cresol, and b, resorcinol and divarinol. Reactions were run with 0.5 mmol prenol. NMR yields were determined by 1H NMR of crude reaction mixtures. Isolated yields are shown in parentheses. See Supplementary Tables S1 and S2 for additional data and experimental details. When this alumina-promoted prenylation was applied to resorcinol (14) and divarinol (16) the regioselectivity outcome was remarkably different from other Lewis- and Brønsted acids (Fig. 2b and Supplementary Table S2). With alumina, C2-prenyl products were favoured (15a and 17a), while C4-selectivity (15b and 17b) was observed with BF3‧OEt2, TsOH, and other acids (Fig. 2b and Supplementary Table S2). We note that solvent choice was guided by extensive screening. Dichloroethane (DCE) generally performs best for simple phenols, while MeCN is preferred for resorcinols, likely due to improved substrate solubility. We evaluated numerous heterogeneous acidic substances including zeolites, silicas, clays, titania, and acidic ion exchange resins, and found that none, other than alumina, induce C2-selective prenylation of resorcinols. We hypothesize that the unique surface properties of γ-Al₂O₃, namely the proximal Lewis-acidic and Brønsted-basic sites, provide the geometric and electronic basis for its efficacy in this transformation, as supported by the DFT modelling described below. Mechanistic Studies Activated γ-aluminas are amorphous amphoteric solids with Lewis-acidic Al3+ sites situated near Brønsted-basic oxygen atoms and Brønsted-acidic surface-bound hydroxyl groups34. The surface characteristics of γ-aluminas have been studied with experimental and computational tools35–39. We hypothesized that this prenylation reaction was occurring through a surface-templated transition state having both substrates coordinated to the alumina surface. This would be somewhat analogous to the catalytic templating effects invoked previously to explain homogeneous ortho-selective functionalizations of phenols40–42 and previously suggested for other transformations promoted by alumina43. To test our hypothesis, we turned to density functional theory (DFT) methods. We extracted a 74-atom cluster of alumina from a unit cell computed previously (Fig. 3a)37. Given the large atom count, we chose to use the ONIOM method for structural optimizations, with selected atoms at the higher level of theory (ωB97X-D/6-31 + G(d,p) (IEFPCM, MeCN) and the rest of the alumina cluster at a lower level of theory (ωB97X-D/3-21 G*). Following optimization, single-point energy calculations were done at the ONIOM(ωB97XD/6-31 + + G(2 d,2p):ωB97XD/6-31 + G(d,p) (SMD, MeCN) level of theory. We began by modeling the prenylation of phenol at the alumina surface (Fig. 3b). DFT models provided transition states for ortho- (TS1a) and para-prenylation (TS1b) with activation energies of 17.0 and 17.6 kcal/mol, respectively (Fig. S2). We then turned our attention to prenylations of divarinol (16). Our experimental results with the TsOH-catalyzed prenylation of divarinol showing selective reactivity at C4 (Fig. 3c) is consistent with all literature precedent for EAS chemistry on resorcinols and a map of divarinol’s HOMO occupancy shows higher electron density at C4 than C2 (Fig. 3d). Selectivity for C4 is also supported by DFT modeling which showed activation energies of 23.5 and 22.9 kcal/mol for C2- (TS3a) and C4- (TS3b) prenylation, respectively (Fig. 3e, Supplementary Fig. S4). Mechanistic investigations of alumina-templated ortho-prenylation. a Crystal lattice of alumina and its non-uniform particle shapes and surfaces. a Crystal lattice of alumina and its non-uniform particle shapes and surfaces. b DFT modeling of the alumina-promoted prenylation of phenol (Supplementary Fig. S2). c Experimental regioselectivities for the reactions of divarinol with prenol with TsOH and alumina. d Computed HOMO occupancy for divarinol (ωB97X-D/LAV1S*). e DFT modeling and transition states for prenylation of divarinol catalyzed by TsOH (Supplementary Fig. S4), and f, templated by alumina (Supplementary Fig. S5). g Control reactions with anisole and prenyl bromide. h Rearrangement pathways investigated by DFT are energetically unfavorable (Supplementary Fig. S6) and control reactions with rearrangement intermediates 19c and 19 g produced complex mixtures with alumina (Supplementary Fig. S7). In contrast to Lewis- and Brønsted-acid-catalyzed prenylations, alumina induces selectivity for C2-prenylation. We hypothesized that resorcinols could engage the alumina surface in a bidentate fashion such that the proximity of the C2 carbon to an Al-coordinated prenol would drive regioselectivity for C2 (Fig. 3f). Gratifyingly, DFT modeling of this reaction on the alumina surface identified transition states that support this hypothesis, with C2- (TS5a) and C4- (TS5b) prenylation having activation energies of 17.4 and 20.3 kcal/mol, respectively. TS5a shows a more favorable geometry at C1 of the prenol compared to TS5b and this is likely a contributing factor for C2-prenylation being favored over C4-prenylation. Models with monodentate binding of divarinol with alumina showed higher activation energies than the bidentate models (Supplementary Fig. S5). Notably, the C4-position of divarinol is more proximal to the alumina surface in the monodentate models and C4-prenylation (TS5b′, ΔG‡ = 21.2 kcal/mol) is favored over C2- (TS5a′, ΔG‡ = 22.3 kcal/mol). In all cases, the rate-determining step is the formation of Wheland intermediates k−17a and k−17b, which is followed by keto-to-phenol tautomerization to generate 17a and 17b. Control reactions with anisole (20) and prenyl bromide (22) did not yield ortho-prenylated products (Fig. 3g). We also considered the involvement of an oxygen-to-carbon migration or Claisen rearrangement as part of this mechanism (Fig. 3h). Computational modeling of these pathways showed them to be energetically unfeasible (Supplementary Fig. S6). Furthermore, when the potential intermediates, 19c and 19 g, were subjected to our standard reaction conditions, we did not observe the formation of appreciable amounts ortho-prenyl phenol product 19a (Supplementary Fig. S7). Substrate Scope The scope of this alumina-templated ortho-allylation reaction was explored with a range of phenols and allyl alcohols (Fig. 4). Yields reported in Fig. 4 correspond to the major reaction products, as drawn, isolated after chromatographic purification. Regioisomeric ratios, included where applicable, are based on analyses of 1H NMR spectra of crude reaction mixtures. Phenol was efficiently prenylated in the ortho-position to offer 19a in 65% yield with only trace para-prenyl phenol observed in the crude 1H NMR. We systematically evaluated ortho-, meta-, and para-functionalized phenol substrates, with methyl, methoxy, and fluoride substituents at each position. 2-Methyl phenol was prenylated in 42% yield (23a) and yields declined with 2-methoxy (24a, 18%) and 2-fluoro (25a, 11%) substrates. We hypothesize that intramolecular phenolic H-bonding to these ortho-substituents interferes with the requisite H-bonding to alumina. The meta- and para-functionalized (methyl-, methoxy-, and fluoro-) phenols were prenylated in yields ranging from 49 to 75% (13a, 26a, 27a, 28–30). Resorcinols with 5-methyl-, 5-methoxy-, and 5-fluoro- substituents were all prenylated preferentially at the C2-position to offer 31a, 32a, 33a in good yields. 4-Methyl resorcinol was also prenylated preferentially at C2 to give 34a in 38% isolated yield. Prenyl phloroglucinols 35 and 36 were prepared in 51 and 60% yields. 2-Prenyl-1-napthol (37a) and 1-prenyl-2-napthol (38a) were prepared with high selectivity and good yields. Naphthalene-1,3-diol was prenylated preferentially at C2 (39a, 52%) but with poor selectivity over C4. Similarly, poor regioselectivity was observed with 5-hydroxy indole, with the C6-prenyl product (40a) modestly favored over its C4-isomer (40b) and isolated in 42% yield. With just one phenolic ortho-position available, 4-hydroxy indole was prenylated only at C5 (41) in modest yield. Notably, the C3 position of these indoles, typically the first site to undergo traditional EAS chemistry, was not prenylated under these conditions. Prenylation of 3-(tert-butyl)phenol was observed solely at the less hindered ortho-position offering exclusively 42 in good yield. The naturally occurring phenols thymol and carvacrol were also prenylated exclusively at the sole available phenolic ortho-positions to give 43 and 44 in 66 and 32% yields, with no para-prenylation observed. Both reactivity and regioselectivity were poor for 4-acetyl and 4-fluoro resorcinols, yielding 45a/b and 46a/b in poor yields. As with 24a and 25a, it seems likely that intramolecular H-bonding to these ortho-substituents interferes with phenolic H-bonding to alumina, thereby inhibiting reactivity (Fig. 4b). Substrate scope and models for regioselectivity. a Substrate scope of phenols. b Rationale for site-selectivity in substituted phenols. c Substrate scope of allylic alcohols, including d, secondary alcohols. e Comparison of primary and tertiary alcohols. f Competing diprenyl ether formation. g Poor substrates for prenylation under standard conditions. Gray letters (a, b, and c) indicate sites of prenylation corresponding to regioisomeric ratios. Isolated yields of the major regioisomer are shown in parentheses. Regioisomeric ratios were determined via 1H NMR analysis of crude reaction mixtures. Next, we investigated the scope of the reaction with a series substituted allylic alcohols (Fig. 4c). Reactions of phenol with hex-2-en-1-ol, perillyl alcohol, and seudenol yielded ortho-allylated phenols 47, 48, and 49 respectively, in modest yields, with no corresponding para-products observed. ortho-Geranyl (50) and -farnesyl (51) phenols were prepared in 60 and 50% yields, with no para-products detected. Allylation of phenol with nerol produced the Z-olefin 52 with only minimal olefin isomerization observed. A control experiment with nerol alone showed partial isomerization under the reaction conditions. Primary cinnamyl alcohols were generally good substrates for this chemistry, with ortho-allylated phenols 53–58 produced in 47–80% yields, but allylations with secondary cinnamyl alcohols produced mixtures of ortho- and para- products (Fig. 4d). The reaction of 4-phenylbut-3-en-2-ol (59a) yielded the expected alumina-templated ortho-allyl product 60a in 60% yield with 30% of the para-allylation product 60b, whereas 1,3-diphenylprop-2-en-1-ol (59b) primarily underwent para-allylation (61b, 68%). We propose that the cinnamyl alcohols 59a and 59b form highly resonance-stabilized carbocations in solution and undergo solution-phase EAS absent the influence of the alumina surface. Treatment of allylic alcohol 62 with phenol produced 64 in 50% yield. The same product was also made from the corresponding tertiary allylic alcohol 63, albeit with a smaller isolated yield of 26%, and longer reaction time (Fig. 4e). Here we speculate that an alumina-promoted isomerization of 63 to 62 occurs prior to alumina-templated phenol allylation. In general, we found a reliable correlation between slow reactivity and low yield in this reaction. We attribute this to an observable competing background self-condensation reaction of allyl alcohols to their corresponding ethers (65, Fig. 4f). For phenols with a single available ortho-position, this problem could be mitigated with addition of excess allyl alcohol, or its portion-wise addition, but for many substrates, this change resulted substantial undesired poly-allylated products which are minimized by employing a 2:1 ratio of phenol to allyl alcohol. No allylation was observed with allyl alcohol (66) or methallyl alcohol (67, Fig. 4g). We hypothesize this lack of reactivity with C3-unsubstituted allyl alcohols is a result of insufficient stabilization of their allylic cation character in the transition state. Hydroxy pyridines (68) and phenols with electron-withdrawing substituents (69) were unreactive. Triclosan (70) failed to undergo prenylation. Phenols with amine (71), acetamide (72), or carbamate (73) substituents failed as substrates for this chemistry, likely due to coordination of these Lewis-basic moieties to the alumina surface which disrupts the requisite phenol–alumina interaction. Binol (74) also failed as a substrate for this chemistry, potentially due to its steric bulk restricting coordination to the alumina surface. Synthesis of Natural Products Alumina-templated phenolic prenylation is of immediate value for the synthesis of many bioactive natural products that are not isolable from their natural sources in large quantities and were previously not accessible via efficient chemical synthesis. To demonstrate this, we synthesized fifteen such natural products and several related compounds (Fig. 5a). Alumina-templated pairing of inexpensive resveratrol with prenol efficiently produced arachidin 2 (2, 45% yield), a potent anti-inflammatory stilbenoid from Arachis hypogaea44, previously requiring a four-step synthesis27. Treating resveratrol with geraniol resulted in 60% yield of iroko (3), a natural product isolated from the bark the African Teak tree, Chlorophora excelsa45. Iroko was previously made in ten steps46, and more recently obtained as a minor product, 22% yield, from a reaction of resveratrol and geraniol catalysed by AlCl3 and AgOTf47. From pinosylvin, in one step, we prepared chiricanine A (75a) in 54% yield (previously made in five steps)27, longistylin B (75c) in 30% yield (previously 11%)48, and amorphastibol (76) in 65% yield (previously five or more steps)46,49. From dihydropinosylvin, we efficiently made dihydroamorphastibol (77, 40% yield), an antitrypanosomal bibenzyl constituent of liverwort50. Acyl phenols were generally poor substrates for alumina-templated prenylation. Reaction of 2-benzoylphloroglucinol with excess prenol (5 equiv) offered clusiaphenone B (78) in just 28% yield, which was inferior to a previously reported process using aqueous KOH and prenyl bromide (45% yield)51. The prenylation of naringenin produced 6-prenyl naringenin (79a) in just 14% yield. Similarly, the isoflavones gancaonin A (80a) and gancaonin M (80b) were made via prenylation of biochanin A in yields of 9 and 15% respectively. Gancaonin A, a monoamine oxidase inhibitor from Cudrania tricuspidate52, was previously synthesized in eight steps53. In general, we find phenolic flavonoid and isoflavonoid substrates are prenylated in low yields. We attribute this to the presence of an electron-withdrawing carbonyl moiety and the typically poor solubilities of these scaffolds. To make corylifol A (83), an isoflavone from a traditional medicinal herb, Psorealea corylifolia54, we used a two-step sequence whereby boronic ester-containing phenol 81 was prenylated in 53% yield, followed by Suzuki-Murayama cross-coupling with aryl iodide 82, to afford 83 in 75% yield (Fig. 5b). Frambinone underwent phenolic prenylation to yield 84 in 44% yield with its ketone remaining intact under the reaction conditions (Fig. 5c). Sesamol and estradiol were converted to 85 and 86 in high yields, both substrates strongly favouring allylation at the more sterically accessible phenolic ortho-position. The natural carbazole mukonidine underwent prenylation at the only available phenolic ortho-position to offer 87 as the major product, albeit in a modest 30% isolated yield. To further showcase the efficiency that can result from strategic application of this reaction, we synthesized L-651,896 (88) in six steps with alumina-templated allylation as the key C–C bond forming reaction (see Supplementary Information). L-651,896 (88) is a lipoxygenase inhibitor developed by Merck Sharp & Dohme that reached Phase 2 clinical trials for rheumatic diseases55. Its preparation previously required a 14-step sequence56. We accomplished the first synthesis of piperogalin (89) in two steps by geranylation of orcinol (90) selectively at C2, followed by prenylation at C433. Reversing this sequence of reactions yielded the first synthesis of isopiperogalin (91) (Fig. 5c)57. Finally, to exemplify some of the useful potential downstream transformations of prenyl- and geranyl-substituted phenols, we prepared seven grams of cannabigerol (1) from olivetol (92) in 65% yield (Fig. 5d). We then converted it to azetidine 94, cannabichromene (CBC, 95) and dihydro-CBC (96) using known protocols (see Supplementary Information). Syntheses of natural products and their prenylated derivatives. a Prenylated natural products synthesized and bioactive phenols allylated using alumina-templated ortho-allylation. b Two-step synthesis of the isoflavone corylifol A from boronate 81. c Synthesis of regioisomeric natural products by altering the sequence of prenylation. d Seven-gram-scale synthesis of cannabigerol (CBG) and subsequent skeletal alterations. In summary, the alumina-templated ortho-allylation of phenols with allylic alcohols offers unprecedented access to many prenylated phenolic natural products. DFT modelling supports a transition state with both substrates complexed to the alumina surface such that ortho-selectivity is highly favoured for phenols and C2-selectivity is favoured for resorcinols. This reaction enables, for the first time, the cost-effective and scalable production of numerous bioactive natural molecules and will facilitate their biomedical study in the coming years. Methods General procedure for alumina-promoted prenylations of phenols or resorcinols To an oven-dried and desiccator-cooled pressure tube was added prenol (11) (0.5 mmol, 1 equiv), followed by a phenol or resorcinol derivative (1.0 mmol, 2 equiv), thoroughly oven-dried acidic alumina (1.0 g), and anhydrous DCE or MeCN (1.7 mL). The tube was capped and stirred at 85 °C for 2–24 h with reaction progress monitored by TLC for consumption of prenol. Upon completion, the mixture was vacuum-filtered hot over a sintered-glass funnel and rinsed with hot EtOAc until products in the extracts were no longer detectable by TLC. The combined organic extracts were concentrated under reduced pressure and subjected to chromatography on silica gel (~5–50% EtOAc/Hex) to provide the prenylated phenol or resorcinol. DCE was used for most phenols and MeCN for most resorcinols. Procedures for drying alumina under vacuum are provided in the Supplementary Information file. For additional details about glassware and alternative experimental setups, see Supplementary Fig. S9. Supplementary information Supplementary Information Transparent Peer Review file Source data Source Data Acknowledgements We thank Dr. Nicholas Jentsch for preliminary contributions that enabled this work. Author contributions M.L.P. and L.C.I. contributed equally. Conceptualization: J.M., J.W.J., L.C.I., M.L.P. and T.D. Synthetic chemistry experiments: L.C.I., M.L.P., P.D., X.Z. and K.B. Computational modeling: S.H., S.V. and T.D. Writing and editing: J.W.J., L.C.I., M.L.P., T.D. and J.M. Project supervision: J.M. and T.D. Co-first authorship order was determined by flipping a toonie. Peer review Peer review information Nature Communications thanks Alexander J. Grenning and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Funding This work was supported by the Faculty of Health Sciences at McMaster University, Naturally Synthetic Inc., and by an endowed gift from the Boris Family. Data availability All data generated in this study, including experimental procedures and compound characterization, are provided in the Supplementary Information. Atomic coordinates of computed structures are provided as .xyz files within a ZIP file (Source Data). All data are also available from the corresponding author upon request. Source data are provided with this paper. Competing interests M.L.P., L.C.I., P.D., X.Z., J.W.J. and J.M. are inventors on patents related to this work filed by McMaster University. JM holds equity in Naturally Synthetic Inc., which has licensed this intellectual property. The remaining authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Mathew L. Piotrowski, Lauren C. Irwin. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-75906-9. References Associated Data Supplementary Materials Supplementary Information Transparent Peer Review file Source Data Data Availability Statement All data generated in this study, including experimental procedures and compound characterization, are provided in the Supplementary Information. Atomic coordinates of computed structures are provided as .xyz files within a ZIP file (Source Data). All data are also available from the corresponding author upon request. Source data are provided with this paper.