Fig 1.
Functional and signaling-competent EPOR in the NSCLC cell line H838 and CFU-E cells.
(A) Left panel: H838 and H838 cells stably overexpressing HA-tagged human EPOR (H838-HA-hEPOR) were either left untreated (-) or stimulated with 10 U/ml EPO beta (+). Cells were lysed after 10 min and hEPOR proteins were subjected to immunoprecipitation (IP, MAB 307, R&D) and phosphorylated (p) EPOR (4G10, Merck Millipore) and total EPOR (C-20, Santa Cruz) were detected by quantitative immunoblotting (IB). The complete immunoblot is shown in S1 Fig. Right panel: murine (m) CFU-E cells were either left untreated (-) or stimulated with 5 U/ml Epo alfa (+). Cells were lysed after 10 min and mEPOR proteins were subjected to IP (M-20, Santa Cruz). pEPOR (4G10, Merck Millipore) and total EPOR (M-20, Santa Cruz) were detected by IB with chemiluminescence by CCD camera. The experiment was performed in biological triplicates. (B) H838 and H838-HA-hEPOR cells were treated with the indicated doses of EPO beta. Cells were lysed after 10 min and JAK2 proteins were subjected to IP. pJAK2 and total JAK2 were detected by IB with chemiluminescence by CCD camera. (C) H838 and H838-HA-hEPOR cells were treated with the indicated doses of EPO beta. Cells were lysed after 20 min and STAT5 proteins were subjected to IP. The degree of phosphorylation of STAT5 was measured with mass spectrometry. For H838 cells, the average degree of STAT5 phosphorylation based on biological duplicates is shown. (D) H838 and H838-HA-hEPOR cells were treated for three days with 5 mg/l cisplatin or left untreated. Additionally, cells were treated with 10 U/ml EPO beta and the cell viability was measured with CellTiter-Blue assay. The error bars represent standard deviation of biological replicates (n ≥ 5). The experiment was performed on two independent days (second replicate in S4 Fig). (E) H838-HA-hEPOR cells expressing the Casper3-GR FRET-based sensor (H838-HA-hEPOR-Casper3-GR) were treated with 5 mg/l cisplatin, EPO beta, a combination of both or left untreated. Staurosporine (10 μM) was used as positive control for induction of apoptosis. Casper3-GR FRET signal was measured by life-cell imaging for 60 hours. Caspase-3 activity was determined based on the green-to-red ratio and normalized to the untreated control (n = 2, second replicate in S5 Fig).
Fig 2.
Example model with simulated data for two different cell types.
The process diagram of a two-step phosphorylation reaction (Protein→pProtein→ppProtein) is shown according to Systems Biology Graphical Notation. The ODE was numerically solved over time for two cell types that differ in the initial protein concentration ([Protein]t = 0) and in one kinetic rate (k3). The initial concentrations of the phosphorylated compounds were set to zero ([pProtein]t = 0 = [ppProtein]t = 0 = 0). The second phosphorylation step is reversible and the dephosphorylation rate (k3) was assumed to be cell type-specific. The parameters for the phosphorylation steps (k1, k2) are the same for both cell types. Simulated data points are depicted as black dots, and the grey shading indicates the standard deviation (σ = 0.1) of the simulated measurement errors. Model trajectories are displayed as black lines.
Fig 3.
L1 regularization recovers cell type-specific parameters.
(A) The dependency of the predicted number of cell type-specific parameters and the regularization weight λ is shown. (B) The likelihood ratio test was performed for each number of cell type-specific parameters. If the test statistics (blue), i.e. the mismatch between data and model, is larger than the statistical threshold (red dashed), the regularization weight λ was too large hence the model was rejected. The crossing of the blue and the dashed red line corresponds to the parsimonious model (dashed black line). (C) The regularization path of the four parameters fold-changes is shown. The regularization-dependent parameter differences are indicated with shades of red (higher in cell type 1) to blue (higher in cell type 2). The asterisks indicate the identified parameter differences. (D) Model trajectories and simulated data for three exemplary scenarios. The dashed green line shows the dynamics for only cell type-specific parameters. In contrast, the blue line displays the model trajectories for no cell type-specific parameters, which is unable to describe the simulated data. Finally, the parsimonious model (solid red) is able to describe the simulated data with only two relevant cell type-specific parameters recovering the simulated parameter differences.
Fig 4.
Generalized mathematical model structure of the EPO-induced JAK2/STAT5 signaling pathway.
The process diagram of the EPO-induced JAK2/STAT5 signaling pathway model is shown according to Systems Biology Graphical Notation. The binding of the ligand EPO to its cognate receptor results in the phosphorylation of first JAK2 and then of the EPOR (pEPORpJAK2). STAT5 is recruited by pEPOR and phosphorylated by pJAK2 and translocates to the nucleus where it induces the transcription of the negative feedback regulators CISH mRNA and SOCS3 mRNA. Protein tyrosine phosphatase (PTP) regulates the dephosphorylation of the EPOR-JAK2 complex.
Fig 5.
The model trajectories for a selection of key pathway components are shown for CFU-E, H838 and H838-HA-hEPOR cells. This includes expression of the EPOR targets CISH mRNA and SOCS3 mRNA measured by qRT-PCR as well as pEPOR, pJAK2 and cytoplasmic STAT5 data measured by quantitative immunoblotting. The amount of pSTAT5 was determined by either mass spectrometry or quantitative immunoblotting. The closed circles represent experimentally measured data in H838 and H838-HA-hEPOR cells. CFU-E data previously published [19] are shown as circles. The lines depict the three applied model strategies: dashed green (only cell type-specific parameters), dashed blue (no cell type-specific parameters) and solid red (parsimonious model, only relevant cell type-specific parameters). The parsimonious model describes the data similarly to the model with only cell type-specific parameters, whereas the trajectories of the model without cell type-specific parameters are not in line with the experimental data, e.g. for SOCS3 mRNA in CFU-E and for pSTAT5 in H838. All data sets, replicates and trajectories of the parsimonious model are shown in S8 and S9 Figs.
Fig 6.
Identification of cell type-specific differences.
(A) The number of cell type-specific parameters in dependency of the regularization weight λ is shown. (B) The likelihood ratio test statistics was calculated for each regularization weight λ, resulting in a nested sub-model with a number of cell type-specific parameters that is smaller compared to the full model. If the test statistics (blue) is larger than the statistical threshold (red dashed), the model reduction step was rejected. The crossing of the blue and the dashed red line corresponds to the parsimonious model (dashed black line). (C) The regularization path of the 26 parameters is shown. The regularization-dependent parameter differences are indicated with shades of red (higher in CFU-E) to blue (higher in H838 & H838-HA-hEPOR). The asterisks depict the identified parameter differences.
Table 1.
Parameters and identified cell type-specific differences.
Fig 7.
Experimental validation of the cell type-specific CISH and SOCS3 mRNA parameters.
The parsimonious model was employed to predict the dynamics of CISH mRNA and SOCS3 mRNA upon treatment with a transcriptional inhibitor. The mRNA dynamics in CFU-E stimulated with 5 U/ml EPO alfa alone (black) or with transcriptional inhibition after 60 min (blue) was predicted. Additionally, the mRNA dynamics in H838-HA-hEPOR stimulated with 10 U/ml EPO beta alone (black) or with transcriptional inhibition after 30 min (blue) was predicted. Shadings surrounded by dotted lines depict uncertainty of the prediction. CFU-E cells were stimulated with 5 U/ml EPO alfa and either additionally treated with 1 μg/ml actinomycin D, to inhibit transcription, at 60 min (blue arrows) or left untreated. The H838-HA-hEPOR cells were either stimulated with 10 U/ml EPO beta alone (black) or additionally with 1 μg/ml actinomycin D at 30 min (blue arrows). The mRNA was extracted at the indicated time and the SOCS3 and CISH mRNA levels were measured with qRT-PCR. Experimental data are depicted as closed circles. The experiment was performed in triplicates and one representative example is shown.
Fig 8.
The process diagram of the EPO-induced JAK2/STAT5 signaling pathway model is shown according to Systems Biology Graphical Notation. Identified parameter fold-changes between CFU-E and H838 cells are shown in red (higher in CFU-E: JAK2actEPO, EPORactJAK2, CISHRNAturn) or purple (higher in H838: STAT5imp, nSTAT5deact, SOCS3prom, SOCS3RNAdelay). Parameters with a more effective inhibition in H838 cells are shown in light blue (JAK2actEPO, EPORactJAK2, STAT5actJAK2, STAT5actEPOR, STAT5imp, STAT5exp, SOCS3RNAturn). The area-under-curve of npSTAT5 at 60 min after stimulation was used as read-out to calculate the sensitivities.