Transcript Kiran2
Novel Dihydrofolate Reductase Inhibitors. StructureBased versus Diversity-Based Design and HighThroughput Synthesis and Screening
Pierre C. Wyss, Paul Gerber, Peter G. Hartman, Christian
Hubschwerlen, Hans Locher, Hans-Peter Marty, Martin Stahl
F. Hoffman-La Roche Ltd, Basel Switzerland
J. Med. Chem. 2003, 46, 2304
Kiran-20060214
Biology
•
•
•
•
•
•
•
•
The spread of antibiotic resistance has led to a renewed / ongoing search
for new targets and new antibacterial compounds
DHFR plays a key role in the synthesis of amino acids and purines. DHFR
null bacteria are not viable
Bacteria are gram +ve or gram –ve depending on whether they stain purple
or pink with the Gram stain
Boils, pimples, strep throat, ear infections, nosocomial infections,
pneumonia etc. are caused bacteria
Trimethoprim (TMP) was already in clinical use at the time this work was
done
Hoffmann-La Roche had RO-64-5781 in hand, it was 190,000 times more
potent than TMP but it was not druggable.
IC50: concentration required to suppress 50% of bacterial growth
MIC: lowest concentration that will inhibit bacterial growth 100% after
overnight incubation (IC100)
Goal and methods
•
Design novel DHFR inhibitors with good drug-like properties
•
Two methods evaluated:
- Structure-based library design
• Crystal structures for DHFR known
• Potent compounds known
- Diversity-based library design
• Known potent compounds had high molecular weights, were highly
plasma protein bound and showed poor solubility
• Design compounds of a different structural class and reduced
molecular weight
•
Selectivity
- Inhibit bacterial DHFRs but not human DHFR
Basic idea for library design
•
Replace the greasy hydrophobic piece (red), keep the diaminopyrimidine
piece (blue)
Testing synthetic feasibility
-
Compound 6 is good practical chemistry, is a key intermediate for library synthesis
Reactions could be done on 0.35 - 0.7 mM scale
Purification by HPLC not required
1392 compounds made in library format
Paradigm for structure-based virtual screening
N
N
complexed with S. aureus DHFR
- Diaminopyrimidine is an ideal fit for a narrow pocket in the active site: “needle”
- Docking would be done with the constraint of a fixed position for the needle fragment
Homology model for structure-based screening
•
•
•
•
Sa: TMP-sensitive S. aureus DHFR
Hum: human DHFR
Spn: TMP-sensitive S. pneumoniae DHFR
Sp1: TMP-resistant S. pneumoniae DHFR
- No significant induced fit effects can be observed, assume receptor is rigid for
docking experiments
- High similarity at the active site, assume that use of a single crystal structure will
be sufficient for screening against all targets
Structure-based compound selection: method
•
9948 secondary amines retrieved as virtual reagents
•
Single 3D conformations were generated using Corina
•
The protonation states were adjusted, the nitrogen on
the aminomethyl substituent was not protonated
•
The single crystal structure of DHFR from TMP-sensitive S. aureus
complexed with RO-62-6091 was used
•
Library members were docked with a fixed position (taken from the crystal
structure) for the 2,4-diaminopyrimidine piece
•
FlexX was used for docking experiments, hydrogen bonds formed at solventexposed regions of the enzyme were penalized.
Structure-based compound selection: results
•
FlexX produced docking results for about half of the library
•
For each compound the solution with the highest binding energy was selected
•
252 of the 300 top-ranked candidates were synthesized (LIBRARY 1)
•
150 candidates were selected randomly from compounds for which no
docking solution could be found.
•
Another 150 lowest-ranked compounds were selected
•
269 of these 300 compounds were synthesized (LIBRARY 2)
Some observations from FlexX docking
•
Hits from LIBRARY 1 were more active against Spn DHFR even though the
crystal structure of Sa DHFR had been used i.e. “the power of virtual
screening lies in its ability to filer out undesirable compounds rather than
identifying specific active ones”
– FlexX does not find docking solutions for compounds with more than one
bulky substituent at the aminomethyl nitrogen.
– FlexX discards very small molecules and compounds with long flexible
moieties.
– FlexX assigns high ranks to conformationally restricted flat rigid polycyclic
motifs: an additional 370 compounds (LIBRARY 4)
Diversity-based compound selection: method & results
•
The library of 9448 virtual reagents was clustered according to chemical
similarity
- Compounds were superimposed in pairs at the newly
formed C-N bond
- The amine substituent was rotated. Conformers with maximum volume
and H-bond-donor and H-bond-acceptor overlap were generated.
- The list of pairwise similarity scores was used for clustering in a binary
tree using a complete linkage algorithm
-- two clusters are combined only if all members of the first are within
the distance threshhold of all members of the second
-- Each iteration reduces the number of clusters by one
Results:
•
About 500 compounds were needed to adequately represent the chemical
space of the library, a set of 501 compounds was synthesized (LIBRARY 3)
Experimental validation and results
•
•
•
Primary high throughput screen against Sa DHFR and Spn DHFR at 10 mM,
select compounds that show >50% inhibition
Screen at 25 mM for antibacterial activity
Also test for thymidine antagonism and activity in the presence of 10% human
serum
IC50s (mM) against Spn DHFR for hits from Library 1
0.006
0.075
0.045
1.1
0.01
0.044
0.007
0.002
0.004
0.012
0.021
0.0098
0.004
Selectivity for hits from Library 1
- Except for compound 11, all the compounds were more active against Spn/Sp1
- Only 10 and 13 showed selectivity for bacterial DHFR over Human DHFR
compared to TMP
- The R isomer of 9 helped identify a new cleft in the enzyme
Crystal Structure of (R)-9
N
N
Summary
•
•
•
Overall, the structure-based method gave better results
Potent and selective inhibitors of wild type and TMP-resistant S. pneumoniae
DHFR were identified
The new class of inhibitors had features for better physicochemical properties
Discussion
•
The same set of 9948 amines was used for both libraries. But the diversity
library gave a poor hit rate. Can that be improved?
•
Is there enough information now available to attempt virtual prediction of
selectivity (human vs bacterial DHFR?)
Back up