A synthetic lethal screen for Snail-induced enzalutamide resistance identifies JAK/STAT signaling as a therapeutic vulnerability in prostate cancer

Despite substantial improvements in the treatment landscape of prostate cancer, the evolution of hormone therapy-resistant and metastatic prostate cancer remains a major cause of cancer-related death globally. The mainstay of treatment for advanced prostate cancer is targeting of androgen receptor signaling, including androgen deprivation therapy plus second-generation androgen receptor blockade (e.g., enzalutamide, apalutamide, darolutamide), and/or androgen synthesis inhibition (abiraterone). While these agents have significantly prolonged the lives of patients with advanced prostate cancer, the evolution of resistance to these treatments in nearly universal. This therapy resistance is mediated by diverse mechanisms, including both androgen receptor-dependent mechanisms, such as androgen receptor mutations, amplifications, alternatively spliced isoforms, and structural rearrangements, as well as non-androgen receptor-mediated mechanisms, such as lineage plasticity toward neuroendocrine-like or epithelial-mesenchymal transition (EMT)-like lineages. Our prior work identified the EMT transcriptional regulator Snail as critical to hormonal therapy resistance and commonly detected in human metastatic prostate cancer. In the current study, we sought to interrogate the actionable landscape of EMT-mediated hormone therapy-resistant prostate cancer to identify synthetic lethality and collateral sensitivity approaches to treating this aggressive disease state. Using a combination of high-throughput drug screens and multi-parameter phenotyping by confluence imaging, ATP production, and phenotypic plasticity reporters of EMT, we identified candidate synthetic lethalities to Snail-mediated EMT in prostate cancer. These analyses identified multiple actionable targets, such as XPO1, PI3K/mTOR, aurora kinases, c-MET, polo-like kinases, and JAK/STAT as synthetic lethalities in Snail+ prostate cancer. We validated these targets in a subsequent validation screen in an LNCaP-derived model of resistance to sequential androgen deprivation and enzalutamide. This follow-up screen provided validation of inhibitors of JAK/STAT and PI3K/mTOR as therapeutic vulnerabilities for Snail+ and enzalutamide-resistant prostate cancer.


58
The treatment landscape of prostate cancer exemplifies the "two truths" of cancer treatment [1]:

59
While tremendous progress has been made to improve patient outcomes, there also remains an 60 urgent need to overcome the significant challenges imposed by the evolution of treatment 61 resistance and metastasis. From the groundbreaking studies of Huggins and Hodges [2] to the 62 development of novel, second-generation androgen receptor inhibitors [3][4][5][6][7][8], and anti-androgens 63 [9,10], much of the existing treatment options for prostate cancer are currently focused on 64 targeting the androgen receptor (AR) signaling axis. These agents have demonstrated 65 significant clinical benefit; however, progression of men treated with these agents in the 66 metastatic, castration-resistant setting is nearly universal.

67
The evolution of resistance to AR signaling inhibitors is mediated by heterogeneous genetic and 68 non-genetic pathways that include both AR-dependent and AR-independent mechanisms 69 (reviewed in [11]). Among these heterogeneous mechanisms, phenotypic plasticity is a central 70 hallmark of AR signaling inhibitor resistance [12]. This phenotypic plasticity occurs along 71 multiple, interconnected cellular lineage axes, such as stemness [13,14], 72 epithelial/mesenchymal [15][16][17][18], luminal/basal [19,20], and neuroendocrine-like lineages or cell 73 states [21,22]. Phenotypic plasticity along these axes often leads to a loss of AR 74 expression/activity and dependency [23], as well as additional aggressive features that promote 75 survival and metastasis [24,25]. New approaches are needed to capitalize on these emerging 76 phenotypic states for therapeutic benefit.

77
Targeted therapy alters the ecological fitness landscapes of cancer in multiple ways [26]. The 78 altered fitness landscape of the drugged environment can promote aggressive biology, but can 79 also induce "collateral sensitivities" to novel agents [27]. This concept, also known as negative 80 cross resistance, has been applied to identify new strategies to treat the evolution of resistance 81 in bacterial infections [28], malaria [29], herbicides [30], and pesticides [31].

82
In the present study, we combined high-throughput screens with multiparameter endpoint 83 measurements from transcription-based reporters, confluence, and cell viability assays to 84 characterize the therapeutic landscapes of Snail-mediated EMT, enzalutamide resistance, and 85 AR activity (Fig. 1A). Our analyses pinpoint histone deacetylases (HDAC), protein kinase A 86 (PKA), PI3K/mTOR, and Janus Kinase (JAK) as key collateral sensitivities to Snail-mediated 87 enzalutamide resistance in prostate cancer cells. Follow-up screens in a model of progressive 88 adaptation to ADT and enzalutamide resistance verified the relevance of these pathways as 89 novel therapeutic vulnerabilities for enzalutamide-resistant prostate cancer (Fig. 1B). These A high-throughput screen was performed in LNCaP95-Snail cells to assess differential response across multiple endpoints of confluence, viability (CellTiter Glo), and EMT status via a fluorescence-based reporter. B. Screen schematic for a collateral sensitivity screen in enzalutamide-resistant CS2 cells. Endpoints included PSA reporter response, confluence, and viability (CellTiter Glo). were measured with and without EMT induction using Snail activation as described above. For 114 PSA-GFP expressing cells, confluence and fluorescence was quantified with and without AR 115 activation using synthetic androgen R1881 at 1 nM.

116
High-throughput drug screening. High-throughput screens were performed in collaboration with 117 the Duke Functional Genomics Shared Resource as previously described [34][35][36][37]. Briefly, 118 compounds from the Bioactives library (SelleckChem) were stamped in triplicate into 384-well 119 plates at a final concentration of 1 μM using an Echo Acoustic Dispenser (Labcyte, Indianapolis, 120 IN, USA). Cells and media were subsequently dispensed into plates using a WellMate (Thermo 121 Fisher, Waltham, MA, USA) at a density of 2,000 cells/well for each cell line. Confluence was 122 quantified using an IncuCyte S3 live cell imaging system. GIIIcI 2 and PSA-GFP readouts were 123 quantified by IncuCyte imaging at 24, 48, 72, and 96 hours. CellTiter Glo was added at 96 124 hours, and luminescence was read using a Clariostar plate reader (BMG, Berlin, Germany).

125
RNA-Seq analysis of EMT scores. Quantification of EMT status for each sample was performed 126 using the following three independent methods: 76GS, KS, MLR, each of which uses a unique 127 algorithm and gene set. The 76GS scores were calculated based on the expression of 76 genes 128 [38]. Higher scores correspond to more epithelial states. A 76GS score > 0 typically indicates an 129 epithelial phenotype and < 0 indicates a mesenchymal phenotype. The score for each sample is

151
Imaging of confluence and GFP was compared using repeated measures ANOVA. Linear 152 regression was used to assess correlations between screen analysis parameters, and outliers 153 were considered to fall outside the 95% confidence interval bands. P-values <0.05 were 154 considered statistically reliable.

156
Fluorescence-based reporters enable real-time monitoring of epithelial plasticity. Prior studies 157 have pinpointed the epithelial plasticity regulator, Snail, as both upregulated during AR inhibition

158
[17] and a mediator of enzalutamide resistance through sustained androgen receptor signaling 159 [15]. In the present work we sought to develop a system to identify novel collateral sensitivities 160 to Snail-induced resistance to enzalutamide. To do this we turned to a Snail inducible LNCaP95 161 cell line system in which Snail is fused to an estrogen receptor mutant (ER mut ) in which 4-162 hydroxy-tamoxifen (4OHT) acts as an agonist ( Fig. 2A). Addition of 4OHT induces estrogen EGFP expression (Fig. 2C). Treatment of LNCaP95-Snail cells with 4OHT leads to a reduction 174 in confluence, consistent with the known relationship between Snail and cell cycle arrest [40] 175 ( Fig. 2D). Similarly, Snail induction also induces robust inhibition of EGFP expression (Fig. 2E) 176 consistent with inclusion of the mesenchymal FGFR2-IIIc exon. A loss of EGFP signal in Snail+ 177 cells is also evident by fluorescence imaging of Snail-(EtOH) and Snail+ (4OHT) cells (Fig. 2F).

178
EGFP expression from the GIIIcI 2 reporter is also consistent with endogenous FGFR2 splicing, 179 in which 4OHT induces a switch from the IIIb to IIIc isoforms, as observed by isoform-specific 180 restriction digestion of FGFR2 RT-PCR products (Fig. 2G).

181
High-throughput screens identify synthetic lethality to Snail-induced epithelial plasticity. We 182 applied this Snail-inducible plasticity reporter system to identify compounds with synthetic 183 lethality for Snail+ prostate cancer that could be subsequently validated for activity in models of 184 enzalutamide resistance given the association between Snail expression and enzalutamide 185 resistance [15]. To do this, we performed a high-throughput small molecule screen using the 186 SelleckChem Bioactives compound library. The Bioactives library contains 2,100 small 187 molecules annotated by target and pathway. The library was designed to include compounds 188 that are structurally diverse, medicinally active, and cell permeable, including both FDA-189 Figure 3. A synthetic lethality screen pinpoints potential therapies for Snail+ prostate cancer. A. Schematic of multi-assay screening strategy. B. Top hits with differential response in Snailand Snail + cells. Below the 1.0 line indicates drug differentially inhibits Snail+ cells; above the line indicates drug differentially inhibits Snail-cells. C. Representative growth slope differences for top candidate agents with differential effects on Snail-and Snail+ cells. D. Top hits grouped by target/pathway ranked by differential slope; color indicates number of drugs per pathway. E. Venn diagram of overlap in compounds that altered both confluence and CellTiter Glo (CTG). F. Overlapping drugs with differential sensitivity in Snail+ cells for both confluence and CTG assays. G. Candidate EMT/MET inducers ranked by GIIIcI 2 induction (higher GFP = more epithelial; lower GFP = more mesenchymal). H. Top 10 candidate MET inducing compounds, as estimated by EGFP expression from the GIIIcI 2 reporter. approved and non-approved compounds [34,36,37]. Screen results were analyzed for cell 190 viability/ATP production by CellTiter Glo at the four-day endpoint, and for cell growth rate and 191 epithelial plasticity status by daily IncuCyte imaging of confluence and GIIIcI 2 EGFP levels, 192 respectively, for four days ( Fig. 3A; Supplemental Table 1). Analysis of CellTiter Glo values for 193 empty wells revealed a significant reduction in growth for Snail+ cells (Supplementary Figure   194 1A), which is consistent with the known role of Snail as a mediator of cell cycle arrest. Across 195 the entire compound library 3.8% of compounds inhibited CellTiter Glo signal for Snail-cells of 196 50% or more, while 22% of the library inhibited Snail+ cells 50% or more (Supplemental Table   197 1). To identify compounds with differential sensitivity based on Snail expression, we analyzed

202
In parallel to CellTiter Glo, we also quantified differences in growth rate for all screen 203 compounds with and without Snail induction. Cell confluence was moderately, but significantly, shown for a subset of compounds in Fig. 1C, with compounds in gray having little to no effect 208 on cell growth of Snail-(EtOH) cells and these same compounds inhibiting growth in Snail+ 209 (4OHT) cells. Subsequent annotation by target enabled identification of targets for which >2 210 drugs hit the same target. Top hits were ranked by their differential slope when comparing 211 Snail+ to Snail-cells. Among these hits were inhibitors targeting signaling molecules and 212 pathways known to be involved in lineage plasticity and prostate cancer therapy resistance, 213 such as aurora kinase, c-MET, and mTOR/PI3K (Fig. 3D). Other targets included inhibitors of 214 CRM1 (XPO1), a nuclear shuttling protein, cyclin-dependent kinases, polo-like kinases, and 215 protein kinase C (Fig. 3D). To identify synthetic lethality for Snail+ cells, we focused on agents 216 with <50% killing in Snail-(EtOH) cells and >50% killing in Snail+ cells by CellTiter Glo. Among 217 these compounds, comparison of drugs that inhibited both CellTiter Glo production and growth 218 rate by greater than 2-fold in Snail+ cells as compared to Snail-cells revealed four candidate

222
We next attempted to identify compounds and pathways that inhibit Snail-induced EMT.

223
To do this we first calculated the fold change in EGFP expression for each compound at day 4 224 as compared to day 1. The fold change in EGFP expression for 4OHT (Snail+) cells was divided 225 by EtOH (Snail-) cells for each compound to identify drugs that were capable of overcoming 226 Snail-mediated EMT. To ensure the gain in EGFP expression was not simply a function of cell 227 growth inhibition or cell death, we compared the EGFP expression to the differential confluence 228 in 4OHT-treated versus EtOH-treated cells. This analysis revealed a subset of compounds that 229 led to differential re-activation of EGFP expression from the GIIIcI 2 EMT/MET reporter while 230 maintaining at least 50% viability or greater (Fig. 3G). These agents included GSK2126458 231 (mTOR/PI3K), three microtubule associated drugs, TAK-875 (GPR40 agonist), PIK-75 (DNA-232 PK, p110α), Sparfloxacin (antibiotic), LY2228820 (p38/MAPK), AUY922 (HSP90), and 233 Edoxaban (Factor Xa) (Fig. 3H).

234
The chemical landscape of collateral sensitivity to enzalutamide-resistant prostate cancer.

235
Given the association between Snail-mediated EMT and enzalutamide resistance, we 236 hypothesized that the evolution of enzalutamide resistance may also enrich for this EMT-like 237 plasticity. To better understand these relationships between phenotypic plasticity and 238 enzalutamide resistance we applied a series of EMT scoring metrics [41-43] to analyze RNA-

239
Seq data from four independent pairs of enzalutamide-sensitive and enzalutamide-resistant cell 240 line models [16]. Consistent with our hypothesis, enzalutamide-resistant cells exhibited a 241 significant shift in scores toward a more mesenchymal-like gene expression signature (Fig. 4A).

242
These overall trends were consistent across scoring metrics, with some exceptions for specific 243 cell line pairs, depending on the scoring metric used (Supplemental Fig. 2A, B). Also 244 consistent with this, treatment of LNCaP95-Snail(-) cells with enzalutamide led to an increase in 245 nuclear localization of Snail (Fig. 4B). The enzalutamide-treated LNCaP95-Snail cells mirrored 246 induction of Snail nuclear localization with 4OHT treatment (Fig. 4C,D). These analyses indicate

249
To further extend the analysis of Snail-specific synthetic lethality, we next attempted to identify 250 potential collateral sensitivities to this EMT-like enzalutamide-resistant phenotype. In order to 251 accomplish this we performed a separate high-throughput compound screen on enzalutamide-252 resistant CS2 cells. The CS2 model is an LNCaP-derived subclone that was generated from androgen responsive elements is inserted upstream of the GFP reading frame (Fig. 5A). These 259 CS2 PSA-GFP enzalutamide-resistant cells were screened using the Bioactives library to interrogate 260 AR signaling (GFP), cell viability (CellTiter Glo), and cell growth (IncuCyte imaging) (Fig. 5A).

261
To ensure the PSA reporter is responsive to androgen receptor signaling, cells were treated 262 with the anabolic-androgenic steroid derivative, R1881, or enzalutamide. R1881 treatment led to 263 a significant increase in GFP signal while enzalutamide had no effect on GFP expression in the 264 enzalutamide-resistant CS2 model (Fig. 5B). The increase in GFP during R1881 treatment was 265 not due to a change in confluence, as these treatments did not significantly alter cell confluence 266 Figure 5. Collateral sensitivity screens identify candidate actionable pathways to treat enzalutamideresistant prostate cancer. A. PSA reporter schematic and screening strategy. B. Validation of the PSA-GFP reporter system. C. Confluence quantification in CS2 enzalutamide-resistant model following exposure to R1881 and enzalutamide. D. Pathway-level analysis of top inhibitors targeting CS2 enzalutamide-resistant cells. E. Activators of PSA reporter activity (green dots); top candidates are labeled by pathway or with drug name. F. Inhibitors of PSA reporter activity (brown dots); top candidates are labeled by pathway. (Fig. 5C). Analysis of cell growth inhibition for the Bioactives screen at the pathway level in the 267 CS2 enzalutamide-resistant cells pinpointed candidate collateral sensitivities of interest, 268 including DNA-PK, cyclin-dependent kinases, histone deacetylases, PI3K, mTOR, CRM1, and 269 PLK (Fig. 5D). Analysis of PSA reporter expression as a function of cell viability also revealed 270 compounds targeting multiple receptors (androgen receptor, estrogen receptor, glucocorticoid 271 receptor, dexamethasone) as inducers of PSA reporter activity (Fig. 5E) and compounds that 272 target epigenetic modifiers as repressors of PSA reporter activity (Fig. 5F).

273
To provide further validation of candidates, we plotted the relative cell viability by CellTiter Glo 274 for compounds in the CS2 enzaR screen by cell viability (CellTiter Glo) in the LNCaP95-Snail 275 screen (Fig. 6A). This analysis revealed a subset of drugs active in both screens. We ranked 276 these top hits by a sum rank statistic that includes the rank of cell death by CellTiter for both 277 screens as well as the differential confluence for Snail-vs. Snail+ cells (Fig. 6B). Top targets 278 from this analysis including PI3K, mTOR, and the proteasome (Fig. 6B). Among this subset, AZ 279 960 (JAK2 inhibitor) and BGT226 (dual PI3K/mTOR inhibitor) were the most effective at 280 inhibiting Snail+ cell confluence (Fig. 6C, D). Consistent with our observations of sensitivity to 281 Figure 6. Comparison of candidate therapies for enzalutamide-resistant and Snail+ prostate cancer. A. Comparison of CS2 enzaR and Snail drug screen hits. B. Top hits for both screens based on a sum rank statistic that includes (CS2 enzaR confluence, Snail+ differential confluence, and Snail+ differential slope of growth rate). C. Growth curves for EtOH (Snail-) and 4OHT (Snail+) cells treated with AZ 960 (JAK inhibitor); and D. BGT226 (PI3K/mTOR inhibitor). E. Quantification of phospho-protein array data for p-STAT1, F. p-JAK1, G. p-STAT2, and H. p-JAK2 in three pairs of enzalutamide-sensitive and enzalutamide-resistant models (Ware et al. Biorxiv).

287
In the present study we sought to characterize the therapeutic vulnerabilities for enzalutamide- inhibitors of mTOR/PI3K, DNA-PK, and p38/MAPK ( Figure 3H). All of these pathways have 296 been previously connected to EMT biology in prostate cancer [54][55][56][57]. We also identified the 297 GPR40 agonist, TAK-875, and Factor Xa inhibitor, Edoxoban, as potential inducers of MET.

298
Consistent with these observations, another GPR40 agonist, GW9508, has been shown to 299 prevent cytokine-induced airway epithelial barriers disruption of claudin, occludin, and ZO-1

300
[58], and Factor Xa inhibition has been shown to reduce EMT in chronic kidney disease [59].

301
These agents represent promising candidates for follow-up studies to inhibit EMT and prevent 302 or delay invasive and metastatic phenotypes associated with hormone therapy resistance.

339
The current study provides a platform to quantify the effects of thousands of compounds across