Strategies for acquisition of resonance assignment spectra of highly dynamic membrane proteins: a GPCR case study

Abstract

In protein nuclear magnetic resonance (NMR), chemical shift assignment provides a wealth of information. However, acquisition of high-quality solid-state NMR spectra depends on protein-specific dynamics. For membrane proteins, bilayer heterogeneity further complicates this observation. Since the efficiency of cross-polarization transfer is strongly entwined with protein dynamics, optimal temperatures for spectral sensitivity and resolution will depend not only on inherent protein dynamics, but temperature-dependent phase properties of the bilayer environment. We acquired 1-, 2-, and 3D homo- and heteronuclear experiments of the chemokine receptor CCR3 in a 7:3 phosphatidylcholine:cholesterol lipid environment. 1D direct polarization, cross polarization (CP), and T₂’ experiments indicate sample temperatures below − 25 °C facilitate higher CP enhancement and longer-lived transverse relaxation times. T_1rho experiments indicate intermediate timescales are minimized below a sample temperature of − 20 °C. 2D DCP NCA experiments indicated optimal CP efficiency and resolution at a sample temperature of − 30 °C, corroborated by linewidth analysis in 3D NCACX at − 30 °C compared to − 5 °C. This optimal temperature is concluded to be directly related the lipid phase transition, measured to be between − 20 and 15 °C based on rINEPT signal of all-trans and trans-gauche lipid acyl conformations. Our results have critical implications in acquisition of SSNMR membrane protein assignment spectra, as we hypothesize that different lipid compositions with different phase transition properties influence protein dynamics and therefore the optimal acquisition temperature.

Introduction

Nuclear Magnetic Resonance (NMR) has continually evolved as a technique to probe protein structure and dynamics (Hu et al. 2021). While solution-state NMR is predominantly applied to water soluble globular proteins, solid-state NMR (SSNMR) evolved to be the technique of choice for amyloid fibrils (Lu et al. 2013; Wickramasinghe et al. 2021; Do et al. 2021; Tuttle et al. 2016) and membrane proteins in a native-like lipid environment (Krug et al. 2020; Jekhmane et al. 2019; Gelenter et al. 2021; Vogel, et al. 2020; Howarth and McDermott 2020; Joedicke et al. 2018). Indeed, use of NMR to study G protein Coupled Receptors (GPCRs) has been critical in understanding the structure–function relationship (Park et al. 2012) involving receptor activation (Ahuja et al. 2009; Kimata et al. 2011b, 2016a, b; Mertz et al. 2012; Struts et al. 2011a; Kubatova et al. 2020; Ray et al. 2023) and conformational dynamics (Krug et al. 2020; Vogel et al. 2020; Thakur et al. 2023). Typically, structure determination via SSNMR first requires chemical shift assignments. This can be quite an undertaking, and usually involves several two-dimensional (2D) and 3D heteronuclear correlation experiments making use of multiple cross polarization steps (CP) (Pines et al. 1972). Originally presented by Schaefer et al. (1979), the principles of broadband double cross-polarization (DCP) were later modified by Baldus et al. to be frequency selective and allow for preferential polarization transfer from the backbone amide to either the carbonyl or Cα, termed Spectrally Induced Filtering in Combination with Cross Polarization, or SPECIFIC-CP (1998). These NCA and NCO heteronuclear correlation experiments are often paired with an additional ¹³C-¹³C mixing step, for example Dipolar Assisted Rotational Resonance (DARR) (Takegoshi et al. 2001), for sidechain identification. Thus, the aspiring spectroscopist can identify a given residue and its connectivity in the primary amino acid sequence. Such an endeavor is typically referred to as a backbone walk. Once chemical shift assignments have been achieved, they can be leveraged for atom–atom distance restraints for protein structure calculation with programs like XPLOR-NIH (Schwieters et al. 2003, 2006).

The ability to probe both structure and dynamics with SSNMR is perhaps the techniques biggest strength, but the influence of temperature-sensitive dynamics on acquisition of protein spectra is not often fully explored. This is especially true for NCO/NCA DCP experiments, as they rely on the dipolar coupling for both polarization transfer steps and the corresponding relaxation in the rotating frame, T_1rho. These observables are heavily attenuated with increasing molecular dynamics. This is especially true for membrane proteins in native-like lipid environments, where temperature-dependent lipid phase properties introduce an additional component to acquisition of high-quality spectra. The effects of bilayer phase properties on spectral transfer efficiency for detection of lipid signals is well characterized (Guo and Hamilton 1995; Warschawski and Devaux 2000; Nowacka et al. 2010, 2013). The influence of bilayer phase on membrane protein spectral resolution is also known, as systemic rigidity in the gel phase is imparted to the protein, facilitating more efficient CP than in the liquid crystalline phase. For example, the drug transporter EmrE was found to have the most DCP efficiency when model phosphatidylcholine bilayers were in the gel phase ~ 15 °C below the phase transition (Banigan et al. 2015). Thus, intrinsic characteristics of the entire system that ultimately dictate protein dynamic timescales should be considered.

Here, we present a comprehensive dynamics investigation of the C–C motif chemokine receptor 3 (CCR3) in cholesterol-containing phospholipid bilayers. CCR3 is a class A GPCR whose ligand and G protein affinity were shown to respond to the presence of cholesterol in a dose-dependent manner (van Aalst and Wylie 2021). For the proposed DCP experiments, we would traditionally select a set temperature of − 20 °C, but the characteristics of the system may not yield the most efficient DCP at this temperature due to protein-specific relaxation parameters that are influenced by the properties of the bilayer, and therefore temperature. Thus, we selected a cholesterol-phosphatidylcholine environment to answer the following question: Considering the dynamics of CCR3 in this environment, what temperature should DCP be performed to yield the most efficient transfer of magnetization and the highest spectral resolution?

We acquired direct polarization (DP), CP, T₂’, and T_1rho experiments to quantify the effects of temperature on the dynamics of CCR3 as they pertain to the proposed DCP experiments. It was found that CCR3 was quite dynamic, evidenced by limited CP enhancement and relative short relaxation time constants. While ¹⁵N and ¹³C CP experiments as a function of temperature resulted in the expected signal attenuation above temperatures of 0 °C, internal dynamics measured by T₂’ persisted in some cases to as low as − 20 °C. Similar observations were made from T1_1rho experiments, where time constants do not become favorable for DCP until − 25 °C and below. We then acquired NCA and ¹³C-¹³C DARR spectra at different temperatures, which hinted at differences in both signal-to-noise (s/n) and resolution based on bulk linewidth analysis. In aggregate, our results indicated that the optimal signal and resolution would be achieved at a sample temperature of − 30 °C. We then acquired NCACX experiments at this temperature, confirming that good signal and resolution are achieved despite inherent GPCR dynamics. Finally, we acquired ¹³C-detected refocused Insensitive Nuclei Enhanced by Polarization Transfer (rINEPT) (Alonso and Massiot 2003; Elena et al. 2005) temperature series experiments to correlate our observations with the phase of bilayer, observing that our optimal temperature falls ~ 10 °C below the start of the phase transition at − 20 °C. In aggregate, our results illustrate the necessity to understand protein-specific dynamics for SSNMR experimentation and identify the potential for rINEPT experiments to easily identify optimal membrane protein DCP temperatures based on the phase transition range of the system.

Materials and methods

CCR3 expression and purification

CCR3 was expressed and purified as previously described (van Aalst and Wylie 2021; van Aalst et al. 2023). Briefly, our Maltose Binding Protein-CCR3 construct was expressed in M9 minimal media supplemented with ¹³C glucose and ¹⁵N Ammonium Chloride (Cambridge Isotopes Laboratories) in C43 (DE3) cells at 18 °C for 24 h. Cell pellets were then harvested and stored at − 80 °C until needed. Cells were lysed via homogenization and CCR3 was detergent extracted with 20 mM n-Dodecyl-β-D-Maltoside (DDM, Inalco) and 2 mM Cholesteryl Hemisuccinate Tris-salt (CHS, Anatrace)). After 3 h, cell debris was removed via ultracentrifugation at 125,000 g for 1 h. The supernatant was collected and MBP-CCR3 was separated from the lysate using a Ni²⁺ affinity column (HisTrap^™, Cytiva) and eluted with 250 mM Imidazole. Imidazole was removed using a desalting column (Cytiva) and Tobacco Etch Virus (TEV) protease, 1 mM EDTA, and 1 mM DTT were added to remove the MBP tag overnight at 4 °C. EDTA was removed via desalting column prior to collection of CCR3 in the flow-through of a second Ni²⁺ affinity column, while the MBP tag and TEV with His-tags were removed via interaction with the column. CCR3 was concentrated using a 30 kDa cutoff membrane filter (Millipore) and separated from aggregate using a 16/600 prep-grade 200 SEC column (Cytiva). The ligand C–C motif chemokine ligand 11 (CCL11) was expressed and purified as previously described (van Aalst and Wylie 2021; van Aalst et al. 2023). Ligand-binding activity of CCR3 was confirmed using a fluorescence polarization assay as previously described (van Aalst and Wylie 2021).

1-pamlitoyl-2-oleoyl-phosphatidylcholine (PC, Avanti Polar Lipids) and Cholesterol (Sigma) were solvated in Chloroform at 10 mg/ml then combined at a molar ratio of 7:3 PC:cholesterol (approximately 30% w/w cholesterol) and dried overnight in vacuo. The dried film was solvated in NMR buffer (20 mM HEPES pH 7.5, 50 mM KCl, 1 mM EDTA) supplemented with 25 mM 3-[(3-Cholamidopropyl) dimethylammonio]-1-propanesulfonate (CHAPS, Anatrace) via mild sonication. Solvated films were added to CCR3 at a 1:1 w/w lipid:protein ratio and annealed for 3 h at room temperature before addition of Bio-Bead SM-2 Resin (Bio-Rad) for detergent removal. The sample was nutated with twice daily additions of Bio-beads for approximately 10–12 days. Complete detergent removal was verified via manual agitation prior to sample collection via ultracentrifugation at approximately 400,000 g for 2 h. The pellet was collected, freeze-thawed via liquid N₂, then centrifuged at 16,250 g to pack the pellet. Excess buffer was removed through consecutive freeze-spin cycles, and finally the wet pellet was packed into a 3.2 mm standard wall PENCIL SSNMR rotor. The lipid-only sample was resuspended in chloroform at the same lipid ratios and dried overnight in the same way. Liposomes were formed by sonication in NMR buffer prior to removal of buffer and packing in a 3.2 mm standard wall PENCIL SSNMR rotor in the same way as the protein-containing sample.

Acquisition of MAS SSNMR spectra

Magic-angle Spinning (MAS)-SSNMR experiments were performed on a 600 MHz Agilent DD2 spectrometer equipped with a 3.2 mm HCN BALUN probe (Agilent Technologies). The ¹³C chemical shift was indirectly referenced to DSS using the downfield adamantane peak at 40.48 ppm (Morcombe and Zilm 2003). Spectra were acquired at a MAS rate of 12 kHz and a recycle delay of 2 s for all experiments. The ¹³C 90° pulse width was 2.7 μs, the ¹H 90° pulse width was 3.1 μs, and the ¹⁵N 90° pulse width was 5.0 μs. Temperature series experiments were acquired at set temperatures of − 50 to 20 °C in steps of 5 °C. This range of set temperatures corresponds to sample temperatures of − 40 to 35 °C, which we previously calibrated for this probe (van Aalst et al. 2022) based on literature corrections for sample heating under decoupling (Gottlieb et al. 1997) and spinning rate (Guan and Stark 2010). All subsequent references to temperature in this text refer to these calibrated sample temperatures. At each temperature, the sample was allowed to equilibrate for ~ 5 min before probe retuning and subsequent spectra were acquired. Tangent-ramped CP spinlock power was set to 57.1 kHz for ¹H and 68.4 kHz for ¹³C during ¹H-¹³C CP. 35.4 kHz for ¹H and 45.7 kHz for ¹⁵N were used during ¹H-¹⁵N CP. CP contact time was 1.15 ms for ¹H-¹³C and 1.0 ms for ¹H-¹⁵N. SPINAL-64 decoupling (Fung et al. 2000) power was set to 80.3 kHz with a pulse width of 5.3 μs for ¹⁵N and 5.6 μs for ¹³C experiments. All CP and DP 1D experiments were acquired using these parameters with 1024 scans.

T₂’ spectra were acquired for both ¹³C and ¹⁵N with the Hahn spin-echo period arrayed in steps of 166 μs (2 × the rotor period) to a maximum of 8.30 ms with 16 scans per row for ¹³C and 13.28 ms with 64 scans per row for ¹⁵N. T_1rho experiments were acquired by arraying the length of the spin lock in steps of 166 μs from 0 to 16.666 ms. Field strength for ¹⁵N 3/2, 5/2 and 7/2 ω_r. conditions were 14 kHz, 28 kHz, and 42 kHz, respectively and for ¹³C were 16 kHz and 32 kHz, respectively for 3/2 and 5/2 ω_r. conditions. All T_1rho experiments were acquired with 16 scans per row. rINEPT spectra were acquired using the parameters outlined above with 256 scans. The first and second delay periods were set to 1.6 and 0.95 ms, respectively. 2D DARR spectra were acquired with 32 scans per row and a sweep width of 48 kHz with 320 complex points in the indirect dimension.

NCA experiments at − 30 °C using 3/2 ω_r. ¹³C and 5/2 ω_r ¹⁵N spin lock field strengths were acquired with optimum N-CA transfer of 16.4 kHz for ¹³C and 26.9 kHz for ¹⁵N with a contact time of 3 ms. All other NCA experiments used 5/2 ω_r ¹³C and 7/2 ω_r ¹⁵N spin lock field strengths. At − 30 °C, spectra were acquired with optimum N-CA transfer at 25.9 kHz for ¹³C and 36.5 kHz for ¹⁵N with a contact time of 3.5 ms. At − 5 °C, spectra were acquired with optimum N-CA transfer at 29.9 kHz for ¹³C and 40.3 kHz for ¹⁵N with a contact time of 4 ms. At 15 °C, spectra were acquired with optimum N-CA transfer at 30.1 kHz for ¹³C and 39.9 kHz for ¹⁵N with a contact time of 5 ms. All NCA spectra were acquired with 64 scans per row and a sweep width of 24 kHz with 160 complex points in the indirect dimension. NCACX spectra were acquired with Non-Uniform Sampling (NUS) of 25% in each indirect dimension. Schedules were derived from T₂’ values described below and generated with NUS-tool. Spectra were acquired with 5/2 ω_r ¹³C and 7/2 ω_r ¹⁵N spin lock field strengths of 25.3 kHz and 36.5 kHz, respectively, at − 30 °C and 29.9 kHz for ¹³C and 40.4 kHz for ¹⁵N at − 5 °C, with 32 scans per row. ¹⁵N-¹³C CP contact time was 6.0 ms. The sweep width was 6,000 Hz for ¹⁵N and 12,000 Hz for ¹³C indirect dimensions. 20 and 84 complex points were acquired for the ¹⁵N and ¹³C indirect dimensions, respectively. The homonuclear mixing time was set to 25 ms. For all NCA and NCACX experiments, the carrier was placed at approximately 55 ppm to facilitate selective transfer. Acquisition parameters for 2D and 3D heteronuclear experiments are summarized in Table 1.

Table 1.

Acquisition parameters for multi-dimensional heteronuclear SSNMR experiments. Unless otherwise stated, each experiment made use of 5/2 ω_r ¹³C and 7/2 ω_r ¹⁵N spin lock field strengths

	13C spin lock field strength (kHz)	15N spin lock field strength (kHz)	15N-13C CP contact time (ms)	Indirect dimension sweep width	Indirect dimension complex points
2D NCA, − 30 °C (3/2 13C, 5/2 15N)	16.4	26.9	3.0	24 kHz	160 points
2D NCA, − 30 °C	25.9	36.5	3.5	24 kHz	160 points
2D NCA, − 5 °C	29.9	40.3	4.0	24 kHz	160 points
2D NCA, 15 °C	30.1	39.9	5.0	24 kHz	160 points
3D NCACX, − 30 °C	25.3	36.5	6.0	6 kHz for 15N, 12 kHz for 13C	20 points for 15N, 84 points for 13C
3D NCACX, − 5 °C	29.9	40.4	6.0	6 kHz for 15N, 12 kHz for 13C	20 points for 15N, 84 points for 13C

NMR data analysis

All spectra were processed using NMRPipe (Delaglio et al. 1995), with the SMILE reconstruction package for NUS data (Ying et al. 2017). 1D spectra were fit using the nonlinear spectral modeling package in NMRpipe with NMRbox computational resources (Maciejewski et al. 2017). Error for a given peak at temperature i was fit as previously described (van Aalst et al. 2022):

$$Normalized1DErro{r}_i=Nois{e}_i\ast \frac{{\left(\frac{FittingError}{FitPeakHeight}\right)}_i}{FitPeakIntensit{y}_i}$$

where Fitting error:Fit Peak Height is the ratio of the spectral modeling package fitting error calculated at each temperature ‘i’ used to modify the estimated spectral noise Noise_i, normalized to the peaks’ intensity. Data was fit using NMR-glue (Helmus and Jaroniec 2013) to an exponential function (Ghosh and Weliky 2021; Gelenter et al. 2021; Thiessen et al. 2018):

$$\frac{I_t}{I_0}=a\ast ex{p}^{-t/Tx}+c$$

where I_t/I₀ is normalized signal intensity, a is the preexponential factor, t is the arrayed decay time in ms, T_x is the time constant for T₂’ or T_1rho, and c is a constant. 2D and 3D spectra were analyzed using NMRFAM-SPARKY (Kneller and Kuntz 1993; Lee et al. 2015), where the noise floor σ is the median of 10,000 estimates. 1D slices from 3D experiments were extracted for analysis in NMRPipe.

Results and discussion

Considering overall cross-polarization signal

Previously, we showed that increased concentration of membrane cholesterol led to increased affinity for the endogenous chemokine C–C motif ligand 11 (CCL11) using a fluorescence polarization (FP) assay (van Aalst and Wylie 2021). Using unassigned chemical shift data, we then filtered atomistic molecular dynamics simulations to provide insight into how cholesterol biases conformational sampling of CCR3 to be more likely to interact with the ligand (van Aalst et al. 2023). To again show functionality, we expressed CCL11 and performed the FP ligand binding assay to good effect (Fig. S1). While previous work was intriguing, further experiments are required to fully understand cholesterol-CCR3 interactions and implications in receptor physiology. To this end, chemical shift assignment is necessary. However, CCR3 is observed to be quite dynamic which adversely affected spectral sensitivity and resolution (van Aalst et al. 2023). Before committing to resonance assignment spectra, we first acquire an exhaustive set of temperature series experiments to identify conditions for appropriately sensitive spectra.

To begin this assessment, we first look at cross-polarizability of the individual components comprising the NCA and NCO experiments. CP efficiency depends in part on dipolar couplings, which are attenuated as a function of internal system dynamics (van Aalst et al. 2022). Temperature effects will therefore play a critical role in CP efficiency. To investigate, ¹⁵N and ¹³C 1D spectra were acquired in a set temperature range from − 50 to 20 °C, corresponding to a dynamic sample temperature range of − 40 to 35 °C after correction for sample heating due to decoupling and spinning rate (see Materials and Methods). Here, and throughout the rest of this study, 1D spectral regions were integrated to obtain peak volume presented in temperature series plots. To analyze bulk carbonyl CO signal and dynamics, the region 174–177 ppm was fit. For the bulk Cα, the main peak between 55 and 57 ppm was probed. Lastly, the maximum region of the backbone amide ¹⁵N peak from 117 to 119 ppm was integrated.

An expected, temperature dependence of ¹³C CO and Cα signal is observed (Fig. 1a, Fig. S2). While Cα CP signal is approximately the same from − 40 to 0 °C (red), there is a slight optimum for CO signal from − 25 to − 5 °C (blue). DP signal is more or less equivalent, relatively, throughout the temperature series (Fig. 1b, Fig. S2, black), therefore CP enhancement plots as a function of temperature (Fig. 1c) presents similar patterns to plots of the CP signal in Fig. 1a. CO enhancement is quite low due to a lack of bound ¹H for cross polarization. Interestingly, Cα enhancement is also quite low compared to previous observations of a maximum ~ 2.0–2.5 CP enhancement for bulk membrane proteins in this temperature range (van Aalst et al. 2022). This observation further cements the inherently dynamic nature of CCR3 and provides further rationalization for this study. Lastly, backbone amide ¹⁵N CP is very similar from − 40 to − 5 °C before decreasing with increasing temperature (Fig. 1, Fig. S3). ¹⁵N DP was not attempted due to expected low signal. From this data it is natural to draw the conclusion that any temperature below 0 °C should be acceptable, but CP efficiency is not the only parameter to consider. Next, we will examine relaxation parameters in the same context to narrow down the range of optimum temperatures for DCP acquisition.

Fig. 1.

Integration of 1D ¹³C a cross-polarization, b direct polarization signal, c cross-polarization enhancement, and d ¹⁵N cross-polarization signal. Raw signal intensity is normalized for each CP and DP dataset per each plot, and data is presented as normalized peak volume ± normalized error-modified spectral noise (see Materials and Methods), which may be overshadowed by scatter points. CP enhancement in c was calculated as raw CP intensity over raw DP intensity, to calculate the extent of enhancement in absolute signal by introduction of CP. A general trend is observed were CP and DP signal decreases with increased temperature

CCR3 relaxation parameters in a PC: cholesterol bilayer: T₂’

The transverse relaxation time constant T₂ defines the decay of transverse magnetization (Bloch 1946). In soft matter such as proteoliposomal samples, the magnitude of T₂ depends predominantly on chemical shift anisotropy (CSA) and dipolar dephasing, which both contribute to signal loss under a spin echo (Schanda and Ernst 2016). T₂ in this instance is often referred to as T₂’ or ‘coherence lifetime’, where a longer T₂’ is directly related to an increase in spectral sensitivity of 2–3D experiments (Paëpe et al. 2003). Since temperature influences dynamics, which in turn influence efficiency of dipolar couplings as above, variation of temperature will ultimately play a role in the magnitude of T₂’. T₂’ under spin echo spectra were acquired for both ¹³C and ¹⁵N by arraying the echo period in steps of 166 μs (2 × the rotor period) to a maximum of 8.30 ms for ¹³C and 13.28 ms for ¹⁵N. T₂’ was then estimated by fitting to a monoexponential decay function. Raw plots giving rise to fit rate constants are available in the SI for Cα T₂’ (Fig. S4), CO T₂’ (Fig. S5), and ¹⁵N T₂’ (Fig. S6). Because data presented herein are estimates, we only interpret relative T₂’ rather than quantitative values, though fits for bulk backbone atoms do reveal surprisingly short T₂’ values (Fig. 2a–c). In all cases, T₂’ values increase with decreasing temperature to a maximum at − 35 or − 40 °C. Temperatures above − 20 °C should be discounted for this reason. Considering all presented thus far, a temperature of − 25 or − 30 °C may be the most favorable temperature for acquisition of DCP spectra.

Fig. 2.

Backbone a Cα, b CO, and c amide ¹⁵N T₂’ measurements at each temperature. T₂’ experiments were acquired under a Hahn Spin Echo by varying the refocusing delay time. Error bars indicate the square-root of the covariance of the monoexponential fit matrix at each temperature

Identification of match conditions and minimization of intermediate timescale dynamics: T_1rho

Typically, SPECIFIC-CP conditions are used for selective transfer from the amide to the Cα or CO, though other options are available (Daviso et al. 2013). However, the transfer efficiency in SPECIFIC-CP is detrimentally sensitive to intermediate timescale molecular motions on the order of μs-ms (Sperling et al. 2010). T_1rho experiments are unique in that they are one of only a few techniques in SSNMR that can be used to probe such motions (Krushelnitsky et al. 2013). Minimization of intermediate timescale motion will lead to longer T_1rho times and thus be directly related to optimized SPECIFIC-CP transfer. In addition, DCP frequency selection requires separation of the field strength to be equivalent to the MAS rate. Here, 3/2, 5/2, and 7/2 ω_r. conditions are selected for this reason, as well as to avoid the rotary resonance condition where the CSA is recoupled, leading to erroneous underestimation of T_1rho (Quinn and McDermott 2012).

To this end, ¹³C and ¹⁵N T_1rho arrays were acquired at each temperature by application of spinlock fields at different field strengths (Fig. 3, Figs. S7–13). As before, the resulting decay of signal was fit to a monoexponential decay to approximate relative T_1rho time constants. Fortunately, the amide ¹⁵N T_1rho at − 25 °C for all field strengths are approximately equivalent (Fig. 3a–b, Figs. S7−9). In general, the values for the amide peak T_1rho are within error for all field strength conditions, allowing for flexibility in choosing optimal ¹³C field strength. For Cα, time constants are quite similar for both field strengths at − 30 °C (Fig. 3c, Figs. S10, 12). Therefore, acquisition of NCA spectra could theoretically proceed with the 3/2 ¹³C and 5/2 ω_r ¹⁵N spin lock field strengths, selected due to slightly more favorable error in fitting.

Fig. 3.

Backbone T_1rho measurements for a ¹⁵N at 3/2 and 5/2 ω_r, b ¹⁵N at 7/2 ω_r, c ¹³C at 3/2 ω_r, and d ¹³C at 5/2 ω_r. Error bars indicate the square-root of the covariance of the monoexponential fit

Temperature dependence of signal and resolution in 2D and 3D SSNMR experiments

We sought to combine observations from 1D experiments in acquisition of DCP spectra in the form of N-Cα selective transfer. We began at − 30 °C but quickly noticed that after extensive parameter optimization transfer efficiency with 5/2 ω_r ¹⁵N and 3/2 ω_r ¹³C conditions were non-ideal, with a maximum at approximately 22.0% compared to the ¹H-¹³C CP 1D signal (Fig. S14a). We next optimized the 7/2 ω_r ¹⁵N and 5/2 ω_r ¹³C DCP conditions to achieve a transfer efficiency of 36.0% (Fig. S14b). The same experiments were reoptimized at − 5 and 15 °C, achieving 28.7 and 22.3% transfer efficiency, respectively (Fig. S14c, d). Impaired transfer as temperature increases is attributed to increased dynamics, thus we do observe enhanced DCP signal at lower temperatures as expected from 1D experiments. However, resolution did not appear to change appreciatively from a qualitative standpoint in 2D NCA experiments that have been signal averaged. Given that our CCR3 construct is 376 amino acids, it is not unexpected that peak crowding would be such that linewidths of individual peaks could not be investigated. This is also observed in single NCA blocks with contours drawn to the same contour levels (Fig. 4a). Here, decreased transfer efficiency is perhaps best exemplified by loss of the Lys sidechain peaks at ~ 80–90 ¹⁵N.

Fig. 4.

Analysis of CCR3 2–3D spectra. a Single block heteronuclear NCA DCP experiments acquired at − 30, − 5, and 15 °C. Transfer efficiency is included compared to 1D ¹³C CP experiments. Results qualitatively suggest temperature-dependence of spectral resolution and sensitivity in multidimensional experiments. Contours in (a) are drawn to the same absolute noise floor (5 σ at − 30 °C) to highlight differences in signal across temperatures. Linewidth analysis is performed in 3D NCACX spectra acquired at b − 30 °C compared to c − 5 °C. Representative peaks in each slice were fit using NMR-FAM-SPARKY. Labels on each peak indicate the fit Cα linewidth. Linewidths of approximately 40–150 Hz (average ± std: 80 ± 48 Hz) are observed at − 40 °C and 40–250 Hz at − 20 °C after accounting for line broadening during processing, indicating 3D resolution is temperature dependent

Recently, we observed significant line broadening in homonuclear DARR spectra of CCR3 at − 5 °C compared to 15°C (van Aalst et al. 2023). Thus, we next acquired ¹³C-¹³C DARR 2Ds with a mixing time of 12 ms at the same temperatures as NCA experiments (Fig. S15) to quantify changes in linewidth. Using the peak integration functionality in NMRFAM-SPARKY (Kneller and Kuntz 1993; Lee et al. 2015), linewidths were estimated for several outlier Cα regions (Fig. S16). Average integrated line widths ± S.D. for each region are plotted in Fig. S16c. While some Cα linewidths (bulk Ala and aromatic residues) indicate no great change in line broadening as a function of temperature, others (bulk Val/Thr and Pro) increase from − 30 to − 5 °C. We next plotted integrated linewidths generated from ¹³C CP 1D spectra in the range of ~ 52–64 ppm (Fig. 1a, Fig. S1), where a similar trend is observed (Fig. S16d). However, caution must be taken when interpreting linewidth analysis of 1D DP and 2D DARR spectra due to a major conceptual flaw: peaks are generally not well isolated due to peak degeneracy as a function of protein size, though resolution is much improved over heteronuclear 2Ds due to differences in ¹³C and ¹⁵N chemical shift ranges. While still a useful exercise especially when considering multiple sources, lack of full isolation is problematic.

Considering this point, we also acquired NCACX spectra at both − 30 °C (Fig. 4b, Fig. S17a) and − 5 °C (Fig. 4c, Fig. S17b) and fit Cα linewidths to gauge spectral resolution in 3D experiments. Results of this analysis indicated the same trend as in 2D DARR spectra occurs. Here, Cα linewidths from 31 to 152 Hz (average ± std: 80 ± 48 Hz) are observed in integrated peaks from three separate ¹⁵N slices at − 30 °C after correction for line broadening (Fig. S17a). Comparatively, fewer peaks are observed in NCACX spectra acquired at − 5 °C (Fig. S17b) and the fit linewidths are also generally broader, on the order of 40–250 Hz (121 ± 86 Hz, ***P < 0.001). Statistical significance between linewidth datasets further suggests a trend despite modest differences. Analysis of 1D slices generated from representative regions of these 2D planes corroborates this observation, where Cα peaks are generally broadened at − 5 °C compared to − 30 °C. This observation is, in some cases, also extended to side chain peaks.

Undeniably, the greatest difference between 3D data acquired at − 5 vs. − 30 °C is loss of signal. Since both datasets were signal averaged for the same timeframe, the estimated noise floors were equivalent. Thus, contours for both spectra are drawn to the same absolute threshold, and direct comparison of signal is possible. Here, presence of ~ 2 × as many peaks at − 30 °C compared to − 5 °C indicates more signal is observed at the lower temperature. Signal attenuation in 3D spectra is estimated to be more than differences in CP enhancement in Fig. 1c would suggest, indicating a possible role of general intermediate timescale dynamics and thus loss of overall signal at − 5 °C. Extracted slices from each 3D further corroborate that overall linewidths are narrower at − 30 °C. Additionally more side-chain resonances are observable and some of these side-chains exhibit intermediate exchange broadening (Fig. S18). From this, we conclude spectra at − 30 °C are better resolved and give a better representation of the residues in CCR3. Our results indicate these spectra are higher quality and are better suited for resonance assignments of CCR3 compared to spectra at − 5 °C. Acquisition of high-quality 3D spectra of a GPCR is a technical achievement in its own right.

Bilayer phase properties dictate the optimal temperature for spectral sensitivity

The goal of this work was to identify a temperature to acquire resonance assignment spectra of CCR3 that were of sufficient signal and resolution to identify and differentiate backbone and sidechain peaks from each other. While this goal was achieved, the differences in linewidths and the possibility of significant intermediate timescale dynamics detailed above was intriguing. Previously Bangin, et al. showed that optimal sensitivity and resolution of the multidrug transporter EmrE was obtained ~ 15 °C below the center of the phase transition range in model PC bilayers (Banigan et al. 2015). The presence of cholesterol is known to influence bilayer phase properties (Warschawski and Devaux 2005a, b), but the effects on optimal resolution are not characterized. CCR3 is extremely dynamic, and we previously reported signal attenuation in DARR spectra of CCR3 in POPC at a set temperature of 20 °C (sample temperature of 35 °C) (van Aalst et al. 2023). Though not measured, we expected this temperature to be above the phase transition of POPC bilayers containing CCR3. We hypothesized this observation to be due to attenuated dipolar couplings at higher temperatures. When spectra were acquired at the same temperature but in 7:3 PC:Cholesterol, magnitude of signal was greatly improved but linewidths were observed to be broadened (van Aalst et al. 2023). That is to say, CCR3 undergoes different dynamic timescales in different lipid environments measured in both SSNMR spectra and molecular dynamics simulations. Thus, it is unclear what effect the presence of cholesterol, and therefore bilayer phase behavior may have on DCP sensitivity compared to current literature experimentation involving PC only. rINEPT transfers magnetization through the ¹H-¹³C J coupling without ¹H decoupling, thus dynamic motion is a prerequisite for efficiency of transfer (Alonso and Massiot 2003; Elena et al. 2005). We have performed several studies using isomer-specific all-trans (AT, ~ 35 ppm) and trans-gauche (TG, ~ 33 ppm) acyl chain chemical shifts in 1D spectra to monitor phase transitions in situ (van Aalst et al. 2022) and in vitro (Borcik et al. 2020; Borcik et al. 2019). Depicted in Fig. 5a is the structure of POPC with labeled acyl conformations. Typically, below the phase transition the bilayer is in the gel phase (L_β) dominated by AT acyl conformations. As the bilayer melts, gel gives way to liquid crystalline (L_α) which can be partitioned into AT Liquid-ordered (L_o) and TG Liquid-disordered (L_d) domains. Additionally, we showed that rINEPT can be used to monitor lipid phase transitions of natural abundance lipids even in a ¹³C-labeled protein background due to extreme difference in molecular dynamics (Borcik et al. 2020), which we implement here to similar effect.

Fig. 5.

Lipid phase rINEPT experiments on natural abundance POPC. a Structure of POPC with lipid acyl conformation labeled. b rINEPT 1D spectral excerpts at temperatures of − 30 °C (black), − 5 °C (red), and 15 °C (purple). Orange highlights indicate the AT (~ 35 ppm) and TG (~ 33 ppm) peaks. In c peak volume of AT vs. TG is plotted as a function of temperature. Signal is first observed at a temperature of approximately − 20 °C, indicating the start of the phase transition

1D ¹³C-detected rINEPT curves were acquired at each temperature (Fig. S19). Spectra at temperatures of interest are presented in Fig. 5b with peaks of interest labeled according to assignments. Data was fit in the same way as DP and CP data above and plotted as a function of temperature (Fig. 5c). For both AT and TG peaks, signal first appears at a temperature of − 20 °C and reaches a maximum at 20–25 °C. The relatively low and broad range of this transition is likely due to the presence of cholesterol and protein which both can have bilayer ordering properties (Borcik et al. 2019, 2020). In the absence of CCR3, the phase transition range is observed from − 5 to 15 °C (Fig. S20), indicating that CCR3 has an ordering effect on the bilayer likely related to direct cholesterol interaction with the receptor (van Aalst et al. 2021). Furthermore, we do observe the same pattern of spectral quality behavior as noted by others: spectra acquired at temperatures within the phase transition range are of lower sensitivity and resolution than those acquired below the phase transition. Ostensibly, this observation is due to phase-specific contributions to protein dynamics that manifest as inhomogeneous line broadening and intermediate timescale dynamics that rectifies above and below the transition range. Coincidently, DCP spectra acquired at a temperature of − 30 °C falls ~ 10–15 °C below the transition temperature range.

Conclusions

In this work we have presented data characterizing the temperature-dependent dynamics of the chemokine receptor CCR3 as they pertain to signal and resolution of SSNMR spectra. While the overabundance of 1D spectra acquired in this work is likely unnecessary in the normal course of experiments acquired by spectroscopists, the results are undeniably intriguing: optimal acquisition of membrane protein spectra depends on how intrinsic protein dynamics are influenced directly by temperature and indirectly by temperature-dependent bilayer properties. A facile approach to accurately predict optimal temperature would simplify the optimization process as SSNMR is applied to more complicated and unexpectedly dynamic membrane protein systems. While more experiments are required to confirm methodology in different lipid environments and for different proteins, the results are promising. We therefore conclude that optimal DCP temperature is not equivalent for different proteins in different lipid environments, but that bilayer fluidity may be a useful tool for easily identifying optimal temperature without the need for multiple temperature series experiments. Use of ¹³C-detected rINEPT experiments to do so offers a quick and easy approach even in a ¹³C-labeled protein background.