High throughput screening identifies dasatinib as synergistic with trametinib in low grade serous ovarian carcinoma

• We performed a high throughput screen of 1610 compounds across six low grade serous ovarian carcinoma cell lines • Of 16 prioritised screen hits, 11 compounds passed dose-response validation using cell viability assays • The most promising hits underwent synergy pro ﬁ ling with the trametinib (dasatinib, disul ﬁ ram, car ﬁ lzomib, romidepsin) • Disul ﬁ ram demonstrated excellent selectivity for LGSOC versus high grade serous ovarian carcinoma comparator lines • Dasatinib demonstrated favourable synergy with the MEK inhibitor trametinib


Introduction
Low grade serous ovarian carcinoma (LGSOC) is a unique and uncommon ovarian cancer type characterised by younger patient age and high levels of intrinsic chemoresistance compared to more common high grade serous ovarian carcinomas (HGSOC) [1].The overall survival time for LGSOC patients is around 10 years from diagnosis [2].While this is longer than many other ovarian cancer histotypes [3] -including HGSOC -this is largely due to prolonged post-relapse survival, and quality of life during this time is significantly impacted by incrementally advancing disease and accumulating toxicity [1].
The objective response rate of LGSOC to first-line platinum-based chemotherapy regimens is ≤25% [1,4,5], and the response rate to common cytotoxic regimens in the recurrent setting is ≤15% [6].Treatment of advanced, progressive and recurrent disease therefore represents a major clinical challenge [6].Poor efficacy of cytotoxic chemotherapy has resulted in a growing interest in molecularly-targeted therapeutics for LGSOC management [1].The most common molecular features of LGSOC include frequent mutational activation of the RAS-RAF-MEK-ERK pathway, with 33%, 10% and 10% of patients demonstrating KRAS, NRAS and BRAF mutations (RAS/RAF-mutant), respectively [7][8][9].Genomic perturbations in USP9X, EIF1AX and CDKN2A are also recurrent in LGSOC, but are present at a lower frequency than RAS/RAF pathway mutations [9].LGSOC are ubiquitously oestrogen receptor (ER) positive, and frequently express progesterone receptor (PR) [10,11].Together, these molecular features have suggested RAS/RAF pathway inhibition and endocrine therapies as potential treatment strategies [6].
Recently, the MEK inhibitor trametinib was shown to improve progression-free survival in recurrent or persistent LGSOC, with a response rate of 26% [12].However, it is clear that not all patients benefit from trametinib, and that not all responses are durable.In particular, response rates in patients whose tumours do not harbour KRAS, BRAF or NRAS mutation appears low (8% in GOG281/NCT02101788) [12].Moreover, trametinib treatment is associated with substantial toxicities, including fatigue, skin rashes and hypertension [12].Endocrine therapy has also demonstrated considerable efficacy in LGSOC [1,13]; however, these agents are typically cytostatic in nature, delaying progression more commonly than inducing objective responses.While both endocrine therapy and MEK inhibitors (trametinib [12], binimetinib [14]) are now included in the recommendations for LGSOC management [15] (for first-line maintenance and treatment of recurrent disease, respectively), there is a clear need for additional treatment strategies in LGSOC.In particular, there is a lack of effective targeted treatment options for patients with RAS/RAF-wildtype tumours.
Repurposing of compounds used in other disease settings -including other cancer types -represents an attractive strategy for fast-tracking agents into clinical studies, as these agents have established toxicity profiles and dosing schedules [16].This strategy is particularly attractive for uncommon disease types such as LGSOC where there is a paucity of representative in vivo models available for extensive pre-clinical testing of novel compounds.Here, we seek to implement a high throughput screening approach across a panel of heterogeneous LGSOC lines to identify novel treatment strategies that may benefit LGSOC patients.

LGSOC cell lines
VOA1056, VOA3723, VOA4627, VOA4698, VOA6406, VOA7681 and VOA13738 were provided by Dr. Carey from the University of British Columbia (table S1); these cells lines were derived from LGSOC patients as previously reported [17,18].HOC7 cells [19] were provided by the MD Anderson Cancer Centre.HO433 was derived from an LGSOC patient treated at the Edinburgh Cancer Centre; ascites drained from the abdomen of a patient with relapsed LGSOC previously treated with platinum-taxane combination therapy was cultured and differentially trypsinised to produce pure malignant epithelial cells.These cells underwent serial passage and were considered to be immortal after >20 passages and repetitive freeze-thaw recovery.

Cell line authentication and mycoplasma testing
All cell lines were authenticated by short tandem repeat profiling prior to experimentation using the Promega GenePrint 10 System (Promega, #B9510).Mycoplasma testing was undertaken every 3-6 months using the Lonza MycoAlert Detection Kit (Lonza #LT07-318) to confirm absence of mycoplasma infection.
Cell cultures were trypsinized with trypLE express (Gibco #12605010), counted using the Countess II Automated Cell Counter (Thermofisher Scientific), resuspended in media to the desired concentration, and then seeded into barcoded 384-well plates for screening using a ViaFill reagent dispenser (Integra).Cell seeding densities were optimised for each cell line to achieve approximately 70% confluence at the end of the treatment period in vehicle-treated conditions (range 300-3500 cells per well).Seeded cells were incubated for 24 h prior to drug treatment for 3 days at a single-dose of 1000 nM per compound.All wells were normalised to 0.1% DMSO (v/v) with controls (negative: 0.1% DMSO; positive: 1 μM staurosporine).After the treatment period, LGSOC cells were fixed in situ by adding an equal volume of 8% formaldehyde for 30 mins before washing twice with PBS.

Screen hit identification
Fixed cells underwent staining to identify cell nuclei using Hoechst-33342 (Thermofisher Scientific).Stained cells were imaged at 20× using the ImageXpress Micro Confocal with robotic plate loader (6 sites per well per channel).Images were analysed using MetaXpress built-in analysis modules and the data were normalised and visualised via TIBCO Spotfire Analyst.Total nuclei counts (sum nuclei counts over 6 sites) were normalised to DMSO vehicle control wells on a plate-byplate basis.Screening QC was carried out (including Z-prime analysis) and screen hits were defined as compounds significantly reducing nuclei count in compound-treated wells versus control wells by >50% (Z-score < −5.5) across more than half of the screened LGSOC cell lines (≥4 of 6 lines).

Hit validation
Drug screen hits were validated in dose-response studies of resupplied compounds in 96-well cell viability assays.Individual compounds were purchased from Selleckchem for use in validation assays (supplementary methods section 2).NXP900 was supplied by the Unciti-Broceta lab at the CRUK Scotland Centre [23,24].Seeding densities were optimised for each cell line to achieve approximately 70% confluence in vehicle-treated wells at the end of the experiment.Cells were seeded into CELLSTAR 96-well plates (Greiner #655180), and incubated for 24 h prior to treatment for 5 days using an 8-point 1:3 serial dilutions of test compounds; the start dose was adjusted as necessary for each cell line based on individual sensitivity.Viability of cells treated at each experimental dose were quantified using the Alamar Blue High Sensitivity Assay (supplementary methods section 3) relative to vehicle-treated control wells.Staurosporine treatment (1000 nM concentration) was used as a positive cell death control.
Dose-response data were used to calculate IC50 values, reflecting overall sensitivity of cells to the compound of interest, using the four- parameter log-logistic function.Compounds with an IC50 < 1000 nM across ≥4 of the 6 models included in the high throughput screen were considered to pass validation.
For validated compounds, dose-response profiling was expanded to a total of 9 LGSOC cell lines (2 KRAS mutant, 2 NRAS mutant, 5 KRAS/ BRAF/NRAS wildtype) (table S1) and 5 HGSOC comparator lines to provide a broader overview of sensitivity ranges and selectivity for LGSOC cells.Presented data are from three independent replicates, with each replicate comprising three technical repeats (N = 9 data points per cell line per compound per dose).
Cells were seeded into 96-well plates; 24 h after seeding, cells were treated in a 6 × 6 dose combination matrix for 5 days using 1:3 serial dilutions of trametinib and the drug of interest.Starting doses were optimised per cell line using the single agent dose-response data as a guide.Three plates were seeded and dosed per biological replicate to provide technical triplicates.After 5 days of treatment, viability was determined using the Alamar Blue High Sensitivity Assay (supplementary methods section 3).

Statistical analyses
All analysis was performed using R 4.2.2 in RStudio 2022.07.2 + 576.Single-agent dose-response data were analysed using the drc R package (version 3.0.1)[25]; IC50 values were calculated using the drm function and four-parameter log-logistic dose response models.Hierarchical clustering of IC50 values was performed using Euclidean distance and Ward's linkage on a matrix of log2 nM concentrations across cell lines.Combination data were analysed using the SynergyFinder R package (version 3.6.3)[26] to calculate zero interaction potency (ZIP) synergy scores for each combination dose.ZIP score >10 was used as a threshold to identify likely synergistic effects, as recommended in the SynergyFinder documentation [26].

High throughput screening of LGSOC cell lines
Six cell lines representing both RAS/RAF-mutant (n = 3: HOC7 KRAS G12A, VOA7681 KRAS G12V, VOA1056 NRAS G61R) and RAS/RAFwildtype (n = 3: HO433, VOA3723, VOA4698; all KRAS, BRAF and NRAS wildtype) (Fig. 1, Fig. 2, table S1) underwent high throughput screening in 384-well format with 1610 compounds from the Prestwick FDA-Approved Compound Library and a TargetMol Anti-Cancer Compound Library (Fig. 1).24 h after seeding, cells were treated with a single dose (1000 nM) of each compound for 3 days, after which cells were fixed and stained to identify cell nuclei.Cell nuclei counts were quantified from images and normalised against vehicle-treated wells.
Univariate analysis of normalised nuclei counts across screening plates, alongside quality control and normalised Z-score distributions, are shown in fig.S1; Z prime analysis was rated good, ranging from 0.46 to 0.92.Z-Z-score plots and hierarchical clustering, comparing hits across cell lines, highlighted VOA1056 as highly resistant to many of the compounds (fig.S1).In total, 72 agents (4.5%) were identified as hits from the screen based on nuclei count (>50% reduction in nuclei count compared to vehicle controls across ≥4 of 6 assayed cell lines; Zscore < −5.5).Of these, 27 represented cytotoxic chemotherapeutics (table S2), 4 were agents of known efficacy in LGSOC (2 MEK inhibitors, trametinib and selumetinib; 2 aromatase inhibitors, letrozole and formestane), and 4 were agents duplicated across the two component libraries, leaving 37 agents of potential interest.Of these, 16 agents were prioritised for further investigation to represent the range of drug classes and mechanisms of action (table S2).

Comparison with HGSOC cells
To identify agents with specificity for LGSOC cells, dose-response analysis was expanded to a larger panel of 9 LGSOC cell lines and IC50 values were compared against a panel of 5 HGSOC cell lines (OVCAR3, OVSAHO, OVCAR4, 59M and SNU119) (Fig. 3, Fig. 4, table S3, appendix A).Analysis of trametinib sensitivity as a benchmark for an agent of known efficacy was also assessed using this approach, demonstrating a range of sensitivity across LGSOC lines, with a large proportion demonstrating greater sensitivity versus to the HGSOC comparator panel (Fig. 3E).The KRAS-mutant models HOC7 and VOA7681 demonstrated marked sensitivity to trametinib (IC50 5 nM and 24 nM) (Fig. 3E).
Of the 11 agents of interest, 7 demonstrated a low dynamic range of sensitivities with poor selectivity for LGSOC over HGSOC cells (selinexor, volasertib, panobinostat, albendazole, ixazomib, homoharringtonine, dinaciclib) (Fig. 4).Disulfiram demonstrated a range of sensitivity across cell lines with favourable selectivity for LGSOC over HGSOC models (P = 0.003) (Fig. 3B, fig.S2).Dasatinib also demonstrated a range of sensitivities, with two broad groups of models (sensitive vs more resistant) (Fig. 3A).Carfilzomib (proteasome inhibitor) and romidepsin (HDAC inhibitor) demonstrated a range of sensitivities across models, with a number of LGSOC cell lines demonstrating high sensitivity to these agents (IC50 < 10 nM) (Fig. 3C and D).
Overall, VOA1056 and HO433 were highly resistant to most agents; hierarchical clustering of IC50 values across compounds grouped these cell lines alongside the HGSOC comparator cells (Fig. 4H).
Carfilzomib did not demonstrate synergy with trametinib across multiple dose combinations in any of the tested models (fig.S3).Romidepsin demonstrated evidence of synergy across multiple dose combinations in HO433 cells (ZIP-max 34.7), but this appeared limited to the highest trametinib doses.No evidence of romidepsin-trametinib synergy was identified in HOC7 or VOA7681 (ZIP-max 2.2 and 6.4) (fig.S4); synergy scores in VOA13738 suggested a romidepsintrametinib combinatory effect at the threshold between additivity and synergy in VOA13738 (ZIP-max 11.5).

Selective SRC inhibition is synergistic with trametinib
As dasatinib inhibits SRC family kinase (SFK) members, ABL, c-KIT and a large number of other tyrosine kinases [27], we sought to further elucidate the mechanism-of-action of synergy between dasatinib and trametinib.Several potent inhibitors of ABL and c-KIT were included in the compound library used in our screen (nilotinib, bosutinib, imatinib, pazopanib), but none of these agents were identified as hits across any of the screened cell lines.We therefore investigated whether combining a selective SFK inhibitor produced a similar synergy profile with trametinib.A novel, highly selective SFK inhibitor -NXP900 (formerly known as eCF506) [23,24] demonstrated markedly similar synergy profiles to dasatinib across LGSOC models (Fig. 6).High levels of synergy were identified in VOA7681 (ZIP-max 41.7), notable synergy was identified in HOC7 (ZIP-max 17.1) and VOA13738 (ZIP-max 18.3), while no synergy was apparent in the trametinib-resistant HO433 cell line (ZIP-max 2.4).

Discussion
LGSOC is an uncommon and chemoresistant ovarian cancer type that typically affects younger women [2].While recent advances have expanded treatment options for LGSOC [15] (endocrine agents [13,28], MEK inhibitors [12,14]), intrinsic and acquired resistance highlights the need for additional treatment options.Moreover, while the MEK inhibitor trametinib represents the only drug with positive randomised late-phase trial data in this disease, treatment is associated with significant toxicity, impacting patient quality of life [12].Identifying agents that act synergistically with trametinib may increase the magnitude and duration of response to this therapy, while also allowing dose reduction to avoid treatment-related toxicities.Notably, poly-(ADP-ribose) polymerase (PARP) inhibitorswhich have represented  S3. a step-change in the management of HGSOC [29] are not considered to be of interest in LGSOC due to the absence of homologous recombination repair defects in this tumour type [2,30].
Our high throughput screen identified a relatively large number of compounds with substantial efficacy (>50% reduction in nuclei counts versus vehicle) across over half of the cell lines included.These hits included multiple agents of known efficacy in LGSOC, including trametinib and selumetinib.Our hits included several drugs with shared mechanisms-of-action, including multiple proteasome inhibitors, multiple MEK inhibitors, and multiple HDAC inhibitors.16 compounds were prioritised for dose-response hit validation studies using resupplied compounds.11 of these compounds passed validation analysis.As our screen included only LGSOC cells without comparator lines, we expanded dose-response profiling of our 11 agents of interest to a total of 9 LGSOC lines and a panel of 5 HGSOC comparator lines to identify agents that may be selective for LGSOC cells.Using the doseresponse profile of trametinib as a reference for a drug of known interest and clinical activity, we identified four compounds of greatest interest based on a spectrum of sensitivity across lines with or without selectivity for LGSOC over comparator cells: dasatinib, a tyrosine kinase inhibitor; disulfiram, an ALDH inhibitor; carfilzomib, a proteasome inhibitor; and romidepsin, an HDAC inhibitor.
Disulfiram showed excellent selectivity for LGSOC, with cell lines showing uniformly higher sensitivity versus the HGSOC comparator cells.Disulfiram is currently used for management of alcohol dependency, irreversibly inhibiting ALDH to induce acute alcohol sensitivity; however, it has recently received research attention for potential utility as an anti-cancer agent [31].While some investigators have attributed disulfiram's anti-cancer action to its ALDH-inhibitory activity [32], many others have suggested alternative anti-cancer mechanisms, such as intracellular copper accumulation and associated cell death [20,31].Other proposed mechanisms include inhibition of angiogenesis, interference with redox homeostasis and inhibition of drug efflux pumps [31].
Synergy analysis with trametinib identified dasatinib as a potentially useful compound for combination treatment, with multiple LGSOC lines demonstrating favourable synergy scores across multiple dose combinations.Synergy was observed in both RAS/RAF-mutant and RAS/ RAF-wildtype models, though the effect was not seen in HO433 cells, which are RAS/RAF-wildtype and demonstrate high intrinsic resistance to trametinib (IC50 > 1000 nM).Dasatinib inhibits a wide range of kinases, including SFKs, ABL, c-KIT and many others.To investigate whether the synergistic relationship to trametinib was related to its SFK inhibitory action, we performed trametinib combination studies with NXP900, a highly selective SFK inhibitor currently undergoing first-in-human phase I dose-escalation studies [23,24].NXP900's selectivity for SFK is in part due to its unique binding mode, as it blocks SRC and its family members in their native inactive conformation [23].The highly divergent structures of inactive kinase conformations make NXP900 only optimal for SFK binding.In contrast, dasatinib binds to its targets in their active conformation, which exhibits significant structural homology across receptor and non-receptor tyrosine kinases.Comparison of dasatinib and NXP900 synergy profiles with trametinib revealed a striking correlation, strongly suggesting SFK inhibition is a key driver of dasatinib's synergy with trametinib.
Combining MEK inhibition with other small molecule inhibitors is already an area of keen interest in LGSOC.A phase I evaluation of combining the dual RAF/MEK inhibitor avutometinib with defactinib, a focal adhesion kinase (FAK) inhibitor, reported an impressive response rate of 46% in 24 LGSOC patients [33], and the response rate of this combination was replicated in the phase II RAMP-201 study [34].Combined FAK and MEK inhibition is of interest across multiple RAS/RAF-driven tumour types [35,36], and FAK/integrin signalling has been implicated in MEK inhibitor resistance [37,38].Activated FAK is pro-tumourigenic, promoting cell survival through a variety of pathways identified across multiple cancer types, including through Hippo and WNT pathway activation, and modulation of the tumour microenvironment [38].Early in FAK signalling activation, autophosphorylated FAK acts as a scaffold, recruiting SRC and other family members into a complex that phosphorylates additional FAK tyrosine residues to mediate downstream signalling from focal adhesions.Together with our in vitro data, the known cooperation between SRC and FAK to mediate protumourigenic signalling [38] alongside the early phase trial data demonstrating favourable responses to combined FAK and MEK inhibition [33] suggest that combined SRC and MEK inhibition may be a feasible strategy of clinical interest in LGSOC.A key consideration for combination therapy is overlapping toxicity; hypertension, fatigue, anaemia, skin rash, nausea and diarrhoea have been identified as the most frequent grade 3 or 4 adverse events associated with trametinib treatment in LGSOC [12].Based on adverse reaction data from chronic myeloid leukaemia [39], haematological toxicity is likely to represent the most relevant potential overlapping toxicity between trametinib and dasatinib.However, identified synergy between these compounds may enable dose reduction to avoid such toxicities.Validation of next generation SFK inhibitors with significantly reduced off-target activity (such as NXP900) may provide future opportunities for refined SRC-MEK inhibitor combination strategies with high tolerability.
A major strength of this study is that we specifically utilized a compound library enriched for FDA-approved agents to identify potential opportunities for drug repurposing.Drugs already in use for other clinical indicationsor that have already been tested in phase I-III trials for other cancer typeshave established dosage and toxicity profiles, facilitating rapid translation into early phase clinical studies.Including a balance of RAS/RAF-mutant and RAS/RAF-wildtype cell lines in our initial screen to represent the two major genomic subgroups of LGSOC (RAS/RAF-mutant, 50%; RAS/RAF-wildtype, 50%) is also a major strength [7].Around 50% of LGSOCs do not harbour a canonical RAS/ RAF pathway mutation, and current data suggest improved response to new agents such as MEK inhibitors in RAS/RAF-mutant cases [12,14].Including models representing the spectrum of LGSOC genomic subtypes in studies seeking to identify new treatment strategies is essential for maximising the likelihood that new candidate treatments have the potential to confer clinical benefit to the wider patient population.A limitation of the study is that we were unable to include models harbouring every known genomic event identified in LGSOC; rare mutations across a large number of genesincluding RAS-RAF-associated genes (NF1, MAP2K1, MAP3K1, MAP4K1; all <5% cases) and others (EIF1AX, PIK3CA; 5% of cases) [8,9] are known to occur in LGSOC, but few LGSOC models harbouring these events are available.Our findings are also limited to in vitro analysis due to the scarcity of more sophisticated and disease-relevant LGSOC model systems.However, given the current lack of preclinical evidence supportive of other treatment strategies, the present study represents a significant advancement in the identification of further candidate therapies for evaluation in clinical studies.

Conclusion
We identify dasatinib and disulfiram as the most promising agents for clinical investigation in LGSOC.The established dosing schedules and toxicity profiles of these agents from their existing indications presents the opportunity to fast-track these drugs into clinical studies.Dasatinib demonstrates synergy with the MEK inhibitor trametinib, and corresponding synergy in a highly selective SFK inhibitor highlights SFK inhibition as the likely driver of this synergy.Synergistic combinations with trametinib may enable dose reduction to avoid toxicities associated with MEK inhibition, while also increasing overall response rates.

Fig. 2 .Fig. 3 .
Fig. 2. Morphology of low grade serous ovarian carcinoma (LGSOC) cell lines included in the high content screen.Images are merged channels showing phalloidin (F-actin) in green, Hoechst-33342 (cell nuclei) in blue and Concanavalin A (endoplasmic reticulum) in red.RAS/RAF-mutant cell lines are in the top row, RAS/RAF-wildtype cell lines are in the bottom row.Scale bar represents 150 μm.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)