Dependence receptor UNC5A restricts luminal to basal breast cancer plasticity and metastasis

Tissue samples were collected with Indiana University Institutional Review Board approval, informed patient consent, and HIPAA compliance. UNC5A and EGFR immunostaining was performed at the CLIA certified Indiana University Health Pathology Laboratory and scoring has been described previously [14]. H scores were calculated using stain intensity (0 to 3) multiplied by percent positive pixels (for UNC5A) or a formula based on stain intensity and number of weak, moderate, or strong positive pixels (for EGFR). For subjects with multiple tumor samples, only those with the highest H score were considered. Statistical analysis was performed on samples from 221 breast cancer patients, but only 196 patient samples (89%) had UNC5A values available. The log-rank test was used to compare patient and tumor variables between those with UNC5A H scores versus those without. The correlations between UNC5A and EGFR were determined by Spearman’s correlation coefficient. For modeling the outcomes of overall survival and disease-free survival, the multivariate covariates used in the multivariate models from the individual reports for EGFR and UNC5A were included. Additionally, the H score information for EGFR and UNC5A were handled in three ways. First, the EGFR and UNC5A were dichotomized using the same optimal cut-points as used in their individual reports. Secondly, the EGFR and UNC5A were dichotomized using their individual medians and cut-points. Finally, the continuous values were used in the models. Since EGFR was not linear, the natural log of EGFR was used in the models. For the models with continuous values, hazard ratios were calculated at the 25th, 50th, and 75th percentile of EGFR. Subgroup analyses were performed where the number of patients available was sufficient.

MCF7 and T-47D cells were obtained from American Tissue Culture Collection and cultured in minimum essential media (MEM) media as described previously [15]. TMCF7 cells correspond to cell lines derived from tumors developed in the mammary fat pad of nude mice implanted with MCF7 cells [16]. Cell lines were authenticated using Short Tandem Repeat Profiling Systems for cell line identification by a commercial vendor (DNAcenter.​com) in August 2012 and cell lines recreated from xenograft tumors were authenticated by Genetica (Burlington, NC, USA).

The human shRNA lentiviral transduction particles for sh5-UNC5A and pLKO.1-puro vector control plasmids (sh-Control) were purchased from Sigma (cat. nos. SHCLNV-NM_133369 and SHC 001, respectively). The lentivector for sh2-UNC5A was obtained from Applied Biological Materials (cat. no. i026703g). CRISPR plasmids to target UNC5A were obtained from Sigma-Aldrich (HS0000509914).

Treatments consisted vehicle, heregulin-β1 (HRG-β1, R&D systems), E2, 4-hydroxy-tamoxifen (OHT), or ICI-182,780 (Sigma-Aldrich). The immunoblotting has been previously described [17] and details of antibodies are provided in Additional file 1. Although the majority of immunoblots were reprobed with antibodies against ACTB (β-actin) as a loading control, only representative data per batch of cell lysates are shown.

cDNA was synthesized from 1 to 2 μg of total RNA using the cDNA Synthesis Kit (Bio-Rad). qRT-PCR in duplicates from at least two biological replicates was performed with either Sybr-Green or TaqMan Universal PCR master mix and transcripts were analyzed in StepOnePlus and TaqMan 7900HT instruments (Applied Biosystems) with β-ACTIN as the normalization control. Fold-change was calculated by the ∆∆Ct method, whereas statistical analysis was performed on ΔCt values. Primers (Integrated DNA Technologies) and TaqMan probe details are shown in Additional file 1. RNA-seq of sh-Control and sh5-UNC5A cells treated for 3 h with vehicle or E2 was performed in triplicate as previously described [18], and raw sequencing data have been submitted to the gene expression omnibus (GEO; accession number GSE89700).

We used STAR RNA-Seq aligner to map all sequence libraries to the human genome (UCSC hg19) [19] followed by the assignment of uniquely mapped reads to individual genes based on annotation of hg19 refGene by featureCounts [20]. After trimmed mean of M values (TMM) normalization, gene expression profiling was summarized on the base-2 logarithmic scale. Genes with an average expression level lower than 1 for all phenotypes in MCF7 and T-47D cells, respectively, were excluded for further analysis. Differential expression (DE) analysis was performed using edgeR [21, 22] for special group comparisons in the study. All p values were corrected by multiple testing false discovery rate (FDR) adjustments. Genes with FDR < 0.05 and absolute value of fold change (FC) larger than 2 were determined as differentially expressed genes (DEGs).

Gene function enrichment analysis was performed using DAVID (http://​david.​abcc.​ncifcrf.​gov/​home.​jsp v6.8) [23, 24]. Significantly overrepresented gene ontology (GO) terms were selected if their q values (p values after FDR multiple test correction) were less than 0.05.

Cells transfected with luciferase constructs were allowed to grow overnight in charcoal-dextran treated fetal calf serum (CCS) containing media followed by a 12-h E2 treatment. The Dual-Luciferase® Reporter assay (Promega) was performed according to the manufacturer’s protocol.

ChIP assays for ERα binding on UNC5A and BCL2 were performed under vehicle or E2 treatment for 45 min and 2 h as described previously [9].

After 24-h plating in regular media, the media was changed to CCS-containing media for 3 days and cells were treated with the indicated drug combinations. Cell proliferation was determined using the bromodeoxyuridine-incorporation enzyme-linked immunosorbent assay (ELISA) kit from Calbiochem after 5–6 days of plating with one media/drug change. Mammosphere assays with 5000 cells were performed as described previously [25].

Cells grown on 35-mm glass-bottom culture dishes were fixed with 4% (w/v) paraformaldehyde for 10 min and permeabilized in phosphate-buffered saline (PBS) containing 0.15% (v/v) Triton X-100, 5% (v/v) donor goat serum (Gibco), and 1% (w/v) bovine serum albumin (Sigma-Aldrich) for 1 h. Cells were incubated with primary antibodies (Additional file 1) diluted in Dako antibody diluent (Dako; Agilent Technologies) for 90 min followed by 1-h incubation with the Alexa Fluor® 488 and 555 conjugated secondary antibodies (ThermoFisher Scientific). Nuclei was counterstained with Hoechst® 33,342.

Cells were stained with the indicated antibodies (Additional file 1) and analyzed in a LSR4 custom-made flow cytometer (BD Biosciences) as described previously [26].

The Indiana University Animal Care and Use Committee approved the use of animals in this study and all procedures were performed as per NIH guidelines. sh-Control, sh2-UNC5A, and sh5-UNC5A TMCF7 cells (2 × 10⁶ in 100 μl serum-free HBSS) were implanted into the mammary fat pad of 7-week-old female nude mice with or without a 60-day slow-release E2 pellet. Tumor growth was measured weekly and tumor volume was calculated as described previously [16]. After 12 weeks, the lungs and primary tumors were collected and processed for hematoxylin and eosin (H&E) and PECAM1 (CD31) staining. The whole slide digital imaging system of Aperio (ScanScope CS) was used for imaging of PECAM1-stained tumors. For the metastatic model, mice were inoculated with 2 × 10⁵ TMCF7 cells into the left cardiac ventricle. Ovaries, spleen, and adrenal glands were collected within 17 weeks and processed as described above.

To determine E2-inducible signaling molecules that may dampen the E2 response or gene-specific E2 regulation and that are expressed at higher levels in luminal breast cancers compared with TNBCs, we first searched our previous microarray data of E2-regulated genes in MCF7 cells for known growth suppressive roles and then determined whether E2 directly regulated their expression by integrating E2-inducible gene expression with ERα ChIP-on-chip and ChIP-seq datasets. UNC5A suited these criteria as its expression was E2 inducible and ERα binding sites for this gene was detectable in ChIP-on-chip and ChIP-seq datasets [9, 27] (Fig. 1a). Furthermore, in 11 out of 12 studies in publicly available Nuclear Receptor Signaling Atlas web resources showed 2- to 35-fold E2-inducible expression of UNC5A in MCF7 cells, uterus, and vagina (Additional file 2). We further confirmed E2-inducible expression of UNC5A in MCF7 cells by qRT-PCR (Fig. 1b) and Western blotting (see below), although induction at mRNA levels in our MCF7 cells was modest. Interestingly, the antiestrogen tamoxifen (OHT) failed to overcome the effect of E2 on UNC5A levels (Fig. 1b) suggesting unique effects of E2 on the expression of UNC5A. We used ChIP assay to verify ERα binding to one of the ERα binding sites (Fig. 1c). R2 Genomic and Visualization Platform (http://​r2.​amc.​nl) analyses revealed a positive correlation between UNC5A and ESR1 mRNA levels in breast cancer cell lines (Fig. 1d). Also, analyses of The Cancer Genome Atlas (TCGA) dataset for the relationship between UNC5A expression and breast cancer subtypes using the UALCAN program [28] revealed highest UNC5A expression in luminal breast cancers, which are usually ERα-positive, compared with TNBCs (Fig. 1e). In contrast, NTN1 expression was higher in TNBCs compared with normal breast or luminal breast cancers (Fig. 1f).

To obtain additional support for our hypothesis that a protein that attenuates ERα signaling has prognostic relevance, we performed immunohistochemical analyses of UNC5A in our previously described breast TMA in which 196 out of 221 tumors had measurable UNC5A expression [14] (Additional file 3). A representative staining pattern of UNC5A in breast tumor is shown in Fig. 2a. Tumor cells were moderate in staining in many of the cases with little to no background staining in the other tissues in the core (vascular endothelial cells, smooth muscle cells, fibroblasts, macrophages, and/or scattered lymphocytes infiltrating the tumor region). In both univariate and multivariate analyses, low UNC5A H score was associated with poor overall survival (Fig. 2b and Additional file 4). In subgroup analyses, in ER⁺/PR⁺/HER2⁻, lower UNC5A H score showed a trend of poor overall survival (p = 0.055) (Fig. 2c). UNC5A had no prognostic relevance when tumors were subgrouped broadly into ER⁺ or ER⁻ subgroups (Fig. 2d, e). Thus, UNC5A is a potential biomarker of outcome in a subgroup of breast cancer patients whose tumors express luminal A markers.

To model low UNC5A levels in cells with intact ERα-dependent signaling, we created shRNA-UNC5A MCF7 and T-47D cells, which express ERα and PR at different levels (Fig. 3a, b). MCF7 cells are more responsive to E2 than T-47D cells and, therefore, most of the experiments were performed in MCF7 cells with a few validation experiments in T-47D cells. UNC5A knockdown did not have an effect on ERα levels and the receptor underwent activation-coupled degradation upon E2 treatment in both sh-Control and sh-UNC5A cells (Fig. 3a, b). Note that E2 increased UNC5A protein in sh-Control but not in sh-UNC5A MCF7 cells (Fig. 3a). We note that this is the only commercially available antibody (Abcam, ab81165) that recognized protein of expected size but showed variability in potency between batches. In transient transfection assay, estrogen response element (ERE)-driven luciferase-reporter gene showed elevated activity in vehicle-treated sh-UNC5A cells compared with sh-Control cells (Fig. 3c). In T-47D cells, which express higher levels of PR than MCF7 cells [29], UNC5A knockdown enhanced E2-inducible expression of PGR (Fig. 3d). In both MCF7 and T-47D cells, UNC5A knockdown substantially increased both basal (vehicle-treated) and, consequently, E2-inducible expression of BCL2 mRNA and protein (Fig. 3e, f). The enhanced BCL2 expression in sh-UNC5A cells correlated with increased E2-independent binding of ERα to the enhancer element of BCL2 to which ERα and JMJD3 bind to create a poised chromatin [30] (Fig. 3g). To determine whether enhanced ERα activity in UNC5A-knockdown cells is due to altered phosphorylation of ERα, we used phospho-specific antibodies to determine the levels of ERα phosphorylated at S118 and S167. Phosphorylation of ERα at these residues is known to confer ligand-independent activity to the receptor [1]. Although we did not find any differences in basal phosphorylation status between sh-Control and sh-UNC5A cells, phosphorylated ERα underwent ligand-coupled degradation in sh-Control cells but not in sh-UNC5A cells (Fig. 3h). As a consequence, there was a modest difference in the rate of degradation of total ERα between clones. These results suggest the need for signaling events downstream of UNC5A in turnover of phosphorylated ERα.

To determine whether the above observations of altered phospho-ERα turnover upon UNC5A knockdown show any relationship with cell proliferation, we measured cell proliferation under vehicle control and E2 ± OHT-treated conditions. In MCF7 cells, sh-UNC5A increased proliferation under vehicle control, E2-, OHT-, and OHT plus E2-treated conditions compared with sh-Control to levels similar to E2-treated sh-Control cells (Fig. 3i, j). Although OHT reduced E2-inducible proliferation of sh-UNC5A cells, the overall proliferation rate of these cells under various treatments remained elevated compared with sh-Control cells. Collectively, these results suggest that UNC5A restricts the proliferation of ERα-positive cells.

To investigate whether enhanced the baseline proliferation of sh-UNC5A cells compared with sh-Control cells is ERα-dependent, we treated cells with ICI-182,780 (Fulvestrant) which degrades ERα. While ICI-182,780 reduced E2-induced proliferation of these cells, it had a minimal effect on baseline proliferation of all cell types (Additional file 5). These negative results can be interpreted in two ways: enhanced basal proliferation of sh-UNC5A cells compared with sh-Control cells is independent of ERα, or ERα in sh-UNC5A cells is less sensitive to ICI-182780-mediated degradation. Surprisingly, although ICI-182,780 caused degradation of total ERα in sh-Control and sh-UNC5A cells to a similar extent, ICI-182,780 increased the levels of ERα phosphorylated at S118 (Additional file 5). This unique effect of ICI-182,780 on phospho-ERα could explain the lack of its effects on the baseline proliferation rate of sh-UNC5A cells. Additional work is needed to clarify the role of ERα in the baseline proliferation rate of sh-UNC5A cells.

The dramatic effect of UNC5A on BCL2 expression was puzzling. To ensure that this increase in BCL2 expression is not due to aberrant integration of shRNAs into the genome, we used the CRISPR/Cas9 system to reduce UNC5A expression and selected single cell clones (Additional file 5). UNC5A protein levels were partially reduced in these single cell clones with an accompanying increase in BCL2 expression. Thus, even a modest decrease in UNC5A protein levels was sufficient to trigger BCL2 expression. We also observed stable BCL2 overexpression in both UNC5A shRNA and CRISPR clones cultured for a prolonged time despite these clones regaining UNC5A protein expression as measured using the available antibody. Thus, it appears that even transient knockdown of UNC5A leads to robust/permanent activation of BCL2, which is similar to previously reported stable activation of cancer germline genes upon transient knockdown of DNA methyltransferase 1 (DNMT1) [31].

We performed RNA-seq of sh-Control and sh5-UNC5A MCF7 and T-47D cells treated with vehicle (basal) or E2 for 3 h and did pairwise comparisons to determine the effect of UNC5A on basal and E2-regulated gene expression (Fig. 4a). Genes were determined as DEGs for comparison if their FDR was < 0.05 and absolute value of fold change |FC| was > 2 (Fig. 4a and Additional file 6). Under basal growth conditions, UNC5A knockdown notably affected the expression of approximately 20% and 7% of genes in MCF7 and T-47D cells, respectively, potentially indicating its role in regulating the transcriptional machinery. For example, APOBEC3B, which is integral to ERα signaling [32], was one of the genes differentially expressed in sh-UNC5A cells compared with sh-Control MCF7 and T-47D cells (Additional file 6). We confirmed elevated expression of APOBEC3B in sh-UNC5A compared with sh-Control MCF7 cells (Fig. 4b). Based on the gene functional analysis using DAVID, genes differentially expressed in UNC5A knockdown MCF7 and T47-D cells were an integral part of the plasma membrane and extracellular region (Fig. 4c and Additional file 7). It is interesting to note that 167 DEGs in MCF7 cells were significantly overrepresented in sequence-specific transcription factor DNA binding activity (Fig. 4c). UNC5A knockdown also had a significant effect on E2-regulated gene expression, particularly in MCF7 cells. A total of 434 genes were recognized as undergoing significant changes in gene expression by E2 in sh-UNC5A MCF7 cells, whereas only 21 genes were identified as DEGs in sh-UNC5A T-47D cells (Fig. 4a and Additional file 6). In sh-UNC5A MCF7 cells but not in sh-Control cells, E2-targeted genes were associated with negative regulation of cell proliferation, extracellular region, and histone demethylase activity (Fig. 4c). For example, E2 induced the expression of histone demethylases KDM4B and KDM7A but reduced the levels of UTY and ARID5B in sh-UNC5A but not in sh-Control MCF7 cells [33] (Additional file 6). JARID2, which regulates the polycomb complex and histone methyltransferases [34], was E2 inducible in sh-UNC5A but not in sh-Control MCF7 cells (Additional file 6). Furthermore, we found that 109 out of these 434 genes (Fig. 4d) were not regulated by E2 in sh-Control MCF7 cells (FDR > 0.5 and |FC| < 1.15), whereas their expression was under E2 control in sh-UNC5A MCF7 cells. These 109 genes are associated with positive regulation of gene expression and affect transcriptional regulation by RNA polymerase II (Fig. 4c). By contrast, no specific GO functions could be assigned to uniquely E2-regulated genes in sh-UNC5A T-47D cells. We note that the effect of UNC5A on E2-regulated genes is gene-specific since sh-Control and sh-UNC5A MCF7 cells showed similar levels of E2-regulated expression of TFF1 and only a modest effect on E2-regulated expression of GREB1, two commonly used genes to measure E2-inducible genes (Fig. 4e). Collectively, these results indicate a cell type-dependent role of UNC5A in controlling basal and E2-regulated gene expression with potential downstream effects ranging from plasma membrane composition to transcriptional output from RNA polymerase II.

UNC5A knockdown increased the levels of oncogenic ΔNp63 isoform mRNA while simultaneously lowering the expression of tumor suppressive TAp63 isoform [35] (Fig. 5a, b). TP63 is an E2-repressed gene and ERα failed to repress ΔNp63 in sh-UNC5A clones with efficient UNC5A knockdown (sh5-UNC5A clones of MCF7 and T-47D; Fig. 5a, b). Overall, ΔNp63 levels ± E2 treatment remained elevated in sh-UNC5A cells compared with sh-Control cells. RNA-seq studies showed lower expression of luminal/alveolar differentiation-associated ELF5 but elevated expression of the pro-oncogenic MECOM (EVI-1) and lymphangiogenic NTN4 [36‐38] in sh-UNC5A cells compared with sh-Control cells (Additional file 6). Indeed, ELF5 levels were significantly lower and NTN4 levels were higher in sh-UNC5A cells compared with sh-Control cells (Fig. 5c, d). MECOM protein was undetectable in MCF7 cells but elevated in sh-UNC5A T-47D cells compared with sh-Control cells (Fig. 5e). In addition, while sh-Control cells expressed mainly KRT19, sh-UNC5A cells expressed either KRT14, KRT19, or both KRT14 and KRT19 (basal and luminal cytokeratins, respectively) [39] (Fig. 5f and Additional file 8). Bipotent luminal progenitor cells are KRT14 and KRT19 double-positive [40]. Note that UNC5A knockdown did not result in the morphologic features of epithelial to mesenchymal transition (EMT), nor did it result in the expression of EMT-associated genes such as SNAI1, SNAI2, ZEB1, or ZEB2 (Additional file 6 and data not shown). However, we observed elevated expression of ITGB6 (Integrin β6) in sh-UNC5A cells compared with sh-Control cells (Additional file 6); ITGB6 is pro-oncogenic and is induced during EMT of colon cancer cells [41].

Since ΔNp63 maintains stem cell phenotype and cancer cells with hybrid luminal/basal/mesenchymal characteristics display enhanced cancer stem cell (CSC) properties [35, 42], we used mammosphere assays and flow cytometry to characterize sh-Control and sh-UNC5A MCF7 cells for stemness. While mammospheres of sh-Control were well organized, sh-UNC5A cells formed irregular mammospheres (Fig. 5g). In addition, while sh-Control cells were predominantly CD49f (ITGA6)⁻/EPCAM⁺, a subpopulation of sh-UNC5A cells showed CD49f⁺/EPCAM⁺ phenotype (Fig. 5h). CD49f⁺/EPCAM⁻, CD49f⁺/EPCAM⁺, and CD49f⁻/EPCAM⁺ cells display stem/basal, luminal progenitor, and differentiated/mature features, respectively [43]. Sh-Control cells showed CD44⁻/CD24⁺ non-CSC phenotype whereas sh-UNC5A cells acquired the features of CSCs as evident from the presence of CD44⁺/CD24⁺ and CD44⁺/CD24⁻ cells [44]. Furthermore, sh-UNC5A MCF7 cells expressed significantly higher levels of stemness-associated SOX2 [45] (Additional file 6).

Two of our observations and one prior report prompted us to investigate whether UNC5A knockdown is associated with altered activity of EGFR, which could explain the effects of UNC5A on E2-regulated gene expression. First, we observed enhanced basal ERE-luciferase activity in sh-UNC5A cells, suggesting ligand-independent activity of ERα which typically involves growth factor receptor–ERα crosstalk [46]. EGFR is forefront in this crosstalk as it can alter ERα cistrome and ERα-regulated gene expression [47]. Second, sh-UNC5A cells showed luminal/basal hybrid phenotype, and EGFR activation is common in cells with basal phenotype [48]. Third, a recent study showed that NTN1, in the absence of UNC5A, increases EGFR at the post-translational level [49]. We first measured EGFR levels in sh-Control and UNC5A knockdown cells. EGFR protein but not mRNA levels were significantly higher in sh-UNC5A cells compared with sh-Control cells (Fig. 6a and data not shown). AKT and ERK are the two major kinases activated downstream of EGFR that can increase ligand-independent activity of ERα [1]. We determined whether UNC5A knockdown had an effect on vehicle, E2-regulated, and HRGβ1-induced activation of these kinases. sh-UNC5A cells showed robust activation of AKT as measured by pAKT-S473 levels but not ERK (Fig. 6b). We recently reported that AKT1 but not AKT2 is active in MCF7 cells [17]. Immunoblotting using isoform-specific phospho-antibodies showed upregulation of pAKT1 but not pAKT2 in sh-UNC5A cells compared with sh-Control cells (Fig. 6c). These results indicate a negative relationship between EGFR and UNC5A expression in cell line models. Similarly, UNC5A and EGFR expression showed negative correlation in breast tumor samples when the analyses included all samples or only ER⁺ samples (Fig. 6d).

We next investigated the role of ERα in negative crosstalk between UNC5A and EGFR/TP63. Treatment of cells with ICI-182,780 results in degradation of ERα and, consequently, elevated expression of genes typically repressed by ERα. Indeed, treatment of sh-Control MCF7 and T-47D cells caused degradation of ERα with a concomitant increase in TP63 (Fig. 6e). Interestingly, ICI-182,780 treatment did not have an effect on EGFR but reduced the level of TP63 in sh-UNC5A cells (Fig. 6e). Similar to TP63, elevated expression of BCL2 upon UNC5A knockdown is ERα-dependent as its levels were lower in ICI-182,780-treated cells compared with untreated sh-UNC5A cells (Fig. 6e). Thus, while elevated EGFR levels in sh-UNC5A cells are ERα-independent, ΔNp63 and BCL2 upregulation in these cells is at least partially ERα-dependent.

MCF7 cells form nonmetastatic tumors in female nude mice when injected with matrigel or when supplemented with E2 pellets, although there is less uniformity between the sizes of tumors between animals [16]. We had previously reported in the MDA-MB-231 model that cell lines derived from tumors that develop in the mammary fat pad upon implantation of parental cells show enhanced and uniform tumorigenicity upon re-implantation [50]. We used this approach to increase uniformity in tumorigenicity and generated TMCF7 sh-Control, sh2-UNC5A, and sh5-UNC5A cells. As with MCF7 cells, sh-UNC5A TMCF7 cells showed elevated BCL2 and ΔNp63 compared with sh-Control cells (Additional file 8). sh-UNC5A but not sh-Control TMCF7 cells displayed KRT14/KRT19 double-positive phenotype (Additional file 8). A large subpopulation of sh-UNC5A TMCF7 cells was of the CD44⁺/CD24⁺ and CD49f⁺/EPCAM⁺ phenotype compared with sh-Control cells (Additional file 8), and mammospheres formed by these cells were irregular compared to mammospheres from sh-Control cells (Additional file 8). In addition, sh-UNC5A TMCF7 cells expressed significantly higher levels of SOX2 despite maintaining ERα expression (Additional file 8). Consistent with RNA-seq data (Additional file 6), a large subpopulation of sh-UNC5A TMCF7 cells were ITGB6⁺ compared with sh-Control cells (Additional file 8).

A significant number of mice injected with sh-UNC5A cells but not sh-Control cells developed tumors in the absence of E2 pellets (Fig. 7a). The size of these tumors was larger than tumors in animals injected with sh-Control cells in the presence of E2 pellet (Fig. 7b). While none of the animals injected with sh-Control cells developed lung metastasis, consistent with our previous study with TMCF7 cells [16], animals that received sh-UNC5A cells showed lung metastasis (Fig. 7c). We next examined whether sh-Control cell- and sh-UNC5A cell-derived tumors differ in angiogenesis because of the differences in NTN4 expression between sh-Control and sh-UNC5A cells noted in Fig. 5. sh-UNC5A cell-derived tumors contained higher numbers of PECAM1⁺ cells compared with sh-Control cell-derived tumors (Fig. 7d), suggesting enhanced angiogenesis in the absence of UNC5A.

To determine whether UNC5A knockdown enhanced the multiorgan homing capacity of tumor cells, sh-Control and sh-UNC5A cells were injected via the intracardiac route into animals supplemented with E2 pellets. Autopsies within 17 weeks of injection revealed growth of tumor cells in ovaries and adrenal glands at a higher frequency in animals injected with sh-UNC5A cells compared with sh-Control cells (Fig. 7e). Histological analysis revealed a severe disruption of the normal architecture of both organs (Fig. 7f). Ovaries were devoid of follicles and corpora lutea, and the very few remaining were undergoing atresia and degeneration. Likewise, adrenals lost a clear differentiation between the cortex and medulla zones with hemorrhagic areas and high vacuolization even in areas where the capsule is still preserved. Spleens of animals that received sh-UNC5A cells showed extramedullary hematopoiesis with enhanced myeloid and erythroid elements and megakaryocytes causing distention of the spleen in the red pulp area (data not shown). Overall, the results presented above clearly indicate the role of UNC5A in regulating the metastasis of ER⁺ tumors and its loss of expression leading to E2-independent growth both in vitro and in vivo.