Introduction

Potassium channels, including voltage-gated (VGKC, Kv), calcium-activated (KCa), inward-rectifier (Kir, IRK), and tandem pore domain (K2P) types, are integral transmembrane (TM) proteins with 2TM, 4TM, and 6TM alpha-helices. They are one of the most diverse and widely distributed classes of ion channels in virtually all organisms and cells. Potassium ion channels switch between closed and opened conformations and regulate the K+ flow across cell membranes. This ability to selectively pass potassium ions through the pore is crucial for their function. For example, potassium channel blockers can physically obstruct the pathway from the intracellular solution into the pore, altering the function of the channel1. Hence, the pore plays an important role in both normal physiological functions and pathophysiological processes2,3. They are crucial for muscle contraction, nerve impulse propagation, cellular activation, and the secretion of biologically active molecules. Additionally, potassium ion channels serve as therapeutic targets for various brain disorders, including brain and spinal cord ischemia, epilepsy, migraine, multiple sclerosis, pain, stroke, Alzheimer’s disease, Parkinson’s disease, and schizophrenia2,3.

In the past two decades, the involvement of potassium ion channels in cancer metabolism, growth, and metastasis has become evident especially for KCNA1, KCNA3, KCNA5, KCNC4, KCND1, KCNH2, KCNH5, KCNJ3, KCNJ8, KCNJ10, KCNJ11, KCNJ12, KCNK2, KCNK5, KCNK9, KCNMA1, KCNN3, and KCNN4 (Table 1)4,5,6,7,8,9. Some potassium ion channels have been found to be involved in various cancer-related processes, for example, cell cycle events, cell proliferation, evasion of apoptosis, sustained angiogenesis, and metastasis7,8,9,10,11. Others are mainly involved either in cell proliferation12,13,14, or cell migration15,16,17,18,19,20. Recently, potassium ion channels have emerged as promising targets for the treatment of different types of cancer21,22,23,24,25,26,27. Thus, it is possible to combine all recently available knowledge of potassium ion channels to formulate medical cocktails to attack cancer cells from many angles in cancer therapies. Structural bioinformatics studies of these potassium channels may yield significant clinical benefits for cancer patients.

Table 1 Eighteen potassium channels involved in various cancers.
Full size table

Several molecular structures of human potassium ion channels have been experimentally (X-ray or Cryo-EM) determined, including KCNN4 (protein Kca3.1) (PDB: 6CNM)28, KCNK2 (K2P2.1) (PDB: 4TWK), KCNA3 (Kv1.3) (PDB: 7EJ1)29, KCNMA1 (Kca1.1) (PDB: 6V22)30, KCNH2 (Kv11.1) (PDB: 5VA2)31, KCNJ11 (Kir6.2) (PDB: 6C3O)32. However, structures of other potassium channels haven’t yet been determined. Since these potassium channels form 2TM, 4TM and 6TM helices, the structural determination of these transporters requires systematic detergent screens before protein purification can be carried out, making the process quite challenging33.

AlphaFold234,35 and RoseTTAFold36,37 were introduced in July 2021 as revolutionary artificial intelligence (AI) computational tools for the accurate protein structure predictions. Since then, both tools have already made a significant impact on our understanding of the molecular structure of numerous proteins that were previously inaccessible. Now we are able to get structural models to perform computational analysis and design of potassium ion channels. This way we can computationally engineer the native channels to make them more suitable for the experimental verification and therapeutic purposes, by, for example, transforming them into hydrophilic analogies (QTY variants).

We previously applied the QTY (Glutamine, Threonine, Tyrosine) code that replaces TM hydrophobic amino acids with structurally similar hydrophilic amino acids (L → Q, I/V → T, F → Y) to design several detergent-free transmembrane (TM) protein chemokine receptors and cytokine receptors for various uses 38,39,40,41. The purified QTY variant proteins exhibited predicted characteristics, stable structures and retained ligand-binding activity 38,39,40,41. Later we prepared QTY variant protein structure predictions using AlphaFold2, achieving results in hours 42 instead of ~ 5 weeks for each molecular simulation using GOMoDo, AMBER and YASARA programs 38,39,40. We already used AlphaFold2 to design water-soluble QTY variants of the 14 human glucose transporters43 and 13 human solute carrier transporters44, and now move to potassium ion channels.

Here we report a structural bioinformatics study of 6 Cryo-EM experimentally determined and 12 AlphaFold2-predicted human potassium ion channels and their water-soluble QTY variants. Native proteins and their water-soluble QTY variants share remarkable structural similarities notwithstanding significant protein sequence differences. Furthermore, different variants have similar characteristics that are responsible for protein stability, thereby preserving the essential sequence and structural parameters associated with protein function. These water-soluble QTY variants of potassium ion channels may be useful as antigens for the discovery and development of therapeutic monoclonal antibodies45.

Results and discussions

The rationale of the QTY code

The QTY code specifically selects three neutrally polar amino acids, namely glutamine (Q), threonine (T), and tyrosine (Y), to replace the hydrophobic amino acids leucine (L), isoleucine (I), valine (V), and phenylalanine (F). According to the electron density maps, we observed significant structural similarities between the original hydrophobic amino acids (L, I, V, F) and their respective replacements (Q, T, Y)38,39. The hydrophobic amino acids within the transmembrane alpha-helices are replaced by Q, T, and Y, resulting in the loss of hydrophobic characteristics in these alpha-helical segments (Fig. 1, Fig. 2, Fig. 5, Fig. 6, Fig. 7, Fig. 8).

Fig. 1
Fig. 1
Full size image

Protein alignments of 6 Cryo-EM structure potassium ion channels with their water-soluble QTY variants. The symbols | and * indicate the identical and different amino acids, respectively. Please note the Q, T and Y amino acid replacement (red). The alpha-helices (blue) are shown above the protein sequences. The loop color codes are internal (yellow) and external (red). Features of natural and QTY variants are pI, molecular weight (MW), total variation (%) and transmembrane variation (%). The alignments are (a) KCNN4 vs KCNN4QTY, (b) KCNK2 vs KCNK2QTY, (c) KCNA3 vs KCNA3QTY, (d) KCNMA1 vs KCNMA1QTY, (e) KCNH2 vs KCNH2QTY, (f) KCNJ11 vs KCNJ11QTY. Although there are significant QTY changes for the TM alpha-helix changes, > 48.4% for KCNN4, > 60% for KCNK2, their changes of molecular weight and pI are insignificant, two have no pI changes at all (KCNA3 and KCNMA1) (Table 2).

Fig. 2
Fig. 2Fig. 2
Full size image

Protein alignments of 12 AlphaFold 2 predicted native potassium ion channels with their water-soluble QTY variants. All characteristics and color codes are referred in Fig. 1. The alignments are (a) KCNA1 vs KCNA1QTY, (b) KCNA5 vs KCNA5QTY, (c) KCNH5 vs KCNH5QTY, (d) KCNJ3 vs KCNJ3QTY, (e) KCNJ8 vs KCNJ8QTY, (f) KCNK5 vs KCNK5QTY, (g) KCNK9vs KCNK9QTY, (h) KCNC4 vs KCNC4QTY, (i) KCND1 vs KCND1QTY, (j) KCNJ10 vs KCNJ10QTY, (k) KCNJ12 vs KCNJ12QTY, (l) KCNN3 vs KCNN3QTY. Although there are significant QTY changes for the TM alpha-helix changes, > 50% for KCNH5, ~ 57% for KCNA1, their changes of pI and molecular weight (MW) are insignificant and some have no pI changes at all (KCNA1, KCNA5, KCNK5, KCNC4, and KCNJ12) (see Table 2).

Protein sequence alignments and other characteristics

We aligned the native potassium ion channels with their QTY variants. The QTY variants exhibit a significant proportion of hydrophobic residues replaced by QTY counterparts (L, I/V, F → Q, T, Y transformation) in the potassium ion channels, both overall (~ 4.8%–15.7%) and particularly in the transmembrane domains (~ 48.4% → 60.7%). For example, the transmembrane domain exhibits differences ranging from 48.4% to > 60.7%, while the overall potassium ion channel proteins differ from their QTY variants by 5.5% to 14.3%, depending on the number of transmembrane alpha-helices present in the proteins (Fig. 1, Fig. 2, Table 2).

Table 2 The protein characteristics of 18 human potassium channels.
Full size table

Molecular weight is one of the factors to consider in comparing the native and QTY variant proteins, as QTY transformation should not cause drastic global structural transfiguration and significant changes in general parameters to let mutated proteins to keep their function. The molecular weight of QTY variants is slightly higher than that of native proteins (Fig. 1, Fig. 2, Table 2). This can be attributed to the fact that nitrogen (14 Daltons) and oxygen (16 Daltons) that are more frequent in hydrophilic amino acids have a higher molecular weight compared to carbon (12 Daltons). Additionally, QTY variants possess water-soluble side chains. The glutamine (Q) side chains form four hydrogen bonds with water: two donors through -NH2 and two acceptors through oxygen on -C = O. Similarly, the side chains -OH of threonine (T) and tyrosine (Y) each form three hydrogen bonds with water: one donor from hydrogen (H) and two acceptors from oxygen (O). Hydrogen bonds directly reflect structural stability, so drastic changes in hydrogen bonds can influence protein stability and integrity.

The isoelectric point (pI) of the potassium ion channels vary, with some falling within the acidic range and others in the basic range. For instance, 7 potassium ion channels, including KCNA1, KCNA3, KCNA5, KCNC4, KCNJ12, KCNK5, and KCNMA1, have acidic pIs ranging from 5.07 to approximately 6.60. KCNH5 has pI close to neutral, around 7.5. On the other hand, 9 potassium ion channels, namely KCNH2, KCND1, KCNJ3, KCNJ8, KCNJ10, KCNJ11, KCNK2, KCNN3, and KCNN4, have basic pIs ranging from pI 8.2 to pI 9.38 (Table 2, Fig. 1, Fig. 2). Notably, despite significant QTY sequence replacements, the pIs remain identical for 7 native channels and their QTY variants, including KCNA1, KCNA3, KCNA5, KCNC4, KCNJ12, KCNK5, and KCNMA1. The pIs in the QTY variants show minimal changes, and in some cases, there are no changes at all. This is because the neutral amino acids glutamine (Q), threonine (T), and tyrosine (Y) do not carry any charges at neutral pH. Consequently, the introduction of these amino acids does not significantly alter the pI of the proteins. This is important as changes in pI can lead to non-specific interactions. Changes in pI can influence protein-solvent and protein–protein interactions, as the protonation state of amino acids depends on the pI values. It can lead to local effects on the stabilization of the protein and even global effects if the number of local disruptions is high enough. From a thermodynamic point of view, when the pI variation increases, more amino acids change their protonation states, the initial interatomic interactions rearrange, leading to higher disorder and an increase in entropy, resulting in positive free energy.

Superposition of native and water-soluble QTY variants

We used experimental Cryo-EM structures that are available for six native potassium ion channels to assess the accuracy of the AlphaFold2 predictions for the native potassium channels. The comparisons revealed RMSD values ranging from ~ 0.5 Å to 2.14 Å (Table 2, Fig. 3), indicating that the models can accurately represent the protein structures. As shown in Fig. 3, these structures superposed well, particularly in the transmembrane alpha-helices (except for some unstructured loops) where the experimentally determined structures (magenta color) and the AlphaFold2-predicted water-soluble QTY variants (cyan color) closely overlap. These observations indicate that these structures share highly similar folds, despite significant QTY amino acid replacements (up to 60% QTY substitutions) in the transmembrane alpha-helices of the water-soluble QTY variants.

Fig. 3
Fig. 3
Full size image

Superposed 6 potassium ion channels Cryo-EM structures with QTY variants predicted by AlphaFold2. The Cryo-EM structures of the native transporters are obtained from the Protein Data Bank (PDB). The Cryo-EM structures (magenta) are superposed with QTY variants (cyan) predicted by AlphaFold2. KCNN4Cryo-EM vs KCNN4QTY (1.790 Å), KCNK2Cryo-EM vs KCNK2QTY (3.358 Å), KCNA3Cryo-EM vs KCNA3QTY (2.236 Å), KCNMA1Cryo-EM vs KCNMA1QTY (1.474 Å), KCNH2Cryo-EM vs KCNH2QTY (2.077 Å), KCNJ11Cryo-EM vs KCNJ11QTY (0.944 Å). These superposed structures display that Cryo-EM structures and their AlphaFold2 predicted native transporters have very similar molecular structures. For clarity of direct comparisons, the N-terminus and C-terminus are removed.

It is important to note that while AlphaFold2 predicts the global protein structure with high accuracy, there may still be some deviations between the predicted structures and the actual structures, particularly in the loops. This is because loops are inherently flexible and challenging to predict. The lower accuracy in loop regions can result in higher RMSD deviations, which can introduce some bias in the arrangement of the alpha-helical regions, even if the overall fold of the protein is correctly predicted. However, it is reasonable to assume that in dynamic environments, these models will undergo relaxation and loop repacking, leading to adjustments in the transmembrane bundle and ultimately reducing the RMSD.

The AlphaFold2-predicted native structures (green color) and experimentally-determined Cryo-EM native structures (magenta color) are similar. The RMSD for the pairwise structures are between ~ 0.40 Å to 1.92 Å, and the seven proteins fall below 1.5 Å (Table 2, Fig. 4). As shown in Fig. 4, these structures superposed well, particularly in the transmembrane alpha-helices where the experimentally determined structures and the AlphaFold2-predicted water-soluble QTY variants closely overlap. These observations indicate that these structures share highly similar folds.

Fig. 4
Fig. 4
Full size image

Superposed 6 potassium ion channels Cryo-EM structures with native structures predicted by AlphaFold2. The Cryo-EM structures of the native transporters are obtained from the Protein Data Bank (PDB). The Cryo-EM structures (magenta) are superposed with native structures (green) predicted by AlphaFold2. The RMSD (Å) for each structure are: KCNN4Cryo-EM vs KCNN4AF2 (1.779 Å), KCNK2Cryo-EM vs KCNK2AF2 (2.146 Å), KCNA3Cryo-EM vs KCNA3AF2 (1.449 Å), KCNMA1Cryo-EM vs KCNMA1AF2 (0.847 Å), KCNH2Cryo-EM vs KCNH2AF2 (1.953 Å), KCNJ11Cryo-EM vs KCNJ11AF2 (0.617 Å). These superposed structures display that Cryo-EM structures and their AlphaFold2 predicted transporters have very similar molecular structures. For clarity of direct comparisons, the N-terminus and C-terminus are removed.

The close superposition of these structures confirms the high accuracy of AlphaFold2’s predictions, as the predicted native structures directly superpose with Cryo-EM structures. These results also suggest that the native potassium ion channels and their water-soluble QTY variants share remarkable structural similarity.

Based on the accurate superpositions of Cryo-EM and predicted native structures, we then used AlphaFold2 to predict the structures of the 12 native potassium channels and 12 QTY variants without Cryo-EM structures (Fig. 5). We anticipate a good resemblance between the native potassium channels and their QTY variants because of the high similarity of 1.5 Å electron density maps between the hydrophobic amino acids L, V/I, F and Q, T, Y hydrophilic amino acids38,39.

Fig. 5
Fig. 5
Full size image

Superposed native potassium ion channels and their QTY variants that were predicted by AlphaFold2. The native structures (green) and their water-soluble variants (cyan). For clarity, N-terminus, C-terminus and large loops are deleted. KCNA1 vs KCNA1QTY (1.565 Å), KCNA3 vs KCNA3QTY (1.092 Å), KCNA5 vs KCNA5QTY (1.551 Å), KCNH5 vs KCNH5QTY (1.543 Å), KCNJ3 vs KCNJ3QTY (1.254 Å), KCNJ8 vs KCNJ8QTY (0.540 Å), KCNK2 vs KCNK2QTY (2.093 Å), KCNK5 vs KCNK5QTY (0.967 Å), KCNK9 vs KCNK9QTY, (0.514 Å), KCNC4 vs KCNC4QTY (1.105 Å), KCND1 vs KCND1QTY (1.025 Å), KCNH2 vs KCNH2QTY (1.547 Å), KCNJ10 vs KCNJ10QTY (1.728 Å), KCNJ11 vs KCNJ11QTY (0.446 Å), KCNJ12 vs KCNJ12QTY (0.397 Å), KCNMA1 vs KCNMA1QTY (0.981 Å), KCNN3 vs KCNN3QTY (1.917 Å), KCNN4 vs KCNN4QTY (1.003 Å). Please find all RMSD values (Å) in Table 2.

Analysis of the hydrophobic surface of native and QTY proteins

The native potassium channels are highly hydrophobic, especially in the 2TM, 4TM or 6TM alpha-helical domains. The 2TM, 4TM, and 6TM domains are directly embedded within the hydrophobic lipid bilayer, allowing the hydrophobic side chains of amino acids leucine (L), isoleucine (I), valine (V), and phenylalanine (F) to interact directly with lipid molecules while excluding water (Fig. 6). Consequently, these domains display highly hydrophobic surfaces. As a result, potassium ion channels are inherently insoluble in water and require detergents to maintain the structure after they are removed from lipid bilayer membranes. In the absence of appropriate detergents, they readily aggregate, precipitate, and lose their biological functionality.

Fig. 6
Fig. 6
Full size image

Hydrophobic surface of 18 potassium ion channels and the designed QTY variants. The native solute carrier transporters have many hydrophobic residues L, I, V and F in the transmembrane helices (yellow color). After Q, T, and Y substitutions of L, I, V, F, the protein surfaces become more hydrophilic (cyan color). Overall, the hydrophobic surface (yellow) of the native transporters become more hydrophilic (cyan). The hydrophobic surface is largely reduced on the transmembrane helices for the QTY variants. These QTY variants converted to water-soluble form. For clarity, the N-terminus and C-terminus are deleted.

Through the QTY conversion, the hydrophobic surfaces observed in the 2TM, 4TM, and 6TM alpha helices are significantly reduced (Fig. 5, Fig. 6). This transformation from hydrophobic to hydrophilic alpha-helices does not lead to significant alterations in the AlphaFold2 predicted structures. We believe that the structures between the native and QTY variants remains similar, based on our previous biochemical experiments38,39,40,41,48. These experiments demonstrated that QTY-designed chemokine receptors and cytokine receptors, which share general similarities with potassium channels as transmembrane alpha-helical proteins, retained their structural integrity, stability, and ligand-binding activities38,39,40,41,48.

Comparison of protein sequence characteristics between native, Cryo-EM, and QTY variant structures

We carried out systematic bioinformatic analyses and compared five characteristics of the predicted native, Cryo-EM, and QTY variant structures of all 18 potassium ion channels. They include: i) stability, ii) grand average of hydropathy (GRAVY), iii) flexibility, iv) solvent-accessible surface area (SASA), v) and solubility. Furthermore, we also compared seven molecular interactions including: i) hydrogen bonds (HBO), ii) polar contacts (PLR), iii) ionic interactions (ION), iv) aromatic contacts (ARO), v) hydrophobic contacts (HDP), vi) van der Waals interactions (VdW), vii) and van der Waals clashes (VCL).

Various sequence and structure parameters can be used to assess the difference in solubility between hydrophilic and hydrophobic proteins. GRAVY’s negative and positive values indicate the relative hydrophilicity and hydrophobicity of amino acids respectively. A GRAVY value < 0 implies a higher degree of hydrophilicity of a protein, whereas a GRAVY value > 0 indicates a hydrophobic nature. In the context of protein flexibility, higher values indicate greater flexibility. Solubility is defined using a threshold of 0.5: a protein is considered insoluble if its value is below 0.5, while values above 0.5 indicate solubility.

By conducting a Kruskal–Wallis H-test, we found that the differences in all sequence parameters were insignificant between the native and Cryo-EM proteins. We utilized this as a baseline to demonstrate the similarity of sequences obtained from experimental structures and AlphaFold2 models. However, we observed significant differences in GRAVY, flexibility, and solubility between the native and QTY structures, as well as the Cryo-EM and QTY structures. These results indicate that QTY sequences have lower GRAVY values, as they are more water-soluble, and higher flexibility and solubility values, which are expected during the transformation of hydrophobic proteins into their hydrophilic counter parts. On the other hand, stability and SASA showed insignificant variations in the same comparisons (Fig. 7). Regarding protein interatomic interactions, the comparison between the AlphaFold2-predicted native models and Cryo-EM structures revealed no significant differences according to the Kruskal–Wallis H-test, similar to the analysis of sequence characteristics. Only the number of HDP exhibits significant variation between the native and QTY variant structures, as well as the Cryo-EM and QTY variant structures, while all other interactions have little changes (Fig. 8).

Fig. 7
Fig. 7
Full size image

Bioinformatic comparison of protein sequence characteristics between native, Cryo-EM, and QTY structures. Grand average of hydropathy (GRAVY) and solvent-accessible surface area (SASA) are important indicators for protein solubility. A GRAVY value < 0 implies a higher degree of hydrophilicity in a protein, while a GRAVY value > 0 indicates a hydrophobic nature. Higher flexibility values reflect greater flexibility. Solubility values below 0.5 correspond to insoluble proteins, while values above 0.5 show soluble proteins. (a) Kruskal–Wallis H-test was performed between native and QTY, native, Cryo-EM, and QTY proteins to compare the differences in stability, GRAVY, flexibility, SASA, and solubility. The red line reflects p-value < 0.05 (significance level). The Kruskal–Wallis H-test is a non-parametric statistical test used to compare the medians of three or more independent groups (on the figure, the groups are stability, GRAVY, flexibility, SASA, and solubility), assessing whether there are significant differences among them. Significant differences between native and QTY proteins can be observed in GRAVY, flexibility, and solubility, while stability and SASA remain similar. (b) Comparison between stability, GRAVY, flexibility, SASA, and solubility of native, QTY, and Cryo-EM proteins is visualized using boxplots. (c) Additional bar plot visualization is presented for each protein including KCNA1, KCNA3, KCNA5, KCNC4, KCND1, KCNH2, KCNH5, KCNJ3, KCNJ8, KCNJ10, KCNJ11, KCNJ12, KCNK2, KCNK5, KCNK9, KCNMA1, KCNN3, and KCNN4 to compare their stability, GRAVY, flexibility, SASA, and solubility.

Fig. 8
Fig. 8
Full size image

Bioinformatic comparison of protein inter-atomic interactions between native, Cryo-EM, and QTY variant structures. Hydrogen bonds (HBO), polar contacts (PLR), ionic interactions (ION), aromatic contacts (ARO), hydrophobic contacts (HDP), van der Waals interactions (VDW), and van der Waals clashes (VCL). HBO are electrostatic attractive forces between hydrogen covalently bound to a more electronegative donor and electronegative acceptor. PLR refer to interactions between polar groups of amino acids. ION involve the attraction between oppositely charged ions. ARO exist in interactions between aromatic rings of phenylalanine, tryptophan and tyrosine. HDP are interactions when hydrophobic groups of amino acids cluster together. VDW are caused by weak attractive forces between nonpolar atoms. VCL represents steric repulsion due to the overlap of van der Waals radii. (a) Kruskal–Wallis H-test was performed between i) native and QTY, ii) native and Cryo-EM, and iii) Cryo-EM and QTY proteins to compare the differences in HBO, PLR, ION, ARO, HDP, VDW, and VCL. The red line reflects p-value < 0.05 (significance level). The Kruskal–Wallis H-test is a non-parametric statistical test used to compare the medians of three or more independent groups on the figure, the groups are HBO, PLR, ION, ARO, HDP, VDW, and VCL, assessing whether there are significant differences among them. The significant difference between native and QTY proteins lies solely in the number of hydrophobic contacts. All other parameters remain similar. (b) Comparisons are visualized using box plots between HBO, PLR, ION, ARO, HDP, VDW, and VCL of native, QTY, and Cryo-EM proteins. Bars represent the number of interactions per protein. (c) Additional bar plot visualization to compare their HBO, PLR, ION, ARO, HDP, VDW, and VCL is presented for each protein including KCNA1, KCNA3, KCNA5, KCNC4, KCND1, KCNH2, KCNH5, KCNJ3, KCNJ8, KCNJ10, KCNJ11, KCNJ12, KCNK2, KCNK5, KCNK9, KCNMA1, KCNN3, and KCNN4.

Molecular dynamics simulations

Molecular dynamics (MD) simulations performed on KCNJ11 and KCNN4 yield results similar but not identical to the previous rigid investigations. We used only the converged part of the trajectories (last 100 ns and 1000 frames) to calculate molecular interactions, as previously done for the rigid structures. All KCNN4 converged around 150 ns in both membrane and water (frame #1500), while KCNJ11 converged at the same time in water but required more time for stabilization in membrane (300 ns). Even though KCNJ11 doesn’t show strong convergence even at the simulation’s end, we still decided to keep this system since, after visual inspection, the protein structures looked reliable (Fig. 9).

Fig. 9
Fig. 9
Full size image

Convergence and stability of MD simulations. We performed molecular dynamics (MD) simulations of 2 proteins shown best in the interaction analysis based on rigid structures (KCNJ11 and KCNN4). In sum, dynamic analysis was made on the converged part of 8 MD systems: KCNN4Cryo-EM, KCNN4Native, KCNN4QTY in membrane, KCNN4QTY in water, KCNJ11Cryo-EM, KCNJ11Native, KCNJ11QTY in membrane, and KCNJ11QTY in water. (a) The converged part is marked by the red line (starting from 300 ns and frame #3,000 for proteins in a membrane, and 200 ns and frame #2,000 for proteins in water). We put the corresponding mean and standard deviation of RMSD values at this part of trajectories above the red line. RMSD values were calculated relative to the first simulation frame of each system. (b) Protein conformations at the beginning (frame #0, 0 ns) and at the end (frame #4,000, 400 ns for proteins in a membrane and frame #3,000, 300 ns for proteins in water) of MD simulations. Each image depicts one variant (Cryo-EM, native, or QTY) of the protein (KCNN4 or KCNJ11) in a membrane (gray lipids with orange phosphates at the membrane edges), surrounded by water and ions (Cl- is green and K+ is violet). All proteins maintain the same global structure at the end of the MD simulations as they had at the beginning.

We separated QTY systems into 2 categories: QTY in membrane and QTY in water. Similar to the rigid analysis, HDP shows significant differences in comparison between both i) native vs. QTY (in membrane and in water), and ii) Cryo-EM vs. QTY (in membrane and in water) systems. However, MD shows that PLR, ARO, and VCL vary between native and QTY in membrane, and HBO, PLR, ARO, and VCL between Cryo-EM and QTY in membrane (in the rigid case only HDP were different). We can observe the same trends in the box- and bar-plots (Fig. 10). The increase of PLR and decrease of ARO in QTY variants additionally confirm that QTY has become more water-soluble. The increase in VCL can be the direct consequence of the QTY amino acid replacements, while the decrease in HBO can imply lower stability.

Fig. 10
Fig. 10
Full size image

Bioinformatic comparison of protein inter-atomic interactions between native, Cryo-EM, and QTY variant structures through MD simulations. We performed molecular dynamics (MD) simulations of 2 proteins shown best in the interaction analysis based on rigid structures (KCNJ11 and KCNN4). In sum, dynamic analysis was made on the converged part of 8 MD systems: KCNN4Cryo-EM, KCNN4Native, KCNN4QTY in membrane, KCNN4QTY in water, KCNJ11Cryo-EM, KCNJ11Native, KCNJ11QTY in membrane, and KCNJ11QTY in water. Other abbreviations mean hydrogen bonds (HBO), polar contacts (PLR), ionic interactions (ION), aromatic contacts (ARO), hydrophobic contacts (HDP), van der Waals interactions (VDW), and van der Waals clashes (VCL). HBO are electrostatic attractive forces between hydrogen covalently bound to a more electronegative donor and electronegative acceptor. PLR refer to interactions between polar groups of amino acids. ION involve the attraction between oppositely charged ions. ARO exist in interactions between aromatic rings of phenylalanine, tryptophan and tyrosine. HDP are interactions when hydrophobic groups of amino acids cluster together. VDW are caused by weak attractive forces between nonpolar atoms. VCL represents steric repulsion due to the overlap of van der Waals radii. (a) Kruskal–Wallis H-test was performed between: i) native and QTY, ii) native and Cryo-EM, and iii) Cryo-EM and QTY proteins to compare the differences in HBO, PLR, ION, ARO, HDP, VDW, and VCL through MD simulations. The Kruskal–Wallis H-test is a non-parametric statistical test used to compare the medians of three or more independent groups on the figure, the groups are HBO, PLR, ION, ARO, HDP, VDW, and VCL, assessing whether there are significant differences among them. The significant difference between native and QTY proteins lies in the number of PLR, ARO, HBP, and VCL. It’s different compared to the rigid study where only hydrophobic contacts are significantly different. All other parameters remain similar. (b) Comparisons are visualized using box plots between HBO, PLR, ION, ARO, HDP, VDW, and VCL of native, QTY, and Cryo-EM proteins. Bars represent the number of interactions per protein. (c) Additional bar plot visualization to compare their HBO, PLR, ION, ARO, HDP, VDW, and VCL through MD simulations is presented for KCNJ11 and KCNN4.

However, the significance of HBO is at the limit of the sensitivity of the test, so it cannot be said with certainty that it is really significant. Moreover, during the visual evaluation of the systems through MD, we can suggest that both HBO and VCL don’t disrupt the global structure of the QTY proteins (Fig. 9). QTY in water behaves similarly to QTY in PLR, HDP, and VCL contacts compared to native systems, but it adds one more significantly different parameter (ARO). Compared to Cryo-EM systems, in addition to PLR, HDP, and VCL, which are significant in QTY in membrane, ION and ARO have also become significant in QTY in water. Higher amounts of ARO in QTY in water can suggest partial rearrangement of surface residues to form more hydrophobic interactions within the protein. An increase in ION can reflect a higher amount of interaction between the remaining surface residues (those not oriented inward to form more ARO) and the surrounding water, maximizing interaction between the residues both within the protein and with the outside environment. In the absence of a membrane, such rearrangements can support protein stability and a thermodynamically favorable state in water. QTY in water has more ARO and ION and less HDP with both proteins compared to QTY in membrane.

Additionally, KCNN4 QTY in membrane has less HBO than QTY in water. The amount of HBO and PLR aren’t significantly different between QTY in membrane and QTY in water. It implies that QTY might be able to exist in both conditions (membrane and water) and keep its hydrophilic properties. Interestingly, the global amount of ARO, ION, and HBO are higher in QTY in water similarly to Cryo-EM and native systems, but differently than QTY in membrane. All inter-atomic interactions remain similar between native and Cryo-EM systems based on the Kruskal–Wallis H-test.

Structural overview of the protein conformation before and after MD shows that all proteins keep their global structure throughout the MD (Fig. 11). The main conformational changes include alpha-helical shifts and loop fluctuations. When we made a pairwise comparison between each pair of systems (10 conformations per protein), we noticed RMSD values up to ~ 9 Å for KCNN4 and up to ~ 8 Å for KCNJ11 (Fig. 11). Considering that the comparisons were based on the backbone of all converged conformations between each pair of systems, such fluctuations are expected due to the dynamic nature of the proteins and their environment and don’t affect the overall arrangement of the proteins. Comparison between converged conformations for each pair of systems (QTY in water vs. QTY in membrane, QTY in water vs. Cryo-EM, QTY in water vs. native, QTY in membrane vs. Cryo-EM, QTY in membrane vs. native, Cryo-EM vs. native) also shows no significant changes in global protein folds (Fig. 11). All of these results suggest the maintenance of the native protein structure after incorporation of QTY mutations.

Fig. 11
Fig. 11
Full size image

Structural comparison of native, Cryo-EM, and QTY variant conformations from MD simulations. We performed molecular dynamics (MD) simulations of 2 proteins which showed best in the interaction analysis based on rigid structures (KCNJ11 and KCNN4). In sum, dynamic analysis was made on the converged part of 8 MD systems: KCNN4Cryo-EM, KCNN4Native, KCNN4QTY in membrane, KCNN4QTY in water, KCNJ11Cryo-EM, KCNJ11Native, KCNJ11QTY in membrane, and KCNJ11QTY in water. (a) Structural alignment between the conformations from the start (gray; 0 ns, frame #0) and end (main variant color; 400 ns, frame #4,000 for proteins in membrane and 300 ns, frame #3,000 for proteins in water) conformations of each trajectory. (b) Structural alignment between Cryo-EM conformations and other variant conformations. Each alignment includes 20 converged conformations (10 per system). (c) General distribution of RMSD values calculated based on pairwise comparison of all converged conformations for each pair of systems.

Structural analysis of oligomer interfaces formed by QTY variants

Potassium ion channels form oligomers to perform their functions (Fig. 12). Due to the limitations of the currently available in silico tools, our study focused on the monomeric forms of these proteins. To check if the QTY forms will still form functional biological assemblies, we placed the last MD frames (400 ns, frame #4,000 for proteins in membrane and 300 ns, frame #3,000 for proteins in water) of the mutated monomers into the available for KCNN4 (PDB: 6CNM)28 and KCNJ11 (PDB: 6C3O)32 Cryo-EM structures of oligomers to see how QTY mutations affect the interface between the domains. We aligned all the in silico monomer models (KCNN4Cryo-EM, KCNN4Native, KCNN4QTY in membrane, KCNN4QTY in water, KCNJ11Cryo-EM, KCNJ11Native, KCNJ11QTY in membrane, and KCNJ11QTY in water) to the corresponding domains of the Cryo-EM structures. The superposition has the same RMSD values as stated in Fig. 11 and Table 2 as the monomeric version of the KCNN4Cryo-EM and KCNJ11Cryo-EM were obtained from the same PDB structures by removing other domains.

Fig. 12
Fig. 12
Full size image

Structural analysis of oligomer interfaces formed by QTY variants. We superimposed the last MD conformations of KCNN4Cryo-EM, KCNN4Native, KCNN4QTY in membrane, KCNN4QTY in water, KCNJ11Cryo-EM, KCNJ11Native, KCNJ11QTY in membrane, and KCNJ11QTY in water monomers on the original Cryo-EM oligomers from PDB. Similar to the previous analysis, the Cryo-EM structures are colored in magenta (both monomers and oligomers), QTY variants are cyan, and native structures are green. (a) The placement of native, QTY, and Cryo-EM monomers in the Cryo-EM oligomer. QTY models fit well into the original Cryo-EM oligomer, though form a small fraction of the oligomer assemblies. (b) Structural interface and positions of the QTY mutations on the interface. The interface is defined as monomer residues that are placed within 5 Å distance from other domains. C-alpha atoms of the interface residues are shown as spheres and the QTY-mutated residues are colored in black. There are only a few QTY mutations on the monomer interfaces, implying no major impact on the functional oligomer.

First, we checked how many Cryo-EM and QTY monomer residues are in the interface with other domains in the oligomer structures. We define as the interface those monomer residues that have at least one atom from other domains in the oligomer within 5 Å distance of any monomer atom. In the oligomer state 28.8% (77 out of 267 residues) of KCNN4 and 82.0% (91 out of 111 residues) of KCNJ11 residues in Cryo-EM monomer structures form the interface between domains. When it comes to QTY variants, KCNN4 has 22.8% (61 out of 267 residues) and 27.3% (73 out of 267 residues) and KCNJ11 has 71.2% (79 out of 111 residues) and 77.5% (86 out of 111 residues) interface residues in water and membrane systems correspondingly.

Next, we looked into how many contacts that the original Cryo-EM monomers have remain the same (unmutated) at the QTY variants. In KCNN4 9 out of 77 residues of the original Cryo-EM interface were mutated to QTY (11.7% out of Cryo-EM interface and 15.0% out of all QTY mutations), while in KCNJ11 it is 7 out of 91 residues (7.7% out of Cryo-EM interface and 21.2% out of all QTY mutations).

Finally, we inspected how many mutations are at the interface in the QTY structures. KCNN4 systems have 5 out of 60 residues in both QTY in membrane (12.2% out of QTY interface) and QTY in water cases (9.4% out of QTY interface) making it 8.3% out of all QTY mutations. KCNJ11QTY in membrane has 5 QTY residues (8.3% out of QTY interface and 15.2% out of all QTY mutations) while KCNJ11QTY in water has 6 QTY residues (8.45% out of QTY interface and 18.18% out of all QTY mutations) both out of 33 interface residues in total.

These numbers highlight a minor interference on the interface that is unlikely to disrupt the oligomer state in the biological assemblies, making it possible for QTY variants to be functional in real life. Our findings confirm, at the level of molecular models, the feasaibility of transforming hydrophobic proteins into hydrophilic analogues while preserving their original functional structure and properties.

Conclusion

Nature has evolved three distinct types of alpha-helices: i) hydrophilic alpha-helices found in water-soluble enzymes including hemoglobin and dehydrogenases, as well as circulating proteins such as growth factors, cytokines, and hormones; ii) hydrophobic alpha-helices found in integral transmembrane proteins such as G protein-coupled receptors, transporters, and various ion channels, including photosynthesis systems; and iii) amphiphilic alpha-helices, which contain both hydrophilic and hydrophobic amino acid residues. Despite their differences in hydrophobicity and hydrophilicity, these three types of alpha-helices share nearly identical molecular structures 39,61,62,63. This is the structural basis of the QTY code.

Applying the simple QTY code, our study presents a straightforward approach to systematically convert hydrophobic alpha-helices in potassium ion channels into their water-soluble variants. After successfully evaluating the AlphaFold2 capability to predict potassium ion channels of 6 Cryo-EM structures, we proceeded to predict the structures of 12 potassium ion channels without the experimentally determined structures. We structurally- and bioinformatically analyzed 42 sequences and structures of transmembrane potassium ion channels including 6 Cryo-EM, 18 native and 18 QTY variants. The structures of QTY variants show a global similarity to native proteins, thus suggesting that the QTY variant proteins are likely to retain their original functions. To validate this assumption, we employed various bioinformatics tools to calculate sequence and structure characteristics associated with protein stability and water-solubility. Our results revealed that the QTY variants maintained comparable structural stability to the native proteins, while displaying water-solubility-related parameters. Furthermore, we observe reduced surface hydrophobicity patches in QTY variants compared to native potassium ion channels. Our findings further support the QTY code as a promising method for modeling water-soluble alpha-helical integral membrane proteins.

Nonetheless, it can be complicated to apply QTY code in the oligomer setting. Though our systems showed little QTY mutations on the interface between domains, it might not be the case in other systems. Considering that oligomers are mostly formed by hydrophobic interaction between monomers the mutations in the interface between monomers can disrupt the stability of the complex or even prevent its formation at all. Therefore, while our results provide insights into the potential effects of QTY mutations on individual monomers, they may not fully capture the consequences of these mutations in the full oligomeric assemblies where such hydrophobic interactions are integral. It is worth noting that keeping native residues at the interface between monomers in the oligomers will increase the probability of proper oligomer complex formation while QTY mutations at other parts will keep the proteins soluble.

We believe that the water-soluble potassium ion channels hold potential for applications including use as soluble antigens to generate therapeutic monoclonal antibodies (mAbs). Currently, there are no such approved therapeutic mAbs to treat a wide range of diseases including cancers. The water-soluble QTY membrane receptors have already been used to design an ultra-sensitive sensing device on dual layer S-layer protein and graphene-based conducting surfaces64. These results underline the practical use of QTY code in therapeutic and biotechnological purposes. 

Methods

Protein sequence characteristics

We use the term “Cryo-EM” for wild-type proteins with experimentally determined structures, while the other two variants (native and QTY) lack experimental structures. Hence, “native” is applied to unmodified sequences (wild type) and “QTY” to modified ones (non-wild type). The “Cryo-EM” structures were obtained from the Protein Data Bank (PDB), while both “native” and “QTY” structures were predicted based on wild-type and mutated sequences, respectively. The native protein sequences for potassium ion channels are obtained from Uniprot (https://www.uniprot.org)46 including KCNA1 (Q09470), KCNA3 (P22001), KCNA5 (P22460), KCNC4 (Q03721), KCND1 (Q9NSA2), KCNH2 (Q12809), KCNH5 (Q8NCM2), KCNJ3 (P48549), KCNJ8 (Q15842), KCNJ10 (P78508), KCNJ11 (Q14654), KCNJ12 (Q14500), KCNK2 (O95069), KCNK5 (O95279), KCNK9 (Q9NPC2), KCNMA1 (Q12791), KCNN3 (Q9UGI6), and KCNN4 (O15554). Native and QTY sequences47 were aligned to obtain the secondary structure corresponding to the sequences38,39,40,42,43,44,48. For further analysis we used sequences extracted from either predicted or available structures by Biopython (1.81)51. We ensured that all sequences have the same length by aligning them according to the Cryo-EM structures in cases where experimental structures are present. In other cases (when only native and QTY sequences are available), the sequences have the same length from the beginning. We used PDBsum49 generate tool (http://www.ebi.ac.uk/thornton-srv/databases/pdbsum/Generate.html) to visualize the secondary structure corresponding to the sequences. The website Expasy (https://web.expasy.org/compute_pi)50 was used to calculate the molecular weights (MW) and pI values of the proteins. The Python library Biopython (1.81)51 was used to compute stability52, grand average of hydropathy (GRAVY)53, and flexibility54 based on sequence information. Protein solubility was predicted by DeepSoluE (http://lab.malab.cn/~wangchao/softs/DeepSoluE/)55.

AlphaFold2 structure prediction

We used AlphaFold234,35 as AlphaFold2_advanced.ipynb notebook from ColabFold (https://github.com/sokrypton/ColabFold)56 for the structure predictions of the native and QTY variants following the instructions at the website on 2 × 20 Intel Xeon Gold 6248 cores, 384 GB RAM, and a Nvidia Volta V100 GPU. Predicted models were relaxed using CplabFold’s relax_amber.ipynb (https://github.com/sokrypton/ColabFold) 56. The initial sequence data was taken from the Uniprot website (https://www.uniprot.org)46 that contains information about each protein ID, entry name, description, and FASTA sequence. The QTY code web server (https://pss.sjtu.edu.cn/) 47 converted the FASTA protein sequences into their water-soluble versions. These steps were optimized using Python libraries for web applications such as requests and splinter. We cleaned the models by adjusting their lengths to match those of the available Cryo-EM structures or, in the case of unavailable experimental structures, the lengths of the native models. We also removed any unstructured portions.

Superposed structures

We performed superposition of native, QTY, and available Cryo-EM structures using PyMOL (https://pymol.org/2/). Experimental structures of potassium ion channels analyzed in this study include KCNN4 (protein Kca3.1) (PDB: 6CNM)28, KCNK2 (K2P2.1) (PDB: 4TWK), KCNA3 (Kv1.3) (PDB: 7EJ1)29, KCNMA1 (Kca1.1) (PDB: 6V22)30, KCNH2 (Kv11.1) (PDB: 5VA2)31, KCNJ11 (Kir6.2) (PDB: 6C3O)32. These structures were obtained from the Protein Data Bank (https://www.rcsb.org)57 in PDB format and processed by removing all atoms considered as HETATM. In cases where experimental structures were unavailable (QTY variants and other native proteins not mentioned above), we used AlphaFold2 models34,35.

Protein inter-atomic interactions

Hydrogens were added to all native, QTY, and Cryo-EM structures by Reduce (3.24.130724) software57,58. After that we converted PDB files into CIF using BioPython (1.79)50. We calculated interatomic contacts by Arpeggio (1.4.1)59 using CIF files as an input.

Structure visualization

Two key programs were used for structure visualization: PyMOL (https://pymol.org/2/) and UCSF Chimera (https://www.rbvi.ucsf.edu/chimera/)60. PyMOL was used for the superposed models, while hydrophobicity patches were visualized using Chimera.

Molecular dynamics simulations

We chose 2 proteins (KCNJ11 and KCNN4) that performed the best in the rigid protein inter-atomic interaction analysis for classical molecular dynamics (MD) simulations. To thoroughly investigate the behavior of all variants (Cryo-EM, native, and QTY) in dynamics we arranged 8 systems: KCNN4Cryo-EM, KCNN4Native, KCNN4QTY in membrane, KCNN4QTY in water, KCNJ11Cryo-EM, KCNJ11Native, KCNJ11QTY in membrane and KCNJ11QTY in water. We started from the protonated structures obtained on the previous steps of this study. Considering 2 different simulation environments (water and membrane) we varied the steps prior to the product MD simulations. For the systems with proteins in membrane, terminus (NTER and CTER) were added to both ends of each protein to make them uncharged. We constructed the bilipid membrane and incorporated the protein into it using CHARM-GUI (https://charmm-gui.org/)65,66,67. All simulation boxes have a rectangular shape. We chose POPC as the main type of membrane lipids based on the previous studies68. All proteins were oriented in the membrane by aligning the first principal axis along Z. We included 0.15 M KCl ion concentration and neutralized the systems’ charge making the conditions close to physiological. Prior to the production MD simulation, we performed energy minimization (EM) and several rounds of NVT and NPT simulations. The initial equilibration was conducted through a 6-step process, utilizing the default scripts from the CHARMM-GUI webserver. Additionally, we performed 2 more NPT equilibration steps (1 ns and 10 ns long respectively) to further stabilize the systems. Systems without a membrane contain the proteins (KCNJ11 and KCNN4), solvent (TIP3P69 water molecules), and ions (0.15 M KCl) and undergo only 1 round of EM, NVT, and NPT. EM was identical to the same with proteins in a membrane. NVT runs 50,000 steps (100 ps), while NPT runs 5,000,000 steps (10 ns). In both water and membrane systems, CHARMM36m69 force field was used for the 400 ns (300 ns for proteins in water) production MD runs with 2 fs integration time step. Energy-related information saved every 10,000 steps and the trajectory information every 50,000 steps, yielding 4,000 frames (3,000 frames for proteins in water) in total. The LINCS70 algorithm was applied to hydrogen bonds with a maximum of one iteration and fourth-order interpolation. Temperature coupling was achieved using the V-rescale algorithm with separate temperature groups for solvent, membrane, and solute at 320 K. Pressure coupling was applied using the Parrinello-Rahman barostat in a semi-isotropic mode. All simulations were made in GROMACS 2023.371. Details about the used parameters can be found on the Zenodo repository in mdp-files. Convergence and RMSD values data was calculated and plotted using Python’s library MDAnalysis72,73.

Oligomer analysis

KCNN4 (PDB: 6CNM)28 and KCNJ11 (PDB: 6C3O)32 oligomers were obtained from PDB. The interface between monomers in the oligomers and their visualizations were made using PyMOL (https://pymol.org/2/). Further analysis of interacting residues on the interface was performed in Python using custom scripts.