A one-dimensional semaglutide-GLP-1R-based mini static GIBAC

1. Introduction

1.1. Intermolecular Binding Affinity in Drug Discovery and Design

Intermolecular interactions are the fabrics underlying almost all processes in living organisms [1,2,3,4,5,6,7,8], where two cornerstone concepts, intermolecular binding affinity (K_d) and free binding energy (ΔG), have long been established to physically describe the strengths of biomolecular interactions [9,10,11,12,13,14,15]. Intermolecular binding affinity, e.g., drug-target binding affinity (K_d and ΔG), is an essential parameter in drug discovery and design to ensure efficacy and specificity [6,11,16,17,18,19,20,21]. First, high binding affinity typically correlates with increased efficacy, i.e., the drug can effectively modulate the target’s activity at lower concentrations [22,23,24,25,26,27,28]. Second, a structural biophysical understanding of the intermolecular binding affinity helps in optimizing the potency and selectivity of drug candidates, reducing off-target effects and adverse reactions [29,30,31,32,33,34].

In August of 2022, therefore, the concept of a general intermolecular binding affinity calculator (GIBAC) was for the first time coined, proposed and defined as $K_{\mathrm{d}}=f(molecules,envPara)$ [35]. Last October, GIBAC was for the first time updated, including its inception, definition, construction, practical applications, technical challenges and limitations and future directions [36]. With the conceptual and practical framework of GIBAC in place [36], this article puts forward a hypothesis that a real GIBAC technically makes sense and is feasible, and aims to test this hypothesis through the construction of a prototype of a real GIBAC [4,35,36].

1.2. Building a GIBAC Prototype with Semaglutide as an Example

In principle, any biomolecule inside Protein Data Bank [37,38,39,40,41] is fit to be used to build a prototype of a real GIBAC [4,35,36]. Here, this article chooses semaglutide due to the availability of abundant experimental data and clinical relevance of semaglutide [42,43]. First, semaglutide is a synthetic long-acting analogue of glucagon-like peptide-1 (GLP-1), which has garnered significant attention for its efficacy in treating type 2 diabetes and obesity [44,45]. Structurally, its direct binding and interaction with the GLP-1 receptor involves intricate mechanisms, where the complex structure has already been experimentally determined with Cryo-EM or X-ray diffraction (Table 1 and Table 2, Figure 1) [42]. From a chemical structure point of view, semaglutide is a peptide-based molecule, including a peptide backbone consisting of 31 amino acids (a modified version of human GLP-1, 7-37), a substitution of alanine with 2-aminoisobutyric acid (an unnatural amino acid [46,47,48]) at position 8 to resist enzyme degradation, and an addition of a C-18 fatty acid chain attached via a spacer (γ-Glu-2x-OEG) at lysine 26 for stronger albumin binding [29,49].

In short, semaglutide is neither small molecule compound, nor monoclonal antibody, nor ADC/PDC, nor an entirely peptide-based drug, making it a reasonable choice as an example for the construction of a GIBAC prototype, i.e., a one-dimensional semaglutide-GLP-1R-based mini static GIBAC, abbreviated below as semaGIBAC and described in Equation 1:

$$K_d=f(ABcomplex,seqA,seqB,center,n,r,T,mList)$$

(1)

where $ABcomplex$ represents the experimental complex structure of $molA$ and $molB$, $seqA$ and $seqB$ represent the amino acid sequences of $molA$ and $molB$, respectively, $center$=$molA$, or $molB$ or both, n represents the number of missense mutations introduced into the sequence of $center$, r represents the number of structural models built by Modeller [50], T represents temperature [51,52,53,54], and $mList$ represents the site-specific mutations introduced into the backbone of semaglutide (PDB entry 4ZGM [42], Table 2).

2. Materials and Methods

To build a prototype of a real GIBAC, as mentioned above, this article chooses semaglutide due partly to the availability of abundant experimental data of semaglutide [42,43]. Among the three experimental complex structures of semaglutide and GLP-1R (Table 2), there is only one experimental structure (X-ray diffraction) of the semaglutide backbone in complex with the extracellular domain of GLP-1R (Figure 1, PDB ID: 4ZGM [42]).

According to PDB entry 4ZGM [42] (Table 2), the sequences of the two chains of semaglutide backbone and GLP-1R are listed in italics in fasta format as below,

>4ZGM_1_Chain A (Figure 1)

RPQGATVSLWETVQKWREYRRQCQRSLTEDPPPATDLFCNRTFDEYACWPDGEPGSFVNVSCPWYLPWASSVPQGHVYRFCTAEGLWLQKDNSSLPWRDLSECEESKRGERSSPEEQLLFLY

>4ZGM_2_Chain B_a modified version of human GLP-1 (7-37) (Figure 1)

HAEGTFTSDVSSYLEGQAAKEFIAWLVRGRG

In 2021, a Val27-Arg28 exchange (Table 3) was for the first time introduced into the backbone of semaglutide to strengthen the semaglutide-GLP-1R binding affinity to ∼ one-third of the K_d between native semaglutide and GLP-1R [6,55,56].

First, with PDB entry 4ZGM [42] (Table 2) in place, Modeller [50] was employed to build 10000 structural models with 100% homology to PDB ID: 4ZGM [42], the binding affinities between semaglutide and GLP-1R were calculated using Prodigy [57,58] 10000 times [59]. Second, with PDB entry 4ZGM [42] (Table 2) in place as an initial input, the process of the construction of a prototype GIBAC (semaGIBAC, Equation 1) subsequently consists of an automated in silico generation of synthetic structural and K_d data, as illustrated in Figure 2 and described previously in detail [60]. Briefly, Modeller [50] was employed to build a total of 11200 ($\frac{28!}{1!\left(27\right)!}\times {20}^1\times 20$) homology structural models with 95.42% ($27/28$) homology to PDB ID: 4ZGM [42]. Afterwards, the binding affinities between semaglutide analogues and GLP-1R were calculated using Prodigy [57,58] for 11200 times [59].

3. Results

With PDB entry 4ZGM [42] (Table 2) as an initial input and an automated in silico generation of synthetic structural and K_d data [60], this article for the first time puts forward a prototype GIBAC, i.e., semaGIBAC, for which everything is included in Table S1 of supplementary file semaGIBAC.pdf:

$$K_{\mathrm{d}}=f(SGcomplex,seqS,seqG,sema,n,r,T,mList)$$

(2)

where $SGcomplex$ represents PDB entry 4ZGM [42] (Table 2), $seqS$ and $seqG$ represent sequences of semaglutide backbone ($molA$) and GLP-1R ($molB$), respectively, $center$=$semaglutide$, n =1, r =20 [50], T = ${37}^{\circ }$C. Of further note, here, $center$=$semaglutide$ indicates that site-specific missense mutation is introduced only into the backbone of semaglutide, but not into GLP-1R (PDB entry 4ZGM [42], Table 2), making semaGIBAC a semaglutide-centered GIBAC prototype [35,36].

In Figure 3 and Table 4, for PDB entry 4ZGM [42] (Table 2), most of the K_d values are located between 2.5 × 10^-6 M and 4.0 × 10^-6 M, with an average at 3.278 × 10^-6 M, which is rather close to the one K_d (3.4 × 10^-6 M) as reported previously [6,59]. In Figure 3 and Table 4, for the 11200 homology structural models (semaGIBAC), most of the K_d values are located between 2.0 × 10^-6 M and 4.0 × 10^-6 M, with minimum and maximum values at 4.1 × 10^-7 M and 9.4 × 10^-6 M, respectively. From Figure 3 and Table 4, it can be seen that the the range of the K_d values gets wider and wider as n (Equations 1 and 2) increases from zero to one and to two.

As described above, the semaGIBAC here is built to be a one-dimensional semaglutide-GLP-1R-based semaglutide-centered mini static GIBAC [35,36]:

• semaGIBAC is a one-dimensional GIBAC, i.e., $n=1$ (Equations 1 and 2).
• semaGIBAC is a semaglutide-GLP-1R-based GIBAC, i.e., PDB entry 4ZGM [42] (Table 2) is the initial input before the construction of semaGIBAC.
• semaGIBAC is a semaglutide-centered GIBAC, i.e., missense mutation is introduced only into the backbone of semaglutide (PDB entry 4ZGM [42]).
• semaGIBAC is a mini GIBAC, i.e., semaGIBAC is able to calculate intermolecular K_d only between GLP-1R and semaglutide (analogues) with one mutation.
• semaGIBAC is a static GIBAC, i.e., semaGIBAC does take conformational dynamics [61,62] into account.

Last October, GIBAC was for the first time updated, including its inception, definition, construction, practical applications, technical challenges and limitations and future directions, including in particular a set of criteria as listed below [35,36]:

• a real GIBAC [4,35,36] needs to take site-specific mutations into account;
• structural biophysics is inextricably linked to and at the core of a real GIBAC [4,35,36], e.g., accurate structural information is necessary [63,64,65];
• for a real GIBAC [4,35,36], a variety of intermolecular K_d-relevant factors need to be taken into account, such as temperature, pH [66,67,68], side chain pKa of protein [51,58,69,70,71,72], ionic strength and buffer conditions [54,58,73,74,75,76], etc;
• a real GIBAC [4,35,36] requires an accurate general forcefield for all atoms [46];
• a real GIBAC [4,35,36] requires a universal linear string/graph-based notation system for all molecular types and drug modalities [61,77];
• a real GIBAC is able to be used the other way around as a search engine for drug discovery & design [4,35,36].

Nonetheless, this article puts forward just only a prototype (i.e., semaGIBAC), instead of a real GIBAC. Specifically,

• semaGIBAC does take one missense mutation into account;
• structural biophysics is at the core of semaGIBAC;
• semaGIBAC takes only some K_d-relevant factors into account, e.g., temperature.
• semaGIBAC does not require an accurate general forcefield for all atoms or a universal linear string/graph-based notation system, as it does not take 8Aib or C-18 fatty acid chain into account;

4. Conclusion

For the first time, this article puts forward a prototype GIBAC, i.e., semaGIBAC, to support the hypothesis that a real GIBAC is technically feasible in practice and of practical use for drug discovery and design:

• semaGIBAC is able to calculate with 95.42% ($27/28$) accuracy the K_d between GLP-1R and any semaglutide analogue with one site-specific missense mutation introduced into its backbone.
• semaGIBAC is able to be used the other way around as a search engine for drug design, i.e., for drug efficacy, semaGIBAC is able to generate a list of semaglutide analogue with one site-specific missense mutation introduced into its backbone, and rank them according to their average K_d (Table 4) values in the range of the minimum and the maximum values of semaGIBAC as included in Table 4, while for drug safety, semaGIBAC is unable to generate a K_d-ranked list of semaglutide analogues with suppressed off-target effects [78,79].
• the construction of semaGIBAC consists of a set of in silico steps of structural and biophysical data generation towards a paradigm shift in precise drug discovery and design, leading to the identification of a promising semaglutide analogue with a K_d of 3.0 × 10^-8 M, in contrast to the K_d of 3.278 × 10^-6 M for native semaglutide and GLP-1R (PDB entry 4ZGM, Table 2 and Table 4) [42].

5. Discussion

5.1. In Silico Generation of Structural and Biophysical Data with Reasonable Accuracy: Expanding Horizons in Precise Drug Discovery and Design

The past three years saw a big step forward in the use of artificial intelligence (AI) in structural biology for protein structure prediction [63,80,81,82,83,84,85], leading to the generation of computational structural data such as AlphaFold database [63,81,82,83,84]. Nonetheless, to train useful AI models for precise drug discovery and design, a huge number of data is needed with reasonable accuracy, buth experimental and synthetic, both structural and biophysical (K_d and G), where a variety of tools are needed, such as molecular docking tools [86,87,88,89], molecular dynamics simulations tools [69,90], side chain placement and energy minimization algorithms [91] to incorporate structural arrangement information of post-translational modifications (PTMs) [92,93,94], post-expression modifications (PEMs) [6,74] into currently available structural models.

In this regard, this article puts forward a set of in silico steps of structural and biophysical data generation towards a paradigm shift in precise drug discovery and design [59,60]. Take semaglutide for instance, a five-dimensional semaGIBAC requires a total of 314496000000 (Table 5) homology structural models with 82.14% ($23/28$) homology to PDB ID: 4ZGM [42] to be built by Modeller [50], and subsequently a total of 314496000000 (Table 5) times of Prodigy-based [57,58] calculations of the binding affinities between semaglutide analogues and GLP-1R. Take MoleculeX (a protein consisting of 100 amino acids) as another example, the number soars from 314496000000 to 240920064000000 (Table 5).

In short, to build a real GIBAC using an entirely structural biophysics-based approach requires an exhaustive exploration of the entire molecular space (Figure 4) [61,77]. In practice, however, this astranomical task is computationally impossible, which explains partly why this article puts forward just only a structural biophysics-based prototype (i.e., semaGIBAC), instead of a real GIBAC. Therefore, this article here again proposes an AI-based open strategy [95] to make it possible and conceivable to generate a vast amount of data to train a real GIBAC with reasonable accuracy and precision, as openness in data acquisition and generation, and AI algorithms is essential for promoting transparency, reproducibility, and collaboration within the community of drug discovery & design, and for facilitating continued improvement of the performance (accuracy, precision and efficiency) of GIBAC in precise drug discovery & design [35,36] in future.

As mentioned above, this article puts forward a set of in silico steps of structural and biophysical data generation towards a paradigm shift in precise drug discovery and design [59,60]. Specifically, here, the generation of synthetic structural and K_d data is akin to the distribution of the electron cloud of a hydrogen atom (Figure 4), where $Radius$ is the distance between the electron (synthetic data) and the proton (i.e., nucleus, experimental data) of the hydrogen atom. With respect to semaGIBAC, PDB entry 4ZGM (Table 2) is the nucleus, while Table S1 of supplementary file semaGIBAC.pdf is one layer of the electron cloud (Figure 4), while semaGIBAC itself is depicted in Figure 4 as one hydrogen atom. In principle, any biomolecule inside Protein Data Bank [37,39], i.e., experimental structural data, is able to be used as a hydrogen nucleus (Figure 4), around which there is an electron cloud representing a vast set of synthetic structural and biophysical data, while the white region of Figure 4 represents structurally and biophysically uncharted territories, highlighting an astranomical task which is computationally impossible, and calling for an AI-based open approach for the construction of a real GIBAC [35,36].

5.2. Designing Semaglutide Analogues with Elevated Binding Affinity and Efficacy through Continued Exploration of the Uncharted Molecular Space of Semaglutide and GLP-1R

The development of semaglutide analogues with increased GLP-1R binding affinity holds significant clinical relevance, offering the potential for enhanced glucose control, weight loss, and cardiovascular benefits in patients with type 2 diabetes and obesity [42,96,97]. Thanks to the continued development of experimental structural biology and the half-a-century old Protein Data Bank (PDB) [37,38,39,40,98], a comprehensive structural biophysical analysis becomes possible [99,100] for specific ligand-receptor complex structures deposited in PDB, such that our understanding of the structural and biophysical basis of their interfacial stability is able to help us modify the binding affinity of certain drug target and its interacting partners [101,102,103].

With semaglutide as an example here [59], one particular analogue (supplementary file semx.pdb, Table 6) stood out with four missense mutations (G13B_A, I20B_Q, L23B_R and V24B_N) through computational structural modeling and biophysics-based rational design, with a Prodigy-calculated K_d as low as 3.0 × 10^-8 M at 37 ^∘C (Table 6), while the K_d is 3.278 × 10^-6 M (Table 4) for the binding of native semaglutide’s backbone to GLP-1 at 37 ^∘C. Technically, the structural biophysics-based rational design of this semaglutide analogue (supplementary file semx.pdb) is a tiny part of the task of the construction of a four-dimensional semaGIBAC, which requires a total of 3276000000 (Table 5) homology structural models with 85.71% ($24/28$) homology to PDB ID: 4ZGM [42] to be built by Modeller [50], and subsequently a total of 3276000000 (Table 5) times of Prodigy-based [57,58] calculations of the binding affinities between semaglutide analogues and GLP-1R. Given the size of the computational task, this article only puts forward one-dimensional semaGIBAC as a GIBAC prototype, making this semaglutide analogue (supplementary file semx.pdb look like a tip of the iceberg floating on the ocean (Figure 4) of a four-dimensional sequence space of semaglutide and GLP-1R (PDB entry 4ZGM [42], Table 2).

Of further note, the K_d of this semaglutide analogue (supplementary file semx.pdb is only a result of the structural biophysics-based calculation of Prodigy [57,58]. As a result, its real K_d and efficacy (in both vitro and vivo) need to be experimentally tested in wet-labs (Figure 5), as part of the drug discovery and design process in the early stage of drug R&D, which to date is still a lengthy, costly, difficult and inefficient yet pivotal process [104]. Given this, this article calls again for the construction of a real GIBAC [35,36] with adequate accuracy, precision and efficiency towards a paradigm shift [105] of precise drug discovery & design, until a real GIBAC comes into being and pushing forward the continued development of the industry [106,107,108].

6. Ethical Statement

No ethical approval is required.

7. Declaration of Generative AI and AI-Assisted Technologies in the Writing Process

During the preparation of this work, the author used OpenAI’s ChatGPT in order to improve the readability of the manuscript, and to make it as concise and short as possible. After using this tool, the author reviewed and edited the content as needed and takes full responsibility for the content of the publication.