Scalable Ligand-Receptor Binding Affinity Landscape: A Case Study with Ziconotide and Ca_V2.2

1. Introduction

The design of peptide-based drugs, particularly in the context of ion channel blockers such as ziconotide, has emerged as a promising therapeutic avenue for the treatment of chronic pain [1,2,3]. ziconotide, a selective blocker of the N-type Ca_V2.2 calcium channel, has been clinically validated for its potent analgesic effects, offering an alternative to traditional opioid therapies [4]. Nonetheless, limitations such as narrow therapeutic windows and specificity [1,5] have necessitated the development of next-generation analogues that retain efficacy while minimizing side effects. Central to these efforts is a deeper understanding of the molecular structural interactions that govern the binding affinity between ziconotide and Ca_V2.2 [6,7,8,9].

To date, drug discovery and design remains a complex multiparameter optimization challenge [10,11]. For instance, the optimization of ligand-receptor binding affinity has relied heavily on experimental approaches such as mutagenesis, crystallography, and biophysical assays [12]. However, these experimental methods are time-consuming, labor-intensive, and often limited in their ability to explore a wide range of analogues and binding conformations [13]. To address these limitations, computational approaches have become increasingly important, providing a high-throughput alternative to experimental workflows. In particular, the advent of in silico methods for generating binding affinity landscapes has opened new possibilities for the rational design of drugs, enabling the exploration of structural modifications with unprecedented efficiency and reasonable accuracy [14].

Here, this article presents a scalable in silico framework for generating and analyzing ligand-receptor binding affinity landscapes [15], with a specific focus on ziconotide and its receptor, Ca_V2.2. By systematically exploring the sequence space at the binding interface between ziconotide and Ca_V2.2, this article aims to build a scalable ziconotide-Ca_V2.2 binding affinity landscape for the design of ziconotide analogues with overall improved therapeutic efficacy and specificity.

2. Materials and Methods

Ziconotide is a synthetic version of a peptide found in the venom of a marine snail, Conus magus [4]. Specifically, it is a peptide consisting of 25 amino acids [5]. As of September 11, 2024, there is a total of three ziconotide-related structures in the Protein Data Bank (PDB) [16,17], as listed in Table 1.

Among the three, there are two ziconotide-Ca_V2.2 complex structures with PDB IDs 7MIX [6,7] and 7VFU [8,9]. Despite the differences in their resolutions, both ziconotide-Ca_V2.2 complex structures are determined experimentally with Cryo-EM [6,7,18] and deposited in Protein Data Bank with a standardized data format for biomolecular structures, making them both suitale to be used as starting structural templates for the scalable structural biophysical workflow [15,19,20,21]. Thus, this study here chooses PDB entry 7MIX (Figure 1) [6,7]) as a structural template to define and build a scalable ligand-receptor binding affinity (K_d) landscape [15] with ziconotide-Ca_V2.2 complex structure as an example.

With PDB entry 7MIX [6,7] (Table 1) as an initial input, subsequent structural modeling (Modeller) [19] and physics-based K_d calculations (Prodigy) [20,21] consists of an automated in silico generation of synthetic homology structural and K_d data, as illustrated in Figure 2 and described previously in detail [14]. Briefly, Modeller [19] was employed to build a total of 500 ($25\times 20$) homology structural models one site-specific missense mutation introduced to the amino acid sequence of ziconotide (PDB entry 7MIX [6,7]).

Afterwards, the binding affinities were calculated using Prodigy [20,21] for all structural models of ziconotide-Ca_V2.2 complex analogues. With PDB entry 7MIX [6,7] as template, all structural modeling [19] and physics-based K_d calculations [20,21] were repeated thirty times ($500\times 30$) on Wuxi Taihu Lake High Performance Computing platform.

3. Results

3.1. A Scalable Ziconotide-Ca_V2.2 Binding Affinity Landscape Based on Computational Structural Biophysics

With Modeller [19] and Prodigy [20,21], a set of structural physics-based calculations were performed for the native experimental ziconotide-Ca_V2.2 complex structure PDB entry 7MIX. As shown in Figure 3, the K_d between native ziconotide and Ca_V2.2 is 4.8 × 10^-8 M (vertical red line in Figure 3), while the K_d values ziconotide analogues and Ca_V2.2 possess a much wider distribution, ranging from 4.6 × 10^-6 M to 5.0 × 10^-9 M, according to the structural biophysics-based Prodigy [20,21] calculations of the 30 × 500 homology structural models of Ca_V2.2 bound to ziconotide analogues.

In short, this study starts from an experimental ziconotide-Ca_V2.2 complex structure (PDB entry 7MIX) to build a scalable ziconotide-Ca_V2.2 binding affinity landscape based on computational structural biophysics (Figure 3). This ziconotide-Ca_V2.2 binding affinity landscape is scalable because:

• the Modigy (Figure 1) workflow [14] is applicable to biomolecular structure databases such as PDB [16] and AFDB [22,23,24,25].
• the Modigy (Figure 1) workflow [14] introduced only one site-specific missense mutation to the sequence of ziconotide (PDB entry 7MIX), where the number could be larger, provided that the overall accuracy is reasonable for the synthetic structural and biophysical data [14].
• the ziconotide-Ca_V2.2 binding affinity landscape (Figure 3) includes only site-specific mutants of ziconotide, but not site-specific mutants of the receptor, i.e., Ca_V2.2, highlighting the use of this in silico workflow [14] in high-throughput generation of synthetic structural and biophysical data for other drug targets (GPCRs [26], ion channels [27], etc.) to train AI models for the discovery and design [28] of not just peptides, but also of small molecule compounds [29,30].
• method-wise, in addition to the structural modeling [19] and physics-based K_d calculations [20,21] employed here, this Modigy (Figure 1) workflow [14] is also able to integrate molecular dynamics simulations [19,31,32,33,34] to further enhance the accuracy of the structural biophysics-based K_d calculations [13,28,35] in drug discovery and design [10,11,12,36,37,38].

3.2. Designing Ziconotide Analogues with Over Two Orders of Magnitude Enhanced Ca_V2.2 Affinity

As described in the Materials and Methods section as above, the key feature of this Modigy workflow [14] lies in its scalability, allowing the generation of homology models for a wide array of site-specific point mutations. This capability is critical in scenarios where the computational design of therapeutic peptides demands extensive exploration of the mutation landscape to optimize therapeutic properties such as improved efficacy and specificity [39,40,41].

Here, in this article, with an automated in silico generation of synthetic structural and K_d data [14], a huge set of ziconotide analogues were designed with the scalable Modigy (Figure 1) workflow [14]. After a ranking of the ligand-receptor K_d values, five ziconotide analogues (Table 2) stood out, with binding affinity to Ca_V2.2 ranging from 1.4 × 10^-10 M to 1.9 × 10^-10 M, according to the structural biophysics-based Prodigy [20,21] calculations of homology structural models of Ca_V2.2 bound to ziconotide analogues with five site-specific mutations introduced to its amino acid sequence. Compared to the K_d between native ziconotide and Ca_V2.2 is 4.8 × 10^-8 M (vertical red line in Figure 3), the binding affinities between the five ziconotide analogues (Table 2) and Ca_V2.2 are at least increased over two orders of magnitude of ligand-receptor binding affinity.

4. Conclusion and Discussion

To sum up, this study reports a scalable ziconotide-Ca_V2.2 binding affinity landscape based on computational structural biophysics, with a structural biophysical workflow [14] integrating structural modeling [19] and physics-based K_d calculations [20,21], tailored for computational drug design and discovery [15]. In addition, the scalable workflow [14] described here presents a technically feasible method for generating synthetic structural and biophysical data, which is of use for enhancing the specificity and efficacy of therapeutic peptides [28,42].

Overall, this scalable synthetic structural and biophysics data serve two purposes: (1), this scalable Modigy (Figure 1) workflow [14] creates a scalable antigen-antibody binding affinity landscape, which acts like a map to guide the design of peptides with improved efficacy and specificity [43,44]; (2), this scalable Modigy (Figure 1) workflow [14] generates useful training data [45,46] for AI-driven drug design (AIDD, Figure 1) models [10,11,28] towards the design of both biomolecule (e.g., peptide) [43,44] and small molecule compounds (Figure 1) with improved efficacy and specificity.

4.1. Implications for Peptide Design and Future Directions

The findings of this study underscore the importance of a precise understanding of ligand-receptor binding dynamics in the structural context of therapeutic peptide design [47]. As ziconotide and its analogues continue to be explored for pain management, the ability to fine-tune their binding affinity to Ca_V2.2 offers significant promise for improving their safety and efficacy profiles, based on the scalable ziconotide-Ca_V2.2 binding affinity landscape based on computational structural biophysics [14]. This scalable framework not only addresses current limitations in experimental throughput but also paves the way for more sophisticated computational peptide design and discovery pipelines.

Moreover, this work contributes also to the broader application of AI-driven drug discovery platforms. While the current framework relies on structural biophysics and binding affinity calculations [20,21], the integration of machine learning models trained on synthetic binding affinity data could accelerate the identification and discovery of optimal ligand-receptor pairs. Furthermore, expanding this approach to other receptor targets, particularly those in complex neural and pain-signaling pathways, could lead to breakthroughs in peptide and protein drug design and discovery in future.