Chemistry Reference
In-Depth Information
to each other. The target pose for PDF represents a particularly challenging case for NOE
matching. An examination of the distribution of the predicted 3D X-filtered NOEs reveals
that, although most predicted protein-ligand NOEs are rich in their information content
in placing the ligand in the correct region of the pocket, they contain little discrimination
power between the two predominant binding modes. These NOEs, which arise from the
ligand's central ring and are to PDF methyl groups that lie directly above the ring, are
readily satisfied in both the correct and the decoy pose that has the ligand flipped by 180º in
the binding pocket. NOEs from the ligand methyl groups at opposite ends of the compound
contain the only true information to distinguish between the poses, and residue types at both
ends of the pocket are similar - each end of the binding pocket contains isoleucines, leucines
and valines. A unique residue in one end of the pocket is a histidine. It is predominantly
this residue that allows the NOE matching to score the correct binding pose with a lower
COST than the decoy pose.
For the three cases shown above using simulated data on protein/fragment complexes,
NOE matching worked with varying degrees of success. As the fragment becomes struc-
turally less complex, the differences in the COST between correct or close to correct poses
and decoy poses becomes smaller. Whereas for the CDK2 case NOE matching readily
identified the correct pose, for the PDF case, the gradation in COST as structures became
more dissimilar to the target pose was very shallow (Figure 5.8A). Nevertheless, the COST
for the poses dissimilar to the target pose (observed at approximately 3.5 Å from the target
pose in Figure 5.8A) is over a factor of two higher than that for the correct poses. Hence it
is evident that, given high-quality data, NOE matching can identify the correct pose even
for fragments.
5.5.4 PDF with Experimental Data
In order to determine whether NOE matching will work on small proteins with fragments
using 'typical'NMR data, we repeated the calculations, but this time using the experimental
NOE cross peak list. A 3D X-filtered NOESY spectrum (
τ
m
=
150 ms) for the PDF/
6
was
acquired on a 1.5 mM sample of PDF with a room temperature probe. The experimentally
determined resonance assignments were used for the ligand resonance assignments. The
3D X-filtered NOESY yielded 109 peaks, which were clustered into 78 protein
1
H
13
C
groups. Trial binding poses were generated with
Poser
as described above for the simulated
example. The NOEmatching protocol was run using BMRB-predicted chemical shifts. This
test case represents a practical application of NOEmatching on this small protein-fragment
complex.
The results obtained from applying NOE matching to PDF/
6
are shown in Figure 5.9.
The pose with the minimum COST value has an RMSD of 1.12 Å to the target pose. The
pose with the closest RMSD to the target pose itself ranks 13 out of 1000 poses. The results
obtained with experimental data are similar to the results obtained with simulated data.
5.5.5 PDF with Experimental Data and SHIFTX Chemical Shifts
In the absence of sequence-specific protein NMR resonance assignments, the data used for
NOE matching are limited to the unassigned experimental protein
1
H and
13
C chemical
shifts, the predicted protein
1
H and
13
C chemical shifts, and the experimental and predicted
