Date post: | 19-Dec-2015 |
Category: |
Documents |
View: | 213 times |
Download: | 0 times |
4. Lecture SS 20005
Cell Simulations 1
V4: Area 3 – interaction networks
Area 1: Systems of coupled differential equations (T. Geyer)
Area 2: Metabolic networks
- flux balance analysis - MILP
- elementary flux modes, linear algebra background
- apply software FluxAnalyzer to test systems
Area 3: Graph networks – interaction networks
- network of proline-rich sequences and adaptor domains
- apply software Cytoscape to test systems
Area 4: Spatial modelling of cellular systems (T. Geyer)
4. Lecture SS 20005
Cell Simulations 2
SH3 as a structural motif in SRC tyrosine kinase
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
The importance of proline-rich motifs in biology is highlighted by the finding that proline-rich regions are the most common sequence motif in the Drosophila genome and the second most common in the Caenorhabditis elegans genome.
The number of defined protein domains that recognize proline-rich motifs has expanded considerably in recent years to include such common motifs as Src homology 3 (SH3), WW (named for a conserved Trp-Trp motif), and Enabled/VASP homology (EVH1, also known as WASP homology 1 or WH1) domains, as well as other proline-binding domains.
The number of domains in an organism roughly corresponds to its perceived complexity (Table 1).
Domains that bind proline-rich motifs are critical to the assembly of many intracellular signaling complexes and pathways.
4. Lecture SS 20005
Cell Simulations 3
First crystal structures of SH3
First X-ray structure of a SH3 domain in 1992. Musacchio,A., Noble, M., Pauptit, R.,Wierenga, R. and Saraste, M. (1992):Crystal structure of a Src-homology 3 (SH3) domain. Nature 359, 851-855
First X-ray structure of a complex of SH3 with proline rich ligand in 1994:Musacchio,A., Saraste, M. andWilmanns, M. (1994): High-resolution crystal structures of tyrosine kinase SH3 domains complexed with proline-rich peptides. Nature Struct. Biol. 1, 546-551
4. Lecture SS 20005
Cell Simulations 4
Function of proline recognition domains
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
Proline recognition domains are usually found in the context of larger
multidomain signaling proteins.
Their binding events often direct the assembly and targeting of protein
complexes involved in - cell growth- cytoskeletal rearrangements- transcription- postsynaptic signaling- and other key cellular processes
In addition, these interactions can play a regulatory role, often through
autoinhibitory interactions that are alleviated by competing binding events.
4. Lecture SS 20005
Cell Simulations 5
Example: negative regulation of T-cell receptor by adaptor domains
Examples of negative regulation by
adaptor molecules and adaptor domains
are depicted.
a Allosteric inhibition by the adaptor
domains of SRC-family kinases. The
SRC-homology 2 (SH2) domain of SRC-
family kinases binds to the carboxy-
terminal phosphotyrosine residue,
thereby restricting substrate accessibility
and kinase activity. The SH3 domain has
also been shown to regulate SRC kinase
activity through intramolecular
interactions that create an inducible
'snap lock', which is dependent on
interdomain hinge regions as well. On
dephosphorylation of the C-terminal
tyrosine by the CD45 phosphatase, the
adaptor domains are released and result
in activation of the kinase.
Nature Reviews Immunology 1; 95-107 (2001)
b Recruitment of negative effector molecules to their
substrates. In unstimulated T cells, raft-associated
PAG/CBP is constitutively tyrosine phosphorylated and
associates with the SH2 domain of CSK, bringing CSK
into close proximity to its substrates (SRC-family PTKs)
at the plasma membrane. Following TCR stimulation,
PAG/CBP is dephosphorylated, resulting in the release
of CSK from the membrane and relieving SRC-family
kinases from CSK phosphorylation-mediated inhibition.
Evidence also indicates that PAG/CBP might also
regulate CSK activity independently of its ability to
recruit CSK to lipid rafts.
4. Lecture SS 20005
Cell Simulations 6
SH3 as a structural motif in SRC tyrosine kinase
http://jkweb.berkeley.edu/external/pdb/1997/hck/hck.html
4. Lecture SS 20005
Cell Simulations 7
A proline-driven conformational switch within the Itk SH2 domain
Mallis, Brazin, Fulton, Andreotti, Structural characterization of a proline-driven conformational switch within the Itk SH2 domain, Nat. Struct. Biol. 9, 900 - 905 (2002)
NMR structures of the cis and trans Itk SH2 conformers. a, Stereo view of 20 low energy structures of the cis (coral) and trans (turquoise) conformations of the Itk SH2 domain. Backbone heavy atoms within the secondary structural elements over the entire sequence were used for superpositions.
b, Ribbon diagrams of the energy minimized average structures of the cis (left) and trans (right) conformers. Secondary structural elements and ligand-binding pockets are labeled in (a,b) according to standard nomenclature for SH2 domains8. Pro 287 is labeled in each structure.
c, Sequence of the Itk SH2 domain and sequence alignment of the CD loop regions in the SH2 domains of several tyrosine kinases. The residues that give rise to nondegenerate chemical shifts2 are bold and underlined, and Pro 287 is labeled.
e, Overlay of the energy minimized average structures of the cis (coral) and trans (turquoise) conformers. Expanded views of the CD loop (left), the central -sheet (right) and the BG loop regions (middle) are shown.
4. Lecture SS 20005
Cell Simulations 8
Structural differences between cis and trans isomers
Mallis, Brazin, Fulton, Andreotti, Structural characterization of a proline-driven conformational switch within the Itk SH2 domain, Nat. Struct. Biol. 9, 900 - 905 (2002)
Structural differences between the cis and trans Itk SH2 domain provide a basis for conformer-specific binding to the Itk SH3 domain.a, A backbone ribbon representation of the Itk cis SH2 domain with the Itk polyproline peptide (KPLPPTP shown in white) superimposed on the structure. The polyproline peptide residues are labeled using the one letter amino acid code and are numbered consecutively. In previously determined peptide–SH3 structures10, 11, Lys 1 (K1), Leu 3 (L3), Pro 4 (P4), Thr 6 (T6) and Pro 7 (P7) directly contact the SH3-binding pocket, whereas Pro 2 (P2) and Pro 5 (P5) do not. SH2 domain residues that are involved in SH3 binding (as determined by chemical shift mapping) are highlighted in yellow and labeled with bold-letter font. Putative correlations between SH2 residues and the canonical polyproline peptide are as follows: Arg 332-Lys 1, Val 330-Leu 3, Thr 279-Pro 4, Cys 288-Thr 6 and Ile 282-Pro 7. This model was arrived at by initial superposition of the basic peptide residue Lys 1 with Arg 332 of the SH2 domain. This assignment is based on the large chemical shift perturbation observed for Arg 332 upon addition of SH3 ( 15N = 0.407 p.p.m. and 1H = 0.196 p.p.m.) and the observation that previously determined SH3–ligand complexes, combined with mutational analyses, have shown that a stabilizing interaction involving a basic amino acid side chain and a conserved acidic site within the SH3 domain is required for SH3 ligand binding32. Subsequently, using Arg 332 as an anchor, the polyproline peptide structure was rotated over the surface of the cis SH2 domain to assess whether the cis SH2 residues that mediate binding to the SH3 domain may be similar in geometric arrangement and chemical nature to the polyproline peptide side chains known to contact the SH3 binding surface in SH3–peptide complexes11, 32. b, Isomerization to the trans SH2 structure disrupts the putative binding site on the cis SH2 domain, which is consistent with the inability of trans SH2 to bind to the SH3 domain. Residue coloring and labeling same as shown in (a).
4. Lecture SS 20005
Cell Simulations 9
Why are proline-rich sequences special?
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
Repetitive proline-rich sequences are found in many proteins and in many cases are
thought to function as docking sites for signaling modules.
Why might proline be singled out for recognition by so many key protein-protein interaction
modules?
Several features of proline distinguish it from the other 19 naturally
occurring amino acids (Fig. 1A): - the unusual shape of its pyrrolidine ring- the conformational constraints on its dihedral angles imposed by
this cyclic side chain- its resulting secondary structural preferences- its substituted amide nitrogen, - and the relative stability of the cis isomer in a peptide bond.
Each recognition domain exploits some combination of these distinctive features of proline
in order to achieve specific binding to proline-rich regions.
4. Lecture SS 20005
Cell Simulations 10
Polyproline type II (PPII) helices
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
One feature of proline-rich motifs that is frequently used in binding to signaling domains is their propensity to form a polyproline type II (PPII) helix.
The PPII helix is an extended left-handed helical structure with three residues per turn and an overall shape resembling a triangular prism (Fig. 1B). A combination of steric and hydrogen-bonding properties of proline-rich motifs is thought to contribute to its preference for this unusual secondary structure.
Two features of the PPII helix make it a useful recognition motif:
First, in this structure both the side chains and the backbone carbonyls point out from the helical axis into solution at regular intervals (Fig. 1B). The lack of intramolecular hydrogen bonds in the PPII structure, due largely to the absence of a backbone hydrogen-bond donor on proline, leaves these carbonyls free to participate in intermolecular hydrogen bonds. Thus, both side chains and carbonyls can easily be “read” by interacting proteins.
Second, because the backbone conformation in a PPII helix is already restricted, the entropic cost of binding is reduced. Nearly all of the domains described here bind their ligands in a PPII conformation. Many of the interactions with the PPII helical ligand involve aromatic residues. The planar structure of aromatic side chains appears to be highly complementary to the ridges and grooves presented on the PPII helix surface.
4. Lecture SS 20005
Cell Simulations 11
Properties of PPII helices
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
(B) Schematic and structural representation of a PPII helix. The helix has twofold pseudosymmetry: A rotation of 180° about a vertical axis leaves the proline rings and the carbonyl oxygens at approximately the same position. The Protein Data Bank (PDB) accession code for the poly-(l)-proline structure shown is 1CF0.
(C) A view down the axis of the PPII helix highlighting the position of the carbons in the xP dipeptide. In the “x” position that requires C-substitution (blue), the primary recognition element is the β carbon, whereas in the “P” position that requires N-substitution (red), the primary recognition element is the δ carbon that is unique to proline.
4. Lecture SS 20005
Cell Simulations 12
Polyproline type II (PPII) helices
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
One interesting structural feature of the PPII helix is that it has twofold rotational
pseudosymmetry: Side chains and backbone carbonyls are displayed with
similar spacing in either of the two N- to C-terminal orientations (Fig. 1B).
This feature may explain why many proline-binding domains are observed to
bind ligands in two possible orientations, a property unique among
characterized peptide recognition modules.
In principle, this orientational flexibility could play an important role in domain
function. For example, one could imagine a complex in which binding in one
orientation could be activating, whereas binding in the opposite orientation could
be inhibitory. However, such an orientational switching role has not been
demonstrated.
4. Lecture SS 20005
Cell Simulations 13
Polyproline type II (PPII) helices
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
Another unique property of proline is that it is the only naturally occurring N-
substituted amino acid. Proteins that recognize the δ carbon on the substituted
amide nitrogen (Fig. 1A) within the context of the otherwise standard peptide
backbone can select precisely for proline at a given position without making
extended contacts with the rest of the side chain (Fig. 1C). Thus, sequence-
specific recognition can be achieved without requiring a particularly high-affinity
interaction.
Interactions that are specific and low-affinity can be quite useful in intracellular
signaling environments where rapidly reversible interactions may be required.
Among proline-binding domains, this phenomenon has been best characterized
for SH3 domains, in which required prolines can be replaced without a significant
loss in binding affinity by a number of nonnatural N-substituted amino acids that
do not resemble proline.
4. Lecture SS 20005
Cell Simulations 14
Polyproline type II (PPII) helices
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
Proline also stands out from other natural amino acids in its ability to exist stably
as a cis isomer about the peptide bond. In an unfolded chain, proline residues
adopt the cis conformation with a probability of ~20% as compared to negligible
amounts for the other amino acids. Moreover, the kinetic barrier for cis-trans
isomerization is higher for proline than for the other amino acids and is even the
rate-limiting step in the folding of certain proteins.
In principle, recognition of cis proline moieties could be a useful way of achieving
regulation, potentially even with some degree of kinetic control.
However, none of the major proline recognition modules discussed here are
known to exploit recognition of cis isomers. Still, the intriguing possibility remains
that cis-trans isomerization could provide a mechanism to modulate such
recognition events.
4. Lecture SS 20005
Cell Simulations 15
Properties of proline
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
Thus, many chemical properties of proline distinguish it from the other 19
naturally occurring amino acids, and proline recognition domains exploit several
of these properties.
If a recognition event involves a property of proline that is sufficiently distinct
among the natural set of 20 amino acids, the interaction does not have to be of
particularly high affinity to be selective.
The benefits of weak, but specific, interactions in intracellular signaling pathways
may help explain the abundance of proline-based recognition motifs.
4. Lecture SS 20005
Cell Simulations 16
Functional roles of SH3 domains
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
(A) Assembly role of SH3 domains. Growth factor stimulation leads to the activation of receptor tyrosine kinases and to the phosphorylation of the receptor tail, of related adaptor proteins (not shown), or of both. The resultant phosphotyrosines form docking sites for the adaptor protein Grb2 (through its SH2 domain). The Grb2 SH3 domains bind proline-rich motifs in SOS, the guanine nucleotide exchange factor for Ras, recruiting SOS to the membrane and colocalizing it with Ras. The resultant stimulation of Ras activates a MAPK cascade, leading to cell growth and differentiation.
(B) Regulatory role of SH3 domains. Intramolecular interactions of the SH2 and SH3 domains of Src kinases hold their kinase domains in an inactive conformation. These autoinhibitory interactions can be disrupted by external SH2 and SH3 ligands, yielding spatial and temporal control of kinase activation.
4. Lecture SS 20005
Cell Simulations 17
Functional roles of SH3 domains
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
Structure and binding mechanism of SH3 domains. The structure of the Sem5 SH3 domain in complex with a proline-rich ligand is shown. A cartoon of the proline-binding surface of these domains docked with a ligand, showing the general mechanism of recognition, is shown below. The core recognition surface has two xP binding grooves formed by aromatic amino acids, shown in yellow, and the adjacent, less conserved specificity pockets are designated in green.The PDB accession code for this structure is 1SEM.
4. Lecture SS 20005
Cell Simulations 18
Structure and binding mechanism of WW domains
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
The structure of the dystrophin WW domain in complex with a proline-rich ligand is shown. A cartoon of the proline-binding surface of these domains docked with a ligand, showing the general mechanism of recognition, is shown on the right. The core recognition surface has one xP binding groove formed by aromatic amino acids (yellow) and adjacent, less conserved specificity pockets (green). The PDB accession code for this structure is 1EG4
4. Lecture SS 20005
Cell Simulations 19
Structure and binding mechanism of EVH1 domains
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
A representative structure of the Mena EVH1 domain in complex with a peptide ligand is shown. Below is a schematic of the recognition mechanism showing the apex of the PPII helix fitting into an aromatic-rich wedge at the binding surface. Although a conserved set of aromatic residues (yellow) also contacts the PPII ligand, the manner in which the PPII helix docks against the domain surface differs from that observed in most other proline-binding domains discussed here. The PDB accession code for this structure is 1EVH.
4. Lecture SS 20005
Cell Simulations 20
Structure and binding mechanism of a GYF domain
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
The structure of the CD2BP2 GYF domain in complex with a proline-rich ligand is shown. A cartoon of the proline-binding surface of these domains docked with a ligand is shown below. The core recognition surface has one xP binding groove formed by aromatic amino acids (yellow) and adjacent, less conserved specificity pockets (green). The PDB accession code for this structure is 1L2Z
4. Lecture SS 20005
Cell Simulations 21
Mechanisms for enhancing the specificity
Zarrinpar, A., Bhattacharyya, R. P., and Lim, W. A. (2003) The structure and function of proline recognition domains, Sci. STKE. RE8.
Potential mechanisms for enhancing the specificity of proline-binding domains. One means of increasing specificity in proline-mediated interactions is by extending the interaction surface with the peptide to include residues beyond the proline-rich core. Another mechanism is to include a nearby sequence on the ligand that interacts with another binding module in the same complex as the proline recognition module. A third mechanism adds a separate recognition surface onto the proline recognition domain that recognizes a distinct peptide.
4. Lecture SS 20005
Cell Simulations 22
Identification of a novel “register-shifted” binding mode
Gu et al. Biochemistry, in press (2005)
NMR structure of GYF domain with wild-
type peptide. The GYF domain is
represented by its molecular surface; the
peptide atoms are drawn as sticks.
Residues forming the binding pocket are
coloured in dark grey and labelled by
their one-letter codes and sequence
numbers.
4. Lecture SS 20005
Cell Simulations 23
What is the conformation of the unbound peptide
Gu et al. Biochemistry, in press (2005)
4. Lecture SS 20005
Cell Simulations 24
Study conformation of unbound peptide
Gu et al. Biochemistry, in press (2005)
Evolution of the backbone dihedral
angles (black: Phi angles; red: Psi
angles) during the MD simulation of the
wild-type peptide (a) and the mutant
peptide (b). Ideal values of the dihedral
angles are shown in solid lines (blue: Phi
angles; green: Psi angles).
4. Lecture SS 20005
Cell Simulations 25
Unbound peptides have PPII helical conformation
Gu et al. Biochemistry, in press (2005)
Superposition of the representative
conformations of simulations of
unbound peptides (from left to right:
WT, WTE, G8W and H9M) onto the
bound peptide in the NMR structure.
Representative conformations are
colored in black while the bound
peptide in the NMR structure is shown
in grey.
4. Lecture SS 20005
Cell Simulations 26
Which residues are crucial for binding?
Gu et al. Biochemistry, in press (2005)
Substitution analysis of the SHRPPPPGHR
peptide binding to the GYF domain. All single
substitution analogues of the peptide were
synthesized on a cellulose membrane. The single
letter code above each column marks the amino
acid that replaces the corresponding wild-type
residue, while the row defines the position of the
substitution within the peptide.
Spots in the most left column (WT) have identical
sequences and represent the wild type peptide.
The membrane was incubated with a GST-GYF
construct of CD2BP2. Bound protein was
detected with an anti-GST primary antibody and a
horse-radish peroxidase coupled secondary
antibody. The relative spot intensities correlate
qualitatively with the binding affinities
WT A C D E F G H I K L M N P R S T V W YQ
H
R
S
P
R
H
P
P
P
W
V
WT A C D E F G H I K L M N P R S T V W YQ
H
R
S
P
R
H
P
P
P
G
V
Conclusion:Two central prolines are criticaland the following glycine.But can this glycine be mutated to Trp?
4. Lecture SS 20005
Cell Simulations 27
Binding analysis of Trp-peptide mutant
Gu et al. Biochemistry, in press (2005)
Binding analysis of the CD2BP2-GYF domain to
the peptide SHRPPPPWHRV in comparison to
the wild-type peptide SHRPPPPGHRV by NMR.
(a) The sum of the weighted geometrical
differences of the chemical shifts (Geometric sum
of chemical shift changes) for assigned peaks,
which could be identified at all applied peptide
concentrations is plotted against the
concentration of the peptide. (b) Mapping of the
binding site of SHRPPPPGHRV and
SHRPPPPWHRV peptides onto the CD2BP2-
GYF domain. Overlay of HSQC spectra of GYF
domain alone (green) and GYF-domain in the
presence of a 10-fold excess of the wild-type
peptide SHRPPPPGHRV (blue) or the mutant
peptide SHRPPPPWHRV (red), respectively.
4. Lecture SS 20005
Cell Simulations 28
MD simulation of GYF:domain complexes
Gu et al. Biochemistry, in press (2005)
Comparison of the binding interfaces
of the GYF domain (NMR and
simulation) for the wild-type complex
(above) and of the H9M mutant
(below). The GYF domain is
represented by its molecular surface
and coloured by position (from orange
to deep blue: completely buried to
completely exposed); the peptide
atoms are drawn as sticks and
coloured according to their
appearance in sequence.
4. Lecture SS 20005
Cell Simulations 29
Trp-peptide mutant shows “register shift”
Gu et al. Biochemistry, in press (2005)
(a) Superposition of the two binding modes found in the simulation
of the G8W mutant complex (starting from the docking results).
The two conformations of the peptide are drawn as sticks (blue:
mode 1, red: mode 2, pink: Pro6 and Pro7 in mode 1, yellow: Pro6
and Pro7 in mode 2). (b) Binding mode of the G8R mutant
complex (representative conformation of the simulation). The
peptide atoms are represented by sticks and coloured according
to their sequence number. In (a) and (b), the GYF domain is
represented by its molecular surface and coloured by position
(from orange to deep blue: completely buried to completely
exposed) and Pro6 and Pro7 are labelled by their one-letter codes
and sequence numbers. Mode 2 is labelled as “(alt)”. (c)
Superposition of the representative conformations of the five
simulations of wild type GYF complex starting from the alternative
binding mode. Pro6 and Pro7 are represented by sticks and are
labelled by their one-letter codes and sequence numbers. Pro6 is
coloured in light grey and Pro7 is coloured in dark grey. (d) The
translation and rotation motions of the peptide between the two
binding modes (blue: mode 1, red: mode 2, pink: Pro4 to Pro7 in
mode 1, yellow: Pro4 to Pro7 in mode 2). For Pro4 to Pro7 a
rotation is the principle component of motion, while for other
residues in the peptide a translation is the principle component of
motion.
4. Lecture SS 20005
Cell Simulations 30
register-shift hypothesis supported by experiments
Gu et al. Biochemistry, in press (2005)
Substitution analysis of the SHRPPPPWHR
peptide binding to the GYF domain. All single
substitution analogues of the peptide were
synthesized on a cellulose membrane. The single
letter code above each column marks the amino
acid that replaces the corresponding wild-type
residue, while the row defines the position of the
substitution within the peptide. Spots in the most
left column (WT) have identical sequences and
represent the wild type peptide. The membrane
was incubated with a GST-GYF construct of
CD2BP2. Bound protein was detected with an
anti-GST primary antibody and a horse-radish
peroxidase coupled secondary antibody. The
relative spot intensities correlate qualitatively with
the binding affinities
WT A C D E F G H I K L M N P R S T V W YQ
H
R
S
P
R
H
P
P
P
W
V
WT A C D E F G H I K L M N P R S T V W YQ
H
R
S
P
R
H
P
P
P
G
V
Figure 7
WT A C D E F G H I K L M N Q R S T WVP
HR
S
PRH
PP
PW
V
Y
CD2BP2-GYF tested with G8W mutant
4. Lecture SS 20005
Cell Simulations 31
Summary
- Complexes of adaptor domains with proline rich sequences form an important
cellular network
- Specificity of interactions vs. Multiplicity of interactions.
- Interactions can be influenced by proline conformation (cis/trans)
- Binding modes may not correspond to simple rigid body docking
(see register shift)
- Next 2 lectures of this module:
set up and analyze interaction network with Cytoscape