Fig 1.
IGF1R phosphotyrosine contact map in HeLa S3 cells.
(A) Ligand-free and inactive IGF1R. Prior to crosslinking of dissimilar sites S1 and S2 of each IGF1R protomer in a dimer, the two kinase domains are not in contact and are thus unable to autophosphorylate receptor tyrosines. (B) IGF1R phosphotyrosine interactions in HeLa S3 cells. Each IGF1R in an IGF1R dimer contains six autophosphorylation tyrosine sites, labeled “Y” followed by the site’s position within the polypeptide chain (according to UniProt numbering). The kinase domains of each IGF1R subunit are labeled accordingly and encompass the region spanning positions 999 to 1274. Autophosphorylation and dephosphorylation of activated (IGF1-crosslinked) receptors in a dimer occur at the tyrosine sites indicated. Signaling proteins containing SH2 and/or PTB domains are recruited to their cognate tyrosine sites in a phosphorylated IGF1R dimer as indicated by the colored lines that begin at a tyrosine site and end at the protein that is recruited to that site.
Fig 2.
Model for IGF1 interaction with IGF1R.
The model considers IGF1R to be present in preformed dimers, where each dimer contains two ligand binding pockets. Within each pocket, there are two sites, which we refer to as sites S1 (light pink) and S2 (light blue). S1 and S2 are bound by IGF1 (black rectangle) as signified by direct contact. Adjacent S1 and S2 sites are crosslinked when they are bound by the same IGF1 molecule, and, when crosslinking occurs, the kinase domains of the two receptor subunits come into contact. The IGF1R dimer is thus activated and can catalyze autophosphorylation. (A) The cyclic reaction scheme of IGF1-IGF1R binding presented here is derived from the model of Kiselyov et. al. [44]. (B) IGF1-IGF1R complexes that arise from the reactions of the cyclic reaction scheme. (C) Phosphorylation of one of the six IGF1R tyrosine sites, Y980. (D) Dephosphorylation of the Y980 site of IGF1R. (E) Recruitment of a representative IGF1R binding partner, IRS1 (pink structure), to phosphorylated Y980 via the IRS1 PTB domain (cyan structure), and subsequent dissociation.
Fig 3.
Parameter values for IGF1-IGF1R binding were obtained by fitting to data from competition experiments and ligand dissociation experiments presented by Kiselyov et al. [44]. Experimental data are shown as points with error bars and the smooth curves were obtained via numerical simulation. For (A), we simulated to steady-state for a range of concentrations of unlabeled IGF1. For (B), we simulated a two-hour incubation with radiolabeled IGF1, then free ligand was washed away (i.e., the abundance of free IGF1 was set to zero), and we simulated the response to ligand addition for either 20 or 60 minutes, as indicated, for a range of concentrations of unlabeled IGF1.
Table 1.
Comparison of parameter values for IGF1-IGF1R binding.
We estimated rate constants (defined as indicated in Fig 2) by fitting to IGF1-IGF1R binding data [44] using the BioNetFit software package [61]. We calculated the 95% confidence interval for each parameter using a bootstrapping procedure (see Material and Methods). The parameter a'2 was taken to be given by Eq 1 and the values of the other rate constants.
Fig 4.
Predicted recruitment of signaling proteins to IGF1R in HeLa S3 cells.
The scale represents the normalized amount of each of the indicated proteins bound in simulations of either (A) time-dependent recruitment at 1 nM IGF1 stimulation, or (B) steady-state recruitment at varying IGF1 doses. (C) Comparison of rank ordering of IGF1R binding partners in HeLa S3 cells across five prediction methods. The top (first) ranked protein indicates the protein predicted to be bound to IGF1R with the highest abundance. Bar length indicates the amount of deviation from the ranking obtained by numerical simulations for estimates obtained by the analytical approach, copy number, by KD, or the ratio of copy number to KD. A negative deviation indicates that the method estimated a lower rank (less binding) than the prediction from numerical simulations; a positive deviation indicates a higher rank (more binding) than in numerical simulations. Calculation and evaluation of rank ordering are described in Materials and Methods.
Fig 5.
Summary of rank ordering of IGF1R binding partners across 45 cell line-specific models.
(A) The average rank across all models is shown for each binding partner. Error bars indicate the standard deviation in rank. The size of a data point indicates the number of cell lines in which the corresponding protein was expressed, as shown in the legend. (B) For the four binding proteins with the highest average rank, we show the percentage of cell lines in which the protein was ranked first, second, third, fourth, fifth, or lower than fifth.
Fig 6.
Hierarchical clustering of cell lines according to predicted signaling protein recruitment.
Heatmap colors indicate the ratio, derived from simulation results, of the number of bound molecules of a given protein (corresponding to protein name on the x-axis) divided by the total number of IGF1R molecules. As multiple copies of a protein can be bound to one IGF1R molecule, the ratio can be greater than one. Cell lines indicated (on the y-axis) include the tissue of origin or type of cancer followed by an underscore and the cell line name. OVAR = ovarian cancer, NSCLC = non-small cell lung cancer, MELAN = melanoma, CERV = cervical cancer, CNS = central nervous system cancer, LEUK = leukemia. For those cell lines evaluated in multiple studies (A549, HeLa S3, and MCF7), the proteomics data from Geiger et al. is indicated with “_G” following the cell line name, the data for the NCI-60 panel from Gholami et al. with the suffix “_N”, and the data from Kulak et al. with “_K”.
Fig 7.
Correlation matrix among the signaling proteins for IGF1R recruitment.
The color map shows the Pearson’s correlation coefficient among pairs of proteins studied in the population-level models for HeLa S3, U251, MCF7, and GAMG cell lines. Red indicates a negative correlation, blue a positive correlation, and white no correlation.