This document presents the results of a study aiming to develop selective inhibitors of BACE2 using computer-aided drug design. BACE2 is homologous to BACE1 but is predominantly expressed in peripheral tissues and may serve as an alternative protease in APP processing. The study validated docking programs for reproducing binding poses of BACE1 and BACE2 inhibitors and evaluated scoring functions for virtual screening. A BACE2 homology model produced better virtual screening results than crystal structures. Future work will involve virtual screening larger libraries and designing new scaffolds for selective BACE2 inhibition.

1. Introduction BACE2 Virtual Screening
Conclusions
Structure of BACE1 vs BACE2
On the Pursuit of New Beta-Secretase 2 Inhibitors Using Structure-
Based De-Novo Design Methods
Boobalan Pachaiyappan, Hongbin Yuan, Pavel A. Petukhov*
Department of Medicinal Chemistry and Pharmacognosy, College of Pharmacy, University of Illinois at Chicago, Chicago IL 60612
University of Illinois at Chicago College of Pharmacy
UIC
* pap4@uic.edu
β-Secretase 2 (BACE2), a homolog of β-Secretase 1 (BACE1), belongs to the family of aspartic
protease and predominantly is expressed in peripheral tissues. There exists a broad consensus in the
Alzheimer’s disease (AD) research community that the key to successful treatment of AD lies in the
specific inhibition of BACE1. BACE1 is crucial for release of the amyloidogenic fragments that later form
the extracellular neuritic amyloid plaques – one of the factors in the pathogenesis of AD. Most of the
known BACE1 inhibitors are relatively non-selective and also target other aspartic proteases including
the highly homologous BACE2. It has been shown that BACE2 may serve as an alternative -secretase,
another crucial enzyme that plays an important role in degradation of the amyloid precursor protein
(APP), inhibition of which is highly undesirable. To better understand the role of BACE2 in human
physiology and APP processing, we believe that selective inhibitors of this enzyme are needed as
research tools. We hypothesize that it should be possible to design and synthesize inhibitors for BACE2
using an integrated multidisciplinary approach based on the combination of computer-aided drug design,
medicinal chemistry, and biology. Herein we present the results of validation of docking and scoring
procedures and highlight the structural differences of the binding sites of BACE1 and BACE2 useful for
further design of active and selective BACE2 inhibitors.
The homologues share 51% overall primary sequence identity and fold in a nearly similar manner. The
side chains of amino acids forming the binding pockets of BACE1 and BACE2 are not oriented exactly
towards the ligand binding region thereby making the selective inhibition a real challenge.
Figure 1: (Left) Overlay of BACE1 and BACE2 protein structures complexed with ligands. (Right) The superimposed binding
regions of BACE1 and BACE2 complexed with ligands. Only heavy atoms are shown here. Green – carbon, red = oxygen,
blue – nitrogen, yellow – sulfur.
Until now, there are no specific inhibitors for BACE2. The ones that are reported are actually BACE1
inhibitors that display BACE2 activity as well, or bind to other members of aspartic protease family.
Experimental Methods
 Protein Structures
 BACE1 (2b8l.pdb,1 2irz.pdb,2 2is0.pdb,3 1w51.pdb4 and 2g94.pdb5)
 BACE2 (2ewy.pdb6 and homology model developed in our lab)
 Simple Docking:
 FlexX,7 FRED8 and GOLD9
 Scoring: FlexX score, CScore, GOLD score, Chemgauss3
and Chemscore
 Virtual screening (VS) experiments using FRED
 Conformer generation: Omega2
 Scoring functions: PLP, Chemgauss3, Screenscore, Shapegauss, CGO, CGT, OEChemscore and
consensus.
 Unless indicated, tripos software (Sybyl7.3)11 was used for modeling and visualization purposes
 All root mean square deviation (r.m.s.d) calculations were done using the tool ‘match’ in Sybyl7.3
Docking Results
Inhibitors
VolSurf Results
 BACE-1 inhibitors must pass BBB. Our modeling results suggest none of the inhibitors reported cross
BBB
 BACE-2 inhibitors must be cell permeable. Our modeling results suggest that none of the inhibitors
reported permeate CACO2 cell model.
Figure 2: (Top left) BACE1 inhibitors projected on BBB model. Red points refer to high ability to cross BBB, whereas the blue points
have low BBB penetrating ability. Black points are BACE1 inhibitors published in literature. (Top right) BACE2 inhibitors projected on
CACO2 model. Red and blue points refer to high and low permeability. Black points are BACE1 inhibitors published in literature.
We evaluated the docking accuracy of BACE1 (5 crystal structures) and BACE2 (2 structures) using
FlexX, FRED and GOLD. The results are summarized in Table 2.
Figure 3: Comparison of poses from crystal structure
(cyan) and GOLD results for a BACE1 inhibitor (2b8l_lig).
Lowest r.m.s.d was less than 1.5Å).
Figure 4: Comparison of poses from crystal structure (cyan)
and GOLD results for a BACE2 inhibitor (2ewy_lig). Lowest
r.m.s.d was less than 1.5Å).
GOLD FlexX FRED
2b8l.pdb 0.34 1.80
0.96*
-
1w51.pdb 1.23 1.82 -
2g94 2.76
0.78*
- 3.68
2irz.pdb 0.60 0.90 2.36
2is0.pdb 0.68 0.82 -
2ewy.pdb 0.41
0.39*
0.46 1.19
Bace2_homology NA NA NA
In order to validate the scoring accuracy, virtual screening was performed on two BACE2 structures
(2EWY.pdb and homology model developed in our lab). A library containing 10,009 non-binders
(obtained from NIH database) and 10 BACE2 binders (see figure) was inputted in omega2 to create
~500 conformers for all ligands. These conformers were rigidly docked to both the BACE2 models
using FRED and analyzed using six scoring functions.
 Simple docking
 Except in one case, original binding modes of ligands are reproduced best using GOLD. The
decreasing order of efficiency of docking programs was found to be: GOLD > FlexX > FRED.
 Inclusion of constraints in GOLD or FlexX resulted in better binding modes
 Docking using FRED is dependent upon number of conformers given as input. Because the
number of conformers increases exponentially with the linear increase of number of rotatable
bonds, conformational sampling is difficult and hence FRED gives inconsistent results.
 Virtual Screening
 BACE2 homology model developed in our lab produced better results than the X-ray crystal
structure itself. This model will be used in future in silico experiments
 Scoring accuracy of ‘Shapegauss’ was found to be better than any other FRED scoring
functions for both BACE2 models
 New scaffolds appeared in top 5% during BACE2 VS will be re-docked (using GOLD and FlexX )
and scored to further explore its utility
 VS of Chembridge library of ~200,000 drug-like compounds will be performed to further diversify
our pursuit.
 VS of Virtual focused combinatorial library will be carried out using the same protocol (alternate
CADD strategy)
Future Directions
Acknowledgments
This research is supported by the National Institute of Health and the National Institute of Aging
grant R21 AG027454 and Hans Vahlteich endowment program grant of College of Pharmacy at
University of Illinois at Chicago.
BACE 2 (2ewy.pdb)
0
200
400
600
800
1000
Rank
2ewy_lig 2b8l_lig M-22 M-23 M-24
M-25 M-26 M-27 M2 M3
CGO PLP
Screen
score
OEChem
score
Shape
gauss
Chem
gauss3
ConsensusCGO
BACE2_Homology_Model
0
100
200
300
400
500
600
700
800
900
1000
Rank
2ewy_lig
2b8l_lig
M-22
M-23
M-24
M-25
M-26
M-27
Merck2
Merck3CGO PLP Screen
score
OEChem
score
Shape
gauss
Chem
gauss3
ConsensusCGO
Figure 5: (top) VS results using 2ewy.pdb.
All the ten binders are ranked within top
171 out of 10,019 compounds using
when score using shapegauss. (Bottom)
VS results using BACE2 homology
model developed in our lab. The
weighted r.m.s.d between the
backbones of X-ray structure and our
homology model is 1.8Å. Major
conformational difference exists only at
the loop regions. Shapegauss performed
best as all the ten binders are ranked
within top 35 out of 10,019 compounds.
The pose of BACE2 inhibitor (2ewy_lig)
was inspected and found to be similar to
the X-ray structure of the ligand.
Table 2: Comparison of r.m.s.d (in Å) between X-crystal poses to
the ones generated by docking programs. * represents docking
under constraints.
Shape H-bonds Metal Aromatic Desolvation
Shapegauss Yes No No No No
Chemgauss3 Yes Yes Yes No Yes
OEchemscore Yes Yes Yes No No
PLP Yes Yes Yes No No
Screenscore Yes Yes Yes No No
M-26
Table 1: Energy terms used in structure-based FRED scoring
functions.
Figure 4. Chemical Structures of Inhibitors1,6,12-13
References:
1. Bioorg.Med.Chem.Lett. v16 pp.641, 2006
2. J Med.Chem. v49 pp.7270, 2006
3. J.Med.Chem. v49 pp.7270, 2006
4. J.Mol.Biol. v343 pp.407, 2004
5. J.Am.Chem.Soc. v128 pp.5310, 2006
6. J.Mol.Biol. v355 pp.249, 2006
7. www.biosolveit.de
8. www.eyesopen.com
9. www.ccdc.cam.ac.uk
10. NIH molecular libraries small molecule repository
(http://pubchem.ncbi.nlm.nih.gov/assay/assay.cgi?aid=375)
11. www.tripos.com
12. J Med Chem. V47 pp6447, 2004
13. J Med Chem. V47, pp6117, 2004
N
S
O O
O
NH
H
N
O
Ph
OH
H
N O
H
N
O Ph
HO H
N
2b8l_lig 2ewy_lig
NR
S
O O
O
NH
H
N
O
Ph
OH
H
NF
M-22 (R=H),
M-23 (R=CH3)
M-25
M-2
M-24
M-27
N
S
O O
O
NH
H
N
O
Ph
OH
NH2F
N
S
O O
O
NH
H
N
O
Ph
OH
NHEtF
N
S
O O
O
NH
H
N
O
Ph
OH
H
N
N
S
N
O O
O
NH
H
N
O
Ph
OH
H
N
O
S
PhH2C
O O
O
NH
OF
H
N
O
NH2
5
M-3
N
S
O O
O
NH
F
O
N
N
NH2
Ph