Phenotypic Consequences of SLC25A40-ABCB1 Fusions beyond Drug Resistance in High-Grade Serous Ovarian Cancer

Simple Summary Among the plethora of malignancies affecting the female reproductive tract, those concerning the ovary are the most frequently fatal. In particular, chemotherapy-resistant High-Grade Serous Ovarian Cancer (HGSOC) remains a clinically intractable disease with a high rate of mortality. We previously identified SLC25A40-ABCB1 transcriptional fusions as the driving force behind drug resistance in HGSOC. As success in the clinical arena will only be achieved by enhancing our fundamental understanding of the drivers that mediate cellular drug resistance, this report sought to elucidate the phenotypic, metabolomic and transcriptional consequences of SLC25A40-ABCB1 fusions beyond drug resistance. High-throughput FDA drug screening was also undertaken to identify new therapeutic avenues against drug-resistant cellular populations. Abstract Despite high response rates to initial chemotherapy, the majority of women diagnosed with High-Grade Serous Ovarian Cancer (HGSOC) ultimately develop drug resistance within 1–2 years of treatment. We previously identified the most common mechanism of acquired resistance in HGSOC to date, transcriptional fusions involving the ATP-binding cassette (ABC) transporter ABCB1, which has well established roles in multidrug resistance. However, the underlying biology of fusion-positive cells, as well as how clonal interactions between fusion-negative and positive populations influences proliferative fitness and therapeutic response remains unknown. Using a panel of fusion-negative and positive HGSOC single-cell clones, we demonstrate that in addition to mediating drug resistance, ABCB1 fusion-positive cells display impaired proliferative capacity, elevated oxidative metabolism, altered actin cellular morphology and an extracellular matrix/inflammatory enriched transcriptional profile. The co-culture of fusion-negative and positive populations had no effect on cellular proliferation but markedly altered drug sensitivity to doxorubicin, paclitaxel and cisplatin. Finally, high-throughput screening of 2907 FDA-approved compounds revealed 36 agents that induce equal cytotoxicity in both pure and mixed ABCB1 fusion populations. Collectively, our findings have unraveled the underlying biology of ABCB1 fusion-positive cells beyond drug resistance and identified novel therapeutic agents that may significantly improve the prognosis of relapsed HGSOC patients.


Introduction
"The law is not the survival of the "better" or the "stronger". It is the survival of those which are constitutionally fittest to thrive under the conditions in which they are placed."

Herbert Spencer
Despite our rapidly evolving knowledge concerning cancer cell heterogeneity, we are far from deciphering the complex dynamics that operate between cellular subpopulations within tumors. Indeed, most of the conventional cytotoxic drug-based strategies which still dominate anticancer treatment do not account for target cell heterogeneity and have reached their limit in terms of efficacy. As growing evidence suggests that cancer cells behave as communities, increasing attention is now being directed towards understanding how the intense selection pressures imposed by cytotoxic therapies influences clonal evolutionary dynamics and thus therapeutic response. With each cubic centimeter of tumor containing up to one billion cancer cells [1], new insights into how clonal interactions and evolutionary dynamics between drug-sensitive and drug-resistant cellular populations influences tumorigenesis and disease progression are urgently required to substantially change therapeutic treatment practices and shift overall survival. This is particularly pertinent for women diagnosed with HGSOC.
To date, ovarian cancer remains the deadliest form of gynecological malignancy, with >140,000 women globally succumbing to disease each year [2,3]. Of these, the HGSOC histosubtype predominates clinical settings and is responsible for a disproportionate share of fatalities from all forms of ovarian cancer (70-80%) [4]. This high mortality is largely due to a lack of effective screening and early detection methods leading to advanced disease at presentation and resistance to available treatments. These genetically highly unstable tumors are characterized by a high degree of genomic copy number change rather than recurrent point mutation, with only TP53 frequently mutated (96.7%) [5][6][7]. As such, the dearth of targetable mutations, with the exception of PARP (poly ADP ribose polymerase) inhibitors for BRCA1/2 mutant cohorts [8], has not fueled profound improvements in overall survival outcomes over the past few decades (overall survival of 40.7 months) [9].
The invariable emergence of drug resistance following initial response to chemotherapy and limited molecularly targeted approaches continue to be the principal limiting factor to achieving cures in patients with cancer. Current HGSOC chemotherapy regimens rely on platinum and taxane-based agents. However, despite initial sensitivity to first-line chemotherapy, 25% of patients will become platinum-refractory in the primary setting, with 20% becoming platinum-resistant with a recurrence within six months of the conclusion of chemotherapy [10,11]. Our recent extensive genomic characterization of post treatment HGSOC patient samples identified the most common mechanism of acquired resistance in HGSOC to date, transcriptional fusions involving the ATP-binding cassette (ABC) efflux transporter ABCB1 (also known as MDR1), which has well established roles in multidrug resistance (MDR) [12], and the solute carrier SLC25A40 [7,13]. Indeed, ABCB1 chromosomal rearrangements driving MDR have also been detected in ALL (acute lymphocytic leukemia) [14] and breast cancer [13]. These resulting fusions place ABCB1 under the control of a strong promoter whilst leaving its open reading frame intact [7,13].
ABC transporters differ from classical selective transporters by way of their promiscuity for structurally and chemically diverse substrates (>200) and contribute to tumor biology independently of their ability to efflux cytotoxic drugs [15]. Indeed, this family of 48 transporters is also involved in lipid export/homeostasis and mediates the release of bioactive lipids (phospholipids and sphingolipids) that activate signaling cascades involved in cellular proliferation, migration and tumorigenesis [16,17]. ABCB1, which encodes the drug efflux transporter P-glycoprotein (P-gp), is the most extensively studied member of the ABC transporter superfamily. ABCB1 couples the hydrolysis of ATP (adenosine triphosphate) to export numerous xenobiotics and toxic metabolites (cationic hydrophobic compounds (300-4000 Da) [18] in a unidirectional path across the phospholipid bilayer of cellular membranes against a chemical gradient [19], including agents that are central to most chemotherapeutic regimens, including anthracyclines (doxorubicin), tyrosine kinase inhibitors (imatinib), taxanes (paclitaxel), epipodophyllotoxins (etoposide) and vinca alkaloids (vincristine) [12,20]. Indeed, ABCB1 fusion events were only detected in HGSOC patients who had been exposed to known ABCB1 substrate chemotherapies (doxorubicin and paclitaxel), with the probability of a fusion event being closely correlated to the number of lines of substrate chemotherapy administered [13]. Interestingly, our genomic analysis revealed that not all cells within the tumor population harbored ABCB1 fusions, raising the question of how fusion-negative cells have survived under chemotherapy-induced selective pressure. Besides the high energy expenditure required to actively export drugs or xenobiotics, the phenotypic consequences of ABCB1 overexpression beyond drug extrusion and how clonal cooperation/dynamics between ABCB1 fusion-positive and negative cells influences cellular behavior, tumor fitness and therapeutic response remains unknown.
Although P-gp was first described in 1976 [21], the potential benefit of inhibiting ABCB1 transporter activity and thus reversing therapy resistance has not come to clinical fruition despite the wealth of knowledge concerning the biochemistry and substrate specificity of ABC transporters. The shortcomings of several third generation P-gp inhibitors in the clinic have primarily been attributed to their distribution to non-target organs, as P-gp is constitutively expressed on epithelial cells of the kidney, liver, pancreas and intestine [22], leading to intolerable side effects [23]. As the emergence of therapy resistance is generally viewed as an evolutionary process in which cancer cells adapt to selection pressures mediated by cytotoxic drugs, this study has specifically focused on elucidating the underlying biology of drug-resistant ABCB1 fusion-positive cells and how this distinctive phenotype, in combination with evolutionary dynamics, can be clinically exploited.
Using a panel of single-cell SLC25A40-ABCB1 fusion-negative and positive clones, our study sought to elucidate the complex underlying biology of fusion-positive cells beyond drug resistance as well as identify therapeutic options that can be utilized to combat this nefarious population of cells.

Immunodetection
Whole-cell lysates were prepared using RIPA lysis buffer containing protease inhibitor cocktail (Roche) followed by sonication. Protein concentrations were determined using the Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA). In total 40 µg of protein was loaded onto Mini-PROTEAN TGX 4-20% gels (Bio-Rad, Hercules, CA, USA) and subjected to gel electrophoresis at 90 V for 10 min followed by 150 V for 90 min and was then transferred onto nitrocellulose membranes using an iBlot2 (Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer's instructions. Membranes were blocked in Odyssey Blocking Buffer (PBS; LI-COR Bioscience) for 1 h at room temperature followed by overnight 4 • C incubation with primary antibodies, GAPDH (Ab8245, 1:10,000) and MDR1 (Ab170904, 1:1000). Immunodetection was achieved after incubation with infrared (IR) dye-conjugated 800CW secondary antibodies (LiCor) with bands visualized using the Odyssey Imaging System. Densitometry analysis was performed using ImageJ software (V.153k), with ABCB1 protein normalized to GAPDH loading control and AOCS18.5 parental cells.

IncuCyte Live Cell Imaging
Cells were seeded (1000-3000 cells/well) in 96 flat clear or black wall (fluorescence assays only) microtiter plates and left to adhere overnight (triplicate wells per condition). Phase contrast and/or green/red fluorescent images (10× magnification) were taken in the IncuCyte ZOOM or SX5 Imaging System (Sartorius, Göttingen, Germany) at 12 or 24 h intervals for a maximum of 264 h, with media changed every 96 h. Cell confluence (phase contrast) or green/red (800 ms) fluorescence area (µm 2 /Image) was evaluated using IncuCyte ZOOM 2016A software (Sartorius). For our purposes, a cell was defined as an entity with an area greater than 450 µm 2 . Data represents mean ± SEM from a minimum of 4 independent experiments.

MDR Efflux Assay
The multidrug resistance direct dye efflux assay was performed according to the manufacturer's protocol (Merck Millipore, Burlington, MA, USA). Briefly, 2.5 × 10 5 cells per cell line were collected and incubated with Rhodamine 123 loading buffer (1 h at 4 • C). Cells were centrifuged (200× g, 5 min), resuspended in cold efflux buffer and separated into 3 treatment groups: (i) 37 • C warmed efflux buffer containing DMSO, (ii) 37 • C warmed efflux buffer containing Vinblastine and (iii) ice-cold efflux buffer. Cells were incubated for 3 h at either 37 • C or 4 • C, followed by 2 washes in ice cold efflux buffer. Cells were then stained with FluoroGold (1:300, Sigma-Aldrich) and assessed for Rhodamine 123 efflux through Flow Cytometry Analysis (FACS) analysis (10,000 cells).

RNA Extraction and qRT-PCR
Total RNA was extracted using the RNeasy kit (Qiagen, Hilden, Germany) with on-column DNase digestion. cDNA was generated using 1000 ng of total RNA (Bioline SensiFast cDNA synthesis kit, (Meridian Bioscience, Cincinnati, OH, USA) with subsequent qRT-PCR performed using SYBR Green PCR mix (Applied Biosytems, Waltham, MA, USA). Reactions were processed on a LightCycler 480 (Roche, Basel, Switzerland) with subsequent gene expression quantified using the ∆∆CT method from triplicate reactions. ABCB1 gene expression and SLC25A40-ABCB1 fusion expression were normalized to the internal housekeeping genes GAPDH and HPRT. Primer sequences as previously described [13].

RNA Sequencing
For cell line analysis, 1 µg of total RNA was submitted for NextSeq 100 bp paired-end, polyA RNA sequencing (Molecular Genomics Core, Peter MacCallum Cancer Centre), 3 independent replicates per clone. Libraries were prepared using the NEBNext Ultra II Directional RNA Library Prep Kit for Illumina, generating 20 million paired reads per sample. Reads were mapped to the human reference GRCh37.92 using the STAR 2-pass method (v2.6.0b). Counts were generated on the ensemble release GRCh37.92 GTF annotation using HTSeq (v0.10.0). Counts were normalized to logged TMM values using edgeR (v3.28.1). TPM expression values were also generated. Differential gene expression analysis was performed using DESeq2 (v1.26.0) on the raw counts. The fGSEA R package (v1.15.1) was run on the genes ranked by DESeq2 log 2 fold changes for the MSigDB Hallmark Pathways (v7.1), generating normalized enrichment scores (NES) for each pathway. Fusion status was confirmed using the Arriba fusion tool (v1.1.0).
Expression profiles from relapsed fusion-negative (n = 21) and fusion-positive (n = 6) HGSOC patient's samples were previously conducted [7,13]. Clinical characteristics of the HGSOC patient cohort are detailed in Christie et al. (2019) and Patch et al. (2015) [7,13]. Women diagnosed with HGSOC were recruited at hospitals across Australia and were recruited through the AOCS. Ethics board approval was obtained at all institutions for patient recruitment, sample collection and research studies (HREC protocol 01/60 and 16/161). Written informed consent was obtained from all participants.

Extracellular Flux Analysis
Extracellular flux analyses were performed on a Seahorse XFe96 Analyzer (Agilent, Santa Clara, CA, USA). Assay medium was prepared using Seahorse XF Base Medium (containing 5.5 mM glucose, 2 mM glutamine and 1 mM sodium pyruvate, adjusted to pH 7.4 and kept at 37 • C; Agilent 102353-100).
The XF Cell Mito Stress Test was performed as previously described [25]. Cells were seeded on Seahorse XF 96 well cell culture plates (5000-14,000 cells per well) 48 h prior to analysis (quadruplicate wells per condition). Cell culture medium was removed and replaced with Seahorse XF medium, with cells equilibrated in a non-CO 2 incubator for 1 h prior to the assay. The XF Cell Mito Stress Test protocol was performed as per manufacturer's directions, using oligomycin (1 µM), FCCP (1 µM) and rotenone/antimycin A (1 µM). The assay was run with repeated cycles of 3 min mix and 3 min measurements following each drug injection with simultaneous measurement of oxygen consumption rate (OCR) and extracellular acidification rate (ECAR). At completion of the assay, cells were injected with Hoescht live-cell nuclear stain and imaged using a Cellomics Arrayscan automated microscope (10× magnification; 4× fields). OCR and ECAR values were subsequently normalized to cell number, and data were analyzed using the Mito Stress Test Report Generator (Agilent). Centre for Functional Genomics) core. Automated drug dispensing was performed using the Sciclone ALH 3000 Workstation (PerkinElmer, Waltham, MA, USA) with end staining facilitated with the BioTek workstation (BioTek Instruments, Winooski, VT, USA).

Viability and Actin Morphology
For cell viability, the entire area of each well was imaged (9 fields) at 10× magnification, and for actin morphology,~300 cells were imaged at 20× magnification using a Cell Insight CX7 High-Content Screening (HCS) Platform (Thermo Fisher Scientific, Waltham, MA, USA). The images were then analyzed for nuclear count and actin morphology using a CellProfiler v3.0 pipeline run on a high-performance computer cluster. To identify cell viability hits, the cell counts were first normalized for batch effects by fold change to the median of the negative control wells on a per plate basis. The normalized values were then robust Z-scored to the entire dataset in order to determine the relative strength of each compound.
To compare actin morphology between fusion-positive and negative clones, 148 measurements encompassing a wide range of features related to cell shape and actin intensity and texture were extracted from untreated cells. These raw values were first normalized and scaled to a reference distribution by robust Z-Scoring each feature to the median and median absolute deviation of the same feature in the fusion-negative cell population. The entire dataset was then passed through an iterative feature reduction process. Features with low variance across fusion-positive and fusion-negative clones, or inactive features, as well as highly correlated (Pearson's correlation coefficient > 0.85) or redundant features, were removed. Upon completion of this process, a final set of 59 features was selected for further analysis.

Targeted Sequencing
Targeted sequencing was performed on AOCS18.5 clone DNA using a customized Agilent SureSelect XT Low Input capture panel (Design ID 3221041) and the SureSelect XT Low Input Target Enrichment System (Agilent). Libraries were sequenced on an Illumina NextSeq 500. The panel covered 63 genes described to play a role in DNA repair or treatment resistance, including ABCB1 and TP53. Sequencing data were aligned to the Genome Reference Consortium human genome assembly (GRCh37 b37) using BWA mem (v0.7.17-r1188). Variants were called using 4 tools: Mutect2 (v4.0.11.0), VarDictJava (v1.5.7), Strelka2 (v2.9.9) and Varscan2 (v2.4.3) and were annotated using the Ensembl Variant Effect Predictor (v92.4). Those variants identified by a single caller were not analyzed further, and variants with a combined variant allele frequency of <10% across all lines were also discarded. The TP53 mutations were manually reviewed in Integrative Genomics Viewer (IGV).

Statistical Analysis
All statistical analyses were performed using GraphPad Prism software (Version 9.0.2) or R (Version 3.6.1). Data are expressed as mean ± standard error of the mean (SEM) as indicated, from a minimum of 3 independent measurements. The minimum threshold for rejecting the null hypothesis was p < 0.05.

SLC25A40-ABCB1 Fusions Mediate an ECM/Inflammatory Enriched Transcriptional Profile
To investigate whether additional genomic changes are associated with the acqu tion of SLC25A40-ABCB1 fusions, targeted sequencing of 63 genes involved in DNA rep or chemotherapy resistance in HGSOC was performed on fusion-negative (n = 5) and p itive clones (n = 5). The parental line and all single-cell clones contained the p.L132M T driver mutation (VAF > 0.99). No mutations were found exclusively in all fusion-posit clones ( Figure S4).
The synthesis and maintenance of ATP-dependent ABC pumps on the apical cell membrane and transport of substrates requires a considerable investment of energy [26]. For cancer cells, this metabolic cost is observed even in the absence of drugs and requires a significant diversion of resources that would ordinarily be devoted to processes such as invasion or proliferation [27]. IncuCyte live-cell proliferation assays were undertaken to assess whether the presence of SLC25A40-ABCB1 fusions alters the proliferative fitness of HGSOC cells. Indeed, fusion-positive AOCS18.5 clones required 2.18-fold more time to reach 50% confluency compared to their fusion-negative counterparts (average 62.46 ± 3.14 versus 135.38 ± 10.99 h, respectively) (p = 0.0002) with ABCB1 mRNA expression levels strongly correlating with cellular proliferative times (R 2 = 0.7779) ( Figure 1B,C and Table S1).
To further examine whether SLC25A40-ABCB1 fusions modulate cellular fitness, the overall fitness of fusion-negative and positive clones was compared. A logistic population growth model was fitted to each clone's 2D growth curves and the estimated intrinsic growth rate parameter provided an absolute fitness measure of each clone. The fitness of fusion-negative clones was 2-fold greater (1.9-2.3) than fusion-positive clones (df = 8, t = −6.1, p < 0.0005) ( Figure 1D). Indeed, the population doubling time for fusionnegative clones was 22 h versus 44 h for fusion-positive clones, highlighting that ABCB1 fusions significantly impair the proliferative fitness of HGSOC cells.
As overexpression of ABCB1 is one of the key factors leading to cancer MDR, we next sought to solidify the association of SLC25A40-ABCB1 fusions with resistance to standard of care HGSOC chemotherapeutic agents. AOCS18.5 parental cells, as well as fusion-negative (n = 5) and fusion-positive clones (n = 5), were treated with P-gp substrate (doxorubicin, paclitaxel) and non-substrate (cisplatin) agents for 72 h, and their viability was determined through DAPI staining. As expected, compared to fusion-negative  Figure 1E) ( Figure S3A). Indeed, cisplatin sensitivity was strongly correlated with proliferative rate (R 2 = 0.878) with slower growing fusion-positive lines (>100 h to reach 50% confluency) requiring significantly higher levels of cisplatin to achieve an IC 50 (p = 0.0003) ( Figure S3B).
Finally, as substrate transport mediated by P-gp is ATP-dependent (P-gp hydrolyzes two ATP for every extruded molecule) [28], we next investigated whether ABCB1 fusions mediate basal changes in mitochondrial respiration. Seahorse Mito Stress Test assays were employed to assess seven parameters of energy expenditure in fusion-negative and positive cells. Of these markers, basal respiration (p = 0.026), ATP production (p = 0.029) and nonmitochondrial oxygen consumption (p = 0.008) were significantly elevated (1.39-1.72 fold) in fusion-positive lines, suggesting that presence of the fusion causes elevated basal oxidative metabolism to compensate for the increased energy demand ( Figure 1F) ( Figure S3C). Together, our results show that in addition to driving drug resistance, SLC25A40-ABCB1 fusions result in proliferative and metabolic fitness penalties.

SLC25A40-ABCB1 Fusions Mediate an ECM/Inflammatory Enriched Transcriptional Profile
To investigate whether additional genomic changes are associated with the acquisition of SLC25A40-ABCB1 fusions, targeted sequencing of 63 genes involved in DNA repair or chemotherapy resistance in HGSOC was performed on fusion-negative (n = 5) and positive clones (n = 5). The parental line and all single-cell clones contained the p.L132M TP53 driver mutation (VAF > 0.99). No mutations were found exclusively in all fusion-positive clones ( Figure S4).
TRRUST (transcriptional regulatory relationships unraveled by sentence-based textmining) analysis for transcriptional regulatory relationships [34] revealed significant enrichment for genes primarily regulated by SP1 (n = 99 genes) and by NKFKB1 (n = 65 genes) in fusion-positive lines ( Figure S5A). The proximal promoter of ABCB1 contains several regulatory regions, including a GC rich element (position −56 to −42) required for constitutive promoter activity. Interestingly, several studies have shown that the transcriptional activator SP1 is the predominant factor that binds to this GC box within the ABCB1 promoter, mediating its activation in the absence and presence of genotoxic stress [35,36].
In addition to ABCB1, at least 16 other ABC transporters from 4 ABC subfamilies have been implicated in the transport of chemotherapeutic agents and/or in mediating MDR [12,37]. We therefore next investigated whether any other ABC family members are differentially expressed in fusion-positive cell lines. Out of the 35 ABC members examined, in addition to ABCB1, 2 ABC family members were significantly upregulated: ABCA3 and ABCC3, with ABCG1 significantly repressed (3.11, 2.71 and −1.65 log 2 fold change respectively) (p < 0.0001) ( Figure 2D). Both ABCA3 and ABCC3 have well-established roles in the efflux of anthracycline and methotrexate/epipodophyllotoxin chemotherapy substrates, respectively [38], with several studies reporting ABCA3 (known for its role in the production of pulmonary surfactant) to be upregulated in cisplatin-resistant ovarian cancer cells [15].
Gene set enrichment analysis (GSEA) of 34 Hallmark pathways revealed 12 pathways significantly enriched across both the cell line and patient RNAseq data sets (Figure 3). These pathways primarily concerned inflammatory response (inflammatory, allograft rejection, interferon alpha/gamma response pathways), myogenesis and epithelial-tomesenchymal transition (EMT). In contrast, only one pathway, fatty acid metabolism, was consistently downregulated across both cohorts. Together, our findings suggest that in addition to the direct upregulation of ABCB1, SLC25A40-ABCB1, fusions result in ECM and inflammatory-enriched transcriptional profiles. significantly enriched across both the cell line and patient RNAseq data sets (Figure 3). These pathways primarily concerned inflammatory response (inflammatory, allograft rejection, interferon alpha/gamma response pathways), myogenesis and epithelial-to-mesenchymal transition (EMT). In contrast, only one pathway, fatty acid metabolism, was consistently downregulated across both cohorts. Together, our findings suggest that in addition to the direct upregulation of ABCB1, SLC25A40-ABCB1, fusions result in ECM and inflammatory-enriched transcriptional profiles.

SLC25A40-ABCB1 Fusion-Positive Cells Display Altered Cellular Morphology
As the presence of SLC25A40-ABCB1 fusions drove a clear ECM/EMT transcriptional profile, we next investigated the morphological consequences of ABCB1 fusions in the cell lines. There was a significant downregulation of 10/16 (62.5%) well-established mesenchymal genes [39] in fusion-positive cell lines, with the zinc finger protein SNAI2 (4.52 log2 fold, p-adj < 0.0001) as the most repressed target ( Figure 4A). In contrast, 4/9 (44.4%) epithelial marker genes [39] were significantly upregulated in fusion-positive clones, with MUC1 (type-I transmembrane glycoprotein) showing the greatest induction (1.54 log2

SLC25A40-ABCB1 Fusion-Positive Cells Display Altered Cellular Morphology
As the presence of SLC25A40-ABCB1 fusions drove a clear ECM/EMT transcriptional profile, we next investigated the morphological consequences of ABCB1 fusions in the cell lines. There was a significant downregulation of 10/16 (62.5%) well-established mesenchymal genes [39] in fusion-positive cell lines, with the zinc finger protein SNAI2 (4.52 log 2 fold, p-adj < 0.0001) as the most repressed target ( Figure 4A). In contrast, 4/9 (44.4%) epithelial marker genes [39] were significantly upregulated in fusion-positive clones, with MUC1 (type-I transmembrane glycoprotein) showing the greatest induction (1.54 log 2 fold, p-adj < 0.0001). MUC1 has been shown to modulate cell-cell and cell-extracellular matrix interactions by steric hindrance, and its overexpression has been suggested to promote metastasis through disruption of these interactions [40].
fold, p-adj < 0.0001). MUC1 has been shown to modulate cell-cell and cell-extracellular matrix interactions by steric hindrance, and its overexpression has been suggested to promote metastasis through disruption of these interactions [40]. Phalloidin staining of F-actin filaments was conducted in AOCS18.5 fusion-negative (Clone D, Clone E) and fusion-positive (Clone 9 and Clone 18B) cell lines to assess phenotypic differences (general intensity, texture and shape features) ( Figure 4B). Of the actin Phalloidin staining of F-actin filaments was conducted in AOCS18.5 fusion-negative (Clone D, Clone E) and fusion-positive (Clone 9 and Clone 18B) cell lines to assess phenotypic differences (general intensity, texture and shape features) ( Figure 4B). Of the actin features assessed, 45/51 (88.2%) were significantly different (p < 0.05) ( Figure 4C). The top 4 features significantly modulated by ABCB1 fusion status were solidity (p = 1.076 × 10 −20 ), texture correlation (p = 7.907 × 10 −20 ), mass displacement (p = 2.727 × 10 −19 ) and compactness (p = 1.511 × 10 −18 ). Furthermore, IncuCyte cell-to-cell analysis revealed that although cell area was not significantly divergent, cellular texture (p = 0.0028) and eccentricity (p = 0.0053) was altered in fusion-positive lines ( Figure 4D). Together, these data suggest that in comparison to ABCB1 fusion-negative cells, drug-resistant fusion-positive cells may possess an epithelial state with stronger cell-cell adhesions and limited migratory potential.

Clonal Co-Operation between Fusion-Negative and Positive Cells Does Not Promote Proliferative Fitness
As we had previously demonstrated that SLC25A40-ABCB1 fusions are frequently subclonal [7,13] and that they endow a proliferative fitness penalty ( Figure 1B), we next investigated which population would outcompete the other when co-cultured in the absence of selective pressure to recapitulate clinical settings when patients are off treatment. Fusion-negative (Clone D) and fusion-positive (Clone 9) clones were transduced with lentiviral gene ontology (LeGO) vectors encoding either Venus Yellow or mCherry, seeded in a 50:50 ratio and co-cultured ( Figure 5A). As expected, fourteen days post seeding, FACS analysis revealed that 99.38% ± 0.069 of the cellular population was enriched for fusion-negative cells. Similarly, one month post co-culture only 0.02% ± 0.01 of the cellular population contained fusion-positive cells, with SLC25A40-ABCB1 mRNA expression undetectable ( Figure S6A-C).
To extend these findings we next determined whether potential cross-talk between fusion-negative and positive cells impedes or enhances the proliferative capacity of either population. Clone D and Clone 9 LeGO-labelled cells were seeded either alone or in three mixed ratios (25:75%, 50:50% and 75:25%) and grown for 11 days. Interestingly, IncuCyte proliferation analysis revealed no significant difference in proliferative capacity (intrinsic population growth rate) when cells were co-cultured for all ratios, possibly suggesting that drug-sensitive and drug-resistant cells do not co-operate to modulate the proliferative fitness of each other ( Figure 5B). However, at higher densities, resistant cells are outcompeted by sensitive cells, whilst sensitive cell abundances are barely impacted by the presence of the resistant competitors. The fitted Lotka-Volterra model of population growth and competition estimated the competitive effect of sensitive cells on resistant cells to be 3.5 times as great as the competitive effects that resistant cells had on them ( Figure 5C).
Finally, to examine how clonal composition influences chemosensitivity, 72 h drug sensitivity assays (doxorubicin, cisplatin and paclitaxel) were conducted with fusionnegative (Clone D and Clone E) and fusion-positive (Clone 9 and Clone 18B) clones seeded alone or in mixed ratios (25:75%, 50:50% and 75:25%). Most surprisingly, no significant differences in IC 50 values were observed in the co-culture assays when compared to 100% fusion-negative controls, with one exception in paclitaxel treated cells at 25:75% ratio (Table 1) ( Figure S7). For example, no significant difference in IC 50 values was demonstrated when doxorubicin assays were performed on mixed cellular populations containing 75% fusion-positive Clone 18B/25% fusion-negative Clone E cells, versus 100% Clone E fusion-negative cells (IC 50 46.33 ± 6.60 versus 43.53 ± 6.878 respectively). These findings highlight the importance of chemotherapeutic protocols, which maintain mixed clonal states to persevere drug sensitivity instead of regimens which drive the whole tumor population towards drug resistance. Cancers 2021, 13, x 16 of 28

High-Throughput Drug Screening Identifies FDA Agents That Induce Cytotoxic Responses Regardless of Fusion Status and Clonal Composition
To date, there remains no standard of care for patients whose tumors harbor ABCB1 fusion-mediated drug-resistant cancer cells. As the original AOCS18.5 patient-derived cell line contained a mixture of both ABCB1 fusion-negative and positive cells, our next goal was to identify rapidly translatable FDA-approved agents that fusion-negative/positive cells were equally susceptible to. High-throughput screening of 2907 FDA-approved compounds was performed in two AOCS18.5 fusion-negative lines (Clone D and Clone E), two fusion-positive lines (Clone 9 and Clone 18B) and a Clone D/Clone 9 (50:50 mix) to recapitulate a mixed population state. Primary screening identified 103 compounds in which normalized cell counts were <50% from control (DMSO) at a minimum of one dose point (0.05, 0.5, 5 µM) in all four cell lines (Table S2). In all, 36/103 (35.0%) compounds reduced viability across 2-3 dose points and included agents such as vincristine, topotecan, docetaxel and camptothecin. Using a viability Z-score criteria < −2 across a minimum of two dose points, 36 agents were nominated for further validation. Confirmatory assays were conducted using a six-point dose curve (0.01-10 µM) and 72 h of treatment.
As cellular division rate is a powerful confounder in the calculation of IC 50 values, the normalized growth rate inhibition (GR) method was used to correct for variation in division rates by estimating the magnitude of drug response on a per division basis [41]. Using this metric, no significant difference in GR 50 values (the concentration at which the effect reaches a growth rate (GR) value of 0.5) was observed between the ABCB1 fusion-negative and fusion-positive lines for 35/36 (97.2%) agents (Table 2) ( Figure S8). Of the 36 compounds, the DNA synthesis inhibitor bleomycin was the only agent identified in which GR 50 values were significantly different between fusion-negative and fusion-positive lines (0.258 µM versus 0.529 µM respectively) (p = 0.028). Due to high potency (>80% reduction in cell viability) at the lowest compound dose tested (0.01 µM), GR 50 were unattainable for three agents (colchicine, docetaxel and podofilox) ( Table 2). Indeed, the top five most potent (GR 50 < 0.02 µM) FDA-approved agents included mitoxanthrone, proscillaridin, triptolide, vinblastine and convallatoxin. Furthermore, regardless of fusion status, cell cycle profiles were consistent across cell lines with strong G0-G1 cell cycle arrest induced by 63.9% (23/36, fraction range 26.9-58.1%, 0.5 and 2.25 µM) of compounds followed by above G2-M arrest (11/36 compounds, fraction range 27.9-44.4%, 0.5 and 2.25 µM) ( Figure S9).
As the selective pressure of chemotherapy will drive the clonal evolutionary development of both drug-sensitive and drug-resistant populations within the same tumor, we next examined whether agents identified from our primary screen were equally cytotoxic for pure ABCB1 fusion-negative (Clone D) and mixed-fusion (Clone D and Clone 9, 50:50 mix) populations. Interestingly, no significant differences in GR 50 values were observed between the two populations for any of our 36 compounds (Table 2) ( Figure S8), highlighting that our identified FDA agents could provide new therapeutic avenues for relapsed HGSOC patients that harbor pure or mixed tumor populations.

Discussion
Cancer is a highly complex, adaptive system that can rapidly evolve new phenotypic and genotypic profiles to circumvent therapy. Indeed, the emergence of drug resistance remains our largest impediment in the quest towards curative cancer treatment, with an estimated 90% of all cancer related deaths attributed to chemoresistance [42]. Although the basis of drug resistance is multifactorial involving both tumor-and drug-related factors, overexpression of ABCB1 is the most common phenomenon employed by cancer cells to diminish the intracellular accumulation of chemotherapeutic agents, with approximately 50% of all anticancer agents used in the clinic effluxed by this transporter [43]. It is well established that high ABC transporter expression is associated with worse survival outcomes [44]. Indeed, several studies, including a meta-analysis of 38 retrospective studies assessing 8067 epithelial ovarian cancer cases, demonstrated that ABCB1 over-expression was a significant risk factor associated with unfavorable overall and progression-free survival [45,46]. As it has become increasingly apparent that MDR is a multifactorial phenomenon, we need to shift our focus from the development of targeted ABCB1 agents to therapeutic strategies that can disrupt ABCB1-mediated MDR-contributing factors.
We show that in addition to driving drug resistance, ABCB1 fusions significantly reduced the proliferative fitness of HGSOC cells in the absence of the selective pressure of chemotherapy. We hypothesize that this is primarily due to the higher bioenergetic demand of fusion-positive cells, as evidenced by elevated basal respiration, necessary for the movement of endogenous molecules such as cholesterol, phospholipids and sphingolipids. Few studies have focused on the role of ABCB1-associated malignant proliferation. The stable knockdown of Mdr1a/1b in mouse colon carcinoma cells significantly inhibited cellular growth both in vitro and in vivo [47]. Similarly, the suppression of intestinal tumorigenesis (small intestinal polyps) was observed in Mdr1a/b knockout mice (APC mutant background) [48]. Interestingly, Mdr1a/1b −/− mice are viable and show no functional deficiency in terms of fertility, and abnormalities across a range of histological, haematological, serum-chemical, and immunological parameters [49]. The clear lack of human-based ABCB1 knockout studies warrants further investigations into ABCB1's role in the regulation of cellular proliferation and how this "fitness" cost can be exploited therapeutically.
The expenditure of energy required to maintain the molecular machinery that governs MDR phenotypes is not evolutionarily favored except when chemotherapy is administered. In this case, the energy demands required to drive ABCB1's drug export function increases survival and therefore confers increased fitness. However, in the absence of chemotherapy, the cost of ABC pumps serves no survival benefit and therefore reduces fitness due to the added energetic cost. For many years, clinical practice has been dominated by the concept of the maximum tolerated dose (MTD), which drives the development and evolution of drug resistance through the intense selection pressures imposed by cytotoxic therapies. This strategy thereby essentially accelerates the proliferation of drug-resistant cells by selecting for resistant clones and eliminating all competing drug-sensitive populations [50,51]. We have clearly demonstrated the proliferative deficiencies of drug-resistant ABCB1 fusion-positive cells, and this can potentially be exploited through adaptive therapy to inhibit population expansion. Adaptive therapy capitalizes on the competitive interactions between drug -sensitive and drug-resistant subclones to maintain a controllable stable cancer population below a certain symptomatic threshold whilst maintaining a substantial population of treatment-sensitive cells [1]. As growing evidence suggests that cancer cells behave as communities, by manipulating Darwinian evolutionary principles, this approach aims to further suppress the proliferation of less fit resistant populations. Indeed, our clonal mixture drug assays demonstrated that even if tumor populations contain as little as 25% ABCB1 fusion-negative drug-sensitive cells (75% fusion-positive), chemotherapeutic IC 50 values (doxorubicin, paclitaxel and cisplatin) were virtually the same as 100% fusionnegative alone, underscoring the importance of maintaining drug-sensitive populations to outcompete and compete with resistant communities during off-treatment periods. Finally, our high-throughput drug screening efforts importantly identified 36 FDA compounds that exert equal cytotoxicity in fusion-negative and positive clones (pure and mixed populations). Further investigations into agents whose primary mode of action is not dependent on cell cycle progression rates (e.g., inhibitors of microtubule polymerization) are warranted. In particular, enrichment for agents that target Na + /K + -ATPase (e.g., digitoxin, convallatoxin) and DNA topoisomerase I (e.g., camptothecin, irinotecan) was observed.
One of the striking findings revealed from our study was the clear enrichment for genes associated with the matrisome (ECM markers) and EMT in fusion-positive cells. This may be the driving mechanism for their observed altered actin feature morphology (compactness, eccentricity and texture etc.). Our RNA-seq analysis revealed the putative tumor suppressor AJAP1 as the most highly expressed gene in fusion-positive lines. AJAP1 is a type-1 transmembrane protein that localizes and interacts with the E-cadherin-catenin complex and is involved in cellular processes such as cell migration and invasion by modulating adherens junctions and remodeling the ECM and cytoskeleton [52]. For example, in nonpolarized, highly migratory and invasive cells, AJAP1 interacts with the transmembrane glycoprotein CD147, an invasion-promoting protein [53]. The stable overexpression of AJAP1 in MCF7 breast cancer cells accelerated cell migration with knockdown decreasing migratory behavior [54]. Paradoxically, the overexpression of AJAP1-attenuated glioblastoma cell line adhesion capacity to extracellular matrix components (laminin, collagen IV and fibronectin), with delayed wound-healing closure only observed on fibronectin-coated plates [55]. Concordant with this study, the knockdown of AJAP1 enhanced the migratory and invasive behavior of primary endothelial HUVEC cells [52].
Cancer invasiveness has long been associated with increased drug resistance, yet little is known about the molecular mechanisms linking these two processes. Indeed, the highly conserved cellular process of EMT has emerged as a major contributor to therapy resistance by permitting polarized, immobile epithelial cells to transform into mesenchymal mobile cells due to loss of apico-basal polarity and cell-cell contacts [56].We observed strong repression of several mesenchymal genes in fusion-positive lines including SNAI1 (SNAIL), SNAI2 (SLUG), Vimentin (VIM) and TWIST. Interestingly, a bioinformatic-based analysis of the promoter regions of 16 ABC transporters by Saxena et al. revealed binding sites for several EMT-inducing transcription factors including SNAI1, SNAI2, TWIST, E12, E47 and FOXC2, with TWIST, SNAI1, and FOXC2 capable of increasing the promoter activity of ABC transporters [57]. However, it must be noted that the fused transcript identified in AOCS18.5 cells is a result of a 250 kb intergenic deletion, fusing the promoter and non-coding exon 1 of SLC25A40 to exon 2 of ABCB1 [7]; hence EMT transcription factor ABCB1 promoter sites are absent.
Although direct evidence linking ABC transporters and metastasis is limited, roles are emerging for these proteins in cell migration and invasion. Attenuation of ABCB1 reduced the migratory capacity of both MCF-7 breast carcinoma and rat brain endothelial cells, with overexpression associated with increased migration [58,59]. Increased migratory ability of MCF-7-ADR-1024 doxorubicin-resistant cells was also exhibited compared to wildtype control lines [60]. Finally, chemotactic response and migration of peripheral dendritic cells to lymph nodes is significantly reduced in Abcc1 −/− mice [61]. Although we demonstrated strong transcriptional repression of mesenchymal programming, it is unknown whether patients whose tumors contain fusion-positive cells exhibit a greater degree of metastatic disease or whether these cells have a proclivity to remain within the ascites peritoneal fluid.

Conclusions
ABC transporters have a pivotal role in host cell detoxification and protection of the body against xenobiotics. However, a considerable body of evidence also points to their fundamental roles in tumor biology. This is the first study to provide significant insights into the phenotypic, metabolomic and transcriptional consequences of SLC25A40-ABCB1 fusions beyond drug resistance. We show that during treatment "holidays", the proliferative fitness deficits of fusion-positive cells will allow fusion-negative cells to outcompete drug-resistant populations. More importantly, we have identified FDA-approved agents that induce equal cellular cytotoxicity regardless of fusion status. Taken together, our findings will have significant implications in guiding changes to current HGSOC treatment regimens, facilitating our ultimate goal of long-term cancer control. From a translational perspective, steering the evolutionary dynamics of ABCB1 acquired resistance through the inhibition of metabolism or ECM may represent a novel therapeutic approach for ABCB1-mediated acquired resistance in HGSOC patients as well as for all ABCB1-driven MDR cancers.