704
CORE TRACING
(Ht-d). Although neither structure was solved using core
tracing, the solution of Ht-d employed core tracing for
the determination of molecular boundaries and in the
initial fitting.
Astacin has a well formed 3 A solvent-flattened MIR-
density map (Gomis-Rtith, Stoker, Huber, Zwilling &
Bode, 1993). The map (their Model 1) is based on six
heavy-atom derivatives. Helices, some /-strands and a
number of bulky side chains are clearly visible when the
map is compared with the final model. Several figures
have been made from this structure in order to have a
wider representation in our examples.
The majority of our examples are taken from Ht-d, a
structure under investigation in our laboratory (Zhang et
al., 1994). Despite extensive effort, we did not solve this
structure from our MIR maps. From the beginning, there
were tantalizing views of helices with enough detail to
verify the handedness and fix the space group (P65).
None of the partial models would successfully refine.
In retrospect, the problem was a poor density, phased
on only three heavy-atom derivatives. The presence of
two molecules in the asymmetric unit related by non-
crystallographic symmetry also caused difficulties in
the phasing. The structure was solved by a graphical
replacement procedure using a closely related model
(Gomis-Rtith, Cress & Bode, 1993) positioned to co-
incide with identifiable parts of the three-derivative
maps (Zn, several helicies). At about the same time,
a reassessment of our data sets found a usable fourth
derivative. This produced a much better MIR map which
fits the refined model but is independent of it. A four-
derivative map was used in most of the illustrations,
resulting in clearer views than those obtainable with
a three-derivative map, but being less faithful to the
structure-solution process.
Helices were discernible at higher thresholds than/3-
sheets in both structures, although there were always
some gaps in the main chain, even at 1.0r. Figs. 3(a)
and 3(b) show the four major helices of Ht-d compared,
at 1.5or, to core tracing and contours. For the illustration,
the core tracing and contours have been restricted to
feature volumes containing the refined atoms and a
single guard layer. This is a four-derivative map; the
three-derivative maps showed only one or two helices
consistently.
fl-strands are not clearly resolved in Ht-d, even with
a four-derivative map. There tends to be more cross
connectivity between strands than connectivity along
the main chain. However, when viewed edge on, the
sheet is separable from the rest of the molecule. Figs.
3(c) and 3(d) show the/3-sheet region of astacin which
is better resolved into strands, and which shows some
of the bulkier side chains. Even this sheet has cross
connectivity at a level low enough (1.0tr) to capture most
of the main-chain connections.
Sometimes a core trace or Greet skeleton is recogniz-
able as helical, but often there are additional connections
along the axis of the helix creating a rod-like bundle of
lines. The sheets with cross connections may look more
like a net than parallel strands. Only some pieces of
random coil consistently appear chain-like. Nonetheless
there are probably characteristic signatures for helices
and sheets which we can learn to recognize.
Volumes containing a few hundred residues are not
too cluttered for viewing. Thus, it is possible to compare
a core tracing to a Co trace of an entire molecule.
The two independent molecules of Ht-d are shown in
Figs. 4(a)-4(d) and 4(h) (202 residues each). There are
some differences between the two molecules but, by
and large, the core tracings are similar. This is more
a test of the quality of the maps and phasing than of the
ability of core tracing to find connectivity. As another
example, the amino domain (residues 1-99) of astacin
is shown in Figs. 4(e)-4(g). For clarity in these printed
figures, the rendering volume has been restricted to the
neighborhood of a single molecule or domain. Although
such a restriction is not possible before a structure is
solved, the dynamic rotation, scaling and clipping on an
interactive display compensate in part.
Finally, we explore the delineation of molecular
boundaries in Fig. 5. A single static view can only
suggest what can be seen interactively. We present
projections perpendicular to the z axis, since the
asymmetric unit in P65 is relatively thin in that direction
(15 A in Ht-d). The densities used are the two MIR maps
for Ht-d analysed in Table 1. The relative quality of the
maps is apparent in the differentiation between molecule
and solvent regions. The selection of long paths for the
core tracing reduces the clutter without degrading the
signal (Figs. 5a and 5b). Some paths may appear short in
the figures because they have been clipped at a boundary
and continued elsewhere. Our structural references are
a dot plot of Ca positions (Fig. 5f) and an augmented
dot plot also containing C6 positions (Fig. 5e) which
fleshes out the molecular volume but does not obscure
the solvent volume. Dot plots of joins and maxima
were also examined as alternatives to core tracing. Figs.
5(c) and 5(d) show the highest 30% of the maxima
for Ht-d (threshold about 2.3tr). The maxima clump in
the molecular volume, especially for the four-derivative
map where only 9% of the dots lie in solvent volume
(Fig. 5d). The difference between solvent and protein is
still visible for the three-derivative map although 25%
of the dots lie in solvent.