Protonation Enhances the Inherent Helix-Forming Propensity of pHLIP.

Cell-penetrating peptides (CPPs) can be potentially used in targeted delivery of therapeutic cargoes. However, their conformation in solution is poorly understood. We employed molecular dynamics simulations to probe the structural fluctuations of an anionic CPP, pH Low Insertion Peptide (pHLIP), in solution to determine the effects of modifications to selected residues on the structure of pHLIP. Two types of modifications were tested: (1) protonation of aspartic acid residues and (2) point mutations known to affect the acid sensitivity of pHLIP. pHLIP samples conformations ranging from coil to helix to sheet, and modifications to pHLIP lead to subtle shifts in the balance between these conformations. In some instances, pHLIP is as likely to form a helical conformation as it is to form an unstructured coil. Understanding the behavior of pHLIP in solution is necessary for determining optimal conditions for administration of pHLIP and design of promising pHLIP variants.

Introduction

Cell-penetrating peptides (CPPs) are a class of molecules with potential applications ranging from antimicrobial agents to vehicles for drug delivery.[1] The pH Low Insertion Peptide (pHLIP) is a fairly unique CPP, in that it is highly anionic (overall charge of −5), long (AEQNPIYWARYADWLFTPLLLLDLALLVDADEGT), and sensitive to changes in pH.[2,3] In solution, pHLIP is in a coiled conformation (state I); when exposed to the cell membrane under alkaline conditions, pHLIP binds to the membrane surface, remaining in a coiled conformation (state II); upon acidification of the environment, the acidic residues in pHLIP are protonated, leading to folding into an α-helix and insertion into the membrane (state III).[4] Although circular dichroism (CD) and fluorescence spectroscopy are commonly used to monitor the transition of pHLIP from state I → state II → state III, they cannot provide atomistic insights into the interactions that characterize each state. In particular, the behavior of pHLIP in solution (i.e., state I) is poorly understood. This lack of understanding is an issue, as most experiments with pHLIP require low peptide concentrations (<20 μM) to avoid aggregation.[5] In addition, point mutations of pHLIP have been utilized in attempts to improve the acid sensitivity of the peptide,[6−8] to varying degrees of success. Often these mutations lead to loss of acid sensitivity or increased aggregation effects. One recent study, in particular,[9] in which non-natural amino acids were substituted at positions 14 and 25 in pHLIP, led to enhanced insertion properties.

In light of these recent developments, we set out to utilize long-time-scale molecular dynamics (MD) simulations of pHLIP in state I to provide a molecular-level characterization of the behavior of pHLIP in solution. By using implicit solvent in our simulations, it is possible to routinely access microsecond time scales to significantly enhance our ability to completely sample the conformational landscape of pHLIP in state I. We used a recently developed fast pairwise Generalized Born (GB) implicit solvent model that has been successfully applied to protein folding on the microsecond time scale.[10,11]

The first set of variables tested was the protonation state of aspartate residues in pHLIP. Earlier work on pHLIP had established that pHLIP undergoes approximately two protonation events during folding and insertion.[2,12] In addition, point mutations of D14 and D25, the two interior aspartate residues, led to complete loss of acid sensitivity.[13] The combination of these results led to a conventional belief that D14 and D25 were the “protonation switches” controlling pHLIP folding and insertion.[14] However, more recent solid-state NMR studies have revealed that the C-terminal residues (D31 and D33) are titrated first under acidic conditions and thus may play a direct role in the function of pHLIP.[15] We thus examined singly protonated residues (D14, D25, D31, and D33) as well as two combinations of titrated residues (D14/D25 and “All”, D14, D25, D31, and D33).

Results and Discussion

Site-Specific Interactions Prevalent in State I

By measuring the average distances between residues, we can identify specific interactions that occur in pHLIP in state I (Figure ). When pHLIP is completely deprotonated (wild-type or WT), a small portion of the N-terminus (residues 6–11) preferentially interacts with the majority of the C-terminal half of the peptide (Figure A). Each individually protonated residue has localized effects, decreasing the interactions between the N-terminal segment and the C-terminal half of the peptide. However, combinations of protonations are somewhat different. When titrating both interior aspartic acid residues, there is a marked increase in interactions (Figure A, D14/D25); when protonating all aspartic acid residues, the interaction between the N-terminus and the C-terminal half of pHLIP is almost completely abolished (Figure A, All).

Figure 1

Modifications to pHLIP lead to site-specific interactions in state I. (A) Residue–residue interactions of Cα atoms in pHLIP. Regardless of the protonation state, increased interactions exist between the C-terminal region (approximately residues 26–33) and the N-terminal region (approximately residues 6–11) of pHLIP. Titration of acidic residues has differing results on these interactions. WT: fully deprotonated pHLIP; D14, D25, D14/D25, D31, D33: protonated residues in pHLIP; and All: all aspartic residues protonated. (B) Residue–residue interactions of Cα atoms in pHLIP for the helix-forming mutant (P20G) and non-natural amino acids that improve pHLIP function.[9] P20G mutation leads to a sharp increase of the N-terminus interacting with most of the peptide. The non-natural amino acids individually lead to greater interactions, but when both are incorporated, these interactions are mainly lost. Aad: α-aminoadipic acid; Gla: γ-carboxyglutamic acid.

The first point mutation (P20G) removes a conformational restriction from the peptide backbone (proline introduces a helical kink in the folded conformation of pHLIP) and should allow residues proximal to position 20 to interact with each other. This is the case, as the N-terminus forms a continual interaction (i.e., <10 Å) with almost all of the C-terminal residues (Figure B). We hypothesized that substitution of non-natural amino acids into pHLIP would lead to increased interactions within the peptide because of the increased negative charge (γ-carboxyglutamic acid (Gla)) or extended side chain (α-aminoadipic acid (Aad)) of each residue. This is only partially the case; substituting Gla at position 14 abolishes interactions with proximal residues while simultaneously increasing interactions with the C-terminal end of pHLIP (residues 24–30). A slight increase in interactions occurs proximal to the Aad substitution at position 25; however, when combining substitutions D14Gla and D25Aad, the only area of increased interactions is with residues between the two substitutions.

Perturbing the C-Terminal Half of pHLIP Leads to Noticeable Increases in Helicity

We next examined the effect that these variations had on the ability of pHLIP to form a helix in solution. Although pHLIP does not form a helix in state I, even at acidic pH, slight variations in the composition of pHLIP are able to form helices at alkaline pH.[13,16] Examination of the different protonation states shows that in almost every case protonation of a single acidic residue in pHLIP (D14, D25, D31, and D33) leads to an increase in helical conformations that are sampled in state I. This effect is not localized; most often, an increase in helicity occurs in the hydrophobic stretch of leucines between P20 and D25 (Figure A).

Figure 2

Increased interactions do not correlate with the helicity of pHLIP in state I. (A) Probability of pHLIP to form a helical segment (as defined by helical φ–Ψ angles for a three-residue sequence) as a function of residue titration. Single protonations in pHLIP increase helicity in state I. It is only when all aspartic acids are protonated that a significant increase in helicity occurs through a majority of the peptide. (B) Probability of pHLIP to form a helical segment as a function of point mutations. P20G leads to localized increase in helicity, due to the removal of proline kink. Non-natural amino acids enhance helicity only at position 25 (D25Aad, D14Gla/D25Aad). Thick black line: helix-forming propensity for the fully deprotonated, wild-type pHLIP.

In addition, multiple protonations are not necessarily cooperative because the protonation of both D14 and D25 leads to a decrease in helicity compared to that of the completely deprotonated pHLIP. (The highest residue–residue interactions also occurred with D14 and D25 both protonated. It appears that there is no direct relationship between interactions and helix-forming propensity.) It is only when all aspartic acids are protonated that a significant increase in helicity occurs through a majority of the peptide.

Point mutations in pHLIP do not have as significant an effect on the increase in helicity as that of titrations of acidic residues (Figure B). P20G leads to an increase in helicity in the interior of the peptide, consistent with observations from CD spectra.[7] D25Aad has the most noticeable increase in helicity near the hydrophobic stretch, possibly due to the charged side chain moving farther away from the peptide backbone. However, combining both point mutations (D14Gla and D25Aad) leads to a much smaller increase in helicity compared to that of the wild-type pHLIP.

pHLIP Transiently Samples Both Major Secondary Conformations in State I

It is only when we consider the entire secondary structure conformational landscape that the true behavior of pHLIP in state I emerges. When none of the acidic residues in pHLIP are titrated (WT), the peptide samples α-helical and β-sheet conformations almost equally (Figures and S2). Titration of the interior acidic residue, D14, leads to an increase in sampling of β-strand-like backbone conformations while maintaining the same level of sampling of helical conformations (Figure A). This trend is most striking when both D14 and D25 are protonated (D14/D25); in this titration state, pHLIP is twice as likely to sample a β-strand than an α-helix (Figures A and S2). The other singly protonated states, corresponding to the C-terminal aspartate residues (D25, D31, and D33), lead to an increase in sampling of an α helix compared to that of a β sheet (Figure ). Finally, when all aspartic residues are titrated (All), pHLIP is twice as likely to sample an α-helix than a β-sheet (Figure ).

Figure 3

pHLIP samples multiple secondary structures in solution. (A) Distributions of the Ψ backbone dihedral angle for different protonation states of pHLIP. The fully deprotonated state (WT, black line) samples the helical region of phase space (−50–0°) more often than the β-sheet region of phase space (120–170°). D31: blue line; D33: blue dash; D14: black dash; D25: red line; D14/D25: red dash; All: green line. (B) Distributions of the Ψ backbone dihedral angle for mutants of pHLIP. All mutants favor helical over sheetlike conformations. Point mutations at position 25 (D25Aad: blue line; D14Gla/D25Aad: green line) increase the likelihood of pHLIP to sample helical instead of sheetlike conformations. WT: black line; P20G: black dash; D14Gla: red line. (C) Ratio of α-helical to β-strand sampling, as determined by the areas under the respective curves in (A) and (B).

Variable effects on secondary structure formation are observed with respect to point mutations in pHLIP. Even though P20G has a localized increase in helicity, the overall helicity is lower than that of WT pHLIP (Figures B and S2). Combined with an increased sampling of β-strand conformations, P20G is more likely to sample strands than helices. Increased sampling of β-strands also occurs with the D14Gla mutant. However, both cases where the non-natural amino acid was introduced at position 25 lead to an increase in helicity and the propensity to sample helical space versus sheet conformations (Figures and S2).

Regardless of the variables tested, pHLIP is capable of sampling conformational space corresponding to β-strands, in over a third of the cases more often than α-helices. There is precedent for these phenomena experimentally: although it was originally suggested that pHLIP adopts a random coil in state I,[2] the corresponding CD spectrum has a negative mean residue ellipticity from 210 to 230 nm. This would indicate that a mixture of secondary structures (helices and sheets) exists[17] and is in agreement with the theory that single-spanning transmembrane helices are in a coil–helix equilibrium in solution.[18] Sheets and helices are often key secondary structural components in protein aggregation associated with neurodegenerative diseases,[19,20] and a recent study showed that amyloid fibrils can form from α-helical transmembrane proteins.[21] The fact that pHLIP aggregates at >10 μM into a tetrameric unit with exciton CD spectra representative of β-sheet formation[4] is similar in nature to the aggregation of peptides associated with amyloid formation.

Low-Energy States in Solution Are Favored by C-Terminal Modifications to pHLIP

Finally, we characterized the energy landscape of pHLIP as a function of the radius of gyration and intramolecular interactions (i.e., contacts). Generally speaking, the most energetically favorable states of pHLIP are when it is the most compact. Nearly all of the variants tested sample their lowest-energy conformations when the radius of gyration is 12 Å or less (Table S1). The one noticeable exception is when D14 and D25 are protonated (Figure A). It appears that neutralizing the two interior aspartic acid residues leads to less favorable intramolecular interactions, as this configuration also has the lowest number of contacts. To compensate for this decrease in interactions, pHLIP will rearrange into a β-sheet conformation. In contrast, when all of the aspartic acids are protonated, pHLIP predominantly lies in a very narrow distribution of the radii of gyration (<10 Å). A common fold for the lowest-energy population had a kinked α-helix from residues 21 to 31 as well as a helix from residues 8 to 13.

Figure 4

Number of contacts is loosely correlated with the radius of gyration in pHLIP. (A) Singly protonated residues in pHLIP (D14, D31, and D33) in general do not have a significant effect on the compactness of pHLIP. The lone exception is when D25 is protonated (D25), which leads to a decrease in the radius of gyration. Protonation of both interior aspartates leads to a significantly less compact conformation (D14/D25). When all aspartates are protonated, pHLIP is the most compact (All). Far left: representative snapshots of WT, D14/D25, and All protonation states of pHLIP. Sticks: deprotonated aspartic acid residues; spheres: protonated aspartic acids. Note the presence of helices for WT and All as well as the formation of β-sheet when D14 and D25 are protonated. (B) Point mutations have little effect on the compactness of pHLIP. The P20G mutation leads to a less compact conformation than that of the wild-type pHLIP (WT). Notably, the double mutant (D14Gla/D25Aad) samples a broader distribution of radii of gyration with an almost equal probability of number of contacts.

Each of the point mutations studied has a slightly different effect on the conformation of pHLIP. D25Aad is the only mutation that results in a slightly smaller radius of gyration than that of WT pHLIP (Figure B). Although P20G introduces greater helicity at position 20, this local ordering is offset by poor folding in the rest of the peptide, manifesting in a much broader distribution of contacts and radii of gyration. Incorporation of the bulkier and more negatively charged Gla side chain at position 14 led to a broader distribution of the radii of gyration. Examination of the simulated structures shows that the D14Gla variant has two predominant populations: (1) the Gla side chain is within the vicinity of the positively charged R11 side chain, and (2) the Gla side chain is >15 Å from R11 (Supporting Information). The D25Aad variant, which has a longer side chain, has no interactions with the R11 side chain. However, when incorporating both non-natural amino acids (D14Gla and D25Aad), the D25Aad side chain forms a salt bridge with the side chain of R11 for nearly half of the simulation. This drastic shift to favor salt bridge formation does not necessarily result in more compact conformations of pHLIP (Figure B); rather, the energetic gain from the salt bridge likely offsets the less energetic conformations with fewer contacts and larger radii of gyration.

The behavior of pHLIP with respect to compactness is also consistent with that of intrinsically disordered proteins. The two-dimensional free energy landscape of pHLIP can be related to recent work studying the effect of topology on the propensity of polyampholytes to aggregate in solution.[22−24] The Pappu and Ghosh groups developed a framework whereby the primary peptide sequence could be used in combination with the fraction of charged residues to relate to the compactness of a monomer in terms of the radius of gyration.[22,23] More recently, Lin and Chan were able to experimentally show that the radius of gyration for a single peptide chain is correlated to the phase behavior of multiple peptide chains.[24] Although pHLIP is an anionic peptide with a highly asymmetric charge distribution, the protonation states that we have tested follow the same relationship between the radius of gyration and sequence charge decoration (i.e., the radius decreases with increasing protonation up to a fully deprotonated pHLIP) that was observed both theoretically and experimentally (Supporting Information).

Conclusions

We have used MD simulations of pHLIP in implicit solvent to characterize the behavior of pHLIP in state I. pHLIP can sample multiple conformations in solution, as observed from previous CD studies. Significantly, we determined that pHLIP is capable of folding into both helices and strands when acidic residues are titrated. These conformational effects appear to be cooperative, as the most noticeable folding occurs when either the interior or all aspartic acid residues become protonated. This phenomenon is relevant to our continued improved understanding of pHLIP function, as it is necessary for determining optimal conditions for soluble administration of pHLIP as a diagnostic imaging or drug-delivery agent as well as its response to fluctuations in environmental pH while in solution.

Methods

System Setup

pHLIP was taken from helix C of bacteriorhodopsin (residues 73–107 of protein data bank (PDB) 2NTU), and GLY73 was mutated to ALA. This results in pHLIP 2–36, as in Karabadzhak et al. (referred to as pHLIP-4).[16] The peptide was then solvated and ionized using visual molecular dynamics (VMD),[25] with the CHARMM36 protein force field[26] used for heating simulations. The peptide was gradually heated to 700 K over 20 ps, followed by 980 ps of production to denature it from the helical conformation. A Langevin thermostat was used to maintain constant temperature, and heating simulation was carried out using NAMD2.9.[27] It was verified that the heating did not lead to any cis conformations of ω in the peptide backbone (Figure S1). The aspartates were protonated in VMD[25] using the psfgen plugin, and these residues were renamed as ASH as per the Amber naming convention. Convpdb[28] was used to convert PDB files into the Amber format, and Amber input files were generated using tleap.[29] ff14SBonlysc[10] and mbondi3 intrinsic radii were used. Unnatural amino acids were parameterized using the R.E.D. server.[30] Briefly, a PDB file for the residue (along with a N-terminal ACE patch and a C-terminal NME patch) was constructed using Avogadro.[31] “α” and “β” conformers were generated by setting the φ/Ψ values to −53/–47 and −119/119, respectively, optimizing each structure using Gaussian09[32] and uploading the results to the R.E.D. server.

Simulation

All simulations were run in Amber16[29] using the GB-Neck2 implicit solvent model.[11] Each peptide system was simulated for 2 μs. Snapshots were taken every 5 ps, providing 400 data points for 2 μs simulation time (similar to ref (10)). This sampling was used for each analysis, except for the free energy and salt bridge analyses, which used snapshots every 100 ps (i.e., 20 000 data points for the 2 μs simulation).

Analysis

Clustering:Trajectories were grouped into 50 clusters using the K-means clustering algorithm in cpptraj of AmberTools.[29] The Cα atoms of residues 10–33 (putative binding domain as per ref (15)) were used for clustering. The top (i.e., most populated) n clusters were selected for analysis, where n was chosen to include ∼40% of the trajectory. Helicity: A residue was defined to be in helical conformation if it simultaneously satisfied the following two conditions: −90 < φ < −30 and −77 < ψ < −17 (as per García and Sanbonmatsu[33]). A stretch was defined to be helical if three or more consecutive residues satisfied the above-mentioned condition. φ/ψ values were calculated in VMD using custom-made tcl scripts. Contact maps: The distance between Cα atoms of residues was used. Distances were calculated in VMD. Free energy analysis: For the number of contacts, the contacts command in cpptraj was used to count the number of Cα atoms within 7 Å of a given Cα atom. The radius of gyration was calculated in VMD. All plots were made using the matplotlib function of Python.[34]