Cyclic sulfone ring remodeling enables molecular shape diversity-oriented synthesis of privileged biaryl and oligoaryl motifs

Abstract

Molecular shape is a key determinant of the structural properties and intermolecular interactions of privileged scaffolds, such as biaryl and oligoaryl compounds, which are widely used in the design of drugs, catalysts, and functional materials. Despite the well-recognized importance of molecular shape diversity and increased three-dimensionality, the chemical space of biaryl compounds remains dominated by linear and disk-like structures featuring para and meta substitution patterns. In contrast, more sphere-like bis-ortho-substituted biaryl scaffolds have been underexplored, because of the dearth of synthetic methods for generating focused libraries of ortho analogs in a synthetically efficient manner. We report herein the development of regioselective nucleophile/electrophile couple-mediated cyclic sulfone ring remodeling that can streamline access to a wide range of biaryl and oligoaryl scaffolds, featuring aliphatic and (hetero)aromatic amines, phenols, (thio)ethers, and phosphines. To facilitate prediction of the regiochemistry of structurally complex biaryl products, a machine learning model based on the previously unexplored Bernoulli Naïve Bayes classifier is developed. The integrated experimental, computational, and data-driven approach charts the course for mitigating structural biases and exploring the underpopulated biaryl chemical space in medicinal chemistry, catalysis, and materials science.

The authors report the development of regioselective nucleophile/electrophile couple-mediated cyclic sulfone ring remodeling that can streamline access to a wide range of biaryl and oligoaryl scaffolds, featuring aliphatic and (hetero)aromatic amines, phenols, (thio)ethers, and phosphines.

Introduction

Molecular shape is a fundamental property central to the interactions of drugs and natural products with biological targets.^1–4 Topological and geometric features also define the physical and chemical properties of advanced functional materials.^5,6 Among the various available molecular scaffolds, biaryl and oligoaryl structures consisting of two or more contiguous aromatic carbocyclic and heterocyclic rings occupy a privileged position in drug discovery (Fig. 1A). They exhibit favorable chemical stability and conformational rigidity and enable precise positioning of peripheral functional groups that guide specific interactions with the molecular targets of interest.^7–9 The biaryl motif is also a key structural feature of a broad array of ligands designed to provide precise control of the reactivity and chemical environment of catalytic metal centers and improve their stability in metal-catalyzed reactions.^10–12 The widespread adoption of biaryl motifs in drug discovery and catalysis has been aided by the development of diverse synthetic methods. These approaches have been particularly successful in improving synthetic access to para- and meta-substituted biaryl structures.⁸ In contrast, the construction of more sterically crowded bis-ortho-substituted biaryl motifs remains challenging, which limits synthetic opportunities for generating focused libraries of diverse ortho-analogs.^13–26 The disparity in the synthetic accessibility of various biaryl substitution patterns has resulted in overpopulation of the chemical space of medicinal biaryl motifs by linear and disk-like structures dominated by para and meta substitution patterns, at the expense of sphere-like biaryl structures that are more equally represented in all three dimensions and more common for ortho-substituted biaryl motifs.⁸ Enhanced three-dimensionality is a hallmark of natural products and has been linked to increased clinical success of medicinal molecules and been recognized as a key feature of drug design for particularly challenging molecular target classes, such as protein–protein interactions.²⁷ In this context, oligoaryl motifs have emerged as important structurally rigid scaffolds that enable precise positioning of peripheral functional groups for interactions with complex protein targets.^28,29 Despite the well-recognized importance of bis-ortho-substituted biaryl and oligoaryl scaffolds in medicinal chemistry,⁸ an analysis of the molecular shape space and distribution of substitution patterns for biaryl structures in the current ChEMBL(v35) database of medicinal compounds indicates that the bias has remained unaddressed (Fig. 1B, C). Moreover, the ortho-substituted biaryl motif has emerged as the most commonly used structural element in ligand design because it facilitates efficient two-point interactions with the catalytic metal center, which is not available for meta and para isomers.^10–12 Thus, methods that enable the construction of focused libraries of ortho-substituted biaryl and oligoaryl structures from common and readily accessible precursors could streamline access to the underpopulated yet critically important shape space and correct the bias in the molecular shape diversity of key structural elements in drug discovery and catalysis.

Fig. 1. Cyclic sulfone ring remodeling for improved access to bis-ortho-substituted biaryl and oligoaryl scaffolds.

A Applications of biaryl and oligoaryl scaffolds. B Distribution of biaryl substitution patterns in ChEMBL(v35); o, ortho; m, meta; p, para. C Molecular shape diversity plot for medicinal biaryl compounds in ChEMBL(v35). D Approaches for the generation of focused libraries of bis-ortho-substituted biaryl compounds. E Construction of bis-ortho-substituted biaryl and oligoaryl scaffolds by predictive regioselective nucleophile/electrophile couple-mediated cyclic sulfone remodeling. ML, machine learning.

Although cross-coupling methods can be used to produce ortho-biaryls, they typically require stepwise preassembly of each coupling partner to synthetically encode a specific ortho-substituent (Fig. 1D).^13–15 This leads to rapid accumulation of undesired synthetic manipulations when a focused library of biaryls bearing diverse ortho-substituents is targeted for the evaluation of the privileged ortho chemical space around a conserved (hetero)biaryl core.^11,30,31 For example, a total of 100 synthetic manipulations will be required to produce a focused library of 20 biaryls, if each coupling partner is synthesized in only two steps. In contrast, a divergent approach, which leverages a common precursor that is readily assembled in several steps can grant rapid access to a focused library of ortho-biaryls in a substantially more efficient manner. In this case, the same focused library can be produced with less than 25 synthetic manipulations, if the common precursor is synthesized in less than 5 steps.

Skeletal editing of cyclic molecular scaffolds has recently gained importance as an efficient synthetic approach for diversification of core frameworks.^32–48 This approach complements the well-established synthetic strategies based on modification of substrate structures via functional group interconversions at the periphery of the scaffold. We posited that remodeling of the central ring in cyclic sulfones 1 by sulfonyl group excision with a nucleophile–electrophile couple (Fig. 1E) could unlock a regioselective approach to bis-ortho-substituted biaryl and oligoaryl scaffolds. A cyclic sulfone remodeling approach that combines nucleophilic ring opening and desulfitative cross-coupling would enable rapid evaluation of the underexplored chemical space by screening diverse nucleophilic and electrophilic coupling partners. Additionally, cyclic sulfones 1 are air- and moisture-stable and readily synthetically accessible from abundant feedstock chemicals.^49–56 Furthermore, a plethora of functional groups can be introduced by simple peripheral modifications,⁴⁹ making sulfones 1 particularly suitable for constructing focused libraries of structural analogs.

This approach would unlock an unexplored area in the chemistry of sulfones 1, which were previously converted to carbazoles, fluorenes, and phenanthrenes with retention of the tricyclic framework in pioneering studies by Yorimitsu.^32,36,39 In a departure from this established precedent, cyclic sulfone remodeling with a nucleophile–electrophile pair would pose the unprecedented challenge of controlling and predicting the regiochemical outcome of the ring deletion process with complex substituted substrates. Furthermore, the facile carbazole formation after ring opening³³ and the thermal instability of intermediate sulfinates²³ could present additional problems for the efficient introduction of electrophilic fragments.

Taken together, the unprecedented nature of nucleophile/electrophile-mediated cyclic sulfone remodeling calls for an integrated strategy that synergizes synthetic, computational, and data-driven techniques to successfully develop this previously unexplored synthetic approach to biaryl and oligoaryl scaffolds.

We report herein the development of a cyclic sulfone remodeling approach that can streamline production of focused libraries of bis-ortho-substituted biaryl scaffolds via regioselective nucleophile/electrophile couple-mediated sulfonyl group excision. This method provides a simple synthetic segue to specifically substituted oligoaryl products via iterative sulfone ring remodeling. We show that this approach enables access to a chemical space that is underpopulated in modern medicinal chemistry. Furthermore, a binary classification model has been developed to facilitate prediction of the regiochemistry of the biaryl products in complex structural settings by leveraging the Bernoulli Naïve Bayes algorithm, which to the best of our knowledge has not been previously used for selectivity prediction of synthetic methods.

Results and discussion

Optimization studies revealed that reaction of sulfone 1a with aniline 2 and aryl bromide 3 as a nucleophile/electrophile couple in MeTHF as a solvent and with Pd(OAc)₂/CyPPh₂ as a catalyst readily produces bis-ortho-biaryl product 4a in good yield (Table 1, entry 1). Lower yields were obtained in other solvents (entries 2 and 3) and with other phosphines (entries 4–8). The combination of KHMDS as a base and potassium carbonate to sequester sulfur dioxide⁵⁷ was optimal. Diminished reaction efficiencies due to the decomposition of the intermediate sulfinate salt and the formation of the carbazole by-product were observed without the basic reagents or with alkoxides, amines, and other inorganic salts (entries 9–14). The interruption of the sulfone remodeling process after the reaction of sulfone 1a and aniline 2 in the presence of the base afforded intermediate 1a-2 in 95% yield in agreement with previous studies³³ and the results presented in Table 1, confirming that the reaction proceeds via an initial nucleophilic ring opening of the sulfone substrate and that the Pd-catalyzed coupling plays a key role in the overall efficiency of the sulfone ring remodeling process. Furthermore, kinetic studies using variable time normalization analysis⁵⁸ (VTNA) and the method of initial rates (Figure S1) revealed that the palladium-catalyzed cross coupling of intermediate 1a-2 is first order in the palladium catalyst and bromide 3 and zero order in 1a-2 and potassium carbonate. These results point to the oxidative addition as the turnover-limiting step.

Table 1.

Optimization studies for cyclic sulfone remodeling ^a

Entry	Variations from standard conditions	Yield of 4a, %
1	None	86 (84)b
2	In toluene	70
3	In dioxane	76
4	With Cy3P instead of CyPPh₂	56
5	With tBuPPh2 instead of CyPPh₂	53
6	With XPhos instead of CyPPh₂	2
7	With BrettPhos instead of CyPPh₂	4
8	With Cy-DPEPhos instead of CyPPh₂	0
9	No KHMDS	0
10	No K2CO3	0
11	With KOtBu instead of KHMDS	4
12	With DBU instead of KHMDS	0
13	With Na2CO3 instead of K2CO3	59
14	With K3PO4 instead of K2CO3	64

^a Reaction conditions: sulfone 1a (0.1 mmol), aniline 2 (0.11 mmol), KHMDS (0.17 mmol), MeTHF (0.4 mL), 40 °C, 12 h; then aryl bromide 3 (0.11 mmol), palladium acetate (0.005 mmol), cyclohexyldiphenylphosphine (0.011 mmol), potassium carbonate (0.3 mmol), MeTHF (1 mL), 120 °C. The yield was determined by ¹H NMR with 1,3,5-trimethoxybenzene as an internal standard. ^b Isolated yield. KHMDS, potassium bis(trimethylsilyl)amide; DBU, diazabicycloundecene; MeTHF, 2-methyltetrahydrofuran.

The scope of aryl and heteroaryl bromide coupling partners was next examined in a reaction with sulfone 1a and aniline 2 (Fig. 2). Aryl groups bearing a number of common carbon- and heteroatom substituents in the ortho, meta, and para positions, including phenyl and fluoro (4c), thio (4d), difluoromethoxy (4e), and ketone (4f) were readily installed. Heteroaryl bromides were also suitable coupling partners. Teraryl systems featuring oxygen and sulfur heterocycles with furan (4g), benzofuran (4h), dibenzofuran (4i), thiophene (4j, 4k) and benzothiophene (4l) could be obtained following the standard procedure. An array of nitrogenous heteroaryl bromides were similarly competent electrophiles, producing indole (4m), pyridine (4n, 4o), and quinoline (4p, 4q) oligoaryl congeners. A pyrazine derivative with an unprotected amino group (4r) was also produced. Other medicinally important nitrogen heterocycles, including benzoimidazole (4s), pyrazolopyridine (4t), and pyrimidine (4u) could also be efficiently introduced, while successful installation of a tetrahydropyridine ring into the ortho,ortho-biaryl framework (4v) demonstrated that vinyl bromides were equally suitable electrophiles. The reaction was also efficient with a wide range of nucleophiles (Fig. 3). Anilines bearing electron-deficient and electron-rich substituents in the ortho, meta, and para positions, including halogen (5a, 5b), thioether (5c), ether (5e, 5f), and alkyl groups, as well as heterocyclic motifs, were readily installed. Notably, the reaction tolerated sterically encumbered ortho-tolyl and mesityl substituents on the reacting nitrogen atom (5d, 5g). Other nitrogen nucleophiles were equally competent coupling partners. Similarly, the heteroaromatic pyrrole moiety was successfully introduced into the oligoaryl framework (5j). Aliphatic amines, such as pyrrolidine and morpholine, also proved to be compatible with the reaction conditions (5k, 5l).

Fig. 2. Scope of aryl and heteroaryl bromides.

Isolated yields are reported. Reaction conditions: sulfone 1a (0.1–0.15 mmol), aniline 2 (0.11–0.15 mmol), KHMDS (0.17–0.25 mmol), MeTHF (0.4 mL), 40 °C, 12 h; then aryl bromide (0.1–0.11 mmol), palladium acetate (0.005 mmol), cyclohexyldiphenylphosphine (0.011 mmol), potassium carbonate (0.3 mmol), MeTHF (1 mL), 120 °C. See the Supplementary Information for further experimental details.

Fig. 3. Scope of the nucleophilic coupling partners.

^a With LiNH₂ as the nucleophile. ^b With KOH as the nucleophile. Isolated yields are reported. Reaction conditions: sulfone 1a (0.1–0.15 mmol), nucleophile (0.11–0.4 mmol), KHMDS (0.17–0.3 mmol) or LiHMDS (0.21 mmol), MeTHF (0.3–0.45 mL), 25–120 °C, 12 h; then aryl bromide (0.11–0.2 mmol), palladium acetate (0.005–0.0075 mmol), cyclohexyldiphenylphosphine (0.011–0.016 mmol), potassium carbonate (0.3–0.45 mmol), MeTHF (1 mL), 120 °C. See the Supplementary Information for further experimental details.

Primary anilines are versatile intermediates, yet direct introduction of an unsubstituted amino group remains a long-standing synthetic challenge. We found that an unprotected amino group could be directly introduced in a coupling reaction with lithium amide (5n). Notably, structural analogues of amine 5n were recently reported as potential glutaminyl cyclase isoenzyme inhibitors for the treatment of cancer.⁵⁹ Furthermore, reactions with other heteroatom-centered nucleophiles were similarly efficient. A thioether group was readily installed in a reaction with a thiol (5o). An unprotected phenol could also be directly accessed with potassium hydroxide as a nucleophile (5p). Alcohols were equally suitable nucleophiles, affording the corresponding teraryl ethers in good yields (5q-5s).

The reaction could be efficiently conducted with combinations of structurally and electronically diverse (hetero)aryl bromides and nucleophiles, enabling simple access to both fully N-substituted anilines 5t and 5u and unsubstituted aniline 5v from the same biaryl precursor. Notably, the reaction provided a straightforward approach to constructing biarylphosphine 5w with a pyridine ortho-substituent, indicating potential applications in phosphine ligand design.

Peripheral substituents play key roles in modulating the target interactions, physicochemical properties, and structural features of biaryl compounds. Substituents in the peripheral benzene rings can also change the favored direction of reactions with the nucleophile, which determines the positions of the ortho substituents installed in the biaryl products.

Therefore, we next examined how peripheral substituents in congeners of cyclic sulfone 1 control the reactivity and positions of the incoming nucleophilic and electrophilic coupling partners (Fig. 4). A symmetrical binaphthyl framework, which is a common structural motif in ligand design, was readily produced (6a). The unsymmetrical mononaphthyl analogs 6b and 6c were isolated as single isomers, demonstrating the strong preference for nucleophilic attack of the naphthyl moiety. The 2-naphthyl congener of sulfone 1 also underwent ring-opening exclusively at the side of the more extended π-system (6d). Interestingly, exclusive selectivity for ring opening at the unsubstituted phenyl ring was observed for sulfones bearing alkyl groups in the para and ortho positions with respect to the sulfonyl group in the sulfone reactants (6e and 6f). In contrast, phenyl substituents in the para and ortho positions directed the incoming nucleophiles to the substituted sulfone ring (6g and 6h). Similarly, a sulfone bearing a pyridyl substituent in the meta position afforded oligoaryl amine 6i, indicating that ring opening occurred at the substituted ring. Notably, methoxy and chloro substituents in the meta (C3) position directed the nucleophile to the heteroatom-substituted ring (6j and 6k). Additionally, a sulfonyl group in the para (C2) position afforded oligoaryl 6l with the amino group in the sulfonyl-substituted ring. Similarly, when indole and azaindole were appended through a C–N linkage in the C3-meta position, the reaction afforded oligoaryl products 6m and 6n, featuring the amino group in the same ring as the nitrogen heterocycles, which are important structural motifs in drug discovery.⁶⁰ Conversely, the nucleophile was installed in the less substituted phenyl ring when para-C2 indole was present (6o). We next explored the directing effects of sulfones featuring different substitution patterns in the opposing phenyl rings (6p to 6r). Nucleophilic ring opening preferentially occurred at the side of the para-C2 trifluoromethyl group over the more electron-rich para-C2 methyl-substituted ring (6p). Likewise, the para-C2 phenyl-substituted side of an unsymmetrically substituted sulfone reactant was also favored over the opposite side bearing the para-C2 alkyl group (6q). Notably, the reaction also afforded oligoaryl 6r as a single regioisomer from the corresponding sulfone reactant bearing methyl groups in different positions (para-C2 and meta-C3) in the side rings. Ring opening occurred in the meta-C3 methyl-substituted ring, indicating that the reaction can be used to control the topology of the oligoaryl system, even for sulfone reactants with small structural differences between the peripheral rings. Furthermore, sulfone reactants featuring peripheral heterocyclic ring systems were also compatible (6s to 6x). A carbazole-containing oligoaryl ether 6s was readily accessed with the nucleophile favoring the heterocyclic side of the sulfone reactant. Other nitrogen heterocycles, including pyridine and quinoline also afforded oligoaryl derivatives produced by the ring opening at the heterocyclic side (6t and 6u). Similarly, the benzothiophene-derived oligoaryl system 6v was efficiently constructed, with the nucleophile favoring the carbocyclic side. Structurally related heteroaromatic ethers 6w and 6x featuring polycyclic sulfur-containing topologies used in advanced materials⁶¹ were also smoothly produced in a reaction facilitated by nucleophilic ring opening at the heterocyclic side. Notably, all of the substituted cyclic sulfones were readily accessed one to three steps from commercially available precursors.

Fig. 4. Regioselective construction of diverse oligoaryl products.

^a Ring opening with 3 equiv. of KOEt at 90 °C. ^b Ring opening at 70 °C. ^c Ring opening at 120 °C. Isolated yields are reported. Aryl bromide 3 was used as the electrophile for all reactions. Reaction conditions: sulfone (0.1 mmol), nucleophile (3 mmol for potassium methoxide or ethoxide), or nucleophile (0.11 mmol) and KHMDS (0.17 mmol), MeTHF (0.4 mL), 40 °C, 12 h; then aryl bromide 3 (0.11 mmol), palladium acetate (0.005 mmol), cyclohexyldiphenylphosphine (0.011 mmol), potassium carbonate (0.3 mmol), MeTHF (1 mL), 120 °C. See the Supplementary Information for further experimental details.

We next sought to demonstrate the applicability of the method in the more complex structural settings of drugs and natural products (Fig. 5). The biaryl analogs of antidiabetic empagliflozin 7a, antitubercular bedaquiline 7b, anti-cancer irosustat 7c, anti-inflammatory naproxen 7d, and antispasmodic hymecromone 7e were expediently accessed from the corresponding aromatic bromide, iodide, and sulfonate precursors. Topologically defined oligoaryl analogs of tyrosine (7f), δ-tocopherol (7g), and estrone (7h) could also be readily constructed. Coupling with the hindered vinyl triflate congener of dehydroepiandrosterone afforded analog 7i bearing a biaryl fragment in the D ring. The reaction also successfully produced derivatives of diosgenin 7j and ergosterol 7k, indicating that the method can be used to construct oligoaryl–polyene conjugates. Furthermore, diosgenin could also be used as the nucleophilic coupling partner (7l). The teraryl analog of the antimicrobial drug sulfadimethoxine 7m was likewise produced in 62% yield. Additionally, estrone and cholesterol congeners 7n and 7o with predefined substitution patterns in the oligoaryl fragments were also regioselectively constructed using the suitably substituted sulfone reactants.

Fig. 5. Structural modification of drugs and natural products.

Isolated yields are reported. (Hetero)aryl bromides (X = Br) were used unless otherwise specified. Reaction conditions: sulfone (0.1–0.15 mmol), nucleophile (0.3 mmol for potassium ethoxide), or nucleophile (0.11–0.225 mmol) and KHMDS (0.17–0.3 mmol), MeTHF (0.4–0.45 mL), 40–90 °C, 12 h; then aryl (pseudo)halide (0.11–0.3 mmol), palladium acetate (0.005–0.0075 mmol), cyclohexyldiphenylphosphine (0.011–0.016 mmol), potassium carbonate (0.3–0.45 mmol), MeTHF (1 mL), 120 °C. See the Supplementary Information for further experimental details.

To demonstrate the synthetic advantages of the sulfone ring remodeling process in enabling efficient biaryl construction, biaryl product 6l was synthesized by a conventional stepwise biaryl elaboration strategy (Fig. 6A and S5). To that end, prerequisite precursors 8 and 9 were synthesized following literature procedures in three and two steps and 77% and 55% yields. The subsequent cross-coupling reaction of intermediates 8 and 9, followed by a base-mediated fluoride displacement reaction with aniline 2 afforded biaryl 6l in a total of seven steps and 26% yield over the longest linear sequence. By comparison, the sulfone ring remodeling approach afforded product 6l in 3 steps from dibenzothiophene 10 and cyclohexanecarboxylic acid⁶² and 49% overall yield, indicating that this approach can provide significant improvement in efficiency and step economy.

Fig. 6. Construction of topologically defined biaryl and oligoaryl congeners and the development of a binary classification model for the prediction of the regiochemistry of the cyclic sulfone ring remodeling.

A Comparison of the sulfone ring remodeling approach with a conventional stepwise biaryl elaboration strategy; NBS, N-bromosuccinimide; mCPBA, meta-chloroperbenzoic acid; l.l.s., longest linear sequence. B Synthesis of oligoaryl congeners. See Table 1 and the Supplementary Information for the experimental details. C Generation of a density functional theory (DFT)-derived dataset for simple model substrates and nucleophiles with binary encoding of the reaction outcomes (1 for the lower ΔG^≠ site and 0 for the higher ΔG^≠ site); HOMO, highest occupied molecular orbital; LUMO, lowest occupied molecular orbital. D Featurization using computationally derived physicochemical descriptors. E Binary classification model selection. NB Naïve Bayes; SVM support vector machine; RF random forest; QD quadratic discriminant; PA passive aggressive. F Model confusion matrix and statistical performance metrics for the training and the experimentally derived external validation (test) sets. TP true positive; FP false positive; TN true negative; FN false negative.

The method is also well suited for efficient construction of topologically defined oligoaryl scaffolds (Fig. 6B). The regiochemically defined ethoxide-initiated cross-coupling of sulfones 11 and 12 afforded sulfone 13, which was subjected to coupling with a different set of nucleophilic and electrophilic partners (methoxide and 3). Product 14 features five consecutively connected aryl fragments with predefined substitution patterns, giving rise to a topologically distinct oligoaryl framework. This iterative approach can be extended to a threefold coupling sequence with three different nucleophile/electrophile pairs. To that end, a regioselective methoxide-initiated coupling with disubstituted sulfone 15 afforded intermediate 16. Subsequent coupling with cyclohexanol and disubstituted sulfone 17 also proceeded regioselectively, producing sulfone 18. Finally, a reaction of sulfone 18 with pyrrole and bromide 3 generated oligoaryl congener 19, which was composed of seven contiguous aryl fragments, featuring ring-specific and preprogrammed substitution patterns. These results indicate that the protocol enables expedient access to topologically defined oligoaryl frameworks that can be used to display various peripheral functional groups in preset positions.

Inspection of the regioselectivity trends in the monosubstituted dibenzothiophene sulfones presented in Fig. 4 reveals that nucleophilic ring opening preferentially takes place in the more electron deficient arene ring. To gain insight into the structural effects, which underlie the observed regioselectivity, computational studies were carried out for the kinetic reference of the ring opening process with 6b, 6j, 6k, 6t-6v as representative substrates (Figures S7–S18). Distortion–interaction activation strain model (ASM) analysis⁶³ revealed that the regioselectivity is governed by a combination of lower distortion in the substrate fragment and a higher interaction energy for the kinetically preferred transition states. Furthermore, energy decomposition analysis⁶⁴ (EDA) (Table S1 and Figures S19–S24) indicated that electrostatic interactions (Coulombic interactions between the reacting fragments) provide the major contribution to the stabilization of the lower-energy transition states, with charge transfer (orbital interactions) and polarization (intrafragment electrostatic interactions) providing additional stabilization. These results suggest that the regioselectivity is dictated by the relative ability of the competing (hetero)aromatic rings on either side of the sulfonyl group to stabilize the incoming negative charge.

Although this chemically intuitive heuristic can be useful when substituent effects are unambiguous (for example, 6e, 6f, and 6j), it becomes less practicable with substrates bearing multiple substituents in different positions and with polycyclic and heteroarene-derived sulfones (for example, 6r, 6s, and 6v).

We posited that a binary classification model^65–69can be developed using computationally inexpensive molecular descriptors, such as orbital coefficients (e.g., c_HOMO and c_LUMO), C–S bond distances, the percent buried volume, and natural bond orbital (NBO) charges of the reactants, to facilitate prediction of the ring-opening site. Because of the complex interplay of structural and electronic effects in the polycyclic conjugated system of the sulfone substrates, additional composite features (e.g., Δc_LUMO(C1,C4), c_HOMO(C1)×c_HOMO(C3), and c_HOMO(C4)/c_HOMO(C2), where C1–C4 denote positions in the sulfone ring with respect to the reactive site α to the sulfone group) were also constructed to better capture the representative substrate characteristics. However, the limited size of the available experimental dataset poses a challenge for training and validation of a machine learning model. We therefore opted to train the model on a more extensive dataset of computationally derived regioselectivities for a range of prototypical cyclic sulfones and nucleophiles (Fig. 6B-E). As an added advantage, this approach enables more robust external validation of the entire experimental dataset of products 6b-6x. To this end, evaluation of an array of classifiers revealed that Bernoulli Naïve Bayes provided the best performance with respect to the F₁ score,⁷⁰ Matthews correlation coefficient (MCC), which was shown to be a more informative and reliable score in evaluating binary classifications than accuracy and F₁ score,⁷¹ and Efron’s pseudo-R^272,73 value on the training dataset.⁷⁴ Bernoulli Naïve Bayes combines computational efficiency and low sensitivity to noise⁷⁵ and, to our knowledge, has not been previously used for the prediction of synthetic method selectivities. The developed classification model maintained excellent performance across an external validation set, indicating that it can be used to predict the outcomes of reactions in diverse structural settings. The use of composite features was key to the successful development of the classification model (Figure S29), underscoring the importance of representative reactivity descriptors for selectivity.

To assess the potential of sulfone ring remodeling to improve the shape diversity of medicinal biaryl compounds, we compared the chemical space of the products presented in Figs. 2–4 with that of a set of 6,000 representative biaryl medicinal compounds from ChEMBL(v35) (Figure S26). Although no attempt to target the underpopulated molecular shape areas by selecting bespoke nucleophile/electrophile couples was made, the plot indicates that the synthetic products provide a more even distribution across the geometric chemical space. Furthermore, the plot also shows that the chemical space of the sulfone ring remodeling products is currently underpopulated by medicinal biaryl compounds from ChEMBL. For example, 84% of the sulfone ring remodeling products are located outside the area occupied by 77% of the biaryl structures in the ChEMBL database. These results highlight the potential of the sulfone ring remodeling methodology to improve synthetic access to underpopulated medicinal chemical space.

In summary, we have developed a cyclic sulfone remodeling method that enables the generation of bis-ortho-substituted biaryl scaffolds by nucleophile/electrophile couple-mediated sulfonyl group excision. The method can streamline the construction of focused libraries for the evaluation of the privileged ortho chemical space. The products feature a wide array of functionalities, including aliphatic and (hetero)aromatic amines, phenols, ethers, thioethers, and phosphines, making the method suitable for rapid construction of focused libraries of various ortho-substituted analogs. Furthermore, cyclic sulfone remodeling provides an iterative approach to the construction of oligoaryl scaffolds with precise positioning of peripheral groups. The method establishes a simple synthetic route to biaryl structures underrepresented in modern medicinal compound libraries. Compared to the currently dominant para- and meta-centered patterns, the ortho-ortho substitution pattern produced by sulfone ring remodeling is associated with increased three-dimensionality and a more uniform distribution in the molecular shape space. The reaction exhibits high regioselectivity with a wide range of substrates and nucleophiles. To facilitate prediction of the regiochemistry of biaryl products for structurally complex cyclic sulfones, a machine learning model based on the Bernoulli Naïve Bayes classifier was trained using computationally derived regioselectivities and descriptors and validated on an experimental set of diverse substrates. Taken together, the integrated experimental, computational, and data-driven approach provides a blueprint for exploring the underpopulated chemical space with potential applications in medicinal chemistry, materials science, and catalysis.

Methods

To a 10 mL screw capped test-tube equipped with a magnetic stirbar were added the sulfone (0.1 mmol) and the nucleophile (0.11 mmol). Potassium hexamethyldisilazide (KHMDS) (0.17 mmol) and degassed anhydrous 2-methyltetrahydrofuran (MeTHF) (0.4 mL) were then added, and the reaction mixture was allowed to stir at the specified temperature for 12 h. Water (100 µL, 5.56 mmol) was added, and the resulting mixture was concentrated under reduced pressure and dried by azeotropic removal of water with acetonitrile or benzene (3 × 0.5 mL). Palladium acetate (1.1 mg, 0.005 mmol, 5 mol%) and cyclohexyldiphenylphosphine (3.0 mg, 0.011 mmol, 11 mol%) were then added, followed by MeTHF (1.0 mL), and the reaction mixture was allowed to stir for 10 min. Dry potassium carbonate (0.3 mmol, 3 equiv.) and aryl halide/iodide/triflate (0.11 mmol) were then added. The reaction tube was sealed with a plastic cap, and the reaction mixture was stirred at 120 °C for 12 h. A saturated aqueous solution of disodium EDTA (2 mL) was added. The organic layer was separated, and the aqueous layer was extracted with ethyl acetate (3 × 2 mL). The combined organic layers were dried over sodium sulfate and concentrated under reduced pressure. The crude mixture was purified by column chromatography on silica gel with a mixture of hexane and ethyl acetate as an eluent to give the o,o-disubstituted oligoaryl coupling product.

Author contributions

O.V.L. conceived and directed the project. V.D.N., S.N., A.P., W.T.T., S.K.D., D.P.M., J.L.-B., C.K.G., and S.G.G. carried out experimental studies. R.T., G.C.H., S.O.F., W.B.H, and P.M.L. performed computational and machine learning studies. H.D.A. performed the X-ray crystallographic analysis. All authors contributed to the manuscript preparation, as well as the analysis and interpretation of the results.

Peer review

Peer review information

Nature Communications thanks Tiezheng jia, and the other, anonymous, reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

O.V.L. discloses support for the research of this work from the National Institutes of Health (GM152123).

Data availability

Experimental procedures, characterization data, computational details, all other data available within the Supplementary Information, and from the corresponding author upon request. The coordinates of the optimized structures are available from the source data. Source data are provided with this paper. The crystallographic data generated in this study have been deposited in the Cambridge Crystallographic Data Centre under accession codes 2259571(S41), 2259553(4b), 2259569(4f), 2474684 (4n), 2259562 (4o), 2259566 (4q), 2259565 (4r), 2259567 (4t), 2259552 (4u), 2259558 (5b), 2259549 (5c), 2474685 (5f), 2259545 (5g), 2259554 (5j), 2259557 (5t), 2259550 (6a), 2259561 (6e), 2259560 (6h), 2259551 (6i), 2259563 (6j), 2259564 (6n), 2260714 (6p), 2259555 (6s), 2268581 (6u), 2266515 (6w), 2259568 (7e), 2260713 (7l), 2262464 (11), 2474020 (13, 2474018 (15), 2474016 (Pd-1).

These data can be obtained free of charge from The Cambridge Crystallographic Data Centre via www.ccdc.cam.ac.uk/data_request/cif. Source data are provided with this paper.

Code availability

The Python code for model training is available on GitHub at (https://github.com/OLGroup-comp/site-selectivity-of-sulfone) or Zenodo at (10.5281/zenodo.17419918).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74669-7.