DNAshape: a method for the high-throughput prediction of DNA structural features on a genomic scale

Tianyin Zhou, Lin Yang, Yan Lu, Iris Dror, Ana Carolina Dantas Machado, Tahereh Ghane, Rosa Di Felice, Remo RohsNucleic Acids Res 2013 · DOI

TL;DR

This paper introduces DNAshape, a novel web server and method for predicting DNA's three-dimensional structural features (like minor groove width and helix twist) at a genomic scale. This high-throughput approach integrates sequence and shape analyses, enabling a deeper understanding of how DNA structure influences protein binding and other biological processes.

What Problem Does This Paper Address?

Understanding the relationship between DNA sequence and its 3D structure (DNA shape) is crucial for deciphering protein-DNA recognition. While experimental methods like X-ray crystallography and NMR spectroscopy provide detailed structural information, they are limited to specific DNA sequences and are not feasible for analyzing entire genomes. Previous computational approaches, such as all-atom molecular dynamics simulations, are too computationally expensive for high-throughput genomic analysis. This leaves a significant gap in our ability to study DNA shape across vast genomic regions and understand its role in biological regulation.

Methods

The core of DNAshape is a "sliding-pentamer model." This approach mines structural information previously obtained from extensive Monte Carlo (MC) simulations of numerous DNA fragments. The method works by analyzing DNA structures in a sliding-window fashion, considering pentanucleotides (five base pairs) as the context for predicting structural features at each base pair or base pair step.

Here's a breakdown of the key methodological aspects:

  • Monte Carlo (MC) Simulations: The foundation of the method relies on pre-computed MC simulations of DNA fragments. These simulations use collective and internal variables, an analytic chain closure, and an implicit solvent model to efficiently sample DNA conformations.
  • Sliding-Pentamer Window: The MC simulation data is then processed using a sliding-pentamer window. For each unique pentanucleotide sequence, the average structural features (minor groove width, roll, propeller twist, and helix twist) of the central base pair or base pair step are calculated.
  • Query Table: These average structural features for each pentamer are stored in a query table, where the pentamer sequence acts as the key.
  • High-Throughput (HT) Prediction: When a new DNA sequence is input into the DNAshape web server, the method uses this query table. It slides the pentamer window across the input sequence and looks up the pre-computed average structural features for each pentamer. This allows for rapid prediction of structural features for any arbitrary DNA sequence.
  • Feature Definitions: Specific definitions for features like minor groove width (MGW) are used, ensuring they are independent of the DNA strand orientation.

Datasets

  • Training Data: The method was trained using MC simulations of 2121 different DNA fragments, ranging from 12 to 27 base pairs in length. This dataset provided coverage for all 512 unique pentanucleotides, with each occurring on average 44 times.
  • Validation Data:
    • Experimental Structures: The predictions were rigorously validated against DNA structures solved by X-ray crystallography and NMR spectroscopy. This included datasets of protein-DNA complexes and unbound DNA structures.
  • Hydroxyl Radical Cleavage Data: Data from the ORCHID2 server, which uses hydroxyl radical cleavage intensity as a proxy for MGW in solution, was used to validate predictions on large datasets of yeast and fly nucleosome binding sites.
  • Molecular Dynamics (MD) Simulations: Comparisons were also made with data from MD simulations.
  • Fis-Binding Sites: Seven DNA binding sites with varying Fis-binding affinities were used to correlate predicted MGW with binding affinity.

Key Insights and Ideas

  • Pentamer Context is Sufficient: The study demonstrates that a pentanucleotide context is sufficient to accurately capture the determinants of DNA shape features, making high-throughput prediction feasible.
  • DNA Shape as a Determinant of Binding Affinity: The strong correlation between predicted minor groove width and the logarithm of binding affinity for Fis-binding sites suggests that DNA shape is a significant factor in determining protein binding strength.
  • Distinguishing Bound vs. Unbound States: The method can predict structural features that reflect intrinsic DNA properties, helping to differentiate between the unbound state and states deformed by protein binding or crystal packing.
  • HT Prediction for Genomic Studies: DNAshape provides a crucial tool for integrating DNA sequence and shape analysis at a genomic scale, which was previously a major bottleneck.

Key Results

  • High Accuracy: The DNAshape web server provides predictions that show high correlation with experimental data (X-ray, NMR, OH-cleavage) and MC simulations, with Spearman's rank correlation coefficients often exceeding 0.8.
  • Speed: The HT methodology allows for the prediction of DNA structural features for the entire yeast genome in less than 1 minute on a single processor.
  • Validation Success: The method successfully predicted structural features for various DNA binding sites, including those for transcription factors like Hox proteins and Fis.
  • Improved HelT Prediction: The method accurately predicts helix twist (HelT) for dinucleotides where MD simulations with certain force fields have struggled.

Limitations and Future Work

  • Crystal Packing Artifacts: The authors acknowledge that experimental structures, particularly from X-ray crystallography, can exhibit deformations due to crystal packing that are not representative of DNA in solution. The method aims to capture intrinsic DNA shape, but direct comparison with experimental data sometimes requires careful interpretation.
  • Dynamic Nature of DNA: While the method predicts preferred conformations, it's noted that protein-DNA recognition is a dynamic process. Future work could explore how these predicted static shapes relate to dynamic interactions.
  • Further Validation: The authors suggest that the rapid progress in making HT sequencing data available can now be coupled with structural analyses, implying future work in applying DNAshape to a wider range of genomic studies.
  • Understanding Binding Specificity: The paper highlights that providing structural information at a genomic scale will be the basis for a better understanding of protein-DNA binding specificity.

Summary updated

BibTeX
@article{zhou2013dnashape,
  author = {Tianyin Zhou and Lin Yang and Yan Lu and Iris Dror and Ana Carolina Dantas Machado and Tahereh Ghane and Rosa Di Felice and Remo Rohs},
  title = {{DNAshape: a method for the high-throughput prediction of DNA structural features on a genomic scale}},
  journal = {Nucleic Acids Res.},
  year = {2013},
  doi = {10.1093/nar/gkt437},
}
Figure