Comparison of Analyses of DNA Curvature

Abstract

Popular programs for characterizing DNA structure include Curves 5.1 (Lavery, R. and Sklenar, H. J. Biomol. Struct. Dyn. 6, 63–91, 1988; Lavery, R. and Sklenar, H. J. Biomol. Struct. Dyn. 6, 655–67, 1989) and Freehelix98 (Dickerson, R. E. Nucleic Acids Res. 26, 1906–1926, 1998), along with the more recent 3DNA (X. J. Lu, Z. Shakked and W. K. Olson. J. Mol. Biol. 300, 819–840 (2000). Given input of structural coordinates, all of these programs return values of the local helical parameters, such as roll, tilt, twist, etc. The first two programs also provide characterization of global curvature. Madbend (Strahs, D. and Schlick, T. J. Mol. Biol. 301, 643–663, 2000), a program that computes global curvature from local roll, tilt, and twist parameters, can be applied to the output of all three structural programs. We have compared the curvature predicted by the three programs with and without the use of Madbend. Global bend magnitudes and directions as well as values of helical kinks were calculated for four high-resolution DNA structures and four model DNA helices. Global curvature determined by Curves 5.1 without Madbend was found to differ from values obtained using Freehelix98 with or without Madbend or 3DNA and Curves 5.1 with Madbend. Using model helices, this difference was attributed the fact that Curves 5.1 is the only program sensitive to changes in axial displacement, such as shift and slide. Madbend produced robust values of bend magnitude and direction, and displayed little sensitivity to axis displacement or the source of local helical parameters. Madbend also appears to be the method of choice for bending comparisons of high-resolution structures with results from cyclization kinetics, a method that measures DNA curvature as a vectorial sum of local roll and tilt angles.