Crowdsourced Ligand Design for Selective Inhibition of the KNa1.2 (Slick) Potassium Channel implicated in Developmental Epileptic Encephalopathies

ABSTRACT

Developmental and epileptic encephalopathies (DEEs) and channelopathies associated with pathogenic variants of KCNT2 represent severe neurological disorders for which targeted therapies remain limited. The sodium-activated potassium channel encoded by KCNT2, KNa1.2 (Slick) regulates neuron excitability, and both gain- and loss-of-function variants have been linked to early onset, treatment-resistant epilepsy. Current pharmacological options such as quinidine are non-specific and produce variable clinical benefits, creating a clear need for more specific inhibitors. This project pursued a hybrid small molecule approach for discovering novel candidates for these specific inhibitors. REvoLd, an evolutionary algorithm, generated scaffold designs that served as starting points for a 12-round Drugit campaign. Drugit, is a citizen science platform that allows players with no prior background in biochemistry to design small molecule drug candidates from the comfort of their own home. Players optimized scaffold molecules in the predicted sodium-binding pocket of a KNa1.2 subunit. A total of 6,015 player-designed ligands were generated and after filtering for accessibility, 3,271 remained for analysis. Following an adjustment of in game incentives, later design rounds produced significantly lower predicted binding energies and concentrated in the top 50 candidate inhibitors. These synthetically accessible (via large make on demand combinatorial libraries) represent strong candidates for downstream validation in vitro. More broadly, this work demonstrates the power of combining computation with citizen science from drug discovery.

INTRODUCTION.

Sodium-activated potassium (KNa) channels play an essential role in regulating neuron electrical activity. These channels respond to increases in intracellular Na⁺ and modulate neuronal firing frequency by contributing to slow afterhyperpolarization after repetitive action potential activity [1-2]. Two KNa channels are expressed in humans: KNa1.1 (Slack, KCNT1) and KNa1.2 (Slick, KCNT2) [1-3]. KNa1.2 shares approximately 74% sequence similarity with KNa1.1 and primarily diverges in the distal C-terminal region [2]. Each of the four subunits that compose a KNa1.2 channel contain six transmembrane helicases and an intracellular C-terminus along with two RCK domains where a sodium ion binding site is located [4-5]. The channel is activated by Na⁺ and contributes to the regulation of neuron excitability [6].

Pathogenic variants in KCNT2 are increasingly recognized as a cause of severe developmental epileptic encephalopathies (DEEs) [4, 7-8]. As defined by the International League Against Epilepsy, DEEs are a group of disorders where developmental factors and frequent epileptic activity significantly impact brain development and functional capabilities. These conditions represent some of the most severe forms of childhood epilepsy, with a collective incidence of approximately 1 in 340 children. Significantly rarer (n=26 documented) pathogenic KCNT2 variants are responsible for severe DEEs that manifest in West syndrome, Lenoxx-Gastaut syndrome, and Epilepsy of Infancy with Migrating Focal Seizures (EIMFS). Affected individuals often suffer from profound psychomotor arrest, severe intellectual disability, and hypotonia [7-8]. In contrast to KCNT1, in which disease variants are predominantly gain-of-function (GoF), KCNT2 variants appear more evenly distributed between gain- and loss- of function mechanisms with GoF producing more severe complications [7-8]. This difference in spite of the 74% sequence similarity suggests the potential for channel specific therapies. But currently, specific therapies are limited as while quinidine has shown benefit in some cases, its non-selective activity complicates its therapeutic use [3].

The absence of selective inhibitors limits both therapeutic strategy and a more thorough understanding of KNa1.2 function. There is a clear incentive for the discovery of novel small-molecule binders that could attach to the Na⁺ binding site of KNa1.2 to inhibit the channel. Despite the presence of loss of function (LoF) KCNT2 variants, the increased impact of GoF variants on patients encourages an emphasis on designing inhibitors of the channel [8].

To navigate the vast chemical space of potential KNa1.2 inhibitors, we utilized Drugit, a special mode within the Foldit citizen-science platform. Foldit is a gamified environment where individuals, who often lack formal biochemical training, solve complex structural problems such as protein folding and drug design after completing a series of interactive training levels [9]. This crowdsourcing approach offers two primary advantages. First, it provides the necessary power to generate a high volume of unique designs. Secondly, while purely computational machine learning (ML) models are efficient at local optimizations, they often rely on gradient minimization techniques that can be trapped in local energy minima. In contrast, human intuition and creativity allow players to explore novel structure modifications that automated systems might overlook. This approach has previously yielded promising small molecule binders such as those for the SARS-CoV-2 Nsp13 CACHE competition and non-peptide VHL inhibitor designs, outperforming other purely ML or computational approaches [9-10]. Because of Drugit’s demonstrated success in the past, we aim to use it for small-molecule drug discovery with a target of the sodium binding site of the KNa1.2 channel.

To complement this citizen science strategy, we incorporate REvoLd (Rosetta Evolutionary Ligand) to efficiently screen ultra-large make-on-demand combinatorial libraries [11]. REvoLd is an evolutionary algorithm that first randomly samples Enamine’s large make on demand combinatorial library. It then mutates these samples, based on a variety of parameters, modifies and then selects optimal designs for propagation to further generations. Integrating these approaches presents an opportunity to identify novel and synthetically accessible small molecules targeting KNa1.2.

This study applied a combined computational and citizen science drug design pipeline to identify candidate inhibitors for the predicted sodium binding pocket of KNa1.2. We hypothesized that our combined approach would yield structurally diverse candidate ligands suited for downstream testing and analysis.

MATERIALS AND METHODS.

REvoLd Screening.

Initial ligand scaffolds were obtained from a virtual screening performed by collaborators using REvoLd which applies an evolutionary algorithm to explore combinatorial make-on-demand chemical libraries [10]. Docking was performed against the sodium ion binding pocket of an AlphaFold2 model of a KNa1.2 subunit. This model was selected due to AlphaFold2’s high accuracy in predicting protein structures in absence of experimental data, the particular Na⁺ binding site used as a target exhibited a pLDDT > 90%, indicating very high local model confidence. Moreover, the predicted model demonstrates structural similarity to a experimental KNa1.1 crystal structure. The top 100 ligands ranked by Rosetta energy (lower being better) were selected for further use in the Drugit campaign.

Drugit Design Campaign.

A 12 round Drugit series named KCNT2 was opened. Players modified starting ligands to optimize binding within the fixed receptor pocket. Players had 1 week to finalize designs for each round until the next round opened with any modifications that may have been needed.

As a scaffold, players received a starting ligand that varied from round to round. Starting ligands were selected from the 100 resulting compounds from the REvoLd screening through visual analysis as shown in figure 1.

Drugit tools allowed for ligand modification and the modification of a small amount of side chains in the binding pocket that upon visual analysis would be elastic during the binding of Na⁺ to the pocket. Players also had access to Enamine’s combinatorial library and could directly insert compounds present in the library. To encourage use of the library, a bonus was awarded to players for design presents in the database. This bonus was doubled in rounds 11-12 to more heavily incentivize the use of the compound library. Outside of the compound library, players were also awarded a variety of bonuses based on drug-likeness as per Lipinski’s rule of 5.

Filtering and Statistical Analysis.

6,015 structures were collected from players. After restricting Enamine-available ligands to make obtaining compounds feasible, 3,271 ligands remained. Energy distributions were examined using visualizations and Shapiro-Wilk tests. Group differences across puzzle rounds were tested using ANOVA, post-hoc Tukey, and Kruskal-Wallis analyses. The 50 ligands with the lowest predicted Rosetta Energies were selected.

RESULTS.

All data handling and filtering were performed using Python. Data was pulled and de-identified before being handed off for analysis. All data analysis was performed in R.

Binding Pocket and Initial Scaffolds.

Distinct binding orientations were observed across starting REvoLd ligands, suggesting multiple viable interaction modes within the sodium ion binding pocket (Figure 1). Multiple high scoring binding modes were also observed among the top 50 ligands, so Drugit players were also able to capture this variety.

Distribution of Game Score and Energies.

Distributions of game score, Rosetta energy, and ligand energy showed heavy-tailed behavior (Figure 2). Upon visual analysis, outliers were associated with steric clashes or highly favorable predicted binding.

Effect of Incentive Modification.

Mean Rosetta energy differed significantly across puzzle rounds (Figure 3, ANOVA, p < 0.005). The compound-library bonus increases in rounds 11-12 produced a substantial improvement (Tukey, p < 0.001; Krusal-Wallis, p < 0.001). The nature of the puzzle rounds being iterative, and players being able to physically recreate previous designs, was also a contributing factor in concentrating high-scoring designs in later rounds.

Relationship Between Bonuses and Rosetta Energy.

To determine whether Rosetta’s energy function inherently captured the effects of our ligand filtering bonuses we analyzed the pairwise correlations between each bonus term and total energy across all (n = 3,271) (Figure 4).

DISCUSSION.

This project demonstrates that combining REvoLd with Foldit crowdsourced designs can generate promising small-molecule inhibitors for KNa1.2. The incentive modifications strongly improved chemical relevance, highlighting the importance of reward design in citizen-science drug discovery.

The selected 50 ligands represent strong candidates for in vitro validation, including potency and selectivity testing against KNa1.1. Given the clinical heterogeneity of KCNT2-related DEEs, selective inhibitors may be required to support future precision therapies.

Correlation patterns indicate partial alignment between Rosetta scoring and drug-likeness metrics. Many though were not captured at all by Rosetta’s energy function and in turn required external incentivization. While giving players bonuses in a Drugit setting may remedy the shortcoming of the Rosetta energy function the same approach is not feasible in a fully computational workflow. In turn, there is a clear need for an improved scoring function for assessing small molecule drug-likeness and binding affinity.

This study has one major limitation: the theoretical nature of the binding target. A predicted AlphaFold2 model was used for both the initial screening and subsequent design refinement periods, and while the model shows extreme confidence in the targeted regions and aligns c loosely with the crystal structure of the highly similar KNa1.1, docking into a predicted model rather than an experimental structure remains a standard computational limitation. Slight variations in backbone positioning could impact the accuracy of the predicted binding energies. Consequently, the selected 50 ligands represent strong candidates that require in-vitro validation to confirm their potency, selectivity, and efficacy against the physical KNa1.2 channel.

CONCLUSION.

This work identified 50 synthetically accessible KNa1.2 inhibitor candidates using a hybrid evolutionary and crowd-sourced design pipeline. If validated, these compounds may guide future therapeutic development for DEEs caused by KCNT2 variants and demonstrate the power of citizen-science driven drug discovery.

ACKNOWLEDGMENTS.

Many thanks are extended to Dr. Popp from the School for Science and Math at Vanderbilt for her mentorship and the Meiler Lab for hosting this research.

REFERENCES

1. A. Bhattacharjee, W. J. Joiner, M. Wu, Y. Yang, F. J. Sigworth, L. K. Kaczmarek, Slick (Slo2.1), a Rapidly-Gating Sodium-Activated Potassium Channel Inhibited by ATP. J. Neurosci. 23, 11681–11691 (2003).
2. L. K. Kaczmarek, Slack, Slick, and Sodium-Activated Potassium Channels. Int. Sch. Res. Not. 2013, 354262 (2013).
3. Y. Tian, et al., Coupling of Slack and NaV1.6 sensitizes Slack to quinidine blockade and guides anti-seizure strategy development. eLife 12 (2024).
4. P. Ambrosino, et al., De novo gain-of-function variants in KCNT2 as a novel cause of developmental and epileptic encephalopathy. Ann. Neurol. 83, 1198–1204 (2018).
5. X. Mao, et al., The Epilepsy of Infancy With Migrating Focal Seizures: Identification of de novo Mutations of the KCNT2 Gene That Exert Inhibitory Effects on the Corresponding Heteromeric KNa1.1/KNa1.2 Potassium Channel. Front. Cell. Neurosci. 14, 1 (2020).
6. S. Rizzi, H.-G. Knaus, C. Schwarzer, Differential distribution of the sodium-activated potassium channels slick and slack in mouse brain. J. Comp. Neurol. 524, 2093–2116 (2016).
7. M. C. Cioclu, et al., KCNT2-Related Disorders: Phenotypes, Functional, and Pharmacological Properties. Ann. Neurol. 94, 332–349 (2023).
8. F. Cui, T. Wulan, Q. Zhang, V. W. Zhang, Y. Jiang, Identification of a novel KCNT2 variant in a family with developmental and epileptic encephalopathies: a case report and literature review. Front. Genet. 15, 1371282 (2024).
9. T. Scott, et al., Drugit: Crowd-sourcing molecular design of non-peptidic VHL binders. Nat. Com. 16, 3548 (2025).
10. O. Herasymenko, et al., CACHE Challenge #2: Targeting the RNA Site of the SARS-CoV-2 Helicase Nsp13. J. Chem. Inf. Model. 65, 6884–6898 (2025).
11. P. Eisenhuth, F. Liessmann, R. Moretti, J. Meiler, REvoLd: Ultra-Large Library Screening with an Evolutionary Algorithm in Rosetta. arXiv:2404.17329 [Preprint] (2024).

---

---

Posted by buchanle on Monday, May 18, 2026 in May 2026.

Tags: Foldit, KCNT2, KNa1.2, Ligand, REvoLd