A simple program, BEND, has been written to calculate the magnitude of
local bending and macroscopic curvature at each point along an arbitr
ary B-DNA sequence, using any desired bending model that specifies val
ues of twist, roll and tilt as a function of sequence. The program has
been used to evaluate six different DNA bending models in three categ
ories. Two are bent non-A-tract models: (a) A new model based on the n
ucleosome positioning data of Satchwell al al 1986 (J. Mol. Biol. 191,
659-675), (b) The model of Calladine et al 1988 (J. Mol. Biol. 201, 1
27-137). Three are bent A-tract models: (c) The wedge model of Bolshoy
at al 1991 (Proc. Natl. Acad. Sci. USA 88, 2312-2316), (d) The model
of Cacchione at al 1989 (Biochem. 28, 8706-8713), (e) A reversed versi
on of model (b). The last is a junction model: (d) The model of Koo an
d Crothers 1988 (Proc. Natl. Acad. Sci. USA 85, 1763-1767). Although t
hey have widely different assumptions and values for twist, roll and t
ilt, all six models correctly predict experimental A-tract curvature a
s measured by gel retardation and cyclization kinetics, but only the n
ew nucleosome positioning model is successful in predicting curvature
in regions containing phased GGGCCC sequences. This model - showing lo
cal bending at mixed sequence DNA, strong bends at the sequence GGC, a
nd straight, rigid A-tracts - is the only model consistent with both s
olution data from gel retardation and cyclization kinetics and structu
ral data from x-ray crystallography.