Terfenadine was purchased from Sigma-Aldrich (catalog number: T9625). Topotecan, paclitaxel, KN62, and KN93 were obtained from Selleck Chemicals (catalog number: S1231, S1150, S7422, and S7423). Rhodamine 123 was purchased from MedChemExpress (catalog number: HY-D0186) Antibodies used in experiments are listed in Table S1 .

All the human ovarian cancer cell lines were purchased from Sigma-Aldrich. Cells were cultured in RPMI 1640 medium with 10% fetal bovine serum (FBS) and 100 U/mL penicillin-streptomycin at 37°C with 5% CO₂.

ATP content assay (Promega) was conducted according to the manufacture’s protocols. Briefly, A2780-ADR cells were plated at 500 cells/well in 5 µL of RPMI 1640 medium with 10% FBS and 100 U/mL penicillin-streptomycin in white, solid-bottom 1,536-well plates and incubated 4 h at 37°C. Four concentrations of compounds from the library of pharmacologically active compounds (LOPAC, Sigma-Aldrich) consists of 1,280 small molecules, the NIH Chemical Genomics Center Pharmaceutical (NPC) collection with 4,265 compounds (28), as well as the Mechanism Interrogation Plate (MIPE) with 1,920 compounds were added to assay plates at 23 nL/well using a pintool station (WAKO Scientific Solutions, San Diego, CA). After a 72-h incubation at 37°C with 5% CO₂, the mixture of ATP LITE assay reagents was added to the assay plates at 5 µL/well. After incubation for the indicated time, the luminescence signal in the plates were detected using a ViewLux plate reader (PerkinElmer).

A2780-ADR cells were seeded onto 96-well plates at a density of 5,000 cells/well. The cells were pretreated with 2.5 to 10 µM terfenadine for different time. After pretreatment, the cells were incubated with 5 µM Rhodamine123 (Rh123) in culture medium and kept in the dark at 37°C with 5% CO₂ for 60 min. Plates were then washed twice with pre-warmed PBS, filled with 100 µl/well PBS, and measured using a Tecan reader at 485 nm excitation and 535 nm emission.

Caspase-3/7 activity assay (Caspase-Glo, Promega) and ATP content cell viability assay (CellTiter-Glo, Promega) were conducted according to the manufactures’ protocols. Ovarian cancer cells were plated at 3,000 to 5,000 cells/well in 100 µL of complete culture medium in white, solid-bottom 96-well plates and incubated overnight at 37°C with 5% CO₂. Compounds were added to the assay plates at indicated concentrations at 100 µL/well diluted in medium. After a 24 h (caspase 3/7 assay and ATP content cell viability assay) incubation at 37°C with 5% CO₂, the mixtures of assay reagents at 100 µL/well were added to the assay plates. After incubation for the indicated times from the protocols, the luminescence signal in assays plates were detected in a ViewLux plate reader.

RNA-sequencing analysis of A2780 and A2780-ADR was performed by Q2 Solutions as previously described (29, 30). RNA was isolated by Qiagen miRNeasy Mini Kit. cDNA libraries were generated using Illumina TruSeq Stranded mRNA sample preparation kit (Illumina # RS-122-2103). Read counts of each sample were normalized with DESeq and ran a negative binomial two sample test to find significant genes in higher transcript abundance in either sample. RNA sequencing data have been deposited in Gene Expression Omnibus (GEO) under accession number GSE177038.

Cells were lysed in RIPA buffer (Cell Signaling Technology) supplemented with protease inhibitors (cOmplete ULTRA Tablets, EDTA-free, Roche) and phosphatase inhibitor cocktail (PhosSTOP, Roche). The cell lysates were centrifuged at 16,000 rpm for 30 min. Supernatant was collected for protein quantitation with a BCA assay kit (Pierce BCA Protein Assay Kit, Thermo Fisher Scientific). The supernatant with similar protein concentrations were subsequently applied to Bis-Tris or Tris-Acetate gels for protein separation. Proteins were transferred to polyvinylidene difluoride (PVDF) membrane by dry transfer (iBlot 2 Gel Transfer Device, both from Thermo Fisher Scientific) or tank wet transfer. Immunoblot analysis was performed with indicated antibodies and the chemiluminescence signal was visualized with Luminata Forte Western HRP substrate (EMD Millipore) in a BioSpectrum system (UVP, LLC). The chemiluminescence intensity of the band was calculated in the VisionWorks LS software (UVP, LLC).

CETSA was performed as previously described (31). A2780-ADR cells were harvested, rinsed with PBS, and re-suspended in detergent-free buffer (25 mM HEPES pH 7.0, 20 mM MgCl₂, 2 mM DTT) supplemented with protease inhibitors and phosphatase inhibitor cocktail. The cell suspensions were lysed via three freeze-thaw cycles with liquid nitrogen. The cell lysates were centrifuged at 16,000 rpm for 20 min at 4 °C to pellet the cell debris from the soluble fraction. The soluble portion were diluted in detergent-free buffer and divided into two aliquots, with or without 600 μM terfenadine treatment. After 60 min incubation at room temperature, each sample was divided into 12 small aliquots in 50 μL/tube and individually heated at different temperatures (37 to 70 °C with 3 °C interval) for 3 min in a thermal cycler (Eppendorf), followed by immediate 3 min cooling cycle on ice. The heated samples were centrifuged at 20,000 × g for 20 min at 4 °C to remove the precipitates from the soluble fractions. The supernatant was examined by western blotting with CAMKII antibody. The relative chemiluminescence intensity of each sample at different temperatures was used to plot the temperature dependent melting curve. The apparent aggregation temperature (T_agg) was calculated by nonlinear regression. The statistically significance between two curves were analyzed by extra sum-of-squares F test. All data represent mean ± SEM of at least 3 replicates.

The primary screen data were analyzed using customized software developed internally (32). All data from the cell-based assays were presented as the mean ± standard error of the mean (SEM) with at least three independent experiments unless otherwise stated. Half maximal inhibitory concentrations (IC₅₀) of doxorubicin or compounds were calculated using Prism software (version 7, GraphPad Software, San Diego, CA). All imaging data are presented as the mean ± SEM and represent data from cells in at least 10 fields from three or more independent experiments. The two-tailed unpaired Student’s test of the mean was used for single comparisons of statistical significance between experimental groups. One-way analysis of variance (ANOVA) with Bonferroni test was used for multiple comparisons. Bliss independence with Prism or SynergyFinder (33) was used to define synergistic or additive effects.

To explore potential novel therapies for ABCB1-mediated MDR ovarian cancer cases in the clinic, we conducted a qHTCS against an ABCB1-overexpressing MDR ovarian cancer cell line, A2780-ADR. Compared to the parental A2780 cells, the A2780-ADR cells exhibited a higher expression and overall activity of ABCB1, as demonstrated by a considerable rise in protein level in Western blot detection and a significantly reduced cellular accumulation of Rho123, an ABCB1 substrate ( Figures 1A, B and Figure S1A ). The IC₅₀ values of three ABCB1 substrates (doxorubicin, topotecan, and paclitaxel) for the A2780-ADR cells were 7.08 uM, 0.0081 uM, and 0.88 uM, which were significantly higher than 14 nM, 0.95 nM, and 0.0016 nM, respectively, for the A2780 cells ( Figure 1C and Figures S1B, C ). In the presence of tariquidar, a specific ABCB1 inhibitor, their anticancer effectiveness against A2780-ADR cells also increased (34) ( Figure 1D and Figures S1D–F ).

The qHTCS was performed in two stages. In the first stage, we examined 6,016 pharmacologically active compounds as single drugs at five doses in a luminescent cell viability assay to narrow down the compound pairs. 246 compounds were found that effectively inhibit the proliferation of A2780-ADR cells with an IC₅₀ < 10 µM ( Figure 1E ). To further identify compounds that showed combination effects with doxorubicin, these 246 compounds were evaluated at 11 concentrations in combination with four doxorubicin concentrations at 0.1, 1, 10, and 25 µM, separately. Consequently, 24 compounds were identified as doxorubicin synergistic compounds in A2780-ASR cells, as indicated by the decreasing IC₅₀ of each drug as the doxorubicin dose rose ( Figure 1F and Table S2 ). Terfenadine was selected for further investigation ( Figure 1G ) since the mechanism of terfenadine and doxorubicin combination is unknown and terfenadine’s anticancer activity has been described (15, 16).

With an IC₅₀ of 4.8 µM, the inhibitory activity of terfenadine as a single agent was confirmed in MDR A2780-ADR cells ( Figure 2A ). As evidenced by the shifted toxicity curve in MDR cells, the combined treatment significantly decreased the IC₅₀ of doxorubicin in a dose-dependent manner, indicating the potential synergistic effect of these two drugs ( Figure 2B ).

To quantify these enhanced anticancer effects, we computed the combinational index (CI) (CI<1, synergism; CI=1, additive; CI>1, antagonism) (35). The mean CI value was 0.35, showing that the interaction in A2780-ADR cells is synergistic ( Figure 2C ). To further evaluate the synergism and determine the best synergistic concentration, effects were investigated using a dose-response matrix and analyzed using the zero interaction potency (ZIP) model (33, 36). As a result, terfenadine exhibited a synergistic effect with doxorubicin. The average ZIP synergy scores for A2780 cells and A2780-ADR cells were 2.874 and 4.403, respectively ( Figures 2D, E ). Notably, in resistant cells, the synergy score reached 49.80 when resistant cells were treated with 3.33 µM terfenadine and 1.11 µM doxorubicin, which was higher than the highest score of 22.39 in sensitive cells, showing a greater synergistic effect in MDR cells than in their parental cells, indicating that terfenadine will target the abnormally activated pathway associated with ABCB1 overexpression in MDR cells. In addition, the combination therapy lowered cell counts in a nuclear staining-based counting assay ( Figure 2F ), enhanced caspase-3/7 activity ( Figure 2G ), and promoted PARP cleavage ( Figure 2H and Figure S2A ), suggesting that the A2780-ADR cells were induced to undergo apoptosis. During the expanded testing, terfenadine had a comparable effect on the toxicity of paclitaxel and topotecan for A2780-ADR cells ( Figures S2B, C ).

To investigate the mechanism underlying this synergistic effect, we first examined two conventional terfenadine targets: the histamine H1 receptor (H1R) (37) and the human ether-a-go-go-related gene (hERG) channel (38). Terfenadine was reported as an antagonist of the H1R and is a prodrug that is converted to fexofenadine in the liver (37). Nevertheless, it was withdrawn from the market due to its ability to inhibit the hERG channel (38). The dose-response curve for doxorubicin as monotherapy was nearly comparable to the dose-response curve for doxorubicin in conjunction with the H1R-specific inhibitor fexofenadine (39) ( Figure 3A ). The CI for doxorubicin and fexofenadine was 0.9819, showing that their effect was additive, not synergistic ( Figures 3A, B ). In a second attempt, tannic acid (TA), a blocker of the hERG channel, was recruited (40). Even while TA decreased cell viability at higher concentrations, the dose-response curve for the combined therapy was similar to that of the doxorubicin treatment alone ( Figure 3C ). The CI value for the combination of doxorubicin and TA was 1.1867, showing that there was no synergy between the two drugs ( Figures 3C, D ). In addition, western blot revealed no difference in H1R and KCNH2 protein expression between A2780 and A2780-ADR cells, indicating that these proteins are not essential for MDR development ( Figure 3E and Figure S3 ). These findings indicated that the synergistic effect of terfenadine and doxorubicin was not the result of terfenadine inhibiting the H1R or the hERG channel.

We questioned whether terfenadine impacts the expression or function of ABCB1 in these MDR ovarian cancer cells, as ABCB1 overexpression was essential for the chemoresistance of A2780-ADR. Indeed, terfenadine decreased the expression of the ABCB1 protein in a dose-dependent manner ( Figure 3F and Figure S4A ). The reduction of ABCB1 was also found following doxorubicin and terfenadine treatment ( Figure 3G and Figure S4B ). In addition, in the ABCB1 activity experiment, terfenadine boosted Rho123 accumulation dose-dependently, indicating the decreased cellular ABCB1 activity. The Rho123 signal achieved a plateau when the concentration exceeded 5 µM ( Figure 3H ). In a time-course study, the intracellular level of Rho123 continued to increase until terfenadine-induced apoptosis occurred ( Figure 3I ). Moreover, combination treatment significantly enhanced doxorubicin levels in A2780-ADR cells ( Figures 3J, K ). These findings imply that terfenadine decreased the expression and activity of the multidrug efflux pump ABCB1 in A2780-ADR cells, which resulted in the accumulation of doxorubicin and apoptosis.

To further investigate the potential targets and mechanisms of drug resistance, we used RNA-seq to assess the transcriptional differences between A2780 and A2780-ADR. A total of 5,694 genes with a fold change of > 2 and a p-value < 0.05 were identified as differentially expressed genes (DEGs) ( Figure 4A ), including 2,755 genes that were over-expressed and 2,939 genes that were under-expressed in the A2780-ADR cells. The gene ontology (GO) enrichment study revealed 13 calcium-related biological processes which caught our attention ( Figure 4B ). The Gene-Pathway network showed that the majority of DEGs were clustered in the cytosolic calcium ion transport, homeostasis, and response processes ( Figure 4C ), indicating this MDR cell line possessed abnormal calcium signaling.

In addition to the preliminary analysis, a gene set enrichment analysis (GSEA) was conducted to identify probable biological pathway enrichment from the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database. This study uncovered the difference in the KEGG calcium signaling pathway (p<0.0001) as well ( Figure 4D ). Using all 177 genes involved in the calcium pathway, unsupervised hierarchical clustering revealed a clear separation of these two cell types ( Figure 4E ), demonstrating a major modification in the calcium homeostasis of MDR cells. Collectively, these findings suggest that the calcium signaling pathway is associated with the MDR phenotype in the A2780-ADR cells.

Among the proteins implicated in the calcium signaling pathway, RNA-seq data revealed dysregulation of calcium/calmodulin-dependent protein kinase II (CaMK2) members. Specifically, CAMK2D is highly up-regulated, and CAMK2B is down-regulated ( Figure 5A ). The rise of CAMK2D and its active form, phosphorylated-CAMK2D (T286), in A2780-ADR cells was confirmed by Western blotting ( Figure 5B and Figure S5A ). In addition, we detected an increase in the phosphorylation of CREB1 at S133, although RNA and protein levels remained unchanged ( Figures 5B, C and Figure S5A ). As revealed by these results, they indicated the CAMK2/CREB1 pathway was overactive in the A2780-ADR cells.

To determine if terfenadine blocked the CAMK2/CREB1 pathway, the expression of related proteins was measured following terfenadine administration. After 24 hours of treatment, dose-dependent reductions in CAMK2D and phosphorylated CAMK2D were observed. Meanwhile, CREB1, ABCB1, and baculoviral inhibitor of apoptosis repeat-containing 5 (BIRC5) were also reduced ( Figure 5D and Figure S5B ). The protein BIRC5, also known as survivin, suppresses apoptosis by inhibiting caspase activation. As CREB1 is a transcription factor located in the nuclei and is activated by direct binding of CAMK2, we also examined their levels in the nuclei and found that terfenadine dose-dependently decreased the CAMK2D and the phosphorylated CREB1 in the nuclei ( Figures 5D, E and Figures S6A, B ), indicating a decrease in the activating and nuclear entry of CAMK2D in the presence of terfenadine.

A cellular thermal shift assay (CETSA) was conducted to assess if terfenadine directly binds to CAMK2D to prevent its activation. CAMK2D’s apparent aggregation temperature (T_agg) was evaluated in the absence or presence of terfenadine in A2780-ADR cell lysates ( Figure 5F and Figure S6C ). The best-fit curve for the terfenadine-treated group shifted significantly from that of the DMSO control (p< 0.001). Figure 5G shows that terfenadine reduced the T_agg of CaMK2D protein from 57.4 to 54.2°C, indicating that it thermally destabilized CAMK2D. Together with the other studies demonstrating that CREB1 regulates ABCB1 expression (41, 42), our findings suggest that terfenadine may prevent cells from apoptosis by regulating the Ca²⁺-mediated CAMK2/CREB1 pathway through binding directly to CAMK2D, thereby causing its destabilization in cells and reducing the activation of CREB1 and subsequent ABCB1 expression.

To confirm further that CAMK2D was a target for MDR combination therapy, we recruited KN62, a CaMK2 specific inhibitor, for our investigation (43). KN62 reduced the expression and activity of ABCB1 in A2780-ADR cells, consistent with the terfenadine therapy ( Figures 6A, B and Figure S7A ). Moreover, KN62 inhibited the expression of BIRC5 and the phosphorylation of both CAMK2D and CREB1 in A2780-ADR cells ( Figure 6A ). The IC₅₀ of doxorubicin dropped from 2.4 µM (doxorubicin alone) to 0.17 µM (doxorubicin paired with 5 µM KN62) in A2780-ADR cells when KN62 was administered in combination with doxorubicin ( Figure 6C ). This was a synergistic combination (CI < 1) ( Figure 6D ). Notably, KN62 had the same effect as terfenadine in the combination with doxorubicin, decreasing ABCB1 and BIRC5 expression and increasing cleaved PARP ( Figures 6E, F ). Although the phosphorylated CAMK2D was lowered in either terfenadine or KN62 treatment alone, there was no difference in CAMK2D phosphorylation in combination treatments with either of them with doxorubicin. In contrast, CREB1 phosphorylation remained lower in the combined treatment, indicating that CAMK2 inhibitors repressed the CAMK2/CREB1 pathway ( Figures 6E. F and Figures S8A, B ). Furthermore, KN93, another CAMK2-specific inhibitor, reduced ABCB1 activity and increased doxorubicin-induced cell death in A2780-ADR cells ( Figures S9A, B ). These findings suggest that inhibiting CAMK2 could resensitize MDR cells to doxorubicin.

To determine if blocking CREB1 would similarly resensitize MDR cells to doxorubicin, A2780-ADR cells were treated with the selective CREB1 inhibitor 3i (also known as 666-15 (44). Consistent with the CAMK2 inhibitors, ABCB1 and BIRC5, along with CREB1 phosphorylation, decreased after treatment with 666-15 ( Figure 6G and Figure S7B ). Under the 666-15 treatment, the total ABCB1 activity similarly dropped in a dose-dependent manner ( Figure 6H ). In addition, the combination of doxorubicin and 666-15 killed MDR cells A2781-ADR synergistically (CI < 1) ( Figure 6I ). Together, inhibition of the Ca²⁺ mediated CAMK2D/CREB1 pathway appears to be a promising therapeutic target for doxorubicin resensitization in ABCB1-mediated MDR ovarian cancer.