METHODS AND SYSTEMS FOR PREDICTING RESPONSE OF CELLS TO A THERAPEUTIC AGENT (2024)

Considerable advances have been made in the development of targeted therapies for the treatment of cancer and other diseases. Such targeted therapies include monoclonal antibodies that bind to antigens that are specifically or preferentially expressed on tumor cells and small molecule drugs that specifically interfere with discrete components of signaling pathways active in tumor cells. For example, cetuximab (Erbitux®) is a monoclonal antibody that targets the epidermal growth factor receptor (EGFR, also known as ErbB1 or HER1) that is expressed in at least certain colon cancers and head and neck cancers. Also for example, imatinib (Gleevec®) is a small molecule that targets the BCR-Abl tyrosine kinase, which is expressed, and acts as an oncogenic factor, in certain chronic myeloid leukemias and is an abnormal variant of a benign cellular protein. While such targeted therapies have been shown to be effective in some patients, the response rate is never 100%. For example, the average response rate for cetuximab monotherapy is only around 15-20% of patients, even when tumors are known to express ErbB1 (EGFR). Thus, mere expression of ErbB1 (the antigen targeted by the cetuximab antibody) in a tumor does not guarantee responsiveness to cetuximab.

Thus, while targeted therapies are very promising, the variable response rate of patients to such therapies, combined with the side effects associated with such therapies and the typical high cost of such therapies, indicates that methods for treating patients which involve predicting which patients are likely to respond to therapeutic treatment and only administering the treatment to patients who are predicted to respond are highly desirable. One approach that has been taken has been to try to identify genetic markers (e.g., mutations or alleles) that correlate with responsiveness to therapy. In this approach, a sample from the patient is genotyped prior to treatment to determine whether the patient carries a genetic marker(s) that is indicative of responsiveness to therapy. Another approach that has been taken is to try to identify protein biomarkers that correlate with responsiveness to therapy. In this approach, protein expression is determined in a sample from the patient prior to treatment to determine whether the patient expresses one or more protein biomarkers that are indicative of responsiveness to therapy.

Both of the aforementioned approaches can be considered to be “direct” marker approaches, wherein the presence (or absence, or level of expression) of the marker(s) (e.g., BCR-Abl or ErbB1) directly being measured has been demonstrated to correlate with responsiveness or non-responsiveness to therapy. Furthermore, both of these approaches rely on the use of markers that are sufficiently stable in cells such that they can be reliably measured or quantitated in a sample that has been isolated from the patient. Given that there may be a considerable time lag between when a sample is isolated from a patient and when the marker(s) is measured in the sample, such “direct” marker approaches described above typically require the use of genetic or protein markers that are not subject to degradation or alteration over time when samples are subjected to conventional processing and handling. While such stable, “direct” markers that are predictive of responsiveness to certain therapeutic agents have been identified, it is unclear whether such markers can be identified for all therapeutic agents.

It is thought that tumors are driven to grow by a set of ligand activated signaling pathways, which are usually activated by ligands binding to their cognate receptors, inducing the phosphorylation of the receptor itself as well as of downstream kinases, leading to further phosphorylation of downstream components of the pathway. These kinases trigger cell survival and proliferation. Accordingly, activation of the signaling pathway leads to alteration of intracellular components, in particular protein phosphorylation. The phosphorylation signature of the receptors expressed on tumor tissue can help to identify the main pathways that drive a particular tumor's progression. However, phosphoproteins can be very labile and the phosphorylation can dissipate quickly after surgery if the tissue sample is not immediately and rapidly frozen (or, in some cases, formalin fixed). Moreover, even where it is possible to reliably measure levels of one or more phosphoproteins in a sample of a particular tumor, the predictive value of the presence or absence of any particular phosphoprotein regarding efficacy of treatment of such a tumor with any particular therapeutic agent is generally unknown. Therefore, while phosphoprotein profiles contain important information about the pathways driving tumor progression, such phosphoprotein profiles currently are not widely used as biomarkers for predicting responsiveness to therapeutic treatment.

Accordingly, new methods for determining levels of various phosphoproteins and of using such levels and other tumor cell characteristics for predicting the responsiveness of individual tumors to particular therapeutic agents are needed to improve the therapeutic and cost effectiveness of cancer therapies.

Herein provided are methods for predicting responsiveness of cells, in particular neoplastic cells such as tumor cells (e.g., benign tumor cells or malignant tumor cells) and malignant cells that are not tumor cells, to therapeutic agents and methods for treating patients having such tumors with therapeutic agents.

In one aspect, methods are provided for treating a patient for a malignancy with an anti-neoplastic therapeutic agent by obtaining a sample of malignant cells from the patient, determining certain biochemical characteristics of the cells in the sample, and subsequently administering at least one anti-neoplastic therapeutic agent to the patient. In certain embodiments, the biochemical characteristics are the level of at least one biomarker; in further embodiments, the level(s) of the biomarker(s) are determined by measuring levels of other, more stable, biochemical compounds and then using a computer modeling paradigm to determine the levels of the biomarker(s) of interest. The therapeutic agent is selected on the basis of the level(s) of the biomarker(s); certain agents are administered only when specific level(s) of biomarker(s) are exceeded.

In certain embodiments, methods are provided for a treating a patient having a neoplastic tumor with an anti-neoplastic therapeutic agent, comprising: obtaining a sample of the tumor (e.g., a biopsy sample or a resected sample) comprising tumor cells, determining a level of phosphorylated ErbB3 (phospho-ErbB3, pErbB3) in the sample, and subsequently administering at least one anti-neoplastic therapeutic agent to the patient. An anti-ErbB3 therapeutic agent is administered if the sample cells are found to contain at least a minimum level of pErbB3; an anti-neoplastic therapeutic agent that is not an anti-ErbB3 therapeutic agent is administered, and an anti-ErbB3 therapeutic agent is not administered to the patient if the sample cells are found not to contain at least the minimum level of pErbB3. Preferred anti-ErbB3 pharmaceutical agents are anti-ErbB3 antibodies. In certain embodiments of this method, the level of ErbB3 in the sample cells is determined inferentially by measuring levels of other, more stable, biomarkers and using a computerized method using a computing system to generate a computer model to compute (based on actual empirically measured levels of other biomarkers in the sample cells) a Network Activation State that determines, by simulation, the levels of pErbB3 in the sample cells.

Within one such embodiment, methods are provided for a treating a patient having a malignant tumor, comprising: obtaining a sample of the tumor, determining a level of pErbB3 in the sample, and subsequently administering at least one anti-neoplastic therapeutic agent to the patient, wherein, if the level of pErbB3 determined in the sample is no lower than 50% of a level of pErbB3 measured in a culture of ACHN renal cancer cells (ATCC No. CRL-1611) following culture for 20-24 hours in serum-free medium (e.g., RPMI) then the at least one anti-neoplastic therapeutic agent subsequently administered to the patient comprises an anti-ErbB3 antibody, and if the level of pErbB3 determined in the sample is lower than 50% of the level of pErbB3 measured in the culture of ACHN renal cancer cells then the at least one anti-neoplastic therapeutic agent subsequently administered to the patient does not comprise an anti-ErbB3 antibody.

In certain aspects, the present invention provides computerized methods using a computing system that comprises at least one input device configured for receiving input and at least one output device configured for rendering output, said methods being for predicting the response of cells (e.g., tumor cells) comprising a cellular network (e.g., an ErbB signaling pathway) to treatment with a therapeutic agent that targets a component within the cellular network, said methods comprising: (a) receiving, through said computing system input device, input that identifies levels of one or more components in the cellular network measured in a sample of the cells; (b) computing from the input, with the computing system, a Network Activation State (NAS) or a Network Inhibition State (NIS) for the cells using a computational model of the cellular network; and (c) generating with the computing system, and thereafter rendering at said output device, a predicted response of the cells to treatment with the therapeutic agent based at least in part on the NAS or the NIS computed in (b).

In further aspects, methods are provided for predicting the response of cells to treatment with a therapeutic agent that targets a component within a cellular network (e.g., an ErbB signaling pathway) comprised by the cells, the methods comprising: (a) measuring the level in a sample of the cells of one or more components of the cellular network; and (b) applying a computer-implemented method comprising: (i) computing a Network Activation State (NAS) or a Network Inhibition State (NIS) for the cells using a computational model of the cellular network input with the one or more measured levels; and (ii) computing and outputting a predicted response of the cells to treatment with the therapeutic agent based at least in part on the NAS or the NIS computed in (i). In certain embodiments, such methods can further comprise treating cells, or a patient from whom the cells are obtained, with a therapeutic agent, based on the predicted responsiveness of the cells to the therapeutic agent.

Also provided herein are methods for predicting the response of cells to treatment with a therapeutic agent that targets a component within a cellular network (e.g., an ErbB signaling pathway), the methods comprising: (a) measuring, in a sample of the cells, levels of one or more components of the cellular network; and (b) applying a computer-implemented method comprising: (i) computing a computational model of the cellular network by applying a statistical classification algorithm to input measured levels and computing a NAS or NIS for the cells therefrom; and (ii) predicting the response of the cells to treatment with the therapeutic agent based at least in part on the computed NAS or NIS. In certain embodiments, such methods can further comprise treating cells, or a subject from whom the cells are obtained, with a therapeutic agent, based on the predicted responsiveness of the cells to the therapeutic agent.

The present invention further provides computerized methods using a computing system that comprises at least one input device configured for receiving input and at least one output device configured for rendering output, said methods being for predicting the response of cells to treatment with a therapeutic agent that targets a component within a cellular network, such methods comprising: (a) receiving, through said computing system input device, input that identifies levels of one or more components in a cellular network measured in a sample of the cells; (b) computing with the computing system a Network Activation State (NAS) or a Network Inhibition State (NIS) for the cells using a computational model of the cellular network; (c) applying, with the computing system, a statistical classification algorithm; and (d) generating with the computing system, and thereafter rendering at said output device, a predicted response of the cells to treatment with the therapeutic agent based at least in part on output of the statistical classification algorithm.

Also provided herein are methods for predicting the response of cells to treatment with a therapeutic agent that targets a component of an ErbB signaling pathway. Certain such methods comprise: (a) measuring, in a sample of the cells, levels of (i) heregulin (HRG) and (ii) at least one receptor selected from ErbB1, ErbB2 and ErbB3; and (b) predicting, using a computer, the response of the cells to treatment with the therapeutic agent based on the levels measured in (a), wherein elevated levels of HRG and the at least one receptor, relative to a control, predict responsiveness to treatment with the therapeutic agent. Other such methods comprise (a) measuring, in a sample of the cells, levels of one or more of ErbB1/ErbB3 heterodimers, ErbB2 monomers, ErbB2/ErbB2 homodimers, phosphorylated ErbB2/ErbB2 homodimers, ErbB2/ErbB3 heterodimers, phosphorylated ErbB1/ErbB3 heterodimers and phosphorylated ErbB2/ErbB3 heterodimers, ErbB2/ErbB4 heterodimers, phosphorylated ErbB2/ErbB4 heterodimers, ErbB3/ErbB4 heterodimers, phosphorylated ErbB3/ErbB4 heterodimers; and (b) predicting, using a computer, the response of the cells to treatment with the therapeutic agent based on the levels measured in (a), wherein a difference in the level of ErbB1/ErbB3 heterodimers, ErbB2 monomers, ErbB2/ErbB2 homodimers, ErbB2/ErbB3 heterodimers, phosphorylated ErbB1/ErbB3 heterodimers or phosphorylated ErbB2/ErbB3 heterodimers, relative to a control, predicts responsiveness to treatment with the therapeutic agent. In certain embodiments, such methods can further comprise treating cells, or a subject from whom the cells are obtained, with a therapeutic agent, based on the predicted responsiveness of the cells to the therapeutic agent.

The present invention further provides computerized methods using a computing system that comprises at least one input device configured for receiving input and at least one output device configured for rendering output, said method being for predicting the response of cells to treatment with a therapeutic agent that targets a component of an ErbB signaling pathway. Certain such methods comprise: (a) receiving, through said computing system input device, input that identifies measured levels of (i) HRG and (ii) at least one receptor selected from ErbB1, ErbB2 and ErbB3, which levels have been measured in a sample of the cells; and (b) generating with the computing system, and thereafter rendering at said output device, a predicted response of the cells to treatment with the therapeutic agent based on the measured levels, wherein elevated levels of HRG and the at least one receptor, relative to a control, predict responsiveness to treatment with the therapeutic agent. Other such methods comprise: (a) receiving, through said computing system input device, input that identifies measured levels of one or more of ErbB1/ErbB3 heterodimers, ErbB2 monomers, ErbB2/ErbB2 homodimers, ErbB2/ErbB3 heterodimers, phosphorylated ErbB1/ErbB3 heterodimers and phosphorylated ErbB2/ErbB3 heterodimers, which levels have been measured in a sample of the cells; and (b) generating and rendering, with the computing system, a predicted response of the cells to treatment with the therapeutic agent based on the measured levels, wherein a difference in the level of ErbB1/ErbB3 heterodimers, ErbB2 monomers, ErbB2/ErbB2 homodimers, ErbB2/ErbB3 heterodimers, phosphorylated ErbB1/ErbB3 heterodimers or phosphorylated ErbB2/ErbB3 heterodimers, relative to a control, predicts responsiveness to treatment with the therapeutic agent.

    • In other aspects, kits are provided herein for predicting the response of cells to treatment with a therapeutic agent that targets a component within a cellular network, the kits comprising: (a) assays for detecting levels of one or more components of the cellular network; and (b) instructions for computing a Network Activation State (NAS) or a Network Inhibition State (NIS) for the cells using a computational model of the cellular network. In certain embodiments, such kits further comprise: (c) instructions for use of the kit to predict the response of the cells to treatment with the therapeutic agent.

The present invention further provides methods for identifying a biomarker for predicting the response of cells to treatment with a therapeutic agent that targets a component within a cellular network, the method comprising: (a) measuring, in a sample of the cells, levels of one or more components of the cellular network; and (b) applying a computer-implemented method comprising: (i) computing levels of one or more additional components of the cellular network using a computational model of the cellular network; and (ii) identifying a component of the cellular network whose computed level predicts response of the cells to treatment with a therapeutic agent to thereby identifying the component as a biomarker for predicting the response of the cells to treatment with the therapeutic agent.

Also provided herein are computerized methods using a computing system that comprises at least one input device configured for receiving input and at least one output device configured for rendering output, said methods being for identifying a biomarker for predicting the response of cells to treatment with a therapeutic agent that targets a component within a cellular network, the methods comprising: (a) receiving, through said computing system input device, input that identifies measured levels of one or more components of a cellular network measured in a sample of the cells; (b) computing, with the computing system, levels of one or more additional components of the cellular network using a computational model of the cellular network; and (c) identifying, with the computing system, a component of the cellular network whose computed level predicts response of the cells to treatment with a therapeutic agent, and thereby identifying the component as a biomarker for predicting a response of the cells to treatment with the therapeutic agent.

Within still further aspects, the present invention provides computer program products comprising one or more computer-readable storage media storing computer-executable instructions that, when executed, implement any of the foregoing methods.

FIGS. 1A-1D are graphs showing the inhibition of xenograft tumor growth by treatment with the Ab #6 antibody. FIG. 1A shows the results for the MALME3M xenograft tumor model. FIG. 1B shows the results for the DU145 xenograft tumor model. FIG. 1C shows the results for the ADRr xenograft tumor model. FIG. 1D shows the results for the ACHN xenograft tumor model.

FIG. 2 is a graph plotting the concentration of phosphorylated ErbB3 (pErbB3) in untreated xenograft tumors (in pg/μg total protein) against the Growth Rate Reduction (%) observed for the xenografts when treated with Ab #6.

FIGS. 3A-3E are bar graphs showing the level of pErbB3 (FIG. 3A) and phosphorylated AKT (pAKT) (FIG. 3B) in ACHN xenograft tumor samples frozen 0, 10, or 60 minutes after xenograft dissection, and levels of ErbB1 (FIG. 3C), ErbB2 (FIG. 3D) and ErbB3 (FIG. 3E) in EKVX xenograft tumor samples frozen 0, 10, 30 or 60 minutes after xenograft dissection.

FIGS. 4A-4D show a schematic diagram of the process of converting a cartoon of a signaling pathway to a computational model. FIG. 4A shows a cartoon of the ErbB signaling pathway comprising different ligands and ErbB receptors. FIG. 4B shows a set of biochemical reactions describing the protein interactions depicted in the cartoon. FIG. 4C shows a set of fluxes derived from the set of biochemical reactions. FIG. 4D shows a set of non-linear ordinary differential equations (ODEs) based on mass action kinetics describing signal transduction networks.

FIGS. 5A-5B are graphs showing the levels of phospho-ErbB3, phospho-ErbB2, phospho-ErbB1 and phospho-AKT over time in cells stimulated with nine different concentrations of heregulin (HRG) (FIG. 5A) or betacellulin (FIG. 5B) in ADRr ovarian cancer cells. The parameters of the computational model were calibrated for the simulation results (solid lines) to describe the experimental data (dots).

FIG. 6 is a bar graph showing the results of the local sensitivity analysis to identify key markers for activation of ErbB3.

FIG. 7 is a graph showing the computed levels of pErbB3 in the MALME3M, DU145, ADRr and ACHN cell lines.

FIGS. 8A-8B are graphs showing use of NAS values to predict responsiveness of 15 cell lines to Ab #6 treatment, based on threshold NAS values established from the 4 training cell lines (MALME3M, DU145, ADRr and ACHN). FIG. 8A is a bar graph plotting simulated pErbB3 levels for the 19 cell lines, from highest to lowest pErbB3 levels. FIG. 8B is a graph ranking the 19 cell lines from highest to lowest NAS value, with cell lines having NAS values below MALME3M being ranked as nonresponders (NR), cell lines having NAS values above ADRr being ranked as responders and cell lines having NAS values between MALME3M and ADRr being ranked as indeterminate.

FIGS. 9A-9C are graphs showing the inhibition of xenograft tumor growth by treatment with the Ab #6 antibody. FIG. 9A shows the results for the IGROV1 xenograft tumor model. FIG. 9B shows the results for the OVCAR8 xenograft tumor model. FIG. 9C shows the results for the SKOV3 xenograft tumor model.

FIGS. 10A-10D are graphs in which the log of the concentration of one ErbB receptor is plotted against the log of the concentration of one or more of the other ErbB receptors. Receptor values are shown for cell lines classified as Ab #6 responders or non-responders. FIG. 10A plots ErbB2 versus ErbB1. FIG. 10B plots ErbB3 versus ErbB1. FIG. 10C plots ErbB2 versus ErbB3. FIG. 10D plots ErbB1 versus ErbB2 versus ErbB3.

FIG. 11 shows a graph in which the log concentration of HRG is plotted against the log concentration of ErbB1. In the graph, responding vs. non-responding cell lines tested in xenograft studies segregate.

FIGS. 12A-12C are graphs in which the log normalized expression levels (in pg/μg, determined by ELISA) of different components of the ErbB signaling pathway in xenograft cell lines and human tumor samples are plotted. FIG. 12A plots ErbB2 versus ErbB1. FIG. 12B plots ErbB4 versus ErbB3. FIG. 12C plots HRG-β1 versus BTC.

FIGS. 13A-13D are graphs showing quantitative immunohistochemistry (qIHC) results for xenograft cell lines and human tumor samples. FIG. 13A shows a cell line standard curve for ErbB1. FIGS. 13B, 13C and 13D are bar graphs plotting the qIHC scores for ErbB1, ErbB2 and ErbB3, respectively, in the xenograft cell lines (red bars) and human tumor samples (blue bars).

FIG. 14 is a bar graph showing the integrated phosphorylated ErbB1:3 heterodimer level (amount of time-integrated heterodimers per cell) computed for eight cell lines, which segregated them into Ab #6 non-responders (MALME3M, BT474, IGROV1 and ADRr) and responders (OVCAR8, SKOV3, DU145 and ACHN).

FIGS. 15A and 15B are flowcharts that include various functional steps and acts that can be performed during implementation of certain embodiments of the invention.

FIG. 16 is one embodiment of a computing environment that can be used to implement certain aspects of the invention.

FIGS. 17A-D are graphs of inhibition curves for cells treated with the bispecific antibody H3×B1D2. FIGS. 17A and 17B show the inhibition curves for pErbB3 levels and pAKT levels, respectively, in OVCAR8 cells. FIGS. 17C and 17D show the inhibition curves for pErbB3 levels and pAKT levels, respectively, in OVCAR8 cells transfected with HER2/ErbB2 (OVCAR8-HER2 cells). The solid line represents the simulated data from the computational model, while the circles represent the experimentally determined data. The simulated IC50 values (DR50sim) and the experimentally determined IC50 values (DR50data) also are shown.

FIGS. 18A-D are graphs of inhibition curves for cells treated with the bispecific antibody H3×B1D2. FIGS. 18A and 18B show the inhibition curves for pErbB3 levels and pAKT levels, respectively, in ADrR cells. The solid line represents the simulated data from the computational model, while the circles represent the experimentally determined data. The simulated IC50 values (DR50sim) and the experimentally determined IC50 values (DR50data) also are shown. FIGS. 18C and 18D show the simulated inhibition curves for pErbB3 levels and pAKT levels, respectively, in ADrR cells with simulated treatment with ErbB1 RNAi.

FIGS. 19A-C are graphs of the in vivo-determined relative growth rates (RGR) for a panel of tumor cells in a xenograft model treated with H3×B1D2 plotted against the computed levels of ErbB2 monomers (FIG. 19A), ErbB2:ErbB2 homodimers (FIG. 19B) and ErbB2:ErbB3 heterodimers (FIG. 19C) in the panel of tumor cells in the absence of H3×B1D2.

FIGS. 20A-B are graphs of the in vivo-determined relative growth rates (RGR) for a panel of tumor cells in a xenograft model treated with H3×B1D2 plotted against the computed relative levels of ErbB2:ErbB3 heterodimers (FIG. 20A) and ErbB1:ErbB3 heterodimers (FIG. 20B) in the panel of tumor cells in the simulated absence and presence of H3×B1D2.

FIG. 21 provides graphic representations of predicted (plotted lines) and actual (data points) HRG-induced pErbB3 and pAKT signaling data from the ErbB2-overexpres sing cell line BT474-M3 at HRG doses as indicated, as detailed in Example 11.

The present invention generally provides methods, systems and computer program products in which the activation state of one or more elements (e.g., ErbB3) in one or more cellular signaling pathways is determined through use of indirect markers rather than through direct measurement of levels of the one or more elements. Such indirect measurements may be used to determine the activation state of elements in a signaling pathway via computer simulation that models the signaling pathway. Using methods provided herein, this information may then be used to predict the responsiveness of cells to therapeutic agents, and thus to select a therapeutic treatment for a patient having a disease or disorder (e.g., cancer).

Through the use of the present invention, determination of a representative phosphorylation signature of a sample of tumor cells can be achieved. This allows for the use of phosphorylated cellular proteins as biomarkers for responsiveness to therapeutic agents without the need to directly measure levels of such phosphoproteins in patient samples, thus avoiding the potential problems associated with phosphoprotein measurement (e.g., instability, unreliability). Such simulation-determined phosphorylation signatures can be used to accurately predict the responsiveness of cells to anti-neoplastic therapeutic agents, and hence can be used to avoid administering cancer drugs to the patient that are ineffective to treat the patient's cancer. Furthermore, the disclosed methods allow for the simulation of the levels of other labile components within a cellular network, such as homo- and/or heterodimers of receptors, or phosphorylated homo- and/or heterodimers, within cellular networks, which components also can be used to predict the responsiveness of cells to therapeutic agents. Still further, the disclosed methods allow for the simulation of the effect of a therapeutic agent on components within a cellular network, which also can be used to predict the responsiveness of cells to the therapeutic agent.

In certain methods, the levels of one or more stable cellular components (such as cell surface receptors, ligands and the like) within a cellular network are measured in a sample of cells (e.g., cells in a tumor biopsy or resected tumor). Based on these measurements, a Network Activation State (or NAS) or a Network Inhibition State (NIS) for the cells is computed using a computational model (e.g., a mechanistic computational model) of one or more signal transduction networks. For example, the NAS can be a numeric value representing the computed level of a phosphorylated protein (that has not been directly measured) within the cellular network. Alternatively, the NIS can be a numeric value representing the computed level of a component within the cellular network in the simulated presence of the therapeutic agent (as compared to the level of the component in the simulated absence of the therapeutic agent). By comparing the computed NAS or NIS to control values representing threshold values for responsiveness or non-responsiveness, one can predict whether the cells are likely to respond to therapeutic treatment or not.

Thus, the use of “indirect” markers of cellular responsiveness to treatment is disclosed, wherein the level of that “indirect” marker (e.g., a phosphorylated protein or dimer within an activated signaling pathway) is not directly measured but rather is computed, or simulated, based on the levels of other cellular components that are directly measured. The methods also can involve the use of statistical classification algorithms, for example in combination with computation of the NAS or NIS. In the context of the present invention and as described further in the Examples section herein, a mechanistic computational model of the ErbB signaling pathway has been successfully used to compute levels of phosphorylated ErbB3 (pErbB3), as an indicator of activation of the ErbB cellular network (or NAS), and the simulated pErbB3 levels were shown to accurately predict the responsiveness of tumor cell lines to treatment with an anti-ErbB3 antibody, Ab #6, in an in vivo xenograft system. Additionally, as described further in the Examples section herein, a mechanistic computational model of the ErbB signaling pathway has been successfully used to compute relative levels of ErbB2/ErbB3 heterodimer and ErbB1/ErbB3 heterodimer in the simulated absence and presence of a therapeutic agent, as an indicator of inhibition of the ErbB cellular network (or NIS), and the simulated relative levels of ErbB2/ErbB3 and ErbB1/ErbB3 heterodimers were shown to accurately predict the responsiveness of tumor cell lines to treatment with an anti-ErbB3×anti-ErbB2 bispecific antibody, H3×B1D2, in an in vivo xenograft system.

So that the invention may be more readily understood, certain terms are first defined.

As used herein, the term “therapeutic agent” is intended to encompass any and all compounds that have an ability to decrease or inhibit the severity of the symptoms of a disease or disorder, or increase the frequency and/or duration of symptom-free or symptom-reduced periods in a disease or disorder, or inhibit or prevent impairment or disability due to a disease or disorder affliction, or inhibit or delay progression of a disease or disorder, or inhibit or delay onset of a disease or disorder, or inhibit or prevent infection in an infectious disease or disorder. Non-limiting examples of therapeutic agents include small organic molecules, monoclonal antibodies, bispecific antibodies, recombinantly engineered biologics, RNAi compounds and the like.

As used herein, a therapeutic agent that “targets a component within a cellular network” refers to an agent whose therapeutic activity results, at least in part, from the agent having a specific direct or indirect effect on the activity of a component within a cellular network. Non-limiting examples of therapeutic agents that target a component within a cellular network include the monoclonal antibody cetuximab, which specifically binds to ErbB1, thus specifically targeting ErbB1 within the ErbB cellular network, and gefitinib, a small molecule that specifically inhibits the tyrosine kinase (TK) domain of ErbB1, thus specifically targeting ErbB1-TK within the ErbB cellular network.

As used herein, the term “Network Activation State” or “NAS” refers to an indicator, typically a numeric value, that reflects, or corresponds to, the level of activation of a cellular network. A NAS typically is computed using a computational model of the cellular network. A NAS can represent, for example, the simulated level of one or more phosphorylated proteins within a cellular network. While one or more phosphoprotein levels are a preferred embodiment for the NAS (e.g., pErbB3 level as described further in Example 7 or phosphorylated ErbB homodimer or heterodimer level, such as pErbB1/ErbB3 heterodimer level as described further in Example 10, or the levels of downstream kinases such as PI3K), other cellular components within an activated signaling pathway can serve as an indicator(s) of network activation and thus can be used as the NAS, including but not limited to receptor dimerization (homodimers and heterodimers, such as levels of ErbB1/ErbB1 or ErbB2/ErbB2 homodimers or levels of ErbB1/ErbB 2, ErbB1/ErbB 3, ErbB1/ErbB 4, ErbB2/ErbB 3 or ErbB2/ErbB 4 heterodimers), protein cleavage, activation of transcription factors and activation of gene expression. The Network Activation State of a cell, e.g., a tumor cell, is an indicator of the dependence of the cell on that signaling pathway, which can be inhibited by a therapeutic agent that targets that particular signaling pathway.

As used herein, the term “Network Inhibition State” or “NIS” refers to an indicator, typically a numeric value, that reflects, or corresponds to, the level of inhibition of a cellular network. A NIS typically is computed using a computational model of the cellular network. A NIS can represent, for example, the simulated level of one or more components within a cellular network in the simulated presence of a therapeutic agent, as compared to (or relative to) the simulated level(s) in the simulated absence of the therapeutic agent. For example, the NIS can be a ratio of the level of one or more components in the simulated presence of a therapeutic agent and the level of those one or more components in the simulated absence of the therapeutic agent. A non-limiting example of a NIS is the computed relative level of one or more homo- or heterodimers (such as relative levels of ErbB2/3 and ErbB1/3 heterodimers) computed in the simulated absence and presence of a therapeutic agent (i.e., the levels in the simulated presence of the therapeutic agent as compared to the levels in the simulated absence of the therapeutic agent). However, it will be appreciated that other cellular components within a signaling pathway whose levels are modulated by a therapeutic agent also can serve as an indicator(s) of network inhibition and thus their levels can be used as indicators of the NIS. The NIS of a cell and the NAS of a cell, e.g., a tumor cell, are indicators of the impact of the therapeutic agent on components within a signaling pathway of the cell, and can be predictive of responsiveness of the cell to the effects of the therapeutic agent.

The term “computational model of a cellular network” refers to a model, such as a computer program, that translates a biological pathway diagram or cartoon (e.g., a set of protein interactions relevant to cancer) into a set of mathematical equations amenable for subsequent simulation and analysis. Certain information (e.g., ligand and/or receptor protein concentrations, rate constants) can be input into the model, which can then simulate additional information that may not be readily measurable (e.g., phosphoprotein levels). A Network Activation State (or NAS) or a Network Inhibition State (or NIS) for a cellular network can be computed using a computational model of the cellular network as described herein.

As used herein, the term “algorithm” generally refers to a set of instructions, or procedures, or formulas, for carrying out a method or solving a problem. The term “statistical classification algorithm” refers to an algorithm that defines a statistical relationship between one or more measurable parameters, or inputs, (e.g., protein levels measured in a tissue sample) and a particular outcome, or output, (e.g., responsiveness to a therapeutic agent) such that a classification, or prediction, can be made (e.g., responder versus non-responder to a therapeutic agent).

As used herein, the term “biomarker” refers to a substance (e.g., protein, mRNA, allele) within, or expressed by, a cell, wherein the biomarker correlates with the responsiveness of the disease to a given treatment.

As used herein, the term “direct biomarker” refers to a substance (e.g., protein, mRNA, allele) within, or expressed by, a cell, wherein the direct biomarker correlates with the responsiveness of the disease to a given treatment, and wherein the presence or level of that substance is directly measured in the cell to thereby predict responsiveness of the disease to a given treatment.

As used herein, the term “indirect biomarker” refers to a substance (e.g., protein, mRNA, allele) within, or expressed by, a cell, wherein the indirect biomarker correlates with the responsiveness of the disease to a given treatment, and wherein the presence or level of that substance is not directly measured in the cell, but rather is determined by indirect means, such as by simulation using a computational model, to thereby predict responsiveness of the disease to a given treatment.

An “antibody,” as used herein is a protein consisting of one or more polypeptides comprising binding domains substantially encoded by immunoglobulin genes or fragments of immunoglobulin genes, wherein the protein immunospecifically binds to an antigen. The recognized immunoglobulin genes include the kappa, lambda, alpha, gamma, delta, epsilon and mu constant region genes, as well as myriad immunoglobulin variable region genes. Light chains are classified as either kappa or lambda. Heavy chains are classified as gamma, mu, alpha, delta, or epsilon, which in turn define the immunoglobulin classes, IgG, IgM, IgA, IgD and IgE, respectively. A typical immunoglobulin structural unit comprises a tetramer that is composed of two identical pairs of polypeptide chains, each pair having one “light” (about 25 kD) and one “heavy” chain (about 50-70 kD). “VL” and VH” refer to these light and heavy chains respectively.

Antibodies include intact immunoglobulins as well as antigen-binding fragments thereof, which may be produced by digestion with various peptidases, or synthesized de novo either chemically or using recombinant DNA technology. Such fragments include, for example, F(ab)2 dimers and Fab monomers. Preferred antibodies include single chain antibodies (antibodies that exist as a single polypeptide chain), more preferably single chain Fv antibodies (scFv) in which a VH and a VL chain are joined together (directly or through a peptide linker) to form a continuous polypeptide. 5,132,405, and 4,956,778).

“Immunospecific” or “immunospecifically” refer to antibodies that bind via domains substantially encoded by immunoglobulin genes or fragments of immunoglobulin genes to one or more epitopes of a protein of interest, but which do not substantially recognize and bind other molecules in a sample containing a mixed population of antigenic molecules. Typically, an antibody binds to a cognate antigen with a Kd of at least 50 nM, as measured by a surface plasmon resonance assay or a cell binding assay. The use of such assays is well known in the art, and is exemplified in Example 13, herein.

An “anti-ErbB3 antibody” is an isolated antibody that immunospecifically binds to the ectodomain of ErbB3. Such binding to ErbB3 exhibits at least Kd of 50 nM as measured by a surface plasmon resonance assay or a cell binding assay. Anti-ErbB3 antibodies that inhibit EGF-like ligand mediated phosphorylation of ErbB3 are preferred. EGF-like ligands include EGF, TGFα, betacellulin, heparin-binding epidermal growth factor, biregulin, epigen, epiregulin, and amphiregulin, which typically bind to ErbB1 and induce heterodimerization of ErbB1 with ErbB3.

The term “bispecific” as used herein refers to a protein comprising two antigen-binding sites, a first binding site having affinity for a first antigen or epitope and a second binding site having binding affinity for a second antigen or epitope distinct from the first.

As used herein, the term “subject” or “patient” includes any human or nonhuman animal having a disease or disorder for which response to treatment with a therapeutic agent can be predicted using the methods of the invention, such as a subject or patient with a tumor. The term “nonhuman animal” includes all vertebrates, e.g., mammals and non-mammals, such as nonhuman primates, sheep, dogs, cats, horses, cows, chickens, etc.

Many embodiments of the present invention comprise one or more computing system(s), such as special purpose and general-purpose computers including various computer hardware, such as input devices, output devices, processor(s), storage media and other corresponding computer components.

Many embodiments of the invention also include computer-readable storage media having computer-executable instructions or data structures stored thereon (including the instructions and data structures defined herein, such as the mechanistic computational models, the measured protein and biomarker levels, classification algorithms, mutation statuses, and so forth) and that are specifically configured for implementing the processes described and claimed herein. Such computer-readable storage media can be any available media that can be accessed by a general purpose or special purpose computer. By way of example, and not limitation, such computer-readable storage media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store desired program code means, modules, and software in the form of computer-executable instructions or data structures and which can be accessed by a general purpose or special purpose computer.

In some instances, transmission media can also be used to carry the computer-executable instructions, such that the present invention also extends to applications, systems and other embodiments incorporating transmission media carrying the computer-executable instructions that are executed to perform one or more of the processes described herein. When information is transferred or provided over a network or another communications connection (either hardwired, wireless, or a combination of hardwired or wireless) to a computer, the computer properly views the connection as a computer-readable transmission medium.

The terms “computer-executable instructions”, “executable instructions”, “modules” and “computing modules”, are sometimes used interchangeably herein to reference the computer code, data structures and software that is accessed and executed by one or more computing processors and processing components of one or more computing systems to implement certain processes of the present invention, as described within this paper and as recited in the claims.

Additional aspects regarding the foregoing and various additional aspects of this disclosure are described in further detail in the following subsections, which should not be construed as limiting.

Mechanistic computational models can be viewed as predictive mathematical descriptions of the molecular interactions in a protein network. In at least certain embodiments, the methods provided herein for predicting responses to therapeutic agents involve the use of a mechanistic computational model. Mechanistic computational models translate a biological pathway diagram or cartoon (e.g., a set of protein interactions occurring along a signal transduction pathway, such as a pathway relevant to cancer) into a set of mathematical equations amenable for subsequent simulation and analysis. Thus, the first step in construction of the model is generation of a detailed diagram, or cartoon, representation of the biological pathway which includes the relevant proteins and molecules involved in the pathway. Critical decisions must be made regarding which proteins and molecules are to be included, as well as the biological reactions that connect them. Information available in the scientific literature about which proteins and molecules are involved in the pathway and which biological reactions connect them is collected and used in the generation of the cartoon representation of the biological pathway.

Once the cartoon representation of the biological pathway has been generated, this information is translated into a system of equations that represent the protein-protein interactions within the pathway, also referred to as a cellular network. The computational models representing the biochemical reaction networks of signal transduction networks are based on, for example, non-linear ordinary differential equations (ODEs), stochastic models, boolean or fuzzy logic models or petri nets. They are composed of differential equations that require two types of parameters that must be experimentally measured or estimated: initial species number (c0,i for the ith species) and the rate constants (kj for the jth rate). A schematic diagram of the process of building a computational model is shown in FIGS. 4A-D, which illustrates a cartoon diagram of the ErbB signaling pathway (FIG. 4A), translation of this pathway into sets of biochemical reaction and fluxes (FIGS. 4B and 4C) and representation of the protein-protein interactions by a set of differential equations (FIG. 4D).

The computational model typically is built as a set of executable instructions written in a computer scripting language, for example using the MATLAB software Simbiology (The MathWorks, Natick, Mass.), optionally in conjunction with The Systems Biology Toolbox 2 for MATLAB (SBtoolbox2.0rg), or JACOBIAN modeling software (Numerica Technology, Cambridge, Mass.). However, it will be appreciated that the scope of the invention also extends to the development and use of computational models, built with software and interfaces other than those built with MATLAB or JACOBIAN software and interfaces. The invention also extends to embodiments that utilize computational models that have been prebuilt by a third party source and downloaded to the computing system implementing other aspects of the invention.

Prior to model calibration, the values for as many parameters as possible are specified and input into the computing system, based on information from the scientific literature, e.g., protein levels, binding affinities, binding rate constants for ligands to their cognate receptors. Parameter values that are not available in the scientific literature can be obtained experimentally.

The model is “trained” by optimizing its output against experimentally obtained data that is input into a computing system performing the training. By fitting the model to experimental data, the optimal set of model parameters is selected. The process of model calibration involves modification of assumptions and parameter estimates. To calibrate the model, one must first identify a subset of proteins and parameters that are especially important biologically for translating a ligand stimulus into a downstream signaling event. This process is termed sensitivity analysis, which, more precisely, is a mathematical tool that measures the change in an output, such as substrate phosphorylation, in response to changes in protein concentrations and kinetic parameters within the pathway. The fully normalized sensitivity (sij(t)) of the ith observable ci(t) with respect to a change in the jth rate constant (kj) is given by the following equation:

s ij ( t ) ln ( c i ( t ) ) ln ( k j )

Model calibration is then performed by a computing system using local and global optimization methods (such as, but not limited to, Genetic Algorithms, simulated annealing, Levenberg-Marquardt optimization, and so forth) that minimize the distance between the experimental data and the simulation results by varying the parameters and initial protein concentrations identified in the sensitivity analysis. The computing system can be configured to automatically vary the parameters during calibration or to vary the parameters only in response to incrementally added user input.

A number of computational models for various signaling pathways have been described in the scientific literature (see e.g., Kholodenko, B. N. et al. (1999) J. Biol. Chem. 274:30169-30181; Schoeberl, B. et al. (2002) Nat. Biotech. 20: 370-375; Hayakeyama, M. et al. (2003) Biochem. J. 373:451-463; Nielsen, U. B. and Schoeberl, B. (2005) IDrugs 8:822-826; Schoeberl, B. et al. (2006) Conf. Proc. IEEE Eng. Med. Biol. Soc. 1:53-54; Schoeberl, B. et al. (2006) IBM J. Res. Dev. 50:645; Fitzgerald, J. B. et al. (2006) Nat. Chem. Biol. 2:458-466; Kholodenko, B. N. (2007) Nat. Cell. Biol. 9:324-330; Birtwistle, M. R. et al. (2007) Molecular Systems Biology 3:144; Hinow, P. et al. (2007) Theoretical Biology and Medical Modelling 4:14). Additionally, the building and use of computational models is reviewed in Kholodenko, R. N. (2006) Nature Reviews: Mol. Cell. Biol. 7:165-176 and Kumar, N. et al. (2006) Drug Discovery Today 11:806-811.

One of the computational models used in certain methods provided herein is a model of an ErbB signaling pathway. Building of a representative computational model of the ErbB signaling pathway is described in detail in Example 4. As used herein, the term “ErbB signaling pathway” is intended to encompass signal transduction pathways that initiate through interaction of a ligand with a receptor of the ErbB family. Components within an ErbB signaling pathway may include: (i) one or more ligands, examples of which include HRG, betacellulin (BTC), epidermal growth factor (EGF), heparin-binding epidermal growth factor (HB-EGF), transforming growth factor alpha (TGFα), amphiregulin (AR), epigen (EPG) and epiregulin (EPR); (ii) one or more receptors, examples of which include ErbB1, ErbB2, ErbB3 and ErbB4; and (iii) intracellular kinases, phosphatases and substrates, examples of which include phosphatidylinositol 3-kinase (PI3K), phosphatidylinositol bisphosphate (PIP2), phosphatidylinositol trisphosphate (PIP3), phosphatase and tensin homolog (PTEN), pyruvate dehydrogenase kinase isozyme 1 (PDK1), AKT, RAS, RAF, MEK, the extracellular signal-regulated kinase (ERK), protein phosphatase 2A (PP2A) and SRC protein tyrosine kinase.

Another computational model used in certain methods provided herein is a model of an IGF1R signaling pathway. As used herein, the term “IGF1R signaling pathway” is intended to encompass signal transduction pathways that initiate through interaction of a ligand with a receptor of the insulin growth factor 1 family. Components within an IGF1R signaling pathway may include: (i) one or more ligands, examples of which include insulin growth factor 1 (IGF1); (ii) one or more receptors, examples of which include IGF1R and the insulin receptor; (iii) one or more IGF binding proteins and (iv) intracellular kinases and substrates, examples of which include insulin receptor substrate 2 (IRS2), PI3K, AKT, Bc1-2 related protein BAD, RAS, RAF, MEK and mitogen-activated protein kinase (MAPK).

Yet another computational model used in certain methods provided herein is a model of a c-Met signaling pathway. As used herein, the term “c-Met signaling pathway” is intended to encompass signal transduction pathways that initiate through interaction of a ligand with a c-Met receptor protein tyrosine kinase. Components within a c-Met signaling pathway may include: (i) one or more ligands, examples of which include hepatocyte growth factor (HGF); (ii) one or more receptors, examples of which include the c-Met receptor protein tyrosine kinase; and (iii) intracellular kinases and substrates, examples of which include PI3K, growth factor receptor-bound protein 2 (GRB2), Src homologous and collagen protein (SHC), SRC protein tyrosine kinase and GAB1 scaffolding protein, as well as RAS, RAF, MEK and mitogen-activated protein kinase (MAPK).

Yet another computational model used in certain methods provided herein is a model comprising any combination of two or more growth factor signaling pathways, such as IGR1R and the ErbB receptor signaling, ErbB receptor signaling and c-Met signaling or IGF1-R, ErbB and c-Met signaling in combination.

Despite the specificity of the foregoing examples, it will be appreciated that other computational models (e.g., for signaling pathways such as TNF, IL-2, PDGF, FGF, TRAIL, integrins, cytokines and virtually any other pathway) can also be incorporated into and utilized by the embodiments of the present invention.

In certain embodiments, the presence of one or more therapeutic agents can be simulated in the computational model. A computational representation of the therapeutic agent(s) can be constructed using mass-action reaction equations that describe the binding of the agent(s) to its cellular target or otherwise describe the effect of the agent(s) on the cellular pathway being modeled. Parameters for the binding events, or other biological effects, can be obtained by direct experimental measurements, as well as by training of the model to match data for the effect of the therapeutic agent on the cell. For example, for antibody agents, the on-rate and off-rate for binding of the antibody to its target antigen can be experimentally determined by standard methods (such as BIACore or KinExA technology) and those parameters can be incorporated into the computational model. Additionally, for example, for bispecific agents, a cross-linking parameter, as a measure of the number of bispecific molecules bound to each individual target of the bispecific or to both targets of the bispecific, can be used as a training parameter for the computational model. The cross-linking parameter can be obtained by taking the overall observed binding affinity (determined by standard FACS analysis) and fitting to a standard logistic binding equation. In addition to the foregoing, additional parameters that may be pharmaceutically relevant for a particular therapeutic agent, and thus to be represented in the computational model, are known to the ordinarily skilled artisan.

When the effect of a therapeutic agent is to be represented in the computational model, a single agent can be modeled or multiple agents can be modeled in combination to thereby simulate the effect of the combination therapy on the cellular responses. For example, in one embodiment, two antibodies that each bind to different target antigens can be represented in the computational model. In another embodiment, an antibody that targets a particular signaling pathway (e.g., an ErbB pathway) and a small molecule inhibitor of that same signaling pathway (e.g., an ErbB pathway) can be simultaneously represented in the computational model to assess the effect of such combination therapy of the signaling pathway in the cell.

In at least certain embodiments, the methods provided herein for predicting responses to therapeutic agents (e.g., generating predicted responses to therapeutic agents with a computer) and methods for treating patients having malignant tumors involve the use of one or more statistical classification algorithms.

One goal of a statistical model is to discern a relationship between, for example, protein levels measured in tissue samples, as well as the activation levels (e.g., the Network Activation State or “NAS”) or the inhibition levels (e.g., the Network Inhibition State or “NIS”) computed by the biochemical model, on the one hand, and the patient's response to a therapeutic agent on the other. Thus, a statistical classification algorithm defines a statistical relationship between one or more measurable parameters, or inputs, (e.g., protein levels measured in a tissue sample, computed NAS or NIS values) and a particular outcome, or output, (e.g., responsiveness to a therapeutic agent) such that a classification, or prediction, can be made (e.g., responder versus non-responder to a therapeutic agent). Accordingly, a statistical model helps to identify the threshold dividing responders and non-responders and also helps to define the uncertainty around the defined threshold.

Various types of statistical classifier systems have been described in the art and may be suitable for use in the methods of the invention, non-limiting examples of which include principal component analysis (PCA), partial least square regression (PLSR), trilinear PLSR, Fuzzy logic and Bayesian inference, random forest (RF), classification and regression tree (C&RT), boosted tree, neural network (NN), support vector machine (SVM), general chi-squared automatic interaction detector model, interactive tree, multiadaptive regression spline, machine learning classifier, and combinations thereof. See for example PCT Publication WO 2007/109571.

To train a statistical classification algorithm, in a preferred embodiment, to use terms of the art: the “machine” or “computing system” (e.g., computer) “learns” the relationship between the protein expression level and network activation state (NAS) with the actual patient response (e.g., responder, non-responder) by examining a number of “training” examples that have been input into the computer by means of the “classifier” (statistical algorithm). This process is known as “supervised learning”, since there is a collection of samples for which both the input (protein levels, measured and computed) and output (e.g., response to a therapeutic agent) are known a priori. An integral part of applying a statistical classification algorithm is the selection of informative features (a feature being any measured or computed protein level or protein activation level), as well as the determination of the optimal threshold for the score produced by the algorithm.

It is generally preferable to validate a statistical classification algorithm on a separate “test set” of samples, for which the output is known, but is not disclosed to the classifier. By comparing predictions with known results, the performance can be measured. When the number of samples is small, this may not be practicable, and there are established techniques to get around this requirement. Principal among them is cross-validation, discussed further below. The output of a classifier is a score which can be translated into a class prediction if desired. Generally, the process of training a classifier also includes methods to determine the optimal thresholds.

In order to test the ability to make predictions, the general procedure in cross-validation is to set aside a portion of the data (a fraction of the samples) as ‘training data’ or the ‘training (data) set’; the remaining data is referred to as the ‘test (data) set’. Typically, this procedure is repeated many times, each time using a different fraction of the data for training and testing. A larger training set is clearly beneficial (the more examples, the better the classifier), but so is a large test set (the more predictions you can verify, the higher your confidence in future predictions). When the sample size is small, this procedure can be amended so that a single sample is set aside as test data, and the classifier is trained using all remaining samples. A prediction is made on the single left-out sample. This is called ‘leave-one-out cross-validation’ (LOOCV), and is well established in the art. This procedure is repeated for each left-out sample in turn. Once all predictions have been made, performance is evaluated.

In the statistical classification algorithms used in the methods of the invention, various parameters, or informative features, or pieces of information, may be used as the input data. Non-limiting examples of such input data include protein levels of one or more components of a cellular network obtained from a cell sample, a Network Activation State (NAS) or Network Inhibition State (NIS) computed using a computational model, the mutation status of one or more proteins in the cell sample, the age of the subject for which responsiveness to treatment is being investigated and the gender of the subject for which responsiveness to treatment is being investigated.

In the methods provided herein for generating predicted responsiveness to treatment with a therapeutic agent and for treating a patient having a malignant tumor, one procedure typically involves measuring, in a cell sample, the levels of one or more components of a cellular network or, alternatively, inputting measured levels of one or more components of a cellular network, which have been obtained from measurements taken from a cell sample, into a computing system. For example, a sample of a tumor can be obtained by standard methods from a patient with the tumor and the levels of one or more components of a cellular network can be measured in the sample of the tumor.

Input can be manually entered into the computing system. Input can also be automatically input or downloaded to the computing system, in some instances, such as when computerized measuring devices are connected to the computing system receiving and utilizing the input to implement features of the invention. In this regard, it will also be appreciated that any measurement information and any status information (e.g., mutation status information), as well as any other tissue/patient data, described in this paper can be obtained through the use of computerized devices that automatically obtain and download the measurement and status information to one or more other computing systems that use the data to perform the described processes of the invention.

As used herein, the term “level” of a component refers to the amount or concentration of the component present in a sample. Component levels can be measured using any of a variety of well known techniques. The level typically is determined by measuring protein levels, but alternatively the level can be determined in some cases by measuring mRNA levels, which may be followed by conversion of the mRNA levels to predicted protein levels. The levels of proteins (e.g., monomers, homodimers, or heterodimers) can be measured using one or more techniques well known in the art, non-limiting examples of which include quantitative fluorescence activated cell sorting (qFACS), enzyme linked immunosorbent assay (ELISA, Luminex), immunohistochemistry (IHC), quantitative immunohistochemistry (qIHC), proximity based methods (e.g., Forster resonance energy transfer-based methods, biomolecular fluorescence complementation (BiFC), VeraTag™ or DNA-Programmed Chemistry™ (DPC™)), mass spectrometry, Western (immunoblot) assay and coimmunoprecipitation. Protein levels may be expressed as pg detected protein/μg total protein

Protein or mRNA levels can be determined in cell lysates. Cell lysates can be prepared, for example, as described in detail in Example 2. Furthermore, representative examples of the use of ELISAs for determining protein levels are described in detail in Examples 2 and 4, a representative example of the use of qFACS for determining protein levels are described in detail in Example 4 and a representative example of the measurement of mRNA levels and conversion to protein levels is described in detail in Example 4 (for HRG-β1).

The tumor sample can be, for example, a fresh cell sample, a fresh frozen sample or a fixed tissue sample. For patient tissue samples, archived tissue blocks may be more easily accessible than fresh frozen samples. Thus, in a preferred embodiment, a formalin-fixed paraffin-embedded (FFPE) tissue sample is used and the level of components (e.g., ligands, receptors) can be determined by semi-quantitative immunohistochemistry (IHC) (described further in Example 9). To convert the semi-quantitative IHC information into a concentration amenable for input into a computational model, a control slide containing cell plugs or xenografts with known receptor and/or ligand expression levels can be compared to the patient sample. Moreover, conversion factors can be determined to convert an expression level obtained in dimensionless units (e.g., protein or mRNA amounts) into a concentration level that can be input into a computational model, described in further detail in Example 4 for BTC and HRG expression levels.

Ligand mRNA and ligand protein levels generally correlate reasonably well. Therefore qRT-PCR can be used to determine the mRNA expression levels in cell lysates from tumor cell lines and xenografts as well from FFPEs.

In certain aspects, the present invention provides methods for predicting the response of cells to treatment with a therapeutic agent that targets a component within a cellular network. Such methods generally comprise the elements indicated below in Predictive Methods 1-5.

    • (a) obtaining measurements of levels of one or more components of the cellular network, by measuring the level(s) present in a sample of the cells of the one or more components of the cellular network; and
    • (b) applying a computer-implemented method comprising:
      • (i) computing a Network Activation State (NAS) or a Network Inhibition State (NIS) for the cells using a computational model of the cellular network input with the measurements; and
      • (ii) computing and outputting a predicted response of the cells to treatment with the therapeutic agent based at least in part on the NAS or the NIS computed in (i).
    • (a) measuring, in a sample of the cells, levels of one or more components of the cellular network; and
    • (b) applying a computer-implemented method comprising:
      • (i) computing a Network Activation State (NAS) or a Network Inhibition State (NIS) for the cells using a computational model of the cellular network;
      • (ii) applying a statistical classification algorithm; and
      • (iii) predicting the response of the cells to treatment with the therapeutic agent based at least in part on output of the statistical classification algorithm.
    • a) a computing system receiving, through an input device, input that identifies levels of one or more components in a cellular network measured in a sample of the cells;
    • b) the computing system computing a Network Activation State (NAS) or a Network Inhibition State (NIS) for the cells using a computational model of the cellular network; and
    • c) the computing system generating, and thereafter rendering at an output device, a predicted response of the cells to treatment with the therapeutic agent based at least in part on the NAS or NIS computed in b).
    • a) measuring, in a sample of the cells, levels of one or more components of the cellular network;
    • b) computing a Network Activation State (NAS) or a Network Inhibition State (NIS) for the cells using a computational model of the cellular network;
    • c) applying a statistical classification algorithm; and
    • d) predicting the response of the cells to treatment with the therapeutic agent based at least in part on output of the statistical classification algorithm.
    • a) a computing system receiving, through an input device, input that identifies levels of one or more components in a cellular network measured in a sample of the cells;
    • b) the computing system computing a Network Activation State (NAS) or a Network Inhibition State (NIS) for the cells using a computational model of the cellular network;
    • c) the computing system applying a statistical classification algorithm; and
    • d) the computing system generating, and thereafter rendering at an output device, a predicted response of the cells to treatment with the therapeutic agent based on output of the statistical classification algorithm.

The above methods can, but need not, further comprise treating cells, or a patient from whom the cells are obtained, with a therapeutic agent, based on the predicted responsiveness of the cells to the therapeutic agent.

In various aspects, the levels of component(s) detected serve to indicate the predicted effectiveness of one or more specific therapeutic agents. In many cases, this means that if the component(s) are detected at a level (or at a particular concentration ratio relative to other specified components) that meets a criterion of being above (or in some cases below) a pre-determined cut-off level or ratio, then a given therapeutic agent is predicted to be effective and is administered to the patient, and if the component(s) are detected at a level or concentration ratio that do not meet the criterion relative to the predetermined cut-off level or ratio, then the therapeutic agent is not administered to the patient. Appropriate cut-off levels may be set using routine practices and as described herein.

To predict the response of the cells, typically NAS or NIS values are computed for a plurality of known responder cells and non-responder cells to the therapeutic agent and these values for the known respondercells and non-responder cells are used to set threshold NAS or NIS values, indicating responsiveness or non-responsiveness to the therapeutic agent (further described in Examples 7 and 10). Thus, the generated and/or predicted response of the cells to treatment can be obtained in c) by comparing the NAS or NIS computed in b) with the threshold NAS or NIS values, indicating responsiveness or non-responsiveness to the therapeutic agent.

To apply the statistical classification algorithm, one or more pieces of information are input into the algorithm or the computing system incorporates one or more pieces of information into the algorithm. For example, this procedure can comprise inputting into the algorithm one or more pieces of information selected from (i) the levels of one or more components of the cellular network (ii) the computed NAS or NIS; (iii) the mutation status of one or more genes in the sample of the cells; (iv) the age of the subject to be treated with the therapeutic agent; (v) the gender of the subject to be treated with the therapeutic agent; (vi) presence or absence of estrogen receptor (ER) on the cells; (vii) presence or absence of progesterone receptor on the cells; and (viii) presence or absence of androgen receptor on the cells. Additionally or alternatively, applying the statistical classification algorithm can comprise the computing system computing the algorithm after inputting one or more of the pieces of information set forth in (i)-(viii) above. Since the statistical classification algorithm defines a statistical relationship between the input information and the responsiveness of the cells (e.g., tumor cells) to treatment, prediction of the response of the cells to treatment can then be based on the output of the statistical classification algorithm.

For the predictive methods and other methods provided herein, a preferred cellular network comprises an ErbB signaling pathway. In one embodiment, the one or more components measured and input into the computing system in the method can comprise one or more ligands involved in the ErbB signaling pathway. Non-limiting examples of such ligands include HRG (including HRG-β1, HRG-β2, HRG-α, HRG-3 and HRG-4), BTC, EGF, HB-EGF, TGFα, AR, EPG and EPR. Additionally or alternatively, the one or more components measured in the method can comprise one or more receptors involved in the ErbB signaling pathway. Non-limiting examples of such receptors include ErbB1, ErbB2, ErbB3 and ErbB4 (also known in the art as HER1, HER2, HER3 and HER4, respectively). Such receptors may be assayed as individual entities (whether they occur in monomers homodimers or heterodimers) and in certain embodiments each homodimer and heterodimer may also be measured as a distinct component and input into the computing system. For example, the components measured may be 1, 2, 3, 4, 5 or 6 or more receptors, ligands, or both, chosen from ErbB1, ErbB2, ErbB3, ErbB4, HRG, and BTC (e.g., ErbB1 and HRG; or ErbB1, ErbB2 and ErbB3). For example, in some methods, the computed NAS simulates the level of ErbB2/ErbB2 homodimer or ErbB2/ErbB3 heterodimer in the absence of the therapeutic agent. In other such methods, the computed NIS simulates the level of ErbB2/ErbB3 heterodimer or ErbB1/ErbB3 heterodimer in the presence of the therapeutic agent as compared to levels of ErbB2/ErbB3 heterodimer or ErbB1/ErbB3 heterodimer in the absence of the therapeutic agent.

In certain of the above methods, the computed NAS simulates levels of one or more phosphorylated proteins in the ErbB3 signaling pathway. For example, the NAS that is computed may simulate pErbB3 levels in the sample of the cells. In alternate embodiments, the NAS that is computed may simulate the level of a phosphorylated ErbB1/ErbB3 heterodimer or a phosphorylated ErbB2/ErbB3 heterodimer in the sample of the cells.

In one embodiment, the therapeutic agent comprises an anti-EGFR (anti-ErbB1) antibody, a representative example of which is the anti-ErbB1 antibody cetuximab (Erbitux®, ImClone Systems). Other examples of anti-ErbB1 antibodies include matuzumab, panitumumab; nimotuzumab and mAb 806 (Mishima, K. et al. (2001) Cancer Res. 61:5349-5354). In another embodiment, the therapeutic agent comprises an anti-ErbB2 antibody, a representative example of which is trastuzumab (Herceptin®, Genentech).

In another embodiment, the therapeutic agent comprises an anti-ErbB3 antibody. In a preferred embodiment, the anti-ErbB3 antibody comprises MM-121, which is currently undergoing Phase I clinical trials. In a preferred embodiment, the anti-ErbB3 antibody comprises Ab #6, described further in WO 2008/100624 and having VH and VL sequences of SEQ ID NOs: 1 and 2, respectively. In another embodiment, the anti-ErbB3 antibody is an antibody comprising the Ab #6VH and VL CDR sequences of SEQ ID NOs: 7-9 (VH CDR1, 2, 3) and 10-12 (VL CDR1, 2, 3), respectively. Other examples of anti-ErbB3 antibodies include Ab #3, Ab #14, Ab #17 and Ab #19, also described further in WO 2008/100624 and having VH and VL sequences of SEQ ID NOs: 3 and 4, 5 and 6, 25 and 26, and 33 and 34, respectively. In another embodiment, the anti-ErbB3 antibody is an antibody comprising the VH and VL CDR sequences of Ab # 3 (SEQ ID NOs: 13-15 and 16-18, respectively) or antibody comprising the VH and VL CDR sequences of Ab # 14 (SEQ ID NOs: 19-21 and 22-24, respectively) or an antibody comprising the VH and VL CDR sequences of Ab # 17 (SEQ ID NOs: 27-29 and 30-32, respectively) or an antibody comprising the VH and VL CDR sequences of Ab # 19 (SEQ ID NOs: 35-37 and 38-40, respectively).

Other examples of anti-ErbB3 antibodies include the antibodies 1B4C3 and 2D1D12 (U3 Pharma AG), both of which are described in US Publication No. 2004/0197332, and the monoclonal antibodies (including humanized versions thereof), such as 8B8, described in U.S. Pat. No. 5,968,511.

In another embodiment, the anti-ErbB3 antibody is a bispecific antibody (e.g., a fusion protein) comprising an anti-ErbB3 antibody linked to a second antibody (e.g., a anti-ErbB2 antibody). A preferred example of such a bispecific antibody is H3×B1D2, the amino acid sequence of which is set forth in SEQ ID NO: 41. In this bispecific antibody, a single chain antibody that binds ErbB3, referred to as H3 (having VH and VL CDRs as shown in SEQ ID NOs: 42-44 and 45-47, respectively) is linked to a single chain antibody that binds ErbB2, referred to as B1D2 (having VH and VL CDRs as shown in SEQ ID NOs: 48-50 and 51-53, respectively). The antibody components of the bispecific antibody H3×B1D2 are described further in U.S. Pat. Nos. 7,332,585 and 7,332,580, as well as PCT Application PCT/US2006/023479 (published as WO 2007/084181) and PCT Application PCT/US2007/024287 (published as WO 2008/140493).

In yet another embodiment, the therapeutic agent comprises two or more anti-ErbB3 antibodies, each of which binds to a different epitope on ErbB3. Preferably, the therapeutic agent comprises three anti-ErbB3 antibodies, each of which binds to a different epitope on ErbB3.

In another embodiment, the therapeutic agent comprises an anti-ErbB4 antibody. In yet another embodiment, the therapeutic agent comprises a pan-ErbB inhibitor or a HER dimerization inhibitor. An example of a HER dimerization inhibitor is the antibody pertuzumab (also known as the 2C4 antibody), which is described further in Agus, D. B. et al. (2005) J. Clin. Oncol. 23:2534-2543.

In yet another embodiment, the therapeutic agent comprises one or more small molecule inhibitors of an ErbB signaling pathway, representative examples of which include gefitinib (Iressa®), which is commercially available from AstraZeneca and Teva, and lapatinib (Tykerb®), commercially available from GlaxoSmithKline. Other examples of small molecule inhibitors of the ErbB signaling pathway include CI-1033 (PD 183805; Pfizer), erlotinib HCL (OSI-774; Tarceva®; OSI Pharma); PKI-166 (Novartis); PD-158780; EKB-569; and Tyrphostin AG 1478 (4-(3-Chloroanillino)-6,7-dimethoxyquinazoline).

In another embodiment, the cellular network comprises a c-Met (mesenchymal epithelial transition factor) signaling pathway. In one embodiment, the one or more components measured and/or input in a) can comprise one or more ligands involved in the c-Met signaling pathway. A non-limiting example of such a ligand is hepatocyte growth factor (HGF). Additionally or alternatively, the one or more components measured in a) can comprise one or more receptors involved in the c-Met signaling pathway. A non-limiting example of such a receptor is the c-Met receptor protein tyrosine kinase.

In view of the above, the predictive methods provided herein allow for prediction of cellular responses, e.g., computer-generated prediction of tumor responses, to therapeutic agents that target components within the c-Met signaling pathway. The therapeutic agent may comprise, for example, an antibody (e.g., a monoclonal antibody) that binds to c-Met. Examples of such anti-cMet antibodies include AV299 (AVEO); AMG102 (Amgen) and 5D5 (OA-5D5; Genentech). A preferred therapeutic agent targeting the c-Met signaling pathway comprises a bispecific monoclonal antibody comprising an anti-ErbB1 antibody linked to an anti-cMet antibody. Examples of such bispecific antibodies are described further in PCT Publications WO 2005/117973 and WO 2006/091209. In another embodiment, the therapeutic agent is a small molecule inhibitor of c-Met signaling, examples of which include ARQ 197 (ArQule) and PHA665752 (Christensen, J. G. et al. (2003) Cancer Res. 63:7345-7355).

In another embodiment, the cellular network comprises an insulin growth factor 1 receptor (IGF1R) signaling pathway. In one embodiment, the one or more components measured in a) can comprise one or more ligands involved in the IGF1R signaling pathway. A non-limiting example of such a ligand is insulin growth factor 1 (IGF1). Additionally or alternatively, the one or more components measured in a) can comprise one or more receptors involved in the IGF1R signaling pathway. A non-limiting example of such a receptor is the IGF1R receptor.

In view of the above, the predictive methods of the invention allow for prediction of cellular responses, e.g., computer-generated prediction of tumor responses, to therapeutic agents that target components within the IGF1R signaling pathway. In a preferred embodiment, the therapeutic agent is an antibody that binds to IGF1R, examples of which include mAb391 (Hailey, J. et al. (2002) Mol. Cancer. Ther. 1:1349-1353); IMC-A12 (Imclone Systems, Inc.), 19D12 (Schering Plough), H7C10 (Goetsch, L. et al. (2005) Int. J. Cancer 113:316-328), CP751,871 (Pfizer), SCV/FC (ImmunoGen, Inc.) and EM/164 (ImmunoGen, Inc.). In a preferred embodiment, the therapeutic agent is a bispecific antibody comprising an anti-IGF1R antibody linked to an anti-ErbB3 antibody. Such bispecific antibodies are described further in PCT Publications WO 2005/117973 and WO 2006/091209. In another embodiment, the therapeutic agent is a small molecule inhibitor of IGF1R (e.g., tyrosine kinase inhibitor), examples of which include NVP-AEW541 (Novartis); NVP-ADW742 (Novartis); NVP-TAE226 (Novartis); BMS-536, 924 (Bristol-Myers Squibb); BMS-554, 417 (Bristol-Myers Squibb); cyclolignans such as picropodophyllin (PPP) (Menu, E. et al. (2006) Blood 107:655-660); and PQ401 (Gable, K. L. et al. (2006) Mol. Cancer. Ther. 5:1079-1086).

In yet another embodiment, the therapeutic agent comprises a combination of therapeutic agents, wherein the combination includes at least one agent that targets a component within the ErbB signaling pathway, such as a combination of agents that includes at least one of the ErbB pathway agents described above. For example, a combination agent can comprise two or more agents that target components within the ErbB signaling pathway. Alternatively, a combination agent can comprise at least one agent that targets a component within the ErbB signaling pathway and at least one agent that targets a component within another signaling pathway, such as a c-Met or IGF1R signaling pathway.

In various other embodiments, the cellular network comprises a combination of two or more signaling pathways, such as an ErbB signaling pathway in combination with a c-Met signaling pathway or an ErbB signaling pathway in combination with an IGF1R signaling pathway.

In another embodiment of the predictive methods provided herein, such methods can further comprise a procedure of determining, in the sample of the cells, the mutation status of one or more genes in the cells. Preferably, the mutation status of at least one gene selected from KRAS (Kirsten rat sarcoma viral oncogene homolog), PI3K and PTEN is determined. Additionally or alternatively, the method can comprise a computer system receiving, through an input device, input that identifies the mutation status of one or more genes in the cells, such as the mutation status of at least one gene selected from KRAS, PI3K and PTEN.

In another embodiment of the predictive methods, the cellular network comprises an ErbB signaling pathway and a) comprises measuring and/or inputting measured levels of BTC and AR and the method further comprises determining the mutation status, or inputting the mutation status, of the KRAS gene. In another embodiment, the cellular network comprises an ErbB signaling pathway and a) comprises measuring and/or inputting measured levels of ErbB1, ErbB2, ErbB3, HRG, BTC, AR, HB-EGF, EGF, TGFα, EPG and EPR, and the method further comprises determining the mutation status, or inputting the mutation status, of the KRAS gene. Preferably for these embodiments, the NAS computed in b) simulates levels of one or more phosphorylated proteins in the ErbB signaling pathway. In another preferred embodiment, the NAS computed in b) simulates levels of phosphorylated ErbB1/ErbB3 heterodimer in the sample of a patient's tumor. In yet another preferred embodiment, the NAS computed in b) simulates levels of phosphorylated ErbB2/ErbB3 heterodimer in the sample of a patient's tumor.

Within certain methods described above, NAS or NIS values computed for each of a plurality of known responder cells and non-responder cells to the therapeutic agent are used to set threshold NAS or NIS values, indicating responsiveness or non-responsiveness to the therapeutic agent. In other methods, the response of the cells to treatment is predicted by comparing the NAS or NIS computed in (b) with the threshold NAS or NIS values, indicating responsiveness or non-responsiveness to the therapeutic agent.

In another aspect, the invention provides direct biomarkers that predict the responsiveness of cells (e.g., tumor cells) to treatment with a therapeutic agent that targets a component of a cellular network (e.g., the ErbB signaling pathway). Such biomarkers may be identified using the computational models provided herein. Methods for identifying such biomarkers generally comprise: (a) measuring, in a sample of the cells, levels of one or more components of the cellular network; and (b) applying a computer-implemented method comprising: (i) computing levels of one or more additional components of the cellular network using a computational model of the cellular network; and (ii) identifying a component of the cellular network whose computed level predicts response of the cells to treatment with a therapeutic agent to thereby identifying the component as a biomarker for predicting the response of the cells to treatment with the therapeutic agent.

Certain methods for identifying a biomarker for predicting the response of cells to treatment with a therapeutic agent that targets a component within a cellular network comprise:

    • a) a computing system receiving, through an input device, input that identifies measured levels of one or more components of a cellular network measured in a sample of the cells;
    • b) the computing system computing levels of one or more additional components of the cellular network using a computational model of the cellular network; and
    • c) the computing system identifying a component of the cellular network whose computed level predicts response of the cells to treatment with a therapeutic agent, and thereby identifying the component as a biomarker for predicting a response of the cells to treatment with the therapeutic agent.

For example, as illustrated in Example 10, a computational model can be used to compute levels of one or more components of a cellular network (e.g., the ErbB signaling pathway) to obtain one or more NAS values (as a measure of activation of the cellular network) and the correlation of the NAS with responsiveness of cells to treatment with a therapeutic agent can be determined. Those components of the cellular network for which the computed levels segregate samples into responders and non-responders can then also be used as direct biomarkers for predicting responsiveness to treatment, in particular when those components are readily measurable by direct means. That is, the computational model/NAS approach described herein can be used to identify (computed) component(s) that predict responsiveness of cells to treatment with a therapeutic agent and then once the component(s) have been identified, they can be directly measured as direct biomarkers for predicting responsiveness. For example, the computational model described herein was used to compute levels of homo- and heterodimers of the ErbB signaling pathway and then those dimers that segregate samples into responders and non-responders can be directly measured as direct biomarkers for predicting responsiveness to tumor treatment.

Further, as described further in Example 8, it has now been demonstrated that the combined measurement of the level of (i) HRG and (ii) at least one ErbB family receptor (e.g., ErbB1, ErbB2 and ErbB3) in a tumor sample effectively stratifies tumors into responders and non-responders with regard to responsiveness to treatment with a therapeutic agent that targets a component of an ErbB signaling pathway, such as an anti-ErbB3 antibody (e.g., Ab #6). Moreover, as described further in Example 10, levels of ErbB1/ErbB3 heterodimer or levels of phosphorylated ErbB1/ErB3 heterodimer can serve as direct markers for responsiveness to treatment with a therapeutic agent that targets a component of an ErbB signaling pathway, such as an anti-ErbB3 antibody (e.g., Ab #6). Moreover, as described further in Example 12, levels of ErbB2 monomer, ErbB2/ErbB2 homodimer and ErbB2/ErbB3 heterodimer effectively stratify tumors into responders and non-responders with regard to responsiveness to treatment with a therapeutic agent that targets a component of an ErbB signaling pathway, such as an anti-ErbB3×anti-ErbB2 bispecific antibody (e.g., H3×B1D2).

As noted above, the present invention provides methods for predicting the response of cells to treatment with a therapeutic agent that targets a component of an ErbB signaling pathway. Certain such methods comprise:

    • (a) measuring, in a sample of the cells, levels of (i) HRG and (ii) at least one receptor selected from ErbB1, ErbB2 and ErbB3; and
    • (b) predicting, using a computer, the response of the cells to treatment with the therapeutic agent based on the levels measured in (a), wherein elevated levels of HRG and the at least one receptor, relative to a control, predict responsiveness to treatment with the therapeutic agent.

In certain situations the levels of HRG and ErbB1 are measured. In other cases, the levels of HRG and ErbB2 are measured, or the levels of HRG and ErbB3 are measured. In other situations, the levels of HRG and at least two receptors selected from ErbB1, ErbB2 and ErbB3 are measured, or the levels of HRG, ErbB1, ErbB2 and ErbB3 are measured. In certain situations, the prediction may be computationally performed, using a method that comprises:

    • (i) a computing system receiving, through an input device, input that identifies measured levels of (i) HRG and (ii) at least one receptor selected from ErbB1, ErbB2 and ErbB3, which levels have been measured in a sample of the cells; and
    • (ii) the computing system generating, and thereafter rendering at an output device, a predicted response of the cells to treatment with the therapeutic agent based on the measured levels, wherein elevated levels of HRG and the at least one receptor, relative to a control, predict responsiveness to treatment with the therapeutic agent.

Preferred therapeutic agents for which responsiveness is predicted include anti-ErbB3 antibodies, more preferably Ab #6 (having VH and VL sequences as shown in SEQ ID NOs: 1 and 2, respectively) or an anti-ErbB3 antibody comprising the VH and VL CDR sequences of Ab #6, which are shown in SEQ ID NOs: 7-9 (VH CDR1, 2, 3) and 10-12 (VL CDR1, 2, 3), respectively.

Other methods for predicting the response of cells to treatment with a therapeutic agent that targets a component of an ErbB signaling pathway comprise:

    • (a) measuring, in a sample of the cells, levels of one or more of ErbB1/ErbB3 heterodimer, ErbB2 monomers, ErbB2 homodimer, ErbB2/ErbB3 heterodimer, phosphorylated ErbB1/ErbB3 heterodimer and phosphorylated ErbB2/ErbB3 heterodimer; and
    • (b) predicting, using a computer, the response of the cells to treatment with the therapeutic agent based on the levels measured in (a), wherein a difference in the level of ErbB1/ErbB3 heterodimer, ErbB2 monomers, ErbB2/ErbB2 homodimer, ErbB2/ErbB3 heterodimer, phosphorylated ErbB1/ErbB3 heterodimer or phosphorylated ErbB2/ErbB3 heterodimer, relative to a control, predicts responsiveness to treatment with the therapeutic agent.

The prediction may be computationally performed, using a method that comprises:

    • (i) a computing system receiving, through an input device, input that identifies measured levels of one or more of ErbB1/ErbB3 heterodimer, ErbB2 monomer, ErbB2/ErbB2 homodimer, ErbB2/ErbB3 heterodimer, phosphorylated ErbB1/ErbB3 heterodimer and phosphorylated ErbB2/ErbB3 heterodimer, which levels have been measured in a sample of the cells; and
    • (ii) the computing system generating, and thereafter rendering at an output device, a predicted response of the cells to treatment with the therapeutic agent based on the measured levels, wherein a difference in the level of ErbB1/ErbB3 heterodimer, ErbB2 monomer, ErbB2/ErbB2 homodimer, ErbB2/ErbB3 heterodimer, phosphorylated ErbB1/ErbB3 heterodimer or phosphorylated ErbB2/ErbB3 heterodimer, relative to a control, predicts responsiveness to treatment with the therapeutic agent.

Within certain embodiments of the above methods, the measured levels are input into a statistical classification algorithm and the response of the cells to treatment is predicted based on output of the algorithm. In further such methods, a Network Activation State (NAS) or a Network Inhibition State (NIS) is computed based on the measured levels using a computational model of an ErbB cellular network; preferably, the response of the cells to treatment with the therapeutic agent is predicted based on the computed NAS or NIS value.

Within certain embodiments, the measured levels are levels of ErbB2 monomer, ErbB2:ErbB2 homodimer and/or ErbB2:ErbB3 heterodimer and the therapeutic agent is a bispecific antibody comprising an anti-ErbB3 antibody linked to an anti-ErbB2 antibody, as described above.

Preferred cells for use in the above methods are tumor cells, including those recited above. Samples include tumor tissue, fine needle aspirate, nipple aspirate, whole blood, serum, plasma, lymph, saliva and urine or shed or circulating tumor cells isolated therefrom.

It will be apparent that, as noted above, the measured levels are protein levels (e.g., monomers, homodimers or heterodimers) or mRNA levels, and may generally be determined as described above.

Any of the above methods may, but need not, further comprise selecting a treatment regimen (e.g., for the therapeutic agent) based on the predicted response of the cells to treatment; and/or may further comprise preparing the therapeutic agent for use based on the predicted response.

In a preferred embodiment, the difference in the level, relative to a control, is an elevated level. The above methods can, but need not, further comprise treating cells, or a subject (patient) from whom the cells are obtained, with a therapeutic agent, based on the predicted responsiveness of the cells to the therapeutic agent.

For embodiments in which levels of ErbB1/ErbB3 heterodimer and/or phosphorylated ErbB1/ErbB3 heterodimer are measured, preferably the therapeutic agent for which responsiveness is predicted is an anti-ErbB3 antibody, more preferably Ab #6 (having VH and VL sequences as shown in SEQ ID NOs: 1 and 2, respectively) or an anti-ErbB3 antibody comprising the VH and VL CDR sequences of Ab #6, which are set forth in SEQ ID NOs: 7-9 (VH CDR1, 2, 3) and 10-12 (VL CDR1, 2, 3), respectively.

For embodiments in which levels of ErbB2 monomer, ErbB2/ErbB2 homodimer and/or ErbB2/ErbB3 heterodimer are measured, preferably the therapeutic agent for which responsiveness is predicted is an anti-ErbB3×anti-ErbB2 bispecific antibody, more preferably bispecific antibody H3×B1D2 (having the amino acid sequence shown in SEQ ID NO: 41) or a bispecific antibody comprising an anti-ErbB3 antibody comprising the CDRs of H3 (SEQ ID NOs: 42-44 and 45-47) linked to an anti-ErbB2 antibody comprising the CDRs of B1D2 (SEQ ID NOs: 48-50 and 51-53).

The levels of receptor homo- or heterodimers or phosphorylated receptor homo- or heterodimer can be measured in the sample of the cells using methods known in the art. For example, such homo- or heterodimers can be measured using dimerization detection methods such as those described in U.S. Pat. No. 7,105,308, U.S. Pat. No. 7,255,999, U.S. Publication No. 20040229380 and U.S. Publication No. 20080187948. These methods employ pairs of probes (e.g., antibodies), one a tagged probe and one a cleaving probe, wherein each probe binds specifically to one component of the dimer. Binding of the two probes to the dimer results in cleavage and release of the molecular tag from the dimer complex, providing a measure of formation of the dimer complex. Such assays are also referred to herein as proximity based methods, a commercially available example of which is the VeraTag™ system (Monogram Biosciences). Alternatively, other methods known in the art for quantitating dimer levels can be used, including but not limited to coimmunoprecipitation of the components within the dimer and use of other proximity based methods such as Forster resonance energy transfer-based methods and biomolecular fluorescence complementation (BiFC) (described further in, for example, Tao, R. H. and Maruyama, I. N. (2008) J. Cell Sci. 121:3207-3217).

In one embodiment of the above direct biomarker methods, the levels measured in a) are input into a statistical classification algorithm stored by a computing system and the response of the cells to treatment is predicted based on output of the algorithm based on calculations and transformation of the data at the computing system with use of the algorithm and measured levels. In another embodiment, a Network Activation State (NAS) or Network Inhibition State (NIS) is computed based on the levels measured in a) using a computational model of an ErbB cellular network. The response of the cells to treatment with the therapeutic agent can be predicted based at least in part on the computed NAS or NIS value.

In various embodiments, the therapeutic agent comprises any combination of one or more of an anti-ErbB3 antibody, an anti-ErbB1 antibody, an anti-ErbB2 antibody, an anti-ErbB4 antibody, a pan-ErbB inhibitor, a HER dimerization inhibitor, and a small molecule inhibitor of an ErbB signaling pathway, each of which is described and exemplified above.

The methods of the invention can be used in predicting the efficacy of treatment for a wide variety of disorders in which therapeutic agents are available that target one or more components of a cellular network involved in the disorder. Still further, the methods of the invention can be used in the selection of a treatment regimen for a subject suffering from the disorder, wherein the methods can further comprise treating the subject according to the selected treatment regimen, which can comprise administering one or more therapeutic agents to the subject. Non-limiting examples of disorders include cancer, autoimmune disorders and inflammatory disorders.

The methods of the invention are particularly useful in predicting, e.g., computationally predicting, the response of a tumor to treatment with a therapeutic agent, i.e., predicting the responsiveness of a patient carrying the tumor to treatment with a therapeutic agent. The predictive methods can be used with any tumor that is dependent upon the signaling pathway that is modeled in the method. For example, in one embodiment, the method is used with tumors that are dependent upon the ErbB signaling pathway (e.g., the ErbB3 signaling pathway). In other embodiments, the method can be used with tumors that are dependent upon the c-Met or IGF1R signaling pathways. In a preferred embodiment, the tumor is a colon cancer tumor. In another preferred embodiment, the tumor is a non-small cell lung cancer (NSCLC) tumor. In another embodiment, the tumor is a solid tumor. In another embodiment, the tumor is a non-solid tumor, such as a clear cell sarcoma. In various other embodiments, the tumor can be, for example, a tumor of a tissue selected from lung, colon, rectum, gall bladder, brain, spinal cord, breast, kidney, pancreas, stomach, liver, bone, skin, spleen, ovary, testis, prostate, head and neck, thyroid and muscle. In yet other embodiments, the tumor is a gastric tumor, a stomach tumor or an oral/pharynx tumor.

To conduct the predictive method, a sample of cells, e.g., cells of the tumor, is obtained from the patient. For example, a preferred sample of the tumor is a sample of tumor tissue. A tumor tissue sample can be obtained by standard methods, such as biopsy of the tumor or surgical resection of the tumor. A fresh, frozen sample of tumor tissue can be used or, alternatively, a formalin-fixed, paraffin-embedded (FFPE) tissue sample is suitable for use as well. Other types of samples from the tumor also may be amenable for use in the methods, wherein the sample contains cells from the tumor and/or cellular components secreted by the tumor. Non-limiting examples of other types of samples of the tumor include fine needle aspirate, nipple aspirate, whole blood, serum, plasma, lymph, saliva and urine urine, or shed or circulating tumor cells isolated therefrom.

In a preferred embodiment, the invention provides for a method of predicting the response of a cancer patient to treatment with a therapeutic agent, wherein the method can be easily and quickly carried out by a diagnostic laboratory to provide rapid information as to the likelihood of a patient's tumor to respond to a particular therapeutic treatment. In this predictive method, a tumor sample, such as a fresh, frozen sample or a FFPE archived tissue sample is obtained from the patient and tumor receptor/ligand levels are measured via semi-quantitative immunohistochemistry (IHC). The receptor(s) and ligand(s) chosen to be measured are based on which cellular network the therapeutic agent targets (e.g., for the ErbB pathway, the following ligands/receptors can be measured: HRG, BTC, ErbB1, ErbB2 and ErbB3). The semi-quantitative IHC measurements are then converted into concentrations using a control slide, which contains cell plugs or xenografts with known receptor and ligand expression levels to compare to the patient sample. In certain situations, the mutations status of one or more genes of interest (e.g., PI3K, PTEN) may be determined in the sample using standard genotyping methods. Next, the data set (ligand and receptor concentrations, gene mutation status if determined) is input into a computational model of the cellular network of interest and a Network Activation State (NAS) is computed. Prediction of responsiveness to the therapeutic agent then can be made based on comparison of the computed NAS value to threshold NAS values for responders and non-responders. Use of a web-based application for entering the protein concentration and mutation data into the computational model, followed by output of the NAS and the predicted response, allows for a diagnostic laboratory to obtain almost instant knowledge of the likelihood of the tumor to respond to treatment with the therapeutic agent.

For any of the predictive methods of the invention described herein, after the response of the cells (e.g., tumor cells) to treatment with a therapeutic agent has been predicted using the method, the method can further comprise selecting a treatment regimen for the subject based on the predicted response of the cells (e.g., tumor cells) to treatment. For example, the methods can further comprise the computing system displaying and manually or automatically recommending and/or selecting a treatment regimen for the subject based on the computationally predicted response(s) of the cells to treatment. Still further, once a treatment regimen has been recommended or selected based on the predicted responsiveness of the cells, the methods of the invention can further comprise treating the subject according to the recommended or selected treatment regimen, which can comprise administering one or more therapeutic agents to the subject.

Also provided herein are kits for predicting the response of cells (e.g., tumor cells) to treatment with a therapeutic agent that targets a component within a cellular network. One such kit comprises: a) an assay or assays for detecting levels of one or more components of the cellular network; b) instructions for computing a Network Activation State (NAS) or Network Inhibition State (NIS) for the cells using a computational model of the cellular network; and c) instructions for use of the kit to predict the response of the cells to treatment with the therapeutic agent. In an additional embodiment, the kit can further comprise instructions for applying a statistical classification algorithm for computing the NAS or NIS.

The cellular network can be, for example, an ErbB signaling pathway, a c-Met signaling pathway or an IGF1R signaling pathway. The therapeutic agent can be, for example, any of the therapeutic agents described above that target components within any of these pathways.

In one embodiment, the means for detecting levels of one or more components of the cellular network is one or more reagents that allow for detection of protein levels of the component(s), such as one or more antibody reagents. In another embodiment, the means for detecting levels of one or more components of the cellular network is one or more reagents that allow for detection of mRNA levels of the component(s), such as one or more nucleic acid reagents (e.g., nucleic acid probes, PCR primers and the like). Such reagents for detection of protein or mRNA levels of cellular components are well known to the ordinarily skilled artisan. Such means for detecting levels can also include computing devices configured to measure protein levels.

Assays suitable for detection of protein levels of cellular components include those described herein, such as quantitative fluorescence activated cell sorting (qFACS), enzyme linked immunosorbent assay (ELISA, Luminex), immunohistochemistry (IHC), quantitative immunohistochemistry (qIHC), mass spectrometry and Western (immunoblot) assay. Assays suitable for detection of mRNA levels of cellular components include, for example, quantitative polymerase chain reaction (qPCR) and Northern blot analysis. The means for detecting levels of one or more components of the cellular network can also include, for example, buffers or other reagents for use in an assay for evaluating the levels of the component(s). The kit can include instructions, (e.g., printed instructions, such as a label or package insert) for performing the assay(s) for detecting the levels of one or more components of the cellular network.

In a preferred embodiment, the cellular network is an ErbB signaling pathway, and the kit includes means for detecting levels of one or more components (e.g., one or more of receptors, receptor homodimers, receptor heterodimers and receptor ligands) of an ErbB signaling pathway selected from ErbB1, ErbB2, ErbB3, ErbB4, HRG (including HRG-β1), BTC, EGF, HB-EGF, TGFα, AR, EPG and EPR. More preferably, the kit includes means for detecting levels of at least one ErbB signaling pathway receptor (e.g., ErbB1, ErbB2, ErbB3, ErbB4) and at least one ErbB signaling pathway ligand (e.g., HRG, BTC, EGF, HB-EGF, TGFα, AR, EPG and EPR). For example, in one embodiment, the kit includes means for detecting levels of ErbB1, ErbB2, ErbB3, HRG and BTC. In another embodiment, the kit includes means for detecting levels of ErbB1 and HRG. In yet another embodiment, the kit includes means for detecting ErbB1, ErbB2 and ErbB3. In still other embodiments, the kit includes means for detecting ErbB2 monomer, ErbB2 homodimer, ErbB2/ErbB3 heterodimer or ErbB1/ErbB3 heterodimer.

The means for computing a Network Activation State (NAS) or a Network Inhibition State (NIS) for the cells using a computational model of the cellular network can be, for example, a computer program product containing executable instructions that when executed cause a processor to perform operations for computing a NAS or NIS for the cells. Alternatively, the means for computing a NAS or NIS can be, for example, a component that allows the user of the kit to interface with an internet-based service that runs a computer program that can compute a NAS or NIS for the cells upon entry by the user of information on the levels of one or more components of the cellular network in the cells. Such a component can include, for example, an interface, a webpage and/or a password to allow access to the internet-based service and instructions, e.g., printed instructions, for use of the service. Computer systems and software established in the art, and described further herein, can be adapted for use in the kits of the invention. Computing devices and the computing components referenced in FIG. 16 can also include means for computing the NAS or NIS, such as computing processors, measuring and input devices, output devices, and so forth.

The instructions for use of the kit to predict the response of the cells to treatment with the therapeutic agent can include computer instructions and computer interfaces, as well as printed publications and manuals. In some instances, the kit is packaged together. In other embodiments, the various components of the kit are maintained in disparate locations. For example, some of the components can be maintained, stored or hosted on one or more remote computing systems and are only made available through a network connection.

Preferably, the kits of the invention are designed for use with a human subject, such as a human patient having a tumor. In such instances, the cells are typically cells obtained from the patient by biopsy or resection of the tumor.

Provided are methods for treating a patient having a malignant tumor. In general, such methods comprise: obtaining a sample (e.g. a biopsy or resection sample) from a tumor from the patient; determining the level of one or more biomarkers in the sample; and administering a therapeutic agent to the patient if the levels of biomarker(s) in the sample match a predetermined profile, e.g., the level of a biomarker is greater than a minimum level. Such methods apply to any solid tumor. Suitable tumor samples are generally as described above. In certain embodiments a biomarker is an ErbB receptor protein.

In certain embodiments, such methods comprise: obtaining a sample of the tumor, assaying the level of pErbB3 in the sample, and subsequently administering at least one anti-neoplastic therapeutic agent to the patient, wherein, if the level of pErbB3 determined in the sample is no lower than a minimum level that is 25%, 50%, 60%, 70%, 80%, 90% or 100% (preferably 50%) of the level of pErbB3 assayed in a culture of ACHN cells (renal cancer cells, ATCC No. CRL-1611) following culture for about 20-24 hours in serum-free medium, then the at least one anti-neoplastic therapeutic agent subsequently administered to the patient comprises an anti-ErbB3 antibody, and if the level of pErbB3 determined in the sample is lower than the minimum level, then the at least one anti-neoplastic therapeutic agent subsequently administered to the patient does not comprise an anti-ErbB3 antibody. Preferred cultures of ACHN cells are those that have been passaged no more than 9 times, e.g., passage 8 ACHN cells. In further aspects of such embodiments, the biomarker is pErbB3, the therapeutic agent is an anti-ErbB3 antibody, and the minimum level is 40%, 50%, 60%, 70%, 80%, 90% or 100% of the level observed in tumor cells from an ACHN xenograft tumor model. In yet further aspects, the minimum level is 0.064 pg/μg total protein, 0.08 pg/μg total protein, 0.096 pg/μg total protein, 0.122 pg/μg total protein, 0.128 pg/μg total protein, 0.144 pg/μg total protein or 0.16 pg/μg total protein.

In the foregoing embodiments, the level of pErbB3 in the sample may, in certain aspects, be determined by: a) measuring levels of at least two components of the ErbB3 signaling pathway in the sample; b) computing a Network Activation State (NAS) that simulates the level of pErbB3 in the sample using the level(s) measured in (a) input into a computational model of the ErbB3 signaling pathway; and (c) determining therefrom the level of pErbB3 in the sample. In certain embodiments, levels of at least three, four, five or six components of the ErbB3 signaling pathway are detected in (a). Suitable components of the ErbB3 signaling pathway include, for example, ErbB1, ErbB2, ErbB3, ErbB4 (and homo- and hetero-dimers of the ErbB proteins), HRG (e.g., HRG-β1), BTC, EGF, HB-EGF, TGFα, AR, EPG and EPR. These components may be assayed as the protein (e.g., a monomer, a homodimer or a heterodimer), or where applicable (e.g., where total level of the protein is measured regardless of phosphorylation state, but not where a phosphoprotein level is measured, e.g., for a monomer, homodimer or heterodimer), as mRNA that encodes the protein. Appropriate assays are well known in the art, and include those described herein.

In further aspects, the method for determining the level of pErbB3 additionally comprises, applying a statistical classification algorithm to generate the computational model of the ErbB3 signaling pathway used in computing the NAS. In further embodiments, the computed NAS simulates levels of a phosphorylated ErbB1/ErbB3 heterodimer and/or a phosphorylated ErbB2/ErbB3 heterodimer in the sample.

Anti-ErbB3 antibodies for use within the present invention include, but are not limited to the anti-ErbB3 antibodies disclosed in International Patent Application No. PCT/US2008/002119, published as International Publication No. WO 2008/100624, which is incorporated herein by reference. A particularly preferred antibody therein disclosed is now known as MM-121, which is now undergoing Phase I clinical trials. Preferred anti-ErbB3 antibodies also include the anti-ErbB3 antibodies described above. Another anti-ErbB3 antibody that may be used in methods disclosed herein is U3-1287 (AMG888) (U3 Pharma AG and Amgen), which is now undergoing Phase I clinical trials.

Tumors amenable to treatment as described herein are generally as described above. Exemplary tumors are of an organ selected from colon, lung, rectum, gall bladder, brain, spinal cord, breast, kidney, pancreas, stomach, liver, bone, skin, spleen, ovary, testis, prostate and muscle. In some aspects, ErbB3 positive tumors or ErbB2 and ErbB3 positive tumors (e.g., breast tumors and non-small cell lung cancer tumors) are preferred.

The anti-neoplastic therapeutic agent may be administered to the patient in any suitable form. Typically, the therapeutic agent is provided in the form of a pharmaceutical composition, which comprises the therapeutic agent in combination with a physiologically acceptable carrier. If desired, other active or inactive ingredients may also be included within the pharmaceutical composition

As used herein, the term “physiologically acceptable” means approved by a regulatory agency of a Federal or a state government (e.g., the U.S. FDA or the EMEA) or listed in the U.S. Pharmacopeia or other generally recognized pharmacopeia for use in animals, and more particularly in humans. The term “carrier” refers to a diluent, adjuvant, excipient, or vehicle with which the anti-neoplastic therapeutic agent is formulated and administered. Physiologically acceptable carriers can be sterile liquids, such as aqueous solutions, which are preferred carriers for intravenous or other parenteral administration. Saline solutions and aqueous dextrose and glycerol solutions are examples of aqueous carriers for injectable solutions. Suitable pharmaceutical excipients include, for example, starch, glucose, lactose, sucrose, gelatin, malt, rice, flour, chalk, silica gel, sodium stearate, glycerol monostearate, talc, sodium chloride, dried skim milk, glycerol, propylene, glycol, water and ethanol. The composition, if desired, can also contain minor amounts of wetting or emulsifying agents, pH buffering agents, or preservatives.

Pharmaceutical compositions may be formulated for any appropriate manner of administration, including, for example, parenteral, intranasal, topical, oral, or local administration, such as by a transdermal means, for prophylactic and/or therapeutic treatment. Examples of suitable pharmaceutical modes of administration and carriers are described in “Remington: The Science and Practice of Pharmacy,” A. R. Gennaro, ed. Lippincott Williams & Wilkins, Philadelphia, Pa. (21st ed., 2005).

Commonly, the pharmaceutical compositions used in the methods provided herein are administered parenterally (e.g., by intravenous, intramuscular, or subcutaneous injection). For parenteral administration, the anti-neoplastic therapeutic agent can either be suspended or dissolved in the carrier. A sterile aqueous carrier is generally preferred, such as water, buffered water, saline or phosphate-buffered saline. In addition, sterile, fixed oils may be employed as a solvent or suspending medium. For this purpose any bland fixed oil may be employed, including synthetic mono- or diglycerides. In addition, fatty acids such as oleic acid find use in the preparation of injectible compositions. Pharmaceutically acceptable auxiliary substances may also be included to approximate physiological conditions, such as pH adjusting and buffering agents, tonicity adjusting agents, dispersing agents, suspending agents, wetting agents, detergents, preservatives, local anesthetics and buffering agents.

In one preferred embodiment, the pharmaceutical composition is formulated for intravenous administration to a patient (e.g., a human). Typically, compositions for intravenous administration are solutions in sterile isotonic aqueous buffer. Where necessary, the composition may also include a solubilizing agent. The ingredients may be supplied either separately or mixed together in unit dosage form, for example, as a dry lyophilized powder or water free concentrate in a hermetically sealed container such as an ampoule. Where the composition is to be administered by infusion, it can be dispensed with an infusion bottle containing sterile pharmaceutical grade water or saline. Where the composition is administered by injection, an ampoule of sterile water for injection or saline can be provided so that the ingredients may be mixed prior to administration.

Pharmaceutical compositions may be sterilized by conventional sterilization techniques, or may be sterile filtered. Sterile aqueous solutions may be packaged for use as is, or lyophilized, the lyophilized preparation being combined with a sterile aqueous carrier prior to administration. The pH of an aqueous pharmaceutical composition typically will be between 3 and 11, more preferably between 5 and 9 or between 6 and 8, and most preferably between 7 and 8, such as 7 to 7.5.

The therapeutic agent is generally present within a pharmaceutical composition at a concentration such that administration of a single dose to a patient delivers a therapeutically effective amount. A therapeutically effective amount is an amount that results in a discernible patient benefit, such as a retardation or cessation of tumor growth or preferably a reduction of tumor size. Therapeutically effective amounts are affected by a variety of factors, including the activity of the anti-neoplastic therapeutic agent employed; the age, body weight, general health, sex and diet of the patient; the time and route of administration; the rate of excretion; any simultaneous treatment, such as a drug combination; and the type and severity of the tissue damage in the patient undergoing treatment. Optimal dosages may be established using routine testing, and procedures that are well known in the art. In general, compositions providing dosage levels ranging from about 1 mg to about 100 mg per kilogram of body weight per day, per week or once every 2 weeks are preferred. Non-limiting examples of suitable dosage ranges and regimens include 2-50 mg/kg (body weight of the subject) administered once a week, or twice a week or once every three days, once every two weeks, or once every three weeks, and 1-100 mg/kg administered once a week, or twice a week or once every three days, or once every two weeks. In various embodiments, a therapeutic agent is administered at a dosage of 3.2 mg/kg, 6 mg/kg, 10 mg/kg, 15 mg/kg, 20 mg/kg, 25 mg/kg, 30 mg/kg, 35 mg/kg or 40 mg/kg at a timing of once a week, or twice a week or once every three days, once every two weeks, or once every three weeks. Additional dosage ranges include: 1-1000 mg/kg, 1-500 mg/kg, 1-400 mg/kg, 1-300 mg/kg and 1-200 mg/kg. Suitable dosage schedules include once every three days, once every five days, once every seven days (i.e., once a week), once every 10 days, once every 14 days (i.e., once every two weeks), once every 21 days (i.e., once every three weeks), once every 28 days (i.e., once every four weeks) and once a month.

Preferably, the therapeutic agent (e.g., anti ErbB3 antibody) is administered to the patient in accordance with the directions in the prescribing information provided by the manufacturer or distributor of the therapeutic agent.

Kits for use in treating patients having a malignant tumor are also provided. Certain such kits typically comprise: a) at least one assay for detecting the level of at least one component of the ErbB3 signaling pathway in a sample; and b) instructions for computing a Network Activation State (NAS) that simulates the level of pErbB3 using a computational model of the ErbB3 signaling pathway input with data obtained from the at least one assay. In certain embodiments, such kits further comprise instructions for applying a statistical classification algorithm. In certain embodiments the kit also comprises an anti-ErbB3 antibody. The assay typically comprises one or more reagents that allow for detection of at least one protein component or at least one mRNA component. In certain embodiments, the instructions for computing a NAS comprise directing the use of a computer program product containing executable instructions that when executed by a computer cause a processor to perform operations for computing a NAS; in such embodiments, the user may be instructed to run the computer program on a local computer or the user may be instructed to interface with an internet-based service that runs the computer program remotely.

Within further embodiments, kits are provided comprising anti-ErbB3 antibodies and instructions (e.g., in the form of labeling, e.g., a package insert) indicating that the anti-ErbB3 antibody is to be administered to the patient if the level of pErbB3 in a tumor biopsy from the patient exceeds a specified minimum value, and that the anti-ErbB3 antibody is not to be administered to the patient if the level of pErbB3 does not exceed the specified minimum value. For example, such instructions may indicate that the anti-ErbB3 antibody is to be administered to a patient having a malignant tumor if the level of phosphorylated ErbB3 determined in the sample is no lower than 50% of a level of phosphorylated ErbB3 measured in a culture of ACHN renal cancer cells (ATCC No. CRL-1611) following culture for about 20-24 hours in serum-free medium; and that an anti-ErbB3 antibody is not to be administered to a patient having a malignant tumor if the level of phosphorylated ErbB3 determined in the sample is lower than 50% of the level of phosphorylated ErbB3 measured in the culture of ACHN renal cancer cells.

As mentioned above, and as should be readily apparent from the disclosure provided in this paper, many of the embodiments of the invention utilize one or more computing systems to perform the various processes described above, including, but not limited to predicting patient responses, generating recommended treatments, identifying biomarkers, computing NAS or NIS values, obtaining computational models of signal pathways, and so forth.

FIGS. 15A and 15B illustrate flowcharts of some of the processes that can be performed by one or more computing systems during implementation of certain embodiments of the invention. As shown, for example, computing systems can be utilized to measure and/or receive input of measured levels of components in a cellular network of a tumor as well as the mutation status of the tumor genes. Computing systems can also be used to obtain computational models, which can be obtained, for example, by receiving, downloading, building, modifying, training and/or accessing computational models from local and remote sources.

Once a computational model is obtained, one or more computing systems compute the NAS or the NIS for the cells, such as, for example, by simulating the relevant levels of phosphorylated proteins, homodimers and/or heterodimers in the cells. The computational model can also be used by the computing system to identify predictive biomarkers through the identification of additional relevant components in the cellular network based on user settings received at the computing system.

The computing system can also be used to identify statistical classification algorithms that can be received, constructed, and/or modified by the computing system as part of the identification process and that can be used in various combinations with the NAS or NIS data, other biomarker data, and patient data, to generate predicted patient responses to treatments, and/or to generate and select recommended treatments.

While the elements illustrated in the flowcharts of FIGS. 15A and 15B infer a suggested sequence or ordering for performing the computing processes of the invention, it will be appreciated that the processes illustrated in FIGS. 15A and 15B can also be performed in sequences having different ordering. For example, a recommended treatment may be identified prior to generating the computational model or computing the NAS or NIS and which may be used in the construction of the computational model. Similarly, the measurement of protein levels can occur prior to or subsequent to the generation of the computational model.

It will also be appreciated that additional processes can be performed as part of the invention, such that the invention is not limited to only methods that include the processes illustrated in the flowchart. For example, the invention can also include processes for obtaining patient information (e.g., age, gender, medical history), and for tracking actual results of recommended treatments, as well as other processes.

It should also be appreciated that the computing systems used to implement the processes of the present invention can include one or more different computing systems, of different types, as well as in different locations.

FIG. 16 illustrates one example of a computing system that can be used to perform certain aspects of the invention (including, for example, at least some of the processes illustrated in FIGS. 15A and 15B, as well as those described throughout this paper). As illustrated, the computing system includes various input devices, output devices, computing modules, processing components and storage media. The input devices can include keyboards, mouse devices, touch pads, touch screens, microphones, as well as any other input devices. The output devices can include speakers, display screens, printing devices, switches, as well as other output devices. The computing modules include the various modules necessary to perform the functionality described in this paper, including modules for receiving and building computational models, modules for computing the NAS or NIS, modules for identifying predictive biomarkers, modules for obtaining, recognizing and storing measured levels of components and for identifying and determining mutation gene status, modules for identifying and applying statistical classification algorithms, modules for generating predicted responses to treatments, modules for generating and selecting recommended treatments, communications modules for interfacing with users and one or more other devices, as well as modules for performing the various other processes described herein.

The computing system also includes one or more processors and other processing components necessary to execute the foregoing modules, as well as storage media for storing the foregoing modules, as well as the various data structures, computational models, and other data described herein. While the storage media is illustrated as being local to the computing system, it will be appreciated that the storage media can also be located remotely from the computing system, or only partially local to the computing system. For example, in some instances, the storage media represents a distributed storage that is shared among a plurality of different computing systems and which includes storage space located in a plurality of computing systems. The storage media can also comprise any combination of persistent and volatile memory.

FIG. 16 also illustrates that the computing system is network connectable to one or more other devices, including measuring devices, remotely located computers, remote services and remote networks. In some instances, one or more of these other devices perform one or more of the processes described in this paper, such that execution of some methods is performed in a distributed network environment involving multiple distributed computing systems and devices.

In view of the foregoing, it will be appreciated that the scope of the present invention can be implemented in various different computing configurations.

The present invention is further illustrated by the following non-limiting examples. The disclosure of each and every US, International, or other patent or patent application or publication referred to herein is hereby incorporated herein by reference in its entirety.

In this example, four xenograft tumor models were used to identify tumor cell lines that responded to treatment with the anti-ErbB3 antibody Ab #6. The four xenograft tumor models studied represent different indications: MALME3M (melanoma cancer line; ATCC No. HTB-64), ADRr (ovarian cancer cell line; NCI-60, cosmic sample ID No. 905987), ACHN (renal cancer cell line; ATCC No. CRL-1611) and DU145 (prostate cancer cell line; ATCC No. HTB-81). As described in further detail below, the MALME3M and ADRr xenografts did not show a response to treatment with Ab #6, whereas the ACHN and DU145 xenografts did show a response to Ab #6 treatment.

In the xenograft tumor models, mice (nu/nu mice: 3-4 week old female mice, T-cell deficient; outbred; Albino background; from Charles River Labs, Wilmington, Mass.) are implanted in the right flank with 3.5×106-3×108 cells/mouse (depending on cell line) in 200 μl via subcutaneous injection. Mice are monitored for initial tumor growth. Tumor cells are allowed to grow for several days until the tumor volume is approximately 200 mm3. The tumor volume is calculated as V=(π/6 (L×W2). The mice are treated with the Ab #6 antibody at a dosage of 600 μg/injection every 3 days (qd3). Control mice are treated with phosphate buffered saline (PBS).

Tumor volume is measured for 60-80 days. Results (obtained using the methods described above or minor variations thereof) of the effect of antibody treatment on tumor growth are summarized in the graphs shown in FIGS. 1A-1D, which demonstrate that Ab #6 treatment inhibited tumor growth in the DU145 and ACHN xenograft models, whereas Ab #6 treatment did not inhibit tumor growth (as compared to the PBS control) in the ADRr and MALME3M xenograft models.

As a measure of responsiveness of the tumor to Ab #6, the exponential growth rate is determined, which described the experimental data best. The following formula is used to describe the exponential growth.


V=Vo*exp(k*t)

wherein V is the tumor volume in mm3, Vo is the tumor volume at time zero, k is the exponential growth rate and t is the time in days.

In order to compare the growth reduction across different xenograft studies, a Growth Rate Reduction (GRR) value is calculated for each cell line tested, which relates the observed Growth Rate in the presence of Ab #6 to the growth rate observed in the PBS control group using the following formula:

Growth Rate Reduction=1−(Ab #6 Growth Rate kAb #6)/(PBS Growth Rate kPBS)

The GRR values for the four cell lines tested (obtained using the methods described above or minor variations thereof) are summarized in Table 1 below. In the case of a negative Growth Rate Reduction, the GRR value is set to zero.

TABLE 1
Summary of Tumor Growth Rate Reduction
for Training Set of Xenograft Studies
Cell LineGRR [%]
MALME3M0
DU14572.6
ADRr0
ACHN28.0

The results demonstrate that the four xenografts show a range of responsiveness to Ab #6 treatment, with ADRr and MALME3M cells showing no response to Ab #6 treatment, ACHN having a mid-range responsiveness and DU145 cells having the highest responsiveness to Ab #6 treatment.

In this example, the concentration of phosphorylated ErbB3 (pErbB3) was measured in vivo in each of the four tumor cell lines studied in Example 1, MALME3M, ADRr, DU145 and ACHN, in a short term pharmacodynamic (PD) study. The OvCAR8 xenograft also was included in this experiment (this xenograft is shown to be responsive to Ab #6 treatment in Example 5 described below).

MALME3M, ADRr, DU145, OvCAR 8 and ACHN cells are grown in culture and harvested for implantation (15×15 cm plates, ˜80% confluency, total # of cells=2-4×108) and kept on ice until implantation. Cells (approximately 2×107 cells/mouse) are implanted into 20 mice (via subcutaneous injection, 200 μl cells/injection/mouse) into the right flank and then the mice are allowed to recover while being monitored for initial tumor growth. Tumors are measured (L×W) by digital caliper measurement. Once the mice reached a tumor volume greater than 100 mm3, they are euthanized by CO2 asphyxiation and tumors from each mouse are excised and snap frozen in liquid nitrogen. Frozen tumor tissue samples are stored at −80° C. for biochemical analysis. The amount of phosphorylated ErbB3 (pErbB3) in the tumor lysates is determined by ELISA using the R&D Systems Human pErbB3 ELISA kit (Catalog #DYC1769). The sample preparation and ELISA protocols are described in further detail below.

For sample preparation and protein extraction, first the frozen tumors are pulverized and transferred to pre-weighed 2 ml VWR cryotubes (VWR International). The pulverized samples are weighed and the weights recorded. After calculating the sample weight, the appropriate amount of ice-cold lysis buffer is added to each tube to a final concentration of 62 mg/ml. The samples are briefly vortexed at low speed and incubated at 4° C. with rotation.

The crude tumor lysate is then transferred to Qiagen Qiashredder and centrifuged at 12000 rpm for 8 minutes for further homogenization of the samples. After transfer of the cleared lysates into a fresh tube, a small amount of each lysate is taken out for BCA protein assay. The rest of the lysate is aliquoted and stored at −80° C. for further ELISA assay analysis.

To quantitate total protein using a BCA Protein Assay Kit (Pierce, Catalog #23225), first a bovine serum albumin (BSA) 8 point standard curve is prepared using the 2 mg/ml BSA standard solution from the BCA kit, starting with the stock concentration of 2 mg/ml. After mixing reagents A and B from the kit (50:1) and preparing 3-fold and 5-fold dilutions of stock tumor lysate with PBS, 20 μl of BSA standard or diluted tumor lysate sample and 160 μl of working reagent are added to each well of a 96 well plate. The plate is incubated at 37° Celsius for 20 minutes. The OD562 is read and the total amount of protein in the tumor lysates is calculated using the BSA standard curve.

To carry out the pErbB3 ELISA, different capture antibodies are diluted with PBS to the working concentration recommended by the kit (R&D Systems DYC1769). After coating black 96-well plates (Nunc Maxisorb) with diluted capture antibodies, all plates are incubated at room temperature (RT) overnight. The plates are then washed 3 times with PBST (PBS+0.05% Tween-20) on a Bio Tek plate washer and blocked for 2 hours at RT with 200 μl of 1% BSA in PBST.

Recombinant proteins for standard curves are prepared with the highest concentration recommended by the kit and 2-fold dilutions for a total of 11 points. The plates are washed 3 times with PBS and 100 μl of tumor lysates is added before incubating for 2 hour at RT. Then, plates are washed 3 times with PBST and 100 μl of primary detection antibody, diluted to the working concentration in PBS/0.1% BSA/0.05% Tween-20, is added. The plates are further incubated at RT for 2 hours. Finally, 100 μl of the mixed SuperSignal ELISA Pico Chemiluminescent Substrate (Pierce, Catalog #37069) is added to each well before reading the plates.

FIG. 2 is a graph plotting the concentration of pErbB3 in the untreated xenograft tumors (in pg/μg tumor lysate) against the Growth Rate Reduction (%—obtained using the methods described above or minor variations thereof) observed for the xenografts when treated with Ab #6. FIG. 2 demonstrates that there is a good correlation between the tumor Growth Rate Reduction and the constitutive pErbB3 levels measured in the short term pharmacodynamic studies. MALME3M and ADRr xenografts, which did not respond to Ab #6 treatment, showed the lowest levels of pErbB3, whereas ACHN, OvCAR8 and DU145 xenografts, which responded to Ab #6 treatment, had significantly higher levels of pErbB3. The results demonstrate that the pErbB3 levels in tumor cells correlate well with the responsiveness of the tumor cells to anti-ErbB3 antibody treatment.

In this example, the stability of pErbB3 and pAKT was assessed, as well as the expression levels of ErbB1, ErbB2 and ErbB3, in tumor lysates as a function of time after resurrection from freezing the tumor.

Untreated ACHN and EKVX xenograft mice are euthanized by CO2 asphyxiation and tumors are dissected and cut into 4 pieces and put into liquid nitrogen at different time points: 0 min, 10 min, 30 min, and 60 min. Then, pErbB3 and pAKT levels, as well as ErbB1-3 levels, are measured in each of the samples after thawing. The results obtained using the methods described above or minor variations thereof are summarized in the bar graphs shown in FIGS. 3A-3E, with FIGS. 3A and 3B showing the levels of pErbB3 and pAKT, respectively, in the ACHN lysates and FIGS. 3C, 3D and 3E showing the levels of ErbB1, ErbB2 and ErbB3, respectively, in the EKVX lysates.

As shown in FIGS. 3A and 3B, in the 10 minute samples, there already was a measurable decrease in pErbB3 and pAKT levels as compared to the 0 minute samples. In the 30 minute samples, a decrease in concentration of 40% was observed for pErbB3 as compared to control (immediate snap freezing of the tumor, 0 minute sample) and a decrease in concentration of 20% was observed for pAKT as compared to control. In contrast, in EKVX and ACHN tumor cell lysates, the total levels of ErbB1-3 remained constant and appeared unaffected by the time to freeze (see FIGS. 3C-3E). Thus, the observed instability of phosphoproteins in tumor samples and the observed stability of total protein measurements demonstrate the advantage of computing the phosphorylation level of ErbB3, rather than directly measuring the level in a tumor cell lysate. This computed level of pErbB3 is referred to as the Network Activation State (NAS) in the following examples using a mechanistic computational model constructed as described below in Example 4.

In the following example, the building of mechanistic computational biochemical models of signal transduction pathways is described. Based on literature knowledge about ErbB signaling, a mechanistic computational model was developed comprising all the protein-protein interactions describing ligand binding to the receptor, dimerization, receptor internalization and degradation as well as the binding of the adapter molecule Gab1 leading to the activation of the PI3K cascade. A cartoon of the ErbB signaling network implemented is depicted in FIG. 4A. The computational model is a set of non-linear ordinary differential equations (ODEs) using mass action kinetics. FIG. 4B shows a set of biochemical reactions from the signaling pathway and FIG. 4C shows a set of fluxes. The biochemical reactions and fluxes are translated into a set of nonlinear ODEs, illustrated in FIG. 4D. In general, the state of change of a protein concentration ci is equal to the rate of the production of the protein vproduction minus the rate of consumption vconsumption of the protein as represented in Equation 1.

ci t = v production - v consumption ( Eqn . 1 )

The computational model used to predict responses to Ab #6 in the following examples consists of the mammalian ErbB network that includes all four receptors (ErbB1-4) and the Akt signal transduction cascade.

ErbB receptors are single-pass Type I transmembrane receptors with extracellular ligand binding domains, an intracellular tyrosine kinase domain and a cytoplasmic tail that acts as a signaling scaffold. ErbB1 and ErbB4 are fully functional in ligand binding and tyrosine kinase activity but ErbB2 does not bind any known ligand, functioning instead as a dimerization-ready signal amplifier (Klapper, L. N. et al. (1999) Proc. Natl. Acad. Sci. USA 96:4995-5000). ErbB3 has a crippled kinase domain (Guy, P. M. et al. (1994) Proc. Natl. Acad. Sci. USA 91:8132-8136) and therefore lacks catalytic activity, instead transducing signals when phosphorylated by other ErbB receptors. Of the 13 known ErbB ligands, BTC and HRG have been implemented as the ligands that induced the highest ErbB3 phosphorylation levels. The 13 known ErbB ligands can be divided into three groups: (i) those that bind specifically to ErbB1, such as EGF, transforming growth factor alpha (TGFα), and amphiregulin (AR), (ii) those that exhibit dual specificity, binding to both ErbB1 and ErbB4, including BTC, HB-EGF, EPG and EPR, and (iii) the neuregulins (NRGs), which fall into two subgroups: NRG1 (also known as GGF2, SMDF or HRG) which binds ErbB3/ErbB4, a property also shared with NRG2, and NRG3/NRG4 which bind ErbB4 alone. Following ligand binding, receptors dimerize and undergo transphosphorylation on residues in their cytoplasmic tails, thereby creating docking sites for SH2-containing adapter molecules such as Shc, Grb2, GAP, Sos, and PI3K. ErbB1 has at least 20 sites of tyrosine phosphorylation on its cytoplasmic tail, 12 of which have been proposed to partner with SH2-containing adapter proteins and enzymes (Schulze, W. X. et al. (2005) Mol. Syst. Biol. 1:2005-2008). Other ErbB receptors undergo equally complex post-translational modification. Receptor-associated adapters such as Grb2 and PI3K activate RAS, and ultimately turn on ERK and AKT. AKT can also be activated in a RAS-independent manner via the direct binding of PI3K-p85 to multiple sites on ErbB3.

Although a number of computational models have been published (see e.g., Kholodenko, B. N. et al. (1999) J. Biol. Chem. 274:30169-30181; Hatakeyama, M. et al. (2003) Biochem. J. 373:451-463; Resat, H. et al. (2003) Biophys. J. 85:730-743; Hendriks, B. S. et al. (2005) J. Biol. Chem. 280:6157-6169); Sasagawa, S. et al. (2005) Nat. Cell. Biol. 7:365-373; Birtwistle, M. R. et al. (2007) Molecular Systems Biology 3:144), the computational model used herein is more extensive and includes all four ErbB receptors and two distinct classes of ligands (HRG and BTC) while nonetheless retaining the rigor of a mass-action formulation based on elementary reactions.

Seven ErbB hetero- and homo-dimers that have been described in the literature were implemented in the model: ErbB1/1, ErbB1/2, ErbB1/3, ErbB1/4, ErbB2/2, ErbB2/3 and ErbB2/4. The majority of these dimers are activated by ligand binding but several arise through a process of “lateral signaling” (or secondary dimerization) in which dimers phosphorylated in a ligand-dependent manner dissociate into monomers that then homo- or hetero-oligomerize with either activated or unactivated monomers to create active dimers. The computational model was trained with a set experimental data that allowed for the identification of the dimers that form in the presence of HRG or BTC using the ADRr cell line.

The computational model is based on non-linear ordinary differential equations, which require two types of parameters that must be measured or estimated: initial species number and rate constants. Prior to model calibration, the values of as many parameters as possible were specified based on literature information, e.g., the binding constants for ligands to their cognate receptors. Using qFACS analysis, the expression levels of the ErbB receptors was quantified across all the cell lines used in this application. Furthermore, ELISAs were used to quantitate levels of BTC and pErbB3. Moreover, mRNA levels of HRG-β1 were determined as compared to mRNA levels in the ZR-75 cell line (ATCC No. CRL-1500). The receptor and ligand expression levels, which information is used in the computational model, are summarized in Table 2 below. The methodologies for obtaining these expression levels are described in further detail below.

Tumor cell lines are obtained from the National Cancer Institute. All cell lines are grown as monolayer cultures in a humidified atmosphere of 5% CO2, 95% air and 37 degrees Celsius in complete media: RPMI-1640 media (Gibco) supplemented with 10% fetal calf serum (FCS) (Hyclone), 2 mM L-glutamine (Gibco) and units/mL Pen-Strep (Gibco).

The receptor expression levels are quantified using the Quantum Simply Cellular Kit 816A (Bangs Laboratories), which allows for quantification of receptor expression levels by qFACS. It contains a series of 4 microsphere populations labeled with varying amounts of goat anti-human IgG plus a blank population. The IgG conjugated to the beads' surface is specific for the Fc portions of the IgG antibodies. The beads are stained just like the cell samples, and with the same antibody. Each of the different populations of microspheres binds a known amount of the labeled monoclonal antibody. By plotting each population's fluorescence intensity versus its assigned Antibody Binding Capacity (ABC) value, a standard ABC curve is generated and the ABC of stained cell samples is easily determined by using the software provided by Bangs Laboratories (QuickCal v 2.3). This program takes into consideration the make of the instrument used, the voltage for that sample and the fluorochrome used.

The BTC expression levels are measured by ELISA using the R&D Systems Dy261 DuoSet-IC human Betacellulin Kit. A 384 well plate is coated with 4 μg/ml capture antibody. The 384 well plate is blocked by adding 50 μl of 2% BSA/1×PBS (no-Tween-20) for 1 hour and a recombinant standard curve is prepared. After washing the plates, 16 μl of cell lysates is added, as well as an Anti-Phospho-Tyrosine-HRP (horse radish peroxidase) detection antibody, followed by incubation for 2 hours. Finally 20 μl of Pico luminescent substrate is added and the plates are read spectrophotemetrically.

As the commercially available assays to measure HRG-β1 protein expression levels were not sufficiently sensitive to obtain reliable data for the cell lines examined, the HRG-β1 mRNA levels are quantitated for the cell lines of interest. RNA is isolated from cell lysates using RNeasy Mini protocol (74104 RNEASY KIT from QUIAGEN). After converting isolated RNA to cDNA and making a master mix for quantitative PCR (QPCR) using the Applied Biosystems 430443 7 Tagman Master mix, QPCR is run and quantified the HRG-β1 mRNA expression relative to the mRNA levels in ZR75-1 cells. Primers for QPCR are purchased from Applied Biosystems.

The ErbB3 phosphorylation levels for the different cell lines shown in Table 2 are determined using the pErbB3 ELISA kit from R&D (Dyc1769-2 DuoSet-IC human phospho-ErbB3), using the methods described above in Example 2 or minor variations thereof.

Table 2 below summarizes the receptor and ligand expression level information obtained for the different cell lines using the methods described above or minor variations thereof, which information was used in the construction of the computational model of the ErbB signaling pathway. Column 1 shows the name of the cell line; column 2 shows the type of tumor; columns 3-5 show the number of receptors per cell for ErbB1, ErbB2 and ErbB3, respectively; column 6 shows the HRG-β1 mRNA levels, expressed as the fold compared to the mRNA levels in ZR-75 cells; and columns 7 and 8 show the amount of BTC and pErbB3 present in the cells, expressed as pg/cell.

TABLE 2
Summary of Receptor and Ligand Expression Levels for the Cell Lines Used
in Xenograft Experiments
HRG b-1
fold
ErbB1ErbB2ErbB3compared toBTCpErbB3
cell linetumor type#/cell#/cell#/cellZR-75pg/cellpg/cell
ACHNrenal448283.545455.75152002544.7943063.34191E−054.15E−
ADRrmammary177817.54079233204.52261.14529E−081.80E−
DU145prostate437841.2569068194222.2191389449.87522E−061.50E−
IGROV1ovarian149031.25158417.55355.2502.61392E−06ND
MALME3Mmelanoma2914.256422673670.4383028616.71181E−087.90E−
OVCAR8ovarian236156.7553272.2531812.75343.3017338 5.8861E−07ND
SKOV3ovarian2641321377660.513693.53401.63191E−05ND
indicates data missing or illegible when filed

For training of the computational model, a data set was used that comprises dose-time matrices in which phosphorylation of ErbB1, ErbB2, ErbB3 and AKT at multiple time points and at nine different concentrations of BTC or HRG stimulation in ADRr cells are measured by ELISA. For stimulation of the cells, the cells are seeded in 100 complete media at 35,000 cells per well in 96 well tissue culture plates and incubated overnight in a humidified atmosphere of 5% CO2, 95% air and 37 degrees Celsius. Cells are then switched to serum free media: RPMI-1640 media (Gibco) supplemented with, 2 mM L-glutamine (Gibco) and units/mL Pen-Strep (Gibco). Starved cells are incubated in a humidified atmosphere of 5% CO2, 95% air and 37 degrees Celsius for 20-24 hours prior to stimulation. For dose-time matrix studies, cells are stimulated with ligand (BTC or HRG) at 0, 1, 2, 3, 4, 5, 7, 10, 20, 30, 60 and 120 minutes. Following stimulation with 9 different concentrations of HRG (0.038 nM-250 nM) and BTC (0-700 nM) for each time course, cells are placed on ice, washed with cold PBS, then lysed in 30 μl cold M-PER Mammalian Protein Extraction Buffer (Thermo Scientific, Catalog #78501) supplemented with protease inhibitor cocktail (Sigma-Aldrich, P2714), 1 mM sodium orthovanadate (Sigma-Aldrich, S6508), 5 mM sodium pyrophosphate (Sigma-Aldrich, 221368), 50 μM oxophenylarsine (EMD Biosciences, 521000) and 10 μM bpV(phen) (EMD Biosciences, 203695).

Levels of protein phosphorylation in the stimulated cells are measured by ELISA. Capture antibodies against ErbB1 (R&D Systems, AF231), ErbB2 (R&D Systems, MAB1129), ErbB3 (R&D Systems, MAB3481) and AKT (Upstate, 05-591MG) are incubated in 384 well black flat-bottom polystyrene high-binding plates (Corning, Catalog #3708) overnight at room temperature. The ELISA plates are blocked with 2% bovine serum albumin (BSA) and phosphate buffered saline (PBS) for one hour then incubated with lysates diluted in 2% BSA, 0.1% Tween-20 and PBS for two hours at room temp. In between each incubation, the plates are washed three times with 0.05% Tween-20 in PBS. ELISAs for measuring phospho-ErbB1, -ErbB2 and -ErbB3 are incubated with phospho-tyrosine horseradish peroxidase (HRP) linked monoclonal antibody (R&D Systems, HAM1676) for two hours. ELISAs measuring phospho-AKT are incubated with primary serine 473 specific anti-phospho AKT mouse monoclonal antibody (Cell Signaling Technologies, Catalog #5102) for 2 hours, then incubated with Streptavidin-HRP (R&D Systems, Catalog #DY998,) for 30 minutes. All ELISAs are visualized with SuperSignal ELISA Pico Chemiluminescent Substrate (Pierce, Catalog #37069) and luminescent signal is measured using a luminometer.

Results (obtained using the methods described above or minor variations thereof) for the data set of the protein phosphorylation at multiple time points and nine different concentrations of BTC or HRG are shown in FIGS. 5A-5B, wherein FIG. 5A shows the levels of phopho-ErbB3, phospho-ErbB2, phospho-ErbB3 and phospho-AKT for HRG-stimulated cells and FIG. 5B shows the levels of phopho-ErbB3, phospho-ErbB2, phospho-ErbB3 and phospho-AKT for BTC-stimulated cells. This data set was used to calibrate the computational model of the ErbB signaling pathway so that the simulation results (lines) describe the experimental data (dots) shown in FIGS. 5A-5B.

Further information on the development of a computational model of the ErbB receptor signaling network is provided below:

Model Structure

The ErbB receptor signaling network model consists of three receptors (ErbB1, ErbB2, ErbB3), a receptor phosphatase, phosphatidylinositol 3-kinase (PI3K, which binds the receptors) and components of the PI3K-AKT cascade (phosphatidylinositol bisphosphate, PIP2, phosphoinositide-dependent protein kinase, PDK1, PTEN deleted from chromosome 10, PTEN, serine, threonine protein kinase also known as protein kinase B, AKT, Protein phosphatase 2A, PP2A) (see Table 8 below). Two ErbB receptor ligands were included: heregulin (HRG1-β), which binds to ErbB3 and ErbB4 (not included in model because experimentally ErbB4 expression levels were very difficult to detect in cell lines (see Table 3 below) and Betacellulin (BTC), which binds primarily to ErbB1 (Beerli, R. R. and Hynes, N. E. (1996) J. Biol. Chem. 271:6071-6076; Jones, J. T. et al. (1999) FEBS Lett. 447:227-231). The mass-action kinetic reactions, listed in full in Table 11 below, were converted to ordinary differential equations using Matlab Simbiology 2.3 (Mathworks, Mass.).

Included in the model were the ligand-induced dimerization, internalization, recycling, and degradation as described in the literature for all homo- and heterodimers (Hendriks, B. S. et al. (2003) J. Biol. Chem. 278:23343-23351; Wang, Z. et al. (1999) Mol. Biol. Cell 10:1621-1636). Receptor dimer stabilities, using the relative scale as published in Shankaran, H. et al. (2008) Biochem. Biophys. Res. Commun. 371:220-224, were implemented, where the coexpression of ErbB1 with ErbB2 or ErbB3 biases signaling to the cell surface and retards signal downregulation and where the simultaneous coexpression of ErbB1-3 leads to an abundance of ErbB2-ErbB3 heterodimers. Therefore, ErbB3-containing dimers internalize and degrade slower than do ErbB1 homodimers or ErbB1-ErbB2 heterodimers. Constitutive dimerization was also included, although ligand-free dimers were assumed not to trigger any downstream signaling (Yu, X. et al. (2002) Mol. Biol. Cell 13:2547-2557) and to remain at the cell surface. ErbB ligands bind with different affinities to ErbB homo- and heterodimers (Teramura, Y. et al. (2006) EMBO J. 25:4215-4222), but to reduce the number of kinetic parameters to be estimated and to describe the experimental data with the simplest model possible, we constrained the binding affinity of the ligands to ErbB homo- and heterodimers to be the same.

Dimerized receptors undergo rapid phosphorylation and activate downstream pathways through binding of signaling adaptors to phosphotyrosine sites on the receptor cytoplasmic tail (Yarden Y. and Sliwkowski, M. X. (2001) Nat. Rev. Mol. Cell. Biol. 2:127-137). Here, a simplified PI3K-AKT cascade was implemented, whereby PI3K binds directly to ligand-bound heterodimers, activates PIP2, which forms a complex with PDK1 and AKT, leading to a two step double phosphorylation of AKT. ErbB3 has six sites for PI3K binding, whereas the other ErbB receptors only have one (Wallasch, C. et al. (1995) EMBO J. 14:4267-4275; Soltoff, S. P. et al. (1994) Mol. Cell. Biol. 14:3550-3558). Although it is not known whether six PI3K molecules can bind simultaneously to ErbB3, PI3K is activated 10-20 times more strongly by ErbB3 (Fedi, P. et al. (1994) Mol. Cell. Biol. 14:492-500) than by the other receptors. There is also evidence that PI3K binds to ErbB3 with greater affinity than it does to the other ErbB receptors (Jones, R. B. et al. (2006) Nature 439:168-174). We incorporated these phenomena into the model and avoided the combinatorial complexity of including six binding sites, by imposing the following stipulations on ErbB3-containing dimers: An enhanced PI3K-binding rate; enhanced PIP2-binding rate to PI3K, and enhanced activation rate of PIP3. PIP2 is not present in endosomes (Haugh, J. M. (2002) Mol. Interv. 2:292-307), thus, only plasma membrane-bound receptor dimers are capable of activating PIP2 in the model. Initial simulations accelerated AKT activation too slowly, making it necessary to almost completely deactivate the AKT phosphatase prior to simulation and incorporate a negative feedback loop whereby AKT activates its own phosphatase, a phenomenon described elsewhere (Camps, M. et al. (1998) Science 280:1262-1265). The MAPK cascade was neglected both for simplicity and because of our recent findings that AKT signaling is relatively insensitive to species and parameters in the MAPK cascade (Chen, W. W. et al. (2009) Mol. Syst. Biol. 5:239); therefore, MAPK-PI3K crosstalk is not required to describe AKT signaling dynamics. Initial species values were directly measured (ErbB receptors, PDK1, AKT), inferred from the literature (Birtwistle, M. R. et al. (2007) Mol. Syst. Biol. 3:144; Hatakeyama, M. et al. (2003) Biochem. J. 373:451-453) or set to non-rate-limiting values.

Model Calibration

Many of the reaction rate parameters are unknown in the cell system we studied and therefore had to be estimated (see Table 10 below). To avoid biasing the starting point, all parameters were set to default values (Aldridge, B. B. et al. (2006) Nat. Cell. Biol. 8:1195-1203). The number of parameters estimated was limited by performing the estimation in two stages: first the model was calibrated against the experimentally observed ErbB receptor phosphorylation data; second, the parameters sensitive to AKT phosphorylation were optimized against the experimental data. Estimated values were only accepted if the parameters were constrained under multiple parameter runs using a genetic algorithm with a population size of 50 and 25 generations (Mathworks, Mass.). The ligand binding rate constants of HRG1-β and BTC were estimated separately using known Kds (Tzahar, E. et al. (1996) Mol. Cell. Biol. 16:5276-5287; Singer, E. et al. (2001) J. Biol. Chem. 276:44266-44274; Jones, J. T. et al. (1999) FEBS Lett. 447:227-231) and initial dose responses curves to approximate a forward binding rate of 1×105M−1s−1 for both ligands. Sensitivity analyses (Mathworks, MA) were performed to identify parameters that strongly influenced the activities of ErbB1, ErbB2, ErbB3, and AKT phosphorylation with HRG1-β or BTC stimulation (see Table 11 below). Each parameter was permitted to vary separately while the system was stimulated with either 1 nM of HRG1-β or BTC. The normalized sensitivities of each species were integrated over the two hour stimulation, and parameters with normalized, integrated sensitivities greater than an arbitrary threshold of 1000 are listed in Table 11 below. As mentioned above, ErbB4 phosphorylation was not included into the parameter estimation because the cell line studied had barely detectable ErbB4 (see Table 3 below). ErbB4 reactions are parameterized according to ErbB3 reactions.

The ErbB receptor phosphorylation profiles were sensitive to dimerization, internalization, recycling, and degradation parameters (enzymatic reactions, dimer dissociation, and phosphatase rate constants were not sensitive). These parameters were fit using a high-density dataset. Each readout was normalized to the maximum activation achieved by either ligand, preserving the relative potency of each ligand. ErbB receptor dimerization rates were tightly constrained and parameter relationships were observed that paralleled literature findings: ErbB2 is the preferred dimerization partner for ligand-bound ErbB1 or ErbB3 (more noticeably with heregulin stimulation). From the available data, it was not possible to constrain both the recycling and internalization rates; therefore recycling rates were set to 0.005 s−1 and only internalization rates were fit: The resulting observations therefore apply equally to recycling rates (in inverse). Internalization rates varied based on ligand stimulus, with HRG1-β-bound dimers exhibiting much slower internalization than BTC-bound homo- and heterodimers, a view supported elsewhere (Sorkin, A. and Goh, L. K. (2008) Exp. Cell Res. 314:3093-3106). Degradation rates were also well constrained, but were similar for all dimers, suggesting that individual dimer degradation rates are not required to explain the observed data. Phosphorylated AKT is sensitive to many parameters within the PI3K cascade and at the receptor level (see Table 11 below). Therefore, we restricted the calibration to model parameters that were in the PI3K-AKT cascade and trained against the time courses of AKT phosphorylation while we locked the parameter values already trained on the ErbB receptor profiles.

Following model training, local, manual adjustment of parameters was performed to decipher the impact of each parameter, to determine if further improvement was possible, and to restrict parameters to biologically plausible values. For standardization, parameters were rounded and condensed to similar values when conserved across a parameter type (as explained for degradation rates).

Inhibitor Implementation

ErbB network inhibitors were included in the model using the simplest interpretation of known mechanisms of action (see Table 12 below): the anti-ErbB3 monoclonal antibody Ab #6 sequesters ErbB3 by preventing ligand binding and induces internalization and degradation; cetuximab sequesters ErbB1 and prevents ligand binding; lapatinib inhibits activation of ErbB receptors but not dimerization or ligand binding; and pertuzumab blocks ErbB2 dimerization. For Ab #6, the rate constants measured by Kinexa were used and for the other inhibitors the rate constants reported in the literature (Wood, E. R. et al. (2004) Cancer Res. 64:6652-6659; Patel, D. et al. (2007) Anticancer Res. 27:3355-3366; Adams, C. W. et al. (2006) Cancer Immunol. Immunother. 55:717-727) and listed in Table 13 below were used; cetuximab parameters were experimentally confirmed by Kinexa.

Various additional information used in the development of the computational model of the ErbB signaling pathway is set forth below in Tables 3-13, as follows:

TABLE 3
Measured ErbB receptor expression and mutation status for investigated cell
lines.
ErbB abundanceMutation
Tumor(molecules/cell)#status*
cell lineTypeErbB1ErbB2ErbB3ErbB4KRASPI3KPTEN
ADRrovarian17781840792332052094nonenonenone
OvCAR8ovarian2361575327231813NDnonenonenone
ACHNrenal44828445456152002086nonenonenone
MALME-melanoma29145642267367 584nonenonenone
3M
DU145prostate4378416906819422NDnonenonenone
ND—non-detectable
#as measured by qFACS
*Web site used to determine the mutation status of the cell lines investigated - http://www.sanger.ac.uk/perl/genetics/CGP/core_line_viewer?action=nci60_list
TABLE 4
Characterization of ErbB1 phosphorylation dose response
curves for ErbB1 binding ligands and HRG1-β.
% stimulation at 5 min
pErbB1NCI-ADRrrelative to maximum
EGFEC50 [nM] 6.0100.0
95% CI2.5-14 
HB-EGFEC50 [nM]21.0 87.0
95% CI12-38
EpigenEC50 [nM]ND 12.0
95% CI
AREC50 [nM]37.0 95.0
95% CI21-68
BTCEC50 [nM]5.7100.0
95% CI2.3-14 
TGFαEC50 [nM]30.0100.0
95% CI10-87
Epiregulin EC50 [nM]ND  9.0
95% CI
HRG1-βEC50 [nM]ND  0.0
ND: could not be determined
CI: 95% Confidence Interval
TABLE 5
Characterization of ErbB2 phosphorylation dose response
curves for ErbB1 binding ligands and HRG1-beta
% stimulation at 5 min
pErbB2NCI-ADRrrelative to maximum
EGFEC50 [nM] 27.0100.0
95% CI15-51
HB-EGFEC50 [nM]230.0 94.0
95% CI 74-730
EpigenEC50 [nM]ND 25.0
95% CI
AREC50 [nM]ND 54.0
95% CI
BTCEC50 [nM] 27.0 94.0
95% CI  4-190
TGFαEC50 [nM] 41.0 36.0
95% CI20-86
Epiregulin EC50 [nM]ND 18.0
95% CI
HRG1-βEC50 [nM]  7.0 43.0
95% CI2.1-23 
ND: could not be determined
CI: 95% Confidence Interval
TABLE 6
Characterization of ErbB3 phosphorylation dose response
curves for ErbB1 binding ligands and HRG1-beta
% stimulation at 5 min
pErbB3NCI-ADRrrelative to maximum
EGFEC50 [nM]ND  7.0
95% CI
HB-EGFEC50 [nM]ND 32.0
95% CI
EpigenEC50 [nM]ND  3.0
95% CI
AREC50 [nM]380.0 17.0
95% CI240-600
BTCEC50 [nM]100.0 35.0
95% CI 47-220
TGFαEC50 [nM]ND  9.0
95% CI
EpiregulinEC50 [nM]ND 22.0
95% CI
HRG1-βEC50 [nM]  9.9100.0
95% CI7.9-12 
ND: could not be determined
CI: 95% Confidence Interval
TABLE 7
Characterization of AKT phosphorylation dose response
curves for ErbB1 binding ligands and HRG1-beta
% stimulation at
5 min relative
pAKTNCI-ADRr to maximum
EGFEC50 [nM]ND26.0
95% CI
HB-EGFEC50 [nM]ND20.0
95% CI
EpigenEC50 [nM]ND11.0
95% CI
AREC50 [nM]0.028.0
95% CI0.064-27  
BTCEC50 [nM]ND22.0
95% CI
TGFαEC50 [nM]ND27.0
95% CI
Epiregulin EC50 [nM]ND11.0
95% CI
HRG1-βEC50 [nM]1.9100.0
95% CI0.23-17.1
ND: could not be determined
CI: 95% Confidence Interval
TABLE 8
Initial amounts of non-zero species in the computational model.
Species name Initial Amount Additional description
HRGvariableHeregulin
BTCvariableBetacellulin
E1178000ErbB1
E241000ErbB2
E333000ErbB3
RTKpase500000Receptor tyrosine
kinase phosphatase
PI3K800000phosphatidylinositol
3-kinase
PIP2700000phosphatidylinositol
bisphosphate
PTEN350000PTEN deleted from
chromosome 10
PDK19500000phosphoinositide-
dependent
protein kinase
AKT900000Serine, threonine
protein kinase
also known as
protein kinase B
PP2A4000Protein
phosphatase 2A
(AKT phosphatase)
PP2Aoff64000Inactive AKT
phosphatase
TABLE 9
Summary of biochemical reactions implemented into the computational model
using mass action kinetics with corresponding parameters
ReactionForwardReverse
NumberReactionparameterparameter
Ligand Binding
v1HRG + E3 <-> [E3:HRG]kf1kr1
v2HRG + [E2:E3] <-> [E2:E3:HRG]kf2kr2
v3BTC + E1 <-> [BTC:E1]kf3kr3
v4BTC + [E1:E1] <-> [BTC:E1:E1]kf4kr3
v5BTC + [E1:E2] <-> [BTC:E1:E2]kf5kr3
v6BTC + [E1:E3] <-> [BTC:E1:E3]kf5kr3
Dimerization
v7[E3:HRG] + E2 <-> [E2:E3:HRG]kf7kr7
v8[E3:HRG_p] + E2_p <-> [E2:E3:HRG_p]kf7kr7
v9[E3:HRG_p] + E2 -> [E2:E3:HRG_p]kf7
v10[E3:HRG] + E1 <-> [E1:E3:HRG]kf10kr7
v11[E3:HRG_p] + E1 <-> [E1:E3:HRG_p]kf10kr7
v12E2_p + E2 -> [E2:E2_p]kf12
v13E2_p + E2_p <-> [E2:E2_p]kf12kr12
v14[BTC:E1] + E1 <-> [BTC:E1:E1]kf14kr14
v15[BTC:E1_p] + E1 <-> [BTC:E1:E1_p]kf14kr14
v16[BTC:E1] + [BTC:E1] <-> [BTC:E1:E1:BTC]kf16kr16
v17[BTC:E1_p] + [BTC:E1] <-> [BTC:E1:E1:BTC_p]kf16kr16
v18[BTC:E1_p] + [BTC:E1_p] <-> [BTC:E1:E1:BTC_p]kf16kr16
v19[BTC:E1] + E2 <-> [BTC:E1:E2]kf19kr19
v20[BTC:E1] + E2_p -> [BTC:E1:E2_p]kf19
v21[BTC:E1_p] + E2_p <-> [BTC:E1:E2_p]kf19kr19
v22[BTC:E1_p] + E2 -> [BTC:E1:E2_p]kf19
v23[BTC:E1] + E3 <-> [BTC:E1:E3]kf23kr19
v24[BTC:E1_p] + E3 <-> [BTC:E1:E3_p]kf23kr19
v25E3 + E2 <-> [E2:E3]kf12kr12
v26E3 + E1 <-> [E1:E3]kf12kr12
v27E1 + E2 <-> [E1:E2]kf12kr12
v28E1 + E1 <-> [E1:E1]kf12kr12
v29E2 + E2 <-> [E2:E2]kf12kr12
Phosphorylation and dephosphorylation
v30[E2:E3:HRG] -> [E2:E3:HRG_p]kf30
v31[E1:E3:HRG] -> [E1:E3:HRG_p]kf30
v32[BTC:E1:E1] -> [BTC:E1:E1_p]kf30
v33[BTC:E1:E1:BTC] -> [BTC:E1:E1:BTC_p]kf30
v34[BTC:E1:E2] -> [BTC:E1:E2_p]kf30
v35[BTC:E1:E3] -> [BTC:El:E3_p]kf30
v36[E3:HRG_p] -> HRG + E3kr1
v37[BTC:E1_p] -> BTC + E1kr3
v38[E2:E3:HRG_p] + RTKpase <-> [E2:E3:HRG_p:RTKpase]kf38kr38
v39[E1:E3:HRG_p] + RTKpase <-> [E1:E3:HRG_p:RTKpase]kf38kr38
v40[BTC:E1:E1_p] + RTKpase <-> [BTC:E1:E1_p:RTKpase]kf38kr38
v41[BTC:E1:E1:BTC_p] + RTKpase <->kf38kr38
[BTC:E1:E1:BTC_p:RTKpase]
v42[BTC:E1:E2_p] + RTKpase <-> [BTC:E1:E2_p:RTKpase]kf38kr38
v43[BTC:E1:E3_p] + RTKpase <-> [BTC:E1:E3_p:RTKpase]kf38kr38
v44[E2:E2_p] + RTKpase <-> [E2:E2_p:RTKpase]kf38kr38
v45[E2:E3:HRG_p:RTKpase] -> [E2:E3:HRG] + RTKpasekf45
v46[E1:E3:HRG_p:RTKpase] -> [E1:E3:HRG] + RTKpasekf45
v47[BTC:E1:E1_p:RTKpase] -> [BTC:E1:E1] + RTKpasekf45
v48[BTC:E1:E1:BTC_p:RTKpase] -> [BTC:E1:E1:BTC] + RTKpasekf45
v49[BTC:E1:E2_p:RTKpase] -> [BTC:E1:E2] + RTKpasekf45
v50[BTC:E1:E3_p:RTKpase] -> [BTC:E1:E3] + RTKpasekf45
v51[E2:E2_p:RTKpase] -> [E2:E2] + RTKpasekf45
PI3K binding and PIP2 activation
v52[E2:E3:HRG_p] + PI3K <-> [E2:E3:HRG_p:PI3K]kf52kr52
v53[E1:E3:HRG_p] + PI3K <-> [E1:E3:HRG_p:PI3K]kf52kr52
v54[E2:E2_p] + PI3K <-> [E2:E2_p:PI3K]kf54kr54
v55[BTC:E1:E1_p] + PI3K <-> [BTC:E1:E1_p:PI3K]kf54kr54
v56[BTC:E1:E1:BTC_p] + PI3K<->kf54kr54
[BTC:E1:E1:BTC_p:PI3K]
v57[BTC:E1:E2_p] + PI3K <-> [BTC:E1:E2_p:PI3K]kf54kr54
v58[BTC:E1:E3_p] + PI3K <-> [BTC:E1:E3_p:PI3K]kf52kr52
v59[E2:E3:HRG_p:PI3K] + PIP2 <->kf59kr59
[E2:E3:HRG_p:PI3K:PIP2]
v60[E1:E3:HRG_p:PI3K] + PIP2 <->kf59kr59
[E1:E3:HRG_p:PI3K:PIP2]
v61[E2:E2_p:PI3K] + PIP2 <-> [E2:E2_p:PI3K:PIP2]kf61kr61
v62[BTC:E1:E1_p:PI3K] + PIP2 <->kf61kr61
[BTC:E1:E1_p:PI3K:PIP2]
v63[BTC:E1:E1:BTC_p:PI3K] + PIP2 <->kf61kr61
[BTC:E1:E1:BTC_p:PI3K:PIP2]
v64[BTC:E1:E2_p:PI3K] + PIP2 <->kf61kr61
[BTC:E1:E2_p:PI3K:PIP2]
v65[BTC:E1:E3_p:PI3K] + PIP2 <->kf59kr59
[BTC:E1:E3_p:PI3K:PIP2]
v66[E2:E3:HRG_p:PI3K:PIP2] -> [E2:E3:HRG_p:PI3K] + PIP3kf66
v67[E1:E3:HRG_p:PI3K:PIP2] -> [E1:E3:HRG_p:PI3K] + PIP3kf66
v68[E2:E2_p:PI3K:PIP2] -> [E2:E2_p:PI3K] + PIP3kf68
v69[BTC:E1:E1_p:PI3K:PIP2] -> [BTC:E1:E1_p:PI3K] + PIP3kf68
v70[BTC:E1:E1:BTC_p:PI3K:PIP2] ->kf68
[BTC:E1:El:BTC_p:PI3K] + PIP3
v71[BTC:E1:E2_p:PI3K:PIP2] -> [BTC:E1:E2_p:PI3K] + PIP3kf68
v72[BTC:E1:E3_p:PI3K:PIP2] -> [BTC:E1:E3_p:PI3K] + PIP3kf66
AKT activation cascade
v73PIP3 + PTEN <-> [PIP3:PTEN]kf73kr73
v74[PIP3:PTEN] -> PIP2 + PTENkf74
v75PIP3 + AKT <-> [PIP3:AKT]kf75kr75
v76[PIP3:AKT] + PDK1 <-> [PIP3:AKT:PDK1]kf76kr76
v77[PIP3:AKT:PDK1] -> AKT_p + [PIP3:PDK1]kf77
v78[PIP3:PDK1] -> PIP3 + PDK1kf78
v79PIP3 + AKT_p <-> [PIP3:AKT_p]kf75kr75
v80[PIP3:AKT_p] + PDK1 <-> [PIP3:AKT_p:PDK1]kf76kr76
v81[PIP3:AKT_p:PDK1] -> AKT_p_p + [PIP3:PDK1]kf81
v82AKT_p_p + PP2A <-> [AKT_p_p:PP2A]kf82kr82
v83[AKT_p_p:PP2A] -> AKT_p + PP2Akf83
v84AKT_p + PP2A <-> [AKT_p:PP2A]kf82kr82
v85[AKT_p:PP2A] -> AKT + PP2Akf83
v86AKT_p_p + PP2Aoff <-> [AKT_p_p:PP2Aoff]kf86kr86
v87[AKT_p_p:PP2Aoff] -> AKT_p_p + PP2Akf87
Internalizaton
v88[E3:HRG] <-> [iE3:HRG]kf88kr88
v89[E3:HRG_p] <-> [iE3:HRG_p]kf88kr88
v90E2_p <-> iE2_pkf88kr88
v91[BTC:E1] <-> [iBTC:E1]kf88kr88
v92[BTC:E1_p] <-> [iBTC:E1_p]kf88kr88
v93[E2:E3:HRG] <-> [iE2:E3:HRG]kf93kr93
v94[E2:E3:HRG_p] <-> [iE2:E3:HRG_p]kf93kr93
v95[E2:E3:HRG_p:RTKpase] <-> [iE2:E3:HRG_p:RTKpase]kf93kr93
v96[E2:E3:HRG_p:PI3K] <-> [iE2:E3:HRG_p:PI3K]kf93kr93
v97[E2:E3:HRG_p:PI3K:PIP2] <-> [iE2:E3:HRG_p:PI3K:PIP2]kf93kr93
v98[E1:E3:HRG] <-> [iE1:E3:HRG]kf98kr98
v99[E1:E3:HRG_p] <-> [iE1:E3:HRG_p]kf98kr98
v100[E1:E3:HRG_p:RTKpase] <-> [iE1:E3:HRG_p:RTKpase]kf98kr98
v101[E1:E3:HRG_p:PI3K] <-> [iE1:E3:HRG_p:PI3K]kf98kr98
v102[E1:E3:HRG_p:PI3K:PIP2] <-> [iE1:E3:HRG_p:PI3K:PIP2]kf98kr98
v103[E2:E2_p] <-> [iE2:E2_p]kf93kr93
v104[E2:E2_p:RTKpase] <-> [iE2:E2_p:RTKpase]kf93kr93
v105[E2:E2_p:PI3K] <-> [iE2:E2_p:PI3K]kf93kr93
v106[E2:E2_p:PI3K:PIP2] <-> [iE2:E2_p:PI3K:PIP2]kf93kr93
v107[BTC:E1:E1] <-> [iBTC:E1:E1]kf107kr107
v108[BTC:E1:E1:BTC] <-> [iBTC:E1:E1:BTC]kf107kr107
v109[BTC:E1:E1_p] <-> [iBTC:E1:E1_p]kf107kr107
v110[BTC:E1:E1:BTC_p] <-> [iBTC:E1:E1:BTC_p]kf107kr107
v111[BTC:E1:E1_p:RTKpase] <-> [iBTC:E1:E1_p:RTKpase]kf107kr107
v112[BTC:E1:E1:BTC_p:RTKpase] <->kf107kr107
[iBTC:E1:E1:BTC_p:RTKpase]
v113[BTC:E1:E1_p:PI3K] <-> [iBTC:E1:E1_p:PI3K]kf107kr107
v114[BTC:E1:E1:BTC_p:PI3K] <-> [iBTC:E1:E1:BTC_p:PI3K]kf107kr107
v115[BTC:E1:E1_p:PI3K:PIP2] <-> [iBTC:E1:E1_p:PI3K:PIP2]kf107kr107
v116[BTC:E1:E1:BTC_p:PI3K:PIP2] <->kf107kr107
[iBTC:E1:E1:BTC_p:PI3K:PIP2]
v117[BTC:E1:E2] <-> [iBTC:E1:E2]kf117kr117
v118[BTC:E1:E2_p] <-> [iBTC:E1:E2_p]kf117kr117
v119[BTC:E1:E2_p:RTKpase] <-> [iBTC:E1:E2_p:RTKpase]kf117kr117
v120[BTC:E1:E2_p:PI3K] <-> [iBTC:E1:E2_p:PI3K]kf117kr117
v121[BTC:E1:E2_p:PI3K:PIP2] <-> [iBTC:E1:E2_p:PI3K:PIP2]kf117kr117
v122[BTC:E1:E3] <-> [iBTC:E1:E3]kf117kr117
v123[BTC:E1:E3_p] <-> [iBTC:E1:E3_p]kf117kr117
v124[BTC:E1:E3_p:RTKpase] <-> [iBTC:E1:E3_p:RTKpase]kf117kr117
v125[BTC:E1:E3_p:PI3K] <-> [iBTC:E1:E3_p:PI3K]kf117kr117
v126[BTC:E1:E3_p:PI3K:PIP2] <-> [iBTC:E1:E3_p:PI3K:PIP2]kf117kr117
Endosomal ligand binding
v127iHRG + iE3 <-> [iE3:HRG]kf127kr1
v128iHRG + [iE2:E3] <-> [iE2:E3:HRG]kf127kr1
v129iBTC + iE1 <-> [iBTC:E1]kf129kr3
v130iBTC + [iE1:E1] <-> [iBTC:E1:E1]kf129kr3
v131iBTC + [iE1:E2] <-> [iBTC:E1:E2]kf129kr3
v132iBTC + [iE1:E3] <-> [iBTC:E1:E3]kf129kr3
Endosomal dimerization
v133[iE3:HRG] + iE2 <-> [iE2:E3:HRG]kf7kr7
v134[iE3:HRG_p] + iE2_p <-> [iE2:E3:HRG_p]kf7kr7
v135[iE3:HRG_p] + iE2 -> [iE2:E3:HRG_p]kf7
v136[iE3:HRG] + iE1 <-> [iE1:E3:HRG]kf10kr7
v137[iE3:HRG_p] + iE1<-> [iE1:E3:HRG_p]kf10kr7
v138iE2_p + iE2 -> [iE2:E2_p]kf12
v139iE2_p + iE2_p <-> [iE2:E2_p]kf12kr12
v140[iBTC:E1] + iE1 <-> [iBTC:E1:E1]kf14kr14
v141[iBTC:E1_p] + iE1 <-> [iBTC:E1:E1_p]kf14kr14
v142[iBTC:E1] + [iBTC:E1] <-> [iBTC:E1:E1:BTC]kf16kr16
v143[iBTC:E1_p] + [iBTC:E1] <-> [iBTC:E1:E1:BTC_p]kf16kr16
v144[iBTC:E1_p] + [iBTC:E1_p] <-> [iBTC:E1:E1:BTC_p]kf16kr16
v145[iBTC:E1] + iE2 <-> [iBTC:E1:E2]kf19kr19
v146[iBTC:E1_p] + iE2 -> [iBTC:E1:E2_p]kf19
v147[iBTC:E1] + iE2_p -> [iBTC:E1:E2_p]kf19
v148[iBTC:E1_p] + iE2_p <-> [iBTC:E1:E2_p]kf19kr19
v149[iBTC:E1] + iE3 <-> [iBTC:E1:E3]kf23kr19
v150[iBTC:El_p] + iE3 <-> [iBTC:E1:E3_p]kf23kr19
v151iE3 + iE2 <-> [iE2:E3]kf12kr12
v152iE3 + iE1 <-> [iE1:E3]kf12kr12
v153iE1 + iE2 <-> [iE1:E2]kf12kr12
v154iE1 + iE1 <-> [iE1:E1]kf12kr12
v155iE2 + iE2 <-> [iE2:E2]kf12kr12
Endosomal phosphorylation and dephosphorylation
v156[iE2:E3:HRG] -> [iE2:E3:HRG_p]kf30
v157[iE1:E3:HRG] -> [iE1:E3:HRG_p]kf30
v158[iBTC:E1:E1] -> [iBTC:E1:El_p]kf30
v159[iBTC:E1:E1:BTC] -> [iBTC:E1:E1:BTC_p]kf30
v160[iBTC:E1:E2] -> [iBTC:E1:E2_p]kf30
v161[iBTC:E1:E3] -> [iBTC:E1:E3_p]kf30
v162[iE3:HRG_p] -> iHRG + iE3kr1
v163[iBTC:E1_p] -> iBTC + iE1kr3
v164[iE2:E3:HRG_p] + RTKpase <-> [iE2:E3:HRG_p:RTKpase]kf38kr38
v165[iE:E3:HRG_p] + RTKpase <-> [iE1:E3:HRG_p:RTKpase]kf38kr38
v166[iE2:E2_p] + RTKpase <-> [iE2:E2_p:RTKpase]kf38kr38
v167[iBTC:E1:E1_p] + RTKpase <-> [iBTC:E1:E1_p:RTKpase]kf38kr38
v168[iBTC:E1:E1 :BTC_p] + RTKpase <->kf38kr38
[iBTC:E1:E1:BTC_p:RTKpase]
v169[iBTC:E1:E2_p] + RTKpase <-> [iBTC:E1:E2_p:RTKpase]kf38kr38
v170[iBTC:E1:E3_p] + RTKpase <-> [iBTC:E1:E3_p:RTKpase]kf38kr38
v171[iE2:E3:HRG_p:RTKpase] -> [iE2:E3:HRG] + RTKpasekf45
v172[iE1:E3:HRG_p:RTKpase] -> [iE1 E3:HRG] + RTKpasekf45
v173[iE2:E2_p:RTKpase] -> [iE2:E2] + RTKpasekf45
v174[iBTC:E1:E1_p:RTKpase] -> [iBTC:E1:E1] + RTKpasekf45
v175[iBTC:E1:E1:BTC_p:RTKpase] -> [iBTC:E1:E1:BTC] + RTKpasekf45
v176[iBTC:E1:E2_p:RTKpase] -> [iBTC:E1:E2] + RTKpasekf45
v177[iBTC:E1:E3_p:RTKpase] -> [iBTC:E1:E3] + RTKpasekf45
Endosomal PI3K binding
v178[iE2:E3:HRG_p] + PI3K <-> [iE2:E3:HRG_p:PI3K]kf52kr52
v179[iE1:E3:HRG_p] + PI3K <-> [iE1:E3:HRG_p:PI3K]kf52kr52
v180[iE2:E2_p] + PI3K <-> [iE2:E2_p:PI3K]kf54kr54
v181[iBTC:E1:El_p] + PI3K <-> [iBTC:E1:E1_p:PI3K]kf54kr54
v182[iBTC:E1:E1:BTC_p] + PI3K <->kf54kr54
[iBTC:E1:E1:BTC_p:PI3K]
v183[iBTC:E1:E2_p] + PI3K <-> [iBTC:E1:E2_p:PI3K]kf54kr54
v184[iBTC:E1:E3_p] + PI3K <-> [iBTC:E1:E3_p:PI3K]kf52kr52
Degradation
v185iHRG -> dHRGkf185
v186iBTC -> dB TCkf185
v187[iE3:HRG] -> [dE3:HRG]kf187
v188[iBTC:E1] -> [dBTC:E1]kf187
v189[iE3:HRG_p] -> [dE3:HRG_p]kf187
v190[iBTC:E1_p] -> [dBTC:E1_p]kf187
v191iE2_p -> dE2_pkf187
v192[iE2:E3:HRG] -> [dE2:E3:HRG]kf192
v193[iE2:E3:HRG_p] -> [dE2:E3:HRG_p]kf192
v194[iE2:E3:HRG_p:RTKpase] -> [dE2:E3:HRG_p:RTKpase]kf192
v195[iE2:E3:HRG_p:PI3K] -> [dE2:E3:HRG_p:PI3K]kf192
v196[iE2:E3:HRG_p:PI3K:PIP2] -> [dE2:E3:HRG_p:PI3K:PIP2]kf192
v197[iE2:E2_p] -> [dE2:E2_p]kf192
v198[iE2:E2_p:RTKpase] -> [dE2:E2_p:RTKpase]kf192
v199[iE2:E2_p:PI3K] -> [dE2:E2_p:PI3K]kf192
v200[iE2:E2_p:PI3K:PIP2] -> [dE2:E2_p:PI3K:PIP2]kf192
v201[iE1:E3:HRG] -> [dE1:E3:HRG]kf201
v202[iE1:E3:HRG_p] -> [dE1:E3:HRG_p]kf201
v203[iE1:E3:HRG_p:RTKpase] -> [dE1:E3:HRG_p:RTKpase]kf201
v204[iE1:E3:HRG_p:PI3K] -> [dE1:E3:HRG_p:PI3K]kf201
v205[iE1:E3:HRG_p:PI3K:PIP2] -> [dE1:E3:HRG_p:PI3K:PIP2]kf201
v206[iBTC:E1:E1] -> [dBTC:E1:E1]kf206
v207[iBTC:E1:E1_p] -> [dBTC:E1:E1_p]kf206
v208[iBTC:E1:E1_p:RTKpase] -> [dBTC:E1:E1_p:RTKpase]kf206
v209[iBTC:E1:E1_p:PI3K] -> [dBTC:E1:E1_p:PI3K]kf206
v210[iBTC:E1:E1_p:PI3K:PIP2] -> [dBTC:E1:E1_p:PI3K:PIP2]kf206
v211[iBTC:E1:E1:BTC] -> [dBTC:E1:E1:BTC]kf206
v212[iBTC:E1:E1:BTC_p] -> [dBTC:E1:E1:BTC_p]kf206
v213[iBTC:E1:E1:BTC_p:RTKpase] ->kf206
[dBTC:E1:E1:BTC_p:RTKpase]
v214[iBTC:E1:E1:BTC_p:PI3K] -> [dBTC:E1:E1:BTC_p:PI3K]kf206
v215[iBTC:E1:E1:BTC_p:PI3K:PIP2] ->kf206
[dBTC:E1:E1:BTC_p:PI3K:PIP2]
v216[iBTC:E1:E2] -> [dBTC:E1:E2]kf201
v217[iBTC:E1:E2_p] -> [dBTC:E1:E2_p]kf201
v218[iBTC:E1:E2_p:RTKpase] -> [dBTC:E1:E2_p:RTKpase]kf201
v219[iBTC:E1:E2_p:PI3K] -> [dBTC:E1:E2_p:PI3K]kf201
v220[iBTC:E1:E2_p:PI3K:PIP2] -> [dBTC:E1:E2_p:PI3K:PIP2]kf201
v221[iBTC:E1:E3] -> [dBTC:E1:E3]kf201
v222[iBTC:E1:E3_p] -> [dBTC:El:E3_p]kf201
v223[iBTC:E1:E3_p:RTKpase] -> [dBTC:E1:E3_p:RTKpase]kf201
v224[iBTC:E1:E3_p:PI3K] -> [dBTC:E1:E3_p:PI3K]kf201
v225[iBTC:E1:E3_p:PI3K:PIP2] -> [dBTC:E1:E3_p:PI3K:PIP2]kf201
Definition of abbreviations used:
: indicates a protein complex e.g. ligand bound to receptor
_p indicates that a protein is phosphorylated
iy indicates that a species y is internalized
<-> indicates a reversible reaction
-> indicates an irreversible reaction
TABLE 10
Description of parameters with values. Parameter number corresponds to
first reaction in which that parameter appears
NameValueUnitsDescription
Av6.0 × 1023Avogadro's number
Vmedia1.00 × 10−04LitersMedia volume per well
Vcell1.00 × 10−12LitersCell volume
Num30000Number of cells per well.
cells
kf15.00 × 10−11molecules−1 HRG binding to E3 or E4.
sec−1Unit conversion calculated
as 1e5M−1s−1/
(Vmedia * Av/Num_cells).
kr10.001sec−1HRG dissociation from E3
kf25.00 × 10−11molecules−1 HRG binding to E3:E2 dimers
sec−1
kr20.001sec−1HRG dissociation from
E3:E2 dimers
kf35.00 × 10−11molecules−1 BTC binding to E1
sec−1
kr30.001sec−1BTC dissociation from E1
kf45.00 × 10−11molecules−1 BTC binding to E1
sec−1homodimers
kf55.00 × 10−11molecules−1 BTC binding to E1
sec−1heterodimers
kf73.00 × 10−06molecules−1 Dimerization of E2 to
sec−1HRG:E3
kr70.001sec−1Dissociation of E2 to
HRG:E3
kf103.00 × 10−08molecules−1 Dimerization of E1 to
sec−1HRG:E3
kf124.20 × 10−09molecules−1 Dimerization-constitutive
sec−1and ligand free
kr120.001sec−1Dissociation-constitutive and
ligand free
kf141.70 × 10−05molecules−1 Dimerization of E1 to
sec−1BTC:E1
kr140.001sec−1Dissociation of E1 to BTC:E1
kf161.70 × 10−05molecules−1 Dimerization of BTC:E1 to
sec−1BTC:E1
kr160.001sec−1Dissociation of BTC:E1 to
BTC:E1
kf193.30 × 10−05molecules−1 Dimerization of E2 to
sec−1BTC:E1
kr190.001sec−1Dissociation of E2 or E3 to
BTC:E1
kf234.70 × 10−06molecules−1 Dimerization of E3 to
sec−1BTC:E1
kf301sec−1Enzymatic auto-
phosphorylation rate
of ligand-bound dimers
kf385.00 × 10−06molecules−1 Receptor phosphatase
sec−1 binding
kr380.1sec−1Receptor phosphatase
dissociation
kf451sec−1Enzymatic rate for receptor
dephosphorylation
kf523.00 × 10−06molecules−1 PI3K binding to E3
sec−1containing dimers
kr520.1sec−1PI3K dissociation from E3
containing dimers
kf547.50 ×10−06molecules−1 PI3K binding to non-E3
sec−1containing dimers
kr540.1sec−1PI3K dissociation from
non-E3 dimers
kf595.00 × 10−06 molecules−1 PIP2 binding to E3
sec−1containing heterodimers
kr590.1sec−1PIP2 dissociation from E3
containing heterodimer
kf615.00 × 10−07 molecules−1 PIP2 binding to non-E3
sec−1containing heterodimers
kr610.1sec−1PIP2 dissociation from non-E3
containing heterodimer
kf660.2sec−1PIP3 activation by E3
containing dimers
kf680.013sec−1PIP3 activation rate by
non E3 containing dimers
kf735.00 × 10−06molecules−1 PIP3 binding to PTEN
sec−1
kr730.1sec−1PIP3 dissociating from PTEN
kf740.1sec−1PIP3 inactivation by PTEN
kf752.60 × 10−04molecules−1 PIP3 binding to Akt or Akt_p
sec−1
kr750.1sec−1PIP3 dissociation from Akt or Akt_p
kf766.70 × 10−05molecules−1 PDK1 binding to PIP3:Akt
sec−1
kr760.1sec−1PDK1 dissociation from PIP3:Akt
kf771sec−1Enzymatic phosphorylation
rate for Akt
kf780.2sec−1PDK1 dissociating from PIP3
kf811sec−1Enzymatic phosphorylation
rate for Akt_p
kf821.70 × 10−06molecules−1 PP2A binding to
sec−1phosphorylated Akt
kr820.1sec−1PP2A dissociation from
phosphorylated Akt
kf831.5sec−1Akt dephosphorylation
and dissociation
kf868.30 × 10−09molecules−1 binding of AKT_p_p
sec−1to PP2Aoff
kr860.5sec−1Dissociation of AKT_p_p
from PP2Aoff
kf870.1sec−1Activation of PP2Aoff
by AKT_p_p
kf880.1sec−1Internalization rate for ligand-
bound or active monomers
kr880.005sec−1Recycling rate ligand-bound
or active monomers
kf930.005sec−1Internalization rate for
HRG-bound E2:E3
heterodimers and
E2 homodimers
kr930.005sec−1Recycling rate for
HRG-bound E2:E3
heterodimers and
E2 homodimers
kf980.005sec−1Internalization rate
for HRG-bound
E1:E3 heterodimers
kr980.005sec−1Recycling rate for HRG-bound
E1:E3 heterodimers
kf1070.1sec−1Internalization rate for
BTC-bound E1 homodimers
kr1070.005sec−1Recycling rate for
BTC-bound E1 homodimers
kf1170.1sec−1Internalization rate for
BTC-bound E1 heterodimers
kr1170.005sec−1Recycling rate for BTC-bound
E1 containing heterodimers
kf1273.8molecules−1 HRG binding in the
sec−1endosome
kf1293.8molecules−1 BTC binding in the
sec−1endosome
kf1850.002sec−1Degradation rate for ligand
kf1870.002sec−1Degradation rate for ligand
bound monomers
kf1920.002sec−1Degradation rate for
ligand-bound E2 containing
homo or heterodimers
kf2010.002sec−1Degradation rate for
ligand-bound E1
containing heterodimers
kf2060.002sec−1Degradation of ligand-bound
E1 homodimers
TABLE 11
ErbB3, ErbB2, ErbB1. and AKT sensitive parameters.
pErbB3pErbB3
sensitivity duringsensitivity during
Heregulin stimulation Betacellulin stimulation
kf14799kf53824
kr932163kf232353
kf71902kf31821
kr881277kf121389
kr121192kr12−1343
kf121111kf14−1742
kf187−1136kf14−6462
kf88−1907
kf93−3122
kf192−4452
pErbB2pErbB2
sensitivity duringsensitivity during
Heregulin stimulationBetacellulin stimulation
kf14747kf33878
kr932245kf193124
kf72136kf51060
kr881267kf14−2768
kr121155kf14−6515
kf12−1071
kf187−1137
kf88−1892
kf93−3239
kf192−4617
pErbB1pErbB1
sensitivity duringsensitivity during
Heregulin stimulation Betacellulin stimulation
kf106990kf32847
kf16222kf41044
kr122224kfl4−1665
kr982174kf206−4983
kr881616
kf187 −423
kf12−2189
kf88−2414
kf98−3155
kf14−4457
kf7−4469
pAKTpAKT
sensitivity duringsensitivity during
Heregulin stimulationBetacellulin stimulation
kf666657kf755462
kf755418kf665395
kf14799kr117 4420
kr933937kf53250
kr732705kr862854
kf72022kr732726
kr861918kf32351
kf781584kf681815
kr881209kf231804
kr121127kf781750
kfl2−1044 kr107 1024
kf187 −1073 kf83−1019
kf87−1929 kf14−1338
kf88−1967 kf107 −1552
kf86−2302 kf74−2720
kf83−2520 kf87−2861
kf74−2719 kf86−3425
kf192−3667kfl4−3890
kf82−4458 kf73−5459
kf73−5416 kf82−5921
kf93−5748 kf117 −6849
TABLE 12
Implementation scheme forAb
ReactionForwardReverse
NumberReactionparameterparameter
Ab #6 implementation
Ab #6 _v1 Ab #6 + E3 <-> [E3:Ab #6]kfAb #6 _1krAb #6 _1
Ab #6 _v2[E3:Ab #6] + E3 <-> [iE3:Ab #6:E3]kfAb #6 _2krAb #6 _2
Ab #6 _v3[E3:Ab #6:E3] <-> [iE3:Ab #6:E3]kfAb #6 _3krAb #6 _3
Ab #6 _v4[iE3:Ab #6:E3] <-> [dE3:Ab #6:E3]kfAb #6 _4
Cetuximab implementation
Cetuximab_v1Cetuximab + E1 <-> [E1:Cetuximab]kfCetuximab_1krCetuximab_1
Cetuximab_v2[E1:Cetuximab] + E1 <-> kfCetuximab_2krCetuxmab_2
[E1:Cetuximab:E1]
Pertuzumab implementation
Pertuzumab_v1[Pertuzumab] + E2 <-> [E2:Pertuzumab]kfPertuzumab_1krPertuzumab_1
Pertuzumab_v2[E2:Pertuzumab] + E2 <-> kfPertuzumab_2krPertuzumab_2
[E2:Pertuzumab:E2]
Lapatinib implementation
Lapatinib_v1Lapatinib + E1 <-> [Lapatinib:E1]kfLapatinib_1krLapatinib_1
Lapatinib_v2Lapatinib + E2 <-> [Lapatinib:E2]kfLapatinib_2krLapatinib_2
Lapatinib_v3Lapatinib + [E1:E1] <-> kfLapatinib_3krLapatinib_1
[Lapatinib:E1:E1]
Lapatinib_v4Lapatinib + [E1:E2] <-> kfLapatinib_3krLapatinib_1
[Lapatinib:E1:E2]
Lapatinib_v5Lapatinib + [E1:E3] <-> kfLapatinib_1krLapatinib_1
[Lapatinib:E1:E3]
Lapatinib_v6Lapatinib + [E2:E2] <->kfLapatinib_4krLapatinib_2
[Lapatinib:E2:E2]
Lapatinib_v7Lapatinib + [E2:E3] <-> kfLapatinib_2krLapatinib_2
[Lapatinib:E2:E3]
Lapatinib_v8Lapatinib + [Lapatinib:E1:E1] <->kfLapatinib_1krLapatinib_1
[Lapatinib:E1:E1:Lapatinib]
Lapatinib_v9Lapatinib + [Lapatinib:E1:E2] <->kfLapatinib_1krLapatinib_1
[Lapatinib:E1:E2:Lapatinib]
Lapatinib_v10Lapatinib + [Lapatinib:E2:E2] <->kfLapatinib_2krLapatinib_2
[Lapatinib:E2:E2:Lapatinib]
Lapatinib_v11Lapatinib + [BTC:E1] <-> kfLapatinib_1krLapatinib_1
[Lapatinib:BTC:E1]
Lapatinib_v12Lapatinib [TC:E1:E1] <->kfLapatinib_3krLapatinib_1
[Lapatinib:B TC:E1:E1]
Lapatinib_v13Lapatinib + [BTC:E1:E1:BTC] <->kfLapatinib_3krLapatinib_1
[Lapatinib:BTC:E1:E1:BTC]
Lapatinib_v14Lapatinib + [BTC:E1:E2] <->kfLapatinib_3krLapatinib_1
[Lapatinib:BTC:E1:E2]
Lapatinib_v15Lapatinib + [BTC:E1:E3] <->kfLapatinib_1krLapatinib_1
[Lapatinib:BTC:E1:E3]
Lapatinib_v16Lapatinib + [E2:E3:HRG] <->kfLapatinib_2krLapatinib_2
[Lapatinib:E2:E3:HRG]
Lapatinib_v17Lapatinib + [E1:E3:HRG] <->kfLapatinib_1krLapatinib_1
[Lapatinib:E1:E3:HRG]
Lapatinib_v18Lapatinib + [Lapatinib:BTC:E1:E1] <->kfLapatinib_1krLapatinib_1
[Lapatinib:BTC:E1:E1:Lapatinib]
Lapatinib_v19Lapatinib + [Lapatinib:BTC:E1:E1:BTC] <->kfLapatinib_1krLapatinib_1
[Lapatinib:BTC:E1:E1:BTC:Lapatinib]
Lapatinib_v20Lapatinib + [Lapatinib:BTC:E1:E2] <->kfLapatinib_1krLapatinib_1
[Lapatinib:BTC:E1:E2:Lapatinib]
Lapatinib_v21[Lapatinib:E1] + E1<-> kf12kr12
[Lapatinib:E1:E1]
Lapatinib_v22[Lapatinib:E1] + E2 <-> kf12kr12
[Lapatinib:E1:E2]
Lapatinib_v23[Lapatinib:E1] + E3 <-> kf12kr12
[Lapatinib:E1:E3]
Lapatinib_v24[Lapatinib:E2] + E1<-> kf12kr12
[Lapatinib:E1:E2]
Lapatinib_v25[Lapatinib:E2] + E3 <-> kf12kr12
[Lapatinib:E2:E3]
Lapatinib_v26[Lapatinib:E1] + [Lapatinib:E1] <->kf12kr12
[Lapatinib:E1:E1:Lapatinib]
Lapatinib_v27[Lapatinib:E1] + [Lapatinib:E2] <->kf12kr12
[Lapatinib:E1:E2:Lapatinib]
Lapatinib_v28[Lapatinib:E2] + [Lapatinib:E2] <->kf12kr12
[Lapatinib:E2:E2:Lapatinib]
Lapatinib_v29[Lapatinib:BTC:E1] + E1 <->kf14kr14
[Lapatinib:BTC:E1:E1]
Lapatinib_v30[Lapatinib:BTC:E1] + E2 <->kf19kr19
[Lapatinib:BTC:E1:E2]
Lapatinib_v31[Lapatinib:BTC:E1] + E3 <->kf23kr19
[Lapatinib:BTC:E1:E3]
Lapatinib_v32[Lapatinib:BTC:E1] + [BTC:E1] <->kf16krl6
[Lapatinib:BTC:E1:E1:BTC]
Lapatinib_v33[Lapatinib:BTC:E1] + [Lapatinib:BTC:E1] <->kf16kr16
[Lapatinib:BTC:E1:E1:BTC:Lapatinib]
Lapatinib_v34[Lapatinib:BTC:E1] + [Lapatinib:E2] <->kf19kr19
[Lapatinib:BTC:E1:E2:Lapatinib]
Lapatinib_v35[Lapatinib:E2] + [HRG:E3] <->kf7kr7
[Lapatinib:E2:E3:HRG]
Lapatinib_v36[Lapatinib:E1] + [HRG:E3] <->kf10kr7
[Lapatinib:E1:E3:HRG]
Lapatinib_v37[Lapatinib:E1] + BTC <-> kf3kr3
[Lapatinib:BTC:E1]
Lapatinib_v38[Lapatinib:E1:E1] + BTC <->kf4kr3
[Lapatinib:BTC:E1:E1]
Lapatinib_v39[Lapatinib:E1:E2] + BTC <->kf5kr3
[Lapatinib:BTC:E1:E2]
Lapatinib_v40[Lapatinib:E1:E3] + BTC <->kf5kr3
[Lapatinib:BTC:E1:E3]
Lapatinib_v41[Lapatinib:BTC:E1:E1] + BTC <->kf4kr4
[Lapatinib:BTC:E1:E1:BTC]
Lapatinib_v42[Lapatinib:E2:E3] + HRG <->kf2kr1
[Lapatinib:E2:E3:HRG]
TABLE 13
Inhibitor parameter values.
NameValueUnitsDescription
Vcell1.0E−12LitersCell volume.
Vshell4.8E−15LitersReduced volume Vshell for 2nd arm of IgG binding.
Vshell = 4/3*pi*((Cell_radius +
Cell_shellheight){circumflex over ( )}3 − cell_radius)
Cell radius (Vcell*3/(4*pi())){circumflex over ( )}(1/3) in decimeters.
Cell_shellheight = 1e−7 decimeters.
Cell_shellhight represents the average distance
between antibody binding sites.
MM-121 parameters
kf Ab #6 _17.15E−11molecules−1Ab #6 inhibitor binding to ErbB3. Unit conversion
sec−1calculated as 1.43e5M−1s−1/(Vmedia*Av/Num_cells).
Num_cells = 30,000 per well.
kr Ab #6 _11.10E−04sec−1Ab #6 inhibitor dissociation from ErbB3
kf Ab #6 _24.96E−05molecules−1Ab #6 inhibitor binding to a second ErbB3. Avidity
sec−1due to volume reduction = kr Ab #6 _1*
(Vmedia/Numcells)/Vshell.
kr Ab #6 _22.20E−04sec−1Ab #6 inhibitor dissociation from a second ErbB3
(2*krmm121_1)
kf Ab #6 _35.56E−04sec−1Internalization rate for Ab #6 bound to two ErbB3
molecules
kr Ab #6 _35.00E−03sec−1Recycling rate for Ab #6 bound to two ErbB3
molecules
kf Ab #6 _42.00E−04sec−1Degradation of Ab #6 bound species
Cetuximab parameters
kfCetuximab_11.10E−10molecules−1Cetuximab inhibitor binding to ErbB1. Unit
sec−1conversion calculated as 2.2e5M−1s−1/
(Vmedia*Av/Num_cells).
krCetuximab_11.10E−03sec−1Cetuximab inhibitor dissociation from ErbB1
kfCetuximab27.64E−05molecules−1Cetuximab inhibitor binding to a second ErbB1.
sec−1Avidity due to volume reduction =
krCetuximab_1*(Vmedia/Numcells)/Vshell.
krCetuximab_22.20E−03sec−1Cetuximab inhibitor dissociation when bound to two
ErbB1 molecules (2*krcetuximab_1)
Pertuzumab parameters
kfPertuzumab_15.60E−11molecules−1Pertuzumab inhibitor binding to ErbB2. Unit
sec−1conversion calculated as 1.12e5M−1s−1/
(Vmedia*Av/Num_cells).
krPertuzumab_19.50E−04sec−1Pertuzumab inhibitor dissociation from ErbB2
kfPertuzumab_23.90E−05molecules−1Pertuzumab inhibitor binding to a second ErbB2.
sec−1Avidity due to volume reduction =
krPertuzumab_1*(Vmedia/Numcells)/Vshell.
krPertuzumab_21.90E−03sec−1Pertuzumab inhibitor dissociation from a second
ErbB2 (2*krpertuzumab_1).
Lapatinib parameters
kfLapatinib-16.40E−12molecules−1Lapatinib inhibitor binding to ErbB1 or 1-3 dimers.
sec−1Unit conversion calculated as 1.28e4M−1s−1/
(Vmedia*Av/Num_cells).
krLapatinib_13.83E−05sec−1Lapatinib inhibitor dissociation from ErbB1 or 1-3
dimers.
kfLapatinib_21.50E−12molecules−1Lapatinib inhibitor binding from ErbB2 or 2-3 dimers.
sec−1Unit conversion calculated as 2.95e3M−1s−1/
(Vmedia*Av/Num_cells).
krLapatinib_23.83E−05sec−1Lapatinib inhibitor dissociation to ErbB2 or 2-3
dimers.
kfLapatinib_31.28E−11molecules−1Lapatinib inhibitor binding to 1-1 or 1-2 dimers
sec−1
kfLapatinib_43.00E−12molecules−1Lapatinib inhibitor binding to 2-2 dimers
sec−1

In this example, a set of protein markers that are predictive for activation of ErbB3, as indicated by pErbB3, were identified using the mechanistic computational model of the ErbB signaling pathway that was described in Example 4.

As the level of ErbB3 phosphorylation was demonstrated in Example 2 to correlate with tumor response rate, a sensitivity analysis was conducted on the trained computational model to identify the key proteins that determine the level of ErbB3 phosphorylation.

In this local sensitivity analysis, cells were virtually stimulated in silico with 0.4 nM of either HRG or BTC, both of which are ligands that activate the ErbB signaling pathway. The sensitivity of pErbB3 with respect to the following cellular receptors, kinases and other proteins was determined: ErbB3, ErbB2, ErbB1, PI3K, PIP2, PTEN, PDK1, PP2A, AKT, RTKpase and the ligands (BTC and HRG).

The local sensitivity analysis is a mathematical tool that measures changes in an output in response to changes in protein concentrations and kinetic parameters within the pathway. The fully normalized sensitivity (sij(t)) of the ith observable ci(t) with respect to a change in the jth rate constant (kj) is given by the following equation:

s ij ( t ) ln ( c i ( t ) ) ln ( k j ) ( Eqn . 2 )

Model calibration was then performed using local and global optimization methods (Genetic Algorithms, simulated annealing, Levenberg-Marquardt optimization) that minimized the distance between the experimental data and the simulation results by varying the parameters and initial protein concentrations identified in the sensitivity analysis.

The results of the local sensitivity analysis are summarized in the bar graph of FIG. 6. The results indicated that the following five proteins are the key set of markers that are predictive for activation of ErbB3 (e.g., formation of pErbB3): ErbB1, ErbB2, ErbB3, HRG and BTC.

Based on the results obtained in Example 5, in which ErbB1, ErbB2, ErbB3, HRG and BTC were identified as the key markers for prediction of pErbB3 using the computational model, the measurements of the protein expression levels depicted in Table 2 were used as inputs into the computational model to compute pErbB3 levels for different tumor cell lines.

Input of BTC and HRG expression levels into the computational model required conversion from dimensionless units or pg/μg into a concentration [M]. Thus, conversion factors needed to be established. The conversion factors that converted HRG mRNA levels and BTC protein expression levels into a molar concentration were extrapolated in the linear range between experimentally measured and predicted pErbB3 levels. For the experimentally measured values, the constitutive ErbB3 phosphorylation levels (pg/μg) were measured in the four cell lines (MALME3M, DU145, ADRr and ACHN) in 10% fetal bovine serum (FBS). These experimentally measured results are shown in Example 4 in Table 2, column 7. For ligand conversion factor training, the normalized predicted pErbB3 signal integrated over time was plotted versus the experimentally measured pErbB3 in 10% FBS in vitro, using a BTC conversion factor of 6.1e-005 and a HRG mRNA conversion factor of 3.1e-013. The ligand conversion factors were trained by optimizing the linear relationship between predicted pErbB3 and measured constitutive pErbB3 levels (by ELISA) in the cell lines ADRr, MALME3M, ACHN and DU145.

Thus, the activation of the pathway by HRG and BTC was simulated using the model and the Network Activation State (NAS), as indicated by computed pErbB3 levels, was obtained as the output. In this case, the NAS was defined as the amount of time-integrated pErbB3 simulated in the model over the first two hours of stimulation by HRG and BTC. The results for the computed pErbB3 levels are shown in the graph of FIG. 7. The simulated NAS for ADRr cells was initially set as the threshold between responder and non-responder to Ab #6 treatment, since of the four cell lines tested in the xenograft models, the ADRr cell line was a non-responder with the highest pErbB3 level.

In this example, the xenograft responses for the four tumor cell lines described in Example 1 were combined with the NAS values (normalized, time-integrated pErbB3 levels) computed as described in Example 6 to set NAS threshold values for responders to Ab #6 treatment and non-responders to Ab #6 treatment.

More specifically, the growth rate reduction (GRR) values determined for the ADRr, ACHN, DU145 and MALME3 cell lines (described further in Example 1) were converted to a binary outcome for stratification training by setting “responders” as having a pErbB3 level that is greater than the level of pErbB3 in ADRr cells (i.e., a pErbB3>pErbB3(ADRr)), wherein “non-responders” were set as having a pErbB3 level that is less than the level of pErbB3 in ADRr cells (i.e., a pErbB3<pErbB3(ADRr)). It should be noted that this was not a model prediction; rather, this was part of the stratification training process.

For the four cell lines, the experimentally determined GRR values were plotted against the computed NAS values (normalized, time-integrated pErbB3). The GRR values, on the x-axis, were divided into the responders (pErbB3>pErbB3(ADRr)) and the non-responders (pErbB3<pErbB3(ADRr)). The NAS training data (obtained as described in Example 4) allowed for division of the Network Activation State y-axis into three categories: Simulated Responder (“Sim R”), Simulated Non-Responder (“Sim NR”) and Simulated Indeterminate (“Sim I”). Two xenograft cell lines were characterized by Growth Rate Reduction values of more than 20% (DU145, ACHN) and of these, the DU145 cell line had the lowest Network Activation State. Consequently, the threshold for classifying a cell line as a Simulated Responder was set at a Network Activation State greater than or equal to the ADRr level. Similarly, the MALME3 cell line xenograft was a non-responder (pErbB3<pErbB3(ADRr)) and, therefore, Network Activation States of cell lines that are lower than the ADRr level were classified as Simulated Non-Responders. Network Activation States between these the ADRr and DU145 thresholds are classified as Simulated Indeterminate.

The Network Activation State, as indicated by computed pErbB3 levels, was simulated for a panel of 15 cell lines for which experimental measurements of HRG, BTC, ErbB1, ErbB2 and ErbB3 were available. The integrated pErbB3 levels computed for the cells were plotted, along with the levels for the 4 training cell lines, in the bar graph shown in FIG. 8A, from highest to lowest pErbB3 levels. The computed NAS values for these 15 cell lines were then ranked against the NAS values previously determined for the four training cell lines ADRr, ACHN, DU145 and MALME3M. The NAS results for the 19 cell lines in total are ranked as shown in the graph of FIG. 8B. NAS values equal to or below that of the MALME3M cell line were set as simulated non-responders (“Sim NR”), NAS values between those of the MALME3M and DU145 cell lines were set as simulated indeterminate (“Sim I”) and NAS values equal to or above that of the DU145 cell line were set as simulated responders (“Sim R”). Thus, as illustrated in FIG. 8B, the IGROV1 (NCI-60, cosmic sample ID No. 905968), MDA-MB-361 (ATCC No. HTB-27), SKMEL-5 (ATCC No. HTB-70), MDA-MB-231 (ATCC No. HTB-26) and T47D (ATCC No. HTB-133) cell lines were predicted to be simulated non-responders, since their computed NAS values were below those of MALME3M. Moreover, the ZR75-1 (ATCC No. CRL-1500), HOP92 (NCI-60, cosmic sample ID No. 905973) and HOP62 (NCI-60, cosmic sample ID No. 905972) cell lines were predicted to be simulated indeterminate, since their computed NAS values were between those of ADRr and DU145. Finally, the SKBR3 (ATCC No. HTB-30), UACC62 (NCI-60, cosmic sample ID No. 905976), EKVX (NCI-60, cosmic sample ID No. 905970), BT474 (ATCC No. HTB-20), SKOV3 (ATCC No. HTB-77), OVCAR8 (obtained from the National Cancer Institute, Division of Cancer Treatment and Diagnostics) and CAK11 (NCI-60, cosmic sample ID No. 905963) cell lines were predicted to be simulated responders, since their computed NAS values were higher than that of ADRr.

To test these model predictions, three additional in vivo xenograft studies were performed. The IGROV1, OVCAR8 and SKOV3 cell lines were used in xenograft studies conducted as described in Example 1, wherein mice were treated with 600 μg of Ab #6 every 3 days or with PBS as a control. The xenograft responses, as determined by changes in tumor volume (in mm3) over time, are summarized in the graphs of FIGS. 9A-9C. Again, the Growth Rate Reduction (GRR) value for each cell line was calculated using the following formula:

Growth Rate Reduction=1−(Ab #6 Growth Rate)/(PBS Growth Rate)

The GRR values for the four cell lines tested are summarized in Table 14 below:

TABLE 14
Summary of Tumor Growth Rate Reduction for
Predicted Set of Xenograft Studies
Cell LineGRR [%]
IGROV16.3
OVCAR891.4
SKOV319.6

Regarding the prediction of Ab #6 responsiveness for each cell line, based on the criteria that a xenograft responder must have a simulated pErbB3 level greater than the simulated pErbB3 for ADRr, the IGROV1 xenograft was categorized as a non-responder, whereas the OVCAR8 and SKOV3 xenografts were categorized as responders.

These data demonstrate that the predictions made based on the computed NAS values precisely corresponded to the experimentally observed responsiveness of the three test cell lines to Ab #6 treatment in vivo in the xenograft studies. More specifically, the IGROV1 cell line was predicted to be a simulated non-responder based on its computed NAS value, and was experimentally observed to be a non-responder based on its GRR value. Similarly, the OVCAR8, and SKOV3 cell lines were predicted to be simulated responders based on their computed NAS values, and were experimentally observed to be responders based on their GRR values.

Accordingly, these experiments confirmed the effectiveness of the computational model, and the computed NAS value of normalized time-integrated pErbB3 levels, as being predictive of responsiveness to Ab #6 treatment in vivo.

In this example, the data obtained for the four cell lines examined in the xenograft studies described in Example 1 (ADRr, ACHN, DU154 and MALME3M) and the three cell lines examined in the xenograft studies described in Example 7 (IGROV1, OVCAR8 and SKOV3) were further examined to determine whether direct biomarkers for Ab #6 responsiveness could be identified.

First, it was examined whether the receptor concentrations for ErbB1, ErbB2 and ErbB3 effectively classified the xenograft data into responders and non-responders. As illustrated in the graphs of FIGS. 10A-10D (in which the log of the concentration of one receptor is plotted against the log of the concentration of one or more of the other receptors), only ErbB1 receptor measurements appeared to classify the xenograft data into responders and non-responders, while none of the other receptors did so.

Next, it was examined whether the concentration of HRG in combination with one receptor concentration (e.g., ErbB1, ErbB3) effectively classified the xenograft data into responders and non-responders. As illustrated in the graph of FIG. 11 (in which the log of the concentration of HRG is plotted against the log of the concentration of ErbB1), these two concentration measurements, HRG and one of the ErbB receptors, were able to accurately classify the xenograft data into responders and non-responders. More specifically, the data for the three non-responders MALME3, ADRr and IGROV1, was separable from the data for the six responders, thereby allowing for classification of non-responders versus responders.

Accordingly, direct biomarkers for Ab #6 responsiveness were identified as HRG in combination with one of the ErbB pathway receptors (e.g., ErbB1, ErbB3).

Table 15 below summarizes the predicted responders and non-responders for the cell lines studied in Example 7 using the direct biomarkers for Ab #6 to make the predictions. The prediction determined using the direct biomarkers corresponded well with the predictions determined using the NAS values (as described in Example 7). For example, the ACHN, DU145, OVCAR8 and SKOV3 cell lines previously had been identified as responders using the NAS values and were also predicted to be responders using the direct biomarkers. Similarly, the ADRr, MALME3M and IGROV1 cell lines previously had been identified as non-responders using the NAS values and were also predicted to be non-responders using the direct biomarkers.

TABLE 15
Predicted Responders and Non-responders
using the Direct Biomarkers for Ab #6
RespondersIn betweenNon-Responders
ACHNSKBR3ADRr
CAKI1BT474
DU145HOP92
EKVXIGROV1
HOP62MALME3M
MDA-MB-231MDA-MB-361
OVCAR8SKMEL5
SKOV3T47D
UACC62
ZR75-1

Comparing the results from Example 7 using the NAS to segregate between responders and non-responders with the results from Example 8 using the direct biomarkers for the segregation, the only discrepancies were for the BT474, MDA-MB-231 and UACC62 cell lines. It is not surprising that there are some differences between these two distinct methods of classification. In this situation, the disputed cell lines are considered responders since, in the context in patient stratification, false positives are preferable to false negatives.

In this example, the similarity or differences in the protein expression profiles observed in xenografts compared to protein expression profiles observed in human tumors were assessed. This was done to determine whether the protein levels observed in the xenografts were comparable to the protein levels observed in human tumors.

In a first set of experiments, protein expression levels are measured by ELISA. Lysates are prepared from snap frozen tumors from human tumor samples or from xenografts, substantially as described in detail in Example 2, and are analyzed by ELISA for protein levels for ErbB1-4, HRG-β1 and BTC, also as substantially described in Example 2. The results (obtained using the methods described above or minor variations thereof) are plotted in the graphs of FIGS. 12A-12C. The results demonstrate that the values obtained for the protein levels in the xenograft samples are largely interspersed with the values obtained for the protein levels in the human tumor samples from different tissue origin, indicating that the protein levels observed in the xenografts are comparable to the protein levels observed in human tumors. Based on these data, it is asserted that the NAS thresholds determined for the prediction of responsiveness in xenografts can be applied to predict responders using human tumor tissue samples.

As frozen tissue samples can be difficult to obtain in a clinical setting, a second set of experiments were performed using measurement techniques that allow for protein quantitation in Formalin Fixed and Paraffin Embedded (FFPE) samples. More specifically, quantitative immunohistochemistry (qIHC) was performed using the AQUA® system (HistoRx, Inc., New Haven, Conn.). Using immunofluorescence and a cell line panel with representative protein expression levels, a cell line standard curve was prepared. The cell line standard curve then allowed for back-calculation of protein expression level in tumor samples and xenografts. The results are shown in FIGS. 13A-13D. FIG. 13A shows a cell line standard curve for ErbB1. FIGS. 13B, 13C and 13D show bar graphs plotting the qIHC scores for ErbB1, ErbB2 and ErbB3, respectively, in the xenograft cell lines (red bars) and human tumor samples (blue bars). The qIHC results demonstrated the similarity in protein expression levels between the human tumor samples and the xenograft samples, which span a wide range of protein expression levels. These results again support the assertion that the NAS thresholds determined for the prediction of responsiveness in xenografts can be applied to predict responders using human tumor tissue samples.

In this example, integrated levels of phosphorylated ErbB homo- and heterodimers were computed as NAS values to determine whether they correlated with responsiveness to Ab #6 treatment.

The same computational model prepared as described in Example 4 was used. This model was generated based on experimentally determined measurements for the levels of ErbB1, ErbB2, ErbB3, HRG-β1 and BTC. As discussed in Example 4, seven ErbB hetero- and homo-dimers that have been described in the literature were implemented in the model: ErbB1/1, ErbB1/2, ErbB1/3, ErbB1/4, ErbB2/2, ErbB2/3 and ErbB2/4. The majority of these dimers are activated by ligand binding but several arise through a process of “lateral signaling” (or secondary dimerization) in which dimers phosphorylated in a ligand-dependent manner dissociate into monomers that then homo- or hetero-oligomerize with either activated or unactivated monomers to create active dimers. The computational model was trained with a set experimental data that allowed for the identification of the dimers that form in the presence of HRG or BTC using the ADRr cell line.

Thus, integrated levels of phosphorylated homo- and heterodimers were computed as a measure of the Network Activation State (NAS) for the following cell lines: MALME3M, BT474, IGROV1, ADRr, OVCAR8, SKOV3, DU145 and ACHN. As shown in the graph of FIG. 14, the computed levels of phosphorylated ErbB1/3 heterodimers (pErbB1:3) segregated the eight cell lines into Ab #6 non-responders (MALME3M, BT474, IGROV1, ADRr) and responders (OVCAR8, SKOV3, DU145 and ACHN). This segregation based on computed pErbB1:3 levels correlated identically with the predicted non-responders and responders determined using the direct biomarkers as described in Example 8. This segregation based on computed pErbB1:3 levels also correlated almost identically with the predicted non-responders and responders determined using the computed pErbB3 level for the NAS value as described in Example 7, with the only difference being for the BT474 cell line, which was identified as a non-responder using both the direct biomarkers and the computed pErbB1:3 levels but was identified as a responder using the computed pErbB3 levels.

The levels of ErbB1/1, ErbB1/2, ErbB1/3, ErbB1/4, ErbB2/2, ErbB2/3 and ErbB2/4 dimers also were computed using the computational model, but none of the levels of any of these homo- or heterodimers segregated the cell lines into responders and non-responders for Ab #6 treatment. Thus, the results observed with the ErbB1/3 heterodimers were unique among the dimers examined.

These results demonstrate that integrated levels of phosphorylated ErbB homo- or heterodimer can be used as the NAS value in the predictive methods of the invention and, more specifically, that the computed level of pErbB1:3 is a preferred NAS value for predicting responsiveness to treatment with Ab #6. These results are interpreted to mean that Ab #6 is particularly effective in cancers with high levels of ErbB1:3 heterodimers and, therefore, that direct measurement of total ErbB1:3 heterodimers or pErbB1:3 heterodimer levels also can be used as direct biomarkers for predicting the efficacy of Ab #6 treatment.

In this example, the approach described in Example 4 for constructing a mechanistic computational model was used to construct and train a model of the ErbB signaling pathway and, furthermore, to develop a computational representation of the mechanism by which a particular therapeutic agent inhibits the signaling pathway.

The therapeutic agent used in this example is the bispecific antibody H3×B1D2 (the amino acid sequence of which is shown in SEQ ID NO: 41 and which is described further in U.S. Pat. No. 7,332,585, U.S. Pat. No. 7,332,580 and PCT Application PCT/US2006/023479, published as WO 2007/084187 and PCT Application PCT/US2007/024287, published as WO 2008/140493). This bispecific antibody is composed of an anti-ErbB3 single chain antibody linked to an anti-ErbB2 single chain antibody.

The H3×B1D2 agent was predicted to preferentially target ErbB2-overexpressing tumors. Thus, a computational model of the ErbB signaling network in the presence of overexpressed ErbB2 was constructed using the methods and model described in Example 4. The model incorporated interactions between HRG and ErbB1, ErbB2, and ErbB3 receptors, leading to receptor trafficking and intracellular signaling downstream to AKT, producing phosphorylated AKT (pAKT). The included interactions were substantially identical to those found in the model of Example 4. In contrast to the model of Example 4, reactions related to the ligand BTC were not included in this model.

This model was calibrated to match experimental data for the ErbB2-overexpressing breast cell line BT474-M3 (the cell line is described in, e.g., Drummond et al. (2005) Clin. Cancer Res. 11:3392; Park et al. (2002) Clin. Cancer Res. 8:1172; Kirpotin et al. (2006) Cancer Res. 66:6732). Model calibration resulted in only minor differences in parameter values as compared to the model of Example 4. The most significant difference was a reduction in the rates of ErbB receptor internalization and degradation. This change is consistent with known data suggesting that ErbB2 overexpression reduces the internalization/trafficking rates of other ErbB receptors, such as ErbB1 (see e.g., Hendriks et al. (2003) J. Biol. Chem. 278:23343-23351; Wang et al. (1999) Mol. Biol. Cell 10:1621-1636; Haslekas et al. (2005) Mol. Biol. Cell 16:5832-5842).

For training the model, a data set was used that comprised dose-time matrices in which phosphorylation of ErbB1, ErbB2, ErbB3 and AKT at multiple time points and at six different concentrations of HRG stimulation in BT474-M3 cells was measured by ELISA.

For stimulation of the cells, the cells are seeded in duplicate wells with 1000 μl complete media at 150,000 cells per well in 12 well tissue culture plates (for 96 half well ELISA) or in duplicate plate of 100 μl complete media at 20,000 cells per well in 96 well tissue culture plates (for 384 well ELISA). These cells are incubated overnight in a humidified atmosphere of 5% CO2, 95% air and 37 degrees Celsius. Cells are then switched to serum free media: RPMI-1640 media (Gibco) supplemented with, 2 mM L-glutamine (Gibco) and units/mL Pen-Strep (Gibco). Starved cells are incubated in a humidified atmosphere of 5% CO2, 95% air and 37 degrees Celsius for 20-24 hours prior to stimulation. For dose-time matrix studies, cells are stimulated with ligand (HRG) at 0, 1, 2, 3, 4, 5, 7, 10, 20, 30, 60 and 120 minutes. Following stimulation with six different concentrations of HRG (0.098 nM-100 nM) for each time course, cells are placed on ice, washed with cold PBS, then lysed in 200 μl for 12 well plates and 45 μl for 96 well plates in cold M-PER Mammalian Protein Extraction Buffer (Thermo Scientific, Catalog #78501) supplemented with protease inhibitor cocktail (Sigma-Aldrich, P2714), 1 mM sodium orthovanadate (Sigma-Aldrich, S6508), 5 mM sodium pyrophosphate (Sigma-Aldrich, 221368), 50 μM oxophenylarsine (EMD Biosciences, 521000) and 10 μM bpV(phen) (EMD Biosciences, 203695).

Levels of protein phosphorylation in the stimulated cells are measured by ELISA. ErbB1, ErbB2, and ErbB3 are measured using R&D Systems Duoset IC kits (ErbB1 DYC1095-E, ErbB2 DYC1768-E, ErbB3 DYC1769-E). Capture antibodies against ErbB1 (R&D Systems, 841402), ErbB2 (R&D Systems, 841425), ErbB3 (R&D Systems, 841428) and AKT (Upstate, 05-591MG) are incubated in 96 half well plates (Greiner, Catalog #82050-046) or 384 well plates (Nunc Cat # 40518) that are black flat-bottom polystyrene high-binding plates overnight at room temperature. The ELISA plates are blocked with 2% bovine serum albumin (BSA) and phosphate buffered saline (PBS) for one hour then incubated with lysates diluted in 2% BSA, 0.1% Tween-20 and PBS for two hours at room temp. In between each incubation, the plates are washed three times with 0.05% Tween-20 in PBS. ELISAs for measuring phospho-ErbB1, -ErbB2 and -ErbB3 are incubated with Anti-Phospho-Tyrosine-HRP Detection antibody (R&D Systems, 841403) for two hours. ELISAs measuring phospho-AKT are incubated with primary serine 473 specific anti-phospho AKT mouse monoclonal antibody (Cell Signaling Technologies, Catalog #5102) for 2 hours, then incubated with Streptavidin-HRP (R&D Systems, Catalog #DY998,) for 20 minutes. All ELISAs are visualized with SuperSignal ELISA Pico Chemiluminescent Substrate (Pierce, Catalog #37069) and luminescent signal is measured using a luminometer.

As shown in FIG. 21, the model matched the HRG-induced pErbB3 signaling data in the ErbB2-overexpres sing cell line BT474-M3 at all doses of HRG examined experimentally. Additionally, the model matched the HRG-induced pAKT signaling data in BT474-M3 cells at HRG doses of approximately 5 nM and lower.

Next, a computational representation of the mechanism by which H3×B1D2 inhibits HRG-dependent signaling of the ErbB pathway was developed. The computational representation of the inhibitor was constructed using mass-action reaction equations that describe the binding of inhibitor to ErbB2 and ErbB3 and subsequent inhibition of HRG-induced signaling. Parameters for the binding events were obtained by a combination of direct measurement (using techniques widely known in the art) and computational training of the model to match data for the inhibition of HRG-induced pErbB3 in cells by H3×B1D2. In particular, the on-rate and off-rate for binding of the H3 single chain arm of the bispecific antibody to ErbB3 and the on-rate and off-rate for the binding of the B1D2 single chain arm of the bispecific antibody to ErbB2 were experimentally determined by standard BIACore and KinExA technology. Reactions and parameters for the computational model of H3×B1D2 appear in Tables 16a and 16b.

TABLE 16a
H3 x B1D2 implementation
ReactionForwardReverse
NumberReactionparameterparameter
H3 x B1D2_v1E2 + (H3 x B1D2) <->h3xb1d2_kf1h3xb1d2_kr1
E2:(H3 x B1D2)
H3 x B1D2_v2(H3 x B1D2) + E3 <-> h3xb1d2_kf2h3xb1d2_kr2
(H3 x B1D2):E3
H3 x B1D2_v3E2 + (H3 x B1D2): h3xb1d2_kf3 h3xb1d2_kr1
E3 <-> E2:(H3 x
B1D2):E3
H3 x B1D2_v4 E2:(H3 x B1D2) + h3xb1d2_kf4h3xb1d2_kr2
E3 <-> E2:(H3 x
B1D2):E3
H3 x B1D2_v5 E2:(H3 x B1D2) + h3xb1d2_kf5
E2:(H3 x B1D2) ->
E2:E2
H3 x B1D2_v6E2:(H3 x B1D2) + h3xb1d2_kf5
E2 -> E2:E2
H3 x B1D2_v7 E2:(H3 x B1D2) + h3xb1d2_kf5
E2_p -> E2:E2_p
H3 x B1D2_v8 E2:(H3 x B1D2) + h3xb1d2_kf6
E1 -> E1:E2
H3 x B1D2_v9 E2:(H3 x B1D2) + h3xb1d2_kf6
E3 -> E2:E3
H3 x B1D2_v10E2 + (H3 x B1D2):h3xb1d2_kf6
E3 -> E2:E3
H3 x B1D2_v11E2:(H3 x B1D2) + h3xb1d2_kf7
E3:HRG -> E2:E3:
HRG
H3 x B1D2_v12E2:(H3 x B1D2) +h3xb1d2_kf7
E3:HRG_p <->
E2:E3:HRG_p
H3 x B1D2_v13E1 + (H3 x B1D2):h3xb1d2_kf8
E3 -> E1:E3
**In this reaction scheme, the amount of free (H3 x B1D2) is held constant.
TABLE 16B
H3 x B1D2 parameters
NameValueUnits
h3xb1d2_kf13.13E+04(mol/L)−1 sec−1
h3xb1d2_kr11.50E−04sec−1
h3xb1d2_kf23.50E+05(mol/L)−1 sec−1
h3xb1d2_kr22.20E−02sec−1
h3xb1d2_kf31.08E−06(molecules/cell)−1 sec−1
h3xb1d2_kf41.20E−05(molecules/cell)−1 sec−1
h3xb1d2_kf51.67E−08 (molecules/cell)−1 sec−1
h3xb1d2_kf65.00E−08 (molecules/cell)−1 sec−1
h3xb1d2_kf75.00E−07(molecules/cell)−1 sec−1
h3xb1d2_kf85.00E−08(molecules/cell)−1 sec−1

The ErbB signaling model was combined with the inhibitor model and used to predict experimental data for the inhibition of pErbB3 and pAKT signaling by H3×B1D2. The experimental data was generated using the same methodology as described above, except that concentrations of H3×B1D2 ranging from 15 pM to 1 μM were added during serum starvation. Additionally, inhibition data was generated using a ten-minute lysis timepoint after 5 nM HRG stimulation. The model successfully recapitulated experimental results showing that the IC50 of inhibition for H3×B1D2 did not greatly change in different cell lines, but the percentage of inhibition did shift greatly. A primary cause of the change of percentage inhibition was the level of expression of ErbB2. This was demonstrated in experiments where ErbB2 was transfected into the OVCAR8 cell line to create an ErbB2-overexpressing cell line (referred to as OVCAR8-HER2). The pErbB3 and pAKT inhibition curves for OVCAR8 cells treated with H3×B1D2 were compared to the same inhibition curves for OVCAR8-HER2 cells treated with H3×B1D2. The results are shown in FIGS. 17A-D, wherein FIGS. 17A and 17B show the inhibition curves for pErbB3 and pAKT, respectively, in OVCAR8 cells treated with H3×B1D2 (either experimentally or simulated in the model) and FIGS. 17C and 17D show the inhibition curves for pErbB3 and pAKT, respectively, in OVCAR8-HER2 cells treated with H3×B1D2 (either experimentally or simulated in the model). The IC50s for the experimentally treated cells (“DR50 data”) and the simulated treated cells (“DR50sim”) are also shown. The data show that higher levels of ErbB2 expression result in greater inhibition percentage by H3×B1D2 treatment.

In addition to the primary role of ErbB2 in modulating percentage inhibition, an unexpected role for ErbB1 was revealed by the computational model. This role for ErbB1 was exemplified by a simulation showing the effect of adding ErbB1 RNAi to ADRr cells, to simulate downregulation of ErbB1. ADRr cells express only low levels of ErbB2 and exhibited a poor percentage of inhibition by H3×B1D2 in both the computational model and in experimentally determined data. This data is shown in FIGS. 18A and 18B, which show the inhibition curves for pErbB3 and pAKT, respectively, in ADRr cells treated with H3×B1D2, either experimentally or simulated in the model. The IC50s for the experimentally treated cells (“DR50 data”) and the simulated treated cells (“DR50sim”) are also shown. However, downregulation of ErbB1 expression (by simulation of RNAi addition) resulted in a greater inhibition percentage in the simulation, as shown in FIGS. 18C and 18D, which show the inhibition curves for pErbB3 and pAKT, respectively, in ADRr cells simulated for treatment with ErbB1 RNA1 and H3×B1D2. The implication of the results from the ErbB1 RNAi simulation is that ErbB1 expression is a negative response biomarker for H3×B1D2.

In summary, the computational model and the experimental data indicated that there are two mechanisms for negatively modulating responsiveness to H3×B1D2 in vitro: (i) an insufficiently high ErbB2 level; and (ii) high ErbB1 levels. Conversely, high levels of ErbB2 expression and low levels of ErbB1 expression correlated with increased responsiveness to H3×B1D2.

In this example, the in vivo responsiveness of tumors to H3×B1D2 treatment was correlated with the computed levels of various components in the ErbB pathway to identify direct and indirect biomarkers for responsiveness to H3×B1D2 treatment.

To characterize the in vivo response of tumors to H3×B1D2 treatment, a panel of tumor cell lines was tested in a xenograft tumor model such as described in Example 1. In the xenograft tumor models, mice (nu/nu mice: 4-5 week old female mice, athymic, nude, outbred background; Albino; purchased from Charles River Labs, Wilmington, Mass.) are implanted in the flank with 5×106-2×107 cells/mouse (depending on cell line) in 200 via subcutaneous injection. Mice are monitored for initial tumor growth. Tumors are allowed to grow for several days until the mean tumor volume is approximately 150-200 mm3. The tumor volume is calculated as V=(π/6(L×W2). The mice are treated with the H3×B1D2 antibody at a dosage of 600 μg/injection every 3 days (q3d). Control mice are treated with phosphate buffered saline (PBS) or with wild-type HSA (human serum albumin). Tumor volume is measured for 40-80 days.

The following twelve tumor cell lines were examined: ACHN, ADRr, IGROV1, LS180, MIA PaCa2, ZR75-1, MDA-MB-361, ADrR-HER2 (ADrR cells transfected to overexpress HER2), NCI-N87, CALU-3, SKOV-3 and BT474-M3. For use of these cell lines in the computational model, the levels of ErbB1, ErbB2 and ErbB3 expression in each cell line was experimentally determined using the methods described above or minor variations thereof, the results of which are shown below in Table 17:

TABLE 17
ErbB ReceptorLevels
Cell lineErbB1ErbB2ErbB3
ACHN4482844545615200
ADrR1778184079233205
BT474-M3129436170660149238
SKOV3264132137766113694
ZR75-13740919913239492
IGROV11490311584185355
OVCAR82361575327231813
MDA-MB-3616585537173132981
NCl-N87417753123347934678
Calu-3161357119697630031
LS18012252014333928841
MIAPaCa-2138563848655735
ADrR-HER227100072200034400

The control and treatment data from the in vivo xenograft experiments was fitted to exponential growth curves, using the following formula: V=Vo*exp(k*t) wherein V is the tumor volume, Vo is the tumor volume at time zero, k is the exponential growth rate and t is the time. The potency of H3×B1D2 in inhibiting tumor growth was represented as the ratio of treatment and control exponential growth rates for each cell line tested. This ratio was denoted “relative growth rate” (RGR) and is represented as:

RGR=kH3×B1D2/kcontrol

A relative growth rate (RGR) of 1 meant the agent had no effect. An RGR of 0 meant that the agent halted tumor growth entirely. A negative RGR meant that the agent caused tumor regression.

Of the twelve cell lines examined, only BT474-M3 had a negative RGR (i.e., H3×B1D2 caused tumor regression only in this cell line). Two cell lines, IGROV1 and LS180, had RGR values greater than 1, indicating H3×B1D2 had no effect on tumor growth for these cell lines. The remaining nine cell lines had RGR values between 0 and 1, indicating that H3×B1D2 partially inhibited tumor growth for these cell lines.

The RGR values for the twelve cell lines were plotted against model-computed levels of ErbB2 monomer, ErbB2:ErbB2 homodimer and ErbB2:ErbB3 heterodimer in the cell lines in the absence of the H3×B1D2 inhibitor, based on the measured levels of ErbB1, ErbB2 and ErbB3 in each of the cell lines. The results are illustrated in FIGS. 19A-C, which show graphs of the in vivo-determined relative growth rates (RGR) for the panel of tumor cells in the xenograft model treated with H3×B1D2 plotted against the computed levels of ErbB2 monomers (FIG. 19A), ErbB2:ErbB2 homodimers (FIG. 19B) and ErbB2:ErbB3 heterodimers (FIG. 19C) in the panel of tumor cells in the absence of H3×B1D2.

The results in FIG. 19 show that there is a linear relationship between RGR and computed levels of ErbB2 monomers, ErbB2:ErbB2 homodimers and ErbB2:ErbB3 heterodimers. In view of this observation, direct measurement of ErbB2 monomers, ErbB2:ErbB2 homodimers and/or ErbB2:ErbB3 heterodimers can be used as direct biomarkers of H3×B1D2 responsiveness. Furthermore, measurement of ErbB1, ErbB2 and ErbB3 in a tumor sample can be used to compute the levels of ErbB2:ErbB2 homodimers and/or ErbB2:ErbB3 heterodimers to stratify tumor responsiveness to H3×B1D2 treatment (i.e., measured levels of ErbB1, ErbB2 and ErbB3 can be used as indirect biomarkers of H3×B1D2 responsiveness, which are used to compute levels of ErbB2:ErbB2 homodimers and/or ErbB2:ErbB3 heterodimers). Thus, computed levels of, for example, ErbB2:ErbB2 homodimers and/or ErbB2:ErbB3 heterodimers can be used as a Network Activation State (NAS) value upon which responsiveness to H3×B1D2 treatment can be predicted.

The RGR values for the twelve cell lines also were plotted against the computed relative levels (e.g., ratio) of ErbB2:ErbB2 heterodimer and ErbB1:ErbB3 heterodimer in the cell lines in the simulated absence and presence of the H3×B1D2 inhibitor. A lower relative level indicates that H3×B1D2 can more potently inhibit formation of that species of heterodimer. The results are illustrated in FIGS. 20A-B, which show graphs of the in vivo-determined relative growth rates (RGR) for a panel of tumor cells in a xenograft model treated with H3×B1D2 plotted against the computed relative levels of ErbB2:ErbB3 heterodimers (FIG. 20A) and ErbB1:ErbB3 heterodimers (FIG. 20B) in the panel of tumor cells in the simulated presence of H3×B1D2, as compared to the simulated levels of the heterodimers in the simulated absence of H3×B1D2.

The results in FIG. 20 demonstrate a correlation between the ability of the inhibitor to disrupt ErbB2:ErbB3 and ErbB1:ErbB3 heterodimers and the relative growth rate of the tumors. That is, tumor cells (e.g., BT474-M3 cells) in which the computed relative levels of ErbB2:ErbB3 and ErbB1:ErbB3 heterodimer are low in the simulated presence of the H3×B1D2 inhibitor (i.e., heterodimer disruption is high by the inhibitor) exhibit lower RGR values (indicating a greater effect of the inhibitor on tumor growth). In contrast, tumor cells (e.g., ACHN cells) in which the computed relative levels of ErbB2:ErbB3 and ErbB1:ErbB3 heterodimer are high in the simulated presence of the H3×B1D2 inhibitor (i.e., heterodimer disruption is low by the inhibitor) exhibit higher RGR values (indicating less of an effect of the inhibitor on tumor growth). These results demonstrate that simulating the presence of the therapeutic agent in the computational model of the signaling pathway, as compared to the simulated absence of the therapeutic agent, allows for generation of a Network Inhibition State (NIS), based on the relative levels of ErbB2:ErbB3 or ErbB1:ErbB3 heterodimer, which NIS can be used as a predictor of the responsiveness of tumor cells to the therapeutic agent in vivo.

The dissociation constants of anti-ErbB antibodies may be measured using either or both of two independent techniques, a Surface Plasmon Resonance Assay and a cell binding assay.

The Surface Plasmon Resonance Assay is performed as described in Wassaf et al. (2006) Analytical Biochem., 351:241-253. A preferred implementation uses a BIACORE 3000 instrument (GE Healthcare) using a recombinant ErbB protein as the analyte and the anti-ErbB antibody as the ligand The KD value is calculated based on the formula KD=Kd/Ka.

A cell binding assay is performed using A-431 cells for ErbB1 binding, ZR-75-1 cells for ErbB2 binding or MALME-3M cells for ErbB3 binding (all from ATCC). The assay is performed substantially as follows.

Cells are detached with 2 mLs trypsin-EDTA+2 mLs RMPI+5 mM EDTA at room temperature for 5 minutes. Complete RPMI (10 mLs) is added immediately to the trypsinized cells, resuspended gently and spun down in a Beckman tabletop centrifuge at 1100 rpm for 5 minutes. Cells are resuspended in BD stain buffer (PBS+2% FBS+0.1% sodium azide, Becton Dickinson) at a concentration of 2×106 cells per ml and 50 μl (1×105 cells) aliquots are plated in a 96-well titer plate.

A 150 μl solution of 200 nM anti-ErbB antibody in BD stain buffer is prepared and serially diluted 2-fold into 75 μl BD stain buffer. The concentrations of the diluted antibody ranged from 200 nM to 0.4 nM. 50 μl aliquots of the different protein dilutions are then added directly to the 50 ul cell suspension giving the final concentrations of 100 nM, 50 nM, 25 nM, 12 nM, 6 nM, 3 nM, 1.5 nM, 0.8 nM, 0.4 nM and 0.2 nM of the antibody.

Aliquoted cells in the 96-well plate are incubated with the protein dilutions for 30 minutes at room temperature on a platform shaker and washed 3 times with 300 μl BD stain buffer. Cells are then incubated with 100 μl of secondary antibody (e.g., a 1:750 dilution of Alexa 647-labeled goat anti-human IgG in BD stain buffer) for 45 minutes on a platform shaker in the cold room. Finally, cells are washed twice, pelleted and resuspended in 250 μl BD stain buffer+0.5 μg/ml propidium iodide. Analysis of 10,000 cells is done in a FACSCALIBUR flow cytometer using the FL4 channel. MFI values and the corresponding concentrations of the anti-ErbB-antibody are plotted on the y-axis and x-axis, respectively. The KD of the molecule is determined using GraphPad PRISM software using the one-site binding model for a non-linear regression curve.

The KD value is calculated based on the formula Y=Bmax*X/KD+X(Bmax=fluorescence at saturation. X=antibody concentration. Y=degree of binding).

METHODS AND SYSTEMS FOR PREDICTING RESPONSE OF CELLS TO A THERAPEUTIC AGENT (2024)
Top Articles
Latest Posts
Recommended Articles
Article information

Author: Twana Towne Ret

Last Updated:

Views: 5513

Rating: 4.3 / 5 (64 voted)

Reviews: 95% of readers found this page helpful

Author information

Name: Twana Towne Ret

Birthday: 1994-03-19

Address: Apt. 990 97439 Corwin Motorway, Port Eliseoburgh, NM 99144-2618

Phone: +5958753152963

Job: National Specialist

Hobby: Kayaking, Photography, Skydiving, Embroidery, Leather crafting, Orienteering, Cooking

Introduction: My name is Twana Towne Ret, I am a famous, talented, joyous, perfect, powerful, inquisitive, lovely person who loves writing and wants to share my knowledge and understanding with you.