Exploring voltage-gated sodium channel conformations and protein-protein interactions using AlphaFold2

Voltage-gated sodium (NaV) channels are vital regulators of electrical activity in excitable cells, playing critical roles in generating and propagating action potentials. Given their importance in physiology, NaV channels are key therapeutic targets for treating numerous conditions, yet developing subtype-selective drugs remains challenging due to the high sequence and structural conservation among NaV family members. Recent advances in cryo-electron microscopy have resolved nearly all human NaV channels, providing valuable insights into their structure and function. However, limitations persist in fully capturing the complex conformational states that underlie NaV channel gating and modulation. This study explores the capability of AlphaFold2 to sample multiple NaV channel conformations and assess AlphaFold Multimer’s accuracy in modeling interactions between the NaV α-subunit and its protein partners, including auxiliary β-subunits and calmodulin. We enhance conformational sampling to explore NaV channel conformations using a subsampled multiple sequence alignment approach and varying the number of recycles. Our results demonstrate that AlphaFold2 models multiple NaV channel conformations, including those observed in experimental structures, states that have not been described experimentally, and potential intermediate states. Furthermore, AlphaFold Multimer models NaV complexes with auxiliary β-subunits and calmodulin with high accuracy, and the presence of protein partners significantly alters the modeled conformational landscape of the NaV α-subunit. These findings highlight the potential of deep learning-based methods to expand our understanding of NaV channel structure, gating, and modulation, while also underscoring the limitations of predicted models that remain hypotheses until validated by experimental data.

Methods

Model generation
All models in this study were generated using colabfold-batch 1.5.0 (localcolabfold) (Mirdita et al., 2022) with a subsampled MSA (Monteiro da Silva et al., 2024). The execution was performed with the following options:

colabfold_batch --num-models 5 --model-type auto --msa-mode mmseqs2_uniref_env \

--num-seeds 20 \

--templates  --max-seq 256 --max-extra-seq 512 \

--num-recycle 6 --save-recycles nav17alphafull.fasta outfiles-6r-256-512-sr

The key aspect that makes this colabfold execution an MSA subsampled implementation is the setting of two parameters: --max-seq and --max-extra-seq. The --max-seq parameter defines the maximum number of sequences randomly selected from the original generated master MSA. The target sequence is always included. Then, the remaining sequences are clustered around the selected sequences using Hamming distance. Finally, the cluster centers (the --max-seq sequences chosen originally) are used alongside a sample from each cluster (as defined by the --max-extra-seq parameter) for the structure prediction network.

Previous studies have shown that adjusting these parameters, particularly by reducing their values, can increase the diversity of conformational sampling (del Alamo et al., 2022). In this study, we used values of 256 for --max-seq and 512 for --max-extra-seq, as these settings enhanced conformational diversity without compromising model accuracy in the original study (Monteiro da Silva et al., 2024). A total of 100 models were generated for each study case, using 6 recycles. Intermediate models were saved at each recycle step, resulting in 7 models per generated structure, ranging from recycle 0 to recycle 6, making a total of 700 models per test case (see Data Availability).

Model Analysis
All models were visually analyzed using UCSF ChimeraX (Goddard et al., 2018). Distances were automatically calculated with custom Python scripts utilizing the PyRosetta package (Chaudhury et al., 2010). All other analyses, calculations, and figure generation were conducted with custom Python scripts. Pore analysis of the channels was conducted using the MOLEonline website (Pravda et al., 2018).