Signalling in multicellular organisms is mediated by complex networks that integrate extracellular and intracellular signals to generate appropriate responses regulating cell proliferation, differentiation and survival. Downstream of many cytokine and growth hormone receptors, receptor-associated JAKs (Janus kinases) activate transcription factors of the STAT (signal transducer and activator of transcription) protein family and thereby mediate signal transduction from the plasma membrane to the nucleus. The JAK/STAT pathway has been shown to be constitutively activated in a wide array of human malignancies. To elucidate mechanisms contributing to tumour formation and identify system properties of the JAK/STAT signalling pathway, a systems biology approach can be employed. So far the majority of studies available have focused on down-regulation of the signalling pathway based on simulations. However, a data-based model of the core module of the JAK2/STAT5 signalling pathway showed that rapid nucleocytoplasmic cycling of STAT5 is an essential pathway property. In the future, combining assays for quantitative analysis at different levels will be important to gain deeper insight into molecular mechanisms regulating intracellular communication mediated by such complex dynamic systems as signalling pathways and their targets.
In multicellular organisms reliable communication between individual cells is crucial to maintain the function of the organism as a whole. Receptors on the surface of cells sense extracellular signals and relay them to the inside of the cell. There they are amplified, integrated with other signals generated by the extracellular and intracellular state of the cell and the organism, and finally transduced to the nucleus where they regulate gene expression.
Dysregulation of any of the signal transduction pathways leads to malfunctions in cellular communication and thereby causes severe diseases such as autoimmunity, diabetes, heart disease and cancer. Therefore these pathways have been of high interest to the medical and biological research community and many of the components involved have been identified. However, multicausal diseases such as cancer are highly complex and their origin and regulation cannot be fully understood by intuition. In these cases, a systems biology approach becomes extremely valuable in the study of signal transduction. Mathematical models derived and adjusted according to knowledge based on quantitative experimental data measured under standardized conditions- this will in the following be referred to as a data-based approach- can help to identify otherwise hidden features of the pathways and thus predict the consequences of perturbations.
One important and well-studied example for a signal transduction pathway activated through many cytokine and growth factor receptors is the signalling cascade mediated by the tyrosine kinase JAK (Janus kinase) and the latent transcription factor STAT (signal transducer and activator of transcription) (Figure 1). The JAK/STAT pathway is conserved in all vertebrates and can also be found in some metazoans. It plays a central role in the regulation of cell proliferation, differentiation, cell migration and apoptosis (reviewed in ). In mammals, four JAK kinases (JAK1, JAK2, JAK3 and Tyk2) and seven STATs (STAT1, STAT2, STAT3, STAT4, STAT5a, STAT5b and STAT6) have been identified. The canonical JAK/STAT pathway is activated by cell-surface receptors that lack cytoplasmic kinase activity. Ligand binding to the extracellular domain of the receptor leads to conformational changes of the intracellular domain, bringing receptor-associated JAK proteins into close proximity and thereby enabling transphosphorylation of the JAKs. The activated JAKs then phosphorylate the receptor, generating binding sites for cytoplasmic proteins with SH2 (Src homology 2) domains such as the STAT proteins. Monomeric STATs bind to specific phosphotyrosine residues of the activated receptor and are subsequently phosphorylated by JAK. Activated STATs dissociate from the receptor, dimerize rapidly and translocate to the nucleus, usually regulated by an importin-dependent mechanism. In the nucleus, STAT dimers, together with cofactors, bind to the DNA and activate transcription of specific target genes.
Three major mechanisms have been identified for the attenuation of the JAK/STAT pathway (reviewed in ). First, several cytoplasmic and nuclear tyrosine phosphatases have been identified to dephosphorylate activated cytokine receptors, JAKs and STATs. Secondly, the expression of SOCS (suppressors of cytokine signalling) are induced by cytokine signalling and therefore act on a slower time scale. They interfere with signalling by binding phosphorylated receptors and JAKs via their SH2 domain. Additionally, they target receptor complexes for ubiquitin-mediated degradation. The third group of negative regulators are the non-inducible PIAS (protein inhibitors of activated STAT).
There have been several studies analysing JAK/STAT signalling by systems biology approaches. The system studied most extensively is the IFNγ (interferon-γ)-activated JAK1/JAK2/STAT1 pathway. Based on computer simulations, nuclear phosphatases were identified as the most important negative regulators in the JAK1/JAK2/STAT1 system . Additionally, multi-parametric sensitivity analysis based on simultaneous variation of several parameter values determined the concentrations of SOCS1, nuclear phosphatase and cytoplasmic STAT1 as well as the corresponding reaction steps influencing those concentrations as most sensitive to perturbation . By employing a multilevel hierarchical paradigm, negative regulation of JAK/STAT1 by SOCS1 was also identified as the co-ordinator of pathway behaviour . Furthermore, the system described by the model in  was shown to be highly robust due to the necessity of STAT1 dimerization for nuclear translocation and activation of gene expression . Mathematical modelling of simultaneous signalling through the JAK/STAT and MAPK (mitogen-activated protein kinase) pathways could show that SHP2 (SH2-domain-containing tyrosine phosphatase) and SOCS3 affect both JAK/STAT and MAPK signalling . All of these studies rely on previously published data and solely make suggestions for experiments to test predictions of the mathematical models. An exception is the study by Swameye et al.  which combines generation of quantitative data with dynamic pathway modelling to study JAK2/STAT5 signalling. The study shows that nucleocytoplasmic shuttling of STAT5 is the most sensitive step of the core module of JAK2/STAT5 signalling. Furthermore, model predictions generated by in silico simulations were verified experimentally.
Alterations of JAK/STAT signalling in cancer
Constitutive activation of STAT3, STAT5 and STAT1 has been detected in a large number of human solid tumours and haematological malignancies (reviewed in ). Although JAK/STAT signalling is tightly regulated in normal cells, these control mechanisms fail in malignant cells, either by overexpression or mutations of one of the regulatory proteins involved. Gain-of-function mutations leading to constitutive activity of tyrosine kinases and therefore causing permanent STAT activation have been found in many malignancies. For example, Bcr-Abl, a constitutively active variant of the tyrosine kinase Abl, was found in CML (chronic myelogenous leukaemia) and the JAK2(V617F) mutation was identified in a large number of Bcr-Abl-negative myeloproliferative diseases such as polycythaemia vera (for a review of oncogenes in myeloproliferative diseases see ).
In the last few years, treatments have focused on blocking the activation of transcription factors through intervention at the level of the respective kinase. For example, the drug imatinib (Glivec) inhibits Abl kinase by binding to the active site of Abl and thereby blocking phosphorylation of Abl substrates, such as JAK2 and STAT5. Imatinib is highly lethal to CML cells, but hardly affects healthy cells. However, the transcription factors themselves have recently been recognized as efficient targets for anti-cancer drugs . Their action in activating certain gene expression patterns is frequently the last step in the development of malignancies since they integrate signals from different pathways. Furthermore, as described above, the most sensitive steps identified in the JAK/STAT pathway lie downstream of the kinase activity and either involve nucleocytoplasmic shuttling of the latent transcription factor STAT or negative feedback regulation of the pathway. Therefore targeting receptor or kinase activity might not be the most efficient approach for intervention. In this case, a systems biology approach can contribute to identifying alternative strategies.
Erythropoietin-induced STAT signalling
In order to identify pathway alterations that promote the onset of diseases it is important to examine dynamic pathway properties under unperturbed conditions. As an example of a data-based systems biological description of JAK/STAT signalling, the investigation of the core module of the JAK2/STAT5 pathway downstream of the EpoR (erythropoietin receptor) is detailed below.
Essentially, the EpoR pathway follows the generic steps for JAK/STAT signalling delineated in the Introduction. In the past it has been assumed that, once a STAT protein has initiated target gene expression in the nucleus, its active role in the pathway has ended. However, Swameye et al.  could show using a data-based mathematical modelling approach that, instead of such a linear cascade, the signalling pathway employs rapid nucleocytoplasmic cycling of STAT5 to constantly monitor the state of receptor activation and quickly relay this information to the nucleus where the output of the system, namely the activation of target genes, is regulated. Furthermore, it was concluded that the cell operates more economically by rapidly cycling STAT5 instead of producing large quantities of STAT5 that then accumulate in large numbers in the nucleus.
Rapid nucleocytoplasmic shuttling of STAT5 as a remote sensor
First, the qualitative biological knowledge of the core module of JAK2/STAT5 signalling (Figure 2) is translated into ordinary differential equations following the concepts of mathematical biology . The model consists of four coupled differential equations (eqns 1–4) describing the ligand-induced activation of the EpoR and subsequent activation of STAT5, assuming a linear reaction cascade based on mass-action kinetics. Although mass-action laws hold true for the dimerization of STAT5, it is only an approximation for STAT5 activation assuming that the kinase mediating STAT5 phosphorylation and the machinery involved in nuclear import are not rate limiting. The activation of JAK2 has been omitted to simplify the model and keep the number of parameters small. The changes in concentration of monomeric unphosphorylated STAT5 over time are described by ẋ1 · x1 decreases depending on the rate constant k1 and the concentration of activated EpoR, EpoRA. Eqn (2) states that monomeric phosphorylated STAT5, x2, is generated with the same rate k1 at which x1 decreases, but also decreases dependent on the square of its own concentration and the rate k2 through a dimerization step. From this follows the first term on the right-hand side of eqn (3), which describes the dynamics of dimerized activated STAT5, x3. The second term in eqn (3) represents the nuclear translocation of dimeric, activated STAT5. Since two molecules of phosphorylated STAT5 yield only one STAT5 dimer, the term for phosphorylated STAT5 has to be squared in eqns (2) and (3). Eqn (4) corresponds to the temporal changes of dimerized STAT5 in the nucleus, x4. In this very basic linear model, STAT5 cannot enter further rounds of activation but remains in the nucleus until the pathway is down-regulated. Based on the biological knowledge available at the time it was assumed that nuclear STAT5 is either degraded  or slowly deactivated and transported out of the nucleus. However, the time scale of these deactivation steps was thought to be much slower than that of the data presented by Swameye et al. . As it is difficult to distinguish the different STAT5 protein states, e.g. monomeric phosphorylated STAT5 and dimerized phosphorylated STAT5, experimentally, it was only possible to obtain quantitative immunoblotting data for the temporal changes of overall STAT5 phosphorylation y1(t), the total amount of STAT5 in the cytoplasm y2(t), as well as receptor phosphorylation y3(t). where k1−k7 represent scaling parameters which are necessary since only relative protein amounts but not absolute amounts can be measured by immunoblotting with sufficient accuracy.
The parameters k1−k7 were estimated based on the variables y1 and y2, whereas y3 serves as input for the inhomogeneous ordinary differential equation system, eqns (1)–(4). Parameter estimation was performed applying the maximum-likelihood approach . A user-friendly implementation for parameter estimation is given by the software PottersWheel (http://www.potterswheel.de; T. Maiwald and J. Timmer, unpublished work).
Applying this model to the quantitative time course data for EpoR and STAT5 activation (Figures 3A and 3B), parameter estimation results in the fits shown in Figure 3(C). The linear model does not describe the experimental data sufficiently well and had to be modified. Different alternative models were tested and it was concluded that the export of STAT5 is necessary to properly describe the data. The export was modelled by a delay term x3τ = x3(t − τ), combining the reactions of STAT5 in the nucleus with the nuclear export. Since no data on the different processing steps in the nucleus were experimentally available, a detailed model could not be established. The input–output relationship of a chain of successive reaction steps can be represented by a delay . In the new cycling model, eqn (8) replaces eqn (1) and eqn (9) replaces eqn (4). The results of a fit of this model to the experimental data are shown in Figure 3(D) and are in good agreement with the data. Surprisingly, the sojourn time ι is only approx. 6 min, describing very rapid cycling of STAT5 between the nucleus and the cytoplasm.
Furthermore, a sensitivity analysis was performed using the overall amount of cycling nuclear STAT5 as output which was calculated from the area under the curve of the time-dependent amounts of nuclear STAT5 determined by in silico simulations (results not shown). For the sensitivity analysis, the parameters of the model were gradually changed and the output of the system was simulated. The results of this in silico simulation are shown in Figure 4(A). The largest effects, revealed to first order by the largest absolute slope at a relative parameter variation of 1, are caused by changes of the parameters involved in nucleocytoplasmic cycling of STAT5, namely the import rate k3, the export rate k4 and the delay time ι. The simulations predict that setting the export rate to zero reduces target gene activation by a factor of two. These predictions could be verified experimentally by inhibiting nuclear export with the drug LMB (leptomycin B) and monitoring induction of CIS (cytokine-inducible SH2-domain-containing protein) protein synthesis as an endogenous target gene of STAT5, which is assumed to be proportional to activated STAT5 in the nucleus (Figure 4B).
The study by Swameye et al.  has shown that it is necessary to investigate more closely the transport processes involved in STAT5 translocation since nuclear import and export of STAT5 were identified as the steps most sensitive to perturbation.
Live-cell imaging of JAK/STAT signalling
Although it is possible to separate cytoplasmic and nucleoplasmic proteins biochemically, the investigation of intracellular spatial changes is still challenging. Furthermore, biochemical procedures require material from a large number of cells and thereby only yield averaged data from a population of cells, possibly losing information on the dynamics of the system. Thus measurements on the single cell level, for example by live-cell imaging, have become an invaluable tool in the study of protein dynamics . GFP (green fluorescent protein) and its variants have been employed to successfully label a wide variety of proteins  and numerous quantitative methods have been developed to study spatio–temporal features of signalling by fluorescence imaging such as FRAP (fluorescence recovery after photobleaching) to investigate subcellular trafficking , FRET (fluorescence resonance energy transfer) to observe protein interactions , as well as FCS (fluorescence correlation spectroscopy) to measure protein mobility and concentrations .
Several STAT proteins have been tagged with GFP and investigated by live-cell imaging in the past. Nuclear accumulation of STAT1–GFP was observed in C243 cells after IFNγ stimulation . For STAT3 it was shown that it constitutively shuttles between the cytoplasm and nucleus, independently of its tyrosine phosphorylation . Recently, constitutive nuclear import and export as well as nuclear accumulation and retention by DNA binding of STAT5a have been studied, by live-cell imaging . Besides nucleocytoplasmic shuttling, other features, such as mobility of STAT proteins in the nucleus and the dependence of STAT1 movement from the cytoplasm to the nucleus on the cytoskeleton, were investigated [25–28]. Even though the information gathered in these studies is valuable for a better understanding of JAK/STAT signalling the data have not been investigated on a systems level.
Although fluorescent proteins are an extremely useful tool, the expression of fluorescently tagged proteins in mammalian cells is prone to artefacts. The protein has to be expressed exogenously, possibly shifting the stoichiometry of the signalling molecules involved in the system under investigation, and the tag itself can alter the function of the protein of interest. For example, it has been shown that tagging STAT1 with GFP alters nucleocytoplasmic cycling of unphosphorylated STAT1 . Therefore fluorescent labelling of proteins has to be critically evaluated prior to use of live-cell imaging data to determine systems properties.
A future challenge in the study of the JAK/STAT pathway will be to combine quantitative data from live-cell imaging with data obtained by other methods to gain more information on the parameters describing nuclear import and export of STAT proteins.
JAK/STAT signalling is one of the central signal transduction pathways in mammalian cells that is frequently dysregulated in malignancies. Using data-based modelling, nucleocytoplasmic shuttling of STAT5 has been identified as a key feature of the pathway implicating it as an efficient target for therapeutic intervention. To further pinpoint potential drug targets within the JAK/STAT pathway it will be necessary to take advantage of novel technical developments of quantitative methods. Currently, many studies analysing signal transduction pathways by mathematical modelling rely on data published from various laboratories working with different model systems and in vitro conditions. This approach is prone to incorrect conclusions since often the data obtained from different sources are not comparable. Therefore it is essential to combine quantitative data generated under standardized conditions with mathematical modelling . In the future, it will become increasingly important to combine various approaches for quantitative data generation such as biochemical time course data, live-cell imaging and FACS analysis, as well as gene expression data, to gather a more complete and quantitative picture of the pathway under investigation. This would allow for a more efficient design of therapeutic components to combat diseases such as cancer.
• Signal transduction through the JAK/STAT pathway is dysregulated in many human cancers.
• Nucleocytoplasmic shuttling of STAT5 has been identified as an important pathway feature of JAK2/STAT5 signalling that is sensitive to perturbations and therefore a potential drug target.
• Studies of negative-feedback regulation of the JAK/STAT pathway might elucidate further targets for therapeutical intervention.
• Quantitative live-cell imaging will become an increasingly important tool for systems biology.
• In systems biology the combination of quantitative data from different biological assays performed under standardized conditions is essential to gain information about as many parameters as possible.
- © The Authors Journal compilation © 2008 Biochemical Society