Drug synergy screen and network modeling in dedifferentiated liposarcoma identifies CDK4 and IGF1R as synergistic drug targets

Abstract

Dedifferentiated liposarcoma (DDLS) is a rare but aggressive cancer with high recurrence and low response rates to targeted therapies. Increasing treatment efficacy may require combinations of targeted agents that counteract the effects of multiple abnormalities. To identify a possible multicomponent therapy, we performed a combinatorial drug screen in a DDLS-derived cell line and identified cyclin-dependent kinase 4 (CDK4) and insulin-like growth factor 1 receptor (IGF1R) as synergistic drug targets. We measured the phosphorylation of multiple proteins and cell viability in response to systematic drug combinations and derived computational models of the signaling network. These models predict that the observed synergy in reducing cell viability with CDK4 and IGF1R inhibitors depends on the activity of the AKT pathway. Experiments confirmed that combined inhibition of CDK4 and IGF1R cooperatively suppresses the activation of proteins within the AKT pathway. Consistent with these findings, synergistic reductions in cell viability were also found when combining CDK4 inhibition with inhibition of either AKT or epidermal growth factor receptor (EGFR), another receptor similar to IGF1R that activates AKT. Thus, network models derived from context-specific proteomic measurements of systematically perturbed cancer cells may reveal cancer-specific signaling mechanisms and aid in the design of effective combination therapies.

Figure 1

Drug combination screen identifies synergistic and antagonistic drug targets in the dedifferentiated liposarcoma cell line, DDLS8817. (A) Example of CI score calculation of paired drug perturbation using the MEK inhibitor, SL327 (MEKi), and the AKT inhibitor, AKT inhibitor VIII (AKTi). Cell viability was estimated after 72 hours of drug treatment by the Resazurin assay that measures cellular metabolic activity. Error bars represent standard deviation of at least four biological replicates. (B) CI scores for a drug synergy screen performed in DDLS8817 cells using 14 targeted inhibitors (“Inh” or “i” in the labels). CI scores were derived as described in (A) and categorized as synergistic (<0.75, red), antagonistic (>1.5, blue), or additive (green). In some cases CI scores could not be calculated (gray). Inset shows distribution of CI scores. A complete list of targets and secondary targets of the drugs used are provided in table S1. Each CI score represents data from seven different drug doses of single and paired drug treatments with at least four biological replicates per condition.

Figure 2

IGF1R and CDK4 are synergistic drug targets in DDLS. (A) Dose-response effects of the CDK4 inhibitor PD0332991 and the IGF1R antibody R1507 on inhibition of cell viability in DDLS8817 cells estimated after six days of drug treatment using the CCK-8 assay that measures cell metabolic activity (upper panel). Each condition represents at least four biological replicates. CI scores were calculated at various effects (lower panel). (B) Performed as in (A) in the DDLS cell line LPS141.

Figure 3

Systematic drug treatments and large-scale proteomic profiling in DDLS8817. (A) Experimental design of 14 individual and 91 pairwise drug perturbations with a set of targeted small molecule drugs. (B) Response of 13 proteins or phosphoproteins [phosphorylated at the indicated residue(s)] after 24 hours of drug treatment as shown in (A) assessed by RPPA. The effect on cell viability was estimated with the resazurin assay (measures cell metabolic activity) after 72 hours of drug treament. All read-outs were z-score normalized. Each condition represents the mean of two biological replicates.

Figure 4

Network inference of proteomic data profiles and prior knowledge interactions provides signaling models specific to DDLS. (A) A network of prior knowledge interactions (edges) between the selected set of protein read-outs (nodes) measured by RPPA and their connection to cell viability, as measured by metabolic activity assays in Fig. 3B. (B) Inferred network models from perturbation-response data and prior knowledge information with the line width reflecting the most probable interactions. The network represents the average network of the 100 lowest error models. Predicted interactions (gray) and prior knowledge interactions that are retained (blue) and rejected are indicated. Nodes that are perturbed but not observed are termed “activity node” and preceded by an “a” (for example, aAKT) and represent the presumed, direct target of the drugs applied.

Figure 5

Synergistic effects are captured in the network models, and the experimentally observed CDK4-IGF1R drug synergy is recapitulated. (A) Example of calculation of model-based synergy scores (S) by in silico perturbation of the 4EBP1_pSer⁶⁵ (phospholyrated at Serine 65) and ERK encoded as an activity node (aERK, external drug node). These nodes were inhibited with eight different perturbation strengths (u) in all possible combinations and the effect was recorded as the response on the cell viability node (z-score). (B), Non-additive synergy effects (Synergy of model) were determined as the difference between the effect of the paired inhibition and the added effects of the two single node inhibitions (S=−0.24 for aERK and 4EBP1_pSer⁶⁵). (C), Computed synergy scores for all node-pairs where each synergy scores was determined from 64 unique perturbations as in (A). Synergy scores were categorized into three classes, where S<−0.20 was considered synergistic (red), >0.20 antagonistic (blue), and otherwise additive (green). The synergy score for IGF1R and CDK4 node inhibition is highlighted (S=−0.23).

Figure 6

Simulation of information flow in network models predicts several important interactions mediating the synergy of CDK4 and IGF1R inhibition, including AKT pathway members. Network edges ranked by their contribution to the model-simulated synergy between CDK4 and IGF1R inhibitors. Each edge was removed in turn and the effect on the cell viability synergy score was recalculated and expressed as the percent suppression of original synergy score. The leave-edge-out calculations were performed using the 100 lowest error models.

Figure 7

The AKT pathway is likely involved in the synergy of CDK4 and IGF1R inhibitors. (A) Western blot of DDLS8817 and LPS141 cells inhibited for 12 hours with 10 μg/mL R1507 (IGF1R antibody), 1uM PD0332991 (CDK4 inhibitor), and siRNA-mediated knockdown of CDK4. (B) Similar to (A), except cells were inhibited for 24 hours with 1 μM NVP-AEW541 (IGF1R inhibitor) and PD0332991. Western blots shown are representative data of at least two independent experiments. (C) Dose-response measurements of cell metabolic activity using the CCK-8 assay (correlates with cell viability) of DDLS8817 cells after drug treatment for 6 days with the AKT inhibitor MK2206 and the CDK4 inhibitor PD0332991. The combination index (CI) score is indicated and was determined at EC50 levels indicated by dashed lines. Error bars represent standard deviation of six biological replicates. (D) Similar to (C) but in LPS141 cells.