Cryo-EM structure of a 40 kDa SAM-IV riboswitch RNA at 3.7 Å resolution

Specimens below 50 kDa have generally been considered too small to be analyzed by single-particle cryo-electron microscopy (cryo-EM). The high flexibility of pure RNAs makes it difficult to obtain high-resolution structures by cryo-EM. In bacteria, riboswitches regulate sulfur metabolism through binding to the S-adenosylmethionine (SAM) ligand and offer compelling targets for new antibiotics. SAM-I, SAM-I/IV, and SAM-IV are the three most commonly found SAM riboswitches, but the structure of SAM-IV is still unknown. Here, we report the structures of apo and SAM-bound SAM-IV riboswitches (119-nt, ~40 kDa) to 3.7 Å and 4.1 Å resolution, respectively, using cryo-EM. The structures illustrate homologies in the ligand-binding core but distinct peripheral tertiary contacts in SAM-IV compared to SAM-I and SAM-I/IV. Our results demonstrate the feasibility of resolving small RNAs with enough detail to enable detection of their ligand-binding pockets and suggest that cryo-EM could play a role in structure-assisted drug design for RNA.

The work by Zhang et al. presents a newly determined structure of a riboswitch regulating sulfur metabolism in bacteria, with potential antibiotic target applications. Furthermore, the authors demonstrate cutting-edge technical performance in cryo-electron microscopy that breaks the current molecular weight paradigm and will encourage other researchers to try similar applications. The results are of high-quality and are clearly presented. I am confident that the paper will be of great interest to researchers from multiple fields and therefore recommend its publication in Nature Communications with a few minor revisions.
Minor points: 1. Line 99: "After a turn containing six nucleotides, …" The figure shows the turn containing four nucleotides. Please correct the text. 2. Lines 107-108 "the preservation of the overall fold is visually apparent in the maps and the automated models (

Reviewed by Radostin Danev
Reviewer #2 (Remarks to the Author): In this manuscript, Zhang et al. used single particle cryo-EM to reveal the structures of a 40-kDa RNAonly SAM-IV riboswitch in different ligand binding states. In these resolved structures, the resolution is high enough to identify/trace the sugar phosphate backbone and some base pairs of RNA. The authors combined various methods to build and validate the atomic models from the cryo-EM results. The information was integrated to further identify the SAM-binding pocket. This work provides important insight to both the mechanism of riboswitch ligand recognition and demonstrates the power of single particle cryo-EM in determining RNA structures in small size. Overall, the manuscript is well-written and presented clearly.
I have a few remarks for the authors to address: Major remark is in the section that describes the identification of ligand binding pocket (from line 121). Without pre-knowledge, the direct ways to locate the binding pocket are: 1) a high enough resolution map shows separated density(s) that define a clear feature of the ligand; 2) compare the apo-/ligand-bound-state structures in the same condition to find significant density(s) corresponding to the ligand. I agree with the authors' statement that the difference in resolution and slight differences in conformations might influence the analysis. However, the difference between apo and ligand-bound map is the "real" data from the cryo-EM reconstruction, while the model-derived map is a deduced one. The authors stated that (line 129) "A comparison between the apo and SAM-bound maps also reveals the same ligand binding site", but they didn't show the difference map between the apo and SAM-bound maps. Such a difference map should be provided in the supplementary information.
I'm curious if the SAM-I crystal structure is unknown, how much confidence the SAM could be locate with the 4.1 angstrom map. In Fig. 3c, at the threshold of 1.5, there are two relatively large densities but both smaller than the size of SAM. Besides, the separated ligand density in Fig. 3c and 3d seems to have different shapes, can the author make an explanation for this? I think the statements "unbiased identification of the SAM binding pocket" and "Notably, the location of the ~ 0.4-kDa ligand could also be determined in our 4.1-Å holo cryo-EM map" should be tuned down.
Minor remarks: 1) In Ext. Fig. 6c, the Q score distribution along NT# seems to have a fluctuation in every ~15 NT. Is there an explanation about it? I'm also curious about whether the Q score distribution is related to the local resolution distribution.
2) As this work is a record-making cryo-EM study of small RNA molecules, the detailed validation results of the maps (i.e. Euler angle distribution, un-masked FSC) should be provided in the supplemental materials for others to understand better of the dataset.
3) The claimed defocus ranges for data collection in Methods section (line 177-178) are different from the ranges provide in Table S1.

Hong-Wei Wang
Reviewer #3 (Remarks to the Author): Single-particle cryo-electron microscopy (cryo-EM) has solved numerous near-atomic or atomic resolution structures of biological molecules recently. Most of the structures are focused on protein or protein-RNA complex. Using cryo-EM to solve the structure of solo RNA molecules, especially small RNA molecules with molecular weight less than 50 KDa will promote the development of the entire RNA research field. The manuscript by Zhang, K. et al., titled "Cryo-EM Structure of a 40-kDa SAM-IV Riboswitch RNA at 3.7 Å Resolution" reports the structures of apo and SAM-bound SAM-IV riboswitches (119 nt, approximately 40 KDa) to 3.7 Å and 4.1 Å resolution with cryo-EM method respectively. Besides, the comparison was made between the structures of SAM-I, SAM-I/IV and SAM-IV riboswitches. SAM-IV adopted similar ligand-binding core but different peripheral tertiary contacts with SAM-I and SAM-I/IV riboswitch. Based on these results, the authors proposed the feasibility of solving the structure of small RNA molecules with cryo-EM, which may facilitate the structure-based drug design for RNA. How about the RNA molecules that have no sequence similarity with other structure-solved RNA molecules? Is it possible to apply this method to a novel RNA molecule and get the same resolution result? The second question is that it's obvious that the current resolution in the manuscript is not high enough to provide the exact structural detail for the structure-based drug design of RNA, please explain how does it assist the drug design of RNA in the current stage. The work by Zhang et al. presents a newly determined structure of a riboswitch regulating sulfur metabolism in bacteria, with potential antibiotic target applications. Furthermore, the authors demonstrate cutting-edge technical performance in cryo-electron microscopy that breaks the current molecular weight paradigm and will encourage other researchers to try similar applications. The results are of high-quality and are clearly presented. I am confident that the paper will be of great interest to researchers from multiple fields and therefore recommend its publication in Nature Communications with a few minor revisions.
Minor points: 1. Line 99: "After a turn containing six nucleotides, …" The figure shows the turn containing four nucleotides. Please correct the text. Response: Thank you, it is 4 nucleotides. This is changed on page 5.
2. Lines 107-108 "the preservation of the overall fold is visually apparent in the maps and the automated models ( In this manuscript, Zhang et al. used single particle cryo-EM to reveal the structures of a 40-kDa RNA-only SAM-IV riboswitch in different ligand binding states. In these resolved structures, the resolution is high enough to identify/trace the sugar phosphate backbone and some base pairs of RNA. The authors combined various methods to build and validate the atomic models from the cryo-EM results. The information was integrated to further identify the SAM-binding pocket. This work provides important insight to both the mechanism of riboswitch ligand recognition and demonstrates the power of single particle cryo-EM in determining RNA structures in small size. Overall, the manuscript is well-written and presented clearly.
I have a few remarks for the authors to address: Major remark is in the section that describes the identification of ligand binding pocket (from line 121). Without pre-knowledge, the direct ways to locate the binding pocket are: 1) a high enough resolution map shows separated density(s) that define a clear feature of the ligand; 2) compare the apo-/ligand-bound-state structures in the same condition to find significant density(s) corresponding to the ligand. I agree with the authors' statement that the difference in resolution and slight differences in conformations might influence the analysis. However, the difference between apo and ligand-bound map is the "real" data from the cryo-EM reconstruction, while the model-derived map is a deduced one. The authors stated that (line 129) "A comparison between the apo and SAM-bound maps also reveals the same ligand binding site", but they didn't show the difference map between the apo and SAM-bound maps. Such a difference map should be provided in the supplementary information. Response: We agree with this reviewer's suggestion. A second difference map, between the apo and SAM-bound cryo-EM maps, is now added in the Supplementary Figure 9e, f. A superimposed map of these two states is also added in Supplementary Fig 9c, showing similar extra density corresponding to the ligand. It should be noted that this difference map also shows other peaks due to the difference in resolution, small changes in the RNA itself, and noise, between the two maps, which are not present in the map minus model difference map in Fig. 3.
The SAM ligand was found to be in the central region of the difference map and coincided with the location of the ligand derived from the SAM-I crystal structure (Fig. 3c). We also further verified the ligand by segmenting the cryo-EM map using Segger (Pintilie, G. et al. J. Struct. Biol. 170, 427-438 (2010)), producing a separable segment similar to the density observed in the difference map at the same location (Fig. 3d).
I'm curious if the SAM-I crystal structure is unknown, how much confidence the SAM could be locate with the 4.1 angstrom map. In Fig. 3c, at the threshold of 1.5, there are two relatively large densities but both smaller than the size of SAM. Besides, the separated ligand density in Fig. 3c and 3d seems to have different shapes, can the author make an explanation for this? I think the statements "unbiased identification of the SAM binding pocket" and "Notably, the location of the ~ 0.4-kDa ligand could also be determined in our 4.1-Å holo cryo-EM map" should be tuned down. Response: We thank the reviewer for this question and thus have undertaken the additional analysis. We found the ligand locations by the difference maps without using the ligand crystal structure (Fig. 3 and Supplement Figure 9). In order to validate whether this central density is indeed attributable to the SAM ligand, we used the ligand crystal structure. To do so, we converted the SAM ligand coordinates to a 4-Å density map, and this density was scanned over the entire difference map using Chimera and Situs (Wriggers et al J Struct Biol 125(2-3):185-195). The highest cross-correlation was found at this putative site. Furthermore, the Q-score (Pintilie, G.. doi:10.1101/722991) of the ligand as placed, 0.35, is close to the expected Q-score in a 4-Å map, 0.40. (In contrast, the Q-score of the same relative position for the ligand in the apo map is -0.24). Our workflow suggests that we can reveal the location of the ligand without prior ligand crystal information. Of course, we cannot characterize the atomic details of the ligand and its interactions with the RNA because of the limited resolution.
We can segment a similar density from the actual cryo-EM SAM-bound map, as shown in Figure  3d. The size and shape of the density corresponding to the ligand depends on the choice of threshold (A new panel with low threshold has been added in Fig. 3c) and the resolution of the map. We think the small difference in shape between the segmented density (Fig. 3d) and the difference map (Fig. 3c) is due to the segmentation accuracy at this resolution. We have revised the manuscript and described appropriately our findings and interpretations concerning the localization of the ligand on pages 7-8.
Minor remarks: 1) In Ext. Fig. 6c, the Q score distribution along NT# seems to have a fluctuation in every ~15 NT. Is there an explanation about it? I'm also curious about whether the Q score distribution is related to the local resolution distribution. Response: It is true that the Q-core can be related to the local resolution variation. We have discussed the relationship of Q-score in reference to the map resolution in our manuscript submitted to BioRxiv (Pintilie, G.. doi:10.1101/722991). This reviewer raised a very interesting observation that the Q-scores fluctuate every ~15 NT. Looking at the model more closely, the low points seem to coincide with turns, loops or kinks where the model may be more stressed or dynamic, resulting in lower resolvability. Where the model, on the other hand, indicates more favorable and stable base pairs, the Q-score is higher. We thank the reviewer for the question and we mention this phenomenon on page 5 in the revised text.
2) As this work is a record-making cryo-EM study of small RNA molecules, the detailed validation results of the maps (i.e. Euler angle distribution, un-masked FSC) should be provided in the supplemental materials for others to understand better of the dataset. Response: We have included the Euler angle distribution and FSC curves derived from the cryoSPARC refinement in the Supplementary Figure 2   3) The claimed defocus ranges for data collection in Methods section (line 177-178) are different from the ranges provide in Table S1. Response: We apologize for this mistake. We have corrected the defocus range in Table S1.

Hong-Wei Wang
Reviewer #3 (Remarks to the Author): Single-particle cryo-electron microscopy (cryo-EM) has solved numerous near-atomic or atomic resolution structures of biological molecules recently. Most of the structures are focused on protein or protein-RNA complex. Using cryo-EM to solve the structure of solo RNA molecules, especially small RNA molecules with molecular weight less than 50 KDa will promote the development of the entire RNA research field. The manuscript by Zhang, K. et al., titled "Cryo-EM Structure of a 40-kDa SAM-IV Riboswitch RNA at 3.7 Å Resolution" reports the structures of apo and SAM-bound SAM-IV riboswitches (119 nt, approximately 40 KDa) to 3.7 Å and 4.1 Å resolution with cryo-EM method respectively. Besides, the comparison was made between the structures of SAM-I, SAM-I/IV and SAM-IV riboswitches. SAM-IV adopted similar ligandbinding core but different peripheral tertiary contacts with SAM-I and SAM-I/IV riboswitch. Based on these results, the authors proposed the feasibility of solving the structure of small RNA molecules with cryo-EM, which may facilitate the structure-based drug design for RNA. riboswitches were solved by X-ray crystallography method already. The above information would definitely assist the structural modeling of SAM-IV riboswitch in this manuscript. How about the RNA molecules that have no sequence similarity with other structure-solved RNA molecules? Is it possible to apply this method to a novel RNA molecule and get the same resolution result? Response: Our map was computed without using prior information from these structures. We judged the quality of the maps using the best practices in cryo-EM in terms of FSC from two independent half-maps and the recently-introduced Q-score. We agree that the existence of crystal structures of other members of SAM riboswitches assisted us to build the models with high level of confidence. Our models are validated by the PDB validation test and are within the acceptable range of model accuracy in terms of stereochemistry.
This reviewer raised a very good question: if there is a completely unknown structure, can we derive an RNA model with confidence? We are pleased to report that it is possible to use cryo-EM maps at 3-10 Å resolution to assist in building RNA atomic models with an estimate of their model accuracy. We have carried out some other studies of RNA molecules ranging from 62-388 nucleotides to demonstrate this integrative approach. These results are documented in two separate manuscripts (doi:https://doi.org/10.1101/717801), which are under review.
The second question is that it's obvious that the current resolution in the manuscript is not high enough to provide the exact structural detail for the structure-based drug design of RNA, please explain how does it assist the drug design of RNA in the current stage. Response: There are multiple steps in the drug discovery process. Once a drug candidate is identified by its medicinal assay, one would like to know where the drug binds and what are the chemical details of the drug-RNA interface. Our studies demonstrate that it is possible to detect a drug-sized molecule interacting with the RNA, though we cannot show the atomic structure detail because of the limited resolution. Our B-factor plot in Supplementary Figure 2d indicates that our resolution is limited by the number of particle images and their inherently low contrast. This can be improved with more particle images and/or possibly with improvement in the instrument (e.g. better DQE of next generation of the detector and better phase plate), which suggests the promise of our approach and the future direction of this project.