RNAs for NMR studies are typically prepared by in vitro transcription (Milligan et al., 1987; Milligan and Uhlenbeck, 1989) or chemical synthesis (Beaucage and Reese, 2009). While chemical synthesis can produce RNA with sophisticated site-specific labeling, this method is limited to RNA of ∼70 nt in length and is not available in most laboratories. On the other hand, in vitro transcription can deliver RNA of any length labeled uniformly or selectively by nucleotide type and is potentially accessible to any laboratory. In this review, we will mostly discuss experiments suitable for the assignment of RNA that is either uniformly labeled or selectively labeled by nucleotide type.

Solution-state NMR studies of RNA have an intrinsic MW limit of ∼40 kDa (120–150 nt) and larger RNA can be assigned only partially; advanced structural studies on large RNA use many differently labeled samples and sophisticated experiments (Lu et al., 2010; Keane et al., 2015; Brown et al., 2020). While ssNMR can, in principle, be applied to RNA of any size, the feasibility of ssNMR studies is determined by the quality of sample preparation, which has a direct impact on the spectral linewidth and therefore on spectral crowding. It is commonly accepted that ¹³C linewidths smaller than 1 ppm are necessary to perform resonance assignment and obtain structural data on non–site-specific labeled samples. Due to the limited number of ssNMR RNA studies, statistics on the quality of different sample preparation techniques are very scarce; nevertheless, some patterns have been identified. Typical linewidths of lyophilized sample preparations are significantly larger than 1 ppm (Olsen et al., 2005; Huang et al., 2011) since insufficient hydration ultimately leads to large inhomogeneous broadening. As stated, linewidths greater than 1 ppm are not sufficient for the resonance assignment, which is, however, not necessary in studies utilizing site-specifically labeled RNAs (Olsen et al., 2005).

The most commonly used sample preparation method of RNA labeled uniformly or selectively by nucleotide type for both ¹³C-detected and ¹H-detected ssNMR studies is micro (nano)-crystallization. This technique was developed for ssNMR studies of proteins (McDermott et al., 2000; Franks et al., 2005; Bertini et al., 2010) and has been successfully applied to study RNA (Huang et al., 2012; Marchanka et al., 2013; Yang et al., 2017). This method provides typical ¹³C linewidths of 1 ppm (29mer HIV TAR RNA) (Huang et al., 2012) or even 0.5 ppm for the 26mer box C/D RNA in complex with L7Ae protein (Marchanka et al., 2013; Marchanka et al., 2015). The same complex shows an ¹H linewidth of separated C1′-H1′ resonances in the protonated ribose of 150 Hz (0.15 ppm) by ¹H-detection at 100 kHz MAS on a 1 GHz spectrometer, while the ¹H linewidth of imino resonances on an 850 MHz spectrometer estimates to 200–300 Hz (0.23–0.35 ppm) (Aguion et al., 2021). In the ¹H-detected imino spectrum of 26mer DIS-HIV-1 RNA acquired at 40 kHz MAS, the linewidths of the ¹H and ¹⁵N dimensions are equal to 500 Hz (0.9 ppm) and 70 Hz (1 ppm), respectively (Yang et al., 2017).

A novel ethanol precipitation method introduced by the Wang group (Zhao et al., 2019) utilizing 75% D₂O/25% H₂O-based buffer yields a promising ¹⁵N linewidth of 80 Hz (1.3 ppm) and 70 Hz (1.2 ppm) for imino resonances of 26mer DIS-HIV-1 RNA and 71mer RiboA71 domain of add adenine riboswitch, respectively. The ¹H linewidth of imino resonances in these two RNAs was generously estimated to 230 Hz (0.38 ppm) and 132 Hz (0.22 ppm), respectively. While RiboA71 showed a good chemical shift dispersion so that different spin systems could be identified and even tentatively assigned based on the known solution-state chemical shifts (Zhao et al., 2019), the chemical shift range for DIS-HIV-1 was very narrow and no identification of individual spin systems was possible. It can be speculated that for the isolated 26mer DIS-HIV-1 RNA, the tertiary structure and the local order are partially lost upon EtOH precipitation, while for the well-folded riboswitch RiboA71, the tertiary structure is preserved. Further investigations into this matter are required, but this approach undoubtedly added a new technique to the repertoire of ssNMR RNA sample preparations. Unfortunately, this method cannot be applied toward protein–RNA complexes due to the instant precipitation of most proteins under such conditions.

While both microcrystallization and ethanol precipitation methods demonstrate the general feasibility of RNA resonance assignment by ssNMR, the typically obtained linewidths are significantly worse than those in solution-state NMR (< 0.1 ppm for ¹H). It implies that nucleotide-type selective labeling is necessary to perform ¹³C-detected ssNMR studies even on short (< 30 nt) RNA (Marchanka et al., 2018a), while ¹H-detection at MAS > 60 kHz allows us to study such RNA using a single uniformly labeled sample (Marchanka et al., 2018b). For larger RNA, site-specific or segmental labeling of short (< 30 nt) RNA stretches should be utilized to make both ¹³C-detected and ¹H-detected ssNMR studies feasible (Marchanka et al., 2018a).

Solution-state NMR spectroscopy provides rapid information about the secondary structure of RNA and identifies canonical Watson–Crick (WC) and non-WC base pairs by observation of characteristic chemical shifts of immobilized amino (NH₂) and imino (NH) groups (Wacker et al., 2020). In solution-state NMR, site-specific assignment of ¹³C resonances is typically not necessary to site-specifically assign imino resonances and identify the RNA secondary structure. 2D ¹H,¹H imino NOESY/3D ¹H,¹⁵N HMQC-¹H,¹H NOESY (Nikonowicz and Pardi, 1993) coupled with HNN-COSY (Dingley and Grzesiek, 1998; Pervushin et al., 1998) and ¹H,¹⁵N-TROSY (Favier and Brutscher, 2011) experiments provide imino assignment and allow imino–imino sequential walk for the nucleotides in the base-paired and/or stacked regions. However, if complete sequential assignment and, especially, determination of the three-dimensional structure of RNA are aimed at, full assignment of RNA resonances is necessary. This task comprises several steps and includes spin system–specific assignment of all ribose atoms, determination of ribose–base connections, and assignment of base resonances, followed by sequential assignment via ¹³C-edited/filtered ¹H,¹H NOEs (Zwahlen et al., 1997; Breeze, 2000). Detailed description of solution-state NMR methods for the RNA assignment can be obtained from the classic work by the Schwalbe group (Fürtig et al., 2003) and from a few recent reviews (Scott and Hennig, 2008; Barnwal et al., 2017).

Since dipolar couplings are preserved in solid-state NMR, they can be utilized to provide rapid insights into the secondary structure of RNA by direct observation of base–base correlations. In contrast to solution-state NMR, ssNMR spectra not only show resonances from the base-paired nucleotides but also from any other immobilized nucleotides, so that observation of amino or imino resonances is not necessarily indicative of a base pairing.

In conventional ¹³C/¹⁵N-detected ssNMR spectroscopy at MAS < 20 kHz, base pair information can be obtained directly by measuring ¹⁵N–¹⁵N correlations between base-paired nucleotides, either through direct dipolar transfers via radiofrequency-driven dipolar recoupling (RFDR) (Bennett et al., 1992) or proton-assisted recoupling (PAR) (Lewandowski et al., 2009), or via spin diffusion (SD)-based experiments, for example, proton-driven SD (PDSD) (Szeverenyi et al., 1982) and dipolar-assisted rotational resonance (DARR) (Takegoshi et al., 2001), or via proton spin diffusion (PSD) NHHN/NHHC experiments (Lange et al., 2002; Riedel et al., 2005a; Herbst et al., 2008). The Görlach laboratory has directly observed canonical WC G:C base pairs in (CUG)₉₇ RNA (Figure 2A), employing both ¹⁵N,¹⁵N RFDR (Leppert et al., 2004) (Figure 2B) and NHHN (Riedel et al., 2005a) experiments (Figure 2C). In our study on 26mer box C/D RNA, we have acquired both RFDR and NHHN spectra that were sufficient for the detection of WC G:C (Figures 2D,E) and A:U base pairs. Site-specific assignment via imino–imino sequential walk was not possible in ¹⁵N-detected spectra due to poor ¹⁵N chemical shift dispersion and typical ¹⁵N linewidths ≥ 1 ppm. Recently, the Goldbourt laboratory has employed ¹⁵N,¹⁵N PDSD and ¹⁵N,¹⁵N PDSD∙RFDR experiments to identify the presence of WC G:C and wobble G:U base pairs in native 1 MDa-sized bacteriophage MS2 RNA and obtained nucleotide-type–specific assignments (Lusky et al., 2021). In all ¹⁵N-detected experiments described above, after the initial ¹H-¹⁵N cross-polarization (CP), magnetization is evolved on ¹⁵N during t ₁ . Following that, ¹⁵N magnetization is spread to nearby nitrogens, either directly by various ¹⁵N–¹⁵N recoupling schemes (Figure 2B) or indirectly through protons via the N→H-PSD-H→N scheme (Figure 2C). Finally, the ¹⁵N magnetization is recorded during t ₂ .

¹H-detected ssNMR on RNA amino and imino groups is possible in two different regimes depending on the MAS frequency used. Since imino and amino protons are exchangeable, high level of deuteration will reduce the network of ¹H, ¹H dipolar couplings and therefore make ¹H resonances observable even at MAS frequencies of ∼20 kHz. The Reif and Carlomagno groups used this approach to acquire an ¹H-¹⁵N dipolar-based CP-HSQC (Zhou et al., 2007) spectrum of deuterated 26mer box C/D RNA in complex with L7Ae protein at 24 kHz MAS in 90% D₂O buffer (Asami et al., 2013). Although the recorded spectrum showed disperse imino resonances of several nucleotides, sequence-specific assignment via sequential walk has not been attempted due to low sensitivity. Recently, the Wang laboratory (Yang et al., 2017) has acquired proton-detected ¹⁵N,¹⁵N PAR correlations at 40 kHz MAS to obtain information about WC G:C and A:U base pairs. In their hCN(PAR)NH experiment, after initial CP transfer from ¹H to ¹³C, a specific ¹³C-¹⁵N CP step is used to transfer magnetization to nitrogen atoms, whose chemical shifts are recorded in t₁. Magnetization transfer across the base pair is achieved by ¹⁵N–¹⁵N PAR transfer with a contact time of 7 ms. A final CP transfer brings the magnetization back to ¹H for detection (Figure 2F). Their approach was successful as they were able to confirm the presence of WC G:C and A:U base pairs in the 26mer DIS-HIV-1. Unfortunately, the low resolution in the ¹H dimension (∼500 Hz for imino protons) did not allow the identification of different spin systems and thus the unambiguous assignment of hydrogen bonds. In our recent study at 100 kHz MAS, in view of good linewidths in the proton dimension (200–300 Hz), spin system–specific detection of base pairs was possible (Aguion et al., 2021), allowing rapid identification of almost all base pairs present in 26mer box C/D RNA. While the 2D/3D (H)N(HH)NH experiment detects WC G:C and non-WC U:U base pairs, the 2D/3D (H)N(HH)CH experiment identifies WC A:U and non-WC G:A base pairs. In the 2D/3D (H)N(HH)CH experiment (Figure 2G), after initial ¹H-¹⁵N CP transfer, ¹⁵N chemical shifts are evolved during t₁. After the second CP has transferred magnetization back to the protons, a ¹H,¹H RFDR of 0.48–0.96 ms spreads magnetization to all nearby protons within a distance of 3–4 Å. The third CP transfers magnetization to ¹³C for an optional evolution (in 3D experiment). Finally, a short read-out CP step transfers magnetization back to the protons for detection during t₂ (t₃ in 3D). In the 2D/3D (H)N(HH)NH experiment, the third CP step transfers magnetization to the directly attached nitrogens, where their chemical shifts evolve during t₂ (in the optional 3D experiment). The final short ¹⁵N-¹H CP read-out step transfers magnetization back to the protons for detection during t₂ (t₃ in 3D). Despite efficient identification of both WC and non-WC base pairs (Figure 2H), sequential imino–imino walk as performed in solution-state NMR was not feasible due to 1) low sensitivity and 2) strong signal overlap.

Presented case studies show that ssNMR allows rapid identification of the type and number of base pairs present in the RNA. However, sequence-specific assignment of base pairs and therefore RNA secondary structure determination based on amino and imino fingerprints alone are not possible and, hence, complete assignment of RNA resonances is necessary. The above will require their correlation with the well-resolved nucleobase carbons C6/C8 (C6-H6/C8-H8 groups) and then with the ribose C1′ (C1′-H1′ groups). Finally, nucleotides should be connected sequentially through ¹H–¹H and/or ¹³C–¹³C correlations (Figure 1).

In the following sections, we will discuss in detail experimental strategies for the resonance assignment of different RNA moieties by ssNMR.

Spin system–specific assignment of ribose resonances by conventional ¹³C-detected ssNMR at MAS frequencies < 20 kHz can be achieved using a multitude of correlation schemes provided a satisfactory spectral linewidth is obtained. In the pioneering study by the Görlach group, mostly intra-ribose and partially ribose–base correlations were obtained using symmetry-based adiabatic ZQ recoupling experiments (Riedel et al., 2004). However, due to the low chemical shift resolution arising from limitations in sample preparation, spin system–specific assignment was not possible despite only three different nucleotides being present in the (CUG)₉₇ RNA. The Drobny group (Huang et al., 2012) acquired ¹³C,¹³C PDSD (Szeverenyi et al., 1982) experiments on the selectively uracil-labeled TAR RNA. Also, there, very narrow carbon chemical shift dispersion together with poor resolution impeded any spin system–specific resonance assignments. In our study on 26mer box C/D RNA, we have exploited 2D ¹³C,¹³C PDSD (Figure 3A) to correlate intra-ribose resonances and even to obtain ribose–base correlations (Marchanka et al., 2013). While 100 ms PDSD mixing was sufficient to obtain a full set of intra-ribose correlations (Figure 3B), 500 ms mixing additionally provided not only intra-base correlations but also an almost complete set of ribose–base and several inter-nucleotide correlations (s. below). Despite good ¹³C linewidths of 0.5 ppm and usage of nucleotide-type selective labeling (Marchanka et al., 2013; Marchanka et al., 2018a), homonuclear ¹³C,¹³C spectra yielded the assignment of less than half of the nucleotides of 26mer box C/D RNA. The assignment process is hampered particularly by significant spectral crowding of C2′/C3′ carbons, and furthermore, the lack of a proton dimension does not help to lift the ambiguity.

A MAS regime of 20 kHz < ω_R < 60 kHz is less suitable for the ribose assignment in RNA labeled uniformly or selectively by nucleotide type. First, since ribose protons are not exchangeable, acquisition of NMR spectra in deuterated buffer does not bear any advantage. Coherence lifetimes of ribose protons are unfavorable at MAS frequencies < 60 kHz (Marchanka et al., 2018b), so that ¹³C-detected ssNMR must be applied. While 3.2 mm probes (maximum MAS frequency = 24 kHz) have optimal ¹³C sensitivity, the smaller rotor size in 2.5 and 1.7 mm probes attenuates ¹³C sensitivity due to a smaller sample volume. Second, pure SD-based experiments (PDSD and DARR) cannot work well at MAS > 20 kHz; therefore, advanced recoupling schemes must be utilized to correlate ribose resonances in this regime. These mixing schemes may include finite-pulse (fp) RFDR (Bennett et al., 1992; Nishiyama et al., 2014), R-symmetry–driven SD (Hou et al., 2011), mixed rotational and rotary resonance (MIRROR) (Scholz et al., 2008), phase-alternated recoupling irradiation (PARIS) (Weingarth et al., 2009), combined R2-symmetry–driven sequences (CORD) (Hou et al., 2013), and many others. A recent review by the Hou group provides a comprehensive description of different recoupling techniques at fast MAS (Ji et al., 2021). Many first-order homonuclear dipolar recoupling sequences (e.g., fpRFDR) suffer from dipolar truncation (Bayro et al., 2009), which may impair observation of long-range correlations, whereas second-order sequences (SD-based, RF-assisted SD, and PAR) are free from this effect (Ji et al., 2021). Regardless of the chosen type of homonuclear recoupling, all ¹³C,¹³C correlation experiments have a similar form. After an initial short CP transfer from ¹H to ¹³C, magnetization is evolved on the starting ¹³C during t₁. Subsequently, carbons are connected by one of the recoupling schemes. Finally, ¹³C chemical shifts are recorded during t₂ (Figure 3A).

Since N1 and N9 nitrogens in pyrimidines and purines, respectively, have very distinct chemical shifts, the ¹⁵N dimension can be added to improve assignment by separation of ribose resonances of different nucleotide types. Optimal separation can be achieved by acquisition of, for example, ¹³C,¹⁵N TEDOR (Jaroniec et al., 2002) or ¹³C,¹⁵N TEDOR-¹³C,¹³C PDSD (Riedel et al., 2005b; Daviso et al., 2013) experiments. In the TEDOR-PDSD experiment, the ¹³C magnetization (C1′) is prepared via a short ¹H,¹³C CP step. During a short ¹³C,¹⁵N TEDOR transfer, the magnetization is propagated to nearby nitrogens (N1 and N9 in pyrimidines and purines, respectively). After t₁ evolution on ¹⁵N, the magnetization is transferred back to ¹³C. A subsequent ¹³C,¹³C PDSD or DARR step spreads the magnetization into ¹³C spins of ribose (C2′-C5′), whose chemical shifts are recorded during t₂ (t₃ in 3D) (Figure 3C). An optional ¹³C evolution step before PDSD (DARR) yields the 3D TEDOR-PDSD experiment which improves resolution at the cost of sensitivity. In addition to improving ribose assignment (Figure 3D), TEDOR-PDSD provides ribose–base connections and improves the assignment of nucleobase carbons (vide infra).

¹H-detected ssNMR at MAS frequencies > 60 kHz and in particularly ≥ 100 kHz opens new avenues for structural characterization of biomolecules, significantly improving resolution and increasing the sensitivity per unit of the sample. Such probeheads operate with significantly smaller rotors that ultimately reduce the sample volume/number of spins packed into the ssNMR rotor and therefore attenuate sensitivity. Due to the optimized coil sensitivity, increased fill factor, and narrowed lines, overall sensitivity is not reduced as a cube of the rotor diameter (Schledorn et al., 2020); nevertheless, exclusively ¹H detection with the sensitivity increased by a factor of ( γ H / γ C ) 3 / 2 = 8 can compensate for the smaller rotor size. In our experience, a MAS regime where ω_R ≥ 100 kHz is optimal for RNA studies by ¹H-detected ssNMR. Coherence lifetimes of both H1′ and H2′-H5′ protons increase significantly at MAS frequencies above 60 kHz and reach 4.2 and 1.7 ms, respectively, at 109 kHz MAS (Marchanka et al., 2018b). In addition to many dipolar mixing schemes that can be used at MAS ≥ 100 kHz, scalar ¹³C couplings can be utilized to correlate all carbons in the ribose with each other in a manner typically utilized in solution-state NMR (Hu et al., 1998). J-coupling–based correlation spectroscopy becomes possible due to the long ¹³C T_1ρ relaxation times of 50 ms at ≥ 100 kHz MAS (Marchanka et al., 2018b). In our study, we have implemented a 4D HCCH-TOCSY experiment utilizing low-power 20 kHz WALTZ-16 (Shaka et al., 1983) mixing of a length of 25 ms, which allows us to fully correlate all CH groups in the ribose ring (Figures 4A,B). In the 4D HCCH-TOCSY, after t₁ evolution on the starting proton, magnetization is transferred by a short CP to directly attached ¹³C. After t₂ evolution on this starting carbon, magnetization is transferred to all carbons in the ribose by WALTZ-16 mixing. After evolution during t₃ on the final carbon, magnetization is transferred by a short CP to ¹H for detection during t₄ (Figure 4A). This experiment has been acquired utilizing non-uniform sampling (NUS) (Paramasivam et al., 2012; Sergeyev et al., 2017) and required 68 h of measurement time. To explore the feasibility of dipolar coupling–based transfer in the ribose at 100 kHz MAS, we have acquired ¹³C,¹³C fpRFDR spectra with 8 and 16 ms of mixing time. A full set of intra-ribose correlations has been obtained from both 3D (H)CCH spectra acquired with high sensitivity using uniform sampling within 40 h.

Despite good resolution in ¹H-detected 4D HCCH-TOCSY or 3D (H)CCH-fpRFDR spectra, three nucleotides in the helical regions were not assigned due to low dispersion of the C1′ and H1′ chemical shifts (Marchanka et al., 2018b). Similar to ¹³C-detected experiments, the N1/N9 dimension can be introduced to resolve spectral crowding at the price of sensitivity. In the 4D (H)NCCH-TOCSY experiment (Figure 4C), after a long ¹H-¹⁵N CP transfer, magnetization is evolved on N1 and N9 during t₁. Subsequently, magnetization is transferred by a specific ¹³C-¹⁵N CP step to the C1′ carbon in the ribose. From here on, the magnetization path follows one of the HCCH-TOCSY experiments and delivers a set of well-resolved N1/N9-C1′-CX′-HX′ correlations after 100 h of measurement time, albeit with low sensitivity (Figure 4D).

The next step of the resonance assignment protocol is the connection of riboses to the nucleobases, which can, in principle, be obtained in ssNMR by long-range carbon–carbon correlations utilizing previously discussed homonuclear recoupling schemes, for example, PDSD (Szeverenyi et al., 1982), DARR (Takegoshi et al., 2001), RFDR/fpRFDR (Bennett et al., 1992), PAR (De Paëpe et al., 2008), and others (Ji et al., 2021) (Figure 5A). Due to the two-bond distance between ribose C1′ and nucleobase C2/C6 and C4/C8 carbons, dipolar truncation (Bayro et al., 2009) can impair the efficiency of first-order recoupling schemes, so second-order dipolar recoupling (e.g., PDSD, DARR, PAR, and CORD) should be preferred depending on the MAS frequency. While in our studies, we have used the PDSD scheme at MAS < 20 kHz (Marchanka et al., 2013) and have employed fpRFDR at 100 kHz, various recoupling schemes (Ji et al., 2021) can be utilized for this purpose. Their efficacies toward RNA have to be evaluated in future studies.

As discussed previously in the ribose assignment section, spectral resolution in ¹³C (and ¹H) dimensions might not be enough for the spin system assignment using homonuclear recoupling schemes. As for ribose assignment experiments, acquisition of ¹⁵N-edited spectra resolves nucleotides by their N1/N9 chemical shifts and facilitates unambiguous ribose–base correlation.

In ¹³C-detected ssNMR spectroscopy, ribose and base resonances can be connected by the CNC-type experiment (Franks et al., 2005; Schuetz et al., 2010) (Figure 5B), which is similar to the HCNCH (Sklenar et al., 1993) experiment in liquids. After initial CP from H1′ to C1′, magnetization is evolved on C1′ during t₁. Next, a long SPECIFIC-CP step (Baldus et al., 1998) transfers magnetization to N1 and N9 in pyrimidines and purines, respectively, where it evolves during t₂. Finally, during a last SPECIFIC-CP step, magnetization is transferred to C2/C6 and C4/C8 in pyrimidines and purines, respectively, for the detection during t₃ (Marchanka et al., 2013). ¹³C detection at MAS < 20 kHz allows straightforward acquisition of ¹³C signals of both protonated and non-protonated carbons and provides well-separated C1′-N1-C2/C4 and C1′-N9-C4/C8 chemical shifts for pyrimidines and purines, respectively (Figure 5C). Due to the very distinct chemical shifts of N1 and N9 nitrogens, just two differently double nucleotide-type selectively labeled samples are sufficient to obtain such correlations for all nucleotides in RNA. This experiment can be acquired in the ribose-to-base or base-to-ribose direction. In our experience, ribose-to-base transfer was more efficient. Moreover, C1′ → N1/N9 → C2/C4 correlations were more intense than C1′ → N1/N9 → C6/C8 due to a shorter coherence lifetime caused by the increased dipolar relaxation in the latter case. Due to the two low-γ specific CP transfers and the intrinsically low sensitivity of ¹³C detection, a single good-quality 3D CNC spectrum required more than 120 h of experimental time on a 700 MHz Bruker spectrometer. However, this experiment provides assignment for C4 and C2 carbons in purines and pyrimidines, respectively, whose assigned chemical shifts are very rarely reported in the BMRB database (Ulrich et al., 2007) (s. below). The same experiment can be acquired at a MAS regime of 20 kHz < ω_R < 60 kHz; however, its sensitivity will be compromised due to the necessity of ¹³C-detection at decreased sample volume (see previous section).

Ribose–nucleobase correlations can be obtained by ¹H-detected experiments at MAS > 60 kHz on a single uniformly labeled ¹³C,¹⁵N RNA sample (Marchanka et al., 2018b). In the ¹H-detected (H)NCH experiment, after long ¹H-¹⁵N CP transfer and evolution on ¹⁵N during t₁, either ribose- or base-tuned band-selective CP transfers the magnetization either to C1′ or to C6/C8 for the evolution during t₂. A final short CP transfers magnetization to directly bound H1′ or H6/H8 protons for the detection during t₃ (Figure 5D). This experiment provides a set of either ribose-specific N1/N9–C1′–H1′ or base-specific N1–C6–H6/N9–C8–H8 correlations in pyrimidines/purines, respectively (Figures 5E,F). Due to the high sensitivity of ¹H detection, good quality ribose- and base-specific spectra have been obtained in 12 and 19 h, respectively (Marchanka et al., 2018b). While the (H)NCH experiment provides unambiguous correlation of ribose and base CH groups with their corresponding N1/N9 nitrogens, ribose–base correlations obtained by these two experiments might be ambiguous, since they share only N1/N9 chemical shifts. The narrow chemical shift range of N1/N9 nitrogens is exacerbated by intrinsically worse linewidths in ssNMR compared to solution-state. To lift the ambiguity, an additional ¹H evolution period can be added before the first ¹H-¹⁵N CP step, yielding the 4D HNCH experiment. In such a spectrum, chemical shifts in the indirect proton dimension (H6 or H8) and N1/N9 represent the nucleobase, while N1/N9, C1′ and H1′ chemical shifts correspond to the ribose.

After resonances of riboses are assigned and correlated with the nucleobase C6-H6/C8-H8 groups, assignment of resonances in the nucleobase is performed. This is not a trivial task, particularly in ¹H-detected NMR spectroscopy, due to the low density of protons in RNA nucleobases. The majority of reported assigned carbons and nitrogen RNA chemical shifts in the BMRB database belong to either protonated nuclei or nuclei directly attached to the protonated carbons. The fraction of assigned chemical shifts for C4 and C5 carbons in purines is very low and estimates to only 1.8% (72/3,921) and 2.0% (79/3,921), respectively, of all assigned base carbon chemical shifts in the BMRB database. Sensitive dipolar coupling–based transfer in ssNMR provides an unprecedented opportunity for the assignment of these otherwise not-easy-to-access carbons.

C2/C6 and C4/C8 carbons in pyrimidines and purines, respectively, are assigned by the 3D CNC experiment acquired at MAS < 20 kHz, as described in the section above. H6-C6 and H8-C8 groups are assigned by the (H)NCH experiment at ultrafast MAS.

Strategies for the assignment of the remaining carbons and nitrogens in the nucleobase depend strongly on if either ¹³C- or ¹H-detection is utilized.

At MAS frequencies < 20 kHz, using ¹³C-detection, ¹³C,¹³C recoupling schemes discussed above for the ribose assignment can be used to obtain carbon assignment in the nucleobase. 500 ms PDSD recoupling (Figure 5A) or 8–16 ms PAR (Figure 6A) were found to be sufficient to obtain many, although overlapping, correlations in the nucleobases of RNA selectively labeled by nucleotide type. As for ribose assignment, inclusion of the nitrogen dimension can reduce spectral crowding. The previously described ¹³C,¹⁵N TEDOR-¹³C,¹³C PDSD experiment (Riedel et al., 2005b; Daviso et al., 2013) can provide assignment of carbon resonances in nucleobases by connecting their chemical shifts to the chemical shifts of assigned carbons in the nucleobase (C6/C8) and the ribose (C1′) (Figure 6B).

The NCC experiment (Pauli et al., 2001; Igumenova et al., 2004; Franks et al., 2005) is employed to connect amino and imino nitrogens with nucleobase and (partially) ribose carbons (Marchanka et al., 2013). After an initial short CP from ¹H to ¹⁵N, ¹⁵N chemical shifts evolve during t₁ and yield the frequencies of N6 (A), N4 (C), N1, N2 (G), and N3 (U). Following this, the magnetization is transferred by SPECIFIC-CP to the nearby carbons, N6→C6 (A), N4→C4 (C), N1→C2/C6 and N2→C2 (G), and N3→C2/C4 (U). Here, magnetization can be optionally evolved during t₂ or spread directly by PDSD to all base carbons and C1′ for the detection during t₂ or t₃ (in the 3D version) (Figures 6C,D). Needless to say, diverse ¹³C,¹³C recoupling schemes can be utilized instead of PDSD. The NCC experiment can, in theory, provide unambiguous assignment for all protonated nitrogens and all carbons in the nucleobase if protonated nitrogens can be uniquely correlated with well-resolved carbons in the nucleobase (C2/C6, C4/C8) or C1′. However, the poor chemical shift dispersion of ¹⁵N amino and imino resonances usually renders assignment inconclusive, especially in helical regions.

Assignment of the remaining non-protonated nitrogens in the base is important for characterization of non-WC base pairs, especially G:A and U:U. Assignment of N7 nitrogen in adenosines and guanosines can be readily obtained from the ¹³C,¹⁵N TEDOR (Jaroniec et al., 2002) experiment that provides (H8)C8-N7 correlations. Assignment of N1 and N3 nitrogens in adenosines is obtained from the same-type TEDOR experiment by acquisition of (H2)C2-N1/N3 correlations (Figure 6E). To obtain site-specific assignment of these N1/N3 nitrogens, C2 carbons have to be correlated with C6/C8 carbons by any of the ¹³C,¹³C correlation schemes discussed above.

The assignment of non-protonated nitrogens (N3) in both cytidines and guanosines is the most challenging task, due to the absence of any protons in their close vicinity. It could, in principle, be obtained from a modified NCC experiment, where a long initial ¹H-¹⁵N CP transfers the magnetization from the distant (r = 2.4 Å) amino protons H41/H42 (C) or H21/H22 (G) to the N3 nitrogen. ¹⁵N chemical shifts evolve during t₁, and then magnetization is transferred to the C2 and C4 carbons, where it is recorded during t₂ (Figure 6F). It is uncertain if 1) T_1ρ during the H41/H42→N3 and H21/H22→N3 transfer is long enough to allow for the efficient long-range magnetization transfer and 2) resolution of ¹⁵N and ¹³C resonances are good enough to allow for the spin system–specific assignment. However, N3 of cytidine can be easily assigned indirectly through base-paired guanosine in the WC G:C base pair. Here, the cross-strand N–N distance of ∼2.0 Å allows the straightforward connection of N3 in cytidine with N1 in guanosine using, for example, ¹⁵N,¹⁵N RFDR or ¹⁵N,¹⁵N PAR correlations (Figure 2B).

While ¹³C-detected ssNMR allows almost complete assignment of carbons and nitrogens in the nucleobases using only two different types of experiments, namely, CNC and NCC, ¹H-detected ssNMR on RNA nucleobases is more challenging. Despite the significant advantages of 1) additional spectral dimensions and 2) higher sensitivity, ¹H detection obviously requires that the magnetization transfer pathway ends on a proton. As is known from solution-state NMR, many different types of experiments are necessary to assign RNA nucleobases (Fürtig et al., 2003).

Very recently, we have addressed this challenge and have published the set of ssNMR experiments that allow assignment of nucleobase resonances (Aguion et al., 2021).

In the first step, all nucleobase carbons are correlated with the previously assigned C6-H6 (pyrimidines) or C8-H8 (purines) groups using ¹³C,¹³C fpRFDR recoupling. This step is accomplished by the 3D (H)CCH experiment, which starts with a long ¹H-¹³C CP step to gain sufficient ¹³C magnetization on both protonated and non-protonated carbons. After t₁ evolution on ¹³C, a phase-cycled selective inversion pulse cancels the signals of the ribose ring. Following this, 8 ms-long ¹³C,¹³C fpRFDR recoupling transfers magnetization to nearby carbons, whose chemical shifts are recorded during t₂. Finally, the ¹³C magnetization is transferred by a short read-out CP to directly attached protons for the detection during t₃ (Figure 7A). This experiment correlates all nucleobase carbons with the protonated C5-H5 and C6-H6 groups in pyrimidines, C8-H8 groups in purines, and C2-H2 groups in adenosines (Figure 7B).

In the second step, ¹⁵N-¹H imino and amino resonances are correlated with the assigned carbons and C-H groups using two different experiments. The 3D (H)CNH experiment (Figure 7C) starts with a long ¹H-¹³C CP step to transfer proton magnetization to non-protonated carbons. After t₁ evolution on ¹³C, magnetization is transferred to directly attached amino or imino nitrogens via a ¹³C,¹⁵N CP step. After t₂ evolution on nitrogens, a phase-cycled selective inversion pulse selects either amino or imino magnetization that is subsequently transferred by a short read-out CP to ¹H for the detection during t₃. This experiment yields C6-H6-H61/H62 correlations for A, C4-H4-H41/H42 for C, C2/C6-N1-H1 and C2-N2-H21/H22 for G, and C2/C4-N3-H3 for U (Figure 7D). The correlation of C2/C4 resonances in uridines and C2/C6 resonances in guanosines with imino groups is sufficient for site-specific assignment of imino resonances if chemical shift dispersion exists in at least one of the correlated carbons. Amino groups, on the other hand, have only one adjacent carbon (C6 in A, C4 in C, and C2 in G) with a very narrow chemical shift range, which often prevents unambiguous assignment of amino resonances. This is aggravated by the poor resolution of amino resonances compared to imino resonances. To resolve the ambiguity in these cases, we have combined the two experiments reported above to develop 3D 1) (H)N(C)CH and 2) H(NC)CH experiments. In experiment 1) (Figure 7E), after an initial short ¹H-¹⁵N CP step, the chemical shifts of amino and imino nitrogens are recorded during t₁. Subsequently, magnetization is transferred to directly attached carbons by a ¹⁵N, ¹³C CP step. The phase-cycled selective inversion pulse selects nucleobase carbon magnetization at ∼ 160 ppm, which is then spread to all nearby carbons by a 14 ms RDFR mixing step. Following this, ¹³C magnetization is evolved during t₂ and finally transferred from protonated carbons by a short read-out CP to directly attached protons for the detection during t₃ (Figure 7E). Despite the low sensitivity of this experiment due to the modest efficiency of both low-γ ¹³C,¹⁵N CP and ¹³C,¹³C RFDR transfers, it provides important information by correlating imino and amino nitrogens directly with the well-resolved C6-H6 and C5-H5 groups in pyrimidines, C8-H8 groups in pyrimidines, and C2-H2 groups in adenosines (Figure 7F). In experiment 2), chemical shifts of amino and imino protons are evolved instead of nitrogens and should provide better spin system separation due to better resolution in the proton dimension. Unfortunately, its sensitivity was poor due to the short coherence lifetimes of imino (∼4 ms) and especially amino (∼1 ms) protons at 100 kHz MAS. Relaxation behavior of RNA amino/imino groups could be improved by 1) increasing the MAS frequency and 2) performing experiments in partial deuterated buffer (∼25–50%), whereas the overall sensitivity of the experiment can be improved by utilizing optimized ¹³C,¹³C recoupling schemes.

While ¹³C,¹³C recoupling-based experiments were successfully utilized for the assignment of most nucleobase resonances, some nuclei can be more efficiently assigned through ¹H,¹H recoupling schemes. Thus, amino groups of guanosines were assigned through correlation with well-resolved imino groups through the 3D (H)N(HH)NH experiment described above (Figures 7G,H). Similarly, amino groups of cytidines were assigned through the previously described 3D (H)N(HH)CH experiment (Figure 7I). This experiment provides N4–C5–H5 correlations for all four cytidines in the structured region of 26mer box C/D RNA (Figure 7J) and, moreover, provides assignment of WC A:U (Figures 2G,H) and non-WC G:A base pairs.

In the third step, assignment of non-protonated nitrogen resonances is performed. In ¹H-detected ssNMR, it can be obtained from a modified version of the 3D (H)NCH experiment described above (Figure 7K). For effective excitation of N1, N3, and N7 resonances, the ¹⁵N carrier frequency should be shifted to a higher ppm (e.g., ∼190 ppm). Moreover, a selective ¹³C refocusing pulse applied after t₂ eliminates unwanted signals of the ribose ring. The experiment correlates N7 resonances with protonated C8-H8 groups in purines and N1 and N3 resonances with protonated C2-H2 groups in adenosines. N3 resonances in guanosines, however, do not have any adjacent CH group and can therefore not be assigned in this experiment. The same applies to N3 resonances in cytidines. As in ¹³C-detected ssNMR, assignment of these resonances remains challenging, but N3 resonances of cytidines can likewise be obtained indirectly through base-paired guanosines. Here, the close distance of cytidine N3 and H1 resonances in the base-paired guanosine (r = 2.0 Å) can be exploited in a simple 2D ¹H-¹⁵N-HSQC experiment with long-range ¹H-¹⁵N/¹⁵N-¹H CP transfer times of 8 ms (Aguion et al., 2021).

At intermediate MAS rates, nucleobase assignment can be performed either by acquisition of ¹³C-detected NC-type experiments (Marchanka et al., 2013; Zhao et al., 2019), taking into account attenuated ¹³C sensitivity, or using ¹H detection of amino/imino protons employing 3D (H)CNH and 3D (H)N(HH)NH experiments. ¹H-detection of nucleobase H6/H8/H2 protons would be unfavorable due to their short coherence lifetimes at MAS < 60 kHz (Marchanka et al., 2018b).

After spin system–specific assignment is achieved, nucleotides are sequentially connected with each other to obtain site-specific assignments. In solution-state NMR, this task is accomplished by measurement of either through-space internucleotide H6/H8(i)-H2′,H1′(i−1) NOEs or through detection of overlapped C_ribose–P correlations (Fürtig et al., 2003). A similar approach has been utilized in ssNMR; however, due to the preservation of direct dipolar couplings, accessible distances are significantly longer and are not limited to ¹H,¹H restraints. The first ever sequential RNA correlations by ssNMR were obtained by the Görlach group. They have utilized a CHHC-type experiment (Lange et al., 2002) and have measured sequential H2′,H3′(i-1)-H6/H8(i) contacts in the (CUG)₉₇ repeat (Figures 8A,B) (Riedel et al., 2006). In our ¹³C- and ¹⁵N-detected study on 26mer box C/D RNA, we have utilized ¹³C,¹³C and ¹³C,³¹P correlations to obtain sequential assignments (Marchanka et al., 2015). Due to severe resonance overlap, eight different double nucleotide-type selective labeled samples were prepared and an ¹⁵N-editing TEDOR step was coupled to the long-range ¹³C,¹³C PDSD recoupling (Figure 8C). Long-range correlations up to 9 Å were obtained at 700 ms of PDSD mixing time. While ¹³C,³¹P correlations acquired by the ¹³C,³¹P TEDOR experiment (Figure 8D) provided important sequential contacts in the non-canonical region of RNA (kink-turn) and corroborated sequential assignments obtained by ¹³C,¹³C correlation experiments, their value for correlating nucleotides in the helical regions was limited due to strong resonance overlap. Additional ¹H,¹H sequential contacts have been obtained from CHHC and NHHC experiments acquired at 200 us of mixing time (Figure 8E). ¹³C,³¹P and CHHC/NHHC correlation experiments were essential to distinguish sequential contacts from the long-distance inter-strand correlations.

While ¹H-detected ssNMR allows spin system–specific assignment of both riboses and nucleobases from single, uniformly ¹³C,¹⁵N labeled RNA samples (Marchanka et al., 2018b; Aguion et al., 2021), no sequence-specific assignment of RNA by ¹H-detected ssNMR has been reported to date. This objective can be achieved by utilizing, for example, ¹H,¹H recoupling coupled with ¹³C-editing either on one (3D) or both sides (4D) to overcome resonance overlap. Furthermore, ¹³C-³¹P and ¹H-³¹P correlation experiments may be beneficial at fast MAS due to effective averaging of CSA. Our group is currently working in this direction, and we hope to present procedures for the sequential assignment and measurement of structural restraints based on ¹H-detected ssNMR in the near future.