All experiments were conducted using six cell line models: two TNBC cell lines (MDA-MB-231 and MDA-MB-468) and two HER2+ BC cell lines (BT474 and SKBR3), either sensitive (Parental, P) or resistant to lapatinib + trastuzumab (LTR). MDA-MB-231 and MDA-MB-468 cell lines were purchased from Baylor College of Medicine Tissue and Cell Culture Core Laboratory. MDA-MB-231 cells were maintained at 37°C and 5% CO₂ in Dulbecco’s modified Eagle medium (DMEM) and supplemented with 10% heat-inactivated fetal bovine serum (HI-FBS) and 1% Penicillin-Streptomycin-Glutamine (PSG). MDA-MB-468 cells were maintained in Leibovitz’s L-15 Medium (in free gas exchange with atmospheric air) supplemented with 10% HI-FBS and 1% PSG. The BT474 cell line was obtained from AstraZeneca (Cheshire, United Kingdom) (Arpino et al., 2007), and maintained in DMEM supplemented with 10% HI-FBS and 1% PSG. SKBR3 cells were obtained from Dr. Joe Gray’s lab (Berkeley Lab, Berkeley, CA, United States) and were grown in McCoy’s 5A supplemented with 10% HI-FBS and 1% PSG (Huang et al., 2011; Wang et al., 2011). Cell lines resistant to HER-targeted therapy (Lapatinib + Trastuzumab) (LTR) were obtained from Dr. Rachel Schiff’s lab. These cells were generated by long-term culture of the cells in their original media with increasing concentrations of trastuzumab (1–50 μg/ml) and lapatinib (0.1–1 μM) as described before (Huang et al., 2011; Wang et al., 2011). MCF10A cell line was maintained at 37°C and 5% CO₂ using MEGM™ Mammary Epithelial Cell Growth Medium BulletKit™ with 100 ng/ml cholera toxin. SUM159 cell line was maintained in DMEM supplemented with 5% HI-FBS and 1% PSG.

Trastuzumab (Herceptin^®, manufactured by Genentech, San Francisco, CA, United States) was purchased from McKesson and was dissolved in sterile, distilled water provided with the drug. Lapatinib was obtained from LC Laboratories (MA, United States) and was dissolved in sterile dimethyl sulfoxide (DMSO). For the screening, a library of 284 drugs/compounds was purchased from Selleck laboratories. Additional 38 FDA-approved drugs were purchased from Tocris Bioscience, 20 were purchased from MedChemExpress, and 8 were from Sigma (Supplementary File S1). Drug dilutions were made in appropriate media such that the final DMSO concentration was less than 0.1%. Anisomysin and BKM-120 were purchased from Selleck and were used as positive controls. For validation studies, nebivolol, carvedilol and metoprolol were purchased from Selleck laboratories, L-748337 was purchased from Tocris Bioscience, and L-NAME was purchased from MilliporeSigma.

A concentration-response analysis of 350 GPCR-targeting compounds was conducted using a quantitative HTS (Inglese et al., 2006) in panel of six BC cell lines as described above. The assay conditions (cell number and incubation time) were optimized for each cell line by plating cells at 5 different cell densities and monitoring growth daily for 5 days during assay development. Cell plating densities were selected that ensure cells are in log phase growth over the course of the experiment while remaining at or below 70–80% confluence on the last day of the assay. For primary screens, each assay plate contained a pre-arrayed experimental drug library, 16 positive controls, an 8-point dose response curve in duplicate of BKM120, and 16 DMSO-treated negative controls wells. Additional Day 0 (pre-drug exposure) and Day 3 (untreated endpoint) were also collected from separate plates and were used for the statistical correction of growth. Each assay plate had a single concentration (0.1, 1.0 or 10 μM) of all the drugs on the plate. Assay plates were then fixed with 4% paraformaldehyde (v/v) after a 72-h drug exposure and the nuclei were labeled with Hoechst 33342 or 4,6-Diamidino-2-phenylindole dihydrochloride prior to imaging. All fluorescent cell images were collected using a 4x/0.2NA Plan Apo lens using a GE IN Cell 6,000 Analyzer at 405 nm/455 nm excitation/emission wavelength.

Each candidate drug/compound was marked for every GPCR it had activity against and its mode of action (e.g., agonist, partial agonist, inverse agonist, or antagonist) using The International Union of Basic and Clinical Pharmacology (IUPHAR)/British Pharmacological Society guide to pharmacology website (Database version 2021.3 August 2022, https://www.guidetopharmacology.org/) (Supplementary File S1). The compounds not found to have direct activity against GPCRs, were excluded from the analysis. We used two cut-offs to identify hits: 1) area under the curve for growth rate inhibition (AUC_GRI) of <0.95 and 2) concentration by which 50% of growth inhibition is achieved (GR₅₀) of <10 μM in two TNBC and two LTR cell lines. FDA-approved GPCR-targeting drugs used for chronic diseases were identified.

To validate the primary HTS results in panel of six BC cell lines as described above, 8-point concentration-response curves were generated. Briefly, cells were seeded at 4,000–6,000 cells/well in a 96-well tissue culture plate for overnight attachment. Then, drugs were added at various concentrations (10 nM–31.6 μM) for 72 h. The plates were scanned using the EnSight Multimode Plate Reader equipped with well-imaging technology (PerkinElmer, MA, United States). Cell count was obtained by digital phase and brightfield imaging. Data was normalized to vehicle 1) and plotted and analyzed using GraphPad Prism version 9. For determination of IC₅₀ values, the data was fitted using non-linear regression analysis and 3-parameter logistic equation with the slope set to 1: Y = Bottom + ((Top–Bottom)/(1 + 10^ ((X–LogIC₅₀))).

Invasion and migration assays were performed with or without Matrigel, respectively, on inserts as described before (Bhat et al., 2018).

To confirm the OXPHOS inhibitory function of Nebivolol, SUM159 TNBC cells (15,000 cells/well) were seeded onto XFp Seahorse cell plate and treated with 10 μM nebivolol. The oxygen conception rate (OCR) was measured using Cell Mito Stress kit (Cat. 103010-100, Agilent Technologies) in a Seahorse XFp Extracellular Flux Analyzers (Agilent Technologies) according to manufacturer’s instructions and as described before (Park et al., 2016).

All cell-based studies were run at least in duplicates and repeated at least two independent times. All data points of cell numbers at different time points (0 vs. 72 h) and under various conditions (vehicle vs. different concentrations of drugs) were used for analysis. The number of nuclei present in each image was counted using IN Cell Developer software version 9.2. Cell growth was determined using the pre-treatment, negative control of the endpoint, and the statistical methods described by the National Cancer Institute (Holbeck et al., 2010; Hafner et al., 2016). Numeric data was analyzed using Pipeline Pilot version 9.5 and GraphPad Prism version 9 to determine the fitness and the level of statistical significance of the assays. To evaluate plate-to-plate variability, data from on-plate controls were used to compare the minimum significant ratio (MSR) (NCGC Assay Guidance Manual) (Coussens et al., 2018). The quality of assay was assessed using the robust z-prime statistics as previously described (Zhang et al., 1999).

A retrospective cohort study was conducted using multi-year SEER-Medicare data from 2009 to 2015 in patients older than 66 years of age. The cohort included patients with BC as the 1st or only cancer between 01/01/2010 and 12/31/2014, continuously enrolled in Medicare parts A, B, and D during the 6 months immediately prior to BC diagnosis, received β-blocker monotherapy (carvedilol, metoprolol, or nebivolol) for at least 6 months prior to BC diagnosis without a gap of more than 30 days in therapy. Patients who used multiple β-blockers, enrolled in Health Maintenance Organization (HMO) or Part C during the 6-month baseline period, or those eligible for Medicare due to reasons other than age, were excluded. Patients were categorized into mutually exclusive and collectively exhaustive cohorts based on the type of β-blocker they used. Covariates like age, sex, race, ethnicity, BC as the first cancer, stage of BC, subtype of BC, use of statins due to their effects on BC mortality (Lv et al., 2020; Hosio et al., 2021; Kim et al., 2022; Zhao et al., 2022), Charlson comorbidity index, and the year of BC diagnosis were measured. The all-cause mortality and BC mortality were the outcomes for this study, and patients were followed from incident BC diagnosis until the earliest of all-cause mortality, discontinuation of the index β-blocker, switching or concomitant use of comparator β-blockers, or the end of the study period (31 December 2015). A sub analysis in patients with TNBC or HER2+ BC subtypes was also conducted. Descriptive statistics were calculated across the exposure groups, mean and standard deviation (sd) was calculated for continuous variables and the frequency and percentages were calculated for categorical variables. The competing risk regression model, adjusted for potential confounding variables, was used to study the association between the β-blocker groups and the risk of all-cause or BC mortality. The Cox proportional hazards model was used to assess the risk of all-cause mortality between the carvedilol, metoprolol, and nebivolol groups, after adjusting for baseline covariates.

Out of 350 drugs/compounds, 216 targeted at least one GPCR based on the IUPHAR/BPS guide to pharmacology website (Database version 2021.3, 2 September 2021) (Supplementary File S1). A total of 85 GPCRs were targeted by at least one drug/compound. The most targeted GPCRs (with ≥20 drugs/compounds) were H1, D2, 5-HT2A, 5-HT1A, 5-HT6, α1D, α1A, β2.

We conducted a cell-count based high-throughput growth assay. All screens were completed in 2 biological replicates run per cell line for all of the compounds, and multiple assay quality and reproducibility metrics were monitored throughout the screening campaign. From this analysis, we showed consistent rates of growth in control wells across all runs and cell line models, a high level of reproducibility in dose response curves across biological batches (Running MSR ≤3 for all cell lines), and a highly robust assay read-out determined by the (Median Z′-factor ≥ 0.70 for all cell lines) (Supplementary File S2). Concentration response curves were then evaluated (Supplementary File S3) to calculate AUC_GRI and GR₅₀ values (Supplementary File S4). Using two cut-offs of <0.95 AUC_GRI and <10 μM GR₅₀ values in two TNBC and two LTR cell lines, we identified six hits in all BC cell lines models (Table 1). Out of these six candidate drugs/compounds, 4 were FDA approved drugs that targeted CaS, mGlu5, β1, β2, β3, 5-HT2A, 5-HT1A, 5-HT1B, 5-HT5A, 5-HT7, and 5-HT6, receptors, with some drugs targeting more than one GPCRs (Table 1). Among these targets, β-adrenergic receptors were identified as commonly targeted GPCRs for various chronic diseases like heart failure and hypertension (Yang A. et al., 2021; Abosamak and Shahin, 2022), and with a role in BC (Barron et al., 2012). β-blockers are typically well tolerated drugs and are used long-term by many patients. Among β-blockers, only nebivolol (β1, β2, and β3 antagonist) inhibited the growth rate of all cell line models with AUC_GRI < 0.95 and GR₅₀ < 10 μM (Table 2). Other β-blockers targeting β1 and/or β2 did not inhibit HER2+ BC and TNBC cell growth (Table 2), suggesting that the effects of nebivolol are independent of its β-blocker role. Nebivolol also had more favorable safety profile when compared with other FDA approved candidates (Supplementary File S5), further justifying its selection for further exploration.

Nebivolol inhibited cell growth measured by the cell count in concentration-dependent manner in all six BC cell lines models (IC₅₀ = MDA-MB-231: 6.57 µM, MDA-MB-468: 4.60 µM, BT474 LTR: 3.47 µM, BT474 P: 7.96 µM, SKBR3 LTR: 8.05 µM, SKBR3 P: 2.55 µM, Figure 1A) similar to our screen data (Figure 1B). In addition, carvedilol and metoprolol showed no significant inhibition in agreement with the HTS data (Figures 1C–F). Nebivolol did not inhibit cell growth of the normal human mammary epithelial cell line MCF10A (Supplementary File S6), suggesting cancer cell-specific effects of nebivolol.

Apart from its β-adrenergic receptors blockade, nebivolol also dilates blood vessels through the L-arginine/NO pathway in the endothelium through β3 agonism (Maffei et al., 2007; Gupta and Wright, 2008; Maffei and Lembo, 2009; Coats and Jain, 2017). Therefore, we tested the effects of β3 antagonist (L-748337, 7 µM) (Rozec et al., 2009) and NO synthase blocker (L-NAME, 1 mM) (Martin et al., 1993) on cell growth by nebivolol in MDA-MB-231 and SKBR3 P cell lines. L-748337 and L-NAME did not reverse the cell growth inhibition effects of nebivolol (Figure 2), suggesting that the effects of nebivolol on cell growth inhibition were independent of β3 receptors and NO pathways.

Nebivolol reduced invasion (IC₅₀ = 208.3 nM) and migration (IC₅₀ = 9.39 nM) of MDA-MB-231 cells in a concentration-dependent manner (Figure 3). Nebivolol (10 µM) also inhibited mitochondrial oxidative phosphorylation (OXPHOS) in TNBC cells (Supplementary File S7), which is consistent with another published report on nebivolol (Nuevo-Tapioles et al., 2020) and high OXPHOS activity of many metastatic TNBC cells (Park et al., 2016; Jia et al., 2018).

The cohort included 4,843 patients, of these 4.79% (n = 232) patients were in the nebivolol group, 20.42% (n = 989) patients were in the carvedilol group, and 74.79% (n = 3,622) patients in the metoprolol group. The mean age of patients in the nebivolol group was 76.54 years (sd = 6.67 years), 78.71 years (sd = 7.60 years) in the carvedilol group, and 77.93 years (sd = 7.59 years) in the metoprolol group. Almost 99% of the cohort were females as expected for the BC study. Most of the patients were White and 95% of the cohort were not of Hispanic descent. Most patients were diagnosed at BC stages 0, 1, or 2 across all three β-blockers: 82.76% (n = 192) in the nebivolol group, 79.07% (n = 782) in carvedilol group, and 81.53% (n = 2,953) in metoprolol group. TNBC and HER2+ BC, accounted for 37.93% (n = 51) in the nebivolol group, 35.79% (n = 180) in the carvedilol group, and 37.71% (n = 681) in the metoprolol group. Baseline characteristics of patients are reported in Table 3.

In the BC cohort, patients using carvedilol (12.74%) had the highest rate of all-cause mortality during the follow-up period, followed by metoprolol (10.77%), and nebivolol (5.60%) of nebivolol. The median time to all-cause mortality in the BC cohort was 306 days for nebivolol, 319.5 days for carvedilol, and 388 days for metoprolol. The median time to BC mortality in this cohort was 250 days for nebivolol, 127.5 days for carvedilol, and 197 days for metoprolol. In the TNBC and HER2+ BC cohort, BC mortality accounted for 7.22% of carvedilol users, and 7.93% of metoprolol users. The event frequency in the nebivolol group was very low and thus the risk of BC mortality in this group could not be assessed using the existing SEER-Medicare data. The median time to BC mortality in the TNBC and HER2+ BC subgroup was 306 days for nebivolol, 386 days for carvedilol, and 281.5 days for metoprolol. The Kaplan Meier plot for the three exposure groups and the time to all-cause mortality is shown in Figure 4. The multivariable Cox proportional hazards model found that there was no significant difference in the risk of all-cause mortality in the nebivolol aHR = 0.71, 95%, confidence interval (CI) = 0.40 to 1.28, p-value = 0.25), and metoprolol groups aHR = 1.00, 95%, CI = 0.82 to 1.24, p-value = 0.97) compared to the carvedilol group. The multivariable competing risk model found that there was no significant difference in the risk of BC mortality in the metoprolol group (aHR = 1.66, 95% CI = 0.86 to 3.18, p-value = 0.13) compared to the carvedilol group (data not shown). The nebivolol group could not be included in this analysis because of low event frequency.