Information Technology Reference
In-Depth Information
1.1 The Wedge Model
The wedge model is called a “nearest neighbour model” since the geometry of a stack of
two base pairs is considered to be defined by the two constituent nucleotides, and the
influence of more distant neighbours is ignored [19]. The model is based on gel-
electrophoresis data, described in terms of dinucleotide parameters, roll and tilt angles.
1.2 The Junction Model
The junction model was proposed based on gel-mobility experiments using
oligonucleotides with “phased” (suitably spaced) adenine tracts [20, 21]. According to this
model, curvature is caused by a deflection at each junction between the axes of the normal
B-DNA and the B'-DNA of the poly dA, poly dT. The model assumes that the deflection at
junction is a result of negative base-pair inclination in adenine tracts and zero inclination in
the intervening B-DNA segments, and that this difference generates the bend [21].
According to Haran et al. [22] the wedge and the junction models are not necessarily
incompatible. It appears, however, that there are events of curvature that neither the
junction model nor the wedge model can sufficiently explain. For example, some GC-rich
motifs, such as GGGCCC and CCCGGG have been showed opposite direction of bending
[3] to those predicted from both models.
1.3 The Elastic Rod Model
The elastic rod model is based on DNAseI digestion data [1]. This enzyme bends the
substrate towards the major grove, so the resulting model allows only one direction of
bending, towards the roll angle. The original method described DNA bending in terms of a
dimensionless parameter, “relative bending propensity” determined for trinucleotides [1, 4].
Subsequently, a physical model of sequence-dependent anisotropic-bendability (SDAB)
was developed [9]. SDAB considers DNA to be an elastic rod, in which the flexibility of
each segment (di- or trinucleotide) is anisotropic, namely, greater towards the major groove
than it is in other directions. As DNAseI cannot distinguish between
a priori
bent and
dynamically “bendable” sites, curvature according to this model is both static as well as
dynamic in nature and can be recognized by the phased distribution of bent/bendable sites
along the sequences. This can be visualized as a vectorial property along the sequence
(Figure 2) which is conceptually analogous to the hydrophobic moment calculations in
protein sequences.
There are a number of computer programs that can predict curvature from sequence.
One of the first algorithms available for curvature calculations was BEND written by
Goodsell and Dickerson [23]. The algorithm can handle both dinucleotide and trinucleotide
descriptions, and uses a simplified procedure wherein the successive deflection angles (roll,
tilt) are summed up as vectors. This is a well-known approximation that is acceptable
however only for low angle values. The BEND algorithm calculates curvature for segments
of 11 nucleotides, and outputs a plot of curvature versus sequence position. The algorithm
was incorporated into the EMBOSS suite of sequence analysis programs [24] under the
name BANANA (which is a reference to curved B-DNA of A and non-A tracts), and is also
available on-line [25]. The Haifa University server [26] for DNA structure calculation is
built around the program Curvature [27]. The DIAMOD program was written by Mensur
Dlakic for PC [28] and handles most curvature models. Finally, several precomputed
