Nucleic acid-protein interfaces studied by MAS solid-state NMR spectroscopy.

Solid-state NMR (ssNMR) has become a well-established technique to study large and insoluble protein assemblies. However, its application to nucleic acid-protein complexes has remained scarce, mainly due to the challenges presented by overlapping nucleic acid signals. In the past decade, several efforts have led to the first structure determination of an RNA molecule by ssNMR. With the establishment of these tools, it has become possible to address the problem of structure determination of nucleic acid-protein complexes by ssNMR. Here we review first and more recent ssNMR methodologies that study nucleic acid-protein interfaces by means of chemical shift and peak intensity perturbations, direct distance measurements and paramagnetic effects. At the end, we review the first structure of an RNA-protein complex that has been determined from ssNMR-derived intermolecular restraints.

Introduction

Besides its long-established role as carrier of genetic information in protein translation, RNA acts in many other cellular contexts, with new roles being discovered regularly (Cech and Steitz, 2014). The vast majority of the RNAs made from the human genome have distinct functions from protein coding (non-coding RNAs, ncRNAs), but many of these functions remain unknown. Both coding RNA (mRNA) and ncRNAs may act in complex with specific RNA-binding proteins (RPBs), which contain well-defined RNA recognition motifs (Corley et al., 2020). Being involved in such high number of biological functions, RNA molecules hold potentials as both therapeutic agents and targets.

As for proteins, the development of RNAs either as therapeutic targets or as protein-targeting agents requires understanding their three-dimensional structures and their interaction modes with proteins. Methods to characterize RNA–protein interactions develop rapidly (Schlundt et al., 2017). X-ray crystallography and cryo electron microscopy (cryo-EM) can be applied to high molecular-weight complexes and have been the methods of choice to study many large ribonucleoprotein complexes (RNPs) in the past years (Ben-Shem et al., 2011, Ghanim et al., 2021, Jackson et al., 2014, Khatter et al., 2015, Nguyen et al., 2016). These techniques work well for RNPs with well-defined structures but fall short when addressing conformational heterogeneity or dynamic processes. RNA molecules can adopt different folds depending on both the environmental conditions and the interacting partners and are often flexible in isolation. The presence of disordered RNA regions can make crystallization of RNPs quite challenging (Blanco and Montoya, 2011), while cryo-EM analysis remains blind to disordered molecular regions. In addition, both X-ray crystallography and cryo-EM are unable to represent the dynamics of the studied system at atomic level, which, especially in enzymes, provides the crucial link between structure and function. Finally, several intermolecular interactions with functional regulatory roles are transient in nature; transient complexes are difficult to crystallize and may dissociate during the preparation of cryo-EM grids, rendering both X-ray crystallography and cryo-EM inapplicable.

NMR spectroscopy is a structural biology technique that is able to provide structural information in the presence of disorder and dynamics. As such, NMR is very useful to study RNPs, and more generally nucleic acid–protein complexes, containing flexible regions, and reveals whether and how these disordered regions contribute to binding specificity and/or modulate affinity. NMR spectroscopy also enables structural studies of transiently forming complexes in a wide range of affinities (Campagne et al., 2011, Carlomagno, 2014, Dominguez et al., 2011, Simon et al., 2011, Yadav and Lukavsky, 2016). Finally, NMR allows the direct observation of hydrogen atoms, which remain inaccessible by X-ray crystallography or cryo-EM but play a crucial role in nucleic acids interactions (Yip et al., 2020).

Notoriously, NMR studies in solution are limited to particles of less than ∼ 50 kDa, due to the direct dependency of line-broadening on the molecular size. For proteins, this limit has been considerably extended by the methyl-TROSY methodology (transverse relaxation-optimized spectroscopy) (Kay, 2011, Rosenzweig and Kay, 2014, Tugarinov et al., 2003) in combination with selective 13CH3 methyl group labeling of highly deuterated proteins. Methyl-TROSY NMR exploits the excellent relaxation properties of methyl groups, which, when present as only hydrogen-bearing groups in otherwise deuterated proteins, retain feasible NMR line widths even in particles as large as 1 MDa (Lapinaite et al., 2013, Mas et al., 2018, Mas et al., 2013, Sprangers and Kay, 2007, Graziadei et al., 2020). Unfortunately, RNAs do not contain any methyl groups and NMR studies of large RNAs rely on a challenging combination of two-dimensional 1H–1H correlation experiments measured on several selectively-deuterated samples (Brown et al., 2020, Keane et al., 2015). Recently, methyl-TROSY has been applied to high-molecular weight DNA, where methyl groups were engineered at the C5 and N6 positions of cytosines (5mC) and adenines (6 mA) (Abramov et al., 2020). However, because artificial methylation of both DNA and RNA can affect their structures, dynamics and interactions with binding partners (Choy et al., 2010, Helm, 2006, Kawai et al., 1992, Ngo et al., 2016, Williams et al., 2001), nucleic acids methylation cannot be generally applicable to study large nucleic acids by NMR. In fact, methylation of both DNA and RNA occurs naturally in the cell, where it exerts a regulatory function by modulating both the structure and the interactome of nucleic acids.

Solid-state NMR spectroscopy (ssNMR) is another form of biomolecular NMR spectroscopy, which has been applied extensively to insoluble and non-crystalline particles, such as membrane proteins (Cady et al., 2010, Lange et al., 2006, Park et al., 2012, Shahid et al., 2012, Shi et al., 2009) and amyloid fibrils (Colvin et al., 2016, Hoop et al., 2016, Tycko, 2011, Van Melckebeke et al., 2010). ssNMR linewidths do not depend on the molecular weight, making the application of ssNMR to large particles feasible, provided that there is enough signal to compensate for the small number of large particles that can be fitted into a rotor of a given size. In addition, because ssNMR lines are intrinsically broader than solution NMR lines, selective isotope labeling is often crucial to resolve spectral overlaps and achieve site-specific assignment also of moderately sized molecules. Despite these limitations, ssNMR has been applied to large viral assemblies (Andreas et al., 2016, Goldbourt et al., 2007, Lusky et al., 2021, Morag et al., 2015, Morag et al., 2014, Sergeyev et al., 2011, Yu and Schaefer, 2008) and site specific structural information was obtained for the 46-residue-long major coat protein subunit of the filamentous bacteriophage Pf1, as part of the 36 MDa virion (Goldbourt et al., 2007), thanks to the fact that the 7300 subunits of the virion all adopt the same conformation. Over the years, the ssNMR toolbox has been extended for the application to RNA in isolation (Leppert et al., 2004, Lusky et al., 2021, Riedel et al., 2006, Riedel et al., 2005a, Riedel et al., 2005b, Yang et al., 2017), RNA bound to short peptides (Huang et al., 2010, Huang et al., 2011, Huang et al., 2017, Olsen et al., 2005, Olsen et al., 2010), RNA as part of RNP complexes (Aguion et al., 2021, Ahmed et al., 2020, Marchanka et al., 2013, Marchanka et al., 2015, Marchanka et al., 2018b) and DNA–protein complexes (Boudet et al., 2019, Lacabanne et al., 2020, Malär et al., 2021b, Wiegand et al., 2020b, Wiegand et al., 2019, Wiegand et al., 2017b, Wiegand et al., 2016, Zehnder et al., 2021).

The main challenge in NMR of RNA both in solution and in solid-state is the overlap of the signals due to the limited chemical diversity of the nucleotides. This is especially true in canonical, and thus homogeneous, tertiary structure elements, such as A-form helices. This challenge can be addressed using nucleotide-type selective and/or segmental isotope labeling (Duss et al., 2010, Nelissen et al., 2008, Tzakos et al., 2007), as well as site-specific labeling (Lu et al., 2010, Marchanka et al., 2018a), which reduce spectral crowding. As mentioned above, selective labeling becomes crucial in ssNMR, where the signal overlaps are significant also for RNAs of medium size in the absence of significant structural diversity (i.e. in helical regions).

The second challenge in NMR of RNA is the unequal proton distribution. Nucleic acids have a high proton density in the ribose ring but only few protons in the nucleobases and no protons at the backbone phosphate. This leads to a limited number of 1H–1H distance restraints available to determine the conformation at both the backbone and the Watson-Crick and Hoogsteen sides of the nucleobases. Fortunately, in ssNMR, the number of distance restraints that can be collected does not directly correlate with the number of protons, as distance restraints can be measured via both heteronuclear and homonuclear correlations mediated by dipolar couplings. Moreover, the distance range of restraints measured in ssNMR experiments such as rotational echo double resonance (REDOR) (Gullion and Schaefer, 1989a, Gullion and Schaefer, 1989b) or proton-driven spin diffusion (PDSD) (Szeverenyi et al., 1982) can exceed the range of those obtained in solution NMR by NOESY experiments (Huang et al., 2011, Huang et al., 2010, Marchanka and Carlomagno, 2019, Olsen et al., 2005, Olsen et al., 2003, Studelska et al., 1996), providing a useful tool for the refinement of global conformations. For example, REDOR experiments have provided long-range distance restraints up to 16 Å in proteins (Studelska et al., 1996) and 13 Å in nucleic acids (Olsen et al., 2003).

Because of its applicability to particles of large size and the versatility it offers in the design of magnetization transfer pathways, MAS ssNMR can adopt an important role in structural biology of RNP complexes. However, the disadvantages caused by signals overlap have long discouraged the application of ssNMR to RNA-containing particles. In the past decade our lab has developed a suite of ssNMR experiments that achieve assignment of RNA 13C,15N (Marchanka et al., 2015, Marchanka et al., 2013, Marchanka and Carlomagno, 2019) and 1H resonances (Aguion et al., 2021, Aguion and Marchanka, 2021, Marchanka et al., 2018b) as well as de novo RNA structure determination using distance restraints obtained solely from ssNMR experiments (Marchanka et al., 2015). These experiments, together with those developed in several other laboratories for the structure determination of proteins using 13C,15N detection (Castellani et al., 2003, Zhao, 2012) at low MAS frequencies and 1H detection (Andreas et al., 2016, Schubeis et al., 2018) at fast MAS frequencies (Penzel et al., 2019, Schledorn et al., 2020), allow the structure determination of the individual RNA and protein components of RNP complexes by ssNMR. Once the structures of the individual components are established, the identification of the intermolecular interfaces as well as the measurement of intermolecular distances yield the structure of the complex.

In this review we present recent advances of ssNMR spectroscopy to determine intermolecular contacts in RNP complexes and discuss advantages and challenges of the individual strategies. Due to the similar nature of the interfaces, we also review ssNMR studies of intermolecular interactions in DNA–protein complexes. We present conventional ssNMR methods that rely on the detection of low γ nuclei, such as 13C, 15N and 31P, as well as novel 1H-detected ssNMR experiments under fast magic-angle-spinning (MAS) rates. Finally, we discuss the first example of structure determination of an RNA–protein complex guided solely by ssNMR-derived restraints.

Sample preparation

NMR of nucleic acids is more challenging than that of proteins due to the poor chemical shift dispersion of their NMR signals. Thus, even for relatively small nucleic acids, advanced isotope labeling strategies may be required to obtain site specific structural information. Nucleic acids for ssNMR studies can be prepared by either chemical (Beaucage and Reese, 2009, Roy and Caruthers, 2013) or enzymatic synthesis; however, the methods available to produce isotope-labeled RNA are considerably more advanced than those available for DNA. An extensive description of isotope labeling strategies for RNA (Marchanka et al., 2018a) and ssDNA/dsDNA (Nelissen et al., 2016) by either chemical or enzymatic synthesis can be found in the literature.

As opposed to X-ray crystallography, ssNMR does not require crystals of any particular size and is therefore applicable to particles with substantial flexible regions, which are difficult to force in well-ordered, large crystals. Common sample preparation techniques in ssNMR include micro (nano)-crystallization (Bertini et al., 2010a, Franks et al., 2005, Huang et al., 2012, Marchanka et al., 2013, McDermott et al., 2000, Yang et al., 2017), ethanol precipitation (Zhao et al., 2019), lyophilization (Huang et al., 2011, Leppert et al., 2004, Olsen et al., 2008, Olsen et al., 2005, Wang et al., 1994), freezing in the presence of a cryoprotectant (Siemer et al., 2012) or sedimentation of soluble macromolecules into the ssNMR rotor using ultracentrifugation (Barbet-Massin et al., 2015, Bertini et al., 2011, Gardiennet et al., 2012, Lacabanne et al., 2020, Wiegand et al., 2016, Wiegand et al., 2020a). Micro-crystallization, ethanol precipitation and sedimentation could all yield sufficiently narrow linewidths to allow for site-specific assignments in individual samples (Aguion and Marchanka, 2021); however, ethanol precipitation is incompatible with the protein component of RNP complexes, while sedimentation has been successfully applied to RNP complexes, such as the prokaryotic ribosome (Barbet-Massin et al., 2015), but no data is available with respect to RNA linewidths in these samples. In contrast, both lyophilization and flash-freezing have been demonstrated to lead to inhomogeneous line broadening (Huang et al., 2011, Olsen et al., 2005, Siemer et al., 2012), impairing site-specific assignments. Nevertheless, structural information can be obtained also for these samples, in combination with site-specific isotopic labeling (Huang et al., 2011, Huang et al., 2010, Olsen et al., 2005).

ssNMR of nucleic acid–protein complexes prepared using either sedimentation or micro-crystallization have been reported to be stable over a long time. A DNA–protein–ATP complex prepared by sedimentation and stored at –20 °C for 3 years has been shown to yield virtually the same 13C–13C DARR spectrum as the freshly prepared sample (Lacabanne et al., 2020, Wiegand et al., 2020a). A ssNMR RNP complex sample prepared by micro-crystallization in our laboratory (Aguion et al., 2021) shows identical 1H–13C and 1H-15N fingerprint spectra after storage at + 4 °C for two years, with only minimal loss of signal intensities (Fig. 1).

Fig. 1

Long-term stability of precipitated RNP samples. (A) Overlay of the 2D CP hCH spectra tailored for the base spectral region of uniformly 13C,15N labeled Pyrococcus furiosus (Pf) 26mer Box C/D RNA in complex with the Pf L7Ae protein (Aguion et al., 2021, Marchanka et al., 2018b) immediately after sample preparation (blue) and two years after storage at + 4 °C (grey). (B) Representative horizontal 1D trace of the C6-H6 peak from residue U20 taken from the 2D spectra at the indicated dashed line in (A). Peaks after two years of storage at + 4 °C show only minimal loss of signal intensity. Spectra were recorded on a Bruker Avance III HD spectrometer at a 1H resonance frequency of 850 MHz, a magic angle spinning (MAS) rate of 100 kHz, and a temperature of 275 K, using a 0.81-mm MAS probe-head developed by the Samoson group (). The measurement time for both experiments was 7.5 h and all acquisition and processing parameters were the same as described in detail in (Aguion et al., 2021).

Finally, ssNMR 1H-detected experiments at MAS rates above 100 kHz require low sample quantities (300–800 μg for rotors of 0.7–0.8-mm size) (Aguion et al., 2021, Lacabanne et al., 2020, Marchanka et al., 2018b), thus limiting the cost and time-demand of sample preparation.

Characterization of nucleic acid–protein interfaces by MAS ssNMR

The approaches used so far to characterize intermolecular contacts in nucleic acid–protein complexes can be divided in three classes, whereby many of the published studies use a combination of these approaches.

Similar to solution-state NMR, the involvement of a molecular surface in interactions with a binding partner can be detected by either chemical shift perturbations (CSPs) or intensity changes of the ssNMR peaks of the surface atoms, when comparing the free and the bound-state of the molecule (Ahmed et al., 2020, Boudet et al., 2019, Lacabanne et al., 2020, Malär et al., 2021b, Wiegand et al., 2020b, Wiegand et al., 2019, Wiegand et al., 2016, Williamson, 2013). For example, the formation of an hydrogen bond at a nucleic acid–protein interface causes a downfield shift of the involved 1H atom (Wagner et al., 1983). Changes in peak intensities can report on changes in the dynamics of one of the binding partners upon complex formation (Lacabanne et al., 2020, Wiegand et al., 2019). In another study, Asami et al. measured a 15N-1H correlation spectrum of a protein in complex with either 1H or 2H-RNA (Asami et al., 2013) and quantified the difference in protein peaks’ intensities that distinguished the RNA–protein interface, owing to the line-broadening caused by the RNA 1H in the 1H-RNA–protein complex. These effects are specific to the surface of the protein in contact with the RNA, while CSPs can also occur in regions other than the intermolecular surface, due to allosteric effects. In any case, both CSPs and changes in peak intensities can be used as ambiguous restraints in docking protocols.

Intermolecular distances can be measured directly through intermolecular dipolar correlation experiments. In solution-state NMR, intermolecular distances are measured through 13C,15N-edited, 12C,14N-filtered 1H–1H NOESY experiments (Breeze, 2000, Zwahlen et al., 1997). In ssNMR, a plethora of methods yield dipolar correlations, such as cross-polarization (CP) (Hartmann and Hahn, 1962), rotational echo double resonance (REDOR) (Gullion and Schaefer, 1989a, Gullion and Schaefer, 1989b) and the closely related transferred echo double resonance (TEDOR) (Hing et al., 1992), as well as dipolar-assisted rotational resonance (DARR) (Takegoshi et al., 2001, Takegoshi et al., 1999), proton-driven spin diffusion (PDSD) (Szeverenyi et al., 1982), proton spin diffusion (PSD) (Lange et al., 2002, Wilhelm et al., 1998), radio frequency-driven dipolar recoupling (RFDR) (Bennett et al., 1992, Sodickson et al., 1993), or combined R2nv-driven (CORD) (Hou et al., 2013) experiments. One of the most popular approaches for the measurement of nucleic acid–protein distances utilizes TEDOR-based 31P–13C and and 31P–15N correlations (Bechinger et al., 2011, Huang et al., 2011, Huang et al., 2010, Jehle et al., 2010, Olsen et al., 2005, Yu and Schaefer, 2008). 31P has a high gyromagnetic ratio (Table 1) and is present in nucleic acids exclusively and with 100 % isotopic abundance; thus, in the presence of a 13C,15N-labeled protein, these correlation experiments are sensitive and report exclusively on intermolecular contacts between the protein and the nucleic acid.

Table 1

Nuclei relevant for the identification of nucleic acid–protein binding interfaces in ssNMR.

Nuclei	Spin quantum number	Natural abundance (%)	Gyromagnetic ratio γ (107 rad T−1 s−1)	NMR transition frequency at 18.8 T (MHz)
1H	½	99.98	26.7519	800
2H	1	0.015	4.1066	123
13C	½	1.1	6.7283	200
15N	½	0.37	−2.7126	80
19F	½	100	25.1815	753
31P	½	100	10.8394	324

The third approach utilizes paramagnetic relaxation enhancement (PRE) effects, which rely on the interaction between nuclei and unpaired electrons. Because the electron spin gyromagnetic ratio is approximately 600-times larger than that of the 1H nucleus, these effects are large and can be used to measure longer distances than internuclear dipolar correlations. In solution, PRE measurements have been often applied to measure long-range intra- and intermolecular distances (up to 30 Å) in nucleic acid–protein complexes (Amrane et al., 2014, Graziadei et al., 2020, Hennig et al., 2015, Lapinaite et al., 2013, Leeper et al., 2010, MacKereth et al., 2011, Martin-Tumasz et al., 2010). For a detailed description of paramagnetic NMR in solution and solid-state, readers are referred to a comprehensive review by Pell and coworkers (Pell et al., 2019). In ssNMR, PRE measurements were first employed to identify residues of metalloproteins in proximity to the paramagnetic metal (Balayssac et al., 2007b, Balayssac et al., 2007a, Pintacuda et al., 2007); recently, PRE experiments have also been used to study nucleic acid–protein complexes, (Ahmed et al., 2020, Wiegand et al., 2017a, Zehnder et al., 2021). Notably, Ahmed et al. report the first structure of an RNP complex obtained solely from ssNMR data.

Fig. 2 gives an overview of the methods developed and utilized to date to probe nucleic acid–protein interfaces, which will be discussed in detail in the next chapters. A representative set of ssNMR studies of nucleic acid–protein complexes is given in Table 2 in chronological order.

Fig. 2

Overview of ssNMR methods utilized to date to probe nucleic acid–protein interfaces. Information derived from paramagnetic effects, cross-interface dipolar correlations, chemical shifts and peak intensities are highly complementary and can be utilized as restraints in a molecular docking protocol that builds the nucleic acid–protein complex from the structures of the individual components. Pseudo contact shifts (PCSs) measure changes in chemical shifts due to the influence of a paramagnetic ion with anisotropic tensor. Paramagnetic relaxation enhancement (PRE) measures the increase in relaxation of nuclei in the vicinity of an unpaired electron. Both effects can be quantified and translated into nucleus-electron distances. Cross-interface dipolar correlations directly measure intermolecular distances through a variety of dipolar recoupling techniques. The ssNMR toolbox for the measurement of intermolecular distances covers transferred echo double resonance (TEDOR)/ rotational echo double resonance (REDOR)-based recoupling experiments, 13C–13C correlation experiments, such as dipolar assisted rotational resonance (DARR) and combined R2nv-driven (CORD), spin-diffusion based CHHP/NHHP experiments, and 1H-detected CP-based hPH experiments. Chemical shift perturbation (CSP) mapping reports on changes in chemical shifts occurring in the protein upon binding of the nucleic acid, while temperature mapping measures the temperature dependence of 1H chemical shifts, which in turn depends on the involvement of the atom in hydrogen bonds. Comparison of temperature coefficients in the free and nucleic acid-bound states can reveal which protein residues are involved in nucleic acid binding. Quantification of peak intensities can also reveal which atoms are close to the partner molecule in the complex, because of binding-induced line-broadening effects.

Table 2

Selection of ssNMR studies of intermolecular interactions in nucleic acid–protein complexes since 2005.

Year	Reference	Complex type	Site specific information	Protein (aa)	NA (nt)	ssNMR technique
2005	(Olsen et al., 2005)	1:1 dsRNA–peptide	Yes	11	29	31P-19F REDOR
2008	(Yu and Schaefer, 2008)	dsDNA–protein (intact bacteriophage)	No	not known	342,000	15N-31P, 31P-15N REDOR
2010	(Jehle et al., 2010)	1:1 dsRNA–protein	Yes	123	26	31P-15N TEDOR
	(Huang et al., 2010)	1:1 dsRNA–peptide	Yes	11	29	13C/15N-19F REDOR
2011	(Huang et al., 2011)	1:1 dsRNA–peptide	Yes	11	29	13C-31P, 13C-19F, 15N-31P, 15N-19F, 31P-19F REDOR
	(Sergeyev et al., 2011)	ssDNA–protein (intact bacteriophage)	Partially (only for coat protein)	46	7349	13C,13C DARR
	(Bechinger et al., 2011)	Oligomeric dsDNA–peptide	No	27	not known	15N-31P REDOR
2013	(Asami et al., 2013)	1:1 dsRNA–protein	Yes	123	26	Intensity mapping (Dipolar-coupling-mediated line broadening)
2014	(Morag et al., 2014)	ssDNA–protein (intact bacteriophage)	Partially (only for coat protein)	50	8233	13C,13C CORD/DARR, PHHC
2016	(Wiegand et al., 2016)	2:12:12 ssDNA–protein–AMP-PNP	Yes	521	20	CSP (1D 1H 31P CP, 13C,13C DARR)
2017	(Wiegand et al., 2017b)	2:12:12 ssDNA–protein–ADP	No	521	20	2D PC-C, 2D PHHC-C
2019	(Wiegand et al., 2019)	2:12:12 ssDNA–protein–AMP-PCP/ ADP:AlF4-/ATP/ADP	Yes	521	20	CSP (1D 1H 31P CP, 13C,13C DARR, 2D CP 13C-15N), Intensity mapping (13C,13C DARR), NHHP
	(Boudet et al., 2019)	1:1:2 ssDNA–-protein–ATP	Yes	331	9	CSP (13C,13C DARR, 2D CP 13C-15N), CHHP, NHHP
2020	(Wiegand et al., 2020b)	2:12:12 ssDNA–protein– ADP:AlF4-/AMP-PCP	Yes	521	20	NHHP, CHHP, CSP (CP-hNH), Intensity mapping(31P,31P DARR)
	(Lacabanne et al., 2020)	1:1:2 ssDNA–protein–ATP	Yes	331	9	CSP (CP-hNH, CP-hCH), Intensity mapping (CP-hNH)
	(Ahmed et al., 2020)	Monomeric dsRNA–protein	Yes	123	26	PRE, CSP (13C,13C DARR, 2D CP 13C-15N)
2021	(Zehnder et al., 2021)	2:12:12 ssDNA–protein–ADP:AlF4-	Yes	521	20	PRE
	(Malär et al., 2021b)	2:12:12 ssDNA–protein–ADP:AlF4-	Yes	521	20	CP-hPH, CSP temperature mapping (CP-hNH)

Chemical shift perturbations and intensity changes

Both CSPs and peak intensities changes can be used to reveal intermolecular surfaces. CSP mapping can be achieved with all types of NMR spectra. In 13C,15N-detected ssNMR, chemical-shift differences measured in 2D 13C–13C DARR spectra (Ahmed et al., 2020, Boudet et al., 2019, Wiegand et al., 2016, Wiegand et al., 2019) and 2D CP 13C,15N spectra (Ahmed et al., 2020, Boudet et al., 2019, Wiegand et al., 2019) of 13C,15N-labeled proteins in the free and complexed forms were used to identify surfaces involved in nucleic-acid binding (Fig. 3A–B). Wiegand et al. used 1D 1H-31P CP spectra to detect binding of a protein to (dT)20 ssDNA, whose 31P peaks shifted from ∼ –1 ppm in the free form to 0–1 ppm in the protein-bound form (Wiegand et al., 2016, Wiegand et al., 2019). However, nucleic acid CSPs have never been used to reveal the nucleic acid residues involved in protein recognition. This task is indeed challenging, especially for RNAs, as these do not always adopt a single conformation in the free form and thus the observed CSPs may report on both folding and binding.

Fig. 3

CSP mapping to determine nucleic acid–protein binding interfaces in ssNMR. (A) Excerpts of overlaid 2D CP NCACX correlation spectra of free (black) and RNA-bound (green) Pf L7Ae protein, with peaks showing noticeable CSPs upon RNA binding. (B) Same as in (C) for 2D 13C–13C DARR correlation spectra. Figures (A–B) are reproduced from (Ahmed et al., 2020) ©2020 with permission from John Wiley and Sons. (C–D) Temperature-dependent proton chemical-shift perturbations. Residue-specific temperature coefficients and corresponding excerpts of the hNH spectra of DnaB helicase complexed with ADP:AlF4- and ssDNA. Residues in (C) show nearly no dependence of their chemical shifts on temperature and are thus likely involved in hydrogen bonds. The chemical shifts of residues in (D) have a larger dependence on temperature and are thus not involved in hydrogen bonds. Figures (C–D) are reproduced from (Malär et al., 2021b) ©2021 with permission from Springer Nature.

Recently, 1H-detected ssNMR methods, such as CP hNH and hCH experiments (Zhou et al., 2007a, Zhou et al., 2007b, Zhou and Rienstra, 2008), have been utilized to map CSPs of a 13C,15N labeled protein in a DNA–protein complex upon nucleotide binding to yield a DNA–protein–nucleotide complex (Lacabanne et al., 2020, Wiegand et al., 2020b). 1H-detection in MAS ssNMR is expected to greatly enhance the resolution and sensitivity of CSP mapping.

Another chemical-shift-based approach was developed by Malär et al. to detect protein side chains involved in hydrogen bonds with the nucleic acid and the nucleotide in a ssDNA–protein–nucleotide complex (Malär et al., 2021b). The method measures the temperature dependence of 1H chemical shifts and correlates a weak dependence with the involvement of the 1H in a hydrogen bond. This correlation was often demonstrated in solution (Baxter and Williamson, 1997, Cierpicki et al., 2002, Cierpicki and Otlewski, 2001) and more recently also in ssNMR (Malär et al., 2021a). Mapping the dependence of chemical shifts on temperature in a CP hNH experiment, Malär et al. were able to confirm that the chemical shifts of 1H protein atoms previously identified as forming hydrogen bonds with the nucleotide and the DNA vary very little with temperature and certainly much less than those not involved in hydrogen bonds (Fig. 3C–D). These experiments, when applied to proteins in both the free and nucleic acid-bound states, can reveal which protein amino acids are involved in nucleic acid recognition.

The first study that employed quantification of ssNMR peak intensities to determine intermolecular surfaces in an RNP complex was communicated by Asami et al. (Asami et al., 2013), who also pioneered 1H-detection in MAS ssNMR of uniformly 13C,15N labeled RNA at moderate MAS frequencies of 20 kHz. The 1H-15N cross peaks of protein and RNA detected in a hNH experiment (Zhou et al., 2007a, Zhou et al., 2007b, Zhou and Rienstra, 2008) do not significantly overlap, with the exception of those belonging to RNA amino groups and protein arginine sidechains (Fig. 4A). This allowed monitoring both components of the complex simultaneously. To obtain narrow linewidths in the hNH experiment at low MAS speed, both RNA and protein were 2H,15N labeled and the labile 2H were back-exchanged to 1H to a 10 % extent. To identify the surface of the protein that binds the RNA, Asami et al. measured protein 1H-15N correlations of two samples, containing 2H,15N-labeled protein and either 2H,13C,15N-labeled RNA or 13C,15N-labeled RNA. By comparing the intensities of the 1H-15N protein cross peaks in the two samples they could identify which protein amino acids form the RNA–protein interface, as the 1H-15N peaks of these amino acids had reduced intensities in the sample containing non-deuterated RNA, due to the line broadening of the protein HNs caused by the dipolar interaction with the nearby RNA 1H atoms. Spectra simulations using the program SIMPSON (Bak et al., 2000) demonstrated that a measurable effect occurred when the distance between the protein HN and any RNA hydrogen was less than 6–8 Å, while the intensities of the protein amide peaks further than 8 Å from the RNA were identical in the two spectra (Fig. 4B). In principle, the ratio of signal intensities in the two samples can be quantified and converted in a distance value; however, as the experiment does not reveal which RNA hydrogen causes the line broadening of a given protein HN, these distances cannot be used as unambiguous restraints in structure calculations.

Fig. 4

Dipolar-coupling-mediated line broadening effects to probe intermolecular interactions. (A) 2D CP hNH correlation spectrum of 2H,15N labeled Pf L7Ae protein in complex with 2H,13C,15N labeled 26mer Box C/D RNA (2H-RNA), showing the characteristic protein and RNA spectral patterns. Labile protons were back exchanged to 1H in a solution containing 10 %/90 % H2O/D2O. (B) Excerpts of 2D CP hNH spectra acquired on RNA–protein complexes containing either 13C,15N-labeled 26mer Box C/D RNA (1H-RNA, left) or 2H-RNA (right). The NMR peaks of protein residues near the RNA (Gly35, Thr36, and Ala97) are less intense in the spectrum of the sample containing 1H-RNA, due to line broadening caused by the 1H(protein)-1H(RNA) dipolar couplings. Peaks from amide groups distant from the RNA hydrogens have similar intensities in the two samples (Thr29 and Ser82). Figures (A–B) are reproduced from (Asami et al., 2013) ©2013 with permission from John Wiley and Sons.

Other studies detected side-chain-specific nucleic acid–protein interactions by monitoring the appearance or disappearance of arginine and lysine side chain peaks in 15N-13C CP correlation spectra or CP hNH spectra upon binding of either nucleic acids or nucleotides (Lacabanne et al., 2020, Wiegand et al., 2019). The appearance of the peaks of these side chains was due to the loss of local dynamics connected with binding. Wiegand et al. also detected the appearance of a 31P-31P correlation peak in the 31P-31P DARR spectrum of the (dT)20 ssDNA upon binding to the DnaB–ADP:AlF4 binary complex, as a result of the rigidification of two of the 20 ssDNA nucleotides when they formed the ssDNA–protein–nucleotide complex (Wiegand et al., 2020b).

Cross-interface dipolar correlations

MAS ssNMR techniques to measure internuclear distances were first applied to an RNA–peptide complex by Olsen et al., who, however, did not measure intermolecular distances but detected peptide binding through the change in the value of a 19F-31P distance in the RNA (Olsen et al., 2005). The 19F atom was introduced as a 2′-fluorine at an individual site, while the 31P belonged to a single phosphorothioate (pS) label in the RNA backbone and thus could be easily assigned (Fig. 5A–B). The distance was measured in a 31P-19F REDOR experiment (Gullion and Schaefer, 1989a, Gullion and Schaefer, 1989b, Merritt et al., 1999) (Fig. 5C). Distance measurements between two unique sites were used before to detect binding in a DNA–protein complex (Yu et al., 2004) and in DNA–small molecule complexes (Mehta et al., 2004, Olsen et al., 2003). The introduction of fluorine atoms at individual position in nucleic acids, either in the ribose as 2′-F, or in the base, as 5-fluorouridine (5FU) and 5-fluorocytidine (5FC), has become popular because of the large chemical shift dispersion of 19F and the high sensitivity of its chemical shift to the structural and chemical environment (Hennig et al., 2007, Marchanka et al., 2018a, Scott et al., 2004). Moreover, substitution of 1H by 19F in nucleic acids has a negligible effect on both structure (Hennig et al., 2007, Merritt et al., 1999) and intermolecular interactions (Olsen et al., 2005).

Fig. 5

Experiments utilizing TEDOR/REDOR recoupling to probe intermolecular contacts. (A–C) Conformational changes in TAR RNA upon binding to the Tat peptide. (A) Conformation of the free TAR RNA (PDB entry #1ANR). U: uridine; A: adenosine (B) Conformation of the TAR RNA bound to the Tat peptide (PDB entry #1ARJ). The RNA phosphate backbone is represented by a grey ribbon. Residues U23 and A27 are shown in sticks. Orange spheres mark the U23 2́-F and A27 pS positions. (C)31P-19F REDOR dephasing curves for free TAR RNA (triangle) and the TAR RNA–Tat peptide complex (rhombus). Solid lines represent simulated dephasing curves corresponding to an internuclei distance of 10.3 Å in the free TAR RNA and 6.6 Å in the TAR RNA–Tat peptide complex. Figures (A–C) are adapted from (Olsen et al., 2005) ©2005 with permission from Oxford University Press. (D–E) Intermolecular cross-peaks between the Pf 26mer Box C/D and L7Ae. (D) Left, solution 1H-15N HSQC spectrum of the Pf L7Ae protein bound to Box C/D RNA. Right, 31P-15N TEDOR spectrum showing cross-peaks between the amide 15N nuclei of the 15N-labelled L7Ae protein and the 31P nuclei of the unlabeled Box C/D RNA. Dashed lines from the left to the right panel connect peaks with the same 15N chemical shift. A: alanine; G: glycine; N: asparagine; S: serine; T: threonine. (E) Intermolecular distances between the L7Ae backbone 15N nuclei and the U20 31P nucleus of the Box C/D RNA measured with the 31P-15N TEDOR experiment in (D). Figures (D–E) are reproduced from (Jehle et al., 2010) ©2010 with permission from American Chemical Society.

The substitution of an individual phosphate group with a thiophosphate has also been widely used in NMR spectroscopy of nucleic acids to unambiguously identify individual 31P atoms thanks to the downfield shift by 50–60 ppm induced by the sulfur atom on the 31P chemical shift (Huang et al., 2011, Huang et al., 2010, Merritt et al., 1999, Olsen et al., 2005, 2003). Also in this case, the pS substitution results in little-to-no perturbation to the global structure, as demonstrated by several high-resolution NMR studies for DNA:RNA hybrids (Bachelin et al., 1998, González et al., 1995, Merritt et al., 1999) and crystal structures of wild-type and pS-substituted DNA duplexes (Cruse et al., 1986). However, alterations of RNA conformation caused by pS substitutions have been reported as well (Smith and Nikonowicz, 2000). In addition, the different charge distribution and polarizability of thiophosphate in comparison to phosphate alters the strength of both ionic interactions and hydrogen bonds and thus, potentially, nucleic acid–protein affinities (Milligan and Uhlenbeck, 1989, Schnitzer and Von Ahsen, 1997). Consequently, it is safer to use methods that do not require chemical modification of the nucleic acid to measure internuclear distances in the context of nucleic acid–protein complexes. With the recent developments in the ssNMR experimental design and sample preparation techniques, it is nowadays possible to obtain site-specific assignments of 31P resonances in the backbone of 13C,15N labeled RNA by, for example, 13C-31P-TEDOR and 13C-31P-TEDOR-13C–13C-PDSD experiments (Marchanka et al., 2015, Marchanka and Carlomagno, 2019). In addition, 1H-detected hPH experiments serve as useful complement to confirm 31P resonance assignments (Malär et al., 2021b).

13C-31P, 15N-31P, 13C-19F and 15N-19F REDOR experiments on an RNA–peptide complex (Huang et al., 2011, Huang et al., 2010), as well as 31P-15N TEDOR experiments on an RNA–protein complex (Jehle et al., 2010) were used to measure the first intermolecular RNA–peptide and RNA–protein distances through the dipolar coupling between the RNA 19F and 31P nuclei and the protein/peptide 13C and 15N nuclei (Fig. 5D–E). These experiments were performed on complexes containing 13C,15N-labeled protein and unlabeled or selectively fluorinated RNA and yielded exclusively intermolecular cross peaks. Huang et al. utilized a selectively labeled peptide, where a single arginine was 13C,15N enriched, and 5FU/pS-modified RNA (Huang et al., 2011, Huang et al., 2010). Besides the problems discussed above with regards to the impact of chemical modifications on RNA structure and interactions, the use of selectively labeled molecules requires these molecules to be available by chemical synthesis, to allow the easy incorporation of labeled or modified building blocks at any position in the primary sequence. This requirement limits the applicability of the strategy. In a more general approach, Jehle et al. measured 31P-15N RNA backbone-protein backbone distances between a uniformly 13C,15N labeled protein and an unlabeled RNA (Jehle et al., 2010). As the assignment of both 31P RNA (Marchanka et al., 2015, Marchanka and Carlomagno, 2019) and 15N protein (Andreas et al., 2016, Castellani et al., 2003, Schubeis et al., 2018, Zhao, 2012) peaks can be obtained by ssNMR experiments, the method is broadly applicable.

Other studies used 15N-31P REDOR experiments to probe intermolecular contacts between either bacteriophage T4 DNA or salmon sperm DNA 31P atoms and 15N atoms of lysine side chains in either a DNA–protein (Yu and Schaefer, 2008) or a DNA–peptide complex (Bechinger et al., 2011). In both cases, the interacting atoms were not assigned, and the experiments were used solely to prove binding.

Following a different approach, Sergeyev et al. (Sergeyev et al., 2011) were able to detect cross-peaks between the uniformly 13C,15N labeled ssDNA of the Pf1 bacteriophage virion and a specific tyrosine residue of the viral coat protein using dynamic nuclear polarization (DNP) enhanced homonuclear 13C–13C DARR correlation experiments (Abragam and Goldman, 1978, Akbey and Oschkinat, 2016, Su et al., 2015, Takegoshi et al., 2001, Takegoshi et al., 1999). The aromatic stacking interactions between the DNA bases and the tyrosine had been previously predicted but had remained unproven, until they were detected by ssNMR (Fig. 6).

Fig. 6

Homonuclear 2D 13C–13C DARR spectrum of uniformly 13C,15N-labeled Pf1 bacteriophage virion. The spectrum shows cross peaks between the ssDNA ribose and base atoms and a specific tyrosine residue (Y40) of the viral coat protein, as highlighted by orange boxes. The figure is reproduced from (Sergeyev et al., 2011) ©2011 with permission from American Chemical Society.

In another study (Morag et al., 2014), fd bacteriophages were prepared on a 13C,15N enriched medium supplemented with non-isotopically labeled aromatic amino acid precursors to obtain particles with uniformly labeled DNA and selectively-unlabeled proteins. Assignment of atom-types of the fd bacteriophage circular ssDNA was obtained using a 13C–13C CORD correlation experiment (Hou et al., 2013). However, due to the large size of the DNA, site-specific resonance assignment was not attempted. The contacts between the ssDNA and the virion capsid proteins were measured in 13C–13C CORD and 13C–13C DARR experiments. The intermolecular correlations could be distinguished from the intramolecular ones due to the partially different chemical shift ranges of 13C atoms in DNA and proteins. DNA phosphate-capsid protein interactions were detected utilizing homonuclear 1H–1H PSD mixing (Lange et al., 2002, Wilhelm et al., 1998) coupled with P—H and H—C CP transfers (Hartmann and Hahn, 1962) in a PHHC experiment. The PHHC experiment yielded intramolecular correlations between the DNA phosphate backbone and 13C nuclei of the DNA riboses and bases as well as intermolecular correlations between the DNA phosphate backbone and side chains of amino acids of the coat protein, in particular lysines (Fig. 7A). This study is remarkable, due to the large size of the particles under investigation. A similar approach could be used to reveal RNA–protein interfaces in RNP complexes, possibly in combination with site-specific assignments. However, neither 13C–13C DARR nor PHHC experiments have been applied to measure RNA-protein distances so far.

Fig. 7

1H–1H PSD-based experiments to probe intermolecular contacts. (A) 2D PHHC correlation spectrum of intact fd bacteriophage to probe intermolecular interactions between DNA phosphate and protein carbon atoms. fd bacteriophages were prepared on a 13C,15N enriched medium supplemented with non-isotopically labeled aromatic amino acid precursors to obtain particles with uniformly 13C,15N labeled DNA and selectively unlabeled proteins. The figure is reproduced from (Morag et al., 2014) ©2014 with permission from American Chemical Society. (B–C) Intermolecular correlations between uniformly 13C,15N-labeled pRN1 archaeal primase and a 9 nt functional DNA sequence, as well as bound ATP. (B) 2D CHHP correlation spectrum. (C) 2D NHHP correlation spectrum. Figures (B–C) are reproduced from (Boudet et al., 2019) ©2019 with permission from Elsevier.

The CHHP experiment, which is conceptually the same as the PHHC experiment, but with detection of 13C and 31P chemical shifts in t1 and t2, respectively, and the NHHP experiment, where polarization is transferred between the protein 15N nuclei and the DNA 31P nuclei through 1H–1H spin diffusion, were utilized in other studies (Boudet et al., 2019, Wiegand et al., 2019, Wiegand et al., 2020b) to observe intermolecular 13C-31P and 15N-31P contacts between a 13C,15N-labeled protein and DNA (Fig. 7B, C).

All these 31P-13C/15N heteronuclear correlation experiments suffer from low signal-to-noise ratio (SNR), in particular for large complexes. The problem can be addressed using DNP, which can significantly enhance the SNR of NMR spectra exploiting the transfer of the large electron polarization to nuclear spins. However, biomolecular MAS-DNP studies are typically performed at cryogenic temperatures, which lead to significant broadening of the NMR lines (Bauer et al., 2017, Siemer et al., 2012, Siemer and McDermott, 2008). Nevertheless, in DNP-enhanced 31P-13C CP and PHHC transfer experiments, Wiegand et al. measured intermolecular contacts between a 13C,15N labeled protein and either ssDNA or ADP in a ssDNA–protein–ADP complex (Wiegand et al., 2017b). In both cases, a 13C–13C DARR mixing step was added after the intermolecular transfer step to reduce spectral overlap, resulting in 2D 13C,13C (P)C—C and (ii) 2D (PHH)C—C correlation spectra. Due to the line broadening induced by cryogenic temperatures, the experiments yielded information only on the protein amino-acid type in contact with either ADP or ssDNA, but allowed neither site-specific protein assignments nor discrimination between the 31P chemical shifts of the ssDNA and ADP. Of note, the 1H–1H spin-diffusion-driven PHHC-C correlation is less selective than the direct CP-driven PC-C correlation but can yield longer 31P-13C distances of the order of 7–10 Å.

Finally, Malär et al. have recently probed spatial proximities between protein 1H nuclei and nucleic acid/nucleotide 31P nuclei in a ssDNA–protein–ADP complex utilizing a 1H-detected 13P-1H correlation experiment under high MAS speed of 105 kHz (Fig. 8) (Malär et al., 2021b). The CP hPH experiment can be performed with unlabeled material and appears to be much more sensitive for probing spatial proximity of hydrogen atoms to phosphate groups than the previously described 1H–1H PSD-based CHHP and NHHP experiments.

Fig. 8

CP-based hPH experiment to probe intermolecular contacts. (A) 2D CP-based hPH correlation spectrum of uniformly 13C,15N-labeled DnaB helicase in complex with ADP:AlF4- and ssDNA. Intramolecular correlations of phosphate groups with ADP protons are highlighted in green; intramolecular correlations of phosphate groups with DNA protons are highlighted in light purple and red. The remaining peaks are assigned to intermolecular correlations of phosphate groups with amide backbone protons, except for those indicated by an asterisk, which are associated with sidechain protons. The figure is reproduced from (Malär et al., 2021b) ©2021 with permission from Springer Nature. (B) Magnetization transfer scheme of the 2D hPH experiment exemplarily shown for a phosphate group in close contact with the amide backbone. Encircled numbers indicate the chemical-shift evolution times corresponding to the two spectral dimensions, and the roman numerals indicate the CP transfer periods.

Paramagnetic relaxation enhancement

The first implementation of PREs in ssNMR of a protein–nucleotide complex took advantage of the fact that ATP-binding is often accompanied by binding of Mg2+ ions, which can be substituted by paramagnetic ions such as Mn2+ (Bonneau and Legault, 2014, Tamaki et al., 2016) and Co2+ (Balayssac et al., 2008, Bertini et al., 2010b) without affecting protein function (Otting, 2010). The presence of a paramagnetic center can cause an increase of the relaxation rates (paramagnetic relaxation enhancement, PRE) and/or a change in the chemical shifts (pseudo-contact shifts, PCS) of close-by nuclei (Jaroniec, 2012). These effects dependent quantitatively on the distance between the nucleus and the paramagnetic center. Through the analysis of signal intensities in 2D 13C,13C DARR spectra of the diamagnetic (in the presence of Mg2+) and paramagnetic (in the presence of Mn2+) protein–nucleotide complex, Wiegand et al. identified protein residues close to the nucleotide binding site (Wiegand et al., 2017a). Recently, Zehnder et al. extended this approach to a ssDNA–protein–nucleotide complex, in which Mn2+ and Co2+ ions were used as paramagnetic centers (Zehnder et al., 2021). 2D 13C,13C DARR, as well as 2D NCA and 3D NCACB spectra (Baldus et al., 1998) were recorded for diamagnetic and paramagnetic ssDNA–protein–nucleotide complexes (Fig. 9A) to localize the metal ion in the complex. Moreover, PCSs of the (dT)20 ssDNA observed in 1D 1H-31P CP spectra allowed to localize the DNA phosphate groups in the complex (Fig. 9B).

Fig. 9

Paramagnetic effects in ssNMR of nucleic acid–protein complexes. (A) Threonine region of 13C–13C correlation spectra of DnaB helicase in complex with ADP:AlF4- and ssDNA in the presence of Mg2+ (purple), paramagnetic Mn2+ (pink) and paramagnetic Co2+ (blue). Peaks which are no longer visible in the paramagnetic spectra, due to PRE effects, are marked with red crosses. Changes in chemical shifts (PCS) are highlighted by red arrows in the sample with Co2+. (B) 1D 1H-31P CPMAS NMR of the ssDNA–DnaB–ADP:AlF4- complex in the presence of either Mg2+ (black) or Co2+ (blue). Figures (A–B) are reproduced from (Malär et al., 2021b) ©2021 with permission from John Wiley and Sons. (C) Overlay of 2D 13C,13C SPC-5 correlation spectra of Alab 26mer Box C/D RNA (top panel) and Ulab 26mer Box C/D RNA (bottom panel) in complex with nitroxide-tagged L7Ae-K32C protein in the diamagnetic (grey) and paramagnetic (blue) state. Peaks of residues in proximity to the paramagnetic tag have reduced signal intensities in the paramagnetic state. (D) Ribbon representation of the L7Ae protein (PDB entry #6TPH) showing the residues selected for coupling with the paramagnetic tag (K32, N38 and I93) represented as blue spheres. Figures (C–D) are adapted from (Ahmed et al., 2020) ©2020 with permission from John Wiley and Sons.

If the molecule of interest does not have a well-defined, individual metal binding site, paramagnetic centers can be introduced in nucleic acid–protein complexes by coupling nitroxide-based tags, one-by-one, to individual cysteine residues engineered at specific protein sites (Nadaud et al., 2007). The PRE effects caused by the paramagnetic tag on the RNA can be used to measure RNA–protein distances in the RNP complex (Ahmed et al., 2020). Ahmed et al. measured PRE effects induced in an RNP complex by protein-coupled paramagnetic tags on 13C,15N-labeled RNA in 2D 13C,13C SPC-5 correlation spectra (Hohwy et al., 2002, Hohwy et al., 1999) (Fig. 9C). Unlike CSPs, which are sensitive also to allosteric effects, PRE-effects are specific reporters of intermolecular contacts. To avoid PRE effects generated by interparticle contacts in the precipitate, the paramagnetic-labeled RNP complex was diluted in 1:3 molar ratio with the diamagnetic, unlabeled RNP complex.

To obtain enough RNA-protein distance restraints by this methodology, the paramagnetic tag should be placed close to the RNA binding site but in such way as not to affect binding (Fig. 9D). CSP data, discussed previously, can guide in the recognition of the RNA binding surface and in the design of the cysteine mutants. The integrity of the complex, after introduction of the nitroxide tag, should be verified by careful inspection of the NMR spectra and of other biophysical properties of the complex. As far as nucleic acids are concerned, paramagnetic tags can be covalently attached to either the phosphate backbone or the ribose or modified bases. In alternative, modified paramagnetic nucleotides can be introduced into the RNA by solid-phase synthesis (Miao et al., 2021). Although paramagnetically tagged RNA has not been used in ssNMR so far, these techniques can be easily applied to increase the number of paramagnetic sites that can be probed in nucleic acid–protein complexes.

Structure determination of an RNP complex by ssNMR only

Ahmed et al. succeeded in determining the structure of the complex formed by the Pf protein L7Ae and the 26mer Box C/D RNA using exclusively ssNMR-derived restraints (Ahmed et al., 2020). They achieved this goal through a docking protocol, which started from the individual structures of protein-bound 26mer RNA determined by ssNMR (Marchanka et al., 2015) and RNA-bound L7Ae protein (Xue et al., 2010). The RNA-bound protein structure was determined by X-ray crystallography, but in principle it would have been accessible by ssNMR through well-established methodology (Schubeis et al., 2018, Zhao, 2012). The docking protocol was implemented in the program Haddock (Dominguez et al., 2003) and used a combination of ssNMR-derived CSP and PRE data to guide complex assembly.

PRE-based restraints were derived from the quantification of PRE effects observed on nucleotide-type specific 13C,15N labeled (Alab and Ulab) RNA in 2D 13C,13C SPC-5 spectra and induced by 3 different paramagnetic tags coupled to the unlabeled protein. PRE effects were quantified as peak volume ratios in spectra measured for the same sample in either the paramagnetic (with the nitroxide radical coupled to a cysteine residue) or the diamagnetic state (after addition of ascorbic acid to reduce the radical). These ratios (V) were converted into distance restraints in a semi-quantitative manner, whereby the intramolecular V ratios of the protein peaks, whose electron-nucleus distances were known from the protein structure, were used to calibrate a linear regression between the V ratio and the electron-nucleus distance. Although not rigorously correct, this linear regression, with the appropriate tolerance bounds, was good enough to deliver broad distance ranges that defined the relative position of the protein and the RNA to a good level of precision.

CSP-derived ambiguous restraints were measured for the protein in 2D 13C–13C DARR and 2D CP 13C,15N spectra of uniformly 13C,15N labeled protein in the free and RNA-bound states.

In total 72 restraints were used for the docking and lead a 26mer Box C/D RNA–L7Ae structure that is very similar to that of an orthologous Box C/D RNA–L7Ae complex determined by crystallography, thus verifying the accuracy of the ssNMR-derived structure (Fig. 10). This exemplary study demonstrates the ability of ssNMR to yield nucleic acid–protein complex structures at atomic resolution.

Fig. 10

Overlay of the MD-refined ssNMR structure (blue) (PDB entry #6TPH) of the Pf Box C/D RNA–L7Ae complex with the crystallographic structure of the orthologous Box C/D RNA–L7Ae complex from Archaeoglobus fulgidus (Af) (gold) (PDB entry #4BW0) (Huang and Lilley, 2013). The figure is reproduced from (Ahmed et al., 2020) ©2020 with permission from John Wiley and Sons.

Conclusions and outlook

The ssNMR methods discussed here build a suite of complementary experiments that identify interaction interfaces and yield intermolecular distance restraints in nucleic acid–protein complexes. They provide a powerful tool to determine the structural basis of intermolecular recognition for those nucleic acids–protein complexes that are not amenable to X-ray crystallography, cryo-EM or solution-state NMR. The structure of the individual components of the complex can be obtained by ssNMR as well as with previously published and often reviewed methodology (Marchanka and Carlomagno, 2019, Schubeis et al., 2018, Zhao, 2012).

In 2021 the program AlphaFold (Jumper et al., 2021) revolutionized structural biology of folded protein domains and their complexes, by demonstrating an unprecedented accuracy in the prediction of protein folding and interactions based on database knowledge. Nucleic acid structures, in particular RNA, are difficult to predict, as they largely depend on the environment and on the binding partners. Thus, it is unclear whether AlphaFold will ever be expanded to this class of polymers. In view of this, NMR spectroscopy, with its power to illuminate intermolecular interactions involving flexible molecules, gains unique relevance.

To date, many studies of nucleic acid–protein complexes have focused on obtaining site specific information for the proteins in the complex. The nucleic acid component has been often neglected due to lack of spectral resolution and limited access to advanced isotope labeling techniques. In the past few years, the ssNMR toolkit for studying nucleic acids and their complexes has grown steadily, including both technical developments in MAS ssNMR and isotope labeling techniques. Consequently, we expect MAS ssNMR of nucleic acid–protein complexes to rapidly grow in relevance and scope, including site specific structural information on both the protein and nucleic acid components of the complex.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.