Network-informed discovery of multidrug combinations for ERα+/HER2-/PI3Kα-mutant breast cancer

Breast cancer is a persistent threat to women worldwide. A large proportion of breast cancers are dependent on the estrogen receptor α (ERα) for tumor progression. Therefore, targeting ERα with antagonists, such as tamoxifen, or estrogen deprivation by aromatase inhibitors remain standard therapies for ERα + breast cancer. The clinical benefits of monotherapy are often counterbalanced by off-target toxicity and development of resistance. Combinations of more than two drugs might be of great therapeutic value to prevent resistance, and to reduce doses, and hence, decrease toxicity. We mined data from the literature and public repositories to construct a network of potential drug targets for synergistic multidrug combinations. With 9 drugs, we performed a phenotypic combinatorial screen with ERα + breast cancer cell lines. We identified two optimized low-dose combinations of 3 and 4 drugs of high therapeutic relevance to the frequent ERα + /HER2-/PI3Kα-mutant subtype of breast cancer. The 3-drug combination targets ERα in combination with PI3Kα and cyclin-dependent kinase inhibitor 1 (p21). In addition, the 4-drug combination contains an inhibitor for poly (ADP-ribose) polymerase 1 (PARP1), which showed benefits in long-term treatments. Moreover, we validated the efficacy of the combinations in tamoxifen-resistant cell lines, patient-derived organoids, and xenograft experiments. Thus, we propose multidrug combinations that have the potential to overcome the standard issues of current monotherapies. Supplementary Information The online version contains supplementary material available at 10.1007/s00018-023-04730-x.


Description of Tables S1 to S3 (available as separate files):
Table S1 Details and doses of the anticancer drugs used throughout the study. The Microsoft Excel workbook contains information about the commercial source of the drugs, biological targets, drug development phase, physical properties, stock concentrations, and doses used for the TGMO-based screens, molecular assays, and xenograft experiments.

Table S2
A list of predicted synergistic 2-, 3-, and 4-drug combinations from "Round 2" of the TGMO-based screen. The Microsoft Excel workbook contains the list of candidate drug combinations for MCF7, MCF7-V, and MCF7/LCC2 cells, and their potential safety profile in MCF10A cells. Combinations of potential therapeutic relevance were selected for "Round 3" assays.

Table S3
Status of PIK3CA gene mutations in different breast cancer cell lines. The Microsoft Excel workbook contains the status of the tested mutations of the PIK3CA gene in exons 9 and 20 in the different breast cancer cell lines used in the study.  Description of the Data files S1 to S5 (available as separate files): Data S1 ERa PPI network. The Cytoscape file contains the network of primary PPIs of ERa with 322 protein interactors and secondary PPIs among them. Based on the analysis of network centrality measures, the top 50 interactors are highlighted in green.

Data S2
List of ERa interactors. The Microsoft Excel workbook contains the list of 322 ERa interactors that were given the feature "DP". The file was imported to Cytoscape to be integrated into the network of the human interactome.
Data S3 Analysis of ERa PPI network centrality measures. The Microsoft Excel workbook of the results of the network analysis done by Cytoscape on the ERa PPI. Results include coefficients for network centrality parameters, such as "Number of directed edges", "Radiality", "Stress", and others.

Data S4
Genes co-expressed with ERa in ERa+ breast cancer datasets. The Microsoft Excel workbook contains the co-expression coefficients obtained from datasets available from the online databases Oncomine (www.oncomine.org) and GOBO (http://co.bmc.lu.se/gobo).
Data S5 Integrated network of SL between interactors of ERa, co-expressed, and frequently mutated genes in ERa+ breast cancer. The Cytoscape file contains two networks; one represents SL between genes in breast cancer, and another network derived from the first one that contains only those genes that are ERa interactors, coexpressed, or frequently mutated genes in ERa+ breast cancer.

Fig. S1
Frequently mutated genes and SL network in ERα+ breast cancer, and crosstalk between potential drug targets. of SL relationships in breast cancer. ERa interactors, genes co-expressed with ERa, and frequently mutated genes in ERa+ breast cancer are highlighted with color codes. Nodes represent genes and edges represent SL relationships between the connected genes. (D) Schematic illustration of relevant signaling pathways and the network of signaling crosstalk between them. Solid lines indicate a direct effect, whereas dotted lines indicate an indirect effect. The 9 drugs selected for "round 1" of the TGMO-based screen and their primary targets are included in the scheme. The scheme was created with Biorender.com based on information from the literature and the KEGG PATHWAY database.

Fig. S2
Dose-response curves of the selected 9 drugs in cancer and non-cancer cell lines. (A-I) Dose-response curves of the 9 selected drugs for "round 1" of the TGMObased screen. The indicated cancer (blue and green) and non-cancer (orange) cell lines were treated with increasing doses of each drug and % ATP levels were calculated as the response. For each curve, a control group treated with vehicle was set to 100%. Data are represented as means ± SD (n = at least 3 independent samples).  (DMSO) control in white, each drug was given at two dose levels, higher dose (dark gray) and lower dose (light gray). Values of the DMSO only controls were set to 100%.

Fig. S4
Regression modeling of the results of "Round 1" of the in vitro TGMO-based screen. (A,B) Estimated regression coefficients obtained by second-order linear regression of the calculated % ATP levels from "round 1" in MCF7-V (panel A) and MCF7/LCC2 (panel B) cells. Blue bars indicate regression coefficients of cancer cells, whereas red bars are show the TW as compared to MCF10A cells. Drugs highlighted in red in the x-axis labels represent drugs or combinations with unfavorable efficacy or toxicity profiles, which were therefore eliminated from "round 2". Statistical significance is shown as * for p ≤ 0.05, ** for p ≤ 0.01, *** for p ≤ 0.001, and "ns" for non-significant results. Data are represented as means ± standard error of the means (SEM) of 3 independent experiments, each containing 3 independent replicates.      , UC2288 (UC2), C188-9 (C18), and OHT, tested alone or in combinations. Red bars show results obtained with MCF10A cells. Note that the same values obtained with MCF10A cells were used in all three panels (A-C) for comparison. The data are presented as means ± SEM of 3 independent experiments, each with 3 independent replicates. The matrices below the graphs represent the scheme of the combinatorial design, in addition to the vehicle (DMSO) control in white, each drug was given at two dose levels, higher dose (dark gray) and lower dose (light gray). Values of the DMSO only controls were set to 100%. CIs calculated by CompuSyn are illustrated below the combinatorial matrices as a red color gradient. Low CI < 0.8 (light red) indicates synergistic combinations, whereas high CI > 1 (dark red) indicates antagonistic combinations. MT, monotherapy; Veh., DMSO control.

Fig. S6
Regression modeling of the results of "Round 2" of the in vitro TGMO-based screen. (A-E) Estimated regression coefficients obtained by second-order linear regression (panels A to C) or third-order linear regression (panels D and E) of the calculated % ATP levels from "round 2" in MCF7 (panel A), MCF7-V (panels B and D), and MCF7/LCC2 (panels C and E) cells. Blue bars indicate regression coefficients of cancer cells (panels A to E), whereas red bars show the TW as compared to MCF10A cells (panels A to C). For panel D, combinations highlighted in green in the x-axis labels have favorable efficacy profiles. Statistical significance is shown as * for p ≤ 0.05, ** for p ≤ 0.01, *** for p ≤ 0.001, and "ns" for non-significant results. Data are represented as means ± SEM of 3 independent experiments, each with 3 independent replicates.  Dose-response curves of the 7 selected drugs from "round 2" to be tested in different combinations in "round 3". The indicated breast cancer cell lines include ERa+ (blue), ERa+ but tamoxifen-tolerant or -resistant (green), and ERa-(red) cell lines. Cells were treated with increasing doses of each drug and % ATP levels were calculated as the response. For each curve, a control group treated with vehicle was set to 100%. Data are represented as means ± SD (n = at least 3 independent samples). Note that the same data (curves) obtained with MCF7, MCF7-V, and MCF7/LCC2 cells were used as in Fig. S2, and plotted here for comparison. whereas high CI > 1 (dark red) indicates antagonistic combinations. Combinations were tested at lower (panels C, E, and G) and higher dose levels (panels D, F, and H).
Red arrows indicate combinations with higher efficacy and synergy in ERa+ breast cancer cell lines and less toxicity in non-cancerous cells.  were normalized to the activities of the internal transfection standard Renilla luciferase. Cells were treated with alpelisib (Alp), UC2288 (UC2), and OHT, tested alone or in the 3-drug combination (red bars). Luciferase activities of the vehicle-treated control groups were set to 1. (D) Luciferase reporter assays for exogenously expressed ERα with HEK 293T cells. Cells were transiently co-transfected with the ERE-Luc reporter and either empty vector pSG5 (EV) or pHEG0 (ERα). Drug treatments were as indicated and mentioned in panels B and C. Relative luciferase activities (RLU) were normalized to the activities of the internal transfection standard Renilla luciferase. Luciferase activities of the control cells transfected with EV and treated with DMSO were set to 1. (E) Immunoblots of the indicated proteins tested in total cell lysates of MCF7-V cell line. Cells were treated for 24 hours with either alpelisib (Alp), UC2288 (UC2), OHT, or talazoparib (Tala) tested alone, or the 3-drug combination (Alp + UC2 + OHT), or the 4-drug combination (Alp + UC2 + OHT + Tala). EGF (10 ng/ml) was added to cells 15 min before harvesting to induce MAPK signaling. GAPDH was used as an internal standard. (F,G) Luciferase reporter assays for the transcriptional activities of endogenous HIF1a activity in T47D (panel F) and MCF7/TamR (panel G) cells. Cells were transiently transfected with the HRE-Luc reporter plasmids. Relative luciferase activities were calculated and cells were treated as in panel B and C. For bar graphs, data are represented as means ± SD (n = 4 independent samples for panels B and C, and n = at least 3 for panels D, F, and G). Statistical significance between the groups were analyzed by one-way ANOVA and p-values ≤ 0.05 were considered statistically significant. Drug doses are listed in Table S1. Bar graph of the % ATP levels of patient-derived mammary tumor organoid (model HUB 056) treated with alpelisib (Alp), UC2288 (UC2), OHT, and talazoparib (Tala) alone or in the combination of 3 drugs (Alp + UC2 + OHT) or 4 drugs (Alp + UC2 + OHT + Tala). ATP levels were tested after one week of treatment. The mean value of the DMSO control was set to 100%. Data are represented as means ± SD (n = at least 7 independent samples). For panels B and C, statistical significance between the groups were analyzed by unpaired student's t-test and p-values ≤ 0.05 were considered statistically significant. (D,E) Line graphs of the % change in tumor volume (panel A) and % change in body weight (panel B) of MCF7 xenografts in Balb/c nude mice treated as indicated and as in Table S1. Data points represent the means and error bars are SEMs. (n = 5 mice / group at the start of the treatments). In the group