Dynamic Nuclear Polarization Study of Inhibitor Binding to the M2_18–60 Proton Transporter from Influenza A

Abstract

We demonstrate the use of dynamic nuclear polarization (DNP) to elucidate ligand binding to a membrane protein using dipolar recoupling magic angle spinning (MAS) NMR. In particular, we detect drug binding in the proton transporter M2_18–60 from influenza A using recoupling experiments at room temperature and with cryogenic DNP. The results indicate that the pore binding site of rimantadine is correlated with previously reported widespread chemical shift changes, suggesting functional binding in the pore. Futhermore, the ¹⁵N labeled ammonium of rimantadine was observed near A30 ¹³Cβ and G34 ¹³Cα suggesting a possible hydrogen bond to A30 Carbonyl. Cryogenic DNP was required to observe the weaker external binding site(s) in a ZF-TEDOR spectrum. This approach is generally applicable, particularly for weakly bound ligands, in which case the application of MAS NMR dipolar recoupling requires the low temperatures to quench dynamic exchange processes. For the fully protonated samples investigated, we observed DNP signal enhancements of ~10 at 400 MHz using only 4–6 mM of the polarizing agent TOTAPOL. At 600 MHz and with DNP, we measured a distance between the drug and the protein to a precision of 0.2 Å.

Magic angle spinning (MAS) NMR is a powerful analytical technique well suited for dynamic and structural characterization of membrane proteins and amyloid fibrils at atomic resolution(1–3). In contrast with solution NMR, the ability to investigate systems of larger molecular mass without an inherent broadening of resonances is a major advantage of MAS NMR. However, the decreased molecular site concentration of larger molecules and detection of smaller magnetic moments of ¹³C and ¹⁵N spins means that sensitivity limits the application of traditional MAS NMR experiments. Dynamic nuclear polarization (DNP) can greatly increase the sensitivity of NMR by transferring the substantially higher polarization found in the electron spin reservoir to nuclear spins. DNP was first demonstrated at a magnetic field of 3.03 mT using a static sample of a metal(4); these seminal studies also experimentally verified the Overhauser effect as an efficient polarization transfer mechanism. In the 1980's, DNP experiments performed on dielectric solids were integrated with MAS at 1.2 T, a low magnetic field by current standards(5–8). The extension of DNP to magnetic field strengths used in contemporary NMR (≥ 5 T) has been achieved with the implementation of gyrotron oscillators as microwave power sources. Additional advances in NMR probe technology and development of stable organic radicals serving as the source of electron polarization have stimulated more widespread adoption of the approach and recently sparked considerable interest in the biological solid state NMR community. DNP in conjunction with MAS NMR has now been proven to yield considerable gains in sensitivity making possible a 25–10,000 fold reduction in experiment time in studies of membrane proteins, amyloid fibrils, and peptide nanocrystals at 9 Tesla(9, 10). Cryogenic MAS studies with improved probe technology on crystalline peptides demonstrated that resolution need not be compromised at low temperature(11, 12). MAS-DNP experiments on bacteriorhodopsin, a helical 7 transmembrane protein, have exploited both the sensitivity available from DNP as well as the cryogenic temperatures employed in such experiments to trap intermediates of its proton pumping cycle to better characterize the mechanism of ion translocation(13, 14). Experiments on the amyloid protein PI3-SH3 have shown that additional cross-peaks appear at low temperatures due to changes in protein dynamics, and the DNP enhanced spectra were useful in determining details of the fibril structure(15). In some, but not all cases, the cryogenic sample temperature leads to inhomogeneous line broadening and reduced resolution. Therefore application of DNP to biological systems of broad interest typically relies on well-resolved chemical shifts from spins found in unique chemical environments, or specific isotopic labeling of a cofactor, inhibitor, or residues to generate site-specific resolution. With the present state of the art and fields of about 9 T to 14 T, successful application will likely involve carefully labeling the protein and ligand to make use of spectra with approximately 1.5 (¹³C) and 3 (¹⁵N) ppm linewidths(11, 16).

In this study, we demonstrate that DNP can be used to elucidate ligand-protein interactions even in the weak (mM) binding regime using the proton transporter M2_18–60 from Influenza A virus and the inhibitor rimantadine. We present direct measurement of membrane protein-ligand interaction using MAS dipolar recoupling and DNP to detect the sites of drug binding in the tetrameric proton channel M2_18–60 in the non-conducting, high pH state. This method is generally applicable to the study of drug or ligand binding, particularly for weak binding because general application of MAS NMR dipolar recoupling techniques to weak binding ligands requires the temperature to be lowered in order to quench dynamic exchange processes(17). Although we stress the importance of low temperatures to quench exchange processes, binding sites can be detected while undergoing exchange using NOE measurements or dephasing experiments in which the exchange term commutes with the Hamiltonian. However, experiments such as ZF-TEDOR(18, 19) that rely on transverse mixing do not commute with exchange terms and are made accessible by the low temperatures and DNP.

The M2 proton channel from Influenza A is inhibited by the antiviral drugs rimantadine (Rmt) and amantadine (Amt). The full protein is 97 amino acids, and a construct comprising residues 21–61 has been shown to retain proton conductivity and inhibition by Amt and Rmt(20). This construct and a similar 18–60 construct contain a single pass transmembrane helix and an amphipathic helix, which has been shown to stabilize the tetrameric assembly. An even shorter construct, M2_22–46, comprising the transmembrane segment, shows reduced inhibition by drug, but remains drug sensitive. A controversy regarding the site of pharmacological inhibition arose from the observation of two different binding sites in various constructs of M2. Prevalence of n terminal resistance mutations(21) and drug induced changes in the pKa of H37(22) suggested binding in the pore of the channel. However, structural studies showed multiple binding sites. An external site near D44 and R45 was observed by NOE in a solution NMR structure(23) of M2_18–60. On the other hand, an internal site in the pore near residues V27, S31, and G34A was proposed in a 3.5 Å diffraction structure of M2_22–46 at pH 5.3(24), and also observed in REDOR recoupling experiments(25). When the drug was first detected in the pore of M2_22–46 in lipid bilayers, it was unclear if the conflicting results were due to truncating the protein near the external site, the particular micelle environment of the solution NMR structure, or other differences in sample preparations. The drug has now been shown to bind to the pore of M2_21–61(26) ruling out problems with protein truncation. It is therefore known that the aminoadamantyl inhibitors Amt and Rmt bind to the pore of fully functional constructs in lipids, and the lack of pore NOE measurements in the solution NMR is likely due to the specific sample environment.

Nevertheless, since the drug has been shown to bind to both sites, the important remaining question was to definitively determine the site(s) responsible for inhibition. The pharmacological significance of each binding site has been supported based on resistance mutations and binding affinities, yet there is not yet consensus regarding interpretation of the functional data(27, 28). Perhaps the strongest support for functional pore binding came from an interesting drug-sensitive chimera protein containing C-terminal residues from influenza B M2 and thus lacking the external binding site(29). A recent study using this AM2-BM2 chimera solved a detailed structure of the tetramer with Rmt bound inside the pore(30).

Given the existing evidence for multiple binding sites, most likely of differing affinity, we investigated M2_18–60 in lipid bilayers to explore the benefits of cryogenic DNP and to determine the locations of the functionally important drug amine group when bound to the protein. We present direct measurement of the drug in both binding sites, and correlate pore binding with widespread chemical shift changes that have been interpreted as an indicator of structural rearrangement and functional binding(17, 31).

We used both wild type (WT) M2 and the D21G, D24G double mutant to study drug binding. The use of the double mutant is justified by noting that the mutations occur outside the lipid embedded channel region. In addition, the mutant shows proton conduction and inhibition similar to the WT protein in liposome assays (see Figure S2 in the supporting information). Additionally, in all samples, cysteins 19 and 50 were changed to serine with no loss of function(27) in order to prevent uncontrolled disulphide bond formation. The two mutations from D to G remove ambiguity in potential cross-peaks arising from D44 in the external binding site. Thus, the WT_18–60 sequence used in this work is RSNDSSDPLVVAASIIGILHLILWILDRLFFKSIYRFFEHGLK.

METHODS

The DNP enhanced ZF-TEDOR experiment

In our DNP enhanced ZF-TEDOR experiment, spin polarization is transferred several times before final detection. First, electron polarization on biradicals(32) such as TOTAPOL(33) is transferred to nearby protons via the cross effect(34) under the application of CW microwave irradiation and concurrently transferred to more distant protons via spin diffusion. The microwave irradiation frequency is set to 28.0498 GHz/Tesla to maximize DNP efficiency using TOTAPOL. Next, ¹H polarization is transferred to ¹³C in a cross polarization step. The previous steps serve to increase the initial ¹³C polarization and increase the sensitivity of the experiment. Finally, ZF-TEDOR recoupling(18) is used to transfer carbon polarization to ¹³C-¹⁵N two-spin order, the nitrogen chemical shift is encoded, and the two-spin order is reconverted into ¹³C single quantum coherence for direct detection. It is in this final ZF-TEDOR step that the interaction between ¹³C labeled protein and ¹⁵N labeled drug is observed.

Optimization of Radical Concentration

A low concentration of 4 mM TOTAPOL was used in order to ensure that the radical induced only minimal paramagnetic relaxation effects on the protein. Particularly, reduction in transverse relaxation during long mixing periods could reduce signal intensity. Average carbonyl ¹³C T₂ relaxation times of only ~2 ms were found in a 40 mM TOTAPOL M2 sample with DNP enhancements of 26. A T₂ of 13 to 14 ms was found using 4 mM, similar to the T₂ value at 278 K. These effects are particularly pronounced in this relatively small protein, and appear to be much less significant in previous studies that used 40 mM TOTAPOL to probe details of the active site of bacteriorhodopsin, which is shielded from close approach of radical by a larger (26 kD) protein(11). General optimization of radical concentration for DNP is the subject of ongoing discussion.

M2 Sample Preparation

Pure M2_18–60 peptide was prepared by expression in E. coli as a fusion protein to TrpLe with a His 9 tag, and purified with a nickel affinity column followed by chemical cleavage with cyanogen bromide in 70 % formic acid and final purification on a C4 reverse phase HPLC column(23, 31). The Udorn strain of influenza was chosen, but with C19S and C50S mutations to prevent unwanted disulphide bond formation. Indicated samples additionally contained the D21G and D24G mutations.

M2 samples were prepared for MAS NMR as described previously(31) with modification to cryoprotect the sample in a glassy matrix(35, 36) containing TOTAPOL for DNP. DPhPC lipid (Avanti Polar Lipids) and lyophilized M2_18–60 were dissolved in denaturing buffer (6M guanidine, 40 mM phosphate, 30 mM glutamate, 3 mM sodium azide, pH 7.8, ≥33mg/mL octyl glucoside (OG) detergent) with a lipid to protein ratio of 2:1 by weight. Buffer components were purchased from Sigma, and detergent from Affymetrix. The resulting solution was then dialyzed in a 3.5kD cutoff dialysis cassette (Thermo) against 1 L of sample buffer (40 mM phosphate, 30 mM glutamate, 3 mM sodium azide, pH 7.8) for 7 days with 2 dialysis buffer changes per day. Lipid/M2 bilayers formed a white precipitate after approximately 24 hours. Solid material was pelleted by centrifugation at ~100 000 × g in a Beckmann ultracentrifuge. For pore-bound samples, a fourfold molar excess of ¹⁵N Rmt was added directly to the membrane pellet and incubated for 2 days at room temperature (~22 °C). The pellet was then mixed with 2 mL of DNP buffer (60:40 by volume glycerol-d₈:75 % deuterated sample buffer and 4 mM TOTAPOL(33)) and separated by centrifugation at 450 000 × g for 12 hours. DNP buffer was drained from below the sample and the sample was packed in a 4mm (Revolution NMR) or 3.2 mm (Bruker) sapphire MAS rotor. For pore-unbound samples, the membrane pellet was first mixed with DNP buffer and centrifuged at 450 000 × g. A fourfold molar excess of ¹⁵N Rmt was then added to the sample immediately before packing into a sapphire MAS rotor.

NMR, Hardware, Referencing, and Processing

Standard NMR spectra were recorded using Cambridge Instruments spectrometers (courtesy of D.J. Ruben) operating at 500, 700, and 750 MHz ¹H frequencies and using triple channel Varian 4 mm (500 MHz), Varian 3.2 mm (700 MHz), and Bruker 3.2 mm e-free (750 MHz) probes. DNP NMR spectra were acquired at 211 MHz using a Cambridge Instruments spectrometer and triple channel HCN probe equipped with a 4 mm MAS stator and waveguide for delivery of microwave irradiation from a gyrotron oscillator at 139.66 GHz(37). Additional DNP spectra were recorded using a commercial DNP spectrometer(38) at 400MHz ¹H frequency and 263 GHz for the microwaves (Bruker, Billerica, MA).

Chemical shifts were referenced using the published shifts of adamantane relative to DSS for ¹³C referencing(39) and the IUPAC relative frequency ratios between DSS (¹³C) and liquid ammonia (¹⁵N)(40, 41) for ¹⁵N referencing. DNP spectra were referenced using the known chemical shifts of G34 at 278 K. Additional experimental details such as pulse sequences, NMR field strengths, and spinning frequencies are indicated in the figure captions. Spectra were processed using the NMRPipe(42) software package and visualized and assigned using Sparky (Goddard and Kneller, University of California, San Francisco). Spectra collected at 600 MHz were processed with 150 Hz Gaussian apodization in both ¹³C and ¹⁵N dimensions to improve the signal to noise ratio of the buildup curve. All other TEDOR spectra were processed with a 100 Hz Gaussian apodization window in the ¹³C dimension, and 150 Hz (crogenic DNP) and 30 Hz (278 K) in the ¹⁵N dimension to improve sensitivity. Spectra in Figure 1 were processed with 30 Hz of Gaussian apodization. TPPM(43) Decoupling field strengths of ~100 kHz were used during TEDOR mixing periods, and either 100 kHz or 83 kHz were used during acquisition. Power levels of 50 to 83 kHz were used for ¹³C pulses, and 20 to 40 kHz for ¹⁵N pulses. Further details of the pulse sequences are provided in the supporting information.

Figure 1.

PDSD spectra of M2 acquired under four sample conditions show that chemical shifts are unperturbed by 60% glycerol if glycerol is the last component added to the sample. In a) and b) are shown cross peaks of G34 and H37, respectively in uniformly ¹³C labeled samples. The first three panels were recorded using a 700 MHz instrument with a Varian 3.2 mm triple channel probe spinning at 12.5 kHz. The fourth panel (4× Rmt then 60% glycerol) was recorded at 750 MHz with a triple channel Bruker e-free probe spinning at 13.4 kHz. These spectra of the D21G and D24G double mutant were recorded at ~278 K using 15 ms of mixing. Peaks of G34 and H37 are displayed because of the excellent resolution in these regions of the spectrum, and a clear change in frequency upon drug binding. Peaks other than those from G34 and H37 are indicated with an asterisk.

RESULTS

The DNP experiments are performed at low temperatures of ~80–100 K where ice crystals would form from buffers commonly used for room temperature experiments. The samples are thus cryoprotected using 60% glycerol by volume. We show in Figure 1 that the chemical shifts are unperturbed by the addition of 60% glycerol, both for the apo and the functionally-bound states at ~278 K. The spectra provide a fingerprint that can be used to distinguish between apo M2 and functionally-bound M2, where functional binding is determined by large chemical shift changes (17, 22, 31). Previous reports found that addition of the aminoadamantyl drugs Rmt and Amt cause widespread chemical shift changes of up to several ppm to occur and a functionally drug-bound spectrum is easily distinguished from an apo spectrum. Based on these fingerprint spectra, we conclude that the glycerol does not cause any significant change in these states of the protein. However, we find that glycerol increases the energy barrier for functional drug binding. In samples with drug added before glycerol, the drug-bound set of shifts was observed. In samples with drug added after glycerol, the apo set of peaks was observed. Since the final composition of both samples is the same, we conclude that the difference is attributed to kinetically trapping the apo state and that the barrier for drug binding is increased by glycerol.

Spectra in Figure 1 were assigned using ZF-TEDOR(18, 19) and PDSD(44, 45) correlation experiments as was reported for WT_18–60(31), and by observation of only minor differences in chemical shift between WT_18–60 and the D21G and D24G double mutant spectra of Figure 1. As with WT, we observe membrane embedded resonances from approximately residue 25 to 50 at 278 K. These observed residues span both proposed binding sites. Residue 24 appears weakly in some spectra, and residues 18–23 and 54–60 are not detected due to unfavorable mobility of this part of the protein. Spectra recorded at low temperature and with DNP were assigned based on the room temperature resonances for G34, and by using the observed range of chemical shifts reported in the BioMagResBank(46) for cross-peaks that do not show up at high temperature. These low temperature cross-peaks could not be uniquely assigned; therefore all possible assignments are indicated.

In order to observe a dipolar coupling between uniformly ¹³C labeled protein and ¹⁵N labeled inhibitor Rmt, we used a ¹³C-¹⁵N ZF-TEDOR experiment with 8.8 ms of mixing. Near room temperature (~278 K), the spectrum shows only two correlations to ¹⁵N labeled drug after 23 days of acquisition (Figure 2, red). In contrast, DNP enhanced TEDOR spectra with 8.7 ms mixing at low (80–105 K) temperatures showed several additional cross-peaks (Figures 2–3, blue) and required only 2 days of acquisition due to the reduction in temperature and a signal enhancement factor of 11. Assignments consistent with the observed cross-peaks are indicated in the figures, and clearly show that at room temperature the drug is observed in the pore near G34 and A30. The G34 cross-peak is unambiguously assigned at 278 K, based on the known unique resonances of G34 at this temperature. The A30 cross-peak is assigned by ruling out the only other alanine, A29, as a possible assignment because this residue is found on the outside of the channel far from G34, and the simplest interpretation of the data is that we are observing a single binding site in the pore.

Figure 2.

ZF-TEDOR spectra acquired at 278 K (red) show one set of cross peaks, and those acquired at ~90 K using DNP (blue) show additional peaks. In red is shown an 8.8 ms TEDOR experiment acquired at 500 MHz and 10 kHz MAS with a sample temperature of 278 K. In blue is shown an 8.72 ms TEDOR experiment acquired at 211 MHz and 4.587 kHz MAS using DNP at 90 K. The 278 K spectrum was recorded in 23 days and the 90 K DNP spectrum in 2 days. Both spectra were acquired using the same sample of D21G-D24G M2 packed in a 4 mm sapphire rotor. ¹⁵N labeled Rmt was added before glycerol, and binding reached ~70% (as indicated by fingerprint spectra such as in Figure 1) before glycerol kinetically trapped the remaining 30% in the pore-unbound state. Possible assignments are indicated.

Figure 3.

ZF TEDOR spectra show that the pore binding site is correlated with chemical shift changes. A 400 MHz DNP enhanced TEDOR spectrum (a) with 12.5 ms mixing is shown in blue. The wild type M2 sequence was used, and ~30% of the sample was drug bound as indicated by chemical shifts. In b), a similar spectrum is shown of D21G, D24G M2 but with the protein kinetically trapped in the apo state. Notably, the pore site at G34 is not detected in the functionally unbound sample shown in blue in (b). In red, the TEDOR spectrum acquired at ~278 K, 500 MHz, and 8.8 ms mixing of D21G, D24G M2 is presented for comparison. All possible assignments are listed for each peak in blue (DNP) and red (non-DNP).

At low temperature and using DNP, the drug is also observed on the periphery of the protein, consistent with the external site near D44 that was previously observed using solution NMR. In addition, cross-peaks are observed that are consistent with drug associating to E56 or N20, which may be another site of weak drug-protein association. The sample in Figure 2 was approximately 70% functionally bound before glycerol was added and the remaining 30% was trapped in the apo state. The amount of functionally bound protein is less than 100% because it can take days for drug to penetrate and fully bind to the thick membrane pellet. Once glycerol is added, further binding is kinetically prevented. We used spectra similar to those shown in Figure 1, and the known chemical shifts of the apo and bound M2 to determine the extent of drug binding.

In Figure 3, we show that if the protein is kinetically trapped in the apo state with drug present in the sample, the drug is observed only outside the pore. Isoleucine and leucine residues were ruled out as possible assignments by reverse labeling of these residue types with ¹²C and observation of the same peaks (see Figure S1 in the supporting information).

The resolution of the DNP enhanced spectrum at 211 MHz is significantly reduced compared to the 278 K spectrum acquired at 500 MHz, but the DNP enhanced spectrum has much higher signal to noise, and the contours are displayed at a lower level. Therefore, the 2D spectra should not be used to infer resolution and we include 1D slices below the 2D spectrum to show the inherent ¹³C resolution in the spectra. We also see that the Ala Cβ resonance does not appear in the DNP spectrum. This is typical of −CH₃ groups at temperatures of ~100–200 K and results from three fold hopping at a rate that interferes with ¹H decoupling. Acquisition at temperatures where the methyl dynamics are much slower than the decoupling field results in a return of signal intensity and ²H labeling of the methyl group has a similar effect by reducing the couplings.(47)

In the 211 MHz DNP enhanced spectrum (Figure 1), the G34Cα resonance is not resolved, but appears as a shoulder on a larger nearby peak. In Figure 2a, we show a similar DNP spectrum, this time acquired at 400 MHz and applying 12.5 ms of mixing. The spectrum clearly shows a peak at G34Cα even though this sample had not equilibrated before addition of glycerol, and only ~1/3 of the sample showed drug binding as measured by the chemical shift changes. The spectrum in Figure 3b shows that when glycerol is added before the inhibitor, only the external binding site(s) are observed, once again indicating that the protein is trapped in the pore-unbound state by 60% glycerol. Pore binding is therefore correlated with chemical shift changes. Coupled with a recent report showing that the drug resistant S31N mutant of M2 does not exhibit chemical shift changes with the addition of rmt,(48) the evidence points convincingly to the pore site as responsible for inhibition in agreement with recent reports(26, 30). Despite other glycines in the sequence of M2, the glycine cross-peaks in the DNP spectra can be uniquely assigned to the pore glycine, G34, because glycine peaks are missing from the spectrum when the protein is trapped in the pore-unbound state; the weaker association of the drug on the periphery of the channel does not result in detection of TEDOR cross-peaks to glycine Cαs.

With this unique ¹⁵N Rmt to G34 Cα assignment under DNP conditions, we can proceed with standard MAS NMR experiments designed to measure internuclear distances. In Fig 4, we show the results of a ZF-TEDOR buildup curve recorded at 600 MHz with DNP using uniformly ¹³C labeled WT M2_18–60 in a sample with nearly saturated binding to the pore. The maximum signal to noise ratio in the DNP enhanced TEDOR experiment was 15 at 11.52 ms of mixing, which is sufficient to measure a distance using an analytical fitting procedure described previously(18). The data are inconsistent with a drug amine centered in the channel, and the fit distance depends somewhat on whether the amine is near two or one helices. A fit of 4.3 or 4.0 ±0.2 Å was found for fits with two and one G34 Cαs, respectively. Details of the fit are described below.

Figure 4.

ZF-TEDOR spectra collected at 600 MHz with DNP are contoured at two mixing times in (a), and displayed for the G34 cross-peak as a function of mixing time in (b). At 3.84 ms mixing (a, left), one bond cross-peaks are detected due to the natural abundance of ~1% ¹³C Rmt drug, and ~0.36% ¹⁵N protein. By 11.52 ms (a, right), these natural abundance cross-peaks have decayed and only the cross-peak of interest remains. In (b), the G34 cross-peak intensity is shown as a function of mixing time (points ± the spectrum rms noise level), with the best-fit simulation as described in the text (smooth curve). The fit resulted in a distance of 4.0 Å ± 0.2 Å. The G34 N-Cα cross-peak was simultaneously fit (b, right) to reduce the degrees of freedom in the fit. A low natural abundance of ¹⁵N (~0.36%) accounts for the reduced signal of this curve. The spectra were recorded in ~4 days at a spinning frequency of 12.5 kHz and a temperature of ~100 K. The DNP enhancement factor was 2.5.

An overall scaling factor, λ, is used in the fit to account for an unknown cross polarization intensity and experimental imperfections. Fortunately, we can reduce the number of free parameters in the fit by using the known distance associated with a 1-bond cross-peak from G34Cα to G34N to determine λ. A natural abundance of 0.36 percent for ¹⁵N was used to scale this peak in the fit, and both the short distance of 1.5 Å, and the long distance of interest were simulated simultaneously using the same value for λ. The fit parameter Γ₂ is an exponential relaxation term that is dominated by ¹³C transverse relaxation and was therefore fit using one value for both curves. The average ¹³C Γ₂ of the sample was determined to be above 100 s⁻¹ using an echo experiment, justifying fitting over a range of Γ₂ values between 50 and 150 s⁻¹, as done for crystalline peptides(18). The one bond J coupling of G34 was varied over a wide range of 40 to 60 Hz in the fit, but influenced the fit distance by less than 0.1 Å. Typical values of ¹J_C′−Cα are about 55 Hz(49). Two percent of the sample was assumed to have no J coupling based on an estimate of 98% enrichment of ¹³C using 99% ¹³C glucose and a ¹²C starter culture. A Monte Carlo approach was used to estimate the measurement error by refitting 1000 times with the addition of the appropriate noise level of the spectrum(18) to each curve. The resulting distribution was used to calculate the rms deviation. All error estimates are reported at 1 rmsd.

The final consideration in properly fitting the distance is to determine how many of the helices in the tetramer are close enough to contribute to the cross-peak. If we assume all four are equidistant from the drug amine, then a fit distance of 4.8 ±0.2 Å is found. With the drug amine positioned between G34 Cα and A30 Cβ, this distance is incompatible with the solution constraints, which allow a minimum distance of about 5.5 Å, and certainly incompatible with the more loosely packed diffraction structures. If we fit the data with two G34 Cαs equidistant from the drug amine, and the other two at a reasonable distance of 6.5 Å or above, we determine a fit distance of 4.3 ± 0.2 Å. Finally, if we assume the ¹⁵N interacts primarily with one helix, with the other G34 Cαs at reasonable distances of 5.5 Å, 6.5 Å, and 6.5 Å, the fit distance is 4.0 ± 0.2 Å. This fit is shown in Figure 4 (b).

In order to visualize the pore binding site for which we found unambiguous constraints, we have performed a simulated annealing using the program XPLOR-NIH(50). In addition to the two new constraints from ZF-TEDOR at ~278 K and with cryogenic DNP, we used structural constraints from a solution NMR structure(23) and a constraint based on previously reported ²H-¹³C REDOR experiments.(25, 51) Based on the DNP enhanced buildup curve, an upper limit of 4.5 Å was allowed for the Rmt amine to G34 Cα. Similar constraints, entered in the calculation as 3.5 ±1 Å, were entered for ¹⁵N-Rmt to A30 Cβ based on the similar intensity of this cross-peak at room temperature. Constraints of 3.5 ±1 Å were also applied between the V27 Cγ1 and Rmt CH₂ groups farthest from the amine in order to orient the adamantyl cage consistent with a recent report using the 22–46 construct(51). Drug constraints were introduced to only one of the four strands of the protein. Given the symmetry of the channel, and the fact that we observe two sets of chemical shifts, the drug amine may hop between 2 or 4 of the helices and thus prevent breaking of the two-fold symmetry of the channel. A representative result from the simulated annealing is shown in Figure 5.

Figure 5.

A structural model of the pore binding site is shown using distance constraints from DNP enhanced TEDOR experiments, 278 K TEDOR experiments, and previously reported data. The structural model was calculated using XPLOR-NIH and converged to a set of similar structures with 0 violations. A representative structure is shown with drug in green and protein in blue. Three distances to the drug ¹⁵N are shown, two of which (A30 Cβ and G34 Cα) are the result of TEDOR measurements. The third distance is a result of the structure calculation and suggests a possible hydrogen bond between the drug amine and the protein at A30 carbonyl oxygen. The constraints deposited in the protein data bank for the solution NMR structure 2RLF(24) were used to define the tetrameric assembly. Distance constraints of 3.5 ±1 Å were introduced between the drug ¹⁵N and A30 Cβ and between drug ¹⁵N and G34Cα based on TEDOR spectra. Finally, constraints of 3.5 ±1 Å were introduced between the V27 Cγ1 and Rmt CH₂ groups farthest from the amine in order to orient the adamantyl cage consistent with previously reports.

DISCUSSION

In previous experiments, two binding sites have been observed, a pore site in M2_22–46 and M2_22–62, and an external site in M2_18–60, M2_22–46. We have confirmed the internal site in the M2_18–60 construct using DNP MAS at 100 K. Additionally, a drug protein association is observed on the periphery of the channel, consistent with the external site, however, the location and affinity are not determined. The pore site was found to have a significantly higher barrier, causing the protein to be kinetically trapped near room temperature in the pore-unbound state using 60% glycerol. Pore-bound Rmt was observed in a standard ZF-TEDOR experiment at ~278 K, indicating that the drug is exchanging with the free population slower than the millisecond timescale. In contrast, the external association was only observed at the low temperatures used in DNP experiments, indicating that drug exchange likely results in loss of signal at 278 K. The external site is observed even when the protein has been kinetically trapped in the pore-unbound state (as determined by chemical shift changes) indicating that this site has a low barrier, is not responsible for chemical shift changes, and is most likely not responsible for pharmacological binding.

Measurements of the external binding site(s) of the protein appear to have some specificity even though they are not unambiguously assigned. For example, we see a cross-peak to E56 or N20, but not to neighboring H57 and G58, or S19, S22, and S23 residues. A completely nonspecific interaction should result in some detection of these external glycine and serine residues. Still, partitioning of the drug into the interfacial region of the membrane may significantly enhance population of these sites. Whether the observation of specificity at these weaker sites requires a slow cooling rate and annealing of the protein remains an open to further investigation.

Although binding to the pore of the channel has been observed previously, the present work is the first to our knowledge to directly determine the position of the drug amine group between residues 30 and 34. A previous report in bilayers determined the correct orientation of the drug in the channel, but placed the amine group about a half helical turn towards the carboxy terminus of the protein compared to our measurements(51). Additionally, our data suggests a hydrogen bond between the drug amine and the A30 carbonyl. Out of 30 calculated structures that did not violate any of the constraints, the distance between the drug amine nitrogen and the A30 carbonyl oxygen was found to vary between 2.4 and 3.5 Å, suggesting a hydrogen bond between drug and protein at this location. Though this agrees with a previously observed conformational heterogeneity at A30's hydrogen bonding donor G34 among drug bound and apo samples(52), and with a recent solution NMR structure of a chimera protein containing C-terminal residues from Influenza-B M2(30), it is the first time this specific interaction has been observed using WT protein.

The structural model presented here is consistent with previous measurements on the Rmt tilt angle in the pore of M2_22–46(51) determined using measurements of the scaled ²H quadrupole splitting of ²H labeled Rmt. The amine was found to point towards the C-terminal end of the channel with a 13° angle between the bilayer normal and the adamantly group. Rmt is tilted by up to ~35° in the structural model shown in Figure 5, with most of the ensemble clustered within about 12°. A large distribution in the tilt angle is not surprising because we have not made measurements to determine this angle, with the adamantyl cage instead being positioned by steric crowding within the channel. Interestingly, the tilt angle of the related drug Amt was previously found to be a much larger 37°(25), suggesting that the shorter link between the adamantyl group and the amine in Amt might force the adamatyl group to rotate so that both the amine and the adamantyl cage can occupy approximately the same locations in the channel for both drugs (see Figure S3 in the supporting information). The evidence for a drug protein hydrogen bond at A30 carbonyl (both the tilt angle of Amt and the TEDOR measurements on Rmt) could help inform molecular dynamics simulations of this channel that have instead proposed hydrogen bonding to water and a small tilt angle(53) or even found the amine group pointing towards the nitrogen terminus of the protein(54).

CONCLUSIONS

We have shown that DNP can be used with MAS for the identification of ligand-protein interactions. The low temperatures used for DNP quenches dynamic processes that interfere with dipolar recoupling experiments, particularly for weakly bound ligands. These effects were demonstrated in M2, in which the pore binding site is evident at 278 K, and weaker binding to the periphery of the protein is only observed at cryogenic temperatures. In addition, we found that glycerol can be used to kinetically prevent the drug from entering the pore, thereby allowing us to make samples with the pore unoccupied, but with drug bound to the external site. With these samples, we have correlated chemical shift changes with binding in the pore. In addition, we located the position of the functionally important drug amine group in relation to specific protein ¹³C atoms, placing the amine close enough to A30 carbonyl to form a hydrogen bond. Further experiments could include BASE TEDOR experiments(18) to improve precision of the distance measurements, as well as unambiguous assignment experiments to precisely determine the external binding site(s).

ABBREVIATIONS

CW

continuous wave

DNP

dynamic nuclear polarization

HPLC

high pressure liquid chromatography

MAS

magic angle spinning

NMR

nuclear magnetic resonance

Rmt

rimantadine

Amt

amantadine

ZF-TEDOR

z-filtered transferred echo double resonance

rms

root mean square

TPPM

two pulse phase modulation