Studying intact bacterial peptidoglycan by proton-detected NMR spectroscopy at 100 kHz MAS frequency

Abstract

The bacterial cell wall is composed of the peptidoglycan (PG), a large polymer that maintains the integrity of the bacterial cell. Due to its multi-gigadalton size, heterogeniety, and dynamics, atomic-resolution studies are inherently complex. Solid-state NMR is an important technique to gain insight into its structure, dynamics and interactions. Here, we explore the possibilities to study the PG with ultra-fast (100 kHz) magic-angle spinning NMR. We demonstrate that highly resolved spectra can be obtained, and show strategies to obtain site-specific resonance assignments and distance information. We also explore the use of proton-proton correlation experiments, thus opening the way for NMR studies of intact cell walls without the need for isotope labeling.

Introduction

Magic-angle spinning (MAS) solid-state NMR (ssNMR) spectroscopy has the unique possibility to provide atom-specific insight into (bio-)molecular systems without the need for crystalline or solubilized samples. Over the last 15 years, ssNMR has rapidly grown, and it has enabled the determination of protein structures at atomic resolution, including crystalline proteins (Castellani et al. (2002); Huber et al. (2011); Linser et al. (2011a); Knight et al. (2012); Bertini et al. (2010)), amyloid fibrils (reviewed by Comellas and Rienstra (2013)), membrane proteins (Wang et al. (2013a); Park et al. (2012); Shahid et al. (2012)) and large assemblies (Loquet et al. (2012); Morag et al. (2015); Andreas et al. (2016)). A particularly attractive property of ssNMR is its ability to work with cell membranes, cell walls or entire cells (Renault et al. (2010); Dick-Perez et al. (2011); Wang et al. (2013b); Cegelski et al. (2010, 2002); Tong et al. (1997); Renault et al. (2012). The focus of the present manuscript is the bacterial peptidoglycan, a giga-dalton large heteropolymer composed of linear glycan strands composed of two alternating carbohydrates cross-linked through short peptide stems. Unique to bacteria, peptidoglycan biosynthesis and maturation is an important target for antibiotics, and is therefore involved in a multitude of complexes formed with antibiotics, cell wall enzymes, virulence factors and immunity defense systems. These complexes nevertheless escape high-resolution characterization by most structural biophysical methods. In this respect MAS NMR stands out as the sole technique that provides insight into structural parameters and dynamics of intact peptidoglycan isolated or in interaction with antibiotics or cell wall components (Kim et al. (2008, 2006); Kern et al. (2010); Cegelski et al. (2002); Tong et al. (1997)), or into its complexes with proteins (Schanda et al. (2014)). An important challenge is thus to improve the NMR methods giving access to these assemblies at atomic scale resolution.

In NMR studies of complex biomolecular systems, such as peptidoglycan, the primary practical challenges are often the limited detection sensitivity, signal overlap and hence limited resolution, as well as difficulties to obtain these biological samples in sufficient quantity. Until recently, biomolecular ssNMR studies have been based primarily on the use of ¹³C-detected experiments at MAS frequencies of <20 kHz, using sample rotors with diameters of 3.2 – 4 mm and sample quantities of >20 mg. Obtaining such large amount of sample is often a major bottleneck, but required to achieve sufficient sensitivity in such experiments. Proton-detection has intrinsically higher sensitivity due to the larger gyromagnetic ratio of ¹H spins compared to ¹³C. However the resulting larger ¹H-¹H dipolar coupling also gives rise to severe line broadening. For example, at MAS frequencies of 20 kHz, typically employed with 3.2 mm rotors, ¹H line widths in fully protonated proteins are of the order of 300-500 Hz (Marchetti et al. (2012)), which generally excludes extraction of site-specific information and resonance assignment.

High-resolution proton-detected ssNMR spectroscopy of biomolecules has become possible through technological advances, namely on the one hand with the introduction of partial deuteration schemes (increasing inter-proton distances and thus decreasing line broadening due to dipolar-dephasing), and on the other hand with the development of hardware allowing for higher MAS frequencies, and thus more efficient averaging of dipolar interactions and smaller line widths. Using deuterated proteins, 100 % back-protonation of the exchangeable amide sites, and MAS frequencies of up to 60 kHz, ¹H line widths of ca. 50-100 Hz were obtained (Zhou et al. (2007a); Lewandowski et al. (2011a); Fraga et al. (2017); Fricke et al. (2017)). Narrower line widths can be achieved by even higher deuteration levels, where the exchangeable sites are back-protonated to only 10-50% in an otherwise fully deuterated environment (Chevelkov et al. (2006); Schanda et al. (2009); Linser et al. (2011b); Reif (2012)). Deuterated samples and 40-60 kHz MAS have enabled well-resolved proton-detected ssNMR experiments for structure determination (Huber et al. (2011); Knight et al. (2012); Zhou et al. (2007b); Linser et al. (2011a)) and detailed measurements of dynamics (Chevelkov et al. (2009); Tollinger et al. (2012); Schanda et al. (2011); Krushelnitsky et al. (2010); Smith et al. (2016)). However, the need for extensive deuteration is often a severe bottleneck, particularly when dealing with complex samples such as entire cell walls from bacteria that are difficult to culture in H₂O-based buffer and may be impossible to grow in D₂O media, or for proteins that are difficult to produce in deuterated form.

Several recent applications have shown the resolution gains of even faster, sometimes termed ”ultra-fast”, MAS, i.e., at spinning frequencies exceeding >100 kHz. This technology is based on sample rotors of only 0.7 mm diameter, containing less than 1µL, i.e., only ca. 0.5 mg of sample. Although deuteration at these spinning frequencies is still beneficial (Agarwal et al. (2014); Xue et al. (2017)), even fully protonated proteins yield relatively narrow ¹H line widths at >100 kHz MAS of the order of 100-150 Hz (Stanek et al. (2016); Andreas et al. (2016); Xue et al. (2017)).

In this manuscript we show that well-resolved ¹H-detected correlation spectra of the intact peptidoglycan from bacterial cell wall of Bacillus subtilis can be obtained in a short experimental time using sub-milligram amounts at 100 kHz MAS frequencies. We present approaches for through-bond correlation experiments, required for assignment, and through-space ¹H-¹H correlation experiments, and explore the possibilities of studying peptidoglycan only with ¹H-¹H correlation experiments, thus removing the need for isotope labeling and opening new avenues to studies of cell walls of bacterial strains that are difficult to culture.

Materials and Methods

Sample preparation

Bacterial peptidoglycan sacculi have been prepared using methods described previously (Severin and Tomasz (1996); Kern et al. (2010)). Briefly, Bacillus subtilis subsp. subtilis strain 168 cells from the American Collection (ATCC 23857) were grown in standard M9 growth medium with ¹³C glucose and ¹⁵N ammonium chloride as sole carbon and nitrogen sources, respectively, or without isotope labeling. At an OD (600 nm) of ca. 0.7, cells were harvested by centrifugation at 5,500 g for 10 min and non covalently bound cell wall components were removed by boiling in 4% w/v SDS for 30 min. The cell suspension was pelleted by ultracentrifugation at 25°C for 45 min at 130,000 g. The pellet was washed twice with 30 ml of 1 M NaCl and repeatedly with milli-Q water until it was free of SDS. The pellet was resuspended in 2 to 4 ml of deionized water, 1/3 volume of acid-washed glass beads (diameter of 0.17 to 0.18 mm) was added, and cells were disrupted. The glass beads were then separated from the cell lysate and the filtrate was centrifuged at 1000g for 5 min, and the supernatant, which contains the disrupted cell wall, was centrifuged at 130,000g for 45 min at 25°C. The pellet was resuspended in 20 ml of 100 mM TrisHCl (pH 7.5) containing 20 mM MgSO₄. DNase A and RNase I were added to final concentrations of 10 and 50 µg/ml, respectively, and the sample was stirred for 2 h at 37°C. CaCl₂ (10 mM) and trypsin (100 µg/ml) were added, and the sample was stirred for 18 h at 37°C. Then, 1% SDS (final concentration) was added, and the sample was incubated for 15 min at 80°C to inactivate the enzymes. The cell wall was recovered by centrifugation for 45 min at 130,000g at 25°C, resuspended with 20 ml of 8 M LiCl, and incubated for 15 min at 37°C. After another centrifugation, the pellet was resuspended in 10 mM ethylenediaminetetraacetic acid (EDTA, pH 7.0) and incubated at 37°C for 15 min. The cell wall was washed with deionized water, acetone, and again with water before finally being resuspended in 2 to 4 ml of deionized water, yielding typically 50 mg of cell wall material for 1 L culture. 5 mg of cell wall was then stirred with 48% hydrofluoric acid (HF) for 48 h at 4°C. The peptidoglycan was recovered by centrifugation at 130,000 g for 45 min at 4°C and washed with ice-cold deionoized water, 100 mM TrisHCl (pH 7.0), and then twice with water and stirred at 4°C. This treatment resulted in intact well-hydrated peptidoglycan sacculi, which were resuspended in 50 mM HEPES buffer (pH 7.2) and filled into a Bruker 0.7 mm rotor using an ultracentrifuge spinning at 55,000 g during ca. 3 hours. To this end, the 0.7 mm rotor was inserted into a 1.3 mm rotor, which was inserted into a device for rotor-filling in an ultracentrifuge (Bruker Biospin).

NMR spectroscopy and data analysis

All experiments were performed on a Bruker Avance 2 spectrometer operating at a ¹H Larmor frequency of 950 MHz. The MAS frequency was set to 100 kHz and stable to within 5 Hz. Sample cooling was achieved with a cooling gas flow (300 L/h) at 273 K, while bearing and drive gas was at ca. 293 K. The resulting sample temperature under these conditions was determined with an external protein sample with the same MAS and temperature setup, using the bulk water resonance frequency and an internal DSS signal. The estimated sample temperature used in this study was ca. 30-32°C. Pulse sequences used in this study are shown in Figure 1. Typical 90°pulse durations were 2.1 µs (¹H, at 9 W), 2.7 µs (¹³C, at 30 W) and 3.8 µs (¹⁵N, at 30 W). Cross-polarization (CP) ¹H-¹⁵N transfer was achieved with a 1 ms long ramped radiofrequency (RF) field strength of 60 to 80 kHz on ¹H and 30 kHz on ¹⁵N. ¹H-¹³C CP steps involved a ramped RF field on ¹H (115 to 140 kHz amplitude) and ca. 30 kHz RF field on ¹³C, for 300 µs. Low-power WALTZ-16 (Shaka et al. (1983b)) decoupling was applied with RF field amplitudes of 10 kHz (¹H), 15 kHz (¹³C) and 5 kHz (¹⁵N). In INEPT-based transfers, the total transfer delay was 3.3 ms (¹H-¹³C) and 4.4 ms (¹H-¹⁵N). Solvent suppression follows the philosophy of the previously proposed MISSISSIPPI scheme (Zhou and Rienstra (2008)), i.e., irradiation on ¹H while heteronuclear magnetization is stored along the z-axis, except that the irradiation scheme involved a train of back-to-back pulses with durations of 0.3 to 2 ms and 20 different phase settings. The aim of this essentially arbitrary pulse trains is to create a ’random’ phase trajectory. We find that this sequence offers superior solvent suppression than with the original MISSISSIPPI scheme or with TPPM-based suppression schemes. The total duration of the solvent suppression was optimized and typically 30-60 ms. Radio-frequency driven ¹H-¹H recoupling (RFDR, Bennett et al. (1995)) elements were applied for 1.5 ms. TOCSY-type scalar-based ¹³C-¹³C transfer, according to the pulse scheme of Figure 1E, was achieved with the DIPSI scheme (Shaka et al. (1988)) at a ¹³C RF field strength of 25 kHz; two experiments were recorded with either 8 DISPI2 cycles (9.2 ms) or 12 cycles (13.8 ms). Spectra were processed with Topspin 3.5 (Bruker Biospin) and analyzed with CcpNmr (Vranken et al. (2005)). ¹H chemical shifts were referenced according to the position of the resolved methyl group of the GlcNAc N-acetylresonance by comparison with the corresponding resonance in the liquid state. ¹³C and ¹⁵N chemical shifts were referenced indirectly using gyromagnetic ratios from the Biological Magnetic Resonance Data Bank.

Figure 1.

Pulse sequences used in this study for recording H-N (A,B), H-C (A,B,D,E) and H-H (E) correlation spectra. Black rectangles and narrow open rectangles correspond to hard 90°and 180°pulses, respectively, and were applied at RF field strengths of 117 kHz (¹H), 90 kHz (¹³C) and 65 kHz (¹⁵N). Ramped cross-polarization (CP) is depicted with wide open symbols. ¹H decoupling was applied during the heteronuclear chemical-shift evolution periods with the WALTZ-16 Shaka et al. (1983a) scheme.

Results and Discussion

A schematic chemical structure of the bacterial peptidoglycan is shown in Figure 2A. While the intact peptidoglycan has a total molecular weight of the order of gigadaltons, the constituting blocks are relatively limited, comprising a glycan backbone and peptide stem, which may be cross-linked to the peptide stem of another glycan strand (in black). The two-dimensional spectra in red in Figure 2B-D show one-bond HN and HC correlations in intact peptidoglycan. Panel (B) reveals backbone and side chain amide resonances from the peptide stems in addition to the N-acetyl resonances of the amino sugars. Panel (C) focuses on the ¹H-¹³C region of the disaccharide motif resonances, while panel (D) shows the corresponding peptide portion of the spectrum. Additionally, a comparison of the CP-based to INEPT-based correlation spectra are shown in Supplementary Figure S1. The relative peak intensities in CP- vs INEPT-based experiments depend on the type of sub-motif studied. Resonances from the glycan strands, which correspond to more rigid portions than the peptide stems (Gansmueller et al. (2013)), are detected with more sensitivity in CP-based esperiments, while peptide resonances are favored in the refocused-INEPT based experiments. Despite the very small amount of sample, these 2D ¹H-¹³C correlation maps could be obtained in an experimental time of ca. 3 hours while the ¹H-¹⁵N spectrum of Figure 2B was collected in 18 hours.

Figure 2.

Proton-detected solid-state NMR spectra of intact Bacillus subtilis peptidoglycan at 100 kHz for resonance assignment with through-bond experiments. (A) Chemical structure of peptidoglycan. Shown are two disaccharide units (top), connected by peptide stems, one of them depicted in black. (B) ¹H-¹⁵N correlation spectrum, based on INEPT transfer. (C) CP-based ¹H-¹³C correlation spectrum focusing on the glycan region is shown in red. A two-dimensional ¹³C-¹³C TOCSY spectrum is shown in black. Different chemical moieties are labeled with different colors, corresponding to the color code in (A). (D) Amino acid aliphatic region of the ¹H-¹³C correlation spectrum (red), and the 2D ¹³C-¹³C TOCSY correlation map, showing through-bond correlations with dashed lines. The full aliphatic and glycan parts of the spectra, with CP and INEPT-based transfer are shown in Figure S1.

The ¹H line widths are in the order of 60-100 Hz for the ¹H-¹⁵N sites, ca. 120 Hz for the carbohydrate ¹H-¹³C sites and 50-150 Hz for the peptide ¹H-¹³C sites (cf Figure 2), which allows resolving the majority of the sites. If the ability of the additional ¹H frequency to differentiate between different sites is not striking for the carbohydrate C1-C6 moieties due to severe overlaps of ¹H resonances on the contrary to ¹³C frequencies (Figure 2C), its potential in raising the degeneracy of DAP backbone (HN or HC at α position) vs side-chain (HN or HC at ϵ position) resonances or in raising the degeneracy of GlcNAc vs MurNAc carbohydrates N-acetyl groups is clearly illustrated in Figure 2 parts B and D. The corresponding resonances are highly difficult to resolve in ¹³C-¹³C correlation maps as highlighted by the red rectangles in Figure S2. The increased resolution brought by the ¹H frequency is particularly noteworthy when it comes to follow the maturation of the peptidoglycan network through the degree of cross-linking at the DAP ϵ site (see DAP with 3-4 subscript labels in panels B and D).

To assign the ¹H-¹³C correlation spectra, we have recorded through-bond correlation spectra. Figure 1E shows the employed pulse sequence which uses TOCSY mixing for efficient ¹³C -¹³C coherence transfer, and the resulting spectra are shown in Figure 2C,D and Supplementary Figure S1. Similarly to the case of proteins (Andreas et al. (2016)), the ¹³C-¹³C mixing scheme provides efficient transfer across multiple bonds, within experimental times of only ca. 1.5 days, despite the small amount of sample.

The ¹H-¹⁵N correlation spectrum of intact peptidoglycan has lower intrinsic sensitivity than its ¹H-¹³C counterpart, rendering triple-resonance experiments for assignments difficult. We tentatively ascribe this observation to chemical exchange of amide hydrogens with the solvent, as expected for flexible peptides at pH 7.5. We, thus, turned to solution-state spectra of peptidoglycan fragments, prepared by enzymatic digestion, to obtain site-specific assignments, which we further confirmed by homonuclear RFDR experiments (see below). Supplementary Figure S3 shows solution-state correlation spectra and experiments enabling the assignment of the ¹H-¹⁵N spectrum. Briefly, the assignment followed well-established solution-state NMR techniques, involving hCconH and HNCAsCB experiments (cf. Figure S3). Figure 2B reports the assignments on the 100 kHz MAS spectrum for those peaks that show similar cross-peak positions in solution and solid-state samples. Interestingly, while showing several similarities, the spectrum of the intact peptidoglycan sacculi differs in peak intensities and peak positions significantly from the fragments. This observation demonstrates that the structure and dynamics of intact PG differs from those of fragments, illustrating the importance of studying intact samples rather than fragments for gaining insight into the native peptidoglycan properties.

Lastly, we have explored the possibilities of obtaining dipolar (i.e., through-space) proximities through ¹H-¹H dipolar recoupling experiments with the RFDR scheme (Figure 1C,D). Figure 3A shows a two-dimensional ¹H-¹H correlation experiment, and Figure 3B shows a two-dimensional ¹³C-edited experiment which encodes ¹H-¹H contacts, based on the pulse scheme shown in Figure 1D. The latter experiment is particularly useful for the glycan sites, which have resonance frequencies close to the one of bulk water, such that the residual solvent signal in the homonuclear experiment makes extraction of these signals difficult.

Figure 3.

Dipolar ¹H-¹H correlation spectra using radio-frequency driven recoupling of the ¹H-¹H interaction. (A) Homonuclear 2D correlation experiment and (B) ¹³C-edited ¹H-¹H correlation, focusing on the glycan part. The full-width spectrum is shown in Suppelmentary Figure S4. Both experiments have been collected with a 1.5 ms RFDR mixing at a MAS frequency of 100 kHz. Assignments of the correlation peaks are indicated and color-coded according to the color scheme shown in Figure 2A.

In these two experiments numerous contacts between protons in spatial proximity are visible. While the majority of the contacts are within the same amino acid or within the same sugar, inter-residue contacts in the amide region of the spectrum in Figure 3A between the DAP amide resonances and D-iGlu Hβ and Hγ yield unambiguous identification of the DAP residues. In the ¹³C-edited in Figure 3B a series of spectrally unambiguous remarkable cross-peaks are the ones connecting the HN, H2 and N-acetyl methyl sites of the GlcNAc sugar moiety with the H^α of the neighboring L-alanine residue on the MurNAc peptide stem. While currently there are no atomic structures of peptidoglycan available, the distance between these nuclei exceeds at least ca. 5 Å.

Conclusions

Recent ssNMR investigations on protein samples (Andreas et al. (2016); Agarwal et al. (2014)), established that high-resolution spectra and even atomic-resolution protein structures can be obtained at >100 kHz MAS. We have shown here that intact bacterial peptidoglycan sacculi yield highly resolved ¹H-detected correlation spectra. Despite the very small amount of sample required (0.5 mg), it is possible to obtain site-specific resonance assignments and information about spatial proximities. The sensitivity of these experiments shall also allow the study of the dynamics of intact cell walls. Indeed, high MAS frequencies suppress artifacts in dynamics measurements, and render quantitative analyses possible (Lewandowski et al. (2011b); Schanda and Ernst (2016)), such that the ultra-fast MAS is doubly favorable, providing high-resolution spectra and facilitating quantitative measurements.

We foresee that ultra-fast MAS NMR experiments will also be instrumental for determining the structures of proteins bound to peptidoglycan. A recent study has determined the binding pose of a protein bound to the cell wall (Schanda et al. (2014)), but provided information primarily for the protein rather than for peptidoglycan. The high sensitivity and resolution with which the peptidoglycan can be observed at ultra-fast MAS may allow NMR experiments to provide a more comprehensive picture of such complexes.

The availability of ¹H-detected spectra may also provide useful information about the structure of peptidoglycan. While the ¹H-¹H distances obtained here with RFDR experiments are rather short-range, and thus not likely to provide much structural information, the ease and sensitivity of these experiments shall enable longer-range distance measurements. For example, the attachment of spin labels to specific sites on the peptidoglycan or to interacting proteins may provide rapid and quantitative distance estimates, possibly contributing to refining the view of the structural organization of this important biopolymer.

This exploratory study also shows that it is possible to obtain ”fingerprint” homonuclear ¹H-¹H correlation spectra, i.e., to study complex biological systems without isotope labeling, which opens new fields of applications to NMR spectroscopy in which so far it has been difficult to obtain any atomic-resolution information.