x3DNA-DSSR: The Next Generation of 3DNA with Unmatched Features for RNA Structural Bioinformatics
(integrated into a 2MB self-contained ANSI C program)

The skmatic.x3dna.org website (see screenshot below) aims to showcase DSSR-enabled cartoon-block schematics of nucleic acid structures using PyMOL. It presents a simple interface to browse pre-calculated PDB entries with a set of default settings: long rectangular blocks for Watson-Crick base-pairs, square blocks for G-tetrads in G-quadruplexes, with minor-groove edges in black. Users can also specify an URL to a PDB- or mmCIF-formatted file or upload such an atomic coordinates file directly, and set several common options to customerize to the rendered image.

Moreover, a web API to DSSR-PyMOL schematics is available to allow for its easy integration into third-party tools.

Input a PDB id

Pre-calculated cartoon-block images together with summary information are available for PDB entries of nucleic-acid-containing structures. Note that gigantic structures like ribosomes that are only represented in mmCIF format are excluded from the resource. The base block images are most effective for small to medium-sized structures.

Here are a few examples:

• 1ehz, the crystal structure of yeast phenylalanine tRNA at 1.93-Å resolution
• 2lx1, the major conformation of the internal loop 5’GAGU/3’UGAG
• 2grb”, the crystal structure of an RNA quadruplex containing inosine-tetrad
• 4da3, the crystal structure of an intramolecular human telomeric DNA G-quadruplex 21-mer bound by the naphthalene diimide compound MM41
• 1oct, crystal structure of the Oct-1 POU domain bound to an octamer site
• 2hoj, the crystal structure of an E. coli thi-box riboswitch bound to thiamine pyrophosphate, manganese ions

Each entry is shown with images in six orthogonal perspectives: front, back, right, left, top, bottom. The ‘front’ image (upper-left in the panel) is oriented into the most-extended view with the DSSR --blocview option. The corresponding PyMOL session file and PDB coordinate file are available for download. One can also visualize the structure interactively via 3Dmol.js.

Sample PDB entries

Users can browse random samples of pre-calculated PDB entries. The number should be between 3 and 99, with a default of 12 entries (see below for an example). Simply click the ‘Submit’ button or the “Random samples (3 to 99)”: http://skmatic.x3dna.org/pdb_entry link to see results of randomly picked 12 PDB entries each time.

Specify a coordinate file

The atomic coordinate file must be in PDB or mmCIF format, and can be optionally gzipped (.gz). One can either specify an URL to or select a coordinate file. Several common options are available to allow for user customizations.

Web API help message

Usage with 'http' (HTTPie):
    http -f http://skmatic.x3dna.org/api [options] url=|model@
    http http://skmatic.x3dna.org/api/pdb/pdb_id  -- for a pre-calculated PDB entry
    http http://skmatic.x3dna.org/api/help        -- display this help message
Options:
    block_file=styles-in-free-text-format [e.g., block_file=wc-minor]
    block_color=nt-selection-and-color    [e.g., block_color='A:pink']
    block_depth=thickness-of-base-block   [e.g., block_depth=1.2]
    r3d_file=true-or-FALSE(default)       [e.g., r3d_file=true]
    raw_xyz=true-or-FALSE(default)        [e.g., raw_xyz=true]
Required parameter
    url=URL-to-coordinate-file [e.g., url=https://files.rcsb.org/download/1ehz.pdb.gz]
    model@coordinate-file      [e.g., model@1ehz.cif]
    # Only one must be specified. 'url' precedes 'model' when both are specified.
    # The coordinate file must be in PDB or PDBx/mmCIF format, optionally gzipped.
Examples
    http -f http://skmatic.x3dna.org/api block_file='wc-minor' model@1ehz.cif r3d_file=t
    http -f http://skmatic.x3dna.org/api url=https://files.rcsb.org/download/1ehz.pdb.gz -d -o 1ehz.png
    http http://skmatic.x3dna.org/api/pdb/1ehz -d -o 1ehz.png
    # with 'curl'
    curl http://skmatic.x3dna.org/api -F 'model=@1msy.pdb' -F 'block_file=wc-minor' -F 'r3d_file=1'
    curl http://skmatic.x3dna.org/api -F 'url=https://files.rcsb.org/download/1ehz.pdb.gz' -o 1ehz.png
    curl http://skmatic.x3dna.org/api/pdb/1ehz -o 1ehz.png

Sample images

While reading DNAproDB: an expanded database and web-based tool for structural analysis of DNA–protein complexes, I noticed SNAP and DSSR being mentioned. The detailed citations are as below:

Information about individual nucleotide–residue interactions is also provided, such as hydrogen bonding, interaction geometry (based on SNAP (10)), buried solvent accessible surface areas and identification of the interacting residue and nucleotide moieties …

DNAproDB assigns a geometry for every nucleotide–residue interaction identified using SNAP, a component of the 3DNA program suite (10). The residues for which probabilities are shown are those with planar side chains so that a stacking conformation can be defined.

Base pairing and base stacking between nucleotides are identified using the program DSSR (20).

SNAP and DSSR are two (relatively) new programs in the 3DNA software suite. As the author, I am always glad to see them being cited explicitly in literature. The fact that SNAP and DSSR are cited together by DNAproDB, however, is especially significant. I am aware of the initial version of DNAproDB, but I definitely like the updated one better. This is what I recently wrote in response to a question on the 3DNA Forum:

Regarding DNA-protein interactions in general, you may want to have a look of DNAproDB from the Remo Rohs laboratory. A new paper has just been published in NAR, ‘DNAproDB: an expanded database and web-based tool for structural analysis of DNA–protein complexes’.

I’ve no doubt that SNAP and DSSR would be widely used in applications related to DNA/RNA structural bioinformatics. DSSR (to a lesser extent, SNAP) represents my view of what a scientific software tool should be.

Recently I read the article Topology-based classification of tetrads and quadruplex structures in Bioinformatics by Popenda et al. In this work, the authors proposed an ONZ classification scheme of G-tetrads in intramolecular G-quadruplexes (G4) as shown below (Fig. 2 in the publication):

I am glad to find that DSSR has been used as a component in their computational tool ElTetrado to automatically identify and classify tetrads and quadruplexes.

Structures from both sets were analysed using self-implemented programs along with DSSR software from the 3DNA suite (Lu et al. (2015)). From DSSR, we acquired the information about base pairs and stacking.

I like the ONZ classification scheme: it is simple in concept yet provides a new perspective for the topologies of G-tetrads in intramolecular G4 structures. So I implemented the idea in DSSR v1.9.8-2019oct16, with this feature available via the --g4-onz option. Note that ElTetrado, according to the authors, is applicable to ONZ classifications of general types of tetrads and quadruplexes. The DSSR implementation of ONZ classifications, on the other hand, is strictly limited to G-tetrads in intramolecular G4 structures.

The DSSR ONZ classification results match the ones reported in Figs. 1, 5, and 6 of the Popenda et al. paper. For example, for PDB entry 6H1K (Fig. 6), the relevant results with the --g4-onz option and without it are listed below:

# x3dna-dssr -i=6h1k.pdb --g4-onz
List of 3 G-tetrads
   1 glyco-bond=s--- groove=w--n planarity=0.149 type=planar Z- nts=4 GGGG A.DG1,A.DG20,A.DG16,A.DG27
   2 glyco-bond=-sss groove=w--n planarity=0.136 type=planar Z+ nts=4 GGGG A.DG2,A.DG19,A.DG15,A.DG26
   3 glyco-bond=--s- groove=-wn- planarity=0.307 type=other  O+ nts=4 GGGG A.DG17,A.DG21,A.DG25,A.DG28
# ---------------------------------------
# x3dna-dssr -i=6h1k.pdb 
#   without option --g4-onz
List of 3 G-tetrads
   1 glyco-bond=s--- groove=w--n planarity=0.149 type=planar nts=4 GGGG A.DG1,A.DG20,A.DG16,A.DG27
   2 glyco-bond=-sss groove=w--n planarity=0.136 type=planar nts=4 GGGG A.DG2,A.DG19,A.DG15,A.DG26
   3 glyco-bond=--s- groove=-wn- planarity=0.307 type=other  nts=4 GGGG A.DG17,A.DG21,A.DG25,A.DG28

With the --json option, the ONZ classification results are always available. An example is shown below for PDB entry 6H1K (Fig. 6):

# x3dna-dssr -i=6h1k.pdb --json | jq -c '.G4tetrads[] | [.nts_long, .topo_class]'
["A.DG1,A.DG20,A.DG16,A.DG27","Z-"]
["A.DG2,A.DG19,A.DG15,A.DG26","Z+"]
["A.DG17,A.DG21,A.DG25,A.DG28","O+"]

I recently read a short communication by Pavel Afonine, titled phenix.hbond: a new tool for annotation hydrogen bonds in the July 2019 issue of the Computational Crystallography Newsletter (CCN). It appears that every bioinformatics tool (e.g., PyMOL or Jmol) has its own implementation of an algorithm on calculating H-bonds, one of the fundamental stabilizing forces of proteins and DNA/RNA structures. So does 3DNA/DSSR, as noted in my 2014-04-11 blogpost Get hydrogen bonds with DSSR.

Both DSSR and SNAP have the --get-hbond option, and they use the same underlying algorithm. However, the default output from the two programs differs: DSSR reports the H-bonds within nucleic acids, and SNAP covers only those at the DNA/RNA-protein interface. Using the PDB entry 1oct as an example, Running DSSR on it with the --get-hbond option gives 33 H-bonds in the DNA duplex, while SNAP reports 38 H-bonds at the DNA-protein interface. By design, the default output caters for the most-common use case of each program.

Under the scene, however, there exist variations in the seemingly simple --get-hbond option. One can attach text ‘nucleic’ (or ‘nuc’, ‘nt’), as in --get-hbond-nucleic, to output H-bonds within nucleic acids. Similarly, --get-hbond-protein (or ‘amino’, ‘aa’) would output H-bonds within proteins. Not surprisingly, the --get-hbond-nt-aa option would list H-bonds in nucleic acids and proteins, including those at their interface. These variations apply to both DSSR and SNAP, even though some are redundant with the default.

Notably, in combination with --json, the --get-hbond option by default would output all H-bonds, as if --get-hbond-nt-aa has been set. For PDB entry 1oct, DSSR or SNAP would report 208 H-bonds. Moreover, the JSON output has a residue_pair field for each identified H-bond, with values like "nt:nt", "nt:aa", or "aa:aa". Using 1oct as an example,

# x3dna-dssr -i=1oct.pdb --get-hbond --json | jq '.hbonds[0]'
{
  "index": 1,
  "atom1_serNum": 34,
  "atom2_serNum": 608,
  "donAcc_type": "standard",
  "distance": 3.304,
  "atom1_id": "O6@A.DG202",
  "atom2_id": "N4@B.DC230",
  "atom_pair": "O:N",
  "residue_pair": "nt:nt"
}
# x3dna-dssr -i=1oct.pdb --get-hbond --json | jq '.hbonds[60]'
{
  "index": 61,
  "atom1_serNum": 462,
  "atom2_serNum": 1187,
  "donAcc_type": "standard",
  "distance": 3.692,
  "atom1_id": "O2@B.DT223",
  "atom2_id": "NH2@C.ARG102",
  "atom_pair": "O:N",
  "residue_pair": "nt:aa"
}
# x3dna-dssr -i=1oct.pdb --get-hbond --json | jq '.hbonds[100]'
{
  "index": 101,
  "atom1_serNum": 791,
  "atom2_serNum": 818,
  "donAcc_type": "standard",
  "distance": 2.871,
  "atom1_id": "N@C.THR26",
  "atom2_id": "OD2@C.ASP29",
  "atom_pair": "N:O",
  "residue_pair": "aa:aa"
}

In the above three cases, using SNAP instead of DSSR would give the same results.

Also, one can take advantage of the residue_pair value to filter H-bonds by type. For example, the following command would extract only H-bonds at the DNA-protein interface (38 occurrences, same as the number noted above):

x3dna-snap -i=1oct.pdb --get-hbond --json | jq '.hbonds[] | select(.residue_pair=="nt:aa")'

Back to the phenix.hbond tool, the author noted that:

Running phenix.hbond requires atomic model in PDB or mmCIF format with all hydrogen atoms added, as well as ligand restraint files if the model contains unknown to the library items.

While there is no particular reason why this should not work for all bio-macromolecules, currently phenix.hbond is only optimized and tested to work with proteins, which is the limitation that will be removed in future.

In contrast, the H-bond identification algorithm in DSSR/SNAP does not require hydrogen atoms. In fact, hydrogen atoms are simply ignored if they exist. As shown above, the H-bond method as implemented in DSSR/SNAP works for DNA, RNA, protein, or their complexes. This does not necessarily mean that the 3DNA way is superior to other similar tools. It just works well in my hand, and it may serve as a pragmatic choice for other users.

Recently I noticed two new citations to DSSR, an integrated software tool for dissecting the spatial structure of RNA. One is from the Yesselman et al. article Computational design of three-dimensional RNA structure and function in Nature Nanotechnology, and the other is from the Wang et al. article 3dRNA v2.0: An Updated Web Server for RNA 3D Structure Prediction in International Journal of Molecular Sciences.

Yesselman et al. has used DSSR in RNAMake for building the motif library. The relevant section is as follows:

We processed each RNA structure to extract every motif with Dissecting the Spatial Structure of RNA (DSSR)54 with the following command:

x3dna-dssr –i file.pdb –o file_dssr.out

We manually checked each extracted motif to confirm that it was the correct type, as DSSR sometimes classifies tertiary contacts as higher-order junctions and vice versa. For each motif collected from DSSR, we ran the X3DNA find_pair and analyze programs to determine the reference frame for the first and last base pair of each motif to allow for the alignment between motifs:

find_pair file.pdb 2> /dev/null stdout | analyze stdin >& /dev/null

It is worth noting the sentence that “DSSR sometimes classifies tertiary contacts as higher-order junctions and vice versa.” Presumably. the authors are referring to the inclusion of ‘isolated canonical pairs’ in junctions by default in DSSR. Overall, the default DSSR settings follow the most common practice in RNA literature. In the meantime, I am aware that the community may not agree on every detail. Thus DSSR provide many options (documented or otherwise) to cater for other potential use cases. See the Stems of junction structure have only one base pair and Junction definition threads on the 3DNA Forum for two examples. In the long run, DSSR is likely to help consolidate RNA nomenclature that can be applied in a pragmatic, consistent manner.

Note also that DSSR provides the reference frame of each identified base pair via the JSON option. Using 1ehz as an example, the following command provides detailed information about base pairs:

x3dna-dssr -i=1ehz.pdb --json --more | jq .pairs

In the 3dRNA 2.0 paper, DSSR is cited as below. This is the first time DSSR is integrated in the 3dRNA pipeline.

The predicted structures are built from the sequence and secondary structure, while the former is obtained from their native structures fetched from PDB (https://www.rcsb.org/), and the latter is calculated from DSSR (Dissecting the Spatial Structure of RNA) [39].

In the PDB, the ligand identifiers 5MC and 5CM all refer to 5-methylcytosine, but differ in the sugar moieties the base is attached to. Chemically, 5CM is 5-methyl-2’-deoxycytidine-5’-monophosphate as in DNA, and 5MC is 5-methylcytidine-5’-monophosphate. See the molecular images shown below.

The 5-methyl group is named C5A in 5CM and CM5 in 5MC, respectively, for non-obvious reasons other than conventions. For comparison, the methyl-group in thymine of DNA is named C7, as for example in PDB id 355d. It is worth noting that DSSR is able to handle all such variations in atom or residue names.

I recently came across a Bioinformatics article VeriNA3d: an R package for nucleic acids data mining by Gallego et al. from IRB Barcelona. VeriNA3d can perform dataset analysis, single-structure analysis, and exploratory data analyses, with an emphasis on complex RNA structures. I am glad to see the DSSR is one of the third-party utilities that have been integrated into VeriNA3d, as shown below

VeriNA3d offers integration with third-party utilities such as the non-redundant lists of RNA structures (Leontis and Zirbel, 2012), the eRMSD suggested to compare RNA structures (Bottaro et al., 2014), a wrapper to the DSSR (Dissecting the Spatial Structure of RNA) software (Lu et al., 2015) and query functions to access the PDBe REST API (Velankar et al., 2016).

I browsed the GitLab repository and read through the supplemental documents. Clearly, VeriNA3d is a handy tool for the R community to perform RNA 3D structural analyses.

To DSSR users, Section “9 The dssr wrapper: getting the base pairs” of the supplemental PDF “VeriNA3d: introduction and use cases” is particularly relevant. The three paragraphs (with minor edits) are excerpted below:

The DSSR software (Dissecting the Spatial Structure of RNA) (Lu, Bussemaker, and Olson 2015) represents an invaluable resource to handle RNA structures. Some of the functions of veriNA3d overlap with the functionalities of DSSR, and both applications provide unique different features. We implement a wrapper to execute DSSR directly from R and get the best of both worlds in one place.

Note that installing veriNA3d does not automatically install DSSR, since we don’t redistribute third-party software. Before any user can use our wrapper, the dssr function, DSSR should be installed separately. To address this installation we redirect you to the DSSR manual, where anyone can find the specific instructions for their system. Once DSSR is installed and working in your computer, you will also be able to use it with our wrapper. If the DSSR executable (named x3dna-dssr) is in your path, dssr will find it automatically. If the wrapper does not find it, you can still use it specifying the absolute path to the executable with the argument exefile. Find more information running ?dssr.

One of the DSSR capabilities that users might be interested in is the detection and classification of base pairs. The following code shows a simple example. The output of the dssr wrapper is an object got from the json DSSR output. From R, json objects are parsed in the form of a tree of lists, with different types of information. Most of the interesting data is under the list models, sublist parameters, as shown herein.

I echo the authors’ policy of not redistributing third-party software with VeriNA3d. DSSR is under active development. Users should always visit the 3DNA Forum for downloading the latest version of DSSR, reporting bugs, and asking questions.

The R interface to DSSR (via JSON output) in VeriNA3d represents one of the intended use cases of DSSR’s many possible applications. No doubt DSSR is being increasingly integrated into other resources of RNA structural bioinformatics. Hopefully, more advanced DSSR features (than the detection and classification of base pairs) will also be widely appreciated in the future. Users would love DSSR better when they gain more experience in structural bioinformatics.

It is a great pleasure to see that our article Web 3DNA 2.0 for the analysis, visualization, and modeling of 3D nucleic acid structures has been highlighted in the cover page of the web server issue of NAR’19. According to the editor, This year, 331 proposals were submitted and 122, or 37%, were approved for manuscript submission. Of those approved, 94, or 77%, were ultimately accepted for publication. Overall, that corresponds to a ~28% acceptance rate.

The cover image and its caption are shown below. Moreover, details on how the cover image was created are available on the 3DNA Forum.

Caption: Examples of customized molecular models that can be generated with 3DNA: (top) a chromatin-like, nucleosome-decorated DNA with the structures of known histone-DNA assemblies placed at user-defined binding sites; (lower left) molecular schematic of a DNA trinucleotide diphosphate illustrating the base planes and reference frames used to construct and analyze 3D nucleic acid-containing structures; (lower right) customized single-stranded tRNA model built from a user-defined base sequence and a set of rigid-body parameters describing the desired placement of successive bases. Color code of base blocks: A, red; C, yellow; G, green; T, blue; U, cyan.

A paper titled DNA Conformational Changes Play a Force-Generating Role during Bacteriophage Genome Packaging has just been officially published in the Biophysical Journal (Volume 116, Issue 11, P2172-2180, June 04, 2019). I am glad to have the opportunity to collaborate with Kim Sharp, Gino Cingolani and Stephen Harvey on this interesting project that has big implications in understanding the mechanism of bacteriophage genome packaging. The abstract of the paper is shown below:

Motors that move DNA, or that move along DNA, play essential roles in DNA replication, transcription, recombination, and chromosome segregation. The mechanisms by which these DNA translocases operate remain largely unknown. Some double-stranded DNA (dsDNA) viruses use an ATP-dependent motor to drive DNA into preformed capsids. These include several human pathogens as well as dsDNA bacteriophages—viruses that infect bacteria. We previously proposed that DNA is not a passive substrate of bacteriophage packaging motors but is instead an active component of the machinery. We carried out computational studies on dsDNA in the channels of viral portal proteins, and they reveal DNA conformational changes consistent with that hypothesis. dsDNA becomes longer (“stretched”) in regions of high negative electrostatic potential and shorter (“scrunched”) in regions of high positive potential. These results suggest a mechanism that electrostatically couples the energy released by ATP hydrolysis to DNA translocation: The chemical cycle of ATP binding, hydrolysis, and product release drives a cycle of protein conformational changes. This produces changes in the electrostatic potential in the channel through the portal, and these drive cyclic changes in the length of dsDNA as the phosphate groups respond to the protein’s electrostatic potential. The DNA motions are captured by a coordinated protein-DNA grip-and-release cycle to produce DNA translocation. In short, the ATPase, portal, and dsDNA work synergistically to promote genome packaging.

Significantly, our work is highlighted in a “New and Notable” article, May the Road Rise to Meet You: DNA Deformation May Drive DNA Translocation by Paul Jardine (Volume 116, Issue 11, Pages 2060-2061, 4 June 2019):

Regardless of what drives conformational change in the portal, the idea that the linear DNA substrate is deformed in a way that makes it an energetic participant in its own movement opens new possibilities for how motors work. Large paddling or rotational motions by motor components may not be required if linear motion can be achieved by stretching or compressing the linear substrate, with rectified, cyclic conformational changes in the DNA rather than lever motions doing the work. If borne out by experiments, further simulation, and more structural information, this proposed mechanism may require a reappraisal of how we think about translocating motors.

For this project, I developed the x3dna-search program to survey similar fragments of single-stranded or double helical structures in the PDB.