Water-Protein Interactions of an Arginine-Rich Membrane Peptide in Lipid Bilayers Investigated by Solid-State NMR Spectroscopy

Abstract

The interaction of an arginine (Arg) residue with water in a transmembrane antimicrobial peptide, PG-1, is investigated by two-dimensional heteronuclear correlation (HETCOR) solid-state NMR spectroscopy. Using ¹³C and ¹⁵N dipolar-edited ¹H-¹⁵N HETCOR experiments, we unambiguously assigned a water-guanidinium cross peak that is distinct from intramolecular protein-protein cross peaks. This water-Arg cross peak was detected within a short ¹H spin diffusion mixing time of 1 ms, indicating that water is in close contact with the membrane-inserted guanidinium. Together with previously observed short guanidinium-phosphate distances, these solid-state NMR data suggest that the Arg sidechains of PG-1 are stabilized by both hydration water and neutralizing lipid headgroups. The membrane deformation that occurs when water and lipid headgroups are pulled into the hydrophobic region of the bilayer is symptomatic of the membrane-disruptive function of this antimicrobial peptide. The water-Arg interactions observed here provide direct experimental evidence for molecular dynamics simulations of the solvation of Arg sidechains of membrane proteins by deeply embedded water in lipid bilayers.

Introduction

Water-protein interactions are ubiquitous in biological systems and play an important role in the folding, stability, and function of proteins. NMR has long been used to investigate protein-water interactions and the mechanisms of polarization transfer from water to proteins ¹. For microcrystalline proteins in the solid-state, magic-angle-spinning (MAS) NMR experiments have shown that water-protein magnetization transfer can be mediated by chemical exchange followed by dipolar spin diffusion ², nuclear Overhauser effects (NOE) between water and the protein ^3,4, and rotating-frame Overhauser effects ⁵. For proteins bound to phospholipid bilayers, the balance between the interaction of polar residues with water and the interaction of hydrophobic residues with lipid acyl chains fundamentally determines the membrane topology of the protein. For membrane proteins that conduct ions across lipid bilayers, protein interactions with both membrane-surface water and channel water are important for the oligomeric structure and the function of the ion channel.

Several spin diffusion solid-state NMR (SSNMR) experiments have been introduced to determine the distances of proteins to membrane-associated water and to lipid chains ^6–8. The focus of these experiments has been to determine the global topology and conformational feature of the membrane proteins: whether they are transmembrane (TM) or surface-bound ⁷, and how the water accessibility of the protein changes with environmental conditions ⁹. Thus, most of these experiments employed relatively long spin diffusion mixing times of tens to hundreds of milliseconds, which detect water or lipid chain magnetization that originated from as far as several nanometers away. At these long mixing times, these experiments do not give site-specific information about water molecules within hydrogen bonding distance to protein residues.

Amino acids with ionizable sidechains such as arginine (Arg) play important roles in the function of many membrane peptides and proteins such as voltage-gated potassium channels ^10–12, antimicrobial peptides ^13,14 and cell-penetrating peptides ¹⁵. There has been great interest in understanding how the positively charged Arg sidechains are inserted into the hydrophobic part of the lipid membrane against high free-energy barriers ¹⁶. A number of molecular dynamics (MD) simulations have been carried out to address this question ^17–22 Freites et al. found that a stabilizing hydrogen-bonded network of water and lipid phosphates around the arginines of the S4 helix of a voltage-gated potassium channel reduced the local thickness of the bilayer hydrocarbon core to about 10 Å ¹⁷. Herce et al. showed that water penetrated the lipid bilayer and solvated the charged arginine and lysine residues during the translocation of the HIV Tat peptide across the membrane ²⁰. Allen et al. found that the Arg sidechain in a poly-Leu helix remained protonated in the membrane as a result of the stabilizing effects of membrane deformations ²³. Compared to MD simulations, direct experimental evidence for Arg interaction with lipids and water remains scarce ^13,24,25, due to the general difficulty of obtaining high-resolution structure information of membrane proteins in the disordered environment of lipid bilayers.

PG-1 is an Arg-rich disulfide-linked β-hairpin antimicrobial peptide found in porcine leukocytes ^26,27. Extensive biochemical assays ^28–30 and neutron diffraction data ³¹ have shown that PG-1 carries out its antimicrobial function by forming pores in the microbial cell membrane, thus disrupting the membrane barrier function. Solid-state NMR ¹H and ¹⁹F spin diffusion data revealed that PG-1 self-assembles into a TM oligomeric β-barrel in bacteria-mimetic POPE/POPG membranes ³². ¹³C–³¹P distance constraints indicate that the Arg residues in these β-barrels are complexed with lipid phosphates, and the guanidinium is within hydrogen bonding distance to the lipid ³¹P ^13,25. These results indicated that charge neutralization by lipid phosphates is a mechanism with which the Arg-rich PG-1 reduces the free energy of insertion into the hydrophobic region of the membrane. However, no experimental data has yet been reported about whether PG-1 uses water to stabilize its structure in the membrane, as suggested by MD simulations of other Arg-rich membrane peptides.

In this work, we investigate the water-Arg interaction of PG-1 in POPE/POPG bilayers using heteronuclear correlation (HETCOR) NMR. We describe how a spectral editing technique that removes protein ¹H signals allows the unambiguous identification of the resonances of water protons in contact with the Arg sidechains. This dipolar editing is essential for non-deuterated proteind, whose backbone Hα and amino protons often resonate within the possible water ¹H chemical shift range. We show how the use of short spin diffusion mixing times combined with dipolar editing allows us to detect water that is specifically associated with a guanidinium in the membrane.

Materials and Methods

Sample preparation

U-¹³C, ¹⁵N-labeled Arg·HCl was purchased from Sigma Aldrich (St. Louis, MO). A PG-1 sample (NH₂-RGGRLCYC RRRFCVCVGR-CONH₂) containing U-¹³C, ¹⁵N-labeled Arg₄ and ¹⁵N-labeled Leu₅ (R4, L5-PG-1) was synthesized and reconstituted into POPE/POPG membranes as described before ²⁵. Briefly, 1-palmitoyl-2-oleoyl-sn-glycero-3- phosphatidylethanolamine (POPE) and 1-palmitoyl-2- oleoyl-sn-glycero-3-phosphatidyl- glycerol (POPG) were mixed at a 3:1 molar ratio in organic solvents and lyophilized. The lipid powder was suspended in water and subjected to five cycles of freeze-thawing to form lipid vesicles. The peptide solution was mixed with the lipid vesicle solution at a peptide–lipid molar ratio of 1: 12.5. The mixture was incubated at 303 K overnight, then centrifuged at 55,000 rpm to obtain a ~40% hydrated pellet, which was packed into a 4 mm MAS rotor for SSNMR experiments.

Solid-state NMR spectroscopy

SSNMR experiments were carried out on a Bruker AVANCE-600 (14.1 T) spectrometer using a 4-mm triple-resonance MAS probe. Temperatures were controlled using a Bruker BCU-Xtreme unit. Radio-frequency pulse lengths were typically 5–6 μs for ¹³C, 6 μs for ¹⁵N and 3.0–4.5 μs for ¹H.

For the model compound U-¹³C, ¹⁵N-Arg · HCl, all ¹H-¹⁵N and ¹H-¹³C Lee-Goldburg (LG) HETCOR experiments (Figure 1a) were carried out at 273 K under 7.5 kHz MAS. ¹H homonuclear decoupling was achieved using the FSLG sequence ³³ with an 80 kHz transverse field. The ¹H chemical shift scaling factor due to the FSLG sequence was directly measured on the model peptide N-formyl-U-¹³C, ¹⁵N-labeled Met-Leu-Phe-OH (MLF) ³⁴ to be 0.560. For ¹H-¹⁵N LG-HETCOR experiments, a LG cross-polarization (CP) contact time of 800 μs and a maximum ¹H evolution time of 5.1 ms were used. For ¹H-¹³C LG-HETCOR experiments, the LG-CP contact time was 300 μs and the maximum t₁ evolution time was 4.1 ms. The MELODI-HETCOR experiment with two rotor periods of dipolar dephasing (Figure 1c) was carried out at 273 K under 6859 Hz MAS. The ¹³C and ¹⁵N 180° pulse lengths were 12 μs. Hartman-Hahn (HH) CP with a contact time of 800 μs was used to allow ¹H spin diffusion. To further promote ¹H spin diffusion, an additional z-filter period of 1 ms was used. Since the ¹H-¹H dipolar coupling during HH-CP is half the strength of laboratory-frame dipolar coupling, the total effective spin diffusion time of this experiment for Arg·HCl was 1.4 ms.

Figure 1.

Pulse sequences for various HETCOR and MELODI-HETCOR experiments. (a) 2D LG-CP HETCOR, (b) 2D HH-HETCOR with ¹H spin diffusion. (c) 2D MELODI-HETCOR with two rotor periods of ¹³C and ¹⁵N dipolar dephasing. (d) 2D MELODI-HETCOR with four rotor periods of ¹⁵N dephasing.

For the membrane-bound R4, L5-PG-1 sample, ¹H-¹³C and ¹H-¹⁵N LG-HETCOR experiments were carried out under 7.5 kHz MAS at 283 K. The LG-CP contact time was 800 μs. For the spin-diffusion-active ¹H-¹⁵N HETCOR experiment (Figure 1b), we used HH-CP with no additional ¹H mixing period. The HH-CP contact time was 2 ms at 283 K, corresponding to an effective spin diffusion time of 1 ms. The ¹⁵N-detected and ¹³C,¹⁵N-dephased MELODI-HETCOR experiment was carried out under 6859 Hz MAS with 2 ms HH-CP without further spin diffusion mixing. Most MELODI-HETCOR experiments used two rotor periods for dipolar dephasing. To better suppress the mobile sidechain proton signals, we also carried out a MELODI-HETCOR experiment with four rotor periods of ¹⁵N dipolar dephasing (Figure 1d). Additional HETCOR experiments at 253 K largely confirmed the 283 K data and thus are not shown here.

¹³C chemical shifts were referenced to the α-Gly C′ signal at 176.49 ppm on the TMS scale, and ¹⁵N chemical shifts were referenced to the ¹⁵N signal of N-acetylvaline at 122.0 ppm on the liquid ammonia scale. The ¹H chemical shifts were externally referenced to those of MLF ³⁵.

Results

2D heteronuclear correlation experiments for detecting water-protein interactions

The ¹H chemical shift of water can vary between 0 and 10 ppm, depending on the extent of hydrogen bonding among water molecules and between water and other polar functional groups ^36,37. Thus, to correctly identify water-protein correlation, it is necessary to distinguish the water and protein ¹H resonances. The protein ¹H chemical shifts can be readily assigned using 2D ¹H-¹⁵N and ¹H-¹³C HETCOR experiments with spin-diffusion-free LG-CP ³⁸ for polarization transfer (Figure 1a). Combined with homonuclear-decoupled ¹H chemical shift evolution, these pulse sequence elements ensure that mostly only one-bond correlation peaks are observed in the 2D spectra. To obtain correlations between non-directly-bonded protons and ¹³C or ¹⁵N, we use a short ¹H spin-diffusion mixing period and/or HH-CP for polarization transfer (Figure 1b), during which ¹H spin diffusion proceeds at half the rate of the laboratory-frame spin diffusion. Thus the HH-HETCOR experiment can exhibit cross peaks of ¹³C/¹⁵N spins with peptide protons several bonds away as well as with neighboring water protons.

We distinguish protein and water protons using a ¹³C/¹⁵N dipolar dephasing period before ¹H evolution (Figure 1c). This technique, termed MELODI ^39,40, was introduced previously for spectral editing of protein HETCOR spectra. During the dipolar dephasing period, ¹H magnetization evolves under ¹³C-¹H and/or ¹⁵N-¹H dipolar couplings, which are recoupled by alternating 180° pulses on the ¹H and heteronuclear channels in the same fashion as REDOR ⁴¹. ¹H homonuclear decoupling and a central ¹H 180° pulse maintain the ¹H magnetization and suppress any ¹H chemical shift evolution. The MELODI dephasing pulses can be applied on one or both heteronuclear spins. Simultaneous ¹³C and ¹⁵N irradiation ensures that the signals of both aliphatic and amino protons of U-¹³C, ¹⁵N-labeled protein residues are suppressed while the signals of water protons survive.

For mobile functional groups with attenuated C-H and N-H dipolar couplings, the two rotor periods of MELODI dephasing may not be sufficient to suppress their signals. In that case, one can readily extend the MELODI dephasing time to four rotor periods or longer to augment the dipolar dephasing effect (Figure 1d).

HETCOR experiments of U-¹³C, ¹⁵N labeled Arg·HCl

Figure 2a, b show 2D ¹H-¹³C and ¹H-¹⁵N LG-HETCOR spectra of U-¹³C, ¹⁵N-labeled Arg·HCl. This crystalline compound has two inequivalent molecules in the asymmetric unit cell ⁴², thus two sets of ¹³C and ¹⁵N chemical shifts were observed in the spectra. These ¹³C and ¹⁵N LG-HETCOR spectra allowed the assignment of all ¹³C, ¹⁵N, and ¹H chemical shifts (Table 1) ⁴³.

Figure 2.

2D HETCOR and MELODI-HETCOR spectra of U-¹³C, ¹⁵N-labeled Arg·HCl. (a) ¹H-¹³C LG-CP HETCOR. (b) ¹H-¹⁵N LG-CP HETCOR. (c) ¹H-¹⁵N HH-HETCOR with 1 ms spin diffusion. (d) ¹H-¹⁵N MELODI-HETCOR with two rotor periods of ¹³C dephasing, (e) ¹H-¹⁵N MELODI-HETCOR with two rotor periods of ¹⁵N dipolar dephasing. (f) ¹H-¹⁵N MELODI-HETCOR with two rotor periods of simultaneous ¹³C and ¹⁵N dephasing.

Table 1.

¹H, ¹³C and ¹⁵N isotropic chemical shifts of U-¹³C,¹⁵N labeled Arg·HCl and R4, L5-PG-1 in POPE/POPG membrane.

	Sample		Arg · HCl	R4, L5-PG-1
	Temperature		273 K	283 K
chemical shifts (ppm)	15N	Nα	44.8	117.6
		Nε	92.2, 90.9	84.1
		Nη1	85.7, 82.6	72.4
		Nη2	69.3, 67.7
	13C	C′	175.9, 174.7	171.3
		Cα	56.8, 54.7	51.5
		Cβ	29.1, 27.3	NA
		Cγ	25.2, 22.1	NA
		Cδ	42.0, 41.0	40.5
		Cζ	156.7	156.6
	1H	HN	7.8	7.9
		Hα	4.6, 4.3	4.7
		Hβ	2.1	1.8
		Hγ	1.9	1.3
		Hδ	3.8	2.9
		Hε	8.1, 8.3	6.4
		Hη1	6.8, 7.5	6.3
		Hη2	5.3, 6.8, 7.7
		H2O	-	4.8

All protons in U-¹³C, ¹⁵N-labeled Arg·HCl are directly bonded to a ¹³C or a ¹⁵N spin, thus ¹³C and ¹⁵N dipolar dephasing should suppress the one-bond peaks of all rigid segments and leave only the signals of mobile segments. The effective dipolar couplings in a two-rotor-period MELODI experiment are ${\delta }_{XH}^{\mathit{eff}}={\delta }_{XH}^{\mathit{rigid}}\times 4\times 0.577\times S_{XH}$ where 0.577 is the theoretical FSLG scaling factor and the rigid-limit couplings were 22.7 kHz for C-H and 10.6 kHz for N-H bonds. The order parameters S_XH account for any segmental motion that may be present and have been previously measured for Arg·HCl to be 0.91, 1.02, 0.96, 1.03 and 0.43 for Cα-Hα, Cδ-Hδ₂, Nε-Hε, Nη-Hη₂ and Nα-H^N₃, respectively ⁴⁴, which give effective dipolar couplings of 47.7, 53.2, 23.5, 25.2 and 10.5 kHz. Since ¹H magnetization evolves during the MELODI dephasing period, the spin dynamics is that of two-spin I-S systems even for CH₂, NH₂ and NH₃ groups. Figure 3a shows the calculated C-H and N-H MELODI dephasing curves as a function of the position of the ¹³C or ¹⁵N 180° pulses ³⁹. When the dephasing pulses are at the center of the rotor period, the Hα, Hδ, Hε and Hη intensities are all near zero (−6% to 1% of the full intensities), thus are negligible in the MELODI-HETCOR spectrum. The only exception is the amino protons H^N, whose three-site jump motions attenuate the ¹H-¹⁵N dipolar couplings so that ~50% of the intensities remain with the 180° pulses at the center of the rotor period.

Figure 3.

Simulated two-rotor-period MELODI-HETCOR dipolar dephasing curves of ¹H spins bonded to ¹³C or ¹⁵N. An ideal FSLG scaling factor of 0.577 ³³ was assumed. The simulated coupling strengths were obtained as ${\delta }_{XH}^{\mathit{rigid}}\times 4\times 0.577\times S_{XH}$, where the rigid-limit couplings are ${\delta }_{NH}^{\mathit{rigid}}=10.6\mathit{kHz}$ and ${\delta }_{CH}^{\mathit{rigid}}=22.7\mathit{kHz}$, and the order parameters were obtained from direct measurements. (a) Simulated curves for Arg.HCl. C-H and N-H order parameters of 0.4 – 1.0 were used. (b) Simulated curves for R4, L5-PG-1 in POPE/POPG membranes. Order parameters varied from 0.36 to 1.0.

Figure 2c shows the ¹H-¹⁵N HETCOR control spectrum in the absence of ¹³C and ¹⁵N dephasing pulses. All one-bond cross peaks and multiple-bond cross peaks such as ¹⁵N to aliphatic correlations are observed. When ¹³C dephasing pulses were turned on, the aliphatic ¹H signals between 1.0 and 5.0 ppm are completely suppressed while the amino ¹H signals between 5.0 and 9.0 ppm remain (Figure 2d). When the ¹⁵N dephasing pulses were turned on, the Hε and Hη peaks between 5 and 7 ppm decrease in intensity while the aliphatic ¹H signals remain. The 7.8-ppm NH₃ signal also remains as expected due to the fast three-site jump motion of the amino group (Figure 2e). Finally, when both ¹³C and ¹⁵N dephasing pulses were turned on, all resonances except for the 7.8-ppm H^N signal are suppressed (Figure 2f). These spectra thus confirm the ability of the MELODI-HETCOR experiment for removing the ¹H signals of rigid segments, leaving only the signals of mobile protons and non-protein protons in the spectra.

Water interaction with an Arg sidechain of membrane-bound PG-1

To investigate whether the Arg₄ sidechain in PG-1 is solvated by water in the lipid membrane, we first verify that Arg₄ is well inserted into the hydrocarbon region of the lipid membrane. Previous paramagnetic relaxation enhancement ⁴⁵ and ¹H spin diffusion experiments ⁴⁶ showed that PG-1 adopts a TM orientation in neutral DLPC and POPC bilayers, with increasing depths of immersion for G2 < F12 < L5 ≈V16. In anionic lipid membranes, ¹³C-³¹P distances ²⁵ and ¹H spin diffusion data ³² showed that Leu₅ is in close contact with both the lipid chains and the headgroups, indicating that the lipid membrane forms a toroidal pore near the peptide. Since Arg₄ is adjacent to Leu₅, it should be in close contact with the acyl chains as well. Figure 4 shows a 2D ¹³C-¹H HETCOR spectrum with 100 ms ¹H spin diffusion. The ¹H dimension was measured without homonuclear decoupling, thus only signals of mobile water and lipids were detected. Weak but clear cross peaks between Arg₄ Cα/Cδ and lipid CH₂ protons can be detected, indicating qualitatively that Arg₄ is inserted to the hydrocarbon region of the membrane. The spectrum also shows strong Arg₄ – water cross peaks. However, at the long ¹H spin diffusion mixing time used, these water peaks may originate from water on the membrane surface rather than hydration water around the guanidinium. Thus, we carried out ¹H homonuclear-decoupled HETCOR experiments with short spin diffusion mixing time to detect neighboring water to Arg₄.

Figure 4.

2D ¹³C-detected ¹H spin diffusion spectrum of R4, L5-PG-1 in POPE/POPG membrane at 283 K. A ¹H mixing time of 100 ms was used.

Figure 5a, b display the 2D ¹H-¹³C and ¹H-¹⁵N LG-HETCOR spectra of membrane-bound R4, L5-PG-1 at 283 K. Both peptide and lipid peaks can be readily assigned. The well resolved lipid ¹H signals in the ¹H-¹³C 2D spectrum provided convenient chemical shift references for the peptide ¹H signals ⁴⁷. Arg₄ Cα shows cross peaks to directly bonded Hα as well as amide H^N, which allows ¹H chemical shift calibration of the 2D ¹H-¹⁵N spectra. When ¹H spin diffusion was activated by the use of 2 ms HH-CP, the 2D ¹H-¹⁵N HH-HETCOR spectrum (Figure 5c) shows multiple-bond cross peaks between the Leu₅ Nα and its aliphatic protons. Arg₄ Nε and Nη exhibit not only directly bonded Hε and Hη peaks, but also weak multi-bond cross peaks with Hδ (2.9 ppm). In addition, a cross peak at 4.8 ppm was detected. In principle, this 4.8-ppm peak may be due to 1) Arg₄ Hα (4.5 ppm in the ¹H-¹³C HETCOR spectrum), 2) a guanidinium proton whose signal is weak in the ¹H-¹⁵N LG-HETCOR spectrum but stronger in the HH-HETCOR spectrum, or 3) water. To distinguish the three possibilities, we implemented ¹⁵N and ¹³C MELODI filters (Figure 6). When ¹⁵N dephasing pulses were turned on for two rotor periods, the amide H^N-Nα peaks of Arg₄ and Leu₅ were both suppressed as expected. The Nε-Hε and Nη-Hη peaks decreased in intensities but did not disappear completely, indicating sidechain motion at 283 K (Figure 6b). This sidechain motion is consistent with previously measured Arg₄ guanidinium S_NH of 0.35–0.50 ⁴⁴, which translate to effective dipole couplings of 25.2, 11.7 and 8.8 kHz for Nα-H^N, Nε-Hε and Nη-Hη during two rotor periods of ¹⁵N MELODI filter. Figure 3b shows the calculated intensities for the three couplings: with the dephasing pulses at the center of each rotor period, the H^N, Hε and Hη intensities are −6%, 43% and 64% of the full intensities. Thus, the complete suppression of the H^N peak and the reduced intensities of the guanidinium peaks in Figure 6b are qualitatively consistent with the calculated intensities. Figure 6b also shows that the multi-bond aliphatic proton – amide ¹⁵N cross peaks of Arg₄ and Leu₅ survived the ¹⁵N MELODI filter, as expected. Moreover, the 4.8-ppm ¹H peak remained in the spectrum. To affect stronger dipolar dephasing, we measured a four-rotor-period MELODI-HETCOR spectrum (Figure 6c), which further decreased the one-bond Nε-Hε and Nη-Hη peak intensities while retaining the intensity of the 4.8-ppm peak. Figure 6e compares the Nε and Nη cross sections of the three 2D spectra, which show that the relative intensity of the 4.8-ppm peak increased from being only 10% of the one-bond peaks in the control spectrum to 30–50% in the two-rotor-period MELODI spectrum, and to 70–100% of the one-bond peaks in the four-rotor-period MELODI spectrum. This trend rules out Hη or Hε as the cause of the 4.8-ppm peak.

Figure 5.

2D HETCOR spectra of R4, L5-PG-1 in POPE/POPG membrane at 283 K. (a) ¹H-¹³C LG-HETCOR. (b) ¹H-¹⁵N LG-HETCOR. (c) ¹H-¹⁵N HH-HETCOR with 2 ms HH-CP and no ¹H mixing period.

Figure 6.

2D ¹H-¹⁵N MELODI-HETCOR spectra of R4, L5-PG-1 in POPE/POPG membranes at 283 K. (a) HH-HETCOR spectrum. (b) MELODI-HETCOR spectrum with two rotor periods of ¹⁵N dephasing. (c) MELODI-HETCOR spectrum with four rotor periods of ¹⁵N dephasing. (d) MELODI-HETCOR spectrum with two rotor periods of simultaneous ¹³C and ¹⁵N dephasing. (e) ¹H cross sections of R4 Nε and Nη extracted from spectra (a–c).

Finally, to test whether the 4.8-ppm peak is due to Arg₄ Hα, we applied simultaneous ¹³C, ¹⁵N-dephasing. The resulting 2D spectrum (Figure 6d) shows the clear retention of the 4.8-ppm peak and the removal of all Arg₄ aliphatic ¹H signals. In particular, the Arg₄ two-bond Nα-Hα peak is suppressed, indicating that ¹³C dephasing is successful for removing the Hα magnetization. Since Leu₅ is not labeled with ¹³C, the Leu₅ aliphatic protons retain weak cross peaks with the Leu₅ amide ¹⁵N. Taken together, these dipolar edited HETCOR spectra indicate that Arg₄ Nη correlates with water protons within a short ¹H mixing time of only 1 ms (due to 2 ms HH-CP). The fact that the backbone Nα’s do not exhibit this 4.8-ppm cross peak confirms the site-specific nature of the water molecules interacting with the guanidinium moiety.

Discussion

The above data demonstrate the feasibility of detecting site-specific interactions between water molecules and membrane proteins without requiring protein perdeuteration. So far all solid-state NMR studies of protein-water interactions by heteronuclear correlation experiments used perdeuterated proteins in order to avoid overlap of protein ¹H signals, primarily Hα and H^N, with the water signal ^2,4. However, protein perdeuration is costly and sometimes unfeasible. The ¹³C and/or ¹⁵N dipolar-edited MELODI-HETCOR experiment obviates the need for perdeuteration by suppressing the signals of all protons that are directly bonded to a ¹³C or ¹⁵N spin, leaving only intermolecular water or lipid ¹H resonances of interest. By restricting ¹H spin diffusion to a short period of 1 ms, one can largely suppress the intensity of remote membrane-surface water ^7,9. The fact that the 4.8-ppm ¹H signal correlates with Arg₄ Nε and Nη but not with the backbone Nα of Arg₄ and Leu₅ (Figure 6d) supports the site-specific nature of the observed water interaction with the guanidinium ion. The Arg₄-bound water is well immersed into the hydrophobic region of the membrane, since Arg₄ and Leu₅ are both within spin diffusion contact with the lipid chains.

The observed Arg₄-water close contact inside the lipid membrane, combined with previous report of short Arg₄ guanidinium-phosphate distances ²⁵, indicate that PG-1 utilizes both lipid phosphates and water molecules to stabilize its Arg sidechains in the hydrophobic core of the membrane (Figure 7). Each guanidinium group provides up to five hydrogen-bond donors, which can interact with the oxygen atoms of several phosphates and water molecules. While the exact stoichiometry of water and phosphates per guanidinium is not known, our data indicate that both species must be present to solvate and neutralize the positively charged Arg₄ in the membrane. The toroidal pore morphology of the membrane near the peptide, caused by lipids that change their orientations to embed their phosphate groups near the arginine sidechains, has been extensively characterized using static ³¹P and ²H NMR lineshapes and MD simulations for PG-1-bound membranes ^48–51 as well as for other membrane-active peptides ⁵².

Figure 7.

Illustration of Arg sidechain interaction with lipids headgroups and water inside the membrane.

The mechanism of water polarization transfer to Arg₄ in membrane-bound PG-1 is probably temperature dependent. At 283 K, the most likely mechanism is chemical exchange, with previously reported exchange rates of ~100 s⁻¹ at pH 7 and 10–20 °C for Arg Hε and Hη ⁵³. We also measured PG-1 HETCOR spectra at 253 K, where weak water cross peaks were also detected (data not shown). At low temperature, chemical exchange becomes negligible (≪ 10 s⁻¹), thus water polarization transfer to guanidinium most likely proceeds by dipolar spin diffusion. Previous solid-state NMR studies of microcrystalline proteins did not detect Arg-water correlations due to inter-residue hydrogen bonding of the arginines with other residues ⁵⁴. PG-1 contains 33% Arg in its amino acid sequence (six Arg residues out of 18 residues) and no acidic residues, thus peptide-peptide interactions are clearly insufficient, if not completely absent, for stabilizing the Arg residues. Thus, it is not surprising that lipid headgroups and water molecules provide the necessary hydrogen-bond partners and solvation shells to these Arg’s, as shown by the current 2D data and previous ¹³C-³¹P distance results.

The observed guanidinium-water correlation provides direct experimental evidence for the proposal, based on MD simulations, that Arg residues in membrane proteins are solvated by water molecules pulled into the membrane. These simulations gave detailed insights about how Arg residues in model peptides and voltage-sensing helices of potassium channels may be stabilized in lipid bilayers. Most simulations agreed that Arg remains largely protonated across most of the lipid membrane. The aqueous pKa of Arg is 12.5 ⁵⁵, and a pKa shift of 4.5 or less has been computed ²³, thus predicting that Arg remains predominantly protonated in the membrane at pH 7 ²¹. Potentials of mean force calculations indicated that the protonated Arg is stabilized mainly by hydration water and to a smaller extent by lipid phosphates, both of which are pulled into the membrane ^18,19. Large water defects that connect bulk water with the charged Arg sidechain all the way to the center of the membrane were observed ²², and the core water molecules stabilize the Arg by as much as 35 kcal/mol ¹⁹. These water defects and embedded lipid headgroups cause significant membrane deformation ^17,19. In the case of PG-1, such membrane deformation is manifested by the ³¹P NMR spectra of PG-1-containing lipid membranes ^49,56. The number of water molecules associated with each guanidinium in the membrane varies with the position of the Arg. For the S4 helix of a voltage-gated potassium channel, 1–3 water-guanidinium hydrogen bonds were detected in MD simulations up to 10 ns ¹⁷. For an Arg-containing poly-Leu helix, five water molecules were found to be coordinated to the Arg throughout the membrane ¹⁸. The Arg-coordinated water is suggested to be more sluggish than bulk water, with 200–300 times longer mean residence times in the case of the S4 helix ¹⁷.

Conclusion

The current HETCOR solid-state NMR study provides the first experimental evidence for site-specific guanidinium-water interaction in an Arg-rich membrane peptide that is well inserted into the lipid membrane. The ¹³C, ¹⁵N-dephased MELODI-HETCOR experiment unambiguously distinguishes intermolecular water-Arg cross peaks from intramolecular peptide cross peaks. Varying the dipolar dephasing time in the MELODI-HETCOR experiment also distinguishes the signals of mobile protein segments from intermolecular water-Arg cross peaks. The membrane-inserted water is able to transfer its polarization to the Arg sidechain within 1 ms, thus it is in close proximity to the guanidinium, and may form transient water-guanidinium hydrogen bonds. Together with previous results of short Arg-phosphate distances ²⁵ indicative of the presence of highly curved toroidal pores in the membrane, these solid-state NMR data indicate that Arg sidechains in the β-hairpin PG-1 utilize water for charge solvation and lipid phosphates for charge neutralization, thus stabilizing the insertion of the peptide across the lipid membrane. The resulting membrane deformation provides the structural basis for the membrane-disruptive function of PG-1.