Novel-Type GABA_B PAMs: Structure–Activity Relationship in Light of the Protein Structure

Abstract

Selecting a known HTS hit with the pyrazolo[1,5-a]pyrimidine core, our project was started from CMPPE, and its optimization was driven by a ligand-based pharmacophore model developed on the basis of published GABA_B positive allosteric modulators (PAMs). Our primary goal was to improve the potency by finding new enthalpic interactions. Therefore, we included the lipophilic ligand efficiency (LLE or LipE) as an objective function in the optimization that led to a carboxylic acid derivative (34). This lead candidate offers the possibility to improve potency without drastically inflating the physicochemical properties. Although the discovery of the novel carboxyl feature was surprising, it turned out to be an important element of the GABA_B PAM pharmacophore that can be perfectly explained based on the new protein structures. Rationalizing the binding mode of 34, we analyzed the intersubunit PAM binding site of GABA_B receptor using the publicly available experimental structures.

GABA_B receptors are the slow and prolonged response-mediating metabotropic counterparts of the fast ionotropic GABA_A receptors, with whom they convey hand in hand the neurophysiological effects of GABA, the principal inhibitory neurotransmitter of the brain. GABA_B receptors belong to Class C of G-protein coupled receptors (GPCRs) together with metabotropic glutamate receptors, calcium-sensing receptors, taste receptors, and a number of orphan receptors, showing the highest homology to the metabotropic glutamate receptors.^1,2

GABA_B receptors are widely expressed in the central nervous system,^3,4 and the GABA_B receptor system seems to play a general role in the regulation of neuronal excitability with consequences on various aspects of behavior. This indicates the potential usefulness of GABA_B receptor stimulation in several medical conditions including anxiety, pain, drug addiction, epilepsy, schizophrenia, and autism.⁵⁻⁸

Allosteric modulation of the GABA_B receptor is an attractive goal for drug development. Positive allosteric modulators (PAM) increase, and negative allosteric modulators (NAM) decrease the efficacy and/or affinity of the orthosteric agonist.^9,10 The character of an allosteric site may greatly differ from the orthosteric site from the drugability aspect, and in general, allosteric modulation is expected to show increased selectivity, fewer side effects, lack of desensitization, and development of tolerance, which has indeed been shown in preclinical models.^5,11,12

Although a variety of GABA_B receptor PAM ligand series have been reported over the years, only a few compounds (e.g., ADX71943, ADX71441, ASP8062, ODM106) have qualified for investigations in clinical settings, and none of these ligands have been approved for human use to date.¹³ This indicates that notwithstanding the potential advantageous and inviting features, the development of druglike GABA_B PAMs is challenging.

GABA_B receptors are obligate heterodimers consisting of two similar, but distinct subunits, the B₁ and B₂. Both subunits are built from an extracellular Venus flytrap domain (VFT), a heptahelical transmembrane domain (TMD) and an intracellular tail.¹⁴ The VFT domain of the B₁ subunit binds GABA as well as other orthosteric agonists (e.g., baclofen, SKF97541) and antagonists (e.g., phaclofen, saclofen, SCH50911), while the B₂ subunit is responsible for G-protein activation.¹³ The binding modes and molecular mechanisms of allosteric modulators are still uncertain, and multiple allosteric binding sites may exist.¹³ However, latest cryo-EM studies suggest that the main binding site of PAMs like (+)-BHFF (1)^15,16 and GS39783 (2)¹⁷ can be found at the interface between the transmembrane domains of GABA_B1 (GB₁) and GABA_B2 (GB₂) site (depicted in Figure 2A). These surprising results may have an essential role in the development of new positive allosteric ligands, since the structural information helps to interpret the earlier ligand-based structure–activity relationships¹⁸ and may point to still unexploited opportunities of PAM development.

Figure 2.

Experimental structures of GABA_B receptor. A) Heterodimeric GABA_B receptor (PDB ID: 6UO8, VFT: Venus flytrap domain, OBS: orthosteric binding site, TMD: transmembrane domain). B) Superimposed binding sites of PAMs (PDB IDs: 6UO8 and 7C7Q). C) SiteMap characterization of the PAM binding site: green areas are lipophilic, and red and blue areas are favorable for hydrogen bond acceptors and donors, respectively. D) Schematic representation of the Tyr-810^6.44(B1)···Asn-698^6.45(B2) and Asn-811^6.45(B1)···Tyr-697^6.44(B2) hydrogen bond network.

At the outset of our work in developing a novel type of GABA_B PAMs, sufficient information was lacking for structure-based modeling, and even the location of the potential binding site was uncertain. Therefore, we started from a small, diverse set of known PAM ligands that were experimentally characterized (Table 1). Based on these data, we built a novel pharmacophore model that was used to design systematic modifications on each region of a selected reference compound, CMPPE (3).¹⁹ The new ligand-based pharmacophore model which was built with PHASE²⁰ using the representative set of GABA_B PAMs is presented in Figure 1. Our analysis supported the high importance of the motif containing the central aromatic site (r) and the surrounding three apolar sites (h1–3). Based on the aligned structures, the a1 site was occupied by an acceptor atom, with at least one of the a2 or a3 atoms also occupied. The a4 site was positioned mainly based on the sulfonyl group of the Roche reference (6), and it was thought that this may be a pharmacophore site that can improve affinity but is not utilized in the other reference compounds. This site was thought to be a possible breakout point that could lead to the discovery of novel compounds.

Table 1. Activity of Reference Compounds.

^aRacemic form.

^bref (26).

Figure 1.

Ligand-based pharmacophore and some reference compounds: compound 3 (with green carbons, (S)-isomer), compound 4 (with cyan carbons), compound 6 (with magenta carbons); h1–3: hydrophobic sites, r: aromatic site, a1–4 hydrogen bond acceptor sites.

Lipophilic ligand efficiency (LLE or LipE) was used as an objective function in this study, and LLE values of selected references²¹⁻²⁴ are presented in Table 1. LLE is a simple index derived from in vitro potency taking lipophilicity into account (see eq 1), which measures the specificity of ligand binding. If molecules with similar conformational, rotational, and translational degrees of freedom are compared, LLE correlates with binding enthalpy.²²⁻²⁴

1

Our primary biological test in this work was the functional activation of rat cortical GABA_B receptors ([³⁵S]GTPγS binding) with a pEC₅₀ end point. Binding affinity (rat GABA_B pK_i) was also determined for selected ligands utilizing a novel competitive binding assay developed based on the novel radioligand, which is a labeled GABA_B PAM from Addex Therapeutics (compound 4.156 in patent WO2008/056257, an analogue of ADX71441). We found a significant correlation between pK_i and pEC₅₀ values, which supports the validity of the hypothesized relationship between LLE (defined with pEC₅₀) and binding modes (see Figure S6 and Table S1). Moreover, in addition to competitive binding studies supporting the hypothesis of the common binding site of all the compounds, receptor modulation has been investigated using binding assay with the orthosteric radioligand [³H]baclofen²⁵ (see Supporting Information). The human and rat GABA_B sequences were compared, and high sequence identity was found for both subunits (>97%) (see Supporting Information).

In light of the recently available structural information, now we have the opportunity to understand the role of the individual pharmacophoric sites in the ligand-based model. Superposition of the PAM bound experimental protein structures available in the Protein Data Bank (PDB) showed only a slight overlap of the binding modes of (S)-BHFF and GS39783 ligands (see Figure 2B), even if the PAMs occupied the same binding site between the two subunits. In parallel, the two different GABA_B PAMs could not be fitted well to a common pharmacophore model, where the individual pharmacophoric sites could be matched (see Figure S1). These suggest a relatively large binding pocket, possibly with multiple sites where specific interactions could stabilize the protein–ligand complex.

In order to gain further insights, the PAM binding site was evaluated by SiteMap.²⁷ This analysis showed that it has a mainly lipophilic character with some spots advantageous for hydrogen bond donor and acceptor groups, allowing more specific interactions to form between the ligand and the protein (see Figure 2C). Regarding the protein–protein interactions, the transmembrane interface of the B₁ and B₂ subunits are highly lipophilic; however, the Tyr-810^6.44(B₁)···Asn-698^6.45(B₂) and Asn-811^6.45(B₁)···Tyr-697^6.44(B₂) hydrogen bonds stabilize the heterodimer (see Figure 2D). This hydrogen bond network may be affected by the PAM binding, since, for instance, GS39783 interacts with Asn-698^6.45(B₂) and π–π stacking can be significant, especially with Tyr-810^6.44(B₁).¹⁷ Overall, π–π stacking and hydrogen bond interactions with these residues seem to be important for binding of ligands having a common pharmacophore with GS39783, and it is expected that PAMs should increase the overall stability of the GABA_B heterodimer. Moving toward the bottom of the binding site, it becomes more hydrophilic since it is closer to the intracellular bulk phase. Polar side chains at the intra- and extracellular side can stabilize the position of the protein inside the cell membrane. In the GABA_B intersubunit PAM site, the Lys-792^5.62(B₁) and Lys-690^6.37(B₂) may be important interaction partners for the ligands having a hydrogen bond acceptor, aromatic ring (cation−π interaction), or even negatively charged group (salt bridge) in this site. Since the side chains of these ionic residues are highly flexible and this binding site does not have a closed rigid structure, the common pharmacophore hypothesis should be handled not as a rigid body with small tolerances as is possible in several cases but as a more flexible pattern. Methods taking the protein flexibility and induced-fit effects into consideration are required.²⁸

Next, these reference compounds were placed at the GABA_B PAM binding site with induced-fit docking protocol (IFD),²⁸ which allows protein relaxation locally at the binding site (see Figure S2) and applies an advanced conformational sampling. After manual inspection and clustering, the obtained various binding poses could be narrowed down to two frequent orientations (Figure 3A). According to the appearance frequency and IFD-scores, one of them was considered as “more likely” (pose A, cyan) while the other reversed orientation (pose B, pink) as “less likely”. In the pose A, the large lipophilic part of the docked ligands was located in the left side of Figure 3A, where SiteMap suggested a large lipophilic area in the binding site under the Asn-811^6.45(B₁) and Tyr-697^6.44(B₂) residues (see Figure 2C).

Figure 3.

Predicted binding poses. A) Most relevant binding poses of (S)-CMPPE in protein (PDB ID: 6UO8). B) Schematic representation of the common pharmacophore; H1–3: hydrophobic sites, R: aromatic site, A1 hydrogen bond acceptor site, and A/D: area favorable for hydrogen bond acceptor and/or donors.

Superposition of the IFD models (see Figure S2) gave a schematic picture of the common pharmacophore of reference compounds, which is quite similar to the ligand-based approach (see Figure 3B and Figure 1). This pharmacophore could complete the SiteMap results shown in Figure 2C. It contains an acceptor site (A1) responsible for the interaction with the Asn-698^6.45(B₂) and a heteroaromatic site (R) present in all reference ligands that is capable of forming a π–π stacking interaction with the Tyr-810^6.44(B₁). The ligand alignment suggests three hydrophobic sites (H1–3) around the heteroaromatic core and show that further hydrogen bond donor and acceptor interactions are possible with the protein. However, the H3 site seems to be closer to the a4 site of the ligand-based model than to its h3 site (methyl group of CMPPE) (Figure S2). Furthermore, hydrogen bonds with the ligand at the acceptor/donor (A/D) region can be formed with the highly flexible Lys-792^5.62(B₁) and Lys-690^6.37(B₂) side chains and with the protein backbone, too. Thus, the optimal position of the acceptor and donor sites in the A/D region of the ligand structure is difficult to predict, and it is expected that the activity may be less sensitive for the position of this pharmacophoric site. Reference compounds are depicted in Table 1 according to this pharmacophore model.

Optimization of the specific protein–ligand interactions has high importance in drug design since this enthalpy-driven approach can increase the binding affinity without the inflation of physicochemical properties. It is advantageous for selectivity, ADME properties, and toxicity. In contrast, entropy-driven improvements of affinity increase the size and the lipophilicity,²³ both contributing to promiscuity and PK problems. Systematic modifications were applied to the selected reference compound CMPPE (3) in each region. Our aim was to increase enthalpic protein–ligand interactions, resulting in increased specificity of binding characterized with LLE. We considered CMPPE (3) as a promising starting point offering space for synthetic chemical modifications. At first, we investigated how the modification of the heteroaromatic core can affect binding specificity. Data summarized in Table 2 suggest that varying heteroatoms and their position in the heteroaromatic system did not increase the LLE. Since the correlation of clog D and pEC₅₀ is high (0.75) in this set of compounds, resulting in similar LLE values, the effect of these modifications on the activity should be mainly entropic instead of enthalpic. Although it should be noted that compounds having a nitrogen atom at the a2 site of the ligand-based pharmacophore (Figure 1) showed slightly higher LLE, but the importance of the a3 site is not supported by the results (3 vs 9 and 7 vs 11). Changing the ring system and, therefore, the charge distribution of the core did not result in any promising analogues. Compounds 9 and 11 which have got slightly better pEC₅₀ than CMPPE (3) are more lipophilic at the same time.

Table 2. Effect of Heteroaromatic Core Modifications.

Substitution of the p-chlorophenyl of CMPPE (3) at the h1/H1 site of the pharmacophore led to a similar conclusion as core modifications: specificity of binding did not seem to be increased considerably (Table 3). Substitution of the chlorine atom by a hydrogen (16) resulted in a slightly higher LLE, while other substitutions (13–15) did not significantly affect the LLE. This suggested that the chlorine atom did not form any specific interaction (e.g., halogen bond) at the binding site. It is in accordance with the modeling results suggesting that ligands can be extended in this direction toward the membrane lipids without any additional specific protein interactions. Interestingly, the position of the chlorine seems to be more important since, in the meta-position (12), the obtained LLE was lower by about 0.9 unit. The explanation might be partly based on entropy change because p-chlorophenyl can be rotated by 180°, which results equivalent conformers, but rotation of the m-chlorophenyl ring results two nonequivalent conformers from which one matches better to the binding site and so the binding event reduces the degrees of freedom of the ligand more in the latter case. Although the application of a benzyl group instead of phenyl breaks the shape of the flat scaffold of CMPPE (3), which may be a fundamental difference, compound 17 was obtained as an active compound with slightly decreased LLE. In summary, it is possible to change lipophilic groups at the h1/H1 site, but we were not able to increase the LLE significantly by the applied modifications.

Table 3. Effect of Modifications at the H1 Site.

In terms of LLE values, substitution of the aliphatic ring at the h2/H2 site of the reference CMPPE (3) led to the most promising derivatives (Table 4). Application of an aromatic ring instead of a piperidine showed that an acetamido-substituted phenyl ring was acceptable in this position (18, 19) since these derivatives had similar LLE values as CMPPE (3). Indeed, ADX71441 (5) similarly contains an aromatic group in this position (Table 1) that further supported our initial pharmacophore hypothesis. At the same time, these data did not suggest that aromatic substituents would yield increased binding specificity, in exchange for extending the delocalized system. Therefore, aromatic side chains at this position, with their potential liability of solubility problems in general,²⁹ were not preferred. Incorporating hydrogen bond donor and acceptor atoms into the ring, such as morpholine, lactam, or open-chain amide function (20, 21, 22, 23), in contrast, increased the LLE significantly by about 0.4–0.9 unit, suggesting more specific binding of these derivatives. In terms of binding specificity, the most promising modification was the introduction of a carboxylic group (25, 26). This modification resulted in much more polar derivatives at the “price” of a moderate loss of biological activity. These findings could be rationalized with the structure-based modeling results, as well. Our model suggests that this part of the ligands is most probably accommodated in the intracellular side of the binding site, where flexible lysine side chains (Lys-792^5.62(B₁), Lys-690^6.37(B₂)) are present. These side chains can form a hydrogen bond or participate in cation−π and even in ionic interactions (salt bridge). Because of the flexibility of these side chains, it is not surprising that pEC₅₀ and LLE are not sensitive for the position of the carboxylic group.

Table 4. Effect of Modifications at the H2 and A/D Sites.

^aRacemic form.

The importance of the a4 site of the ligand-based pharmacophore, which seems to be rather a hydrophobic site (H3) based on the protein structure, was investigated via substitution of the hydrogen atom in the 6-position (27) of the CMPPE (3) core (Table 5). In the case of the methoxy group, the LLE was increased by about one unit parallel with the biological activity. This suggested that substitution in this position may highly be favorable for increased binding specificity. The nitrile substitution (28) and the effect of an ethoxycarbonyl group (29) were tested in combination with the morpholine group at the H2 site, which increased LLE slightly in the first and nonsignificantly in the second case compared to compound 20. Thus, the small gain in the pEC₅₀ value of compound 29 (from 5.61 to 6.09) might have resulted from the change in lipophilicity, but it suggests that ligands can be extended in this direction in accordance with the docking results. Derivatives with methylthio, methanesulfinyl, and methanesulfonyl groups (30–32) at the a4/H3 site and 3-acetamidophenyl at the H2 site were also prepared, inspired by the Roche reference compound (6). In this series, the methylthio substitution (30) displayed an exceptional leap in potency and affinity (pK_i of 7.81) resulting in both biological activity and its LLE increased compared to both compounds 18 and 3 (CMPPE). In line with the increase in polarity, the oxidated derivatives (31, 32) showed lower activities, but even higher LLE, which may be the consequence of proximal highly flexible and charged Lys-792^5.62(B₁) and Lys-690^6.37(B₂) side chains, which are available for contacting with methanesulfinyl and methanesulfonyl groups (see Figure S3). This also explains the role of acceptor site a4 in the ligand-based pharmacophore model (Figure 1).

Table 5. Effect of Modifications at the H3 Site in Combination with Modification at the H2 and A/D Sites.

^aRacemic form.

Since carboxylic acid function at the A/D site led to highly emerged LLE and lipophilic groups at the H3 site seem to be able to increase the LLE, we hypothesized that the combination of these two types of substitution may be advantageous. Since the highly polar carboxyl group decreased the PAM activity of compounds 25 and 26, it is expected that lipophilic groups at the H3 site can lead to an optimal balance of biological activity and lipophilicity with high binding specificity. The in vitro activity data of compound 33 (pEC₅₀ = 5.69) and especially compound 34 (pEC₅₀ = 6.93) (Table 5) supported this hypothesis. Although their pEC₅₀ values were somewhat lower than that of compound 30 (pEC₅₀ = 7.75), their clog D values were more than three units lower. Consequently, the obtained LLE of compounds 33 and 34 surpassed that of both 25 and 30, suggesting an additive or even supra-additive effect of these modifications on binding specificity.

The predicted binding poses of compounds 30 and the more active R-isomer of 34 are depicted in Figure 4. In this binding mode, the ligands form a hydrogen bond with the Asn-698^6.45(B₂) residue, while the heteroaromatic core and the apolar aromatic p-chlorophenyl ring (H1 site in the pharmacophore) are close to the aromatic side chains of Tyr-810^6.44(B₁) and Tyr-697^6.44(B₂) residues, respectively. The acetamido and carboxyl groups (A/D site in the pharmacophore) are accommodated by the lower, intracellular side of the PAM binding site where interactions are possible either with the backbone or with the flexible Lys-792^5.62(B₁) and Lys-690^6.37(B₂) side chains. In addition, the methylthio groups (H3 site in the pharmacophore) are located in the same apolar site where one of the cyclopentyl groups of the GS39783 reference is accommodated according to the cryo-EM structure (PDB ID: 6UO8, see Figure 2, Figure 3, and Figure S4). Chirality at the carboxyl group has little effect on the activity, which can be explained with flexibility of the Lys-792^5.62(B₁) and Lys-690^6.37(B₂) side chains, since forming a salt bridge is possible for both (R)-34 and (S)-34 isomers (see Figure S5).

Figure 4.

Predicted binding pose for compound 30 (A) and (R)-34 (B); protein structure: PDB ID: 6UO8.

Finally, to elaborate further on the positive cooperativity exerted by GABA_B PAMs, a receptor binding assay for the orthosteric site was established by using [³H]baclofen. PAM ligands CMPPE and compound 34 enhanced [³H]baclofen binding (see Figure S7). Analysis using the ternary complex model²⁵ yielded alpha values of 5.5 and 5.2, respectively, indicating positive modulation of agonist binding by both compounds (see Table S2).

In summary, the latest cryo-EM results support the finding that the most relevant PAM binding site is located at the heterodimer interface of the transmembrane domain of B₁ and B₂ subunits. This binding site is common for all the compounds investigated in this study suggested by our novel competitive binding assay. Based on the recently published experimental structures containing either BHFF (1) or GS39783 (2) ligands at this site, a diverse set of GABA_B PAM-s were docked into the receptor, and based on the docking-based alignment, a common schematic pharmacophore model was created (Figure 3B), supporting a ligand-based model. Systematic modification of the selected CMPPE (3) starting structure and monitoring LLE led to three main findings: (i) carboxyl-group-containing substituents at the H2 and A/D site may be favored for the binding specificity, but due to the highly polar character, decrease the PAM activity of the ligands; (ii) lipophilic substituents at the H3 site can increase not only the PAM activity but also the LLE, which suggests higher binding specificity; (iii) these two types of modifications can be applied in a combination, whose effects were summed additively (or even supra-additively) based on our data. The effect of the systematic modifications is summarized in the corresponding pEC₅₀-clog D diagram (Figure 5). Following the optimization steps discussed above, this diagram traces the improvements achieved in LLE from 2.3 of CMPPE (3) to 5.8 of (R)-34.

Figure 5.

LLE diagram.

Our results suggest that compound 34 is a promising candidate offering the possibility to improve potency without drastically inflating the physicochemical properties in further optimizations. The unexpected discovery of the carboxyl group as an important element of the GABA_B PAM pharmacophore preceded the time when the experimental structures of PAM binding the GABA_B receptor were published. However, it can be perfectly explained based on the new protein structures, which clearly show the possibility of forming a salt bridge with the flexible side chains of Lys-792^5.62(B₁) and Lys-690^6.37(B₂).

Glossary

ABBREVIATIONS

PAM

positive allosteric modulator

NAM

negative allosteric modulators

GPCR

G-protein coupled receptor

GABA

γ-aminobutyric acid

MV

mean value

SD

corrected sample standard deviation

ND

not determined

SAR

structure–activity relationship

IFD

induced-fit docking

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.3c00560.

• Details of modeling, biological assays, and synthetic procedures (PDF)

Author Contributions

All authors have given approval to the final version of the manuscript.

The authors declare no competing financial interest.