Metastasis is the major cause of mortality for patients with cancer, and dysregulation of developmental signaling pathways can significantly contribute to the metastatic process. The Sine oculis homeobox homolog 1 (SIX1)/eyes absent (EYA) transcriptional complex plays a critical role in the development of multiple organs and is typically downregulated after development is complete. In breast cancer, aberrant expression of SIX1 has been demonstrated to stimulate metastasis through activation of TGFβ signaling and subsequent induction of epithelial–mesenchymal transition (EMT). In addition, SIX1 can induce metastasis via non-cell autonomous means, including activation of GLI-signaling in neighboring tumor cells and activation of VEGFC–induced lymphangiogenesis. Thus, targeting SIX1 would be expected to inhibit metastasis while conferring limited side effects. However, transcription factors are notoriously difficult to target, and thus novel approaches to inhibit their action must be taken. Here we identified a novel small molecule compound, NCGC00378430 (abbreviated as 8430), that reduces the SIX1/EYA2 interaction. 8430 partially reversed transcriptional and metabolic profiles mediated by SIX1 overexpression and reversed SIX1-induced TGFβ signaling and EMT. 8430 was well tolerated when delivered to mice and significantly suppressed breast cancer–associated metastasis in vivo without significantly altering primary tumor growth. Thus, we have demonstrated for the first time that pharmacologic inhibition of the SIX1/EYA2 complex and associated phenotypes is sufficient to suppress breast cancer metastasis.
These findings identify and characterize a novel inhibitor of the SIX1/EYA2 complex that reverses EMT phenotypes suppressing breast cancer metastasis.
High toxicity and low specificity of standard chemotherapy dramatically affects quality-of-life for patients with cancer, necessitating the development of targeted therapeutics. To date, significant advances have been made in the development and use of inhibitors that disrupt kinases that mediate oncogenic signaling, in large part due to the feasibility of targeting ATP-binding pockets in protein kinases. Kinase inhibitors have been shown to improve clinical outcomes in many cancer types. However, resistance frequently develops in patients treated with such inhibitors (1, 2), making it important to develop novel, additional targeted therapies. Recently, oncology has moved toward drugs that activate the immune system as a means to inhibit tumor progression. Unfortunately, such therapies are effective only in a small population of patients, and can cause severe side effects (3). In contrast, few therapies are being developed that target metastasis specifically, which is the main cause of the vast majority of cancer deaths. It is thus critical to continue to explore novel approaches that inhibit tumor progression, with a specific emphasis on the development of drugs that target metastatic disease.
The fundamental molecular principles of cancer, including loss of growth inhibition, sustained proliferative signaling, unlimited replicative potential, and evasion of apoptosis, often result from dysregulated transcription (4). Although specific transcription factors (TF) have been identified as essential to the induction of tumorigenesis and metastasis, few drugs target these critical regulators. Instead, most anticancer drugs target binding pockets in enzymes or on protein receptors. These drugs often target upstream pathways that can cripple tumors, but may fail to hit the central node (such as a TF) that responds to multiple oncogenic signals. Thus, it is essential to investigate the more difficult task of targeting the TFs themselves (either directly or via targeting their regulation), particularly those that are involved in the metastatic process.
The Sine oculis homeobox homolog 1 (SIX1) is a homeoprotein (5, 6) that transcriptionally controls the expansion of progenitor cell populations through its effects on proliferation and survival, and that also controls cell migration, invasion, and epithelial-to-mesenchymal transitions (EMT) during the embryonic development of numerous organs (5, 7, 8). SIX1 is largely downregulated post-embryogenesis, but is reexpressed in a number of cancers where it is known to contribute to tumor growth and progression (9–13), likely through reinstating embryonic properties out of context. Intriguingly, gain-of-function mutations in SIX1 were recently described in Wilms tumors (14, 15), with mutations increasing the ability of SIX1 to transcriptionally activate glycolytic genes, thereby potentiating the Warburg effect (16). While gain-of-function SIX1 mutations have not been observed in breast cancer (13), it is amplified in a small percentage of breast tumors (17), and is overexpressed in up to 90% of all breast cancers regardless of subtype when compared with normal breast tissue (13, 17, 18). Using xenograft and transgenic mouse models, we and others have shown that SIX1 is a critical mediator of cancer onset and progression (11, 19–23), and that it enhances metastasis via multiple mechanisms, both tumor cell autonomous and non-cell autonomous (24). It is becoming increasingly appreciated that targeting TFs will be critical in the fight against cancer, and SIX1 has been highlighted as a key candidate to target to inhibit EMT and tumor progression (25).
SIX1 contains a DNA-binding homeodomain, but does not harbor an intrinsic transactivation domain (5). Developmentally, SIX1 partners with eyes absent (EYA) family members, which function as transcriptional coactivators to form a bipartite transcriptional complex (8). Importantly, SIX1 requires its interaction with EYA proteins to mediate metastasis (26, 27). Like SIX1, EYA proteins are largely downregulated postembryogenesis (28–30) but are reexpressed in cancers (5, 26, 27, 31). Thus, targeting the SIX1–EYA transcriptional complex may be a means to inhibit tumor progression without inducing significant side effects given the low expression of these proteins after development is complete. Herein, we describe our identification and development of a novel small molecule compound, NCGC00378430 (abbreviated as 8430), that reduces the SIX1–EYA2 interaction in cells. Compound 8430 partially reverses the transcriptional and metabolic effects observed by SIX1 overexpression (OE), and also reverses SIX1-induced TGFβ signaling and EMT in vitro in breast cancer cells. Most importantly, 8430 had little toxicity in vivo, yet dramatically impeded SIX1-induced breast cancer metastasis in a mouse model of breast cancer. Taken together, our work provides proof-of-concept data that SIX1, and thus likely other TFs, can be promising pharmacologic targets to suppress tumor progression.
Materials and Methods
Biological and technical replicates in experiments
With the exception of RNA sequencing (RNA-seq), metabolomics, and animal studies, all experiments were independently conducted in triplicates three or more times. Images and quantification from individual representative experiments were shown to reflect the intrinsic variability of each assay. Biological triplicates were used for RNA-seq and metabolomics analyses. The animal study using MCF7 tumor xenografts in NSG mice was performed twice using exactly the same conditions (n = number of mice/group/experiment), and results were combined for analysis.
Survival analysis in The Cancer Genome Atlas BRCA patients
Median-centered log2-transformed mRNA expression and corresponding clinical BRCA data were obtained from the curated The Cancer Genome Atlas (TCGA) PanCancer Atlas study dataset via the cBioPortal cgdsr R package (https://cran.r-project.org/web/packages/cgdsr/index.html). Patients were stratified on the basis of histologic subtype, and further stratified by combined SIX1 and EYA2 expression status according to percentile cutoffs. Survival analysis was performed using the R packages “survival” and “survminer” (https://cran.r-project.org/web/packages/survival/index.html). Patients exhibiting high SIX1 and EYA2 expression were assigned “BOTH HIGH” status and patients exhibiting any other combination of SIX1 and EYA2 (both low, only high SIX1, and only high EYA2) were assigned “OTHER” status. To prevent the arbitrary selection of a percentile from which to stratify patients into the aforementioned SIX1/EYA2 covariate groups, we iterated a log-rank test (constructed under a Cox proportional hazards model) and obtained P values with SIX1/EYA2 status as the covariate at all discrete percentiles from the 1 to 100th percentiles. Representative Kaplan–Meier curves reported in this manuscript are based on a range of percentile cutoffs that yielded adjusted P values ≤0.05 when the log-rank test was performed.
Protein expression and purification
Human SIX1 (residues 1–259) was subcloned into the pET21a vector modified to include the DNA sequence for GST and expressed in the BL21(DE3) E. coli strain (Novagen). GST-fused SIX1 was purified from the lysate using Glutathione Sepharose 4B resin, and washed with Buffer L (100 mmol/L Tris-HCl pH 8.0, 250 mmol/L NaCl, 5% glycerol, and 1 mmol/L TCEP) before elution by Buffer L containing 30 mmol/L glutathione and 0.1% Triton X. After concentration, SIX1 was further purified on a Superdex 200 size exclusion column (GE Healthcare).
Human EYA2ED (residues 253–538) was subcloned into the pET15b vector. 6xHis-EYA2 ED was expressed in the BL21(DE3) E. coli strain (EMD Biosciences). Cells were lysed by sonication in Buffer L. 6xHis-EYA2 ED was purified using Ni2+-Sepharose HP resin (GE Healthcare). Bound protein was washed with 60 mmol/L imidazole in Buffer L and eluted using an imidazole gradient up to 1 mol/L imidazole. Eluate from the Ni2+ resin was concentrated and further purified on a Superdex 200 size exclusion column (GE Healthcare) in Buffer L.
The AlphaScreen assay was carried out using 6xHis-EYA2 ED and GST-SIX1 proteins as described previously (32). Briefly, the assay was carried out in white 384-well plates (PerkinElmer) at 25°C in assay buffer [50 mmol/L Tris (pH 8.0), 250 mmol/L NaCl, 0.05% BSA, and 0.02% Tween-20]. Compounds of varying concentrations (781 nmol/L–100 μmol/L) were incubated for 2 hours with 200 nmol/L 6xHis-EYA2 and 200 nmol/L GST-SIX1 before the assay plate was read in an EnVision Multilabel Reader (PerkinElmer) in AlphaScreen detection mode. The IC50 value of each compound was calculated using nonlinear regression (dose-response inhibition module) in the GraphPad Prism software.
The synthesis of NCGC00378430 was initiated by the nucleophilic substitution of 2-methoxy-5-nitrobenzenesulfonyl chloride with morpholine followed by a palladium-on-carbon catalyzed hydrogenation of the nitro group to the aniline. This aniline was then subjected to HATU-mediated amide coupling with commercially available 3-(1H-pyrrol-1-yl)benzoic acid to provide NCGC00378430 (N-(4-methoxy-3-(morpholinosulfonyl)phenyl)-3-(1H-pyrrol-1-yl)benzamide). For a more detailed analysis of this synthesis, please see Supplementary Methods.
Detailed cell culture conditions for all breast cancer cell lines are described in the Supplementary Methods. HMLER cell line is a kind gift from Robert Weinberg at MIT. MCF7, T47D, and MDA-MB-231 cells have been used in the Ford lab since its inception, the same lines can be obtained from ATCC HTB-22, ATCC HTB-133, and ATCC HTB-26, respectively. MCF7-Ctrl and MCF7-SIX1 cells were generated as described before (27). All cell lines were authenticated by University of Colorado Cancer Center Protein Production, Monoclonal Antibody, Tissue Culture Shared Resource using short tandem repeat analysis (latest authentication was performed on October 25, 2019). All cells are examined by MycoAlert Mycoplasma Detection Kit (Lonza LT07-118) bimonthly and only negative cells were used for experiments.
Cells were detached and whole-cell protein extracts were isolated by lysing cell pellets using RIPA buffer. Collected protein lysates (30–50 μg) were run through polyacrylamide gel electrophoresis, size-separated proteins were then transferred to nitrocellulose membranes followed by the blocking step using 5% nonfat milk. Next, membranes were incubated with the primary antibodies for overnight at 4°C. The following day, membranes were washed with TBST and secondary antibodies were applied. Finally, membranes were washed and examined using chemiluminescence, which were then imaged on the Odyssey Fc Imaging System (LI-COR Biosciences). Antibodies and concentrations that were used are outlined in Supplementary Table S1.
Cells were lysed using ELB buffer (250 mmol/L NaCl, 50 mmol/L Hepes pH 7.0, 5 mmol/L EDTA, and 0.1% NP40), and protein samples were precleared using TrueBlot anti-Rabbit IgG Magnetic beads (Rockland p/n 00-1800-50). Samples were then incubated with 2 μg of antibody targeting EYA2: anti-EYA2 IgG (Sigma, HPA027024) and 50 μL of Trueblot magnetic beads overnight at 4°C, while gently rocking. The following day, beads were washed using TBS, and proteins were dissociated from the beads by boiling the sample with loading buffer prior to the Western blotting analyses. Of note, secondary antibodies used during Western blotting analyses only recognize nondenatured IgG (Rabbit Trueblot Anti-Rabbit IgG HRP, Rockland, p/n 18-8818-31) and Mouse Trueblot ULTRA Anti-Mouse IgG HRP, Rockland, p/n 18-8817-33).
Proximity ligation assay
Cells were incubated in 4% paraformaldehyde for 15 minutes, and were then permeabilized using 2% Triton in PBS. The Duolink In Situ Red Starter Kit Mouse/Rabbit (Sigma-Aldrich, DUO92101) was used to perform proximity ligation assay (PLA) following the manufacturer's instructions. The SIX1–EYA2 interaction was detected using the antibody pairs listed below, and examined by fluorescent microscopy. Antibody pairs: anti-SIX1 IgG (Cell Signaling Technology, D4A8K, 1:50) and anti-EYA2 IgG (Sigma-Aldrich, WH0002139M4, 1:50). Finally, PLA signal was quantified using Cellprofiler software and calculated as the ratio of red fluorescence over DAPI signal. One-way ANOVA with Tukey posttest was applied for statistics analysis. For quantification, more than 200 cells from 5 fields per condition were examined in each experiment.
MEF3 Firefly/Renilla dual luciferase assay
Cells were transfected with the PGL3 Renilla reporter vector (as a control for transfection efficiency) along with either PGL3 empty vector (EV) or PGL3 MEF3 Firefly reporter (33) for 48 hours in triplicate. After 48 hours, vehicle (DMSO) or 8430 treatment at the mentioned doses was administered for 1 day prior to luciferase activity measurement. Renilla and Firefly luciferase activity was measured following protocols provided in the Dual-Luciferase Reporter Assay System (Promega, E1960), and data were collected using a Modulus II Microplate Multimode Reader (Promega, 998-9375). All Firefly/Renilla ratios were normalized to the EV with vehicle treatment condition.
Cells were fixed in 4% paraformaldehyde for 15 minutes and permeabilized using 2% Triton in PBS for 10 minutes, after which, they were blocked with 10% goat serum for 30 minutes. Primary antibodies were applied for an overnight incubation at 4°C. The following day, primary antibody was removed and three times of PBS wash was performed. Then, secondary antibodies were applied for 1 hour followed by PBS wash and slides were mounted using Fluoroshield with DAPI (Sigma-Aldrich, F6057-20ML) prior to quantitation using fluorescent microscopy. Quantification was performed using the Cellprofiler software, where the percentage of high junctional E-cadherin (E-CAD) was assessed using number of cells with strong membranous E-CAD expression (fluorescent intensity at 30% greater than cytosolic signal) over total cell number, while Fibronectin expression was assessed by ratio of total red fluorescence intensity over DAPI. Similar to PLA, more than 200 cells from fields per condition were examined in each experiment. Antibodies and concentrations that were used are outlined in Supplementary Table S1.
Total RNA was isolated from MCF7 cells using Direct-zol RNA Miniprep Kit (Zymo Research, R2052), RNA integrity and purity were checked by agarose gel electrophoresis and NanoDrop 2000 spectrophotometer (Thermo Fisher scientific, ND-2000) prior to sample submission. RNA samples with high integrity (>95%) in biological triplicates were submitted to submitted to Novogene for library preparation and highthroughput sequencing, using the NEBNext Ultra RNA library Prep kit and Illumina NovaSeq 6000 S4 platform.
RNA-seq data analysis
NovaSeq Control Software (Illumina) was used for signal processing, base-calling and removal of barcode, and adapter sequences. Data quality was assessed using FASTQC (version 0.11.5, https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and FastQ Screen (v0.9.1) was used to check for common sequencing contaminants (https://www.bioinformatics.babraham.ac.uk/projects/fastq_screen/). Low-quality bases (Q < 10) were trimmed from the 3′ end of reads and reads (<30 nt after trimming) were removed using the Fastx toolkit. Reads were aligned to a GRCh37/hg19 Human reference using TopHat2 (v2.1.1, –b2-sensitive –keep-fasta-order –no-coverage-search –max-multihits 10 –library-type fr-firststrand) with the UCSC hg19 Gene Transfer Format (GTF) annotation file provided in the iGenomes UCSC hg19 bundle (https://support.illumina.com/sequencing/sequencing_software/igenome.html). Aligned reads were then filtered to remove low-quality mapped reads (MAPQ < 10) using SAMtools (v1.5). Quality assessment of final mapped reads was conducted using RSeQC (v2.6.4). Gene-level counts were obtained using HTSeq (v0.8.0) with the following options (–stranded = reverse –minaqual = 10 –type = exon –idattr = gene_id –mode = intersection-nonempty) using the iGenomes UCSC hg19 GTF annotation file. Differential gene expression was evaluated using DESeq2 (version 1.6.3; ref. 34) in R (version 3.1.0) using Padj < 0.1 as the cutoff.
Heatmap, overrepresentation analysis, GSEA, and Gene Ontology analysis
To generate heatmaps, significantly differentially expressed genes from between the wild-type (WT) and SIX1 OE condition, as defined by Padj < 0.1 and |log2fold-change| > 1, were plotted as heatmaps across all RNA-seq samples. Gene expression was converted into z-scores prior to plotting, and heatmaps were subsequently created using the package pheatmaps in R (https://cran.r-project.org/web/packages/pheatmap/pheatmap.pdf). Hierarchical clustering of z-score-converted expression values indicates three predominant and distinct clusters of differential gene expression exist. Pathway analysis was performed using gene sets from the Hallmarks collection of the Molecular Signatures Database (35). Overrepresentation analysis was performed on all genes with a Padj <0.1 using the clusterProfiler R package (36). Gene set enrichment analysis (GSEA) was performed on the fold-change of the indicated comparisons using the fgsea R package (https://doi.org/10.1101/060012). Gene ontology enrichment analysis was performed on genes influenced by both SIX1 OE and 8430 treatment (Padj < 0.1, and changed in opposite directions) using PANTHER Overrepresentation Test against Homo sapiens Gene Ontology biological process complete annotation dataset (http://geneontology.org/, test type: Fisher exact, correction: calculate FDR).
Metabolomics analyses were performed via ultrahigh pressure-liquid chromatography-mass spectrometry (UHPLC-MS—Vanquish and Q Exactive; Thermo Fisher Scientific) as reported previously (37). Briefly, cells (in biological triplicates) were extracted in ice cold methanol:acetonitrile:water (5:3:2 vol/vol) at a concentration of 2 million cells/mL. After vortexing for 30 minutes at 4°C, samples were centrifuged at 15,000 g for 10 minutes at 4°C and supernatants processed for metabolomics analyses. Ten microliters of sample extracts were loaded onto a Kinetex XB-C18 column (150 × 2.1 mm i.d., 1.7 μm—Phenomenex). A 5-minute gradient was used to elute metabolites for both negative mode (phase A: 5% acetonitrile, 95% water with 1 mmol/L ammonium acetate; B: 95% acetonitrile, 5% water with 1 mmol/L ammonium acetate, 0%–100% B) and positive mode (phase A: 0.1% formic acid in water; phase B: 0.1% formic acid in acetonitrile, 5%–95% B). The mass spectrometer scanned in Full MS mode at 70,000 resolutions in the 65–975 m/z range, 4 kV spray voltage, 45 sheath gas, and 15 auxiliary gas, operated in negative and then positive ion mode (separate runs). Metabolite assignment was performed against an in-house standard library, as reported previously (37).
IHC staining was performed by the University of Colorado Cancer Center Histology Shared Resource using a previously described procedure (33). Sections were examined through bright-field microscopy and signal positive regions (staining intensity 50% greater than background) were quantified using ImageJ software. More than 1,000 cells from 4 fields per tumor were examined. Antibodies and concentrations that were used are outlined in Supplementary Table S1.
Evaluation of 8430 pharmacokinetics
For a detailed analysis of pharmacokinetic evaluation, please see Supplementary Methods.
Evaluation of 8430 toxicity
8430 was formulated in a mixed solution containing N-methyl-2-pyrolidone (NMP; Sigma-Aldrich, 328634), polyethylene glycol (PEG400;Sigma-Aldrich, 202398) and 30% Solutol HS 15 w/w in water (Sigma-Aldrich, 42966), in a 1:2:3 ratio. NMP, PEG400, and Solutol HS 15 mixed solvent was used as vehicle control. 8430 was introduced to 4th mammary fat pad of CD-1 mice at 0 (vehicle), 15, 20, and 25 mg/kg every other day for 3 continuous weeks. Animal weight was measured every week. In addition, animals were euthanized, both gross and histopathologic evaluation was performed, including histologic assessment of liver, heart, pancreas, small, and large intestine tissues by a board-certified veterinary pathologist (D.P. Regan) at the Comparative Pathology Laboratory at Flint Animal Cancer Center, Colorado State University (Fort Collina, CO).
Six- to 8-week old NSG mice were used for MCF7 tumor cell implantation. Estrogen pellets were embedded subcutaneously in 6–8 week old NSG mice one day before breast tumor cell injection as described previously (18). The following day, 3 × 105 turboRFP labeled MCF7 SIX1 cells were orthotopically introduced into the 4th mammary fat pad of NSG mice. Vehicle or 25 mg/kg of 8430 treatment (adjacent to the injection site of the tumor cells) started 2 days after the injection (day 3), and was continued every other day until the day 21. Fluorescent signal was examined and quantified every week using IVIS 200 (PerkinElmer)-imaging until the end of study (week 9), signal mean was plotted across time, and statistical significance was evaluated using a functional longitudinal mixed-effects model. At the end of study, mice were euthanized while primary tumors were harvested for further IHC analysis. In addition, multiple organs of the mice were examined during necropsy. However, in this analysis, we only observed lymph node metastases (metastases were confirmed by examination of hematoxylin and eosin sections by a pathologist, Dr. Paul Jedlicka). All work involving animals was approved by Institutional Animal Care and Use Committee at University of Colorado, Anschutz Medical Campus.
SIX1/EYA2 complex is a potential therapeutic target for breast cancer
Previous analysis of the van de Vijver gene expression dataset (295 patients) demonstrated that coexpression of SIX1 and EYA2 is significantly associated with poor prognosis in human breast cancer (27). Using the current larger TCGA PanCancer Atlas BRCA dataset, which contains expression data from 943 patients with breast cancer of identified disease subtypes, we asked whether simultaneous high expression of SIX1 and EYA2 are particularly relevant to specific subtypes of the disease. To that end, we used a Cox regression model, stratifying patients based on combined high SIX1 and EYA2 expression (strata definitions described in Materials and Methods section) as covariates. Intriguingly, we discovered high expression of SIX1 and EYA2 shows strong predicting power only in luminal A and basal subtypes, but not in other breast cancer subtypes. Particularly, at the most significant cut-off percentile (lowest P value), simultaneous high expression of SIX1 and EYA2 strongly associates with shorter progression-free survival in patients with luminal A and basal breast cancer patients (Fig. 1). Taken together, these data suggest that the SIX1/EYA2 complex may be an attractive therapeutic target for more than one breast cancer subtype.
Identification of a small-molecule compound targeting the SIX1–EYA2 complex
To monitor the ability of SIX1 to interact with its protein partner, EYA2, we developed an AlphaScreen assay amenable to miniaturization for a high-throughput screen (38). Human EYA2 (residues 253–538) was expressed and purified with a His6 tag for binding on Ni2+-chelated acceptor beads, while SIX1 (residues 1–259) was expressed and purified as a GST-fusion protein for immobilization on GSH-linked donor beads. The interaction of SIX1 and EYA2 brings the donor and acceptor beads close together and produces strong fluorescent signal. Untagged EYA2, but not an unrelated protein, Brr2, inhibits this signal, indicating that the fluorescent signal is specific to SIX1–EYA2 interaction (Fig. 2A). Screening of ∼370,000 compounds from the NIH's Molecular Library Initiative using this AlphaScreen assay identified a class of compounds that inhibited SIX1-mediated transcription by 50% at 50 μmol/L concentration in a MEF3-luciferase reporter assay (39) in MCF7 cells (Supplementary Fig. S1). Our medicinal chemistry effort around the parent compound generated one analog (NCGC00378430, abbreviated as 8430; Fig. 2B; Supplementary Fig. S2A and S2B) that has a moderate IC50 of 52 μmol/L in the Alphascreen assay (Fig. 2C), but improved inhibitory activity against SIX1-mediated transcription (80% inhibition at a concentration of 50 μmol/L in the luciferase assay; Fig. 2D). Intrigued by this apparent cellular activity, we decided to further characterize compound 8430 in cell-based assays.
Compound 8430 disrupts the SIX1–EYA2 interaction and transcriptional activity in tumor cells
Because 8430 was able to reduce SIX1-mediated transcription, we set out to evaluate the effects of this compound in a well-characterized breast cancer cell line model, MCF7, where we had previously shown that SIX1 can induce metastasis in a manner dependent on EYA2, via stimulating TGFβ signaling and EMT (26). In addition, in this same MCF7 model, we had demonstrated genetically that disruption of the SIX1–EYA2 interaction, via expression of an EYA-binding deficient mutant of SIX1 (SIX1V17E), led to a significant reduction in SIX1-induced breast cancer metastasis (27). Together, these data provide clear evidence that in the context of MCF7 cells, SIX1 requires EYA2 to mediate metastasis, and thus provides a perfect model in which to test the effect of our inhibitor on metastasis.
To this end, previously generated stable SIX1 OE MCF7 cells (MCF7-SIX1) were used, while chloramphenicol transferase-OE MCF7 cells (MCF7-Ctrl) were used as control (27). As additional models to examine the efficacy of 8430, we chose two breast cancer cell lines (one hormone-positive, T47D, and one triple-negative, MDA-MB-231) that expressed high endogenous levels of SIX1 and EYA2 compared with MCF7 and HMLER cells (Supplementary Fig. S3). All models were treated with either vehicle (DMSO) or 8430 (10 μmol/L for MCF7 and 20 μmol/L for T47D as well as MDA-MB-231 cells), and the SIX1–EYA2 interaction was evaluated using a PLA. Strikingly, 8430 effectively decreased the SIX1–EYA2 interaction as measured in PLA, when compared with the vehicle control, in all three systems (Fig. 3A–C).
Because PLA is an indication of proximity, and cannot be used to definitively measure protein–protein interactions (40, 41), we also performed immunoprecipitation (IP) assays to validate the effect of 8430 on the SIX1/EYA2 complex. In corroboration of the PLA results, 8430 dramatically decreased EYA2-associated SIX1 across all three systems (Fig. 3D–F). It should be noted that, in addition to disrupting the interaction, it is possible that 8430 indirectly influences the complex by also affecting protein levels. Importantly, we found that SIX1/EYA2 transcriptional activity is suppressed by 8430 across all three systems (Fig. 3G–I). Thus, our data demonstrate that 8430 can inhibit the SIX1–EYA2 complex, and its transcriptional activity, in breast cancer cells.
Compound 8430 partially reverses SIX1-mediated transcriptional and metabolic signatures in breast tumor cells
Because 8430 reduces the SIX1–EYA2 interaction in cells, we next examined whether this inhibitor could reverse global gene expression changes induced by SIX1 OE. To this end, we treated MCF7-Ctrl cells with vehicle (DMSO, abbreviated as Veh), and MCF7-SIX1 cells with vehicle or 10 μmol/L 8430 for 3 days, after which, we performed RNA isolation and RNA-seq analysis. Comparison of transcriptional profiles between Ctrl+Veh, SIX1+Veh, and SIX1+8430 showed that 8191 genes (4048 up and 4143 down) were significantly altered (adjusted P < 0.1) by SIX1 OE (Supplementary Fig. S4A, left), whereas 8430 treatment, compared with Veh treatment, significantly changed 1082 genes (351 up and 731 down; Supplementary Fig. S4A, middle). Of note, when comparing SIX1+8430 with Ctrl+Veh, there are 7836 genes (3884 up and 3952 down) significantly changed (Supplementary Fig. S4A, right). Although SIX1 long-term OE influences the transcriptional profile of breast cancer cells more profoundly than 8430 treatment, there was a substantial overlap between genes altered by SIX1 OE and 8430 treatment (Fig. 4A; Supplementary Fig. S4B). Specifically, 534 of the 1082 genes that were dramatically altered after 8430 treatment were also impacted by SIX1 OE in the opposite direction (up in SIX1 OE and down in 8430 treatment, or vice versa; Fig. 4A). While 8430 was unable to reverse the full spectrum of gene changes induced by SIX1, it is likely that long-term expression of SIX1 in a stable cell line imparts more stable changes (such as epigenetic alterations) that are not easily reversed by short-term treatment of the compound.
Despite 8430 affecting a smaller number of genes than SIX1 OE, we observed marked similarities in the processes/pathways induced by SIX1 and reversed by 8430 treatment. For example, biological processes relevant to development such as cell differentiation, cellular developmental process, and tissue development were particularly enriched both with SIX1 OE and 8430 treatment (Supplementary Fig. S4C), strongly indicating 8430 specifically influences developmental processes that are known to be regulated by the SIX1/EYA2 complex. Through over-representation analysis (ORA), we found substantial overlap (11 of top 15 enriched pathways) between pathways impacted by SIX1 OE and 8430 treatment (Fig. 4B). In particular, EMT, known to be regulated by the SIX1/EYA2 complex (11, 26, 27, 42), ranked high in the overlapping pathways list. In agreement with ORA pathway analysis, GSEA also demonstrated that SIX1 OE and 8430 treatment influenced EMT in opposite directions (Fig. 4C). Together, through interrogating global transcriptional signatures, we demonstrate that compound 8430 is able to partially reverse SIX1-mediated transcriptional changes, including SIX1-induced EMT gene expression signatures.
Recent studies demonstrated a novel role for SIX1 in enhancing aerobic glycolysis, and thus the Warburg effect, in cancer cells, as a means to promote tumor progression (16, 43, 44). We thus asked whether 8430 treatment would alter the metabolic profile of breast cancer cells by performing metabolomics analysis on MCF7-Ctrl cells treated with vehicle, and MCF7-SIX1 cells treated with vehicle or 8430 at 10 μmol/L for 3 days. This analysis revealed that 8430 treatment could partially reverse the metabolic signature induced by SIX1 (Fig. 5A; only top 50 most significantly altered metabolites are shown in heatmap). In particular, tricarboxylic acid cycle intermediates such as malate and fumarate as well as essential amino acids including glutamine, threonine, and aspartate were reversed by 8430 (Fig. 5A; black-bracketed regions). Interestingly, 8430 treatment seemed to trigger a strong signature of increased acylcarnitines, which was not observed in other conditions (Fig. 5A; yellow-bracketed region), this could be attributed to 8430-impeded tricarboxylic acid cycle, alternatively, unappreciated effects from 8430 could result in mitochondrial dysfunction that hinders catabolism of fatty acids. In addition, via metabolites set enrichment analysis, we discovered a substantial overlap (13 of 20 top enriched pathways) between pathways altered by SIX1 OE and 8430 treatment (Fig. 5B). Pathways regulated by both SIX1 and 8430 include multiple amino acid synthesis pathways as well as glycolytic pathway. Collectively, 8430 treatment impinges on metabolic pathways regulated by SIX1 OE, and partially reverses the metabolite profile back to that seen in the control condition.
8430 suppresses TGFβ signaling and EMT in breast cancer cells
It is well accepted that SIX1 induces EMT and metastasis, at least in part via its ability to enhance TGFβ signaling (11, 42), and that this effect of SIX1 is mediated by its interaction with EYA2 (26, 27). Furthermore, as mentioned above, our RNA-seq analysis demonstrated that EMT, a major contributor to metastasis (45), is one of top pathways induced by SIX1 and reversed by 8430 treatment. Thus, we asked whether 8430 is able to phenotypically reverse SIX1/EYA2-mediated induction of TGFβ signaling and EMT in breast cancer cells. To this end, we examined the effects of 8430 in MCF7-SIX1 OE cells, where SIX1 OE had previously been shown to increase TGFβ signaling and EMT (11, 26, 27, 42). We also tested 8430 in T47D cells treated with TGFβ to induce an EMT (46, 47).
Strikingly, treatment of MCF7-SIX1 cells with 10 μmol/L 8430 reversed the SIX1-induced increase in p-SMAD3 (Fig. 6A), and also led to an alteration in FN1, a well-known mesenchymal marker (Fig. 6B and C). Interestingly, we found that although 8430 treatment did not alter total E-CAD levels (Fig. 6B), it restored membranous E-CAD in MCF7-SIX1 cells, along with inhibiting FN1 expression (Fig. 6C). This phenotype fits well with our previous observation that SIX1 OE does not change total E-CAD levels, but rather changes its localization (11). In T47D cells, 20 μmol/L 8430 treatment blocked TGFβ-induced activation of p-Smad3, upregulation of FN1, and downregulation of E-CAD (Fig. 6D and E). Similar to what was observed in MCF7 cells, 8430 treatment increased membranous E-CAD expression while decreasing FN1 levels in T47D cells (Fig. 6F). Taken together, 8430 dramatically inhibits TGFβ signaling and EMT in SIX1-high tumor cells.
8430 inhibits metastasis of SIX1 OE tumor cells
Because 8430 reversed the EMT phenotypes induced by SIX1 in breast cancer cells in culture, we examined its effects on SIX1/EYA2-induced metastasis in vivo. To this end, we first determined the half-life of 8430 in vivo, and found that it was rapidly metabolized, with a half-life under an hour (Supplementary Fig. S5A and S5B). Therefore, we sought out to test whether frequent local delivery would be practical. We examined the maximal dose that could be tolerated in mice, and found no severe toxicity at a dose of 25 mg/kg (delivered to 4th mammary fat pad every other day until day 21, performed to mimic our experimental protocol in tumor-bearing mice; Supplementary Fig. S5C). Histologic examination of organs from nontumor-bearing animals treated with various doses of 8430 did not reveal any major toxicities. The most prominent and frequently observed histologic change was the presence of multifocal, random, single-cell hepatocyte necrosis in some, but not all, evaluated animals. These foci of necrosis were often associated with a minimal inflammatory response, and rarely, mild replacement fibrosis (Supplementary Fig. S5D). The random nature of these lesions suggests that they may be embolic in nature, and were interpreted to either reflect an underlying husbandry-related issue or a treatment-related liver toxicity. In those animals bearing histologic liver lesions, these lesions were minimal in severity and would be expected to be subclinical.
Because we determined that 8430 was relatively safe, we injected 3 × 105 turbo RFP-tagged MCF7-SIX1 cells into the 4th mammary fat pad of NOD scid gamma (NSG) mice. Two days after tumor cell injection, we treated the mice with vehicle [NMP, PEG400, and 30% Kolliphor HS15 (Solutol) in 1:2:3 ratio], or 8430 dissolved in the same vehicle. Given the short half-life, the vehicle and compound were delivered to the area adjacent to the site of tumor cell injection every other day until day 21. Although we observed no growth inhibitory effect of 8430 on the primary tumor, 8430 treatment dramatically decreased distant metastatic burden compared with vehicle treatment (Fig. 7A; Supplementary Fig. S6A and S6B), demonstrating the efficacy of this compound as an antimetastatic agent. Consistently, IHC staining of primary tumor sections revealed no proliferative or apoptotic phenotypes when comparing vehicle with 8430-treated tumors (Fig. 7B). Surprisingly, we did observe a slight, yet significant, difference in E-CAD and FN1 expression in the 8430-treated tumors versus their control counterparts (Fig. 7C), even though treatment was stopped at week 3 and the analysis in the tumors was performed 6 weeks after the final injection. These data suggest a prolonged suppressive effect on EMT after short-term treatment of 8430. Thus, although 8430 does not impede primary tumor growth, it greatly hinders metastases development, potentially through long-lasting EMT inhibitory effects. In summary, we have identified a novel compound 8430 that disrupts the SIX1/EYA2 function in cells. Compound 8430 globally acts on numerous transcriptional and metabolic pathways on which SIX1 impinges. Furthermore, consistent with our RNA-seq data, we show that 8430 blocks TGFβ signaling and EMT in high SIX1-expressing cells, and most importantly is able to reverse SIX1-induced metastasis in vivo with limited to no toxicity. Thus, our study provides proof-of-concept that pharmacologic inhibition of SIX1 is a promising approach to inhibit breast cancer metastasis.
Recent advances in targeted therapy have led to improved survival of patients with cancer, and have significantly benefitted the breast cancer patient population. However, resistance often develops and cancer re-emerges. Unfortunately, there have been few successful efforts targeting metastasis, and thus many patients succumb to metastatic disease. SIX1, a homeobox TF, requires an interaction with EYA family proteins to transactivate downstream target genes. Concerted high SIX1 and EYA2 expression predicts poor clinical outcome in patients with breast cancer harboring tumors of different subtypes (Fig. 1). Importantly, SIX1 promotes tumor progression and metastasis through multiple means, such as inducing TGFβ signaling (11, 42, 48), enhancing VEGFC production (23, 48), increasing matrix metalloproteinase (MMP) secretion (49, 50), promoting glycolysis (16), and activating GLI signaling in neighboring cells (24). In addition, EYA proteins are known to enhance tumor progression through regulation of migration and invasion of tumor cells (51) as well as through regulating the tumor microenvironment. Because of the many ways by which SIX1 and/or EYA enhance tumor progression, investigating the potential to pharmacologically inhibit their function is of significant value. Because SIX1 and EYA are both developmental genes that are down-regulated in most normal adult tissues (5), inhibiting the SIX1–EYA complex may have limited side effects.
Herein, we identified a promising compound, 8430, that can disrupt SIX1–EYA2 function in cells, and can reverse SIX1-mediated cellular phenotypes and partially restore transcriptional and metabolic signatures that are induced by high SIX1 expression. Most importantly, 8430 inhibits SIX1-mediated breast cancer metastasis in a mouse model. The cellular and in vivo effect of 8430 is SIX1-specific, suggesting that the action of the compound is likely to be at least in part through SIX1. However, due to the low potency and solubility of 8430, we have not been able to conclusively demonstrate whether 8430 directly binds to SIX1 and/or EYA2 through in vitro biophysical approaches, which typically require high concentrations of interacting partners. Although we demonstrated 8430 disrupts the SIX1–EYA2 interaction in cells via both PLA and IP analysis (Fig. 3), it remains possible that 8430 reduces the SIX1/EYA2 interaction indirectly in cell culture, for example, by affecting the levels of one or both of the proteins. In the future, it will be important to unambiguously identify 8430s direct protein target, either through in vitro biochemical analyses as we obtain more soluble and potent compounds, or via the use of affinity tagged analogs for pulldown experiments in cells.
RNA-seq analysis revealed that other than known SIX1-regulated processes such as myogenesis, EMT, TGFβ signaling and Hedgehog signaling, additional pathways are under the control of SIX1 and are targetable through pharmacologic inhibition of SIX1 (5). These pathways include estrogen responses, KRAS signaling, apical junction alterations, and hypoxia-associated pathways (Fig. 4B). Furthermore, our metabolomics analysis identified additional metabolic pathways that are under the control of SIX1, beyond its previously reported effects on glycolysis and its association with the Warburg effect (16). For example, multiple amino acid and nucleic acid metabolism pathways were enriched in both SIX1 OE and 8430-treated conditions (Fig. 5B), suggesting SIX1 is involved in these metabolic processes.
Compound 8430 exhibits promising antimetastatic potency in vivo, but the mechanism remains to be determined. In vivo target engagement and time of residency as well as the effect in other breast cancer models might be necessary to fully assess the therapeutic value of 8430. Remarkably, although treatment of 8430 was carried out only until day 21 in our in vivo metastasis experiment, enhanced junctional E-CAD was detected 6 weeks after 8430 treatment was discontinued, suggesting that 8430 imposed long-term (or irreversible) changes on tumor cells preventing them from readily going through EMT. SIX1 has been shown to mediate metastasis via both cell autonomous and non-cell autonomous means (11, 24, 27, 42), and its inhibition may thus change not only the tumor cells themselves, but also the tumor microenvironment, leading to more lasting effects. For example, high SIX1 expression in tumor cells induces lymphangiogenesis (23, 48) and angiogenesis (50), where the latter process can recruit tumor-associated macrophages to further foster tumor progression (50). In addition, high SIX1 expression in macrophages promotes hepatocellular carcinoma growth and invasion (52). It is possible that 8430′s antimetastatic efficacy is partially attributed to creating an antimetastatic microenvironment, which has a prolonged effects on blocking tumor cell dissemination and colonization. Alternatively, loss of SIX1 may disable tumor cell plasticity, locking the cells in a more differentiated and less metastatic state.
Before 8430 can be moved into the clinical realm, significant chemical alterations are needed to improve the compound, including increased solubility and increased half-life in vivo. We anticipate that pharmacologically targeting the SIX1/EYA2 complex can potentially benefit patients whose cancers express high levels of SIX1 and EYA2 together. Because many studies have associated SIX1 with the induction of an EMT signature (11, 24, 42, 53) that is known to contribute to resistance to therapeutic regimens including tyrosine kinase inhibitors and immunotherapy (54), SIX1/EYA2 inhibitors may also serve as a promising adjuvant therapy to prolong patients' responses to other chemotherapeutic regimens.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: H. Zhou, M. Ferrer, S. Patnaik, J. Marugan, R. Zhao, H.L. Ford
Development of methodology: H. Zhou, M.A. Blevins, D. Kong, L. Zhang, L. Mathews Griner, M. Ferrer, A. D'Alessandro, S. Patnaik, J. Marugan, R. Zhao, H.L. Ford
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.A. Blevins, R. Culp-Hill, D. Ramirez, L. Zhang, J. Trinidad-Pineiro, L. Mathews Griner, R. King, E. Barnaeva, M. Ferrer, D.L. Gustafson, A. D'Alessandro, J. Marugan, H.L. Ford
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Zhou, M.A. Blevins, J.Y. Hsu, M.D. Galbraith, A. Goodspeed, R. Culp-Hill, M.U.J. Oliphant, J. Trinidad-Pineiro, L. Mathews Griner, E. Barnaeva, X. Hu, N.T. Southall, M. Ferrer, D.L. Gustafson, D.P. Regan, J.C. Costello, S. Patnaik, J. Marugan, R. Zhao, H.L. Ford
Writing, review, and/or revision of the manuscript: H. Zhou, J.Y. Hsu, M.D. Galbraith, R. Culp-Hill, M. Ferrer, D.L. Gustafson, J.C. Costello, S. Patnaik, J. Marugan, R. Zhao, H.L. Ford
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Kong, D. Ramirez, E. Barnaeva, N.T. Southall, M. Ferrer, D.P. Regan, A. D'Alessandro
Study supervision: M. Ferrer, J. Marugan, R. Zhao, H.L. Ford
Other (conducted experiment): L. Zhang
Other (discovered compound used for POC. scaled, formulated compound for in vivo studies): S. Patnaik
We thank the Tissue Biobanking and Histology Shared Resource for IHC and advice. We thank the Cancer Center Support Grant (P30CA046934) for supporting the following shared resources used in this work: Tissue Biobanking and Histology, Animal Imaging, Biostatistics and Bioinformatics, and Cell Technologies. We thank Marisa Elise Wagley and Emily Duncan for help making Fig. 2C and Dr. Paul Jedlicka for confirming metastases in the described animal study. This work was supported in whole or in part by the following grants: NIH R01CA224867 (to H.L. Ford and M.T. Lewis), NIH R01NS108396 (to H.L. Ford), NIH R03DA033174 (to R. Zhao and H.L. Ford), NIH R21CA185752 (to R. Zhao and H.L. Ford), AACR-BCRF grant 10-60-26-FORD (to H.L. Ford), Alex's Lemonade Stand Foundation Innovation Grant (to H.L. Ford), DOD Synergistic grant BC084105P1 (to R. Zhao and H.L. Ford), the Colorado Cancer Translational Research Accelerator award (to H.L. Ford and R. Zhao), Colorado Bioscience Discovery and Evaluation Grant (to R. Zhao and H.L. Ford), Cancer League of Colorado Research Grant (to R. Zhao and H.L. Ford), The V Foundation for Cancer Research Award (to R. Zhao), Kimmel Foundation for Cancer Research grant (to R. Zhao), and an NCI fellowship 1F99CA234940-01 to H. Zhou. Work at NCATS was supported by the NIH Intramural Research Program.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.