A protein-DNA docking benchmark.

We present a protein-DNA docking benchmark containing 47 unbound-unbound test cases of which 13 are classified as easy, 22 as intermediate and 12 as difficult cases. The latter shows considerable structural rearrangement upon complex formation. DNA-specific modifications such as flipped out bases and base modifications are included. The benchmark covers all major groups of DNA-binding proteins according to the classification of Luscombe et al., except for the zipper-type group. The variety in test cases make this non-redundant benchmark a useful tool for comparison and development of protein-DNA docking methods. The benchmark is freely available as download from the internet.

INTRODUCTION

Biomolecular docking has become a mature discipline within structural biology (1). Docking aims at predicting the structure of a complex given the 3D structures of its components. The field of protein–protein docking in particular has seen extensive progress over the last decade as witnessed by recent CAPRI (Critical Assessment of Predicted Interactions) results, a community-wide blind docking experiment (2). For protein–DNA docking, however, progress lags behind. The scarcity of information for a proper identification of interaction surfaces on DNA and its inherent flexibility have hampered the development of effective docking methods. The field of protein–DNA docking is, however, receiving increased attention and efforts are put into the development of docking methods that address the above mentioned limitations (3). Considering the importance of biomolecular interactions in system biology, gaining insight into the biochemistry of recognition and gene expression is highly relevant (4). New developments in protein–DNA docking approaches are therefore expected.

A set of well-defined test cases that form a common ground for validating and comparing the different docking methods would facilitate the development of effective protein–DNA docking methods. Such a benchmark should contain the native structures of both protein and DNA in their unbound form together with the reference structure of the complex.

We have constructed a benchmark of 47 protein–DNA test cases in a similar manner as has been done for protein–protein docking (5). The benchmark covers all major groups of protein–DNA complexes according to the classification proposed by Luscombe et al. (6) except for the zipper-type group. It contains a variety of challenging systems in terms of size of the interaction interface, number of individual components present in the complex and conformational changes that the unbound components undergo upon complex formation. Its diversity makes it a comparison tool for different docking methods as their performance may vary depending on the type of complexes. This benchmark should benefit the entire docking community and offer a starting-point for the improvement of various algorithms.

MATERIALS AND METHODS

RCSB Protein Data Bank (PDB) query

A non-redundant benchmark was generated from structures deposited in the RCSB PDB (7). The PDB (as of September 2007) was queried for all entries containing X-ray crystallographic structures with a resolution better than 3.0 Å containing both protein and DNA. Complexes containing DNA structures with a sequence length smaller than 8 bp and protein structures containing mutations in the core and or interface region were removed.

For the resulting complexes, the PDB was queried for unbound protein entries. Structures resolved using NMR or X-ray crystallography with a resolution better than 3.0 Å were retrieved. Structures with a sequence similarity larger than or equal to 90% were removed. Structures were regarded as redundant if the raw alignment score is positive, >80% of their sequences are aligned and >60% of the sequences are identical. Sequence alignments were performed using the Needleman–Wunsch algorithm as implemented in the LSQMAN software package (8) with a gap penalty of 5.

Generation of unbound DNA models

Models for unbound DNA were generated using the DNA analysis and rebuilding program 3DNA (9) with the base-pair sequence of the DNA in the reference complex. The models were generated in canonical B-DNA conformation (fiber model 4) using the nucleotide building blocks as determined in the fiber diffraction studies of Chandrasekaran and Arnott (10). Structures with overhanging base-pairs were converted to all-paired structures by adding their Watson–Crick counterparts.

Structure post-processing

The residue numbering of the bound and unbound components was matched to allow for easy comparison. The DNA was assigned one chain identifier and renumbered. Structures of unbound proteins that contain more than one chain were assigned a single chain identifier instead of being separated into their individual components; residues were renumbered to avoid overlap in numbering. Atom and residue names were matched to the topallhdg5.3.pro (11) and dna-rna_allatom.top topology files (12) naming for direct use in HADDOCK (13).

Analysis

The size of the interaction interface between protein and DNA is expressed in terms of the buried surface area (BSA, Table 1) of the DNA in the complex. The BSA was calculated using NACCESS (Hubbard, S. J., Thornton, J. M. 1993) with a probe radius of 1.4 Å. The conformational changes between the unbound and the bound states are expressed in terms of the root mean square deviation (RMSD) calculated using ProFit (Martin, A.C.R., http://www.bioinf.org.uk/software/profit/). These were calculated in three different ways:

Table 1.

The protein–DNA benchmark

Complex		Protein		DNA			RMSD
PDB ida	Cat.b	PDB ida	Description	Sequence 5′-3′c	Nr.d	BSAe	Inter.f	DNAg	Proth
‘Easy’ targets
2c5r	1	2bnkX	Phage PHI29 replication organizer protein P16.7	TCCACCGG	4	402	0.49	0.49	0.82
1pt3 (A:C:D)	8	1m08X	Col-E7 nuclease domain	GCGATCGC	2	730	1.35	2.09	1.36
1mnn	1	1mn4X	Sporulation specific transcription factor NDT80	TGCGACACAAAAACT	2	1292	1.48	1.81	0.83
1fok	1	2fokX	Restriction endonuclease FOKI	TCGGATGATAACGCTAGTCAT	2	1920	1.53	2.51	1.09
1ksy (A:C:D:F)	4	1f08X	Papillomavirus replication initiation domain E-1	ATAATTGTTGTCAACAATTAT	3	1020	1.58	2.56	0.52
3cro	1	1zugN	Phage 434 CRO	AAGTACAAACTTTCTTGTAT	3	1473	1.58	2.66	1.17
1emh	8	1akzX	Human uracil-DNA glucosylase	TGT(P2U)ATCTTT	2	869	1.62	4.53	1.46
1h9t	1	1e2xX	FADR, fatty acid responsive transcription factor	CATCTGGTACGACCAGATC	3	1622	1.68	3.88	0.77
1tro (A:C:I:J)	1	3wrpX	TRP repressor	TGTACTAGTTAACTAGTACA	3	1540	1.70	3.08	1.42
1by4 (A:B:E:F)	2	1rxrN	Retinoid X receptor DNA binding domain	TAGGTCAAAGGTCAG	3	1480	1.77	1.46	2.23
1hjc (A:B:C)	5	1eanX	RUNX1 runt domain	GAACTCTGTGGTTGCGG	2	634	1.80	2.88	0.97
1diz (A:E:F)	8	1mpgX	E. coli 3-methyladenine DNA glycosylase II	TGACATGA(NRI)TGCCT	2	805	1.82	5.80	0.46
1rpe	1	1r63N	Phage 434 repressor	ACAAACAAGATACATTGTATA	3	1430	1.87	2.97	0.94
‘Intermediate’ targets
1vrr	8	1sdoX	Restriction endonuclease BSTYI	TTATAGATCTATAA	3	2098	2.08	2.11	2.22
1f4k	1	1bm9X	Replication terminator protein	CTATGAACATAATGTTCATAG	3	1741	2.26	1.94	2.29
1k79 (A:B:C)	1	1gvjX	ETS-1 DNA binding and autoinhibitory domain	TAGTGCCGGAAATGTG	2	912	2.37	3.82	0.80
1kc6 (A:B:E:F)	8	2audX	Restriction endonuclease HINCII	CCGGTCGACCGG	3	2658	2.38	4.67	1.38
1ea4 (D:E:F:G:W:X)	6	2cpgX	Transcription repressor COPG	TAACCGTGCACTCAATGCAATC	3	1473	2.43	4.48	0.64
1z63 (A:C:D)	8	1z6aX	Sulfolobus solfataricus SWI2/SNF2 ATPase core domain	ATTGCCGAAGACGAAAAAAA	2	603	2.51	2.74	2.27
1r4o	2	1gdcN	Glucocorticoid receptor	CCAGAACATCGATGTTCTGT	3	1401	2.61	3.05	1.91
1azp	6	1sapN	Hyperthermophile chromosomal protein SAC7D	GCGATCGC	2	778	2.70	3.77	2.76
1w0t	1	1ba5N	HTRF1 DNA-binding domain	CTGTTAGGGTTAGGGTTAGA	3	1545	2.78	3.20	2.47
1cma	6	1mjkX	Methionine repressor	TTAGACGTCT	2	775	2.81	2.60	2.05
1jj4	4	1f9fX	Papillomavirus type 18 E2	CAACCGAATTCGGTTG	2	1169	2.83	3.32	2.25
1vas	8	1eniX	T4 pyrimidine dimer specific excision repair	ATCGCGTTGCGCT	2	1445	3.04	6.99	1.42
4ktq	8	1ktqX	DNA polymerase I	GACCACGGCGC(DOC)	2	1685	3.23	3.64	1.97
1z9c (A:C:D)	1	1z91X	Organic hydroperoxide resistence transcription regulator	TACAATTTAATTGTATACAATT TAATTGTA	3	2107	3.24	4.26	4.18
1ddn	1	2tdxX	Diphtheria TOX repressor	ATATAATTAGGATAGCTTTACC TAATTATTTTAA	5	2877	3.26	7.25	0.50
2irf	1	1irgN	Interferon Regulatory Factor 2	AAGTGAAAGUGA	2	898	3.35	2.23	3.83
1jt0	1	1jusX	Multidrug binding transcription factor QACR	CTTATAGACCGATCGATCGG TCTATAAG	2	2484	3.49	4.58	3.53
1g9z	8	2o7mX	I-CreI endonuclease	GCAAAACGTCGTGAGACAGTTTCG	2	3255	3.67	5.02	4.21
1a73	8	1evxX	Intron-encoded homing endonuclease I-PPOI	TTGACTCTCTTAAGAGAGTCA	2	2076	4.26	8.22	1.20
2fio	4	2fibX	Phage PHI29 transcription regulator P4	AAAAACGTCAACATTTTATA AAAAAGTCTTGCAAAAAGT	2	1114	4.41	8.03	0.67
1qne (A:C:D)	5	1vokX	Adenovirus major late promotor TBP	GCTATAAAAGGGCA	2	1487	4.57	8.54	0.89
1zs4	1	1zpqX	Phage lambda CII	CCTCGTTGCGTTTGTTTGCACGAAT	2	1358	4.71	2.97	3.77
‘Difficult’ targets
1qrv	4	1hmaN	High mobility group protein D	GCGATATCGC	3	1204	5.19	7.68	3.91
1o3t	1	1g6nX	CAP-CAMP	GCTTTTTACGCTAGATCTA GCGTAAAAAGCGC	2	1277	5.20	10.6	2.55
1b3t	4	1vhiX	Epstein-Barr virus nuclear antigen-1	GGAAGCATATGCTTCCC	2	2627	5.32	3.91	3.53
3bam	8	1bamX	Restriction endonuclease BAMHI	TATGGATCCATA	3	2208	5.55	2.19	4.50
1rva	8	1rveX	Eco RV endonuclease	AAAGATATCTTT	2	2350	5.68	9.78	3.88
1zme	2	1ajyN	Proline utilization transcription activator PUT3	ACGGGAAGCCAACTCCGT	2	1362	5.76	4.68	8.64
1dfm	8	1es8X	Restriction endonuclease BGLII	TATTATAGATCTATAAAT	3	2735	6.31	3.04	4.68
1bdt	6	1arqN	Phage P22 Arc gene regulating protein	TATAGTAGAGTGCTTCTATCATT	3	2109	6.45	4.90	5.20
7mht	8	2hmyX	HHAI methyltransferase	GTCAGCGCATGG	2	1613	6.71	2.55	3.84
2fl3	8	1ynmX	Restriction endonuclease HINP1I	CCAGCGCTGG	2	1670	6.71	2.95	4.37
1eyu	8	1pvuX	PVUII endonuclease	TGACCAGCTGGTCA	2	2068	6.82	4.49	6.36
2oaa	8	2oa9X	Restriction endonuclease MVAI	GGTACCTGGATG	2	2009	8.95	8.15	8.02

aThe RCSB PDB accession number for the structures used. Specific chains are in parenthesis. Structures for the unbound protein were either solved by X-ray crystallography (X) or NMR spectroscopy (N).

bThe classification of the protein–DNA complexes in eight different groups according to the scheme of Luscombe et al. (6).

cThe base sequence of the DNA in the bound complex also used for generating the unbound DNA structure. Some sequences contain modified bases. These are: DOC (2′,3′-dideoxycytidine-5′-monophosphate), NRI (phosphoric acid mono-(4-hydroxy-pyrrolidin-3-ylmethyl) ester) and P2U (2′-deoxy-pseudouridine-5′monophosphate).

dThe number of individual biomolecules that need to be docked to reconstruct the complex.

eBuried surface area of the DNA upon complex formation in Å2.

fThe RMSD (Å) from the bound form calculated over the interface Cα and phosphate atoms of the unbound protein structure after superposition onto the reference complex.

gThe RMSD (Å) from the bound form calculated over all phosphate atoms of the unbound DNA after superposition onto the reference complex.

hThe RMSD (Å) from the bound form calculated over Cα atoms of the unbound protein after superposition onto the reference complex.

Conformational change of the protein–DNA interface was calculated by superimposition of all Cα and phosphate atoms at the interface. Residues belonging to the interface are identified as those having atoms within 5.0 Å intermolecular distance of one another (RMSD Inter., Table 1). The interface RMSD values were used to classify the test cases as ‘easy’, ‘intermediate’ or ‘difficult’ (see below).

As the conformational change in the DNA tends to affect the complete molecule, the RMSD of the DNA was calculated by superimposition of all phosphate atoms (RMSD DNA, Table 1).

Conformational changes in the protein, such as global domain reorientations and flexible segments not located at the interface are represented by means of the RMSD calculated over all Cα atoms of the protein (RMSD Prot, Table 1).

COMPOSITION OF THE BENCHMARK

The protein–DNA benchmark version 1.0 (Table 1) contains 47 test cases. For all test cases, the unbound structures of both protein and DNA are available. In addition, the reference complexes have been separated into their DNA and protein bound forms. This should allow to evaluate the performance of a docking method for bound–bound, bound–unbound and unbound–unbound cases. Although the reference structure is always from X-ray crystallography, the unbound proteins contain both solution NMR and X-ray structures. The use of an ensemble of NMR structures as starting point for the docking provides an easy way for various docking algorithms to sample additional conformational space. The benchmark contains members of all major structural groups described by Luscombe et al. (6) apart from the zipper-type group. These are: 16 helix–turn–helix (group 1), three zinc-coordinating (group 2), five other α-helix (group 4), two β-sheet (group 5), four β-hairpin/ribbon (group 6) and 17 enzyme (group 8) complexes.

Each test case in the benchmark poses its own challenges for a docking algorithm. A common theme throughout the benchmark is ‘conformational changes’ either in the protein, the DNA or both. This benchmark differs from its protein–protein counterpart by the omnipresence of conformation changes. To provide some structure in the test cases, we classified them as ‘easy’, ‘intermediate’ or ‘difficult’. This classification is based on the interface RMSD values between the bound and unbound components of the complex:

‘easy’ test case: interface RMSD between 0.0 Å and 2.0 Å

‘intermediate’ test case: interface RMSD between 2.0 Å and 5.0 Å

‘difficult’ test case: interface RMSD above 5.0 Å.

An ‘easy’ test case

The individual components from this group of complexes do not change significantly the conformation of their interface upon binding. Conformational changes at the interface of the protein are mostly brought about by small flexible loop rearrangements. This does not mean that the components can always be regarded as rigid. Conformational changes at the interface of the DNA often cause the DNA to bend and twist in the interface region (see DNA RMSD values in Table 1). A representative example from this group is the Papillomavirus replication initiation domain E-1 (PDB entry 1ksy, Figure 1A).

Figure 1.

Illustration of ‘easy’ (interface RMSD < 2.0 Å), ‘intermediate’ (2.0 Å ≤ interface RMSD < 5.0 Å) and ‘difficult’ (interface RMSD ≥ 5.0 Å) test cases from the protein–DNA benchmark. ‘Easy’ test case: the Papillomavirus replication initiation domain E-1 (PDB id 1ksy) (interface RMSD = 1.6 Å) (A). ‘Intermediate’ test case: the intron-encoded homing endonuclease I-PPOI complex (PDB id 1a73) (interface RMSD = 4.3 Å) (B). ‘Difficult’ test cases: the proline utilization transcription activator (PDB id 1zme) (interface RMSD = 5.8 Å) (C) and the PVUII endonuclease complex (PDB id 1eyu) (interface RMSD = 6.8 Å) (D). The bound form of the complex is shown in yellow and the unbound protein in blue. The bound- and canonical B-form DNA structures are shown as insets to highlight the conformational changes in the DNA.

An ‘intermediate’ test case

Unbound components of this group undergo more pronounced structural rearrangements in their interface upon complex formation. The type of conformational changes involves global and local domain rearrangements in the protein and global conformational change in the DNA. An example is the intron-encoded homing endonuclease I-PPOI complex (PDB entry 1a73, Figure 1B), the protein shows little conformational change upon binding but the DNA is heavily kinked in its centre.

A ‘difficult’ test case

In the difficult cases, the extent of structural rearrangement upon complex formation increases even further. In addition to the conformational changes occurring in the ‘intermediate’ test cases, the ‘difficult’ group contains complexes with features like structural transitions and major domain reorientations in the protein. An example is the proline utilization transcription activator (PDB entry 1zme, Figure 1C), a protein that has two DNA interaction domains linked together by a long highly flexible loop; the dimerization interface connecting the two DNA interaction domains show a loop to sheet transition upon DNA binding. In the PVUII endonuclease complex (PDB entry 1eyu, Figure 1D), the individual protein chains do not show much conformational changes but a hinge point connecting them facilitates a ‘clamping’ motion upon binding. This results in a large RMSD between bound and unbound structures. This is an example of global domain motions upon binding.

The benchmark also contains several structures with special features such as strand breaks (PDB entries 1g9z, 1o3t and 3bam) and flipped out bases in the DNA (PDB entries 1diz, 1emh, 1vas and 7mht).

We constructed this benchmark as a test base to stimulate developments in the field of protein–DNA docking and will use it in particular for further developing our own protein–DNA docking approach (3). Ideally, the classification of ‘easy’, ‘intermediate’ or ‘difficult’ could have been based on docking results; at this stage, however, we chose to purely base it on conformational changes as measured by the RMSDs between bound and unbound form. Basing the classification on HADDOCK results would have introduced a bias not only toward the amount of conformational changes, but also toward our ability to predict protein–DNA interfaces since HADDOCK requires some kind of input to drive the docking process. We will of course proceed with evaluating our performance on this benchmark, but this is outside the scope of this article.

In conclusion, allowing for structural rearrangements in both protein and DNA during docking, while maintaining the helical character of DNA is a major challenge in protein–DNA docking. The large variety of protein–DNA complexes in the benchmark should provide a valuable test set to evaluate and improve docking algorithms. Version 1.0 of the benchmark is available from the web site: http://haddock.chem.uu.nl/dna/benchmark.html