Parallels between the chloro and methoxy groups for potency optimization

Abstract

Small substituents such as chloro (Cl), fluoro, methyl, and methoxy (OCH₃) are often used in drug discovery to optimize ligand–protein interactions. Even though Cl is an electron-withdrawing group and OCH₃ is an electron-donating group exerting opposite effects on an aromatic ring, Cl and OCH₃ also display similarities in that they both exhibit dual electrostatic behavior. In a C–Cl bond, Cl is electronegative and adopts negative electrostatic potential, but at the same time, it has a σ-hole that is depleted of electron density and has an area of positive electrostatic potential. Similarly, the oxygen atom of OCH₃ displays negative electrostatic potential, but inductive electron-withdrawing effects bestow positive electrostatic potential at the terminal methyl group. This dual nature allows a versatile interaction with partially positive (δ⁺) and partially negative (δ⁻) regions of a protein pocket. In this study, four main types of intermolecular interactions are discussed from the vantage point of Cl and OCH₃ substituents: 1) hydrogen bonding, 2) orthogonal multipolar interactions, 3) halogen bonding and CH–O hydrogen bonding, and 4) Cl–π bonding and CH–π bonding. A comprehensive search of the PDB and analysis of X-ray co-crystal structures for each type of interaction unveiled parallels between Cl and OCH₃ in the manner in which these substituents engage with amino acid residues. The opposing electronic effects of Cl and OCH₃ substituents on an aromatic ring, along with the dual electrostatic versatility of these two groups, render them useful scouts to probe protein pockets for potency optimization.

There are parallels between Cl and OCH₃ in the types of ligand–protein interactions they engage in. The dual electrostatic versatility of these two substituents renders them useful scouts to probe protein pockets for potency optimization.

1. Introduction

In drug discovery, medicinal chemists often elucidate the pharmacophore by studying the structure–activity relationship (SAR) of a small-molecule ligand that binds to a particular protein target. This is often achieved by adding or removing several atoms from the small-molecule ligand and monitoring the effects that those structural changes have on bioactivity. The most common substituents have been tabulated and classified according to stereoelectronic and lipophilicity parameters, most notably by Louis Hammett and Corwin Hansch.^1–4 These pioneers in physical organic chemistry have important parameters named after them, notably the Hammett parameter σ^1,2 and the Hansch lipophilicity parameter π.^3,4 Hammett and Hansch parameters were essential in establishing the concept of quantitative structure–activity relationship (QSAR), which estimates the bioactivity of a molecule by numerically assessing its structural components.^5,6

The Hammett parameter σ is an electronic determinant^1,2 that quantifies the amount of electron withdrawal or electron donation of a substituent on an aromatic ring (Fig. 1A). Very strong electron-withdrawing groups (EWGs) such as nitro (σ_para = +0.78 and σ_meta = +0.71) and very strong electron-donating groups (EDGs) such as dimethylamino (σ_para = −0.83 and σ_meta = −0.15) are often avoided in medicinal chemistry exploration due to concerns in toxicity.⁷ Therefore, the most commonly used substituents are milder electronic modifiers that occupy the middle range of the table: chloro (Cl)⁸ and fluoro (F)^9–12 as EWGs, and methyl (CH₃)^13–15 and methoxy (OCH₃, or CH₃O)^16,17 as EDGs.

Fig. 1. A) Selected examples of Hammett parameter σ (σ_para for para substituents and σ_meta for meta substituents)^1,2 and Hansch parameter π for substituents on an aromatic ring.^3,4 B) Topliss' batchwise approach²⁰ suggesting the first compounds to synthesize when exploring initial SAR on a phenyl ring. C) Electrostatic potential maps showing the negative (in red) and positive (in blue) nature of Cl and OCH₃ substituents, generated using the B3LYP/6-31G** basis set.

The Hansch parameter π can be described as a lipophilicity determinant^3,4 derived from the octanol–water partition coefficient, log P (Fig. 1A).¹⁸ Very lipophilic groups with large positive values of π such as adamantyl (π = Δ(log P) = +3.30) or trimethylsilyl (π = +2.59), and very hydrophilic groups with large negative values of π such as trimethylammonium (π = −5.96) or sulfate (π = −4.76) are usually avoided because these substituents could incur dramatically large changes in the bioactivity of the parent molecule, oftentimes leading to inactive compounds. Therefore, the most commonly used substituents are milder log P modifiers that occupy the middle range of the table: Cl, CH₃, F, and OCH₃.

Although SAR exploration is conducted in a seemingly subjective manner that often relies on an individual medicinal chemist's skillset, knowledge, and experience, there are literature guidelines on how to develop SAR around a small-molecule scaffold. One of the most known SAR flowcharts was developed approximately 50 years ago by John Topliss, whose contributions include the “Topliss tree”¹⁹ as well as the “Topliss batchwise approach”²⁰ (Fig. 1B). According to Topliss, the first substituents to consider in an SAR exploration around a phenyl ring are the above-mentioned Cl, CH₃, and OCH₃ groups. Although the F atom is rarely mentioned in the Topliss tree, this is likely due to the low commercial availability of fluorine-containing building blocks in the 1970s, and the flowchart was strongly influenced by synthetic feasibility at the time.^19,20 As such, the importance of the F atom as a key substituent for medicinal chemistry has only been described more recently, within the past two decades.^9–12

Using benzene as a core scaffold, electrostatic potential maps of these substituents allow for a side-by-side comparison (Fig. 1C). On top of Cl and F being electronically similar atoms that are isosteric in the periodic table, they are similar in the manner in which they render the center of the ring less electron-rich compared to unsubstituted benzene. Furthermore, these substituents attract electron density to themselves, and adopt a negative electrostatic potential (shown in red color). Although Cl displays negative electrostatic potential, it can also exhibit a positive electrostatic potential at the terminus of the C–Cl axis that is called a “σ-hole”,^21,22 resulting in dual, opposing, electrostatic behavior on the same atom.⁸ This is in contrast with the F atom, which only displays negative electrostatic potential. The other two substituents, OCH₃ and CH₃, are similar because the former contains the latter from a structural point of view. These substituents are also similar in the manner in which they render the center of the ring more electron-rich compared to unsubstituted benzene. Furthermore, both these substituents adopt positive electrostatic potential (shown in blue color). Although OCH₃ has a component of positive electrostatic potential at its terminal CH₃ position, it also exhibits negative electrostatic potential at its O atom, resulting in dual electrostatic behavior.^16,17 This is in contrast with the CH₃ substituent, which only displays positive electrostatic potential.

For both the Cl and OCH₃ substituents, these dual electrostatic potentials allow for versatile interactions with the partially positive (δ⁺) and partially negative (δ⁻) regions of a protein binding pocket.^23,24 Although there is a fundamental difference between the Cl and OCH₃ substituents in that Cl is a para-EWG (i.e., σ_para > 0) and OCH₃ is a para-EDG (i.e., σ_para < 0), parallels could be drawn between the ways in which these substituents engage a protein pocket. Four types of molecular interactions²³ are discussed to illustrate the similarities in the binding of a Cl-containing ligand^8,24 and an OCH₃-containing ligand^16,17 to its protein target (Fig. 2):

Fig. 2. Parallels between Cl and OCH₃ substituents in ligand–protein intermolecular interactions.

1) hydrogen bonding²⁵ (QH⋯Cl and QH⋯OCH₃), in which the δ⁻ region of Cl and OCH₃ interacts with a δ⁺ H atom bonded to a heteroatom (Q);

2) orthogonal multipolar interactions²⁶ (Cl⋯C O and CH₃O⋯C O), in which the δ⁻ region of Cl and OCH₃ interacts with the δ⁺ carbon (C) atom of a carbonyl (C O);

3) halogen bonding (Cl⋯O C)^27–31 and CH–O hydrogen bonding (OCH₃⋯O C),³² in which the δ⁺ region of Cl and OCH₃ interacts with the δ⁻ oxygen (O) of a carbonyl (O C), or occasionally with the δ⁻ sulfur (S) atom of a Met side chain;

4) Cl–π bonding (Cl⋯π)^33–35 and CH–π bonding (OCH₃⋯π),³⁶ in which the δ⁺ region of Cl and OCH₃ interacts with the δ⁻ electron cloud of an aromatic ring.

Other directional interactions such as quadrupolar interactions³⁷ (Cl⋯CH_edge or CH₃O⋯CH_edge¹⁶) and π–π interactions^38,39 were not included in this discussion because the binding energies contributed directly by the Cl⁴⁰ or OCH₃ ⁴¹ substituent were deemed small (<0.3 kcal mol⁻¹); nondirectional interactions such as van der Waals interactions were also excluded from this study.^23,42

Although F and CH₃ can only participate in a subset of these interactions (interactions 1 and 2 for F, and interactions 3 and 4 for CH₃), Cl and OCH₃ can engage in all four interactions and are therefore more suitable as “scouts” for protein pocket finding.^8,16,17 For both Cl and OCH₃, the four types of intermolecular interactions²³ are discussed herein through an analysis of X-ray co-crystal structures of ligand–protein interactions.⁴² Each type of intermolecular interaction illustrates the parallels between the Cl and OCH₃ groups, which is the focus of the current study. The versatile electrostatic nature of Cl and OCH₃ is a desirable feature in exploratory SAR for potency optimization campaigns.

2. Methods

Ligand–protein interactions in the Research Collaboratory for Structural Bioinformatics (RCSB) Protein Data Bank (PDB)⁴³ were analyzed (among 241 922 experimental PDB structures registered as of Sep 21, 2025) using PSILO (version 2024.0602), which is a database and search tool by the Chemical Computing Group.⁴⁴ The structural resolution limit was set at < 3.00 Å^29,31 because the RCSB considers resolutions of 3 Å or greater to be of low resolution,⁴⁵ and only high- or medium-resolution structures were desired for this analysis. The obtained PDB files were analyzed without protein or ligand preparation. Most of the PDB searches were set at an upper limit of 4.00 Å for the interatomic distance (measured from heavy atom to heavy atom)⁴² because this number is large enough to encompass the various sums of the atomic van der Waals radii^46,47 and allows for the inclusion of weaker intermolecular forces,^24,48,49 yet is small enough to minimize the inclusion of van der Waals interactions and randomly distributed non-attractive interactions.^50–52 As an exception, the bond distances for interactions with aromatic π clouds (Cl⋯π and OCH₃⋯π) were set at an upper limit of 4.50 Å.³⁴ Bond angles used in the PSILO search were restricted with recommended literature values (often with a more inclusive limit or a more restrictive limit), and these parameters are described in each section. Explicit hydrogen atoms were not used in the PSILO search, and therefore OCH₃-containing small-molecule ligands were identified by searching for all ether (and ester) C–O–C bond-containing molecules, and filtering at a later stage for OCH₃-containing molecules only. Solvent (e.g., dichloromethane, dichloroethane, ethylene glycol monomethyl ether) and solvent-like molecules (e.g., naturally abundant cofactors like coenzyme Q) were removed during the filtering stage because the high concentrations of these molecules used to prepare the ligand–protein co-crystal might have led to nonspecific interactions that skew the current analysis. Data entries were further curated by visual inspection. The data were filtered using Vortex (version 6.1.1136-s), a chemical data analysis program by Dotmatics.⁵³ PDB files were rendered using Maestro (version 13.8.132 release 2023-4), a structural visualization program by Schrödinger.⁵⁴ Electrostatic potential maps were generated using Jaguar (version 12.2), a quantum chemistry software by Schrödinger.⁵⁵ Chemical structures were illustrated using ChemDraw (version 22.2.0), a chemical structure drawing software by Revvity Signals.⁵⁶

3. Discussion

3.1. Hydrogen bond (QH⋯Cl and QH⋯OCH₃)

Optimizing the binding of a small-molecule ligand to a protein pocket relies on attractive molecular interactions.²³ To design a molecule that selectively interacts with a protein of interest, directionally sensitive interactions such as hydrogen bonding are required.^23,25,57,58 Hydrogen bonding is the most frequently encountered form of directionally sensitive, noncovalent interactions in the PDB.⁴² Although many types of hydrogen bonding are essential in ligand–protein interfaces, there is a wide range in the strength of the hydrogen bond:⁵⁹ for example, for Q = N or O, QH⋯O C bonds are often considered “strong” hydrogen bonds, whereas both QH⋯Cl and QH⋯OCH₃ are generally considered “weak” hydrogen bonds.

Hydrogen bonding is an attractive interaction between two electronegative atoms with a hydrogen (H) covalently bonded to one of the two atoms.⁶⁰ Since the electronegativity value of Cl is between that of N and O,^61,62 Cl-containing molecules might be expected to participate in hydrogen bonding interactions as strongly as N- and O-containing molecules. However, among the various Cl-containing species, the C–Cl group is the weakest hydrogen bond acceptor (HBA) based on measured QH⋯Cl bond lengths: Cl⁻ > metal–Cl ≫ C–Cl.⁶³ In a theoretical, isolated system of one molecule of chlorobenzene interacting with a compound that approximates an amino acid's hydrogen bond donor (HBD), the peak interaction energy is between −2 and −4 kcal mol⁻¹.⁶⁴ These values are rather similar to the interaction energy of chlorobenzene with water of −2 kcal mol^−1 64 (independently calculated as −1.72 kcal mol^−1 65), which is an important consideration because a hydrogen bond with water must first be broken in a desolvation event before ligand–protein binding can occur. As such, QH⋯Cl–C attractive forces in biological systems are known to be weak;^66,67 however, these still make a favorable contribution to ligand–protein binding and should actively be considered during molecular design.⁶⁴ Much like the Cl substituent, an OCH₃ group on a small-molecule ligand could act as an HBA to complement the HBD present inside a protein pocket. Although hydrogen bonding is often considered strong for O-containing HBA, ethers are considered weaker acceptors than carbonyl compounds.⁶⁸ For aromatic ethers such as anisole (methoxybenzene), not only is hydrogen bonding to OH and NH infrequently observed,⁶⁹ but its HBA strength is also weaker than aliphatic ethers.⁷⁰ Comparison of HBA strengths is often accomplished with a parameter called hydrogen bond basicity (pK_BHX),^71–75 which shows that anisole (pK_BHX = 0.09)⁷¹ is a much worse HBA than aliphatic ethers (pK_BHX = 1.01 for diethyl ether).^72,75 An exception is the HBA strength of 1,2-dimethoxybenzene (pK_BHX = 1.16),⁷² where there is cooperativity of the two O atoms engaging one HBD. Regarding hydrogen bonding energy, gas-phase interactions between dimethyl ether as the HBA and N-methylacetamide as the HBD were calculated to be −6.2 kcal mol⁻¹; when anisole was the HBA, hydrogen bonding interactions were typically 1 kcal mol⁻¹ smaller in magnitude and was approximately −5 kcal mol⁻¹.⁵⁸

A Cl atom contains a “σ-hole” with δ⁺ character at the terminus of the C–Cl axis and a “negative belt” with δ⁻ character in a plane that is orthogonal to the C–Cl axis (see Fig. 1C).^21,22 Since angle θ = ∠(C–Cl⋯Q) = 165–180° defines the σ-hole and leads to halogen bonding interactions,^24,27 all bond angles θ ≤ 165° defining the negative belt were included for hydrogen bonding (Fig. 3A). Although θ ≤ 165° is a criterion that allows for a more inclusive dataset (resulting in 5338 examples), a more restrictive dataset using θ ≤ 140° as a criterion would ensure that the hydrogen bonding is not obscured by a component of halogen bonding^24,28,64 (this would lead to a slightly smaller set of 4460 examples). A literature study of the Cambridge Structural Database (CSD) has shown that QH⋯Cl hydrogen bonding is most frequently observed between 90° and 130°,^24,63 and a previous analysis of the PDB has shown that this hydrogen bonding is most probable between 80° and 110°.⁶⁴ In our analysis, a histogram displaying the range of hydrogen bond angles peaks between θ = 90° and 120°, matching the trends in the literature (Fig. 3B, green bars). Regarding the hydrogen bond length d (measured between heavy atoms Q⋯Cl), the search in the PDB was set with a more inclusive upper limit of d = 4.00 Å for the interatomic distance, regardless of the nature of Q being N, O, or S (resulting in 5338 examples).⁶⁴ For a more restrictive upper limit of d(Q⋯Cl), the sum of the van der Waals radii (Σ_vdW)^46,47 is often used in the literature:^{23,57,59,63,68} the Σ_vdW value for N⋯Cl is (1.55 Å + 1.75 Å =) 3.30 Å, O⋯Cl is (1.52 Å + 1.75 Å =) 3.27 Å, and S⋯Cl is (1.80 Å + 1.75 Å =) 3.55 Å. However, using these restrictive upper limits for distance would lead to a much smaller set for analysis (714 examples). A histogram showing the range of hydrogen bond lengths displays a first peak around 3.55–3.70 Å, and then increases in frequency as random interactions (including van der Waals interactions) start to populate above 3.80 Å (Fig. 3B, green bars). Although a more restrictive dataset using d ≤ Σ_vdW as a criterion could potentially identify examples with energetically stronger hydrogen bonds, a large percentage of examples (87%) would be deleted if this restriction were applied. Simultaneously applying all the restrictive limits for bond angle and distance would significantly reduce the number of examples from 5353 to 510 (Fig. 3B, orange bars), which would render analysis of trends difficult.

Fig. 3. Hydrogen bonding (QH⋯Cl). A) General schematic of the molecular interaction, along with the parameters set in the PDB search for the indicated bond angle and distance. B) Histograms depicting the frequency of the observed angle and bond distance (green bars = inclusive set with parameters shown in part A; orange bars = restrictive set with additional restriction parameters shown in part B). C) Frequency of various HBDs engaging with Cl-containing HBA. D) A matched molecular pair (MMP) of compounds 1 and 2 showing the effect of the Cl substituent on potency, and E) a closely related molecule 3 engaging in QH⋯Cl hydrogen bonding with the target protein YTHDC1 (X-ray co-crystal PDB 8Q2U).⁷⁶.

All the QH⋯Cl hydrogen bonding interactions in the PDB that satisfy the more inclusive criteria for bond angle (θ = 0–165°) and distance (d = 2.00–4.00 Å) were classified according to the nature of the HBD (amide backbone NH or side chain functional group; Fig. 3C and SI). The largest occurrence of QH⋯Cl hydrogen bonding in this study is the amide backbone NH (2354 examples), which is the most frequent protein HBD in the literature,²⁴ due to the abundant presence of one NH motif per amino acid residue. The Tyr side chain OH (550 examples), Ser side chain OH (446 examples), and Asn side chain NH₂ (349 examples) were also found to be common HBDs for the C–Cl HBA. In an example of Asn binding, the fragment molecule 1 was rendered 8.3-fold more potent upon chlorination to give 2 (Fig. 3D). The purine Cl atom of closely related analog 3 was found to engage the YTHDC1 protein's Asn364 and Asn367 side chain NH₂ groups with short hydrogen bonding interactions (d ≤ 3.30 Å; PDB 8Q2U; Fig. 3E).⁷⁶ Within the more inclusive distance limit (d ≤ 4.00 Å), additional hydrogen bonding interactions with the Ser362 side chain OH and with the Asn363 backbone NH were observed in this X-ray co-crystal structure.

For aliphatic ethers, the hydrogen bonding interaction mostly occurs in the plane of the lone pairs, but otherwise, there is little additional restriction of vector or directionality.^23,25,68 Regarding the hydrogen bond directionality of aromatic OCH₃ groups, an HBD that is located in the plane of the aromatic ring to engage the O atom is 2 kcal mol⁻¹ more favorable than an HBD placed above or below the ring, and the optimal hydrogen bonding angle bifurcates the aryl–O–C bond.⁷⁰ In our analysis of X-ray crystal structures in the PDB, the bond angle θ = ∠(C–O⋯Q) of the ligand–protein interaction was recorded, but not constrained (i.e., 0° ≤ θ ≤ 180°; Fig. 4A). Upon analysis of bond angles, a frequency peak was observed at θ = 120° (Fig. 4B, blue bars), which is reasonable given that most examples involve aromatic OCH₃ groups, and trigonal planar geometry approximates bond angles of 120°. Regarding the hydrogen bond length, although the search was set with a more inclusive criterion of d(Q⋯O) ≤ 4.00 Å (resulting in 2343 examples), the more restrictive criterion of d ≤ Σ_vdW(Q⋯O) for N⋯O (1.55 Å + 1.52 Å = 3.07 Å), O⋯O (1.52 Å + 1.52 Å = 3.04 Å), and S⋯O (1.80 Å + 1.52 Å = 3.32 Å) could be used. However, using these restrictive upper limits for distance would lead to a smaller set for analysis (504 examples). A histogram showing the range of hydrogen bond distances displays a first peak around 3.00–3.20 Å, and then increases in frequency especially above 3.80 Å (Fig. 4B, blue bars). Since OCH₃ is also a weak HBA, the upper limits set by Σ_vdW might be too restrictive (Fig. 4B, orange bars), leading to a deletion of a large percentage of examples (78%) if this restriction were applied.

Fig. 4. Hydrogen bonding (QH⋯OCH₃). A) General schematic of the molecular interaction, along with the parameters set in the PDB search for the indicated bond angle and distance. B) Histograms depicting the frequency of the observed angle and bond distance (blue bars = inclusive set with parameters shown in part A; orange bars = restrictive set with additional restriction parameters shown in part B). C) Frequency of various HBDs engaging with the OCH₃-containing HBA. D) An MMP of compounds 4 and 5 that shows the effect of the OCH₃ substituent on potency, and E) a closely related molecule, infigratinib (6), which engages in QH⋯OCH₃ hydrogen bonding with the target protein FGFR1 (X-ray co-crystal PDB 3TT0).⁷⁷.

All the QH⋯OCH₃ hydrogen bonding interactions in the PDB with θ = 0–180° and d = 2.00–4.00 Å were tabulated (Fig. 4C and SI). This showed that the amide backbone NH serves most frequently as the HBD (597 examples), followed by the Gln side chain NH₂ (266 examples), Lys side chain NH₃⁺ (240 examples), and the Ser side chain OH (236 examples). An example of the effect of the OCH₃ group is shown with fibroblast growth factor receptor (FGFR) inhibitors, wherein an MMP comparison of 4 and 5 shows that addition of an OCH₃ group exerts a 13-fold improvement in potency (Fig. 4D).⁷⁷ The enhanced kinase inhibitory ability bestowed by this OCH₃ group led to the discovery of infigratinib (6), a drug developed to treat cholangiocarcinoma. In the X-ray co-crystal structure of 6 in FGFR1 (PDB 3TT0), one of the OCH₃ groups engages in a hydrogen bond with the amide backbone NH of Asp641 (Fig. 4E).⁷⁷ This hydrogen bond has a length (d = 3.21 Å) that nears the Σ_vdW value (3.07 Å), and the backbone carboxamide HBD is coplanar with the OCH₃ HBA. This 3,5-dimethoxybenzene unit is a privileged motif in FGFR inhibitors, since approved drugs such as erdafitinib (PDB 5EW8), pemigatinib (PDB 7WCL), and futibatinib (PDB 6MZW) all contain the same motif, and their binding modes all involve an OCH₃ hydrogen bond to the amide backbone of Asp641.⁷⁸

As such, when comparing the hydrogen bond of Cl (QH⋯Cl) and OCH₃ (QH⋯OCH₃), several parallels could be observed: 1) the intermolecular interaction energies are similar, typically ranging from −0.5 to −5 kcal mol⁻¹, but a desolvation penalty of 2 kcal mol⁻¹ must be taken into account; 2) the histograms for the bond angles are similar, peaking in frequency between 90–120°; 3) the histograms for the bond distances are similar, with a first frequency peak that ends at 3.65 Å, and with an increase in frequency above 3.80 Å; 4) frequent ligand–protein interactions occur with the amide backbone NH, as well as with the Ser side chain OH group.

3.2. Orthogonal multipolar interactions (Cl⋯C O and CH₃O⋯C O)

Interaction between the δ⁻ region of a heteroatom and the δ⁺ region of a carbonyl group was first coined by Diederich as “orthogonal multipolar interactions”.²⁶ This term describes the ∠(heteroatom⋯C O) angle tending toward 90° (i.e., orthogonal), and the interactions between a pair of dipoles (i.e., four poles, or multipolar) that are often observed in ligand–protein binding. Although this interaction was first identified for a fluorine-containing ligand,^79–81 the number of occurrences of Cl⋯C O multipolar interactions was greater than for F⋯C O in a search of the CSD.²⁶ Although the attractive interaction energy for the Cl⋯C O interaction has not been quantified, the corresponding energy for fluorine's F⋯C O interaction has been estimated at −0.2 to −0.4 kcal mol⁻¹,^23,26,80 and the energy for a carbonyl engaging in a C O⋯C O interaction has been estimated at −1.9 kcal mol^−1 26 (−0.65 kcal mol⁻¹ for amide carbonyl groups).⁸² Similarly to the Cl substituent, an OCH₃ group could act as the δ⁻ component to complement the δ⁺ region of a protein backbone carbonyl's carbon atom. Regarding the nature of the O atom, although an ether O is usually a weaker electron pair donor than a carbonyl O for hydrogen bonding, the number of occurrences of an ether engaging in multipolar interactions (O⋯C O) and the number of occurrences of a carbonyl engaging in multipolar interactions (C O⋯C O) was found to be the same within a distance limit of 3.25 Å in the CSD.²⁶

Since the δ⁻ region of a Cl atom is strongest in the anionic belt that is orthogonal to the C–Cl axis, the Cl atom of a small-molecule ligand should approach the δ⁺ carbon atom as a “side-on interaction” above the plane of the carbonyl group.^23,24 Since multipolar interactions involve an electrostatic attraction between two dipoles (C–Cl bond and C O bond), there are two bond angles to consider (Fig. 5A): θ¹ = ∠(C–Cl⋯C), which is centered on the geometry around the C–Cl bond, and θ² = ∠(Cl⋯C O), which is centered on the geometry around the C O group. For θ¹, similarly to the hydrogen bonding section, the criterion of θ¹ ≤ 165° was applied (resulting in 2991 examples), but a more restrictive criterion of θ¹ ≤ 140° could be applied to exclude the possibility of halogen bonding confounding the search (this would lead to 2337 examples).⁴² For θ², although the exact nature of this interaction (dipole–dipole, electrostatic, or n → π* interactions) has been debated,⁸³ the most commonly reported optimal bond angle is 90°, as opposed to a more obtuse Bürgi–Dunitz angle of ≈110°.⁸⁴ In our search of the PDB, the θ² range was not constrained in order to allow for a larger inclusion of data (2991 examples), as opposed to a more restricted range of θ² = 70–110° ^26,42 (this would lead to 1725 examples). In slight deviation to the optimal angle of θ² being 90°, our analysis has revealed a peak of 65–85° as the most common θ² angle (Fig. 5B, green bars). Regarding the bond distance, although the search was set with a more inclusive criterion of d(Cl⋯C) ≤ 4.00 Å (2991 examples),⁸³ the more restrictive limit of Σ_vdW(Cl⋯C) = 3.45 Å could potentially be used⁸³ (this would lead to only 502 examples). Although the restrictive limits for bond angles could be acceptable, setting the bond distance criterion of d ≤ 3.45 Å would be impractical, because even the first frequency peak of 3.45–3.55 Å would be excluded, and most examples (83%) would be discarded (Fig. 5B). Simultaneously applying all the restrictive limits for bond angle and distance would significantly reduce the number of examples from 2991 to 307 (Fig. 5B, orange bars), which would render analysis of trends difficult.

Fig. 5. Orthogonal multipolar interactions (Cl⋯C O). A) General schematic of the molecular interaction, along with the parameters set in the PDB search for the indicated bond angle and distance. B) Histograms depicting the frequency of the observed angle and bond distance (green bars = inclusive set with parameters shown in part A; orange bars = restrictive set with additional restriction parameters shown in part B). C) Frequency of various carbonyl groups engaging with the Cl-containing ligand. D) A matched series 7–10 showing the effect of the Cl substituent on potency, and E) a representative molecule 8 engaging in Cl⋯C O interactions with the target protein TNKS (X-ray co-crystal PDB 6A84).⁸⁵.

All the Cl⋯C O multipolar interactions in the PDB that satisfy the more inclusive criteria for bond angles (θ¹ = 0–165° and θ² = 0–180°) and distance (d = 2.00–4.00 Å) were tabulated (Fig. 5C and SI). Among the carbonyl groups surveyed, the amide backbone C O was found to be the most likely to engage in multipolar orthogonal interactions (2328 examples), as opposed to side chain C O groups, owing to the difference in their abundance within a protein. An example of such an interaction is illustrated for tankyrase (TNKS) inhibitors 7–10 (Fig. 5D), for which the 3- to 5-fold potency enhancement arising from the Cl substituent could be partly ascribed to the ideal bond angles (θ¹ = 126.0° and θ² = 105.8°) and to the short 3.04 Å engagement of the Cl atom of 8 to the protein backbone's C O (PDB 6A84; Fig. 5E).⁸⁵

Much like for Cl multipolar interactions, two bond angles were examined for OCH₃ (Fig. 6A): θ¹ = ∠(C–O⋯C) and θ² = ∠(O⋯C O). For θ¹, similarly to the OCH₃ hydrogen bonding section, the bond angle was recorded, but not constrained; a peak of 100–115° was observed as the most common bond angle, with a second peak at 120–130° (Fig. 6B, blue bars). For θ², the range was also not specified in order to allow for a larger inclusion of data (resulting in 1108 examples), as opposed to the oft-used restricted range of θ² = 70–110° for multipolar interactions^26,42 (this would lead to 593 examples). Our analysis of the θ² angle has revealed a peak of 65–85° as the most common bond angle. Regarding the bond distance, although the search was set with a more inclusive criterion of d(O⋯C) ≤ 4.00 Å (1108 examples), the more restrictive limit of Σ_vdW(O⋯C) = 3.22 Å could potentially be used (this would lead to only 84 examples). Much like in the histograms for Cl multipolar interactions, the histograms for OCH₃ multipolar interactions show that the restrictive limits for bond angles could be acceptable, but setting the restrictive criterion of d ≤ 3.22 Å would be impractical, because even the first frequency peak of 3.30–3.45 Å would be excluded from the study and most examples (92%) would be filtered out (Fig. 6B). Simultaneously applying all the restrictive limits for bond angle and distance would almost entirely eliminate the number of examples (from 1108 to 70; Fig. 6B, orange bars), which would be impractical for the analysis of trends.

Fig. 6. Orthogonal multipolar interactions (CH₃O⋯C O). A) General schematic of the molecular interaction, along with the parameters set in the PDB search for the indicated bond angle and distance. B) Histograms depicting the frequency of the observed angle and bond distance (blue bars = inclusive set with parameters shown in part A; orange bars = restrictive set with additional restriction parameters shown in part B). C) Frequency of various carbonyl groups engaging with the OCH₃-containing ligand. D) An MMP of compounds 11 and 12 that shows the effect of the OCH₃ substituent on potency, and E) PF-00835231 (12) that engages in CH₃O⋯C O interactions with the target protein SARS-CoV-1 (X-ray co-crystal PDB 6XHL).⁸⁶.

All the CH₃O⋯C O multipolar interactions in the PDB that satisfy the more inclusive criteria for bond angle (θ¹ = 0–180° and θ² = 0–180°) and distance (d = 2.00–4.00 Å) were tabulated (Fig. 6C and SI). Among the carbonyl groups surveyed, the amide backbone C O is most likely to engage in orthogonal multipolar interactions (676 examples). An example of orthogonal multipolar interaction using an OCH₃ group was described in a historically important compound, PF-00835231 (12; Fig. 6D), which covalently binds to the 3C-like cysteine protease (3CL^pro) domain of severe acute respiratory syndrome coronavirus 1 (SARS-CoV-1).⁸⁶ Originally, Pfizer had developed PF-00835231 (12) during the SARS outbreak of 2002–2004 that was caused by SARS-CoV-1.⁸⁷ Importantly, the 3CL^pro domain is conserved between SARS-CoV-1 and SARS-CoV-2, and therefore 12 was used as a critical starting point in the discovery of PF-07321332 (nirmatrelvir; a component of the antiviral drug Paxlovid®), used against the SARS-CoV-2 virus that is responsible for the COVID-19 pandemic.^87,88 The OCH₃ group of PF-00835231 (12) effected a 9.5-fold improvement in antiviral potency compared to the desmethoxy analog 11. This potency enhancement could be partly attributed to the nearly ideal bond angles (θ¹ = 126.9° and θ² = 78.9°) and bond distance (d = 3.37 Å) between the OCH₃ group and the SARS-CoV-1 protein's Gln189 backbone carbonyl (PDB 6XHL; Fig. 6E).⁸⁶ Furthermore, in geometrically optimized multipolar interactions such as this one, the O–C bond of the methoxy group is parallel to the C O bond of the carbonyl to align the local dipole moments of the ligand and protein in “side-on” antiparallel fashion to benefit from further dipole–dipole interactions.¹⁶ Several X-ray co-crystal structures of 12 bound to SARS-CoV-2 were obtained as well,^86,89 demonstrating how the same interaction with Gln189 is present in both strands of the coronavirus protease domain.

As such, when comparing the orthogonal multipolar interactions of Cl (Cl⋯C O) and OCH₃ (CH₃O⋯C O), several parallels could be drawn: 1) the intermolecular interaction energies are similar, typically ranging from −0.1 to −1 kcal mol⁻¹; 2) the histograms for the bond angles are similar, in which θ¹ has two frequency peaks at 110° and at 125°, and θ² has a main frequency peak between 65–85°; 3) the histograms for the bond distances are similar, with the greatest frequency occurring at d > 3.70 Å, and therefore the Σ_vdW limits were deemed too restrictive; 4) most interactions involve the amide backbone CO, with a minor involvement from side chain carbonyl groups. These observations could potentially be useful when trying to optimize ligand–protein binding with orthogonal multipolar interactions in mind.

3.3. Halogen bonding (Cl⋯O C) and CH–O hydrogen bonding (OCH₃⋯O C)

Although halogen (X) atoms are electronegative and have overall δ⁻ character,^60–62 there is anisotropy of electron density in the heavier halogen atoms (X = Cl, Br, and I), wherein the region on the hind side of X along the C–X bond axis has a σ-hole with positive electrostatic potential (see Fig. 1C).^21,22 The interaction of this halogen's positive region with an available electron pair is called halogen bonding, which has been a topic of literature reviews both for small-molecule interactions^21,22 and for ligand–protein interactions.^27–31,64 In the context of proteins, this electron pair could be an amide backbone carbonyl's oxygen atom, or a side chain's lone pair of electrons (including carboxamides, carboxylate ions, or a sulfur atom).

The Cl atom in a C–Cl bond has dual electrostatic character^23,24,66,67 and can engage with a protein's amide backbone in many ways, including hydrogen bonding with the amide NH (vide supra), and halogen bonding with the amide O. For halogen bonding strength, the trend is F ≪ Cl < Br < I,^21,22,29 and although Cl can be a halogen bond donor, it is considered to be at the weaker end of the spectrum. Furthermore, halogen bond energy is generally weaker than hydrogen bond energy:^24,28 specifically for chlorobenzene, halogen bond energies of acetone binding was calculated to be between −1.3 to −2.1 kcal mol⁻¹,^22,28,90N-methylacetamide binding was calculated as −1.3 kcal mol⁻¹,^28,29,91,92 and acetamide binding was calculated as −0.7 kcal mol⁻¹;⁶⁴ in contrast, the hydrogen bond energy of binding chlorobenzene to the acetamide NH at the most optimal angle of interaction (≈70°) was calculated as −3.0 kcal mol⁻¹.⁶⁴ Of note, halogen bonding interactions of chlorobenzene with an anionic species such as acetate was calculated to be slightly stronger, at −2.5 kcal mol⁻¹,⁶⁴ demonstrating that Cl⋯O C interactions with amino acid side chains such as Asp and Glu are important. Despite the generally weaker energetics of halogen bonds compared to hydrogen bonds, halogen bonds have a lower cost of desolvation than hydrogen bonds, which is not taken into account in quantum mechanical calculations,²⁸ and this is advantageous when considering the overall free energy of the binding event.²³ As such, halogen bonding has been recognized as an important intermolecular interaction presented by a Cl-containing organic molecule,²⁴ and should be a routine element of medicinal chemistry design.²³ The equivalent of Cl halogen bonding (Cl⋯O C) for the OCH₃ group is CH–O hydrogen bonding (OCH₃⋯O C), where the δ⁺ region of the terminal methyl group in an OCH₃-containing ligand engages the δ⁻ portion of a protein backbone's carbonyl group. These CH–O interactions are classified as weak hydrogen bonds,^93–97 whose energy is approximated as −0.5 to −2.5 kcal mol⁻¹,^95,96 and is similar to that of a halogen bond.²³ Although CH–O hydrogen bonds are typically weaker than conventional hydrogen bonds, they occur frequently^42,59 and are penalized less by desolvation,²³ thereby contributing significantly in the context of intermolecular interactions.⁹⁷

Halogen bonding is a highly directional molecular interaction,^98,99 and therefore the bond angle of interaction is very important when searching the PDB. Two bond angles were examined for the halogen bond (Fig. 7A): θ¹ = ∠(C–Cl⋯O) and θ² = ∠(Cl⋯O C). For θ¹, its value was set at 140° as the more inclusive lower limit (θ¹ = 140–180°; resulting in 1545 examples),^{24,28–30,100–102} or potentially 165° as the more restrictive lower limit (θ¹ = 165–180°; this would lead to only 311 examples).^24,28 For θ², an inclusive range encompasses all angles (θ² = 0–180°; 1545 examples). However, angle ranges have been previously suggested as θ² = 90–150° (this would lead to 1117 examples),⁴² since the largest probability of electron density lies in the n electron orbitals in the plane of the carbonyl at 120°.^29,31 Histograms for the bond angles θ¹ and θ² were generated (Fig. 7B, green bars), and agree with literature findings regarding the most frequently observed angles for the halogen bond.²⁹ For θ¹, although the halogen bonding interaction might be the most energetically optimal at 175–180°, it is the least frequent bin because the surface area of the cone defined by this angle is the smallest. In contrast, the surface area of the cone defined by the angle θ¹ = 140–145° is larger, and the frequency of encountering this geometric arrangement is higher. Setting the restrictive criterion of θ¹ ≥ 165° would eliminate too many examples and would hamper the analysis. For θ², the frequency is already high for the 90–150° range, and therefore using this more restrictive range would not eliminate the majority of the observed examples.

Fig. 7. Halogen bonding (Cl⋯O C). A) General schematic of the molecular interaction, along with the parameters set in the PDB search for the indicated bond angles and distance. B) Histograms depicting the frequency of the observed angle and bond distance (green bars = inclusive set with parameters shown in part A; orange bars = restrictive set with additional restriction parameters shown in part B). C) Frequency of various carbonyl groups (or sulfur atom) engaging with the Cl-containing ligand. D) A matched series 13–17 showing the effect of the halogen substituent on potency. E) Cl-containing molecule (15) engaging in Cl⋯O C interactions with the target protein cathepsin L (X-ray co-crystal PDB 2YJC).¹⁰¹.

For bond distance, although the search was set with a more inclusive criterion of d(Cl⋯O) ≤ 4.00 Å (resulting in 1545 examples),^42,102 the more restrictive limit of Σ_vdW(Cl⋯O) = 3.27 Å^{29–31,100,103} could be used (this would lead to only 543 examples); furthermore, the “ideal” halogen bond distance has previously been suggested as 3.0 Å.²⁷ It is of note that calculated complex formation energies decrease to about 50% when the bond distance is increased by 1.0 Å, and therefore halogen bonding is highly sensitive to interatomic distance.²⁸ Histograms for d(Cl⋯O) have been generated in our analysis, showing a first peak at around 3.20 Å (Fig. 7B, green bars). A second peak was observed at 3.40–3.60 Å, and therefore, if the more restrictive criterion of d ≤ 3.27 Å were used, a large percentage of the examples (65%) would be filtered out. For Cl halogen bonding, simultaneously applying all the restrictive limits for bond angle and distance would almost entirely eliminate the number of examples (from 1545 to 109; Fig. 7B, orange bars), which would render the analysis of trends less meaningful.

All the Cl⋯O C halogen bonding interactions in the PDB that satisfy the more inclusive criteria for bond angle (θ¹ = 140–180° and θ² = 0–180°) and distance (d = 2.00–4.00 Å) were tabulated (Fig. 7C and SI). Although not a carbonyl O atom, the S atom of the Met side chain is also known to engage in halogen bonding (Cl⋯S),^28,92,104 and was therefore included in this PDB search as well (150 examples; Fig. 7A and C). Other heteroatom-containing amino acid side chains (e.g., Ser and Cys) were excluded because it is often unclear when the interaction consists of halogen bonding or hydrogen bonding due to the lack of protonation states in the PDB.^24,28 Among the carbonyl groups surveyed, the amide backbone oxygen atom is most likely to engage in halogen bonding (1315 examples), and many examples of Met engagement through Cl⋯S were observed as well (Fig. 7C).

When the introduction of a Cl atom onto a molecule leads to potency improvement, it could be difficult to identify the specific type of molecular interaction that contributes most to the potency change, since the Cl atom can potentially engage in many types of interactions. However, a case study by Diederich suggested halogen bonding to a backbone carbonyl's oxygen atom to be at the origin of the increase in ligand–protein binding affinity (Fig. 7D).¹⁰¹ Introduction of a Cl atom at the para position of the phenyl group in 13 led to 15, resulting in a 13-fold improvement in potency against human cathepsin L (h catL). This improvement in potency was attributed to halogen bonding for three reasons: 1) the para-F compound 14 led to slightly worse potency than 13, owing to electrostatic δ⁻–δ⁻ repulsions between the F and O atoms; 2) the size of the σ-hole and magnitude of the positive electrostatic potential is in the order of Cl < Br < I, which correlates well with the order of binding affinity of 15, 16, and 17; 3) X-ray co-crystal structures were obtained for 15 (PDB 2YJC), 16 (PDB 2YJ2), and 17 (PDB 2YJ8), all demonstrating halogen bonding geometries. For Cl compound 15, the halogen bond to the Gly61 carbonyl has nearly ideal parameters of θ¹ = 171.0°, θ² = 147.3°, and d = 3.02 Å, satisfying even the restrictive ranges for bond angles and distance (Fig. 7E). Even with multiple aspects of supporting evidence, Diederich had concluded for this example that it is not completely clear whether halogen bond enthalpy alone leads to the observed improvement in potency, since water displacement is also a possible hypothesis for this ligand–protein system.¹⁰¹

The bond angle of interaction is also important for CH–O hydrogen bonds, and therefore two bond angles were recorded (Fig. 8A): θ¹ = ∠(O–CH₃⋯O) and θ² = ∠(CH₃⋯O C). For θ¹, its value was not restricted (θ¹ = 0–180°). Even though there are recommended values for bond angles in CH–O hydrogen bonds, literature ranges were derived from small-molecule X-ray crystal structures from the CSD and thus the locations of the H atoms had been specified;^32,42,95 the locations of the H atoms are not often identified in ligand–protein co-crystal structures, and therefore the same bond angle restrictions could not be set in our analysis. For θ², the bond angle was also recorded, but not constrained, to allow for an inclusive range (θ² = 0–180°; resulting in 3894 examples). However, analogously to the halogen bond, the θ² angle range for the carbonyl portion could be set at θ² = 90–150° (this would lead to 3170 examples), to approximate the vector of the n electron orbitals in the plane of the carbonyl at ≈120°.⁹⁵ Histograms for the bond angles θ¹ and θ² have been generated in our analysis: for both θ¹ and θ², a symmetric and somewhat broad frequency distribution around a maximum peak of 115° was obtained (Fig. 8B, blue bars).

Fig. 8. CH–O hydrogen bonding (OCH₃⋯O C). A) General schematic of the molecular interaction, along with the parameters set in the PDB search for the indicated bond angles and distance. B) Histograms depicting the frequency of the observed angle and bond distance (blue bars = inclusive set with parameters shown in part A; orange bars = restrictive set with additional restriction parameters shown in part B). C) Frequency of various carbonyl groups (or sulfur atom) engaging with the OCH₃-containing ligand. D) A matched pair 18–19 showing the effect of the OCH₃ group on potency. E) OCH₃-containing molecule (19) engaging in CH–O hydrogen bonding with the target protein BACE1 (X-ray co-crystal PDB 2VA7).¹⁰⁶.

For bond distance, although the search was set with a more inclusive criterion of d(CH₃⋯O) ≤ 4.00 Å (resulting in 3894 examples), the more restrictive upper limit of Σ_vdW(C⋯O) = 3.22 Å could be used (however, this would lead to only 525 examples). Other upper limits such as 3.6 Å have been previously used,⁴² and CH–O hydrogen bonds were reported to have an experimental range of 3.0–3.8 Å.²³ Histograms for d(CH₃⋯O) have been generated in our analysis, showing that the frequency of these interactions peaks at 3.50 Å (Fig. 8B). If the restrictive cutoff of 3.22 Å were used, most examples (87%) would be filtered out. For CH–O hydrogen bonding, simultaneously applying all the restrictive limits for bond angle and distance would significantly reduce the number of examples from 3894 to 449 (Fig. 8B, orange bars), which would render the current analysis much more limited.

All the CH–O hydrogen bonding interactions in the PDB that satisfy the more inclusive criteria for bond angle (θ¹ = 0–180° and θ² = 0–180°) and distance (d = 2.00–4.00 Å) were tabulated (Fig. 8C and SI). Although not a carbonyl O atom, the S atom of the Met side chain is also known to engage in CH–S hydrogen bonding,¹⁰⁵ which was included in this PDB search as well (171 examples; Fig. 8A and C). Interactions with other heteroatom-containing amino acid side chains (e.g., Ser and Cys) were excluded from this search due to the difficulty of differentiating CH–O hydrogen bonding from a classical hydrogen bonding with OCH₃ serving as the HBA. Among the carbonyl groups surveyed, the amide backbone oxygen atom is most likely to engage in CH–O hydrogen bonding (2402 examples), and examples of Met engagement through CH–S hydrogen bonding were observed as well (Fig. 8C). It is of note that the number of examples shown in Fig. 8C is superior to that of Fig. 7C, likely due to having less angle constraints in CH–O hydrogen bonding.

In the exploration of β-secretase (BACE1) inhibition as a hypothesis toward the treatment of Alzheimer's disease, an amidine fragment was elaborated into biphenyl compound 18 (Fig. 8D).¹⁰⁶ An OCH₃ substituent on the phenyl ring led to 19, which benefited from an 8-fold improvement in potency. The role of the OCH₃ group was identified through an X-ray co-crystal structure of 19 bound to BACE1 (PDB 2VA7; Fig. 8E). The most proximal amino acid residues to the OCH₃ group in question are Ser10 and Ser229, whose carbonyl O atoms are located at 3.56 Å and 3.27 Å, respectively. Although these OCH₃⋯O C interactions are slightly longer than Σ_vdW(C⋯O) = 3.22 Å, the two CH–O hydrogen bonds combine together to allow for potency improvement of the scaffold.

As such, when comparing halogen bonding (Cl⋯O C) and CH–O hydrogen bonding (OCH₃⋯O C), several parallels could be seen: 1) the intermolecular interaction energies are similar, typically ranging from −0.3 to −3 kcal mol⁻¹; 2) the histograms for the bond angle θ² are similar, as the frequency peaks between 90–130°; 3) the histograms for the bond distances are similar, since the frequency continues to increase until d = 3.55 Å, before decreasing; 4) most interactions involve the amide backbone carbonyl, but a minor involvement from the Met S atom could be observed. These observations could potentially be useful when trying to optimize ligand–protein binding with halogen bonding and CH–O interactions in mind.

3.4. Chlorine–π bonding (Cl⋯π) and CH–π bonding (OCH₃⋯π)

The positive electrostatic potential at the terminus of Cl and OCH₃ can engage with the negative electrostatic potential of an amide O atom, resulting in halogen bonding and CH–O hydrogen bonding, respectively (vide supra). Instead, when these δ⁺ regions engage with a δ⁻ electron cloud of an aromatic ring on an amino acid side chain, this leads to chlorine–π bonding (Cl⋯π)^33–35 and CH–π bonding (OCH₃⋯π).³⁶ Conceptually, these types of noncovalent bonds are similar to the interaction of HBDs with aromatic π systems, and to cation–π interactions.^107–110 However, chlorine–π bonding is much weaker than cation–π interactions,¹¹⁰ and in general, it is weaker than halogen bonding (binding energy magnitudes are 10–25% lower):³⁵ gas-phase binding energies for the Cl–π interaction were calculated to be only −1.82 kcal mol⁻¹,⁶⁷ −2.01 kcal mol⁻¹,³⁴ or −2.66 kcal mol⁻¹.³⁵ Similarly, CH–π bonding has been compared to CH–O hydrogen bonding, but bonding energies for CH–π bonding have generally been considered weaker (−0.5 to −1.0 kcal mol⁻¹ in one estimation,¹¹¹ or −1.5 to −2.5 kcal mol⁻¹ in others^112–114).

In a search of the PDB, the angles of Cl–π bonding were restricted in order to minimize the inclusion of quadrupolar Cl–CH_edge interactions (where CH_edge is the C–H bond found at the periphery of an aromatic ring with positive electrostatic potential).³⁴ For the bond angle centered on the Cl atom (Fig. 9A), much like for halogen bonding, θ¹ = ∠(C–Cl⋯ring centroid) was set at 140° as the more inclusive lower limit (θ¹ = 140–180°; resulting in 644 examples), and 165° as the more restrictive lower limit (θ¹ = 165–180°; this would lead to 208 examples). A second bond angle, θ², was defined as ∠(Cl⋯ring atom⋯ring centroid), where the ring atom closest to the Cl atom was used for this angle measurement. Limiting this angle increases the likelihood that the Cl atom is in a region of space above the aromatic ring to engage with the π cloud, as opposed to being in plane with the aromatic ring to engage with the CH_edge. A more inclusive upper limit of 140° has been previously used as a criterion for Cl–π interactions³⁴ (θ² = 0–140°; resulting in 644 examples), but a more restrictive upper limit of 90° would ensure that the Cl atom is strictly placed within the boundaries set by the ring atoms³³ (θ² = 50–90°; this would lead to only 83 examples). However, this strict angle limit would filter out most (87%) of the search hits, since the frequency of search hits peaked at θ² = 90–95° (Fig. 9B, green bars). For bond distance, although a range of distance thresholds of up to 6.0 Å has been previously employed,¹¹⁰ it was deemed important to use a shorter distance limit to minimize confounding interactions, e.g., van der Waals forces. As such, an upper limit of d(Cl⋯ring centroid) was set at 4.50 ų⁴ (resulting in 644 examples). More restrictive limits of 4.20 Å,¹⁰⁰ or even as short as 3.80 Å,³³ have been previously used (however, the criterion of d ≤ 3.80 Å would lead to only 144 examples). For Cl–π bonding, simultaneously applying all the most restrictive limits for bond angle and distance would almost entirely eliminate the number of examples (from 644 to a mere 19; Fig. 9B, orange bars), which would make the analysis of trends fruitless.

Fig. 9. Chlorine–π bonding (Cl⋯π). A) General schematic of the molecular interaction, along with the parameters set in the PDB search for the indicated bond angles and distance. B) Histograms depicting the frequency of the observed angle and bond distance (green bars = inclusive set with parameters shown in part A; orange bars = restrictive set with additional restriction parameters shown in part B). C) Frequency of various (hetero)arene-containing amino acids engaging with the Cl-containing ligand. D) An MMP of compounds 20 and 21 showing the effect of the Cl substituent on potency. E) Cl-containing molecule (21) engaging in chlorine–π bonding with the target protein ERK2 (X-ray co-crystal PDB 5BVF).¹¹⁵.

As such, all the Cl–π interactions in the PDB that satisfy the more inclusive criteria for bond angle (θ¹ = 140–180° and θ² = 0–140°) and distance (d = 2.00–4.50 Å) were tabulated (Fig. 9C and SI). This led to the identification of Tyr (325 examples) and Phe (225 examples) as the most common amino acids that the small-molecule ligand could bind to with this type of interaction.

In a lead optimization program geared at inhibiting extracellular signal-regulated kinase 2 (ERK2), inhibitor 20 was chlorinated at the meta-position of the phenyl group to give 21 with 12-fold improved potency (Fig. 9D).¹¹⁵ The Cl–π interaction is evident in the X-ray co-crystal structure of 21 bound to ERK2 (PDB 5BVF; Fig. 9E), wherein the C–Cl bond is located directly above the ring centroid of Tyr34 and is near-orthogonal to the plane of the aromatic ring (θ¹ = 169.9° and θ² = 71.0°). The very short bond distance d(Cl⋯ring centroid) = 3.27 Å is the closest interaction that this Cl atom experiences inside the protein pocket. However, whether the Cl–π interaction alone is responsible for the improvement in potency is debatable, since the backbone NH of Tyr34 is also in the vicinity of the Cl atom (d = 3.61 Å), with this hydrogen bond possibly contributing to a small portion of the binding interaction energy.

The bond angle of interaction is also important for CH–π bonding (OCH₃⋯π), and therefore two bond angles were monitored during the PDB search (Fig. 10A): θ¹ = ∠(O–CH₃⋯ring centroid) and θ² = ∠(CH₃⋯ring atom⋯ring centroid). For θ¹, its value was not restricted (θ¹ = 0–180°) because the orientation dependence of the interaction energies of the OCH₃⋯π complex is very small.¹¹² The range for θ² was set at a more inclusive upper limit of 140° much like for Cl–π interactions (θ² = 0–140°; resulting in 828 examples), but a more restrictive upper limit of 90° could potentially be used (θ² = 50–90°; this would lead to 291 examples). A limited angle would indicate that the OCH₃ methyl group is located directly above (or below) the aromatic ring, and not in plane with the ring.¹¹¹ The frequency of search hits was analyzed in a histogram for both θ¹ and θ², indicating that CH–π interactions are frequent between θ² = 80–110°, and that applying a cutoff at 90° would be too limiting (Fig. 10B, blue bars). For bond distance, much like for Cl–π interactions, the more inclusive upper limit of d(CH₃⋯ring centroid) was set at 4.50 Å^111,114 (resulting in 828 examples), but the more restrictive limit of 3.80 Å could be employed (this would lead to 260 examples). A histogram analysis showed that the first frequency peak occurs at 3.60–3.80 Å, which agrees with the previous findings in the literature;¹¹¹ however, a significant portion of the examples were found with greater intermolecular bond distances (Fig. 10B). For CH–π bonding, simultaneously applying all the restrictive limits for bond angle and distance would significantly reduce the number of examples from 828 to 207 (Fig. 10B, orange bars), which would obscure our understanding of the trends in frequency.

Fig. 10. CH–π bonding (OCH₃⋯π). A) General schematic of the molecular interaction, along with the parameters set in the PDB search for the indicated bond angles and distance. B) Histograms depicting the frequency of the observed angle and bond distance (blue bars = inclusive set with parameters shown in part A; orange bars = restrictive set with additional restriction parameters shown in part B). C) Frequency of various (hetero)arene-containing amino acids engaging with the OCH₃-containing ligand. D) A matched series of compounds 22–26 showing the superior effect of the OCH₃ substituent on potency compared to other substituents in this benzoxazole series. E) OCH₃-containing molecule (25) engaging in CH–π bonding with the target protein thrombin (X-ray co-crystal PDB 6ZUG).¹¹⁷.

All the CH–π interactions in the PDB that satisfy the more inclusive criteria for bond angle (θ¹ = 0–180° and θ² = 0–140°) and distance (d = 2.00–4.50 Å) were tabulated (Fig. 10C and SI). Much like in the case of Cl–π bonding, Phe (268 examples) and Tyr (230 examples) were identified as the most common amino acids that the small-molecule ligand could bind to using CH–π interactions.

There are two distinct protein targets to treat blood coagulation (also known as thrombosis): Factor Xa and thrombin. Although methoxy-containing molecules like apixaban are blockbuster drugs for Factor Xa inhibition,¹⁶ direct thrombin inhibition could potentially be a better mechanism of action,¹¹⁶ and therefore active research is being conducted to provide patients with more options for small-molecule inhibitors of thrombin. In a late-stage preclinical program (Fig. 10D),¹¹⁷ the C7 position of the benzoxazole core in 22 (IC₅₀: 214 nM) was substituted with Cl, CH₃, or OCH₃ for protein pocket finding in a manner reminiscent of the Topliss approach.^19,20 A one-heavy-atom substituent like Cl (23) or CH₃ (24) led to a slight improvement in potency, but a two-heavy-atom substituent like OCH₃ (25) was deemed to fit perfectly in the thrombin “S2 pocket”, resulting in an 18-fold potency improvement (IC₅₀ of 25: 12 nM).¹¹⁷ Interestingly, another two-heavy-atom substituent like ethyl (CH₂CH₃; 26) was found to be worse for potency (IC₅₀ of 26: 104 nM), since the terminal CH₃ group rotates out of plane with respect to the benzoxazole ring. The X-ray co-crystal structure of 25 bound to thrombin (PDB 6ZUG) shows that the CH₃ group needs to be in plane with the benzoxazole ring in order to project toward the Tyr60A aromatic ring, and therefore an OCH₃ group was more optimal than CH₂CH₃ (Fig. 10E).¹¹⁷ The CH–π interaction is evident, wherein the CH₃ group with δ⁺ character is located directly above the ring centroid of the benzene of Tyr60A, and the O–CH₃ bond is orthogonal to the plane of the aromatic ring (θ¹ = 162.4° and θ² = 82.4°), with a very short bond distance d(CH₃⋯ring centroid) = 3.36 Å. A second CH–π interaction to the benzene ring of Trp60D (d = 3.97 Å) can also be observed. Moreover, the conformation of the OCH₃ group shown in the X-ray is more stable by 1.5 kcal mol⁻¹ compared to a conformation where the CH₃ group is in the vicinity the benzoxazole O atom, further aiding the conformation that facilitates CH–π interactions.¹¹⁷ This C7-methoxybenzoxazole became a key structural element that was retained in the candidate molecule, BAY 1217224, which is currently in clinical trials to investigate the treatment of thrombosis.¹¹⁷

As such, when comparing chlorine–π bonding (Cl⋯π) and CH–π bonding (OCH₃⋯π), several parallels could be noted: 1) the intermolecular interaction energies are similar, typically ranging from −0.25 to −2.5 kcal mol⁻¹; 2) the histograms for the bond angle θ² have similarities in that the 90° limit is too restrictive, and the definition of the interaction should include at least up to θ² = 120°; 3) the histograms for the bond distances are similar, with the greatest frequency occurring at d > 4.00 Å, and the d = 3.80 Å limit is likely to be too restrictive; 4) most interactions involve the Tyr and Phe benzene rings. These observations could potentially be useful when trying to optimize ligand–protein binding with chlorine–π bonding and CH–π bonding in mind.

3.5. Cooperative hydrogen bonding

Upon inspection of numerous X-ray co-crystal structures of Cl- and OCH₃-containing ligands bound to proteins, a recurring motif was apparent from the ligand perspective: 1,2-disubstituted heteroarenes, where the substituents at positions 1 and 2 are both Cl, both OCH₃, or a combination of each. Even though an individual hydrogen bond arising from a Cl atom or an OCH₃ group might be weak, multiple simultaneous interactions could magnify this binding energy. From the protein perspective, the amino acid side chain that acts as the HBD is often a “double HBD” such as NH₂ in Asn and Gln, and occasionally a “triple HBD” such as NH₃⁺ in Arg and Lys. This could be rationalized by the fact that Asn and Gln have a short carbon linker in the side chain, and therefore engaging these amino acid residues is enthalpically productive without significant entropic loss. Furthermore, these carboxamide-containing amino acids have two HBDs at defined angles in the same plane, with restricted rotational freedom around the CO–NH₂ bond. Thus, when two HBAs are placed in the same plane by virtue of 1,2-disubstitution on an arene, an optimal, cooperative interaction results.

Typically, 1,2-dichloroarenes engage an Asn or Gln residue with an in-plane approach (Fig. 11A). The side chain carboxamide's NH₂ group is often equidistant from the two Cl atoms to maximize both hydrogen bonding interactions. Simultaneous hydrogen bonds that are parallel to each other have been reported to be stronger than “bifurcated” or “zigzag” hydrogen bonding networks.¹¹⁸ Although a Cl-containing ligand could potentially rotate in place when only a single hydrogen bond interaction is available, a second hydrogen bond would lock the small-molecule's rotation and secure the ligand inside the protein pocket.

Fig. 11. Cooperative hydrogen bonding. A) General schematic of combined intermolecular interactions from two Cl substituents, wherein two hydrogen bonding interactions to a single amino acid side chain stabilize the ligand–protein interaction. B) X-ray co-crystal structure of a bis-Cl-containing compound in the protein pocket of 8-oxoguanine DNA glycosylase 1, showing both hydrogen bonds taking effect at similar bond angles and distances (PDB 9HLF).¹¹⁹ C) General schematic of combined intermolecular interactions from two OCH₃ groups, wherein two hydrogen bonding interactions to a single amino acid side chain stabilize the ligand–protein interaction. D) X-ray co-crystal structure of a bis-OCH₃-containing compound in the protein pocket of BRD4, showing both hydrogen bonds taking effect at similar bond angles and distances (PDB 6VUJ).¹²¹.

An X-ray co-crystal structure (PDB: 9HLF) of a dichloroaniline-containing fragment molecule bound to 8-oxoguanine DNA glycosylase 1 (OGG1) illustrates this cooperative hydrogen bonding interaction (Fig. 11B).¹¹⁹ Although the hydrogen atoms are not resolved in this X-ray and are not apparent, the carboxamide side chain N atom of Gln315 is located at bond distances of 3.29 Å and 2.91 Å to the Cl atoms, which are both shorter than Σ_vdw(Cl⋯N) = 3.30 Å. The bond angles of 110.8° and 122.1° indicate engagement of the anionic belt of the Cl atoms, suggesting an energetically favorable hydrogen bonding. Although the direct potency effect of the two Cl atoms in this molecule was not disclosed, this dichloroaniline moiety is a privileged motif for OGG1 binding, and an earlier report showed a 32-fold potency improvement from the introduction of 1,2-disubstituted Cl atoms on a similar scaffold.¹²⁰

A 1,2-dimethoxyarene motif adopts a specific conformation where the two OCH₃ groups are coplanar with the arene ring, with the CH₃ moieties projecting away from each other to avoid a steric clash (Fig. 11C).¹⁶ This allows the available lone pairs of electrons to orient themselves in the same direction, resulting in a cooperative interaction where the HBA strength of a 1,2-dimethoxyarene system is amplified (vide supra comparing the pK_BHX = 0.09 of anisole⁷¹ to the pK_BHX = 1.16 of 1,2-dimethoxybenzene⁷²).

An X-ray co-crystal structure (PDB: 6VUJ) of a dimethoxybenzene-containing molecule bound to bromodomain 4 (BRD4) illustrates this cooperative hydrogen bonding interaction (Fig. 11D).¹²¹ The Asn140 side chain's NH₂ group is located at short bond distances of 3.17 Å and 3.01 Å from the OCH₃ oxygen atoms, and with ideal bond angles of 104.9° and 109.8°, indicating an equidistant, cooperative hydrogen bonding. In this example, the biological effect of the two OCH₃ groups in this molecule is known: the desmethoxy benzene analog was found to have IC₅₀ = 3200 μM, whereas the corresponding 1,2-dimethoxybenzene displayed IC₅₀ = 52 μM. Since the OCH₃ group is not a contributor to a molecule's lipophilicity,^16,17 this 62-fold improvement in potency directly resulted in almost 2 units of improvement in lipophilic ligand efficiency (LLE).¹²¹

The effects of cooperative hydrogen bonding interactions extend beyond Cl- and OCH₃-containing small-molecule ligands. For example, a 1-alkoxy-2-chloroarene system or a 1-alkoxy-2-methoxyarene can benefit from similar cooperativity. As such, installation of a Cl or OCH₃ substituent can be even more effective when placed vicinal to an existing alkoxy group in the molecule.

3.6. Dual electrostatic interactions

All the noncovalent, intermolecular interactions presented in this article have been considered weak interactions:

1) for hydrogen bonding, the interaction energy could potentially reach −5.0 kcal mol⁻¹ for Cl or OCH₃, but desolvation penalties need to be taken into account;

2) for orthogonal multipolar interactions, the binding energy could potentially reach −1.0 kcal mol⁻¹;

3) for halogen bonding or CH–O hydrogen bonding, the interaction energy could potentially reach −3.0 kcal mol⁻¹;

4) for chlorine–π and CH–π bonding, the interaction energy could potentially reach −2.5 kcal mol⁻¹.

All these calculated values are for highly isolated interactions of model molecules in the gas phase, but in ligand–protein interactions, the perfectly ideal bond angles and bond distances are extremely difficult to design, and therefore these maximal values are typically not reached. Given that −1.4 kcal mol⁻¹ of binding energy translates to a 10-fold improvement in ligand–protein binding affinity at physiological temperatures,¹²² a single Cl or OCH₃ substituent does not often result in a potency improvement of multiple orders of magnitude. However, in the best-case scenarios, >1000-fold improvements in potency have been observed upon introduction of a Cl substituent,⁸ and >100-fold improvements have been observed with an OCH₃ substituent.¹⁶ These significant boosts in potency likely arise from a combination of conformational effects and additive enthalpic effects that result from the dual electrostatic nature of Cl and OCH₃ substituents.

This dual electrostatic advantage is well-documented for chlorine as well as other heavier halogen atoms:^24,66,67 hydrogen bonding could take place in the “side-on” orientation with its negative electrostatic potential, and halogen bonding could take place in the “head-on” orientation with its positive electrostatic potential (Fig. 12A). For example, an X-ray co-crystal structure of a Cl-containing compound bound to the BACE1 protein pocket¹²³ indicates hydrogen bonding to the Gly135 backbone NH with θ = ∠(C–Cl⋯N) = 117.6° and d(Cl⋯N) = 3.51 Å, as well as halogen bonding to the same Gly135 backbone C O with θ¹ = ∠(C–Cl⋯O) = 166.3°, θ² = ∠(Cl⋯O C) = 116.1°, and d(Cl⋯O) = 3.45 Å (PDB 4FRS; Fig. 12B). The effect of the Cl substituent on potency was elucidated in an MMP, in which a 19-fold improvement in potency arose from this tactful combination of hydrogen and halogen bonding.¹²³

Fig. 12. Dual electrostatic interactions. A) General schematic of intermolecular interactions from a Cl substituent, whose dual electrostatic nature could participate in both hydrogen bonding and halogen bonding. B) X-ray co-crystal structure of a Cl-containing compound in the protein pocket of BACE1, showing both hydrogen bonding and halogen bonding (PDB 4FRS).¹²³ C) General schematic of intermolecular interactions from an OCH₃ substituent, where its dual electrostatic nature can engage in both hydrogen bonding and CH–O hydrogen bonding. D) X-ray co-crystal structure of an OCH₃-containing compound in the protein pocket of hTRK-A, showing both hydrogen bonding and CH–O hydrogen bonding (PDB 6PMA).¹²⁴.

The dual electrostatic nature of an OCH₃ group could be exploited in similar fashion: hydrogen bonding could take place from the O atom with its δ⁻ component, and CH–O hydrogen bonding could take place from the CH₃ group with its δ⁺ component (Fig. 12C). An X-ray co-crystal structure of an OCH₃-containing compound bound to human tropomyosin receptor kinase A (hTRK-A) indicates hydrogen bonding to the Met592 backbone NH with θ = ∠(C–O⋯N) = 132.9° and d(O⋯N) = 2.97 Å, and CH–O hydrogen bonding to the same Met592 backbone C O with θ² = ∠(CH₃⋯O C) = 127.3°, and d(CH₃⋯O) = 3.03 Å (PDB 6PMA; Fig. 12D).¹²⁴ This compound arose from a hit identification effort, and therefore synthesis was not undertaken to clarify the role of the OCH₃ group on potency. However, the interaction of the OCH₃ group with Met592 was found to be critical in the binding with hTRK-A because other hit molecules with binding to Met592 were found to be OCH₃ bioisosteres: the OCH₃ group's O atom could be considered bioisosteric to a pyridine N atom (PDB 6PMB), a carbonyl O atom (PDB 6PME), and an indazole N atom (PDB 5KVT).¹²⁴

As described above, the dual electrostatic role of the Cl substituent manifests itself most commonly in a combination of hydrogen bonding and halogen bonding,^24,66,67 and an OCH₃ substituent results primarily in a combination of hydrogen bonding and CH–O hydrogen bonding. Although combinations could potentially include multipolar interactions or Cl–π/CH–π interactions (e.g., in Fig. 9D, where Cl–π bonding is combined with hydrogen bonding), these types of interactions have many restrictions regarding bond angles, and are therefore difficult to combine, especially at shorter intermolecular distances. Highly directional interactions are difficult to implement because subtle design errors could immediately lead to a loss in binding energy,¹²⁵ and especially, combining directional interactions using substituents with dual electrostatic character would likely necessitate many iterations of structure-based drug design.

4. Conclusions

Among small substituents in medicinal chemistry exploration, Cl and OCH₃ are arguably the most versatile thanks to the dual presence of both negative and positive electrostatic potentials. When exploring SAR to understand a small-molecule ligand's pharmacophore, this electrostatic versatility could favor Cl and OCH₃ over other substituents like F and CH₃. There are parallels between the Cl and OCH₃ groups regarding their intermolecular interactions: the dual electrostatic nature of these substituents can engage with both δ⁺ and δ⁻ regions of a protein pocket, resulting in 1) hydrogen bonding, 2) orthogonal multipolar interactions, 3) halogen bonding or CH–O hydrogen bonding, and 4) Cl–π bonding or CH–π bonding.

A comparison of Cl and OCH₃ was accomplished using bond angles and distances collected in a comprehensive search of the PDB. Although lessons were derived through the lens of structure-based drug discovery, the versatility of these substituents could benefit drug design even when the target protein's structure is not known. Performing a scan of Cl and OCH₃ substituents on the core scaffold and employing these groups as scouts for protein pocket finding could be a potency improvement strategy.¹⁷ Although numerous similarities have been presented in this article, the opposite electronic characters of these two groups, i.e., the electron-withdrawing nature of Cl and the electron-donating character of OCH₃, make both groups worth employing in parallel in the early stages of drug discovery.

Although Cl and OCH₃ are oft-used substituents, with >200 approved drugs each for Cl-containing drugs¹²⁶ and OCH₃-containing drugs,¹⁶ there could be disadvantages when installing numerous Cl and OCH₃ groups onto a molecule. For Cl, the increased lipophilicity and associated ADMET problems (lowered aqueous solubility, increased hERG liabilities, etc.) are deterrents, and therefore a maximum of three Cl atoms in a preclinical molecule is advisable.⁸ For OCH₃, the lowered metabolic stability due to O-demethylation is a concern, and therefore a maximum of three OCH₃ groups in a molecule is recommended.^16,17 To address these concerns, a Cl or OCH₃ substituent could be first used as a scout to identify a protein pocket, but then replaced at a later stage of the drug discovery process. For example, in the discovery of the approved (but later withdrawn) drug betrixaban, a Cl scan on a benzene ring led to an 8-fold potency improvement, but the Cl was replaced with an OCH₃ group to alleviate hERG liabilities.¹²⁷ In the discovery of the anti-HIV drug fostemsavir, an OCH₃ scan on an indole core led to a 23-fold potency improvement, but the metabolically labile OCH₃ was replaced with a carboxamide, and then finally with a triazole.¹²⁸ Thus, Cl and OCH₃ substituents are employed often in early discovery, even in fragment libraries,¹²⁹ but bioisosteric replacement strategies¹³⁰ such as Cl to F⁸ or CN,¹³¹ and OCH₃ to F¹³² or OCD₃,^16,17 could be considered in the later stages of drug discovery.

Finally, it is of note that all the intermolecular forces directly engaging with the Cl or OCH₃ groups are weak. The four interactions discussed in this article, as well as Cl–CH_edge, O–CH_edge, π–π interactions, and van der Waals interactions, rarely result in a significant potency improvement if presented individually. However, these noncovalent forces could contribute cooperatively in ligand–protein binding, which as a summation, could translate into a significant enhancement in potency toward a desired biological target.²³ Thus, small substituents like Cl and OCH₃ that can engage in a multitude of specific and directional interactions are practical functional groups in medicinal chemistry, both for structure-based design as well as for activity-based design, and should be considered in parallel as a strategy for potency optimization.

Conflicts of interest

There are no conflicts to declare.

Data availability

For each type of molecular interaction discussed in this article, data tables with chemical structure, PDB identifier, X-ray crystal structure resolution, intermolecular bond distance, relevant bond angles, and target amino acid residue have been compiled and are available in Supplementary Information files (a total of 8 tables in .csv format).

Supplementary information (SI) is available. See DOI: https://doi.org/10.1039/d5md00848d.