Letters in Drug Design & Discovery, 2006, 3, 675-682 675

Prediction of the Conformation of the Human P2X7 Receptor

Peter P. Mager* , Peter Illes and Horst Walther

Institute of Pharmacology and Toxicology, University of Leipzig, Saxony, Germany

Received June 15, 2006: Revised July 25, 2006: Accepted July 26, 2006
Abstract: There is a hope that inhibitors of the purinoreceptor P2X7, a ligand-gated ion channel, would have a
beneficial role in the therapy of inflammatory processes. Therefore, the P2X7 channel is a putative drug target
but neither X-ray nor NMR analyses have been made. It was attempted to predict the conformations of P2X
receptor subunits using homology-based comparative modelling and threading. The modelling could not be
carried out because of missing template proteins. The paper reviews a new method to predict the conformation
of proteins. The method is exemplified on the human P2X7 (h-P2X7) receptor. The coordinates of the spatial
structure of the h-P2X7 protein were determined by a profile-based neural network prediction. The
conformation was geometry optimised using the quantum chemistry RHF/3-21G minimal basic set and all-
atom molecular mechanics AMBER force field. A dielectric constant of ε = 3.5 was used to simulate the
lipophilic environment of the membrane-bound glycoprotein. The h-P2X7 protein has a number of interacting
peptide modules. The discovery of peptide module raises the possibility of exploiting structure-based
strategies to design h-P2X7 inhibitors.
Keywords: Structural bioinformatics, Proteomics, Drug target, Human P2X7 receptor, Ligand-gated ion channel.

INTRODUCTION State-of-the-art homology modelling techniques and

automatic ab initio prediction methods could also not solve
The membrane-embedded purinoreceptor P2X family
the problem (see Appendix). Therefore, a hybrid technique
belongs to the large class of membrane-embedded
has been proposed which consists of structural
glycoproteins. It has two transmembrane domains and a core
bioinformatics, secondary structure prediction, subsequent
of five extracellularly occurring disulfide bonds. Neither X-
geometry optimisation, and comparative sequence-function
ray nor NMR analyses have been made because of technical
analysis [6].
difficulties. Currently, seven different P2X receptor proteins
(P2X1 to P2X7) have been cloned. This review summarises the method using the h-P2X7
receptor subunit. It is shown that the new technique led to
The P2X7 subunit is an unusually large P2X protein
the discovery of interacting peptide modules. The modules
which can operate as trimeric, membrane-embedded,
may provide new insights for further experimentation and
monovalent-bivalent (Na+ , K+ , Ca2+ ) cation channel. It
serve as an explanatory guide for designing selectively acting
opens on binding of extracellularly occurring adenosine 5'-
h-P2X7 inhibitors.
triphosphate (ATP) to produce postsynaptic depolarisation
and excitation [1,2]. Additionally, agonist stimulation for
more than a few seconds causes a reversible expansion of the METHODOLOGY
P2X7 selectivity filter. The nonselective pore (with
The amino acid sequences of the human P2X7 (h-P2X7)
unknown stoichiometry and function) allows for the passage
receptor subunit have been taken from the Swiss Protein
of large molecules (≤ 900 Da). The human P2X7 (h-P2X7)
Sequence Database (Swiss-Prot, Swiss Institute of
protein is an ubiquitous receptor in many tissue types. It
Bioinformatics, and European Bioinformatics Institute,
differs from the other P2X subtypes in additionally coupling
EMBL). The primary accession number is Q99572. Among
to numerous other signalling cascades (e.g., a rapid release
the large number of approaches of secondary structure
of inflammatory cytokines), and influences actin
prediction, the PROFsec method was applied [7-11]. The
cytoskeleton organisation for many macrophage functions
results were used as starting point for a subsequent analysis.
including cell fusion, phagocytosis and apoptosis. The
To convert the two-dimensional sketch into a reasonable
P2X7 subunit plays a role as global thermistor of immune
conformational structure, the PROFsec results were uploaded
function, e.g., in astrocyte-mediated inflammation [3] of
into HyperChem program (Hypercube Inc., 419 Phillip
neurodegenerative diseases such as Alzheimer’s disease. The
Street, Waterloo, Ontario N2L 3X2, Canada). The side
misfolded Aβ(1-42) amyloid peptide of Alzheimer’s disease
chains of the h-P2X7 protein were added by building the
[4] can modulate the h-P2X7 receptor [5].
model in stereochemically preferred conformers using a
Unfortunately, sufficiently close targets to confidently library of protein bonds, angles, and torsion-angles
model the sequence of P2X proteins were not available. distributions. The tertiary and supersecondary folds are build
and then optimised from an assembly of secondary units by
*Address correspondence to this author at the Research Group of producing a single low-energy conformer.
Pharmacochemistry, Institute of Pharmacology and Toxicology, University
of Leipzig, D-04107 Leipzig, Härtelstr. 16-18, Saxony, Germany; The bonding states of the cysteines of h-P2X7 subunit
Tel: +49(0)041974636-2; Fax: +49(0)041974636-9, were determined by the CYSPRED program [12] and
E-mail: magp@medizin.uni-leipzig.de multiple alignment analysis [13,14] in combination with

1570-1808/06 $50.00+.00 © 2006 Bentham Science Publishers Ltd.

676 Letters in Drug Design & Discovery, 2006, Vol. 3, No. 10
experimentally obtained results reviewed elsewhere [6].
Multiple alignments of P2X sequences were carried out in
three steps: all sequences were compared to each other
(pairwise alignments), a dendogram was constructed
describing an approximate grouping of sequences by
similarity, and the final multiple alignment was carried out
using the dendogram as guide. The approaches used here are
the ClustalV and ClustalW algorithms. The PAM100
Dayhoff Mutation Data Matrix (ClustalV) and the BLOSUM
series (ClustalW) were applied to weight different pairs of
aligned amino acids. The results suggested five-disulfide
bond pairs of the h-P2X7 receptor: pair 1: C119-C168, pair
2: C129-C152, pair 3: C135-C162, pair 4: 216-226, pair 5:
C260-C269 [6]. Geometry optimisation occurred by the
HyperChem program (see above). It was assumed that the
geometry optimised h-P2X7 receptor subunit is monomeric,
unoccupied with any ligand, and charged. The intramolecular
hydrogen bonding forces were included in the calculation
process. A hybrid QC/MM-model was used. It consists of
the quantum chemistry (QC) RHF/3-21G minimal basic set
at the restricted Hartee-Fock level and all-atom molecular
mechanics (MM) force field. The partial charges were derived
from restrained electrostatic potential fits to ab initio
distributions obtained at the RHF/3-21G level (Gaussian94,
Gaussian Inc., Carnegie Office Park, Pittsburgh, PA 15106).
As MM force field, the AMBER96 force field (Assisted
Model Building and Energy Refinement 96) was applied.
The dielectric constant of ε = 3.5 simulated the behaviour of
the membrane-embedded h-P2X7 subunit, and the interior of
the protein. To avoid interactions between the nonbonded
atoms by a sharp cutoff which can cause discontinuities in
the potential surface, a smoothing applied from the inner
radius (10 Angstroms) to the outer radius (14 Angstroms)
was used. The structure was refined using a conjugate
gradient minimiser (Fletcher-Reeves modification of the
Polak-Ribiere method). Convergence was obtained when the
gradient root mean square RMS was RMS ≤ 0.05
kcal/Angstroms x mol. The results of the geometry
optimisation were checked by the Ramanchandran plot,
rotamer analysis, all-atom contact dots, and C(β) deviation
[6].
Naturally occurring posttranslational phosphorylation and
attachment sites were screened by using the knowledge-based
ScanPROSITE regular expression search via PROSITE's
interface [16]. The one-letter amino acid code is here used.
The general PROSITE signature SY-x(0,2)-[AL]-K-{P}
describes a subpeptide where SY must be present, it is
followed by no amino acid or two unspecified amino acids,
then it follows A or L, the subsequent K must be present,
and P is not allowed.
DOMAINS
The categorisation of the domains into two intracellular
cytoplasmic domains (ICDs), two transmembrane domains
(TMDs), and an extracellular domain (ECD) is as follows
[6]: the N-terminal ICD1: amino acid positions 1-25; the
TMD1: 26-48; the ECD: 49-330 (a globular glycoprotein
domain with an asymmetrical charge distribution and a
twisted loop, four putative nucleotide binding domains, and
five disulfide bridges); the TMD2: 331-350; and the C-
terminal ICD2: 351-595. The ICD2 includes a large, bulky,
Mager et al.
conformationally flexible fragment (amino acid subsequences
398-595) which may be termed "winking-cloud“. It is only
observed in P2X7 subunits.
The four putative nucleotide binding domains (NBDs)
with positively charged Arg and Lys residues that are
completely conserved in all species-dependent P2X subunits
(with exception of NDB2 in mouse P2X4) are as follows in
h-P2X7 [6]: NBD1: 63-67, NBD2: 188-194, NBD3: 292-
299, NBD4: 307-314.
The putative protein kinase C (PKC) phosphorylation
regulator domains are as follows: PKC1: 15-17 (conserved
in all species-dependent P2X subunits), PKC2: 28-30,
PKC3: 47-49, PKC4: 124-126, PKC5: 131-133, PKC6:
149-151, PKC7: 204-206, PKC8: 274-276, PKC9: 397-
399, PKC10: 467-469, PKC11: 490-492, and PKC12: 555-
557. The h-P2X7 subunit does not include tyrosine protein
kinase phosphorylation sites but a number of kinase-II
phosphorylation sites (6-9, 86-89, 390-393, 402-405, 420-
423, 508-511).
The extracellularly occurring N-glycosylation attachment
sites are N-gly1: 187-190 (strictly conserved in all species-
dependent P2X subunits with exception of mouse P2X4), N-
gly2: 202-205, N-gly3: 213-216, N-gly4: 241-244, and N-
gly5: 284-287 (Fig. 1).
Some regions in proteins represent classes of typically
aligned sequence motifs that identify potentially interesting
motifs of a protein family. They are used to characterise the
protein family because they encode protein folds and
functionalities more flexibly and powerfully than can single
motifs. The general signature of P2X receptors represents the
identifier code of evolutionarly selected functions and for the
maintenance of its spatial environment. The fingerprint is
commonly found in the whole P2X receptor family and
observed in the extracellular domain. Related to the h-P2X7
protein, the extracellular subsequence 249-275 represents the
ATP-gated P2X receptor signature [6]. Pro271 is a part of
the conserved sequence motif. Three neighboured residues of
Pro271 are also completely conserved: C-x(1)-P-y(1)-Y-z(1)-
F where x(1), y(1), and x(1) are unspecified amino acids.
Pro271 maintains probably an anti-misfolding capacity of
the P2X receptor signature.
Fig. 1 illustrates the predicted backbone structure of the
h-P2X7 protein and the location of important domains.
COMPARISON OF STRUCTURAL FOLDS WITH
KcsA
The framework for the molecular architecture of ion
channels is the KcsA protein. It has been found that the
prokaryotic two-transmembrane K+ -KcsA channel and the
P2X1, P2X3, and P2X4 receptor subtypes share a similar
structural fold [6,17,18] although the primary structures and
functions are quite different. The conclusion is also
supported by the result of the h-P2X7 receptor. It appears
that the common folds of these cation channels represent an
evolutionarily selected structure, independently of their
specific volume and function, that is, it seems that overall
three-dimensional shapes are better conserved during
evolution than sequences.
Conformation of h-P2X7 Letters in Drug Design & Discovery, 2006, Vol. 3, No. 10 677

Fig. (1). Result of structural bioinformatics applied to the membrane-embedded, monomeric h-P2X7 receptor. The result from
computational tools is a prediction and subject to further experimental verification. The structure is viewed parallel to the plane of
the membrane. The fully functional receptor is a trimeric ion channel. For visual clarity, the backbone structure of the amino acids is
illustrated. There are two intracellular domains, a N-terminal ICD1 (amino acid of the positions 1-25, and a C-terminal ICD2 (351-595,
it includes the "winking-cloud" in positions 398-595 which is a large, bulky, conformationally flexible fragment), two
transmembrane domains called TMD1 (26-48, and TMD2 (331-350), and an extracellular domain ECD (49-330, which is a globular,
twisted glycoprotein loop with an asymmetrical charge distribution and a disulfide core (marked by short yellow lines; pair 1: C119-
C168, pair 2: C129-C152, pair 3: C135-C162, pair 4: 216-226, pair 5: C260-C269). Colors: red, nucleotide binding domains NBD1
(63-67), NBD2 (188-194), NBD3 (292-299), NDB4 (307-314); yellow: PKC phosphorylation sites PKC1 (completely conserved, 15-
17), PKC2 (28-30), PKC3 (47-44), PKC4 (124-126), PKC5 (131-133), PKC6 (149-151), PKC7 (204-206), PKC8 (274-276), PKC9
(397-399), PKC10 (467-469), PKC11 (490-492), PKC12 (555-557); green: N-glycosylation attachment sites N-gly1 (completely
conserved, 187-190, not shown here due to visual clarity), N-gly2 (202-205, not shown here due to visual clarity), N-gly3 (213-216),
N-gly4 (241-244), N-gly5 (284-287); violet (left): signalling peptide (1-44); violet (right): cavity and gate/filter of the channel-
forming motif (332-344, 345-353); white: trafficking motif (383-387); dark blue: fingerprint (249-275). The kinase II
phosphorylation sites are not shown here.
PEPTIDE MODULES such interacting peptide modules of h-P2X7, some should
be mentioned:
The prototype of a protein that requires interacting
peptide modules is the acetylcholinesterase [15]. It was (1) There are many differences between the "orthodox ion
suggested that a functional P2X channel requires also a channels" and the membrane-embedded P2X ion
precise coupling between distinct regulator modules that are channel. The filter and the gate of P2X receptors are
important for signal transduction [6]. Among the number of commonly localised by an area at the end of the
678 Letters in Drug Design & Discovery, 2006, Vol. 3, No. 10
TMD2 and the beginning of the ICD2. It has been
proposed that the door/cavity module of the h-P2X7
channel is formed by the amino acids in positions
332-344, and the gate/filter module is located in the
positions 345-353 [6]. The channel of the P2X
receptors is gated by ATP but the ATP binding center
is outside the proper channel. The octapeptide module
NFRYAKYY of the NBD3 (292-299, the completely
conserved residues are marked in bold-face letters) lies
directly over the top of the door of the putative
channel-forming motif. It was suggested that ATP-
complexed NBD3 opens the channel, and an unbound
NBD3 closes the channel, that is, a periodic
conformational switch of the channel forming motif
may occur. The residues which put into the cavity
and contribute to hydrogen-bonding forces are
probably involved in the control of the transport of
hydrated cations through the channel. Dimeric models
for P2X2 and P2X3 channel constructs have been
proposed in literature [6,19].
(2) The second example deals with NBD2 (FTVLIKN,
188-194) which is in direct neighbourhood to the
extracellularly occurring regulator domain PKC7
(TTR, 204-206) and two N-glycosylation attachment
sites, the conserved N-gly1 (NFTV, 187-190) and N-
gly2 (NYTT, 202-205). Fig. 1 illustrates the areas.
Experimental results have shown that the priming
effect of nucleotides might be attributable to a
phosphorylation of the PKC sites of the ectodomain
[20].
It is commonly accepted that only the extracellular
domain of the P2X receptor has N-glycosylation
addition sites. The following rules were here used to
predict putative N-glycosylation sites. The peptide N-
x(1)-[ST] represents the general signature of N-
glycosylation with the site Asn. The presence of the
imino acid proline between Asn and Ser/Thr will
inhibit N-glycosylation. The hydrophilic
oligosaccharide chains are bound to the Asn residues
of N-glycosylation addition sites (Fig. 1) and modify
the surface properties, and consequently, the energy of
solvation, transport, and membrane-binding
properties of the protein. They may shield a large
selection of the protein surface and are probably
important for the expression at the cell surface of P2X
channels. It appears that the N-glycosylation addition
sites N-gly1 and N-gly2 of h-P2X7 are required for
the proper folding and arrangement of the
neighboured peptide modules, that is, they have a
stabilising effect on the structure of the NBD2 and
PKC7. As the N-glycosylation addition sites
contribute in part to NBD2 and PKC7, the effect of
single-point mutants can hardly be controlled. The
Asn residue N187 of the N-glycosylation attachment
site of h-P2X7 is homologous with N182 of rat
P2X2. Replacement of the double mutants 182/239
and 182/298 of Asn residues of the N-glycosylation
addition sites of rat P2X2 by Gln led to a
nonfunctional P2X2 subunit [21].
(3) The non-functionality of P2X mutants is also due to
trafficking problems. The conserved C-terminal
trafficking motif Y-x(2)-[KRQT]-K of the h-P2X7
Mager et al.
protein includes the amino acid positions YYRKK in
positions 383-387. The motif is in neighbourhood to
the PKC9 (TLK, 397-399) and kinase-II
phosphorylation sites (SIVE, 390-393) and probably
involved in the maintenance of cell-surface
expression. A number of other C-terminal trafficking
motifs that are not conserved must be envisaged, too.
(4) An example of a triad of sterically interacting peptide
modules is the nucleotide binding domain NBD4
(RTLIKAFG, 307-314), the N-glycosylation
attachment site N-gly5 (NVSL, 284-287), and the
PKC5 phosphorylation site (SDR, 131-133). Fig. 1
illustrates the modules. The interaction of ATP with
the NBD4 of the h-P2X7 receptor takes place in the
depth of the NBD4 pocket. The positive charge of the
Arg and Lys residues supply charge shielding of the
negatively charged phosphate groups. Further details
of an interaction of ATP with nucleotide binding
domains see literature [6,22]. As expected, R307Q
polymorphism causes loss of function of the h-P2X7
protein [23].
INTERNALISATION
Subdomains are also involved in internalisation of the
complete protein. Regulation of the protein by ligand-
mediated internalisation consists of the formation of a
complex between the target protein and other proteins (e.g.,
GRK-3-, clathrin-, β-arrestin-2-, and dynamin-mediated
internalisation of h-P2X7 into clathrin domains), formation
of vesicles, transport to endosomes, and release of the target
protein. Then, fully assembled proteins are delivered back to
their functional location, while misfolded proteins are
directed to lysosomes where their degradation occurs.
Unfortunately, the various proposed peptide clusters
[2,24,25] for an endocytosis of internalised P2X receptors
are not conserved.
The h-P2X7 protein starts probably its endocytosis with
the N-terminal intracellular ICD1 (Fig. 1). It includes the
only completely conserved PKC phosphorylation site
(PKC1, TNK, positions 15-17), the nonconserved PKC2
(TIK, 28-30), and a kinase-II phosphorylation site (SCSD,
6-9) which are parts of the signalling peptide (1-44). The
signalling peptide starts with the known methionine residue.
The most likely cleavage site of signalling peptides is
identified by an artificial neural network and a hidden
Marcovian model between the positions 44 and 45 [6]. Such
cleavage sites are one of the targets for designing specific
and efficient inhibitors. The T18 of P2X5 is homologous
with the T12 of P2X7, and it was shown that the T18A
mutant blocks P2X5 internalisation did not show ATP-
activated currents [26].
An example that extracellular domains may also
contribute to internalisation is probably the N-glycosylation
addition site N-gly4 (NFSD, 241-244; Fig. 1). Due to the
twisted nature of the globular coil of the extracellular
domain, the PKC3 phosphorylation site SDK (47-49) and
the signalling peptide are in steric neighbourhood to N-gly4
(Fig. 1).
Quite recently, it was reported that ATP downregulates
the P2X7 protein by ligand-mediated internalisation and a
Conformation of h-P2X7
pathway that does not depend on cation flux [27,28]. In
homology to the results of other proteins, it may be
speculated that an agonist-promoted down-regulation of P2X
receptors is also related to the N-gly4 subdomain. The N-
glycosylation site (N187) of the β2-adrenergic receptor
contributes to the internalisation during a long-term agonist
exposure, and the mutant N187D failed to display long-term
agonist-promoted down-regulation [29]. N-linked glycosyla-
tion sites of the H+ -K+ -ATPase β-subunit are critical for
folding and protect the internalised protein from targeting to
the degradation pathways [30].
THE "WINKING-CLOUD" OF THE P2X7 PROTEIN
Although the C-terminal portion might to be essential
for multiple functions, little information is known regarding
the structural motifs that are involved in receptor
localisation, trafficking, and signal transduction. It was
suggested that the fragment 573-590 of the ICD2 of h-P2X7
also contains several protein-protein and protein-lipid
interaction motifs [31], which may interact with areas that
are involved in ligand-induced gating of the channel [32] and
promote lipid-dependent endocytosis [33].
The subdomain 398-595 ("winking-cloud", lipid-
interaction motif) includes three PKC phosphorylation sites
(PKC10: 467-469, PKC11: 490-492, and PKC12: 555-557)
as regulator domains, three calcium/cyclic nucleotide-
independent kinase-II phosphorylation sites (402-405, 420-
423, 508-511), and two putative N-myristoylation addition
sites (504-509, 586-591), which are probably responsible for
irreversible lipid binding. Once added, the N-myristoyl
group increases the lifetime of the protein. Thus, the
myristoylated and surrounding motifs may contribute to the
targeting to specific intracellular compartments, reversible
interactions with transmembrane lipids and proteins, and
transport. Some of these peptide modules are able to interact
(Fig. 1). The intracellularly occurring P474 located within
the ICD2) s only present in the "winking cloud" of P2X7
(human, rat, mouse, frog). The signature is [SP]-P-x(1)-WC-
[QK].
The localisation of P2X7 to the cell surface is tightly
regulated by a C-terminal interaction motif resembling a
binding site for the subcellularly located human
lipopolysaccharide receptor [34]. The lipopolysaccharide
receptor (LPS-R) regulates upon nucleotide stimulation a
variety of cell signalling mechanisms. Experimental studies
have shown that h-P2X7 modulates the production of
numerous LPS-R-stimulated inflammatory mediators, e.g.,
proinflammatory cytokines, NF-κB-dependent inflammatory
cytokines, and macrophage production of TNF-α [34,35].
The "winking-cloud" of the P2X7 family (studied species:
human, rat, mice, frog) includes conserved fragments that
match with important areas of the LPS-R. Related to the
amino acid positions of human P2X7, the subdomains are as
follows: [LF]-[KR]-x(1,2)-F-x(1)-D (398-405), V-x(5)-G-
x(2,3,4)-Q-x(3)-G (412-426), [EQ]-L-x(6)-G (496-504), and
W-x(8,9)-FA-x(0,7)-[IVG] (556-568; x = 0 is valid for the
P2X7 family, V of [IVG] is present in frog P2X7, x = 7 and
G are found in PLS-R). The mutant E496A (replacement of
the charged, hydrophilic Asp in position 496 by the
hydrophobic Ala residue) or the mutant I568N (replacement
of the lipophilic Ile in position 568 by the hydrophilic Asn
Letters in Drug Design & Discovery, 2006, Vol. 3, No. 10 679
residue) lead to loss-function polymorphisms of the human
P2X7 protein [36,37]. The mutant E496A impairs the ATP-
induced IL-1β and IL-18 release from monocytes [38].
The LPS-binding domain was identified between
residues 573 and 590 which controls trafficking of h-P2X7
and, there fore, modulates availability and activity during an
inflammatory action [35]. The mutants R578E and K579E
of h-P2X7 led to trafficking defects [39] but it must be
stated that the dibasic amino acids R578 and K579
(positions related to h-P2X7) are only conserved in human,
rat and mouse P2X7, but in frog P2X7 they are missing and
replaced by E544 and T545. However, the number of
matches of the "winking-cloud" of the P2X7 family with
proinflammatory cytokines, such as the interleukins IL-1β,
IL-4, IL-6, IL-10, IL-13, IL-18, 1β (which must be activated
by caspase-1), is negligible.
PROLINE RESIDUES
The hydrophobic, nonpolar imino acid proline plays a
particular role in proteins, it forms a peptidyl-prolyl bond
that is incompatible with the peptide bond geometries. The
following extracellularly occuring proline residues are
strictly conserved in all species-dependent P2X subunits (the
position is related to h-P2X7): P96, P169, P227 (with
exception of mouse-P2X7 where Pro is replaced by Ser), and
P271. Pro169 of h-P2X7 deserves special attention. It is
included in the completely conserved subsequence WCP-
x(1)-E and contains the "proline-switch motif" E-x(1)-P that
has the potential to mediate conformer-specific ligand
binding resulting in an allosteric activation of the active site.
Like Gly, Asn, Ser and Asp, proline residues generally
prefer turns or loops and have α- and β-breaking influence
onto the local conformation [6]. An analysis of the
configuration of the modelled P2X3 has shown that the
peptide bonds nearly always have a trans configuration
which is energetically more favourable than cis [6]. The
frequency of a cis and trans configuration of proline residues
depends, at least in part, on the neighboured residues. In the
modelled P2X3 protein, the cis configuration of proline was
found more frequently [6].
The high activation barrier (14-24 kcal/mol) for the
slowly occuring cis/trans-isomerisation (time constants of 10
to 100 sec) through a planar transition state could be
involved in the rate-limiting step of slow-folding kinetics of
proteins, adds probably to their flexibility, and assists in
conformational changes due to changes in the solvent-
accessible, extracellular surface area [40-43]. While most
structures reveal peptidyl-prolyl bonds that adopt either cis
or trans configuration, other exhibit conformational
heterogeneity about one or more peptidyl-prolyl bonds. It
should be added here that the possible existence of both
isomeric forms of the native conformation may only be
observed in solution and in membrane-embedded protein but
not in the crystal structures. Furthermore, crystal structures
may not be of sufficient resolution to provide evidence for
occurring cis/trans-isomerisation. An inhibitor of the
isomerisation is the azetidine-2-carboxylic acid.
The peptidyl-prolyl-cis/trans isomerase family isomerises
the [TS]-x(0)-P motif which is a target for a phosphorylation
of Thr/Ser by proline-directed protein kinases. The reversible
680 Letters in Drug Design & Discovery, 2006, Vol. 3, No. 10
phosphorylation is also called proline-directed
phosphorylation, it is a major signalling mechanism in
diverse cellular processes such as cell proliferation and
differentiation [40,41,43]. The P2X7 subunit does not
include SP, TP, or the reverse Pro-Thr/Ser, in contrast to the
other P2X subunits (S289P290 in P2X4, S86P287 in
P2X1; T18P19 in P2X1, T30P31 in P2X2, T17P18 in
P2X4, T110P11 in P2X5, T109P11 in P2X6; P330T331 in
P2X1, P340T341 in P2X2; the positions are related to
human proteins). However, the strictly conserved Pro96
(related to h-P2X7) of the complete P2X family is part of
the [TS]-x(1)-P signature of the P2X7 subfamily where x(1)
denotes a single apolar amino acid. The function of x(1) is
still an open question (hydrophobic distance spacer?).
PERSPECTIVES IN MEDICINAL CHEMISTRY
This review dealt with a hybrid approach consisting of
structural bioinformatics, secondary structure prediction,
subsequent geometry optimisation, comparative sequence-
function analysis, and experimental findings. It was
surprising to see that the predicted structural P2X1, P2X2,
P2X3, and P2X4 structures [6,17-19,22] agreed with the
P2X models proposed from biochemical, neurophysio-
logical, pharmacological, and mutation experiments [1-
2,20,44-46]. The same conclusion can be drawn with respect
to P2X7 [47].
Lead molecules for h-P2X7 inhibitors are reviewed
elsewhere [50]. Originally developed as inhibitor of the
calcium/calmodulin-dependent protein kinase-II, 1-N,O-
bis(1,5-isoquinolinesulfonyl)-N-methyl-L-tyrosyl)-4-phenyl-
piperazine derivatives (KN-62) have been recognised as
selectively acting antagonists of the h-P2X7 protein. KN-62
analogues block currents in cells expressing the h-P2X7
receptor but have little effect on rat P2X7 [48-51]. However,
neither of their actions on h-P2X7 appears to be related to
calmodulin or the calcium/calmodulin-dependent protein
kinase-II. Quantitative structure-activity relationships of KN-
62 analogues [6] have shown the role of steric, electronic,
and hydrogen-bonding properties of the substituents while
lipophilic parameter seem to play a less important role.
Potent h-P2X7 inhibitors are also 4,5-diarylimidazolines.
The search for novel 4,5-diarylimidazolines (Fig. 2) occurred
by an extensive use of high throughout screening
approaches, utilising diverse compound library sources [52].
As regressand, the unscaled 50% inhibitor concentration was
employed (Y = IC50). Because of the presence of flagged
observations, a major problem in developing a QSAR
equation was to select suitable chemical descriptors from a
pool of variables. The "best way" was to employ a
simultaneous one-regression/one-observation leaving-out
resampling of regression analysis [53,54]. The final
regression equation was equal to
IC50 = -0.35 - 1.20(σ+ ) - 1.18(pKH) - 1.42(Es(ortho))
(sample size of the training set: n = 26) where σ+ is an
electronic substituent constant, pKH is the negative
logarithm of the equilibrium constant KH of hydrogen
binding between substituted benzenes and phenol as
hydrogen donor, and Es(ortho) is Taft's steric constant for
ortho substituents. The squared multiple correlation
coefficient R2 = 0.62 was larger than the critical quantile g
Mager et al.
of the maximum-likelihood criterion (g = 0.29, 5%
significance level). The standard error of the regressand was
0.67, and the test statistics for examining the null
hypothesis of zero regression coefficients were TS1 = 2.68,
TS 2 = 2.93, and TS3 = 2.32 (critical quantile g = 2.07 at
the 5% significance level). There was no significant
collinearity and multicollinearity, no high-leverage point,
and no influential point. Other parameters, such as the
logarithm of the distribution coefficient P (octanol/water, pH
7.4), and an indicator variable XD which parameterises the
bridges X = CH2CH2 (XD = 1) and CH2 (XD = 0) of the
lead structure (Fig. 1), were insignificant. There was a hope
to improve the fit by a generalised-regressionartificial-neural
network (GRNN) using an unconstrained genetic algorithm.
In contrast to the regression analysis, the genetic GRNN
ignored relevant and added noisy descriptors although the
goodness-of-fit criterion was strongly improved due to
overfitting (R2 = 0.98). Consequently, the genetic GRNN
was considered as supplementary tool but could not replace
the traditional QSAR methodology.
NH
N
X
R
Fig. (2). 4,5-diarylimidazolines as inhibitors of the human
P2X7 receptor, a potential target for a therapy of Alzheimer’s
disease and rheumatoid arthritis.
Although the so-called computational molecular docking
problem is far from being solved and there are limitations in
each docking strategy, significant progress has been made
[55], and the predicted conformation of the ligand-gated h-
P2X7 protein may be useful for designing experiments
directed towards the understanding of blocking mechanisms.
For the large h-P2X7 protein, subdomain docking is
probably the only rational way. In particular, the interacting
peptide modules are potential targets. Their discovery raises
the possibility of exploiting structure-based strategies to
develop h-P2X7 inhibitors that are of paramount
pharmacological importance.
APPENDIX: MODEL LIMITATIONS
To avoid semantic confusion, it should be noted that
"prediction" is defined in a probabilistic sense. Matches to
generic rules do not mean "this is true" but rather "this
might be true". Only biological and chemical knowledge can
determine whether or not these predictions are meaningful.
Methods of secondary structure prediction are based on
algorithms that concentrate on predicting the most suitable
conformation on the basis of the properties and interactions
of amino acids. As underlying theories are usually applied:
information and statistical theories, homology method, k-
nearest-neighbours (kNN), hierarchical (HNN) and
bidirectional recurrent (BRNN) methodologies of artificial
Conformation of h-P2X7
neural network (ANN) analyses, deterministic methods of
combinatorial optimisation, empirical rules, and hybrid
techniques. The approaches vary in their average accuracy but
the majority of the predicted secondary structure cluster do
not differ markedly.
The following methods were used for comparing the
results of secondary structure prediction: GOR (versions I
and IV of the Garnier approach), SOPMA (self-optimised
prediction with alignments of homologous sequences),
APSSP and APSSP2 (combination of nearest neighbour
with ANNs and multiple alignment analysis), SSpro2 and
SSpro8 (BRNN), PSIpred (prediction based on position-
specific scoring matrices incorporating four feedforward
ANNs), SAM-HMM and SAM-T99-2d (prediction by
combining kNN and multiple sequence alignment),
PROFsec (a profile-based ANN prediction from multiple
sequence alignments), PSA (probabilistic discrete state-space
model with optimal filtering and smoothing methods), and
DSSP (pattern recognition of hydrogen-bonded and
geometrical features). An example of hybrid methods is a
multivariate linear regression combination (MLRC) which
consists of the SOPMA, GORIV, and SIMPA96 methods.
The majority of the programs tended to overestimate the
role of α-helices in transmembrane domains because the
generic algorithms were intentionally constructed from a set
of a priori model assumptions. For example, low-resolution
experiments suggested that most helices span over 17-25
residues and most loops between two helices are longer than
15 residues. For high-resolution data it was found that only
50% fall into this interval and that 50% of the loops were
shorter than 10 residues. Secondary structure prediction
methods use often low-resolution input data, however.
Furthermore, the hypothesis that 80% of proteins have α-
helical transmembrane domains have been generalised. Thus,
the frequency of helices in TMDs is overestimated. The
PROFsec method simulates the influence of α-helical
breaking amino acids in a more satisfactory degree.
It was believed at first glance that ATP-binding proteins
with two transmembrane domains can be used as putative
templates in a homology-based molecular modelling
process. Many modern tools of homology-based molecular
modelling are Web server based as opposed to installable
software for local use key. Unfortunately, sufficiently close
targets were not available (e.g., ProModII of the
SwissModel Server, Aug 22, 2001, and Jun 02, 2003; 3D-
JIGSAW, PSI-BLAST, and CPHmodels Servers, Nov 05,
2002; SDSC Server, Nov 13, 2002; Superfamily HMM
Library and Genome Assignments Server, Jun 02, 2003).
Other programs presented outputs but indicated that the fold
recognition results were probably not worthy of attention
(3D-PSSM Server, Nov 05, 2002; mGenTHREADER, Nov
06, 2002). Modelling techniques such as Modeller or the
respective modules of SYBYL or MOE also could not solve
the problem.
A referee has stated that some ab initio methods deal
successfully with the 3D packing of secondary structure
elements. The Monte Carlo fragment insertion method for
tertiary structure prediction (ROSETTA) and its improved
version did not give results, however. This first fully
automatic ab initio prediction method of the I-
sites/HMMstr/ROSETTA server has been used (Jun 12,
Letters in Drug Design & Discovery, 2006, Vol. 3, No. 10 681
2006) but there was no success because the input sequence
was too long so that Rosetta simulations and HMMstr-CM
contact maps have been disabled. The limit is less than 150
amino acids.
The P2X subunits can form oligomers that are viewed as
probably trimeric assemblies consisting of homomeric
and/or heteromeric subunits which differ in their biological
and chemical properties. With respect to the complete P2X7
subunit, the modelling of trimeric protein is not possible
due to the large number of amino acids. Furthermore, the
sites of contacts are unknown. Therefore, the proposal of a
referee to imagine the possible trimeric structure cannot be
realised at the present state of knowledge.
Molecular dynamics (MD) simulation can add explicit
water molecules around a protein, the use of a periodic box
prevents solvent molecules from diffusing away or
evaporating during a MD calculation. Unfortunately, the
P2X7 molecule is too large (bounding box 103, 92, 274
Angstroms) to solvate in a box containing the TIP3P model
of water molecules, this is no drawback, however. Compared
with the use of continuum electrostatics, MD is often
insufficient to describe solvation effects due to the lack of
description for electronic polarisations. Although the
constant-dielectric and distance-dependent dielectric
functions are not based on rigorous physics, they provide a
relatively fast evaluation of solvation effects and the second
derivatives of the electrostatic energy. The native fold in
solutions was simulated by varying the dielectric constant ε
(1-4 scale factors: electrostatic 0.5, van-der-Waals 0.5). The
protein is dehydrated if ε = 1 is used. A model that captures
the ionisation properties of surface residues of proteins (ε =
42) is the solvent-accessibility model. In the Poison-
Boltzmann model, the dielectric constant is much lower (ε =
4). A dielectric constant of ε = 3.5 simulates an apolar
environment, the behaviour of the membrane-embedded
P2X7 subunit, and the interior of the protein. When
conformational relaxation is treated explicitly, ε = 3.5 yields
reasonable results, especially, when the neutral and ionised
states of the groups are treated separately. The buried charged
groups make minimal contact with side chain or backbone
polarised atoms. Thus, ε = 3.5 was thought to include
contributions by relaxation of permanent dipoles in addition
to electronic polarisability.
To build dimeric P2X channel constructs of the 22
amino acids that form the proper channel, distance restraint-
based protein-protein docking was applied [6,18]. The
resulting constructs were again geometry optimised [6]. The
proposed dimeric model satisfied the rules of a general
architecture (length, diameters) of P2X ion channels but the
transport of hydrated cations and counterions in the central
cavity could not be modelled. However, the building of
trimeric channel constructs failed. Furthermore, to simulate a
cation flux, the dielectric constant of the open P2X ion
channel (ε = 3.5) must be converted to the local dielectric
behaviour of intra-channel water (ε = 80) ad bulk solution
conditions of hydrated cations (ε = 82, ionic strength 0.1M;
ε = 97, ionic strength 0.01M). As consequence, a channel
construct must be viewed with at least two dielectric
regions. When cations are included in the gate, the pattern of
proton sharing is altered, and the ionisation state of the
channel side chains is changed. The solution of these
682 Letters in Drug Design & Discovery, 2006, Vol. 3, No. 10 Mager et al.

problems is computationally difficult and further research is [28] Kochukov, M. Y.; Ritchie, A. K. Am. J. Physiol. Cell Physiol.,
2004, 287, C992-C1002.
needed to develop suitable algorithms. [29] Mialet-Perez, J.; Green, S. A.; Miller, W. E.; Liggett, S. B. J.
In summary: template proteins that confidently model Biol. Chem., 2004, 279, 38603-38607.
[30] Vagin, O.; Turdikulova, S.; Sachs, G. J. Biol. Chem., 2004, 279,
the sequence of P2X7 receptor protein, and novel ab initio 39026-39034.
methods that were able to present predictions of medium- [31] Denlinger, L. C.; Fisette, P. L.; Sommer, J. A.; Watters, J. J.;
sized and larger proteins, are not available up-to-now. Under Prabhu, U.; Dubyak, G. R.; Proctor, R. A.; Bertics, P. J. J.
these circumstances, it is sometimes better to use a simple Immunol., 2001, 167, 1871-1876.
method that works (PROFsec and subsequent geometry [32] Klapperstück, M.; Büttner, C.; Schmalzing, G.; Markwardt, F. J.
Physiol. (Lond.), 2001, 534, 25-35.
optimisation) than state-of-the-art methods that cannot work [33] Markwardt, F. Letter to the author, Sep 11, 2003.
up-to-now for large proteins. [34] Denlinger, L. C.; Angelini, G.; Schell, K.; Green, D. N.;
Guadarrmara, A. G.; Prasbhu, U.; Coursin, D. B.; Bertics, P.;
Hogan, K. J. Immunol., 2005, 174, 4424-4431.
REFERENCES [35] Ferrari, D.; Pizzirani, C.; Adinolfi, E.; Memoli, R.; Curti, A.;
Idzko, M.; Panther, E.; Di Virgilio, F. J. Immunol., 2006, 176,
[1] North, R. A. Physiol. Rev., 2002, 82, 1013-1067. 3877-3883.
[2] Vial, C.; Roberts, J. A.; Evans, R. J. Trends Pharmacol. Sci., 2004, [36] G, B. J.; Zhang, W. Y.; Worthington, R. A.; Sluyter, R.; Dao-
25, 487-493. Ung, L. P.; Petrou, S.; Barden, J. A.; Wiley, J. S. J. Biol. Chem.,
[3] DiVirgilio, F. Immunol. Today, 1995, 16, 524-528. 2001, 276, 11135-11142.
[4] Mager, P. P. Med. Res. Rev., 1998, 18, 403-430. [37] Wiley, J. S.; Dao-Ung, L. P.; Li, C.; Shemon, A. N.; Gu, B. J.;
[5] Rampe, D.; Wang, L.; Ringheim, G. E. J. Neuroimmunol., 2004, Smart, M. L.; Fuller, S. J.; Barden, J. A.; Petrou, S.; Sluyter, R. J.
147, C571-C581. Biol. Chem., 2003, 278, 17108-17113.
[6] Mager, P. P.; Weber, A.; Illes, P. Curr. Topics Med. Chem., 2004, [38] Sluyter, R.; Dalitz, J. G.; Wiley, J. S. Genes Immun., 2004, 5, 588-
4, 1657-1705. 591.
[7] Pollastri, G.; Przybylski, D.; Rost, B.; Baldi, P. Proteins, 2002, 47, [39] Denlinger, L. C.; Sommer, J. A.; Parker, K.; Gudipaty, L.;
228-235. Fisette, P. L.; Watters, J. W.; Proctor, R. A.; Dubyak, G. R.;
[8] Chen, C. P.; Rost, B. Protein Sci., 2002, 11, 2766-2773. Bertics, P. J. J. Immunol., 2003, 171, 1304-1311.
[9] Chen, C. P.; Rost, B. Appl. Bioinformatics, 2002, 1, 21-35. [40] Lu, K. P.; Liou, Y. C.; Zhou, X. Z. Trends Cell Biol. 2002, 12,
[10] Rost, B. J. Struct. Biol., 2001, 134, 204-218. 164-172.
[11] Rost, B.; Sander, C. Proteins, 1994, 19, 55-72. [41] Lu, K. P. Trend Biochem. Sci., 2004, 29, 200-209.
[12] Fariselli, P.; Riccobelli, P.; Casadio, R. Proteins, 1999, 36, 340- [42] Andreotti, A. H. Biochemistry, 2003, 42, 9515-9524.
346. [43] Schutkowski, M.; Bernhard, A.; Zhou, X. Z.; Shen, M.; Reimer,
[13] Dayhoff, M. O.; Barker, W. C.; Hunt, L. T. Meth Enzymol., 1983, U.; Rahfeld, J.-U.; Lu, K. P.; Fischer, G. Biochemistry, 1998, 37,
91, 524-545. 5566-5575.
[14] Corpet, F. Nucl. Acids Res., 1988, 16, 10881-10890. [44] Khakh, B. S. Nature Rev.Neurosci ., 2001, 2, 165-174.
[15] Mager, P. P.; Weber, A. Drug Des. Discov., 2003, 18, 127-150. [45] Ennion S. J.; Evans, R. J. Mol. Pharmacol., 2002, 61, 303-311.
[16] Falquet, L.; Pagni, M.; Bucher, P.; Hulo, N.; Sigrist, C. J.; [46] Clyne, J. D.; Wang, L.-F.; Hume, R. I. J. Neurosci., 2002, 22,
Hofmann, K.; Bairoch, A. Nucl. Acids Res., 2002, 30, 235-238. 3873-3880.
[17] Mager, P. P.; Weber, A.; Sanchez, L.; Wirkner, K.; Illes, P. [47] Kim, M.; Jiang, L.-H.; Wilson, H. L.; North, R. A.; Surprenant,
Protein Pept. Lett., 2006, 13, 77-81. A. EMBO J., 2001, 20, 6347-6358.
[18] Mager, P. P.; Weber, A.; Walther, H.; Wirkner, K.; Illes, P. [48] Gargett, C. E.; Wiley, J. S. Br. J. Pharmacol., 1997, 120, 1483-
Curr. Proteomics, 2005, 2, 319-324. 1490.
[19] Mager, P. P.; Weber A.; Illes, P. Med. Chem., 2005, 1, 109-115. [49] Baraldi, P. G.; Del Carmen Nunez, M.; Morelli, A.; Falzoni, S.;
[20] Wirkner, K.; Stanchev, D.; Köles, L.; Klebinger, M.; Dihazi, Di Virgilio, F.; Romagnoli, R. J. Med. Chem., 2003, 46, 1318-
H.; Fleming, V.; Evans, R. J.; Fürst, S.; Mager, P. P.; Eschrich, 1329.
K.; Illes, P. J. Neurosci., 2005, 25, 7734. [50] Baraldi, P. G.; Di Virgilio, F.; Romagnoli, R. Curr. Topics Med.
[21] Torres, G. E.; Egan, T. M.; Voigt, M. M. Biochemistry, 1998, 37, Chem., 2004, 4, 1707-1717.
14845-14851. [51] Baxter, A.; Bent, J.; Bowers, K.; Braddock, M.; Brough, S.;
[22] Mager, P. P.; Illes, P. Expert Opin. Drug Discov., 2006, in press. Fagura, M.; Lawson, M.; McInally, T.; Mortimore, M.;
[23] Gu, B. J.; Sluyter, R.; Skarratt, K. K.; Shemon, A. N.; Dao-Ung, Robertson, M.; Weaver, R.; Webborn, P. Bioorg. Med. Chem.
L. P.; Fuller, S. J.; Barden, J. A.; Clarke, A. L.; Petrou, S.; Lett., 2003, 13, 4047-4050.
Wiley, J. S. J. Biol. Chem., 2004, 279, 31287-31295. [52] Merriman, G. H.; Ma, L.; Shum, P.; McGarry, D.; Volz, F.;
[24] Smith, F. M.; Humphrey, P. A.; Murrell-Lagnado, R. D. J. Sabol, J. S.; Gross, A.; Zhao, Z.; Rampe, D.; Wang, L.; Wirtz-
Physiol., 1999, 520, 91-99. Brugger, F.; Harris, B. A.; Macdonald, D. Bioorg. Med. Chem.
[25] Feng, Y.-H.; Wang, L.; Wang, Q.; Li, X.; Gorodeski, G. I. Am. Lett., 2005, 15, 435-438.
J. Physiol. Cell. Physiol., 2005, 288, C1342-C1356. [53] Mager, P. P.; Sanchez, L. Curr. Comput.-Aided Drug Design,
[26] Jensik, P.; Thomas, C. Pflügers Arch. Eur. J. Physiol., 2002, 444, 2005, 1, 163-177.
795-800. [54] Mager, P. P. John Wiley & Sons Inc., New York, 1991.
[27] Hiken, J. F.; Steinberg, T. H. Am. J. Physiol. Cell Physiol., 2004, [55] Halperin, I.; Ma, B.; Wolfson, H.; Nussinov, R. Proteins Struct.
287, C403-C412. Funct. Genetics, 2002, 47, 409-443.