Sequence-dependent structural properties of B-DNA: what have we learned in 40 years?

Abstract

The structure of B-DNA, the physiological form of the DNA molecule, has been a central topic in biology, chemistry and physics. Far from uniform and rigid, the double helix was revealed as a flexible and structurally polymorphic molecule. Conformational changes that lead to local and global changes in the helix geometry are mediated by a complex choreography of base and backbone rearrangements affecting the ability of the B-DNA to recognize ligands and consequently on its functionality. In this sense, the knowledge obtained from the sequence-dependent structural properties of B-DNA has always been thought crucial to rationalize how ligands and, most notably, proteins recognize B-DNA and modulate its activity, i.e. the structural basis of gene regulation. Honouring the anniversary of the first high-resolution X-ray structure of a B-DNA molecule, in this contribution, we present the most important discoveries of the last 40 years on the sequence-dependent structural and dynamical properties of B-DNA, from the early beginnings to the current frontiers in the field.

The early beginnings

The Drew-Dickerson first B-DNA crystal structure

After the Watson and Crick double helix theoretical model for B-DNA was described (Watson and Crick 1953), the direct evidence for its structure was based on “ideal” representations using standard bond parameters obtained from low-resolution X-ray diffraction data of crystalline fibres (Langridge et al. 1960; Fuller et al. 1965; Arnott and Hukins 1972; Arnott et al. 1980). It took 28 years from Watson–Crick discovery and significant advances in the DNA synthesis field and X-ray diffraction techniques to make possible the first B-DNA crystal structure at atomic resolution (Drew et al. 1981). In 1981, Dickerson’s group described the conformation and dynamics of a complete turn of right-handed B-DNA with the sequence d(CpGpCpGpApApTpTpCpGpCpG) (Fig. 1a). To date, this famous sequence known as the Drew-Dickerson dodecamer (DDD) has been determined experimentally hundreds of times. The non-uniform motion of bases and phosphate groups in the helix, the description of some local helical parameters, torsion angles of the backbone and sugar conformations, led to the foundations 40 years ago (Drew et al. 1981) for the study of the sequence-dependent structural properties of B-DNA.

Fig. 1.

The early beginnings. a Representation of the first B-DNA crystal structure, the Drew-Dickerson dodecamer (DDD),

adapted from Drew et al. 1981. b Cross-chain stacking of purines (vertical dashed lines) and steric clashes (*) could explain the sequence-dependent values adopted by slide, roll and propeller twist parameters, adapted from Calladine et al. 2004. c Prediction of the twist parameter for the DDD by applying Calladine rules, adapted from Dickerson 1983. d Helical parameters approved by the Cambridge convention, adapted from Lu and Olson 2003

Calladine rules and Zhurkin’s contributions

Based on the findings from Dickerson’s group, the Journal of Molecular Biology published in 1982 a letter sent by Calladine (Calladine 1982), suggesting that the specific B conformation observed in the DDD structure was a consequence of steric repulsive forces between purine bases in consecutive basepairs on opposite strands (Fig. 1b). With simple steric considerations and geometric constraints based on a few structural parameters at the base (propeller twist) and basepair (roll and slide) level, a set of rules to describe the 3D structure of B-DNA from sequence was proposed (Calladine 1982; Fratini et al. 1982). One year later, Calladine rules were extended by including additional aspects of the overall helix geometry, sequence-structure relationships, and the effect of hydration on DNA stability (Fig. 1c) (Dickerson 1983; Calladine et al. 2004). In parallel to these initial findings, Zhurkin and colleagues were also pioneer in pointing to differences on DNA’s structure/deformability associated with sequence effects (Zhurkin et al. 1979). In those early times, the studies on the sequence-dependent anisotropic flexibility of B-DNA were even able to explain the structural behaviour of a few selected tetranucleotides (Ulyanov and Zhurkin 1984a, 1984b).

Lavery-Sklenar generalized helical parameters

The increasing need in the community for tools able to standardize the structural analysis of nucleic acids led to the definition of a set of generalized helical parameters and an axis of curvature (Lavery and Sklenar 1988). From the description of the overall DNA conformation to the way of distributing its sequence “irregularities”, Lavery and Sklenar presented a unique algorithm solving the problem of obtaining a rigorous helical description for any given segment. The associated computer program termed Curves (Lavery and Sklenar 1988), nowadays Curves + (Lavery et al. 2009; Blanchet et al. 2011), was the first algorithm of its kind and is still one of the most used tools in the field. A few years later, Lu and Olson published another comprehensive software package termed 3DNA (Lu and Olson 2003).

Cambridge and Tsukuba conventions

One year after the definitions of Lavery and Sklenar were implemented in Curves, an EMBO workshop was held in Cambridge to reach international consensus in key concepts and define a common structural nomenclature for base, basepair and basepair step parameters in the helical space (Fig. 1d) (Dickerson 1989). The standardization of these parameters was revisited in 1999 at the Tsukuba workshop held in Japan (Olson et al. 2001).

The nearest neighbour approximation

DNA sequence-dependent deformability from crystal structures

In a seminal work, Olson and coworkers studied the DNA deformability associated with the identity of adjacent basepairs from a set of 92 DNA–protein complexes resolved by X-ray diffraction (Olson et al. 1998). The behaviour of DNA in basepair steps was interrogated by analysing six helical parameters that affect the planes of successive basepairs: the twist, roll and tilt rotations and the shift, slide and rise translations (Fig. 1d). Their work marked the beginning of several systematic studies focused on uncovering sequence effects in the helical space. At that time, due to the scarcity of structures deposited in the PDB, Olson’s main assumption was that different proteins and different environments impose different forces on the DNA, so if averaged over a large number of DNA–protein complexes and environments, the external-induced distortions would cancel each other and the conformational response of DNA would emerge. In this reduced helical space, the deformation energy of the 10 unique basepair steps was expressed for the first time using a harmonic approximation. Among other findings (Fig. 2a), the pyrimidine-purine (YR) steps emerged as the more flexible ones, acting as “hinges” at the protein-DNA interface. Some key correlations and conformational sub-states interdependences at the dinucleotide level were also described in Olson’s PNAS article, leading a few years later to an initial “theory of sequence-dependent DNA elasticity” (Coleman et al. 2003). Also in 1998 and using a similar approach based on the analysis of X-ray crystal structures, Packer and Hunter provided a comprehensive view of the coupling between the conformational properties of the sugar-phosphate backbone and the basepair step parameters (Packer and Hunter 1998). The knowledge gained was then used to build a simple model to compute potential energy surface maps for dinucleotide and tetranucleotide steps as a function of slide and shift (Packer et al. 2000a, 2000b).

Fig. 2.

The nearest neighbour approximation. a Roll-twist correlation for YR steps as revealed by Olson and coworkers in 1998,

adapted from Olson et al. (1998). b Twist and roll average values for the 10 unique dinucleotides comparing MD simulations and X-ray crystal structures, adapted from Pérez et al. (2008). c Average twist values (black circles) for the 10 unique dinucleotides. The analysis of the 136 unique combinations emerging from considering all nearest neighbour nucleotides has revealed the existence of stable conformational sub-states (black triangles), adapted from Lavery et al. (2010). d A consensus view on the existence of low-twist (LT) and high-twist (HT) structural polymorphism for d(CpG) steps and the twist degree of freedom. The Bayesian information criterion was used for the deconvolution of twist distributions in LT and HT populations, adapted from Dans et al. (2012)

DNA sequence-dependent flexibility from MD simulations

Since the experimental data on sequence-dependent DNA deformability was rather limited 20 years ago, Lankas, Pérez, Fujii and their collaborators were among the first to investigate the sequence-dependent flexibility of DNA using molecular dynamics (MD) simulations (Lankas et al. 2000, 2003; Fujii et al. 2007; Pérez et al. 2008). Inspired by Olson and Hunter approach to analyse the helical space of DNA, a library of sequences containing all unique basepair steps was designed and simulated. Significant sequence-dependence on the torsional stiffness and stretching of DNA segments was observed (Lankas et al. 2000, 2003), leading nevertheless to a simpler description of the conformational landscape than the one previously reported using crystallographic data (Olson et al. 1998), all assuming a rigid basepair approximation with two, uncoupled, nearest neighbour interactions. In parallel, and rooted in the same ideas, Gonzalez and Maddocks introduced a rigorous method to compute expectations from MD data for models with rigid units while properly accounting for the statistical mechanics of the rotational degrees of freedom (Gonzalez and Maddocks 2001). A few years later, these ideas were implemented by Lankas and coworkers (Lankas et al. 2009), suggesting that a rigid base approximation, with its associated five nearest neighbour interactions, was a remarkably good fit to the MD data, thereby providing detailed, quantitative support for the earlier ideas of Calladine on base interactions. The quality of the fit of the rigid base, nearest neighbour approximation to MD data, in turn lead to the predictive, sequence-dependent, rigid base, coarse-grained cgDNA model which can efficiently predict Gaussian statistics (i.e. first and second moments) for double-stranded B-DNA fragments of arbitrary length and sequence (Gonzalez et al. 2013; Petkevičiūtė et al. 2014). In 2007, Sarai’s group was among the first to provide a global view of the conformational similarities of all steps in their tetranucleotide environment (Fujii et al. 2007). One year later, Orozco’s group published a comparative study showing that MD simulations performed with the Parmbsc0 force field (Pérez et al. 2007) were able to predict the dynamic and sequence-dependent physical properties of DNA with an accuracy, in terms of local and global descriptors, reasonably close to that obtained with gold-standard NMR and X-ray experiments (Pérez et al. 2008), providing a first consensus view between theory and experiments (Fig. 2b).

The Ascona B-DNA Consortium (ABC)

In June 2001, during an informal discussion in a conference on Atomistic to Continuum Models for Long Molecules organized by Maddocks and held in Ascona, Switzerland, it began an initiative that involved 17 researchers from nine independent laboratories (Beveridge et al. 2012). The original objective of the ABC project was to produce a database of state-of-the-art MD of DNA in which all the tetranucleotides were represented, developing standards and simulation protocols and obtaining information for the comprehensive study and improved understanding of basepair sequence effects on structure and dynamics. This collaborative and massive simulation effort allowed for a systematic view, highlighting the sequence-induced effects produced by the nearest neighbours (NN) of the d(CpG) basepair step (Beveridge et al. 2004) and later of the 10 unique dinucleotides (Lavery et al. 2010). Using the trajectories made available by the ABC in 2010, a rigid basepair coarse-grained model derived elastic force constants to build an interactive chromatin modelling web server (Stolz and Bishop 2010), while other colleagues focused on the detailed analysis of the sequence-dependent solvation and ion atmosphere of B-DNA (Dixit et al. 2012).

Challenges to the NN approximation

The ABC simulations, based on a library of 39 duplexes containing multiple copies of the 136 unique tetranucleotides, revealed that some helical parameters at the basepair step level were greatly affected by their nearest neighbours and that certain of them did not exhibit a normal distribution (which is an evidence for the existence of a unique stable conformational sub-state for that specific degree of freedom) but rather showed a bimodal behaviour (Fig. 2c) (Lavery et al. 2010). Furthermore, this bimodality seemed to be intrinsic to some dinucleotide steps in certain tetranucleotide environments and was particularly notable in the case of d(CpG) steps, where this structural polymorphism emerged from both theoretical and experimental evidence (Fig. 2d) (Dans et al. 2012). These results established the limitations of Olson’s original approach proposed more than 10 years earlier based on dinucleotides and even challenged the current harmonic-based NN models by introducing the existence of bimodal degrees of freedom.

Structural polymorphisms at the base and backbone level

Experimental evidence from high-resolution X-ray structures

It was soon warned that the experimental evidence supporting the existence of structural polymorphisms had been obtained indirectly, by analysing DNA structures deposited in the PDB (Dans et al. 2012; Pérez et al. 2012). This bioinformatics approach, despite being valid, did not prove the simultaneous existence of the two detected sub-states nor the reversible transitions observed between them in the simulations. In this context, the high-resolution crystal structure published by Hud and Williams groups from Georgia Tech in 2012 (that same year, PDB id 3GGB) was particularly relevant (Maehigashi et al. 2012). It was the first observation by X-ray diffraction of full positional heterogeneity in DNA at the base level and, hence, the first direct evidence supporting previous theoretical findings.

The paradigmatic d(CpG) step

A detailed study of the twist degree of freedom for the most polymorphic step, d(CpG), allowed for an in-depth understanding of the structural transitions coupled with the observed “low twist” (LT) and a “high twist” (HT) sub-states, and the dramatic influence exerted by the nearest neighbours (Dans et al. 2014). It was shown that the twist polymorphism was coupled with transitions at the backbone level leading to BI/BII conformations at the 3’-side junction of the d(CpG) step, where the canonical HT sub-state is associated with BI, while the LT state correlates with BII. The BI/BII sub-states were defined following the established conventions based on the value of ε − ζ torsions (Fig. 3a) (Hartmann et al. 1993) but measured at the 3’-side of the d(CpG) step, i.e. in the next backbone junction on each strand (Fig. 3b). In this context, RCGY flanking sequences tended to favour almost exclusively the canonical conformation (HT/BI), YCGR mostly favoured the non-canonical sub-state (LT/BII), while RCGR sequences could sample both. The spontaneous twist polymorphism of the d(CpG) step also provided some clues about its extreme prevalence in intercalation complexes (Boer et al. 2009; Dans et al. 2014).

Fig. 3.

Definition of backbone sub-states and CH–-O interactions. a BI/BII sub-states are defined according to the established conventions based on the value of ε − ζ torsions. b HT/LT transitions in YR steps (step i) are coupled to backbone transitions in the following 3’-junction (step i + 1), denoted BI_3’/BII_3’. c Formation of the C8H8–-O3’ hydrogen bond in RR steps in a BII conformation. d Formation of the C6H6–-O3’ hydrogen bond in RY steps in a BII conformation

The significant stabilization of the LT/BII sub-state observed for YCGR tetranucleotides was connected to the capacity of these sequences to form non-conventional hydrogen bonds (Hbond) at the 3’-sides (on both strands) of the d(CpG) step (Wahl and Sundaralingam 1997). These Hbonds, which are established by the interaction of the C8H8 group of the R base and the O3* atom of the backbone at the 3’-side (i.e. at the level of the following d(GpR) step), were confirmed for all d(RpR) tetranucleotides (Pasi et al. 2014) and later extended to all d(RpY) steps between the same oxygen atom in the backbone and the corresponding C6H6 group of pyrimidines (Balaceanu et al. 2017). Another key finding was the relationship determined between these coupled sub-states and the sequence-dependent dynamic presence of cations in the minor groove of the d(CpG) step. The presence of Na⁺/K⁺ was exclusively observed in the minor groove during the LT/BII conformation, following the same sequence-dependent trends described and, thus, reaching the highest concentration for the most polymorphic tetranucleotide: TCGA. This complex choreography of conformational changes was shown to be connected by a cause-and-effect relationship, where the entrance of cations in the minor groove of the d(CpG) step triggers the transition from HT/BI/no-CHO Hbond to LT/BII/CHO Hbond in a few picoseconds (Fig. 4a) (Dans et al. 2014). A largely partial but direct experimental validation of the unequivocal relationship between the backbone polymorphism and the interaction with cations was provided once again by high-resolution crystals of B-DNA (PDB id 1D8G and 1EN9) (Kielkopf et al. 2000; Chiu and Dickerson, 2000; Hud and Engelhart 2009). Two years later, this molecular choreography was described again for the GCGA tetranucleotide, by analysing a multi-microsecond long simulation of the DDD (Dans et al. 2016) performed with the latest Parmbsc1 force field for DNA (Ivani et al. 2015). The corrections introduced in the Parmbsc1 force field led to the first sequence-dependent consensus view between MD simulations (Balaceanu et al. 2017) and a vast amount of information on BI/BII structural heterogeneity that was obtained using NMR (nuclear magnetic resonance) techniques. Hartmann’s group has been particularly active in this field, providing a sequence-dependent view of BI/BII at the dinucleotide and even tetranucleotide level (Heddi et al. 2006, 2008, 2010; Abi-Ghanem et al. 2010) but also putting into perspective NMR results with values obtained from crystal structures or MD simulations (Imeddourene et al. 2015).

Fig. 4.

Structural polymorphisms at the nearest neighbour level and beyond. a Conformational changes triggered by the entrance of cations in the minor groove of the d(CpG) step that leads to the HT/BI ↔ LT/BII transition,

adapted from Dans et al. (2014). b ABC representation of the full polymorphic helical space obtained from the analysis of MD simulations of all the unique 136 tetranucleotides. All unique tetranucleotides are grouped according to the nature (R = purine, Y = pyrimidine) of the central bps. BI_3’ and BII_3’ refers to the presence of a BI/BII sub-state at the 3’ junction of a given YR step (see Fig. 3), adapted from Pasi et al. (2014). c Patterns and correlations between helical parameters and backbone sub-states extracted from MD, enabled the formulation of extended Calladine-Dickerson rules. The rules allow for the structural prediction of complex conformational sub-states that could occur simultaneously for a given tetranucleotide, as observed experimentally, adapted from Dans et al. (2019). d Multi-states observed for d(TpA) and the BI (CH-O broken)/BII (CH-O formed) correlation between consecutive basepair steps. Adapted from Balaceanu et al. 2019

A systematic study of tetranucleotide sequence effects

In 2014, after a new ABC collaborative round termed µABC, the detailed study of the complete conformational space at the nearest neighbours level including all structural polymorphisms at the backbone, sugar and base level was published (Pasi et al. 2014). The co-existence of more than one stable conformational sub-state was confirmed in RR steps for shift and twist degrees of freedom, while YR steps displayed bimodal distributions for slide and twist. RY steps appeared to be unimodal for all helical parameters, adopting, in the case of the three polymorphic parameters, typical canonical values (Fig. 4b). The relationship between HT/LT sub-states and BI/BII transitions at the backbone level was also confirmed for all tetranucleotide contexts, highlighting complex sequence-dependent patterns of correlated and anti-correlated movements between base stacking and the backbone conformations, in perfect agreement with early surveys of high-resolution X-ray structures and solid-state ³¹P-NMR experiments (Gorin et al. 1995; van Dam and Levitt 2000; Tolstorukov et al. 2007; Olson and Zhurkin 2011). The coupling between backbone conformations and LT ↔ HT transitions was revisited 1 year later by Zgarbova and coworkers (Zgarbova et al. 2017) using a library of longer B-DNA oligomers and the refined OL15 force field (Zgarbova et al. 2015). All six nearest phosphates were found to influence the twist of the central step, and when an LT/BII sub-state emerges at a certain place, it generates twist changes in its neighbourhoods that tend to compensate locally the produced under-twisting. All this conformational information extracted from MD was integrated along with X-ray data and then used to predict, based on DNA shape features, binding specificities of transcription factors at the genome-scale (Li et al. 2017). Other colleagues used the µABC trajectories to analyse ion distributions around DNA (Pasi et al. 2015) using an innovative curvilinear helicoidal coordinate (Lavery et al. 2014; Dans et al. 2014).

One key aspect of these transitions is the timescale on which they take place. B-DNA breathing (fast, spontaneous and reversible opening/closing of WC Hbonds), transitions between backbone sub-states, sugar re-puckering, moderate bending and sampling of a wide variety of helical values are fast motions occurring clearly below the µs timescale (Dans et al. 2014; Galindo-Murillo et al. 2014). Beyond the μs timescale, with the exception of terminal basepair fraying (opening of the terminal WC basepairs) and rare sequence-specific kink events (Dans et al. 2019), these faster movements are averaged out, and no other motions are actually expected for “naked” B-DNA up to the ms timescale (Galindo-Murillo et al. 2014).

Extending Calladine-Dickerson rules

Integrating, unifying and putting in perspective all findings of the last decade made possible for a new and comprehensive extension of Calladine-Dickerson rules (Dans et al. 2019). By defining a small library of B-DNA sequences, termed miniABC, all tetranucleotide contexts were analysed in different conditions employing microsecond-long MD simulations with Parmbsc1. A systematic analysis of 16 helical parameters and the associated backbone/sugar torsional changes was conducted, which included the first prediction of anharmonicity based on sequence context (Fig. 4c). These extended rules were implemented in a web server that predicts the average conformation of any B-DNA sequence, highlighting those steps where structural polymorphisms could occur (Fig. 4c). The knowledge gained and the extensive volume of raw trajectories made available to the community were then used to develop state-of-the-art coarse-grained models of B-DNA based on rigid bases (De Bruin and Maddocks 2018), rigid basepairs coupled to a multi-modal Hamiltonian and a Metropolis Monte Carlo algorithm (Walther et al. 2020) and a multiscale model of chromatin describing B-DNA at the basepair level and proteins at single amino acid resolution (Farr et al. 2021).

Beyond nearest neighbours?

As discussed above, the structural and dynamical study of all the unique tetranucleotides requires, to reach a complete view, the analysis of 136 sequence combinations. Pushing the conformational analysis of B-DNA statistical mechanical properties to the next level would imply the systematic study of 2,080 unique hexanucleotide sequences, which could be used to detect specific next-to-nearest neighbour (NNN) sequence effects. In the same way, the analysis of possible “long-range” structural correlations at the octanucleotide level would require the simulation of 32,826 unique sequence combinations. The huge number of sequences that must be considered to move the paradigm to the next stage can hardly be covered by a single research group and is even challenging for a whole consortium. Nevertheless, a partial view can be obtained by carefully selecting a given sub-set of representative sequences from the complete sequence space. Using this approach, Balaceanu and coworkers studied using MD simulations the NNN sequence effects affecting the d(TpA) step when embedded in an NCTAGN environment detecting the existence of multiple simultaneous sub-states (Fig. 4d) (Balaceanu et al. 2019). The results seemed to confirm the existence of complex and correlated patterns through consecutive steps at the NNN level and even beyond (Fig. 4d), capable to modulate on both strands the relative populations of BI/BII sub-states observed at the central d(TpA). This “proof-of-principle” study showed that there is indeed interesting “long-range” structural and dynamical correlations to be discovered, therefore, making the grand challenge project to explore it all a worthwhile endeavour.

Closing remarks

Forty years after the first B-DNA crystal structure and 20 years since the beginning of the Ascona B-DNA Consortium seem to be the perfect time to celebrate and embark on the journey through the major discoveries on the sequence-dependent structural effects in B-DNA that we covered in this review. From an “ideal” B-DNA model to the description of subtle conformational sub-states at the base, sugar and backbone level, the gained knowledge has influenced several fields. The new initiative from the ABC, aimed to unravel the sequence-dependent physical properties of B-DNA at the complete NNN level (https://bit.ly/3ywgGV6) and the work of many other colleagues from the field, herald new and exciting chapters in the days to come.

Funding

This work was funded by the Agencia Nacional de Investigación e Innovación (ANII, Uruguay) [FCE_3_2018_1_748945, to PDD], the Comisión Sectorial de Investigación Científica (CSIC, UdelaR, Uruguay) and the Programa de Desarrollo de las Ciencias Básicas (PEDECIBA, UdelaR, Uruguay). This work was also supported by FOCEM (Fondo para la Convergencia Estructural del Mercosur) [COF 03/11 to Institut Pasteur de Montevideo] and the Swiss National Science Foundation Grant [200020-18218, to JHM]. VC and PDD are SNI (Sistema Nacional de Investigadores, ANII, Uruguay) and PEDECIBA Química researchers.

Declarations

Conflict of interest

The authors declare no competing interests.