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

From early on, DSSR-derived nucleic acid secondary structures have been written in the compact dot-bracket notation (.dbn) with pseudo-knot information. To better connect DSSR to the 2D world, I recently looked into the connect (.ct) format, which was first introduced by Zuker’s mfold program. Over time, the .ct format has become one of the most commonly used RNA secondary structure formats, and it is more expressive than the .dbn format (see below).

As of v1.0, for each analyzed structure, DSSR produces two secondary structure files with default names dssr-2ndstrs.dbn and dssr-2ndstrs.ct, in .dbn and .ct formats, respectively. Using the 27-nucleotides (nt) RNA fragment 1msy as an example, the DSSR-derived secondary structure in .dbn and .ct formats are shown below:

In dot-bracket notation (.dbn) [dssr-2ndstrs.dbn]
------------------------------------------------------
>1msy nts=27 DSSR-derived secondary structure
UGCUCCUAGUACGUAAGGACCGGAGUG
.(((((.....(....)....))))).
------------------------------------------------------

In connect format (.ct) [dssr-2ndstrs.ct]
------------------------------------------------------
   27 DSSR-derived secondary structure in '1msy'
    1 U     0     2     0  2647 # name=A.U2647
    2 G     1     3    26  2648 # name=A.G2648, pairedNt=A.U2672
    3 C     2     4    25  2649 # name=A.C2649, pairedNt=A.G2671
    4 U     3     5    24  2650 # name=A.U2650, pairedNt=A.A2670
    5 C     4     6    23  2651 # name=A.C2651, pairedNt=A.G2669
    6 C     5     7    22  2652 # name=A.C2652, pairedNt=A.G2668
    7 U     6     8     0  2653 # name=A.U2653
    8 A     7     9     0  2654 # name=A.A2654
    9 G     8    10     0  2655 # name=A.G2655
   10 U     9    11     0  2656 # name=A.U2656
   11 A    10    12     0  2657 # name=A.A2657
   12 C    11    13    17  2658 # name=A.C2658, pairedNt=A.G2663
   13 G    12    14     0  2659 # name=A.G2659
   14 U    13    15     0  2660 # name=A.U2660
   15 A    14    16     0  2661 # name=A.A2661
   16 A    15    17     0  2662 # name=A.A2662
   17 G    16    18    12  2663 # name=A.G2663, pairedNt=A.C2658
   18 G    17    19     0  2664 # name=A.G2664
   19 A    18    20     0  2665 # name=A.A2665
   20 C    19    21     0  2666 # name=A.C2666
   21 C    20    22     0  2667 # name=A.C2667
   22 G    21    23     6  2668 # name=A.G2668, pairedNt=A.C2652
   23 G    22    24     5  2669 # name=A.G2669, pairedNt=A.C2651
   24 A    23    25     4  2670 # name=A.A2670, pairedNt=A.U2650
   25 G    24    26     3  2671 # name=A.G2671, pairedNt=A.C2649
   26 U    25    27     2  2672 # name=A.U2672, pairedNt=A.G2648
   27 G    26     0     0  2673 # name=A.G2673
------------------------------------------------------

Presumably, the .ct format is very simple, and examining a sample file as shown above would give one a pretty good sense of what each column is about. While there exist many oversimplified descriptions of the .ct format on the web, the most detailed and accurate explanation is from the mfold manual:

The ``ct’‘ file (connect table) contains the sequence and base pair information, and is meant to be an input file for a structure drawing program. In addition to containing base pair information, it also lists the 5′ and 3′ neighbor of each base, allowing for the representation of circular RNA or multiple molecules. The ct file also lists the historical base numbering in the original sequence, as bases and base pairs are numbered according from 1 to the size of the folded segment. A portion of a ct file is displayed in Figure 12.

Figure 12: The ct file for the second and final folding of S. cerevisiae Phe-tRNA at 37°, with default parameters. The first record displays the fragment size (76), ΔG and sequence name. The i^th subsequent record contains, in order, i, r_i, the index of the 5′-connecting base, the index of the 3′-connecting base, the index of the paired base and the historical numbering of the i^th base in the original sequence. The 5′, 3′ and base pair indices are 0 when there is no connection or base pair.

Specifically, the 3rd, 4th, and 6th columns in the .ct format convey specific information; by design, they are not redundant to information contained in the 1st column. Note that in the above ‘1msy’ example, the 6th column gives the nt sequence numbers (as in the PDB datafile) instead of the serial numbers (as in the 1st column). The DSSR produced .ct files also contain extra information after ‘#’, in the comma separated key=value format.

As an example of the usefulness of the 3rd and 4th columns, have a look of the DSSR-derived .ct file for the Dickerson DNA dodecamer duplex with sequence CGCGAATTCGCG:

   24 DSSR-derived secondary structure in '355d'
    1 C     0     2    24     1 # name=A.DC1, pairedNt=B.DG24
    2 G     1     3    23     2 # name=A.DG2, pairedNt=B.DC23
    3 C     2     4    22     3 # name=A.DC3, pairedNt=B.DG22
    4 G     3     5    21     4 # name=A.DG4, pairedNt=B.DC21
    5 A     4     6    20     5 # name=A.DA5, pairedNt=B.DT20
    6 A     5     7    19     6 # name=A.DA6, pairedNt=B.DT19
    7 T     6     8    18     7 # name=A.DT7, pairedNt=B.DA18
    8 T     7     9    17     8 # name=A.DT8, pairedNt=B.DA17
    9 C     8    10    16     9 # name=A.DC9, pairedNt=B.DG16
   10 G     9    11    15    10 # name=A.DG10, pairedNt=B.DC15
   11 C    10    12    14    11 # name=A.DC11, pairedNt=B.DG14
   12 G    11     0    13    12 # name=A.DG12, pairedNt=B.DC13
   13 C     0    14    12    13 # name=B.DC13, pairedNt=A.DG12
   14 G    13    15    11    14 # name=B.DG14, pairedNt=A.DC11
   15 C    14    16    10    15 # name=B.DC15, pairedNt=A.DG10
   16 G    15    17     9    16 # name=B.DG16, pairedNt=A.DC9
   17 A    16    18     8    17 # name=B.DA17, pairedNt=A.DT8
   18 A    17    19     7    18 # name=B.DA18, pairedNt=A.DT7
   19 T    18    20     6    19 # name=B.DT19, pairedNt=A.DA6
   20 T    19    21     5    20 # name=B.DT20, pairedNt=A.DA5
   21 C    20    22     4    21 # name=B.DC21, pairedNt=A.DG4
   22 G    21    23     3    22 # name=B.DG22, pairedNt=A.DC3
   23 C    22    24     2    23 # name=B.DC23, pairedNt=A.DG2
   24 G    23     0     1    24 # name=B.DG24, pairedNt=A.DC1

Note the 0 at the 4th column for A.DG12 which is at the 3′ end of chain A, and the 0 at 3rd column for B.DC13 which is at the 5′ end of chain B.

From early on, 3DNA calculates the Zp parameter to separate A- and B-DNA double helical steps. First introduced in the paper A-form conformational motifs in ligand-bound DNA structures (see figure below), Zp is the mean projection of the two phosphorus atoms onto the z-axis of the dimer ‘middle frame’. Zp is greater than 1.5 Å for A-DNA, and it is less than 0.5 Å for B-DNA. As noted in the 3DNA NAR paper, other parameters such as slide should also be examined to confirm conformational assignments based on Zp.

As of v2.1, 3DNA has introduced the single-stranded variant for the Zp parameter (ssZp) as a more robust substitute for the Richardson phosphorus-glycosidic bond distance parameter (Dp) to characterize sugar puckers. See post Sugar pucker correlates with phosphorus-base distance for more details. In 3DNA/DSSR, ssZp is defined as the z-coordinate of the 3′ phosphorus atom expressed in the standard reference frame of the preceding base; it is positive when phosphorus lies on the +z-axis side (base in anti conformation) and negative if phosphorus is on the –z-axis side (base in syn conformation). Note that by definition, Dp should always be positive.

As in the previous post, here I am using G175 and U176 of PDB entry 1jj2 (the large ribosomal subunit of Haloarcula marismortui) as examples to illustrate how the ssZp parameters are calculated. The GpU forms a dinucleotide platform, where the sugar of G175 adopts a C2′-endo conformation, and that of U176 C3′-endo. For verification, here is the PDB data file for fragment 1jj2-G175-U176-A177.pdb (note A177 is included for its phosphorus atom). Run the following 3DNA commands:

find_pair -s 1jj2-G175-U176-A177.pdb stdout
frame_mol -1 ref_frames.dat 1jj2-G175-U176-A177.pdb ref-G175.pdb
frame_mol -2 ref_frames.dat 1jj2-G175-U176-A177.pdb ref-U176.pdb

File ref-G175.pdb contains the following line:

ATOM     24  P     U 0 176      -5.624   6.937   1.918  1.00 24.19           P 

The z-coordinate of U176 (which is 3′ to G175) is 1.918, which is the ssZp for G175. It is less than 2.9 Å, corresponding to the C2′-endo sugar conformation of G175.

Similarly, file ref-U176.pdb contains the following line:

ATOM     44  P     A 0 177      -3.841   6.592   4.377  1.00 25.91           P

So the ssZp for U176 is 4.377, which is greater than 2.9 Å, corresponding to the C3′-endo sugar conformation of U176.

To sum up, the double-stranded Zp as originally available from 3DNA can be used for discriminating A- and B-DNA double-helical steps: Zp > 1.5 Å for A-DNA, and Zp < 0.5 Å for B-DNA. The newly introduced single-stranded Zp is intended for characterizing sugar puckers: Zp > 2.9 Å for C3′-endo, and Zp < 2.9 Å for C2′-endo. Since A-DNA has predominately C3′-endo sugar conformation and B-DNA has C2′-endo sugar, the ssZp parameter would be helpful in classifying a dinucleotide into A- or B-like conformation. A survey of ssZp in well-defined A- and B-DNA structures (as performed for double-stranded Zp) should prove useful.

Realizing the naming confusions of double-stranded Zp vs single-stranded Zp, I am considering to rename single-stranded Zp as ssZp in future releases of 3DNA and DSSR. Do you have any comments or suggestions? Please let me know by leaving a comment!

Recently I was surprised by some cases of nucleotides with missing atoms in PDB entry 1pns. The story started like this: 3DNA/DSSR maps various nucleotide names to one-letter codes, based on the data file baselist.dat (see post Modified nucleotides in the PDB). In the meantime, 3DNA/DSSR internally assigns a nucleotide as either purine or pyrimidine, by virtue of coordinates of base atoms. Be definition, purines should only include A/a/G/g/I/i, and pyrimidines C/c/T/t/U/u/P/p. However, no consistency check has been implemented in DSSR until just now.

I first noticed the inconsistency between residue name and atom coordinates for nucleotide A6 on chain U (hereafter referred to as U.A6) in 1pns. The nucleotide has standard name ‘  A’, obviously a purine. However, somehow DSSR classified it as a pyrimidine based on atomic coordinates. Upon further check of the PDB data file, I found the following remarks:

REMARK 470 MISSING ATOM                                                         
REMARK 470 THE FOLLOWING RESIDUES HAVE MISSING ATOMS(M=MODEL NUMBER;            
REMARK 470 RES=RESIDUE NAME; C=CHAIN IDENTIFIER; SSEQ=SEQUENCE NUMBER;          
REMARK 470 I=INSERTION CODE):                                                   
REMARK 470   M RES CSSEQI  ATOMS                                                
REMARK 470       A U   6    N9   C8   N7                                        
REMARK 470       G U   8    N9   C8   N7                                        
REMARK 470       A U  12    N9   C8   N7                                        
REMARK 470       A U  13    N9   C8   N7                                        
REMARK 470       A U  14    N9   C8   N7                                        

The atomic coordinates for U.A6 are as below:

ATOM  34447  P     A U   6      81.861  37.210  78.651  1.00378.87           P  
ATOM  34448  OP1   A U   6      80.631  37.121  77.831  1.00378.87           O  
ATOM  34449  OP2   A U   6      81.665  37.221  80.119  1.00378.87           O  
ATOM  34450  O5'   A U   6      82.707  38.495  78.212  1.00378.87           O  
ATOM  34451  C5'   A U   6      83.948  38.777  78.887  1.00378.87           C  
ATOM  34452  C4'   A U   6      84.600  40.000  78.276  1.00378.87           C  
ATOM  34453  O4'   A U   6      84.975  39.698  76.901  1.00378.87           O  
ATOM  34454  C3'   A U   6      83.714  41.239  78.153  1.00378.87           C  
ATOM  34455  O3'   A U   6      83.654  41.968  79.369  1.00378.87           O  
ATOM  34456  C2'   A U   6      84.403  42.015  77.020  1.00378.87           C  
ATOM  34457  O2'   A U   6      85.564  42.655  77.474  1.00378.87           O  
ATOM  34458  C1'   A U   6      84.834  40.864  76.105  1.00378.87           C  
ATOM  34459  C5    A U   6      82.033  39.296  74.209  1.00378.87           C  
ATOM  34460  C6    A U   6      82.941  39.553  75.166  1.00378.87           C  
ATOM  34461  N6    A U   6      81.170  39.949  72.090  1.00378.87           N  
ATOM  34462  N1    A U   6      83.830  40.588  75.041  1.00378.87           N  
ATOM  34463  C2    A U   6      83.843  41.410  73.939  1.00378.87           C  
ATOM  34464  N3    A U   6      82.899  41.124  72.974  1.00378.87           N  
ATOM  34465  C4    A U   6      81.968  40.108  73.016  1.00378.87           C  

No atom records for N7, C8 and N9. So far, so good. However, surprise came when I visualized U.A6 in Jmol, as shown in the following image. Note here atom N1 is connected to C1’ as in pyrimidines, and N6 is bonded to C4!

The same issue also exists for U.G8 (see figure below), U.A12, U.A13, and U.A14.

It is beyond my imagination to understand why such weird cases exist in the PDB, even given the lousy resolution (8.7 Å) of 1pns.

I recently upgraded my Macs to OS X Mavericks to check if 3DNA/DSSR works in the new operating system. I am glad to report that both run without a hitch, as expected.

Since OS X Mavericks is free from the Mac App Store, it will quickly become the de facto version virtually all Mac users would use. I also noticed that Ruby on Mavericks has been upgraded to ruby 2.0.0p247 (2013-06-27 revision 41674), a major step forward from the now retiring Ruby 1.8.7 distributed in previous versions of Mac OS X.

As a rule, I’d ensure that 3DNA/DSSR executes properly in major releases of the commonly used operating systems — Mac, Windows, and Linux.

While having not used DOS for ages, I am glad to find that the DSSR version compiled for MinGW/MSYS on Windows works perfectly under this operating system (see screenshot below). The DSSR DOS command-line interface functions exactly the same as for Linux, Mac OS X, MinGW/MSYS, and CygWin. Among other possible usages, it allows for batch files to take advantage of DSSR.

Implementing DSSR in strict ANSI C as a self-contained and zero-dependent command-line program pays off enormously: it simplifies code maintenance and ensures that the program is applicable wherever a C compiler exists. The easy web interface to DSSR makes the program universally accessible.

Aside from its extensive functionality for RNA structural analyses, DSSR also introduces a consistent and flexible way to process command-line options. Here, each option can be specified via a --key[=value] pair (or -key[=value] or key[=value]; i.e., two/one/zero preceding dashes are all accepted), key can be in either lower, UPPER or MiXed case, and value is optional for Boolean switches. Furthermore, options can be put in any order; if the same key is repeated more than once, the value specified last overwrites corresponding previous settings.

As always, the rules are best illustrated with concrete examples. Some typical use-cases are given below:

#1 analyze PDB entry '1msy', with default output to stdout
x3dna-dssr --input=1msy.pdb

#2 same as #1, with output directed to file '1msy.out'
x3dna-dssr --input=1msy.pdb --output=1msy.out

#3-6, same as #2
x3dna-dssr --output=1msy.out --input=1msy.pdb
x3dna-dssr --OUTPUT=1msy.out --Input=1msy.pdb
x3dna-dssr -output=1msy.out input=1msy.pdb
x3dna-dssr output=1msy.out --input=1msy.pdb

#7 the value '1ehz.pdb' overwrites '1msy.pdb'
x3dna-dssr --input=1msy.pdb input=1ehz.pdb

#8-12 with the switch --more set to true
x3dna-dssr -input=1msy.pdb --more
x3dna-dssr -input=1msy.pdb --more=true
x3dna-dssr -input=1msy.pdb --more=yes
x3dna-dssr -input=1msy.pdb --more=on
x3dna-dssr -input=1msy.pdb --more=1

#13 same as without specifying --more,
#      or with values set to false/no/0
x3dna-dssr -input=1msy.pdb --more=off

#14 shorthand forms for --input and --output
x3dna-dssr -i=1msy.pdb -o=1msy.out

#15 it can also be more verbose
x3dna-dssr --input-pdb-file=1msy.pdb

#16-18 within a key, separator dash(-) and underscore (_)
#      are treated the same, and can be omitted
x3dna-dssr -i=1msy.pdb -non-pair
x3dna-dssr -i=1msy.pdb -non_pair
x3dna-dssr -i=1msy.pdb -nonpair

By allowing for 2/1/0 dashes to precede each key and a dash/underscore character or none to separate words within the key, DSSR provides users with great flexibility in specifying command-line options to fit into their preferred styles. Not surprisingly, new programs to be added into 3DNA, or the version 3 release of the software will all follow the same convention.

In addition to the five canonical bases (A, C, G, T, and U), nucleic acid structures in the PDB contains numerous modified variants (natural or engineered) in the nucleobase, sugar, or the phosphate. For instance, the 76-nt (nucleotide) long yeast phenylalanine tRNA (1ehz) contains 14 modified bases: 2MG10, H2U16, H2U17, M2G26, OMC32, OMG34, YYG37, PSU39, 5MC40, 7MG46, 5MC49, 5MU54, PSU55, and 1MA58. Among which, the most prevalent and best-known example is pseudouridine. Note that in the PDB, each residue (including modified nt) is named with an up to three-letter identifier, e.g., PSU for pseudouridine. For a comprehensive list (with chemical and structural information) of small molecules, including modified nts, please refer to the Ligand Expo website hosted by the RCSB PDB.

Given the widespread occurrences of modified bases in nucleic acid structures, any practical structural bioinformatics software should be able to treat them effectively, as with the canonical bases. In 3DNA, from the very beginning, modified bases are mapped to standard counterparts, e.g. 5‐iodouracil (5IU) to uracil (U) and 1‐methyladenine (1MA) to adenine (A), allowing for easy analysis of unusual DNA and RNA structures (see the NAR03 reference). Specifically, in the 3DNA distribution the file baselist.dat contains the mappings explicitly.

As of v2.1, 3DNA automatically maps a new modified base not available in the file baselist.dat. Yet, I have continuously updated the list in line with new DNA/RNA entries released by the PDB. The process is automated with a Ruby script which calls find_pair -s on each nucleic-acid-containing structure to output unknown bases. As an extreme, the baselist.dat file below comprises only canonical bases:

  A   A
  C   C
  G   G
  T   T
  U   U
 DA   A
 DC   C
 DG   G
 DT   T

With the above minimum mapping list, running the command find_pair -s on 1ehz.pdb identifies all the 14 modified bases. A sample case for 2MG is shown below:

Match '2MG' to 'g' for residue 2MG   10  on chain A [#10]
    check it & consider to add line '2MG     g' to file <baselist.dat>

By parsing the output of a batch run on all DNA/RNA-containing entries in the PDB as of October 18, 2013, I identified a total of 596 modified bases. The top portion is as below:

02I     a
08Q     c
08T     a
0AD     g
 0C     c
0DC     c
0DG     g
0DT     t
 0G     g
0KL     u
0KX     c
0KZ     t

An explicit list of base mapping makes the correspondence transparent, and helps avoid ambiguous cases as to which canonical base a modified nt matches to. DSSR uses the same list internally. Hopefully, the information would also be useful to other related projects.

Recently I was a bit surprised to find that the methyl group is named differently in the PDB: C7 in DT8 (thymine) of B-DNA 355d, CM5 in 5MC40 (5-methylated C) of tRNA 1ehz, and C5M in 5MU54 (5-methylated U, i.e., T) of the same tRNA 1ehz. See the three figures below for details.

I know that the previously named C5M of thymine in DNA has been renamed C7 as a result of the 2007 remediation effort (PDB v3). However, browsing through the wwPDB Remediation website and reading carefully the article Remediation of the protein data bank archive, I failed to see explanations of the obvious inconsistency of CM5 (5MC40) vs C5M (5MU54) in the nomenclature of the 5-methyl group in the same tRNA entry 1ehz, except for the following note:

As with the Chemical Component Dictionary, names for standard amino acids and nucleotides follow IUPAC recommendations (10) with the exception of the well-established convention for C-terminal atoms OXT and HXT. These nomenclature changes have been applied to standard polymeric chemical components only.

5-methyl is named C7 in DT8 of the DNA entry 355d

5-methyl is named CM5 in 5MC40 of the RNA entry 1ehz

5-methyl is named C5M in 5MU54 of the RNA entry 1ehz

Am I missing something obvious? If you have any further information, please leave a comment. Whatever the case, it helps (at least won’t hurt) to know the naming discrepancy for those who care about the small methyl group in nucleic acid structures.

Recently, I upgraded my local ViennaRNA package installation from v2.0.7 to v2.1.3 on my Mac. Following Quickstart in the INSTALL file, I ran ./configure successfully, but make aborted with error messages. Since I previously had a working copy of the software, it must be configuration issues when I compiled this new version. After a few iterations of checking the error message and reading through the INSTALL file, I came up with the following settings:

./configure --disable-openmp --without-perl
make
sudo make install

Apart from some warning messages, the above make command ran successfully.

This post serves mainly as a note for my own reference. Hopefully, the information may prove useful to others who try to install the versatile ViennaRNA package on a Mac OS X machine.