X3DNA-DSSR: a resource for structural bioinformatics of nucleic acids
(An NIGMS National Resource supported by NIH grant R24GM153869)

As of beta-r20-on-20130830, DSSR is able to detect two types of U-turns (see the figure below), the UNR-type (left) originally identified by Quigley and Rich [1976] in yeast phenylalanine tRNA, and the GNRA-type (right) later on established by Jucker and Pardi [1995] in GNRA tetra loops. See the Gutell et al. paper Predicting U-turns in Ribosomal RNA with Comparative Sequence Analysis for a more extensive account of U-turns.

As its name implies, a U-turn is characterized by a reversal of the RNA backbone direction within a few nucleotides. Among other factors, the U-turn is stabilized by two key H-bonding interactions, illustrated in dotted lines in the figure below.

UNR-type (1ehz)	GNRA-type (1msy)

Applying DSSR to 1jj2 (the crystal structure of the Haloarcula marismortui large ribosomal subunit) led to the identification of over 30 cases. In addition to the well-documented UNR- and GNRA-type U-turns, the program also finds other variants. An example is shown below, where the U-turn is within a GCA triloop instead of a GNRA tetraloop. Here, the N1 (not N2) atom of G1809 forms an H-bond with OP2 of G1812. The G1809 N2 atom is H-bonded to G1812 O5′ to further stabilize the U-turn.

An examination of the chemical structure of the nitrogenous bases (see figure below) shows clearly other possibilities to connect RNA base donors to the phosphate oxygen acceptors. DSSR allows for the exploration of such variations, and more.

3DNA can build DNA/RNA structures with a precise base but approximate sugar-phosphate backbone geometry. In the 2003 3DNA-NAR paper, Table 3 of the section “Structures built with sugar–phosphate backbone” lists “root mean square deviation (in Å) between rebuilt 3DNA models and experimental DNA structures” for three representative DNA structures (in A-form, B-form, and a protein-DNA complex). It was noted that The RMSD of reconstructed versus observed base positions is virtually zero and that for both base and backbone coordinates is <0.85 Å, even for the 146 bp nucleosomal DNA structure.

The backbone geometry is approximate because 3DNA uses a fixed sugar-phosphate conformation (in A-DNA, B-DNA or RNA) that is attached to the corresponding bases in the model building process. The most noticeable effect is the long O3′(i)···P(i+1) bond that connects consecutive nucleotides along a chain. The imprecise structure was intended as a starting point for other objectives (e.g., all-atom molecular dynamics simulations) that are out of the design scope of 3DNA. Nevertheless, over the years, I have been concerned with the overlong O3′—P distance issue. I tried but failed to find a satisfying third-party (command-line driven) tool that can perform restraint optimization of the sugar-phosphate backbone geometry while keeping base atoms fixed.

The problem was finally solved after I attended the 43rd Mid-Atlantic Macromolecular Crystallography Meeting held at Duke University a few months ago. At the meeting, I had the opportunities to talk to several members of the PHENIX team. Particularly, Jeff Headd revised the geometry_minimization component of PHENIX to do the trick. Here is the mail reply from Jeff, using a 3DNA-generated DNA duplex (355d-3dna.pdb) as an example (see full details below):

Here’s a first go at refining just the backbone atoms of you input DNA model. You’ll need the most recently nightly build of Phenix (dev-1395 would work) and then run:

phenix.geometry_minimization 355d-3dna.pdb min.params

using the attached min.params file.

What I specify in the params file is to only move the backbone atoms, which I’ve done with a selection. You can modify the atoms that are allowed to move to your liking.

The only other change was to allow longer distance linkages, as some of the backbone linkages start quite far apart.

The content of file min.params is:

pdb_interpretation {
  link_distance_cutoff = 7.0
}
selection = name " P  " or name " OP1" or name " OP2" or \
            name " O5'" or name " C5'" or name " C4'" or \
            name " O4'" or name " C3'" or name " O3'" or \
            name " C2'"

To make the story complete, given below is the step-by-step procedure, using 355d, a B-DNA dodecamer at 1.4 Å resolution as an example. The corresponding PDB file is named 355d.pdb.

find_pair 355d.pdb stdout | analyze stdin
x3dna_utils cp_std bdna
rebuild -atomic bp_step.par 355d-3dna.pdb
# the rebuilt structure is called '355d-3dna.pdb'

# with Phenix dev-1395 and above
phenix.geometry_minimization 355d-3dna.pdb min.params
# the optimized structure is called '355d-3dna_minimized.pdb'

# to verify:
find_pair 355d-3dna.pdb stdout | analyze stdin
find_pair 355d-3dna_minimized.pdb stdout | analyze stdin
# check files '355d-3dna.out' and '355d-3dna_minimized.out'

The three key files mentioned above are provided here for your verification:

• 355d-3dna.pdb
• 355d-3dna_minimized.pdb
• min.params

Finally, the following figure illustrates the B-DNA dodecamer duplex in experimental (left), 3DNA-generated (middle) and PHENIX-optimized (right) coordinates. Note that disconnected O3′—P linkages (marked by red dots for two cases, see bottom of the middle image) due to overlong distances in 3DNA-rebuilt structure are fixed following the restraint PHENIX optimization.

355d-experimental	3DNA-rebuilt	PHENIX-optimized

Note added on 2016-11-11: In the min.params file, the selection is in one long line. For illustration purpose, the selection section (see below) is split into serveral short lines in the blog post. However, PHENIX requires ending backslashes (\) to combine the split lines into a single grammatical unit. I was not aware of this strict rule, and missed to add the ending \s in the original post. Thanks to Oleg Sobolev from the PHENIX team for pointing out this omission to my attention. Note that the content of min.params did not have a problem, and thus no change is made.

pdb_interpretation {
  link_distance_cutoff = 7.0
}
selection = name " P  " or name " OP1" or name " OP2" or \
            name " O5'" or name " C5'" or name " C4'" or \
            name " O4'" or name " C3'" or name " O3'" or \
            name " C2'"

One of DSSR’s noteworthy features is the auto-detection of helical junctions in nucleic acids structures, be it RNA, DNA, or chimeric DNA/RNA, consisting of one or multiple chains. Helical junctions are created at the interface of three and more stems composed of canonical pairs (Watson-Crick A—T/U and G—C, or wobble G—U). A three-way junction model is illustrated below (copied from Figure 1 of the Bindewald et al. RNAJunction paper). Note that the three chains are each continuous (i.e., consecutive nts are covalently connected), and together with the three inner bps, forming a loop in the middle. Here, the three-way junction is of type [3×2×3], and the loop is composed of a total of 3×2+3+2+3 = 14 nts.

DSSR automatically detects all existing helical junctions in a nucleic acid structure, as illustrated by the following examples.

1l6b [all DNA Holliday junction structure of d(CCGGTACm5CGG)]

This is a simple four-way junction of type [0×0×0×0], where all bases are paired, leaving no connecting nts. The related portion of DSSR output is:

List of 1 junction(s)
   1 4-way junctions: 8 nts; [0x0x0x0]; linked by [#1, #2, #4, #3]
       1:A.DA6+1:A.DC7+2:B.DG14+2:B.DT15+2:A.DA6+2:A.DC7+1:B.DG14+1:B.DT15 [ACGTACGT]
       0 nts junction ; 1:A.DA6-->1:A.DC7 [AC]
       0 nts junction ; 2:B.DG14-->2:B.DT15 [GT]
       0 nts junction ; 2:A.DA6-->2:A.DC7 [AC]
       0 nts junction ; 1:B.DG14-->1:B.DT15 [GT]

Technically, note the following points:

• The four-way junction is derived from the biological assembly 1 (PDB file 1l6b.pdb1), which contains two copies of the asymmetric unit, delineated by MODEL/ENDMDL. By default, DSSR/3DNA works one structure at a time, corresponding to the first structure/model in a given PDB or mmCIF file. To take the biological assembly as a whole, and to avoid confusions with MODEL/ENDMDL delineated NMR entries, the ENDMDL record of the first model is commented out in the file (1l6b.pdb1), as below:

#ENDMDL                                                                          
MODEL        2                                                                  

• With the modified PDB file 1l6b.pdb1, the DSSR command can be run as x3dna-dssr -i=1l6b.pdb1, with the output going to stdout.

• The simplified schematic block png image was generated with the command below to create the Raster3D .r3d file (1l6b.r3d), which was then ray-traced using PyMOL.

blocview -r 1l6b.r3d 1l6b.pdb1

1egk [a four-way DNA/RNA junction]

This four-way junction consists of both DNA and RNA chains. Here the helical junction may not be that obvious by directly looking at the 3D image.

List of 1 junction(s)
   1 4-way junctions: 10 nts; [0x0x1x1]; linked by [#3, #-1, #4, #5]
       B.DC37+B.DT38+B.DA45+B.DC46+C.G109+C.A110+C.U111+D.DA130+D.DG131+D.DG132 [CTACGAUAGG]
       0 nts junction ; B.DC37-->B.DT38 [CT]
       0 nts junction ; B.DA45-->B.DC46 [AC]
       1 nts junction C.A110 [A]; C.G109-->C.U111 [GAU]
       1 nts junction D.DG131 [G]; D.DA130-->D.DG132 [AGG]

1ehz [yeast phenylalanine tRNA]

As shown below, DSSR correctly detects the classic L-shaped 3D structure and the cloverleaf 2D structure of a tRNA.

List of 1 junction(s)
   1 4-way junctions: 16 nts; [2x1x5x0]; linked by [#1, #2, #3, #4]
       A.U7+A.U8+A.A9+A.2MG10+A.C25+A.M2G26+A.C27+A.G43+A.A44+A.G45+A.7MG46+A.U47+A.C48+A.5MC49+A.G65+A.A66 [UUAgCgCGAGgUCcGA]
       2 nts junction A.U8+A.A9 [UA]; A.U7-->A.2MG10 [UUAg]
       1 nts junction A.M2G26 [g]; A.C25-->A.C27 [CgC]
       5 nts junction A.A44+A.G45+A.7MG46+A.U47+A.C48 [AGgUC]; A.G43-->A.5MC49 [GAGgUCc]
       0 nts junction ; A.G65-->A.A66 [GA]

2fk6 [RNAse Z/tRNA(Thr) complex]

In a recent paper Predicting Helical Topologies in RNA Junctions as Tree Graphs by Laing et al., this PDB entry was selected in Table 1 as containing a three-way junction. However, DSSR fails to detect any junction in this structure, even though the program does find co-axial stacks. It turns out that the PDB entry 2fk6 does not possess the anti-codon stem/loop, thus nts C25 and G46 are not covalently connected. While three-way junctions may be defined differently, the DSSR result follows the above mentioned chain-continuity requirement.

Overall, DSSR can consistently find all helical junctions in a given nucleic acid structure. Try DSSR on a ribosomal structure, you may well appreciate what it reveals. Moreover, it is straightforward to apply the program to all RNA/DNA-containing entries in the PDB via a script.

Given the primary sequence of an RNA molecule, there are numerous methods for predicting its secondary (2D) structures. To judge their accuracy, three-dimensional (3D) RNA structures solved experimentally by X-ray or NMR as deposited in the PDB are often used as benchmarks. DSSR is a handy tool to derive an RNA 2D structure from its 3D coordinates in PDB or mmCIF format. The 2D structure is specified in the dot-bracket notation (dbn), which can be fed directly into drawing programs such as VARNA for interactive display and easy generation of publication quality 2D diagrams.

Over the past few months, I’ve been asked a few times on the details of how the diagrams in the DSSR post were created. The answer is really simple, and has already been mentioned above and in the post. Here are two concrete examples to show how the process works.

1zc5 (structure of the RNA signal essential for translational frame shifting in HIV-1)

This is the structure used in the VARNA paper. Let the PDB file be named 1zc5.pdb, the DSSR program can be run like this:

x3dna-dssr -i=1zc5.pdb

The output is sent to stdout by default, with the following three lines towards the end:

>1zc5-A #1 RNA with 41 nts
GGCGAUCUGGCCUUCCUACAAGGGAAGGCCAGGGAAUUGCC
(((((((((((((((((....)))))))))))...))))))

Simply copy and paste the last two lines (sequence and the 2D structure in dbn notation) into the Seq: and Str: fields of the VARNA demo page, the diagram will be updated automatically, as shown in the screenshot:

1ehz (crystal structure of yeast phenylalanine tRNA at 1.93 Å resolution)

This example (1ehz.pdb) is used to illustrate tRNA’s classic cloverleaf 2D structure. The related command and result are:

x3dna-dssr -i=1ehz.pdb -o=1ehz.out

# the output is sent to file '1ehz.out'
# towards its end are the following 3 lines

>1ehz-A #1 RNA with 76 nts
GCGGAUUUAgCUCAGuuGGGAGAGCgCCAGAcUgAAgAPcUGGAGgUCcUGUGuPCGaUCCACAGAAUUCGCACCA
(((((((..((((.....[..)))).((((.........)))).....(((((..]....))))))))))))....

I’ve used a local copy of the JAVA web start version of VARNA (VARNA-WebStart.jnlp) to generate the following 2D diagram. Here, in addition to the customized title, I have set the number period to 5 nts, adopted the simple base-pair style, and manually adjusted the T arm (upper right corner) to make the long line connecting G19 and C56 a bit more unobtrusive. Right-click to see the context menu.

Note that the G19—C56 pair creates a pseudo-knot (specified by the matching [] pair in the dbn notation above) in tRNA. I was not aware of this salient feature from previous knowledge of relevant literature. It was indeed a surprise when I first saw it in the 2D diagram.

As illustrated above, DSSR serves well as a bridge from RNA 3D to 2D structures. Give DSSR a try, you will find the program actually has much more to offer!

As of June 24, 2013, the number of 3DNA Forum registrations has passed the 1000 mark. On September 16, 2012, I wrote the post The number of 3DNA forum registrations has reached 500. Thus, in slightly over 9 months, the number has doubled, with approximately 2 registrations per day.

I am glad to see the steady increase of the 3DNA user base. Over the time, I have strived to be responsive to user questions, and made every effort to keep the forum spam free. By and large, employing simple 3DNA-related questions has turned out to be an effective anti-spam strategy. Since the launch of the new forum.x3dna.org in March 2012, I’ve received less than five requests (to the best of my memory) asking for help on registrations. As a recently example, a potential user got stuck with the question about what ‘w’ means in w3DNA. Based on user feedback, I have added hints to some questions to make their answers more obvious. Whatever the reasons, each reported issue has been promptly resolved.

With the release of DSSR and the continuous support of an enthusiastic user community, I have every reason to believe that 3DNA will gain more popularity in the years to come.

As of the beta-r14-on-20130626 release, DSSR has the functionality to identify kink-turns and reverse k-turns given an RNA structure in PDB format.

The k-turn motif was first described by Klein et al. (2001) in the paper The kink-turn: a new RNA secondary structure motif, based on analyses of the H. marismortui large ribosomal unit. It turns out to be a widespread structural motif, now with a dedicated k-turn database hosted by the Lilley laboratory.

Geometrically, k-turn is composed of an asymmetric internal loop, with a sharp kink between the two framing helices and characteristic loop features (including at least one sheared G-A pair and A-minor interactions). Overall, k-turn is a complicated motif, and I am not aware of any published method or available software for its auto-detection.

Previous releases of DSSR has built up all the necessary components to detect key features of a k-turn. Over the past few weeks, I have been focusing on connecting the dots to implement an algorithm for its auto-identification. As of beta-r14-on-20130626, DSSR can locate ‘simple’ k-turns or reverse k-turns from an RNA structure in PDB format. I understand the subtleties and variations of k-turns, and will refine the algorithm in future releases of DSSR.

Without putting k-turns under its umbrella, DSSR appears incomplete in its functionality. Hopefully, detection of k-turns will help DSSR gain more attention from the RNA structure community.

A new paper titled Analyzing and Building Nucleic Acid Structures with 3DNA has been published in JoVE (Journal of Visualized Experiments). Specifically, the article illustrates 3DNA’s unique capability to characterize and modify DNA structures at the level of the constituent base-pair steps, and highlights a new feature in v2.1 to analyze and align an ensemble of related structures determined with NMR or generated by MD simulations.

Here is the abstract:

The 3DNA software package is a popular and versatile bioinformatics tool with capabilities to analyze, construct, and visualize three-dimensional nucleic acid structures. This article presents detailed protocols for a subset of new and popular features available in 3DNA, applicable to both individual structures and ensembles of related structures. Protocol 1 lists the set of instructions needed to download and install the software. This is followed, in Protocol 2, by the analysis of a nucleic acid structure, including the assignment of base pairs and the determination of rigid-body parameters that describe the structure and, in Protocol 3, by a description of the reconstruction of an atomic model of a structure from its rigid-body parameters. The most recent version of 3DNA, version 2.1, has new features for the analysis and manipulation of ensembles of structures, such as those deduced from nuclear magnetic resonance (NMR) measurements and molecular dynamic (MD) simulations; these features are presented in Protocols 4 and 5. In addition to the 3DNA stand-alone software package, the w3DNA web server, located at http://w3dna.rutgers.edu, provides a user-friendly interface to selected features of the software. Protocol 6 demonstrates a novel feature of the site for building models of long DNA molecules decorated with bound proteins at user-specified locations.

A new section dedicated to the JoVE paper will be set up on the 3DNA Forum soon. It will contain all the data files and scripts so our published results can be strictly reproduced. The section should also serve as a platform for open discussions of related protocols.

Over the past six months or so¹, I’ve been focusing mostly on developing DSSR, a new addition to the 3DNA suite of programs. So what is DSSR, specifically? Why did I bother to create it? How would it be relevant to the nucleic acid structure community?

Literally, DSSR stands for Defining the (Secondary) Structures of RNA². Starting from an RNA structure in PDB format, DSSR employs a set of simple criteria to identify all existent base pairs (bp): both canonical Watson–Crick (WC) pairs and non-canonical pairs with at least one H-bond, made up of normal or modified bases, regardless of tautomeric or protonation state. The classification is based on the six standard rigid-body bp parameters (shear, stretch, stagger, propeller, buckle, and opening), which together rigorously quantify the spatial disposition of any two interacting bases. Moreover, the program characterizes each bp by commonly used names (WC, reverse WC, Hoogsteen, reverse Hoogsteen, wobble, sheared, imino, Calcutta, and dinucleotide platform), the Saenger classification scheme of 28 types, and the Leontis-Westhof nomenclature of 12 basic geometric classes. DSSR also checks for non-pairing interactions (H-bonds or base stacking).

DSSR detects triplets and even higher-order base associations by searching horizontally in the plane of the associated bp for further H-bonding interactions. The program determines helical regions by exploring each bp’s neighborhood vertically for base-stacking interactions, regardless of backbone connection (e.g., coaxial stacking of helices or pseudo helices). Moreover, each helix/stem is characterized by a least-squares fitted helical axis to allow for easy quantification of relative helical geometry. DSSR calculates commonly used backbone (including the virtual η/θ) torsion angles, classifies the main chain backbone into BI/BII conformation and the sugar into C2’/C3’-endo like pucker, identifies A-minor interactions (types I and II), ribose zippers, G quartets, hairpin loops, kissing loops, bulges, internal loops and multi-branch loops (junctions). It also detects the existence of pseudo-knots, and outputs RNA secondary structure in the dot-bracket notation.

Experienced 3DNA users may notice that some of the above outlined functionality (e.g., calculation of torsion angles, identification of all pairs, higher order base associations, and helices) have existed for over a decade. Over the years, I have written several posts (see What can 3DNA do for RNA structures?, and links therein) to advocate 3DNA’s applications in RNA structural analysis. Nevertheless, 3DNA has never been widely used in the RNA structure community, for various possible reasons: (1) the misconception that 3DNA is only for DNA (but not RNA); (2) the basic functionality is split into two programs (find_pair and analyze), and needs to be run several times with different options (default find_pair, and with -s, or -p). Thus even though 3DNA is applicable to RNA structures, it is unnecessarily complicated and confusing (especially to new 3DNA users); (3) 3DNA is command-line driven, consisting of many C programs and scripts, with different styles in specifying options. It has the ‘reputation’ of being powerful, but cryptic and hard to use.

I’ve created DSSR from scratch to take consideration of these factors, by employing my extensive experience in supporting 3DNA, an increased knowledge in RNA structures and refined C programming skills. Implemented in ANSI C as a stand-alone command-line program, DSSR is self-contained. Its executables (on MacOS X, Linux and Windows) have zero runtime dependencies. No setup is necessary; simply put the program into a folder of your choice (preferably one on your command PATH), and it should work. DSSR has sensible default settings and an intuitive output, making it directly accessible to a much broader audience than 3DNA per se. Since its initial release on March 3, 2013, I’ve yet to hear any installation or usage problem. So far, all reported bugs have been verified and fixed promptly. The latest beta release has been checked against all nucleic-acid-containing entries in the PDB, without any known issues.

Overall, DSSR consolidates, refines, and significantly extends 3DNA’s functionality for RNA structural analysis. There are more in DSSR than its simple interface suggests. Piecewise, DSSR may appear nothing new, yet combined together, it has unique features not available anywhere else. Its value will be gradually appreciated as DSSR becomes more widely used by the community. Want to know if your structure contains any Hoogsteen pair, sheared G•A pair, or a dinucleotide platform? DSSR can check it for you, easily.

DSSR-beta already possesses all the basic functionality and has been well tested to serve as a handy tool for RNA structural analysis. I stand firmly behind DSSR, and strive to continuously improve the program. Give it a try, and report back on the 3DNA Forum any issues you have. As always, I respond quickly and concretely to all questions posted there. I hope you enjoying using DSSR as much as I enjoy creating and supporting it!

¹ This post was published on March 29, 2013, shortly after the beta releases of DSSR [note added on March 15, 2014].

² DSSR also works for DNA, or DNA-protein complexes, as far as the basic functionality is concerned. Moreover, the acronym could have two other possible interpretations, as would be obvious when the program gains a wider recognition.

Early on when I started on DNA structures, I read Saenger’s book Principles of Nucleic Acid Structure and became familiar with his classification of the 28 possible base-pairs (bps) for A, G, U(T), and C involving at least two (cyclic) hydrogen bonds (see figure below).

Later on, I read from the 2nd edition of The RNA World book a list of 29 bps compiled by Burkard, Turner & Tinoco. While the one bp discrepancy (28 vs 29) has been in my mind for quite a long while, I had never paid much attention to the issue until recently while adding classifications of RNA bps (among many other functionalities) to 3DNA. A Google search did not help solve the puzzle, so I decided to dig it out by comparing the two lists.

The Burkard et al. list is titled Structures of Base Pairs Involving at Least Two Hydrogen Bonds and it mentions specifically Saenger’s list:

The structures of 29 possible base pairs that involve at least two hydrogen bonds are given in Figures 1–5 (for further descriptions, see Saenger, in Principles of nucleic acid structure, p. 120. Springer-Verlag [1984]).

However, in the five figures, Burkard et al. do not provide the corresponding Saenger numbers (I to XXVIII, 1—28) for the 28 common bps; thus it is not immediately obvious which one (i.e., the new addition by Burkard et al.) is missing from Saenger’s list. Under careful scrutiny, the absent bp turns out to be the “G•C N3-amino, amino-N3” pair in Figure 3: “Six possible flipped purine-pyrimidine mismatches.” One example of such G+C pair is found in the 5S ribosomal RNA (chain 9, G3022—C3026) of Haloarcula marismortui in PDB entry 1vq8.

The above figure shows clearly that the G+C bp does indeed have two canonical H-bonds between base atoms, and it is difficult to speculate how it escaped Saenger’s selection criteria. In the upcoming new 3DNA component, I am listing this bp as number XXIX (29), along with the other 28 base pairs.