close

Вход

Забыли?

вход по аккаунту

?

Generation and Evaluation of a Homology Model of PfGSK-3.

код для вставкиСкачать
Arch. Pharm. Chem. Life Sci. 2009, 342, 327 – 332
S. Kruggel and T. Lemcke
327
Full Paper
Generation and Evaluation of a Homology Model of PfGSK-3
Sebastian Kruggel and Thomas Lemcke
Institute of Pharmacy, University of Hamburg, Hamburg, Germany
Plasmodial GSK-3 is a potential new target for malaria therapy. For a structure-based design project, the three-dimensional information of the designated target is needed. Unfortunately, experimental structure data for plasmodial GSK-3 is not yet available. Homology building can be used
to generate such three-dimensional structure data using structure information of a homologous
protein. GSK-3 possesses a very flexible ATP-binding site, a fact reflected in the variety of X-ray
structures of the human GSK-3b which are deposited in the protein data base and are crystallized with different ligands. We used ten different HsGSK-3b templates for the model building of
plasmodial GSK-3 and generated 200 models for each template with different modeling protocols. The quality of the models was evaluated with different tools. The results of these evaluations were used to calculate a rank-by-rank consensus score. The top models of this were used to
compile an ensemble of PfGSK-3 models that reflect the flexibility of the ATP-binding site and
that will be used for the structure-based design of potential ATP-binding site inhibitors of PfGSK3.
Keywords: GSK-3 / Homology building / Malaria structure / Scoring /
Received: September 4, 2008; accepted: February 5, 2009
DOI 10.1002/ardp.200800158
Introduction
Malaria is still one of the most serious diseases in the
world. More than one million people die of malaria every
year, among them, a child dying of malaria nearly every
30 seconds [1]. Malaria infections are caused by parasites
of different Plasmodium species, of which P. falciparum is
the most vicious one. It causes malaria tropica, a variation of the disease that very often ends lethally if the
patient is not treated adequately. A major problem in
malaria therapy is the rapid development of parasite
strains that are resistant against the most often used antimalarial drugs. In some areas of Africa or southeast Asia,
multiresistant parasite strains are making the therapy of
Correspondence: Dr. Thomas Lemcke, University of Hamburg, Institute
of Pharmacy, Bundesstr. 45, D-20146 Hamburg, Germany.
E-mail: lemcke@chemie.uni-hamburg.de
Fax: +49-40-42838-6573
Abbreviations: Glycogen synthase kinase-3b (GSK-3b); GSK-3 of P. falciparum (PfGSK-3); molecular dynamics (MD); rank-by-rank strategy
(RCS)
i
2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
malaria infections very difficult [2]. Therefore, the search
for new biochemical targets of the parasite is an important task.
Protein kinases have evolved as promising drug targets
in the last few years. Glycogen synthase kinase-3b (GSK3b), also known as tau-protein kinase I, is a serine / threonine protein kinase known to be involved in multiple cellular signal transduction pathways (WnT and insulin signalling, glycogen and protein synthesis, regulation of
transcription factors, embryonic development, cell proliferation and adhesion, tumorgenesis, etc.) [3 – 7]. Inhibition of the phosphorylation of glycogen synthase and
tau-protein are concepts that can be used in the treatment of diabetes and Morbus Alzheimer [8]. A number of
selective inhibitors of the ATP-binding site of human
GSK-3b (HsGSK-3b) are described in the literature and
some might be used as lead substances in drug-design
processes [9].
Droucheau et al. identified and cloned a gene homologue of the GSK-3 of P. falciparum (PfGSK-3). Subsequent
studies proved partially divergent sensitivity of inhibitors of PfGSK-3 and HsGSK-3b, suggesting the PfGSK-3 as a
328
S. Kruggel and T. Lemcke
potential antimalaria target [10]. To pursue this finding,
the design of selective PfGSK-3 inhibitors might be a
promising project.
A prerequisite for structure-based drug design is structural information about the intended biological target.
Several X-ray structures of HsGSK-3b are deposited in the
Research Collaboratory for Structural Bioinformatics
(RCSB) Protein Data Base [11], crystallized with different
inhibitors. A comparison of the sequences of human and
plasmodial GSK-3 shows subtle differences in the ATPbinding site, which might be used in the design of selective inhibitors. That a selective inhibition is actually possible can be demonstrated by the paullones [12], a class of
kinase inhibitors showing antiproliferative activity
against human tumor cells [13]. Paullones inhibit mammalian GSK-3 with a potency exceeding PfGSK-3 inhibition by one to two orders of magnitude [10]. This example
shows that the structural ATP-binding site differences are
sufficient for selective inhibition, although a directly
opposed selectivity type is mandatory for the development of PfGSK-3 inhibitors useful as antimalarial agents.
Unfortunately, three-dimensional structural information of PfGSK-3 is lacking. The benefit of using homology
models in structure-based design of kinase inhibitors
[14, 15] and other small molecule inhibitors has been
demonstrated in the past on several occasions, although
varying validation procedures were used [16 – 19]. Schafferhans et al. [18] demonstrated that the use of different
templates for the generation of a set of homology models
was beneficial for the identification of the correct binding modes of potential ligands by docking into an averaged binding site representation.
In this paper, we describe the generation and evaluation of a number of homology models of PfGSK-3 based
on several different crystal structures of HsGSK-3b. Afterwards, different tools were used to evaluate the quality of
the produced homology models. A consensus-scoring
scheme was developed to rank the calculated models by
their quality. This ranking will be helpful in the decision
to choose appropriate protein models that could be used
in subsequent ligand docking and structure-based design
studies.
Results
Homology modeling
Homology building is a technique that is used to model
the three-dimensional structure of a protein in the
absence of experimental data. To successfully use homology models in structure-based design procedures, a high
sequence similarity in the binding site on the one hand
i
2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Arch. Pharm. Chem. Life Sci. 2009, 342, 327 – 332
Table 1. PDB codes of the HsGSK-3b used as templates in the
model building.
PDB
Resolution
Ligand
Consecutive
Literature
1j1b (B)
1q3d (B)
1q3w (A)
1q41 (B)
1q4l (A)
1q5k (A)
1r0e (A)
1uv5
2ow3 (B)
2o5k
1.8
2.2
2.3
2.1
2.77
1.95
2.25
2.8
2.8
3.2
AMPPNP
STU
ATU
IXM
679
TMU
DFN
BRW
BIM
HBM
+
–
–
–
–
–
–
+
–
+
Aoki et al. [36]
Bertrand et al. [37]
Bertrand et al. [37]
Bertrand et al. [37]
Bertrand et al. [37]
Bhat et al. [38]
Allard et al. [39]
Meijer et al. [40]
Zhang et al. [41]
Shin et al. [42]
Abbreviations of ligands according to RCSB.
and the possibility to capture the conformational space
of the side chains on the other hand are of great importance.
The overall sequence identity between human (SwissProt entry P49841, 420 aa) and plasmodial GSK-3 (PlasmoDB entry pfc0525c, 440 aa) is 44.3%, similarity is about
61.4%. To cover the structural variability of the ATP-binding site of GSK-3, we used the information of ten X-ray
structures of the HsGSK-3b co-crystallized with different
ligands (see Table 1). The model was built from Ser58 to
Phe413 of the plasmodial GSK-3 sequence because of
existing human 3-D templates for this part of the target.
In the modeled region, the sequence identity is 53.1%
(similarity 69.7%). Figure 1 shows the alignment of the
sequences, active site residues are marked up; these 35
amino acids are aligned identically in each of the ten
templates used for the homology modeling, and the
sequence identity in the active site is 85.7%.
After aligning every template with the sequence of
pfc0525c, an automated homology building with MODELLER [20] was performed. To consider conformational
space, for each of the ten templates 200 models were calculated. Models were built up in two ways: (1) the standard MODELLER procedure was accomplished, and (2)
information of the co-crystallized ligand was taken into
account.
This resulted in a total of 4000 different models. The
goal of the model building was to create a set of highquality PfGSK-3 models that would represent the flexible
ATP-binding site of this kinase and could be used for subsequent structure-based design projects. To decide which
models should be included in the set, all models had to
be ranked accordingly. Ranking criteria were obtained
by evaluating all models with PROCHECK [21] and ProSa
[22, 23]. From PROCHECK, stereo chemical parameters
like the distribution in the Ramachandran plot, G-factor,
and bad contacts were used for the evaluation, from
ProSa z-score as energetical parameter was taken into
www.archpharm.com
Arch. Pharm. Chem. Life Sci. 2009, 342, 327 – 332
Generation and Evaluation of a Homology Model of PfGSK-3
329
Figure 1. Alignment of the modelled region, ruler refers to PfGSK-3 (pfc0525c) numbering.
account, additionally, the MODELLER probability density
function was examined. Because of contradictory or, in
other respect, very similar results, we calculated a consensus scoring including the abovementioned parameters in a rank-by-rank strategy (RCS).
Table 2. Criteria contributing to RCS.
Consensus scoring
The RCS is considering five parameters (see Table 2) and is
calculated as
2 G-factor
N
P
RCS ¼
Ri
i¼1
N
ð1Þ
In this equation, N is the number of parameters, i the
corresponding parameter index. Ri is given by the ranking of a model according to parameter i. Thus, RCS gives
the mean ranking of a single model. With respects to bad
contacts, Ramachandran distribution and G-factor, several models were quite similar. In these cases, the corresponding parameters were extended: Models were first
ranked by their preferable small number of residues in
the disallowed regions. Those ones with the same rank in
this aspect were sorted according to their preferably high
number of residues in the core region. The G-factor was
further subdivided in the same manner. First criterion
was the overall, second the dihedral, and third the covalent G-factor. Ranks regarding bad contacts were also subdivided. Models should have (1) as few as possible bad
contacts in the ATP-binding site and (2) as few as possible
i
2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Index Parameter
Exact composition
1 Ramachandran
1. As few residues in disallowed region
as possible.
2. As many residues in core region as
possible.
1. As high overall G-factor as possible.
2. As high G-factor dihedral as possible.
3. As high G-factor covalent as possible.
1. As few bc in the active site as possible.
2. As few bc in general as possible.
Low value of probability density function.
Low z-score.
3 Bad contacts (bc)
4 MODELLER
5 ProSa
Contributions 1 to 3 are further subdivided.
bad contacts in general. The workflow was established
with KNIME [24].
The consensus-score ranking was calculated twice:
From all 4000 models the overall top model was determined and in the second approach the top models resulting from the 20 different MODELLER calculations were
investigated. The overall top model was pfgsk_1j1b_128,
derived from the 1j1b, which had ANP as ligand. This is a
ligand that represents the natural substrate best. Several
other models of 1j1b performed very well and models
derived from 1q41 also had quite high scores. Figure 2
illustrates on the one hand the critical influence of the
template. The different templates induce different models, which can be clearly differentiated by the scoring
www.archpharm.com
330
S. Kruggel and T. Lemcke
RCS as mean ranking is the better the lower it is estimated. The left hand crosses are
models constructed taking into account ligand information, the right hand circles
belong to models constructed with the standard protocol.
Figure 2. Models based on the X-ray structures mentioned on
the x-axis (numbered consecutively by MODELLER) and their
RCS on the y-axis.
Arch. Pharm. Chem. Life Sci. 2009, 342, 327 – 332
were monitored over a simulation time of 2 ns. For every
template, two models were examined, the RCS top model
with and without taking ligand information into
account. All simulations exhibited similar rms deviations, relative to the starting geometry, and energy distributions over the observed time (the profile for
pfgsk_1j1b_128 is depicted as an example in Fig. 3). However, the MD runs of two models that were calculated on
the basis of template and ligand information (1uv5_BRW
and 2o5k_HBM), aborted after a short period due to close
contacts in the protein. These close contacts clearly
resulted in the additional constraints introduced into
the model-building protocol. The results of the MD simulations confirmed the reliability of the top models resulting from the different templates and supported the
somewhat inferior RCS score of the ligand-derived models, compared to the ones derived without ligand information.
Discussion
Rms deviations to the starting geometry were calculated after least-square fit of the
active site Ca to simulation beginning. It shows that active-site residues behave very
well, especially the Ca with constant rmsd less equal 0.5 .
Figure 3. The potential energy trend shows a stable plateau
after the first 100 ps.
scheme. On the other hand, the impact of the homologymodeling protocol on the scoring is illustrated. Models
generated using the constraints deduced from the cocrystallized ligands generally performed worse than the
ones that did not use any additional constraints.
Molecular dynamics
To further ensure the quality of the PfGSK-3 models,
molecular dynamics (MD) simulations were performed at
300 K and the spatial stability and the overall energy
i
2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
One major topic of our approach was to capture the flexibility of the ATP-binding site of the target structure by
using different HsGSK-3b templates for the model building. In Table 3, distances across the binding site for some
important residues are listed for the models and the corresponding templates. The height of the binding site
does not vary much in both models and templates (distance Leu213-Val93/Leu188-Val70). For the other dimensions (depth (lys166-lys108/arg141-lys85) and width
(gln210-ile160/gln185-val135)), the variability between
the individual templates is reflected very well in the
different models (see Fig. 4).
The absolute quality of our top RCS models has to be
estimated not only among the models themselves but,
even more importantly, to be compared to the underlying experimental structures. Regarding the Ramachandran parameters and bad contacts calculated with PROCHECK, the models were superior to the PDBs. The PROCHECK G-score and the z-score, derived from ProSa, of the
models performed slightly worse in comparison with the
templates but still exhibited reasonable values.
From Fig. 2 it is obvious that the models which were
generated using constrains deduced from the co-crystallized ligands generally performed worse than the ones
that did not use any additional constrains.
The additional steric restraints in the ATP-binding site
introduced into the homology modeling process by
including fixed distances between protein atoms and
atoms from the co-crystallized ligand apparently led to
difficulties in constructing a reasonable 3-D structure.
www.archpharm.com
Arch. Pharm. Chem. Life Sci. 2009, 342, 327 – 332
Generation and Evaluation of a Homology Model of PfGSK-3
331
Table 3. Distances for height, depth, and width across the ATP-binding sites of the top-scoring models and the corresponding templates.
Model
gln210-ile160a) lys166-lys108a) leu213-val93a)
Template
gln185-val135a) arg141-lys85a)
leu188-val70a)
pfgsk_1j1b
pfgsk_1q3d
pfgsk_1q3w
pfgsk_1q41
pfgsk_1q4l
pfgsk_1q5k
pfgsk_1r0e
pfgsk_1uv5
pfgsk_2o5k
pfgsk_2ow3
mean
min
max
D min/max
14.09
15.69
13.75
14.89
12.38
13.74
15.18
15.41
16.15
15.8
14.71
12.38
16.15
3.77
1J1B
1Q3D
1Q3W
1Q41
1Q4L
1Q5K
1R0E
1UV5
2O5K
2OW3
15.52
15.23
15.14
15.33
11.88
15.39
15.3
15.52
16.1
16.12
15.15
11.88
16.12
4.24
9.75
10.7
10.81
9.65
9.96
9.68
10.41
9.65
9.22
11.2
10.10
9.22
11.2
1.98
a)
14.03
17.78
15.55
17.63
17.8
15.59
17.04
19.49
13.65
20.01
16.86
13.65
20.01
6.36
9.55
10.66
11.01
9.76
9.92
9.74
10.41
9.78
9.46
11.19
10.15
9.46
11.19
1.73
16.35
19.47
17.25
17.56
16.51
17.17
19.29
17.79
10.71
21.25
17.34
10.71
21.25
10.54
Distances are indicated in .
The presented results enable us to compile a set of high
quality homology models, which captures the flexibility
of the ATP-binding site of glycogen synthase kinases and
will be used in an ongoing structure-based design study
to find selective inhibitors of PfGSK-3.
Computational details
Distances according to Table 3.
Figure 4. ATP-binding site of pfgsk_1j1b.
These difficulties seemed to end in unfavorable geometries not only in the ATP-binding site but also in adjacent
regions of the protein, because the optimization is
restricted by the mentioned conditions. Finally, this
resulted in models inferior to the models without these
restrains. The above-mentioned results of the aborted MD
simulations support this observation. One reason for this
behavior could be the “hardness” of the used constraints.
A more “soft” constraint might be more effective and
would probably not cause unfavorable geometries.
This leads us to the assumption that the additional
constraints provided by co-crystallized ligand information did not provide advantages in the model-building
process. Thus, in our further examination, we will concentrate on the models generated without ligand information.
i
2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Homology modeling
All homology models were generated using MODELLER
version 9v2, for MODELLER alignments the implemented
Needlemann-Wunsch algorithm [25] was used. Alignment shown in Fig. 1 was calculated with ClustalW
[26, 27], values for identity and similarity with EMBOSS
[28].
The used MODELLER standard procedure (automodel)
is a comparative protein structure modeling where spatial restraints are derived from a sequence alignment. In
the second approach, ligand distance restraints are additionally considered. For this purpose, the co-crystallized
ligand HETATM coordinates were transferred into the
model with automodel.nonstd_restraints hence residueprotein CA distances are constrained to template values
if these are 10 or less.
Molecular dynamics simulations
MD simulations were performed with the GROMACS
package version 3.2.1 [29 – 31], the opls all atom force
field [32, 33] and the single point charge (spc) water
model were used.
After box definition (dodecahedron box of 510 nm3)
equilibration of the system was achieved by a 100 steps of
steepest minimization (initial step size 0.01 nm, converwww.archpharm.com
332
S. Kruggel and T. Lemcke
gence criterion 1000 kJ N mol – 1 N nm – 1) and a further
position restraint MD simulation for 30 ps with fixed protein.
The MD simulation itself was performed again with
time step of 0.002 ps for 2 ns. Berendsen-thermostat [34]
was used to keep a constant temperature of about 300 K,
LINear Constraint Solver LINCS [35] to solve the major
MD problem of bond vibrations.
References
[1] WHO, Malaysia Fact Sheet, No. 94, 2007.
[2] C. Wongsrichanalai, A. L. Pickard, W. H. Wernsdorfer, S. R.
Meshnick, Lancet Infect. Dis. 2002, 2, 209 – 218.
[3] J. E. Forde, T. C. Dale, Cell Mol. Life Sci. 2007, 64, 1930 –
1944.
[4] B. W. Doble, J. R. Woodgett, J. Cell Sci. 2003, 116, 1175 –
1186.
[5] P. Cohen, S. Frame, Nat. Rev. Mol. Cell Biol. 2001, 2, 769 –
776.
[6] S. Frame, P. Cohen, Biochem. J. 2001, 359, 1 – 16.
[7] A. J. Harwood, Cell 2001, 105, 821 – 824.
[8] H. Eldar – Finkelman, Trends Mol. Med. 2002, 8, 126 – 132.
[9] L. Meijer, M. Flajolet, P. Greengard, Trends Pharmacol. Sci.
2004, 25, 471 – 480.
[10] E. Droucheau, A. Primot, V. Thomas, D. Mattei, et al., Biochim. Biophys. Acta 2004, 1697, 181 – 196.
[11] H. M. Berman, J. Westbrook, Z. Feng, G. Gilliland, et al.,
Nucleic Acids Res. 2000, 28, 235 – 242.
[12] K. Wieking, M. Knockaert, M. Leost, D. W. Zaharevitz, et al.,
Arch. Pharm. Pharm. Med. Chem. 2002, 335, 311 – 317.
[13] T. Lahusen, A. De Siervi, C. Kunick, A. Senderowicz, Mol.
Carcinog. 2003, 36, 183 – 194.
[14] D. J. Diller, R. Li, J. Med. Chem. 2003, 46, 4638 – 4647.
[15] R. H. Gunby, S. Ahmed, R. Sottocornola, M. Gasser, et al., J.
Med. Chem. 2006, 49, 5759 – 5768.
[16] A. Evers, G. Klebe, J. Med. Chem. 2004, 47, 5381 – 5392.
[17] S. Grneberg, QSAR Comb. Sci. 2005, 24, 517 – 526.
[18] A. Schafferhans, G. Klebe, J. Mol. Biol. 2001, 307, 407 – 427.
[19] Z. Zhou, M. Bates, J. D. Madura, Proteins 2006, 65, 580 – 592.
[20] A. Sali, T. L. Blundel, J. Mol. Biol. 1993, 234, 779 – 815.
i
2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Arch. Pharm. Chem. Life Sci. 2009, 342, 327 – 332
[21] R. A. Laskowski, M. W. MacArthur, D. S. Moss, J. M. Thornton, J. Appl. Crystallogr. 1993, 26, 283 – 291.
[22] M. J. Sippl, Proteins 1993, 17, 355 – 362.
[23] M. Wiederstein, M. J. Sippl, Nucleic Acids Res. 2007, 35,
W407 – 410.
[24] M. R. Berthold, N. Cebron, F. Dill, G. D. Fatta, et al., KNIME:
The Konstanz Information Miner; www.knime.org, 2006.
[25] S. B. Needleman, C. D. Wunsch, J. Mol. Biol. 1970, 48, 443 –
453.
[26] M. A. Larkin, G. Blackshields, N. P. Brown, R. Chenna, et al.,
Bioinformatics 2007, 23, 2947 – 2948.
[27] J. D. Thompson, D. G. Higgins, T. J. Gibson, Nucleic Acid Res.
1994, 22, 4673 – 4680.
[28] P. Rice, I. Longden, A. Bleasby, Trends Genet. 2000, 16, 276 –
277.
[29] H. J. C. Berendsen, D. van der Spoel, R. van Drunen, Comput. Phys. Commun. 1995, 91, 43 – 56.
[30] E. Lindahl, B. Hess, D. V. D. Spoel, J. Mol. Model. 2001, 7,
306 – 317.
[31] D. V. D. Spoel, E. Lindahl, B. Hess, G. Groenhof, et al., J.
Comput. Chem. 2005, 26, 1701 – 1718.
[32] W. L. Jorgensen, D. S. Maxwell, J. Tirado – Rives, J. Am.
Chem. Soc. 1996, 118, 11225 – 11236.
[33] W. L. Jorgensen, J. Tirado-Rives, J. Am. Chem. Soc. 1988, 110,
1657 – 1666.
[34] H. J. C. Berendsen, J. C. M. Postma, A. DiNola, J. R. Haak, J.
Chem. Phys 1984, 81, 3684 – 3690.
[35] B. Hess, H. Bekker, H. J. C. Berendsen, J. G. E. M. Fraaije, J.
Comput. Chem. 1997, 18, 1463 – 1472.
[36] M. Aoki, T. Yokota, I. Sugiura, C. Sasaki, et al., Acta Crystallogr. D Biol. Crystallogr. 2004, 60, 439 – 446.
[37] J. A. Bertrand, S. Thieffine, A. Vulpetti, C. Cristiani, et al., J.
Mol. Biol. 2003, 333, 393 – 407.
[38] R. Bhat, Y. Xue, S. Berg, S. Hellberg, et al., J. Biol. Chem.
2003, 278, 45937 – 45945.
[39] J. Allard, T. Nikolcheva, L. Gong, J. Wang, et al., “From genetics to therapeutics: the Wnt pathway and osteoporosis”. 1r0e
PDB entry, 2004.
[40] L. Meijer, A.-L. Skaltsounis, P. Magiatis, P. Polychronopoulos, et al., Chem. Biol. 2003, 10, 1255 – 1266.
[41] H. C. Zhang, L. V. Bonaga, H. Ye, C. K. Derian, et al., Bioorg.
Med. Chem. Lett. 2007, 17, 2863 – 2868.
[42] D. Shin, S. C. Lee, Y. S. Heo, W. Y. Lee, et al., Bioorg. Med.
Chem. Lett. 2007, 17, 5686 – 5689.
www.archpharm.com
Документ
Категория
Без категории
Просмотров
0
Размер файла
669 Кб
Теги
homology, mode, generation, evaluation, pfgsk
1/--страниц
Пожаловаться на содержимое документа