Re-epithelialization of cancer cells increases autophagy and DNA damage: Implications for breast cancer dormancy and relapse

Abstract

Cellular plasticity mediates tissue development as well as cancer growth and progression. In breast cancer, a shift to a more epithelial phenotype (epithelialization) underlies a state of reversible cell growth arrest called tumor dormancy, which enables drug resistance, tumor recurrence, and metastasis. Here, we explored the mechanisms driving epithelialization and dormancy in aggressive mesenchymal-like breast cancer cells in three-dimensional cultures. Overexpressing either of the epithelial lineage-associated transcription factors OVOL1 or OVOL2 suppressed cell proliferation and migration and promoted transition to an epithelial morphology. The expression of OVOL1 (and of OVOL2 to a lesser extent) was regulated by steroid hormones and growth factors and was more abundant in tumors than in normal mammary cells. An uncharacterized and indirect target of OVOL1/2, C1ORF116, exhibited genetic and epigenetic aberrations in breast tumors, and its expression correlated with poor prognosis in patients. We further found that C1ORF116 was an autophagy receptor that directed the degradation of antioxidant proteins, including thioredoxin. Through C1ORF116 and unidentified mediators, OVOL1 expression dysregulated both redox homeostasis (in association with increased ROS, decreased glutathione, and redistribution of the transcription factor NRF2) and DNA damage and repair (in association with increased DNA oxidation and double-strand breaks and an altered interplay among the kinases p38-MAPK, ATM, and others). Because these effects, as they accumulate in cells, can promote metastasis and dormancy escape, the findings suggest that OVOLs promote dormancy entry and maintenance in breast cancer, but also may ultimately drive dormancy exit and tumor recurrence.

Introduction

Epithelial lineage determination and differentiation shape the mammary epithelium during embryonic development as well as later, during gland development (1). Both mesenchymal-to-epithelial transition (MET) and the reverse transition, epithelial-to-mesenchymal (EMT), are associated with normal mammary development during embryogenesis, and with cancer. The EMT program depletes epithelial features, such as cell polarity, while conferring mesenchymal features. However, rather than dualistically switching between epithelial and mesenchymal states, cells can acquire a spectrum of epithelial/mesenchymal (E/M) states (2). Nevertheless, the relevance of EMT and especially the hybrid states to clinical metastasis is still open (3, 4). Likewise, although it is clear that disseminated tumor cells (DTCs) may enter prolonged quiescence, how these cells later relapse remains incompletely understood.

Relapses of dormant metastases might occur in patients as long as 15 to 20 years after surgical removal of the primary tumor (5). Once arrested in the host parenchyma, DTCs acquire a reversible phenotype that includes reduced growth and motility, along with resistance to treatment (6, 7). Poor vascularization might instigate angiogenic dormancy (8), whereas suppressive immune cells may control immunogenic dormancy (9, 10). Both mechanisms likely lead to balanced proliferation and apoptosis (‘tumor mass dormancy’). Alternatively, disseminated solitary or clustered cancer cells might adopt a state of reversible growth arrest (11). Resolving the molecular details of entry into and escape from dormancy is limited by the relatively short duration of conventional experiments and dearth of suitable models. Early studies identified increased active p38, relative to active ERK, as a marker of tumor dormancy (12), and later work showed that endothelium-derived angiogenesis suppressors, such as thrombospondin-1 (TSP1), as well as phosphorylation of myosin light chain (MLC) (13), control dormancy. Likewise, through combined signaling by the tumor necrosis factor and interferon-gamma, tumor-specific T cells prevent angiogenesis and cell proliferation, to promote dormancy (14). STING (Stimulator of Interferon Genes) also acts as a suppressor of reawakened metastatic cells (15).

Resolving the action of specific EMT-inducing transcription factors (such as ZEB, SNAIL and TWIST) has greatly contributed to the understanding of metastasis (16). To similarly simulate the reciprocal transition, MET, and gain insights into tumor dormancy, we induced epithelialization of highly mesenchymal breast cancer cells by means of doxycycline-inducible alleles of two ovo-like (OVOL) proteins. The corresponding transcriptional regulators have repeatedly been implicated in epithelial lineage differentiation (17) and hybrid E/M states (18–22). For instance, OVOLs regulate epithelial differentiation during hair formation and spermatogenesis (23), and mathematical models identified OVOLs as stabilizers of hybrid E/M states (24). According to one study, OVOL1 suppresses EMT by accelerating the degradation of the TGF-β receptor (25). In addition, OVOL1 is involved in somatic cell reprogramming (26), whereas loss of OVOL2 confers stemness characteristics (27).

Here, when overexpressed in highly aggressive, mesenchymal breast cancer cells, OVOLs both decreased the abundance of EMT-associated transcription factors and increased the expression of the genes encoding E-cadherin and C1ORF116, a poorly characterized OVOL target. Furthermore, we found that C1ORF116 functions as a putative autophagy receptor that physically controls two redox modulating proteins, thioredoxin and GCLC (glutamate-cysteine ligase catalytic), thereby increasing reactive oxygen species (ROS). In parallel to the inactivation of antioxidants at the protein level, OVOLs can inhibit the ability of NRF2 (nuclear factor erythroid 2-related factor 2) to transcriptionally up-regulate antioxidant proteins. Consistent with this, we found that OVOL1 alters DNA repair by controlling the kinases involved in damage response, as well as by increasing DNA oxidation and double strand breaks. These findings are reminiscent of previously reported studies showing that DTCs fail colonizing oxidized tissues (28), that autophagy supports survival of dormant cancer cells (29, 30), and that dormant breast cancer metastases keep accumulating new mutations (31). Together, our results raise the possibility that epithelialization and acquisition of the dormant state are controlled by epithelial transcription factors, including OVOL1, which is hormone-dependent, regulates redox homeostasis, and permits accumulation of DNA damage.

Results

An in vitro 3-D model of breast cancer dormancy points towards dormancy induction by OVOL proteins

Because the OVOL transcriptional repressors have been implicated in epithelial lineage differentiation and the establishment of hybrid E/M states (18–22), we assumed that their overexpression in mesenchymal breast cancer cells would promote epithelial differentiation and simulate phenotypes sharing functional features with dormant DTCs. To test this prediction, we first established inducible overexpression (iOE) of OVOLs in the highly mesenchymal breast cancer cell line MDA-MB-231, carrying mutant forms of p53, BRAF, KRAS and NF1 (fig. S1A). Next, we adopted a well-established model of cancer dormancy based on the observation that plating dormant breast cancer cells on a 3-dimensional extracellular matrix, called 3D-BME, decelerates proliferation due to nuclear expression of cyclin-dependent kinases, but non-dormant cells retain robust proliferation under the same conditions (13). In line with these observations, cells displaying growth arrest in the 3D BME system were found to be dormant in vivo (13, 32–34), and the experimental model has been adopted by several laboratories, including researchers who study dormancy of breast cancer (35) and osteosarcoma (36).

After calibrating the minimal concentration of doxycycline (DOX) that was sufficient for long-term OVOL1/2 induction, we diluted the cells, including the control EV (empty vector) cells, in media containing Basement Membrane Extract (BME). Next, the cells were overlaid on a gelled BME layer and visualized. Notably, a previous study that made use of this 3-dimensional system attributed important roles in the transition from dormancy to metastatic growth to collagen, actin fibers and integrin (32). In contrast to the actin stress fibers displayed by un-induced MDA-MB-231 cells, F-actin displayed cortical distribution in cells overexpressing either OVOL1 or OVOL2 (fig. S1B), and the cells changed their morphology from an elongated to a more rounded appearance (fig. S1C) characteristic to arrested non-metastatic cells (13). Consistent with the observed cytoskeletal reorganization, the rapidly proliferating MDA-MB-231 cells almost fully arrested their growth in the 3-dimensional matrix once the expression of OVOL1 or OVOL2 was induced, but the empty vector cells and the DOX-untreated cells growth was unaffected (fig. S1, D and E). This demonstration that OVOLs can confer dormancy traits, in vitro, to highly aggressive breast cancer cells prompted our subsequent efforts to resolve the mechanisms underlying epithelialization and dormancy.

OVOL1 functions as a mammary epithelium differentiation marker regulated by growth factors, hormones, and malignancy

In similarity to the SNAIL family of EMT-TFs, OVOLs are zinc-finger transcriptional repressors that bind with epigenetic regulators through N-terminal SNAG domains ((37); Fig. 1A). Despite homologous SNAG motifs and shared domain configurations, SNAILs repress but OVOLs increase the expression of E-cadherin (38). In line with distinct roles, mRNA analysis of >50 mammary cell lines (39) clearly separated OVOL1/2 from SNAIL1/2 and other mesenchymal markers (fig. S1F). Further analysis confirmed the absence of OVOL1 in claudin-low/ZEB1-positive cell lines (fig. S1G), which are enriched for EMT markers (40). In addition, evaluation of data retrieved from the Cancer Genome Atlas Breast Invasive Carcinoma Collection (TCGA-BRCA; 1,084 patients), well separated the OVOL family from the SNAILs and other mesenchymal markers, such as ZEB1 and TWIST2 (Fig. 1B). Aside from cell lines and patients, we analyzed OVOL1/2 in a large biobank of patient-derived tumour xenograft (PDTX) models representing the diversity of human breast cancer (41) (fig. S1H). This uncovered a similar pattern in which OVOL1/2 were expressed in many models, whereas the EMT markers were, in general, weakly expressed. Of note, in both PDTX and cell line models, OVOL1 was highly correlated to epithelial lineage markers (fig. S1I). In conclusion, although the OVOL and SNAIL families engage DNA response elements and epigenetic modifiers through homologous motifs, they respectively belong to distinct epithelial and mesenchymal groups of genes.

Figure 1: OVO-L1 is expressed in human and murine mammary cancers and acts as an EGF- and steroid hormone-inducible epithelial marker.

(A) The domain structures of the OVO-like (OVOL) and the SNAIL (SNAI) families are shown. (B) Transcriptional profiles of mesenchymal genes (blue) and epithelial genes (red) were retrieved from the TCGA-BRCA resource (Cancer Genome Atlas Breast Invasive Carcinoma; 1,084 patients) and subjected to principal component analysis (PCA). OVOL and SNAI family members are highlighted. Contours correspond to the estimated gene density distribution. (C) Densities and box plots comparing the expression of OVOL1 and OVOL2 in human normal breast tissue (144 samples) and tumor samples (1980 samples; METABRIC dataset). The Welch two-sample t-test (parametric) and Mann-Whitney test (non-parametric, two-sample Wilcoxon rank-sum test) were used. (D) Shown are β-galactosidase–stained tissues from female mice expressing a murine Ovol1 promoter-reporter construct. The sections were counterstained with eosin. Scale bars, 100 μm. Images are representative of 4 mice per group. (E and F) MCF10A cells were treated with EGF (50 ng/ml) for the indicated time intervals. The mRNA levels of OVOL1 and OVOL2 were determined using RT-PCR (E). Alternatively, cell extracts were subjected to an immunoblotting assay that used antibodies specific to the indicated proteins (F). Data are representative of 3 independent experiments. (G) OVOL1 promoter activity was measured in HEK293T cells after 3 hours of treatment with vehicle or EGF (50 ng/ml). The luciferase luminescence signal was normalized to secreted alkaline phosphatase (SEAP) to control for transfection efficiency. Shown are the results of 4 independent transfection experiments; error bars represent the mean ± SEM. (H) Inducible overexpression (iOE) of an OVOL1 protein tagged with a V5 peptide was pre-established in MCF10A cells. These cells, along with the control (empty vector; EV) cells, were treated for the indicated time intervals with the inducer, doxycycline (1μg/ml). Lysates were subjected to immunoblot analysis, as indicated. Vinculin was the reference control for equal gel loading. Blots are representative of 2 independent experiments. (I) MCF7 and T47D cells were treated for 24 hours with estrogen, progesterone, testosterone (DHT), dexamethasone and isoproterenol (each at 100 nM). Cell extracts were analyzed using the indicated antibodies. Blots are representative of 4 independent experiments.

Analysis of METABRIC (42), a breast cancer dataset (1980 patients) that includes 144 normal mammary samples, revealed that the expression levels of OVOL1 and OVOL2 are higher in tumors relative to normal mammary tissues (Fig. 1C; p<2.2e-16). This was confirmed by analyzing OVOL1 in another dataset, TCGA-BRCA (The Cancer Genome Atlas - Breast Invasive Carcinoma), which showed that OVOL1’s levels are significantly lower in normal breast tissue compared to all subtypes of breast cancer, including the basal and the luminal B class. To independently establish these observations, we examined the expression of the murine form of OVOL1. An Ovol1 promoter fused to the β-galactosidase gene was first examined in Ovol1^{tm1a(KOMP)Wtsi} transgenic mice. As previously reported (43), the promotor was active in sebaceous glands and the interfollicular epidermis of the skin (Fig. 1D). However, we detected weaker or no signals in the mammary glands of both virgin and lactating mice. Next, we bred the transgenic mice with MMTV^PyMT/+ animals, an accepted model of invasive breast cancer, which expresses the polyomavirus middle T oncogene in the mammary gland (44). All mammary gland tumors observed in the MMTV^PyMT/+/Ovol1^{tm1a(KOMP)Wtsi} animals stained positively for β-galactosidase (Fig. 1D), suggesting that undefined oncogenic cues increase the expression of OVOL1 in both human and murine mammary glands.

Common oncogenic cues might include growth factors and steroid hormones. Hence, we first asked if EGF can regulate OVOL1/2 expression. Treating untransformed MCF10A mammary cells with EGF revealed that OVOL1’s mRNA and protein levels peaked 1 to 2 hours after stimulation, but those of OVOL2 remained undetectable (Fig. 1, E and F), likely because OVOL1 represses the promoter of OVOL2 (45). By means of an OVOL1 promoter-reporter construct, we confirmed the ability of EGF to up-regulate OVOL1 transcription (Fig. 1G). Next, we used OVOL1’s cDNA tagged with a V5-peptide to establish MCF10A derivatives that inducibly overexpress an allele of OVOL1 (Fig. 1H). Note that tagging OVOL1 with V5 does not affect binding to DNA elements and activation of transcription (46). In line with epithelial characteristics, overexpression of OVOL1 was paralleled by decreased abundance of vimentin. Because MCF10A cells are devoid of estrogen and progesterone receptors (ER and PR, respectively), we used luminal cell lines and found that estrogen and glucocorticoids can respectively down- and up-regulate the abundance of OVOL1, but OVOL2 was less responsive, and other agents, including isoproterenol, which stimulates the β-adrenergic receptor, displayed less consistent effects (Fig. 1I). In summary, OVOL1 and OVOL2 are lowly expressed in mesenchymal breast cancer cell lines and in PDTXs, but expression of OVOL1, more than OVOL2, can be controlled by growth factors and hormones. Moreover, malignant transformation increases OVOL1 expression in the murine mammary gland.

High OVOL1 predicts poor prognosis of patients with ER-low tumors and associates with reduced cell proliferation, migration, and matrix degradation

To uncover the potential clinical significance of OVOL1/2, we divided the METABRIC dataset (42) into two groups according to OVOL1 transcript levels. This showed only small differences in terms of patient survival time (Fig. 2A). However, when patients were grouped into ER-low and ER-high cohorts, we observed better curve separation, especially in the ER-low cohort (fig. S2A), in line with an association between low ER, high OVOL1, and an aggressive disease. This conclusion was strengthened by similar analyses we performed with data from TCGA-BRCA, as well as by immunohistochemical (IHC) analyses of OVOL1 levels (Fig. 2B). However, similar analyses of OVOL2 were less informative.

Figure 2: OVOL1 is associated with poor survival of patients with ER-negative breast cancer; its overexpression reduces viability, invasiveness and migration of breast cancer cells.

(A) The METABRIC mRNA dataset (1980 patients) was divided into two equal groups, high OVOL1 expression (>50^th percentile) and low OVOL1 expression (<50^th percentile). Shown are analyses of overall patient survival per group. (B) The METABRIC dataset was stratified according to the expression of the estrogen receptor (ER), and disease-specific patient survival (DSS) probability was calculated. The abundance of the OVOL1 protein was based on immunohistochemical (IHC) analyses. The p and n values are indicated. (C) MDA-MB-231 and BT-549 cells that inducibly overexpress OVOL1 or OVOL2 were plated in 96-well plates and incubated overnight. At the indicated time points, the medium was replaced with full medium containing doxycycline (1 μg/ml). Thereafter, cells were incubated for 2 hours at 37°C with MTT (0.5 mg/ml) containing media and the formazan crystals formed by metabolically active cells were dissolved. Absorbance was determined at 570 nm. The histograms present the signals determined at the end of the 5-day interval. The experiment was performed using 3 technical replicates in each of 4 biological replicates. (D and E) MDA-MB-231 cells stably expressing OVOL1 or an empty vector were starved for serum factors for 20 hours. Thereafter, 40,000 cells were seeded on 0.4 μ slides (from Ibidi) that were pre-coated with fibronectin. After incubating for 2 hours to let the cells adhere, the medium was replaced with a medium containing albumin and EGF, without or with TGF-β. Cells were imaged every 15 min for 6 hours. The movement and velocity of 80 individual cells from 3 independent experiments were quantified and analyzed using Fiji and the Chemotaxis and Migration Tool from Ibidi. (F and H) MDA-MB-231 cells (40,000 per well; 12-well trays) were plated on Oregon-488 conjugated gelatin. Following a 24-hours long incubation either untreated or treated with doxycycline (DOX; 0.25 mg/ml), cells were fixed and stained with Alexa 568-phalloidin and DAPI. Data are from biological triplicates. Statistical significance of the numbers of invadopodia per cell (G) and the area of degraded gelatin (H) was analyzed using one-way ANOVA. EV, empty vector.

To permit additional analyses of OVOL1/2, we established yet another iOE derivative, of BT-549 cells, which harbor partial RB gene deletions and p53 mutations. Consistent with the possibility that OVOLs confer quiescence, a distinctive hallmark of tumor dormancy (7, 47), DOX-induced overexpression of either OVOL gene inhibited proliferation of both BT-549 and MDA-MB231 cells (fig. S2B; note the stronger effect of cytosine arabinoside, Ara-C, a reference chemotherapeutic agent). Two additional assays were consistent with OVOL-induced growth arrest: MTT [3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide] assays, which measure mitochondrial activity (Fig. 2C), and colony formation tests (fig. S2C).

To examine the prediction that OVOL1 can reduce cell migration, in line with the immotile phenotype of DTCs that reached dormancy (48), we began by establishing MDA-MB-231 cells that stably expressed OVOL1. The cells were starved for serum factors and thereafter they were placed in fibronectin-coated chambers suitable for migration assays. Notably, overexpression of OVOL1 did not change the rate of spontaneous (basal) cell migration. In contrast, the ability of EGF, either alone or in combination with the transforming growth factor beta (TGF-β), to accelerate cell migration, which was clearly manifested by the control cells, was nearly lost when OVOL1-overexpressing cells were examined (Fig. 2, D and 2E). These observations implied that the upstream signals regulating cell migration, rather than the migration machinery, are controlled by OVOL1. To address matrix degradation, we assayed the formation of invadopodia, protrusive structures that remodel the extracellular matrix (ECM) (49). DOX-inducible MDA-MB-231 cells were plated on fluorescent gelatin, and after 24 hours they were fixed and stained to detect actin-rich puncta co-localizing with gelatin degradation spots. The results presented in Figures 2F–H indicated that inducible overexpression of either OVOL1 or OVOL2 significantly reduced the formation of invadopodia, as well as diminished areas of matrix degradation, implying that not only inducible motility but also matrix degradation is inhibited by OVOLs.

Next, we examined in animals the effects of an overexpressed OVOL1 on tumorigenic growth and metastatic spread of MDA-MB-231 cells. Prior to implanting iOE cells in the fat pad of female mice, we either untreated or treated cells with DOX. In addition, by supplementing the animals’ drinking water with DOX, treatment of a second group of mice was continued post-implantation. A third group received DOX only after implantation. Regardless of the mode of DOX treatment, overexpression of OVOL1 strongly inhibited tumor growth (fig. S2, D and E). As expected, analysis of tumor extracts confirmed induction of OVOL1, along with DOX-induced increased abundance of ZEB1 and vimentin (fig. S2F). Alongside, we observed increased abundance of two forms of C1ORF116, a transcriptional target of OVOLs (see below). Although control mice displayed more metastatic lung nodules, normalization to primary tumor weight weakened the possibility that OVOL1 markedly altered metastasis in this experiment (fig. S2, G and H). In conclusion, OVOL1 is associated with poor prognosis of ER-low patients but its overexpression can reduce growth rates in cell cultures, as well as in a xenograft model.

RNA sequencing and protein mass spectrometry implicate OVOL1 in hormone response and epithelialization, as well as uncover induction of C1ORF116

RNA sequencing and proteomic analyses of MDA-MB-231 cells overexpressing OVOL1 (Fig. 3A and data files S1 and S2) concordantly uncovered increased expression of genes involved in a broad spectrum of cellular functions, such as RHOD, an atypical Rho GTPase that controls actin filaments (50), and SH2D3A, which is induced by hypoxia (51). Further analysis of the RNA data indicated that genes involved in the response to estrogen were enriched in OVOL1-overexpressing cells (fig. S3A), whereas mesenchymal genes were enriched in the control cells (Figs. 3, B and C). In accordance, RT-PCR confirmed decreased abundance of vimentin and increased E-cadherin in the OVOL1-overexpressing cells (Fig. 3B). Likewise, one of the characteristic epithelial genes that underwent increased abundance encodes an EGFR ligand, amphiregulin (AREG), which was verified using culture media (Fig. 3D). We also noted a DOX-induced decrease and increase in the abundance of two additional transcripts encoding NR2F1 (nuclear receptor subfamily 2, group F) and C1ORF116, respectively. It has been reported that NR2F1 induces a partial EMT program and acts as a barrier to dissemination (52), but C1ORF116 has not previously been implicated in cancer dormancy or epithelial differentiation. We decided to focus on this poorly characterized protein because it includes no recognizable protein domain, has no family members, and contains an AR (androgen receptor) response element (53), reminiscent of our results linking OVOLs to steroid hormones. Inducible overexpression of OVOL1/2 confirmed increased expression of two C1ORF116 forms (Fig. 3E). These forms displayed variable ratios in mammary cell lines (fig. S3B). Along with C1ORF116, overexpression of OVOL1/2 drove increased expression of E-cadherin and KLF4, a protein implicated in epithelial differentiation (26, 54). Immunofluorescence analyses confirmed induction of C1ORF116 in the cytoplasm of MDA-MB-231 cells expressing an inducible allele of OVOL1 (fig. S3C), enhanced E-cadherin and EpCAM, and decreased levels of N-cadherin, a mesenchymal marker (Fig. 3F and fig. S3D). In summary, the uncovered transcriptional targets of OVOLs are associated with increased and decreased expression of epithelial and mesenchymal markers, respectively. Notably, because OVOLs act as transcriptional repressors, the observed increased abundance of C1ORF116 likely represents a secondary effect.

Figure 3: Overexpression of OVOL1 is likely associated with induction of the epithelial lineage and breast cancer dormancy.

(A) RNA and proteins were isolated from control MDA-MB-231 cells and derivatives overexpressing OVOL1. Shown are the results of both RNA sequencing analysis (log2 fold change of at least +/−1, an adjusted p-value of < 0.05, and a base mean > 5; data file S1) and mass spectrometry analysis of extracted proteins (K-nearest neighbor algorithm, KMN; data file S2). The abundance of genes/proteins indicated in red is significantly increased, whereas genes/proteins colored in blue are significantly downregulated in OVOL1-overexpressing cells. The names of relevant upregulated and downregulated genes are framed in the plot. RNA-Seq data was generated from 4 biological replicates, and the proteomics data from 5 biological replicates. (B) qPCR was applied on RNA isolated from control and OVOL1-overexpressing MDA-MB-231 cells. The experiment was performed in 4 biological replicates with 4 technical replicates. (C) Shown are gene set enrichment analyses (GSEA) and the enriched hallmarks that were derived from the RNA-seq data in the OVOL1 overexpressing cells (left) and in the empty vector (EV) control cells (right). (D) Media conditioned for 3 days by MDA-MB-231 cells that were treated, or not, with doxycycline (DOX, 1 μg/ml) were subjected to an ELISA specific to AREG. The results are summarized in histograms. The experiment was performed three times in triplicates. (E) Control MDA-MB-231 and BT-549 cells (EV), along with the corresponding DOX-inducible overexpressing cells (OVOL1-iOE and OVOL2-iOE) were treated with DOX (1 μg/ml) for the indicated time intervals. Cleared cell extracts were subjected to immunoblotting with the indicated antibodies; vinculin was used as a loading control. Blots are representative of 2 independent experiments. (F) Shown are the results of immunostaining for E-cadherin and N-cadherin in OVOL1-iOE, OVOL2-iOE and control (EV) cells. Cells were treated with DOX (1 μg/ml) for 72 hours and later they were seeded on glass slides and incubated for 72 additional hours. At the end of the incubation, cells were fixed, probed using the indicated antibodies and imaged using a spinning disk microscope. Scale bar, 50 μm. Framed areas in the left-column images are enlarged in the right column. Images are representative of 5 fields analyzed from each condition in each of 3 independent experiments. (G) The indicated iOE derivatives of MDA-MB-231 cells, including the control empty vector (EV) cells, were treated for 72 hours with the inducer, doxycycline (DOX; 1 μg/ml). Cleared cell lysates were subjected to immunoblot analysis using the indicated antibodies. Vinculin was to the loading control reference. Blots are representative of 3 independent experiments. (H) RNA expression data derived from the METABRIC database of patients with breast cancer was analyzed for correlation between C1ORF116 and expression of the indicated transcripts. Patient data were stratified according to ER status. Correlations with p-values >0.05 are not shown. Correlation scores corresponding to ER-negative tumors are shown in red, and correlations corresponding to ER-positive tumors are shown in blue. (I) Bulk RNA sequencing data from Rosano et al. (56) were analyzed for the abundance of transcripts corresponding to OVOL1 and C1ORF116. They employed two MCF7 treatment models of drug-induced cancer dormancy: TAM (tamoxifen) treatment and treatment with aromatase inhibitors (estrogen deprivation). Messenger RNA was harvested at the indicated time intervals and subjected to bulk sequencing. Transcript levels were compared to the untreated arm. Note that latency refers to the time between treatment onset and dormancy entry. The data were analyzed using DESeq2.

Oncogenic RAS mutants downregulate OVOL1, the expression of which is associated with C1ORF116, and both are increased in models of cancer dormancy

To further investigate the ability of OVOLs to promote an epithelial phenotype, we examined the prediction that an oncogenic RAS mutant, widely known as a promoter of mesenchymal phenotypes (55), can downregulate OVOLs in human mammary cells. To this end, we stably expressed a mutant allele of HRAS in MCF10A cells and confirmed, using PCR, that the epithelial markers E-cadherin and p63 were downregulated, whereas the mesenchymal markers ZEB1, vimentin and fibronectin were upregulated (fig. S3E). As predicted, we observed HRAS-induced increased abundance of both OVOL1 and OVOL2, in line with their inferred roles as epithelialization genes. Two additional lines of evidence lent support to the ability of OVOLs to bias epithelialization: (i) DOX-induced OVOL1 and OVOL2 increased expression of two of the major markers of the epithelial phenotype, HER2 and ER (Fig. 3G), and (ii) analysis of RNA expression data from METABRIC validated positive association between OVOLs and C1ORF116, as opposed to negative association with several epithelial markers (Fig. 3H).

To examine potential associations between cancer dormancy and both OVOL1 and C1ORF116, we analyzed a dataset published by Rosano and colleagues (56). These authors proposed that endocrine therapies can induce nongenetic cell state transitions into dormancy. To demonstrate this, they expanded barcoded MCF7 cells and created multiple replicates assigned to long-term estrogen deprivation (“-E2”, by means of a simulating treatment with aromatase inhibitors) or tamoxifen treatment (“TAM”). Critically, both the -E2 and TAM groups entered a period of presumed dormancy after a few weeks of treatment, and they were maintained under pharmacological pressure until suspected awakening (early progression). We re-analyzed the data from Rosano and colleagues and summarized the results (Fig. 3I) as the average transcript levels of OVOL1 and C1ORF116 in the -E2 and TAM replicates. Notably, both transcripts exhibited significantly increased abundance relative to baseline levels during the latency period (the time between treatment initiation and the onset of dormancy). Furthermore, the levels of C1ORF116 continued to rise in both groups during dormancy and prior to the awakening event. Thus, along with supporting the functional link between OVOLs and OVOL1, these findings are consistent with the role we attribute here to OVOL1 and C1ORF116 in tumor dormancy.

C1ORF116 acts as an intrinsically disordered, putative autophagy receptor

In similarity to the hormonal control of OVOL1/2, we observed increased and decreased abundance of C1ORF116’s protein bands following treatment of breast cancer cells, especially T47D cells, with testosterone (or dexamethasone) and estrogen, respectively (Fig. 4A) (53). Structure prediction uncovered an intrinsically disordered 3-dimensional configuration of C1ORF116, which includes three alpha-fold clusters embedded in highly conserved regions (Fig. 4B and fig. S4A). Although no recognizable protein domain was identified, our search for short motifs, which used iLIR (57), found three putative LC3-interacting regions (LIRs; Fig. 4C). Further analysis of extended LIRs (xLIRs) against a custom position-specific scoring matrix (PSSM) assigned to the SSYDFL sequence the highest score. Notably, LIR motif-containing proteins are involved in autophagy, a degradation process that selectively targets intracellular components (58). The autophagosomes, double-membrane vesicles, are trafficked to lysosomes along with their payloads. This process involves several autophagy-related (ATG) proteins, including yeast ATG8 (LC3 and GABARAPs in mammals) and a unique type of LIR-harboring adaptor proteins, termed autophagy receptors. Consistent with the possibility that C1ORF116 functions as an autophagy receptor, sequence alignments identified homologies between the N-terminal LIR motif of C1ORF116 and the LIRs of several autophagy receptors, including p62/SQSTM1 (fig. S4B).

Figure 4: C1ORF116, a hormone-inducible intrinsically disordered protein, might act as an autophagy receptor.

(A) MCF7 and T47D breast cancer cells were incubated for 24 hours with the indicated hormones or with isoproterenol (100 nM). Cleared cell lysates were probed for C1ORF116 and a loading-control protein (vinculin or tubulin). Blots are representative of 4 independent experiments. (B) The following parameters of C1ORF116 were analyzed: (i) Evolutionary conservation based on analysis of 73 homologous sequences (ConSurf server; https://consurf.tau.ac.il/consurf_credits.php), (ii) 3D structure, including helix (orange) and loop (yellow) regions predicted by AlphaFold (https://alphafold.ebi.ac.uk/about), and (iii) prediction of disorder using two energy estimation approaches: IUPRED (blue line; https://iupred2a.elte.hu/) and Anchor (red line; http://anchor.elte.hu/). (C) The iLIR resource (57) was used for the analysis. The summary table lists the three putative LC3-interacting regions (LIRs) of C1ORF116. Also shown are the results of scoring extended LIRs (xLIRs) against a custom position-specific scoring matrix (PSSM). PSSM scores of 13–17 are considered highly accurate, but some verified LIRs might have scores as low as 7 (https://ilir.warwick.ac.uk/index.php). (D) The indicated cells were pre-incubated for 72 hours in a medium supplemented or not with DOX. Afterwards, 40,000 cells were seeded in 8-well glass bottom chambers and incubated overnight. An autophagy assay kit (from Abcam) was used to visualize cells undergoing autophagy (green). Images (400X magnification) were obtained using a spinning disk microscope (scale bar, 50 μm) and are representative of 5 fields analyzed from each condition in each of 2 independent experiments. (E) Listed are the results of a yeast 2-hybrid screen that used a human breast tumor epithelial cell library. A fragment of C1ORF116 (amino acids 1–601) was used as bait. The screen identified 136 positive clones among 129 million interactions. The number of prey fragments identified is indicated, along with the Predicted Biological (PB) score, in which “A” indicates the highest interaction confidence. (F) The 30 uppermost putative interaction partners of C1ORF116 from the yeast 2-hybrid screen (data file S3) were subjected to pathway enrichment analysis using EnrichR. The dot plot presents the corresponding GO processes. The size of each dot represents the number of genes, whereas the color of each dot represents the corresponding database analyzed.

Because ubiquitin serves as a signal for selective autophagy and the autophagy receptor p62/SQSTM1 both binds with ubiquitin and undergoes ubiquitination (59), we examined the possibility that C1ORF116 can undergo ubiquitination. As predicted, probing C1ORF116 immunoprecipitates with an anti-ubiquitin antibody confirmed that both forms of C1ORF116 undergo ubiquitination (fig. S4C). Next, we utilized ubiquitin-agarose beads that were incubated with extracts of cells that were pre-transfected with a V5-tagged C1ORF116 plasmid. In similarity to p62/SQSTM1, C1ORF116 was specifically pulled down by the immobilized ubiquitin (fig. S4, D and E). Because these results potentially implicated C1ORF116 in autophagy, we utilized a dye that labels autophagic vacuoles. This revealed that overexpression of either C1ORF116, OVOL1 or OVOL2 enhanced autophagy in both MDA-MB-231 and BT-549 cells (Fig. 4D). In conclusion, the poorly characterized C1ORF116 protein might serve as an autophagy receptor downstream of OVOL1/2. Notably, it has been reported that autophagy prolongs breast cancer cell dormancy (60) because it permits long-term cell survival under nutritional stress (48, 61).

C1ORF116 binds with LC3/GABARAP family proteins and two key redox enzymes

To identify the target proteins of C1ORF116, we performed yeast two-hybrid (Y2H) screens. The N-terminal portion of C1ORF116 was used as bait while screening breast cancer cDNA libraries. The screen identified 136 positive clones (data file S3). The uppermost interactors identified were GABARAPL2, a member of the LC3 family, the catalytic subunit of glutamate cysteine ligase (GCLC), and TRX (thioredoxin; Fig. 4E). Notably, TRX and GCLC control the cellular redox potential by respectively acting as a major scavenger of reactive oxygen species (ROS) and an enzyme that mediates a rate-limiting step in glutathione biosynthesis (62). These observations are consistent with the identification of C1ORF116 as a putative autophagy receptor and indeed pathway enrichment analysis of the 30 uppermost C1ORF116’s interaction partners from the Y2H screen putatively identified autophagy, along with AR signaling, as the most relevant pathways (Fig. 4F).

To validate the results of the Y2H screens, we first analyzed the subcellular localization of C1ORF116, GABARAPs, and GCLC. These immunofluorescence analyses indicated that all three proteins are primarily cytoplasmic (Fig. 5A). In addition, we detected areas of co-staining as well as co-localization with C1ORF116 in some puncta. Next, we performed co-immunoprecipitation experiments that used a plasmid encoding a tagged C1ORF116. The results validated specific interactions between C1ORF116 and the endogenous forms of both GABARAPL2 (Fig. 5B) and GCLC (Fig. 5C). To address TRX, we immunoprecipitated this relatively small but abundant protein from lysates of cells that were pre-transfected with increasing amounts of the V5-C1ORF116 plasmid (Fig. 5D). This confirmed the existence of TRX-C1ORF116 complexes and revealed gradual diminution of TRX. However, we detected no parallel diminution when analyzing the rather high GCLC levels. Conceivably, by acting as an autophagy receptor, C1ORF116 physically binds with both GCLC and TRX, but only the latter is detectably sorted for degradation. Assuming autophagic degradation of TRX, we compared the effects of the proteasome inhibitor, MG132, and bafilomycin A1, which prevents maturation of autophagic vacuoles. Unlike MG132-treated cells, in cells treated with bafilomycin A1, we detected a ubiquitinated form of TRX, which disappeared following overexpression of C1ORF116 (Fig. 5E). Thus, by binding with and enhancing ubiquitination of TRX, C1ORF116 likely sorts this antioxidant protein to autophagic degradation. Notably, in line with the inferred model attributing to OVOL1/2 and C1ORF116 the ability to harness autophagy and redox homeostasis, it has previously been proposed that redox mechanisms control tissue colonization by dormant breast cancer cells (28, 63).

Figure 5: C1ORF116 binds with GABARAPL2, GCLC, and TRX, and its overexpression increases intracellular ROS and decreases glutathione.

(A) MDA-MB-231 cells overexpressing an inducible V5-C1ORF116 were plated in round-bottom glass wells. After treating the cells with DOX for 3 days, cells were fixed, and immunolabelled using anti-V5 (red), anti-GABARAP (green), anti-GCLC (green) or anti-thioredoxin (green) antibodies. DAPI was used to stain nuclei (blue). Scale bar, 20 μm. Images are representative of 5 fields analyzed from each condition in each of 3 independent experiments. (B) MDA-MB-231 cells stably transfected with either an empty vector (EV) or a plasmid encoding V5-C1ORF116 were lysed. The cleared extracts were subjected to immunoprecipitation (IP) using protein A/G magnetic beads coated with either normal mouse immunoglobulin G (IgG) or an anti-V5 antibody. A fraction of the lysate (5%) was separately analyzed (Input). After protein transfer to membranes, we performed immunoblotting (IB) with antibodies specific to the indicated proteins. (C) HEK293T cells grown in 90-mm dishes were un-transfected, transfected with an empty vector (EV) or with a C1ORF116-V5 plasmid. Forty-eight hours later, the cells were extracted, and the cleared extracts were subjected to immunoprecipitation using protein A/G beads and antibodies specific to GCLC. As a control, we mixed extracts from the EV- and the C1ORF116-transfected cells and used control beads. (D) HEK293T cells were transfected with an empty vector or with increasing amounts of the C1ORF116-V5 plasmid (0, 0.5, 0.75, 1.5, and 3 μg per plate). Forty-eight hours later, the cells were extracted, and a portion of each cleared extract was subjected to immunoprecipitation (IP) with either control immunoglobulins or an anti-TRX antibody. Whole extracts (input) along with the immunoprecipitates were immunoblotted (IB) using the indicated antibodies, including antibodies specific to TRX and GCLC. (E) HEK293T cells grown in 10-cm plates were transfected with a control plasmid (EV) or with plasmids encoding HA-ubiquitin and C1ORF116-V5. Following 48 hours of incubation, cells were treated with either bafilomycin (20 nM; 16 hours) or MG132 (10 μM; 8 hours). Thereafter, we extracted all cells and immunoprecipitated TRX using agarose protein A/G beads pre-coated with an anti-TRX antibody. The washed immunoprecipitates or cleared extracts were subjected to immunoblotting using the indicated antibodies. (F) Inducible overexpression (iOE) of the C1ORF116 protein tagged by a V5 peptide was pre-established in both MDA-MB-231 and BT-549 cells. These cells were treated for the indicated time intervals (in hours) with doxycycline (1 μg/ml). Lysates were subjected to immunoblot analysis using antibodies specific to the indicated proteins. GAPDH was used to ensure equal gel loading. (G) The indicated derivatives of MDA-MB-231 cells were treated with doxycycline (1 μg/ml) for 72 hours to induce the overexpression of OVOL1, OVOL2, or C1ORF116. The corresponding cell extracts were subjected to an assay that determined the intracellular levels of reduced glutathione (GSH). Each column represents the mean ± SEM of triplicates of at least three independent experiments. (H) Shown are the results of a hydrogen peroxide assay that employed DCFDA (2ʹ,7ʹ-dichlorofluorescin diacetate). Representative images of the inducible MDA-MB-231 cells overexpressing OVOL1, OVOL2 or C1ORF116 are presented. The images were obtained using epifluorescence microscopy (original magnification X100). Scale bar, 200 μm. The histogram presents the quantification of hydrogen peroxide levels by densitometric analysis. NAC (N-acetyl-l-cysteine; 10 mM) was used as a ROS scavenger, and oltipraz (50 μM) was used to increase ROS. Data are mean +SEM of triplicates of three independent experiments. Blots in (B to F) are each representative of 2 or more independent experiments. Dashed lines in (D and E) indicate the position of the EV control in (D) and the boundary between EV and C1ORF116 cell lines in (E).

The OVOL-C1ORF116 axis controls ROS and glutathione levels

Following the results of the Y2H screens, we predicted that overexpression of either C1ORF116 or OVOL1/2 would downregulate antioxidants and increase ROS levels. To examine this scenario, we established inducible overexpression of C1ORF116 in MDA-MB-231 and BT-549 cells (Fig. 5F) and determined the levels of reduced glutathione (GSH) following induction treatment with doxycycline. The assays utilized two references: oltipraz, which increases superoxide radicals, and N-acetyl cysteine (NAC), which quenches ROS. Relative to the control cells, C1ORF116-overexpressing cells displayed lower basal levels of GSH, and a further reduction was observed after induction (Fig. 5G and fig. S5A). Similar analyses of OVOL1-iOE and OVOL2-iOE cells confirmed the inducibly increased abundance of GSH (Fig. 5G).

We then applied a bioluminescence test that assayed hydrogen peroxide (H₂O₂). The assay was based on a cell-permeant fluorogenic dye, DCFDA (2ʹ,7ʹ-dichlorofluorescin diacetate), that probes hydroxyl, peroxyl, and other ROS activities. The results (Fig. 5H and fig. S5B) demonstrated the ability of both C1ORF116 and OVOL1 to upregulate ROS levels. Note, however, that OVOL2 exerted relatively small effects in MDA-MB-231 cells. To ensure its pivotal role in the induction of ROS, we knocked-down C1ORF116 using siRNA. As we expected, pre-treatment with the specific siRNA reduced the abundance of ROS in all our iOE cell lines (fig. S5C). However, siRNA-treated OVOL1-iOE cells still displayed substantial ROS signal, suggesting that OVOL1 can modify the redox potential through additional mechanisms. In line with this prediction, re-analysis of the RNA-seq data revealed changes in the expression of genes encoding glutathione peroxidase 8 and several glutathione S-transferases as well as NCF2, which generates superoxide bursts, and cytochrome P450 1B1, which is involved in hyperoxic toxicity. In conclusion, in line with the ability of C1ORF116 to bind with both TRX and GCLC, the overexpression of either C1ORF116 or OVOL1/2 increased ROS and reduced GSH levels. Still, additional transcriptional targets of OVOLs likely regulate the redox potential of mesenchymal mammary cells once they acquire an epithelial phenotype.

The OVOL-C1ORF116 axis controls NRF2, metabolites, and antioxidants

Assuming that the effect of OVOLs on the cellular redox potential is accompanied by changes in additional metabolites, other than ROS and glutathione, we analyzed polar small molecules using mass spectrometry (data file S4). The resulting heatmap of the normalized abundance of selected metabolites (Fig. 6A) shows that, as expected, the metabolic effects induced by OVOLs were more varied and greater than those induced by overexpression of C1ORF116, the downstream target of OVOLs. Among the top differentially downregulated metabolites in both OVOL2- and OVOL1- overexpressing cells we identified taurine, an abundant non-essential amino acid that maintains glutathione stores, and both the oxidized and the reduced forms of glutathione. In addition, among other alterations, several amino acids were upregulated in OVOL-overexpressing cells, including glutamine and the cysteine-glycine dipeptide, a derivative of glutathione. Together, the observed effects of OVOLs on metabolism might complement the action of C1ORF116 on the cellular redox potential and reflect the proposed ability of autophagy to recycle amino acids in dormant cancer cells.

Figure 6: C1ORF116 displays genetic and epigenetic aberrations in breast cancer; the upstream OVOL axis is active in epithelial E/M hybrid states and regulates both metabolism and NRF2.

(A) Stable derivatives of MDA-MB-231 cells that overexpress OVOL1, OVOL2 or C1ORF116 (bold type), along with Empty Vector (EV; normal type) control cells, were treated for 72 hours with doxycycline or vehicle. Thereafter, 5 million cells were snap frozen, and polar metabolites were determined using mass spectrometry. The heatmap shows the normalized abundance of selected metabolites (4 biological replicates per group). (B) The indicated DOX-inducible derivatives of BT-549 cells were untreated or treated with DOX for 72 hours. Thereafter, cells were counted, seeded in glass slides, and treated (or not) for an additional 72 hours. This was followed by fixation and confocal microscopy analysis of the subcellular distribution of NRF2 using a specific antibody. Hoechst was used to visualize nuclei. Scale bars, 20 μm. Blots are representative of 5 fields obtained from 2 independent experiments. (C) Shown are mRNA expression levels (upper panel) and methylation status (lower panel) of C1ORF116 in the ten Intrinsic Clusters (ICs) of breast cancer. ICs were derived from the invasive breast cancer cohort of TCGA. Statistical testing of RNA and methylation levels of C1ORF116 across the IntClust 10 groups was performed using ANOVA. (D) The TCGA dataset of breast cancer was used to derive total copy number (TCN) information for the indicated genes. Deletions were defined as TCN<2, Normal as TCN = 2, and Gains as TCN>2. (E) Shown are the levels of OVOL1 and OVOL2 expression in the previously reported stable cellular states that underlie EMT (71). TN refers to a triple negative state lacking expression of the epithelial cell adhesion molecule (EpCAM), as well as lacking expression of CD106 (Vascular Cell Adhesion Molecule 1; VCAM-1), CD51 (integrin alpha V; ITGAV) and CD61 (integrin beta-3; ITGB3). The data were derived from GEO record GSE110587. Error bars refer to ± SEM from 3 independent experiments.

The capacity of the OVOL-C1ORF116 axis to modify redox homeostasis raised the possibility that this would involve NRF2 (nuclear factor erythroid 2-related factor 2), a transcription factor capable of protecting cells against oxidative damage. Under unstressed conditions, NRF2 is unstable and retained in the cytoplasm, but in response to oxidative stress it translocates to the nucleus to activate antioxidant genes (64). We found that NRF2 resides mainly in the nuclei of control BT-549 (Fig. 6B) and MDA-MB-231 (fig. S5D) cells in the absence or presence of DOX. This is attributable to genetic or epigenetic aberrations characteristic of breast cancer cells, including frequent aberrations in PTEN (65, 66). Nevertheless, overexpression of either OVOL1 or OVOL2 translocated NRF2 to the cytoplasm (Fig. 6B and fig. S5D) and the following NRF2 target genes were transcriptionally repressed in OVOL1-overexpressing cells: GCLC, GSTP1, CAT and GPX4 (Fig. S5E). In contrast, NRF2 was retained in the nuclei of DOX-induced cells overexpressing C1ORF116, further implying that non-C1ORF116 mediated mechanisms contribute to the enhanced oxidative stress of OVOL-overexpressing cells. In summary, several mechanisms underlie the ability of OVOLs to enhance oxidative stress, and they include increased abundance of C1ORF116, inactivation of several antioxidants, and simultaneous suppression of the ubiquitous NRF2-centered protective machinery.

C1ORF116 controls growth of breast cancer cells and undergoes genetic and epigenetic aberrations in mammary tumors

It has recently been reported that C1ORF116 overexpression in thyroid cancer cells enhanced their proliferation (67). Consistent with this report, overexpression of C1ORF116 in MDA-MB-231 cells accelerated proliferation, but weaker effects were observed when analyzing the less rapidly proliferating BT-549 cells (fig. S6A). Notably, whereas colony formation assays detected only small effects in both cell lines (fig. S6B), CRISPR-CAS9 mediated ablation of C1ORF116 in MDA-MB-231 cells clearly reduced their rate of proliferation (fig. S6C). In line with growth stimulation, high expression levels of C1ORF116 are associated, albeit weakly, with shorter survival of patients with breast cancer (fig. S6D). Furthermore, we noted that this prognostic effect was more significant in the group of AR-negative patients, which is reminiscent of the ability of androgens to control C1ORF116’s promoter (53).

To better understand the clinical significance of C1ORF116, we surveyed the 10 integrative breast cancer clusters, which classify tumors according to their gene expression and copy number aberrations (68). The results (Fig. 6C, upper panel) display C1ORF116 expression by each integrative cluster (IntClust). The highest level was displayed by IntClust10, which is dominated by ER negativity and multiple chromosomal aberrations. In line with the ability of estrogen to inhibit the OVOL-C1ORF116 axis, the lowest C1ORF116 level was displayed by IntClust6, comprising ER-positive/HER2-negative tumors (luminal B). Assuming that additional mechanisms can downregulate C1ORF116 in luminal B tumors, we analyzed DNA methylation profiles of >1500 breast tumors (69). This revealed that IntClust6 is characterized by the highest C1ORF116 promoter methylation (Fig. 6C; lower panel), indicating that this gene is epigenetically repressed in IntClust6. In addition to aberrant methylation, we found that most tumors included in METABRIC exhibited significant copy number gains of the C1ORF116 gene (Fig. 6D). This includes MUC1, which is one of the most aberrantly overexpressed genes in breast cancer (70), with similarly extensive gains as C1ORF116. In conclusion, we identified C1ORF116 as a poor prognosis gene susceptible to genetic and epigenetic aberrations, as well as uncovered an association between AR negativity, high C1ORF116, and breast cancer aggressiveness.

OVOLs are highly expressed in two stable E/M hybrid states, EpCAM-positive and triple-negative cells

Mathematical modeling predicts that OVOL1 stabilizes a hybrid E/M phenotype characterized by cells exhibiting both epithelial and mesenchymal traits (24). Accordingly, OVOL1 overexpression can reprogram fibroblasts to epithelial cells and help breast cancer cells maintain the intermediate E/M phenotype while retarding the transition to mesenchymal states (17). Because this mechanism could explain the poor prognosis significance of OVOLs and their potential contribution to metastasis (22), we addressed OVOL’s relevance to the hybrid E/M states. To this end, we made use of a dataset derived from mouse tumors undergoing EMT (71). Originally, this analysis identified in animals several stable intermediary (hybrid) EMT states based on three surface proteins that together define all EMT transition states: CD106 (Vcam1), CD51 (ItgaV) and CD61 (Itgb3). The use of the three markers, in addition to the epithelial cell adhesion molecule, EpCAM, uncovered that OVOL1/2’s levels are highest in the fully epithelial state, but they steeply decrease as the mesenchymal component increases (Fig. 6E). As references, our analysis included three basic/neutral epithelial keratins: K5, K6A and K6B, which displayed patterns similar to the distribution of OVOLs. This contrasted with ZEB1, a characteristic mesenchymal marker, which displayed gradually increasing levels towards the fully mesenchymal state. Notably, in line with the results of our in vitro models and the pathology analyses, the data from animals demonstrated that the triple-negative and CD106⁺ E/M states are endowed with the highest metastatic potential (71).

OVOL-mediated epithelialization induces DNA double strand breaks, as well as alters the interactions among the kinases regulating the DNA damage response (DDR)

Stimulation of stress signaling might explain how DTCs survive post cytotoxic treatments (6, 47). For example, the stress kinases p38 mitogen activated protein kinase (p38-MAPK) and PERK (encoded by EIF2AK3) have been implicated in dormancy (12, 72). In line with these reports, re-analysis of our RNA-sequencing data revealed that OVOL1 overexpression increased the expression of transcripts corresponding to p38-delta (MAPK13) and EIF2AK3 (fig. S6E). We also observed concurrent decreased levels of mRNAs encoding ATM, a master regulator of the cellular response to DNA double-strand breaks (DSBs) (73, 74). Because ATM regulates autophagy (75) and inhibits epithelialization (76), we focused on this kinase, along with the two other members of the PIKK family, ATR and DNA-PK, which act as effectors of the DNA damage response (DDR). Notably, ATM and DNA-PK recognize DSBs, whereas ATR responds to single stranded regions. In addition, all three PIKK kinases, as well as p38-MAPK (77, 78), phosphorylate the tail of histone variant H2AX, and this phosphorylated form, γH2AX, marks DSBs (79).

To stimulate ATM, we treated the inducible derivatives of MDA-MB-231 cells with DOX and later exposed them to a chemotherapeutic agent, carboplatin. Under these conditions, we observed phosphorylation of not only ATM and H2AX but also DNA-PKcs and ATR (Fig. 7A). Notably, overexpression of OVOL1 and OVOL2 reduced ATM’s abundance and, correspondingly, diminished the amount of auto-phosphorylated ATM (pATM). Unexpectedly, overexpression of either OVOL1 or OVOL2 increased phosphorylation of H2AX also in the absence of external stress (meaning, in chemotherapy-naïve cells; Fig. 7A). This, however, did not occur in C1ORF116-overexpressing cells (fig. S6F). In addition, overexpression of OVOLs increased the phosphorylation of AKT, but, as we expected, carboplatin partly decreased this effect, in line with the reported ability of DTCs to survive cytotoxic treatments (7, 80). Together, these observations implied that OVOLs activate a cell survival pathway despite their induction of ROS and growth arrest. Along this line, our results also proposed that OVOLs can alter the well-studied kinase interplay that regulates DDR. For example, we noted that OVOL2 suppressed the expression of ATM and increased phosphorylation of DNA-PKcs, which might complement the bi-directional inhibitory interactions between DNA-PKcs and ATM (81, 82).

Figure 7: OVOL1 and OVOL2 regulate an interplay among p38-MAPK and the three major kinases involved in the DNA damage response.

(A) The indicated derivatives of MDA-MB-231 cells, which overexpress V5-tagged OVOL1 and OVOL2, or control (EV) cells, were induced for 3 days with DOX. Thereafter, cells were untreated or treated for 48 hours with carboplatin (15 μM) in the presence of DOX. Next, cell lysates were prepared and subjected to immunoblot analysis that made use of antibodies specific to V5 or the indicated proteins. Vinculin was used to ensure equal gel loading. (B) The indicated derivatives of MDA-MB-231 cells, including cells expressing an inducible allele of C1ORF116 and an empty vector (EV), were untreated or treated for 5 days with DOX. Cell extracts were later cleared and analyzed as in A. (C) MDA-MB-231 cells expressing inducible alleles of OVOL1 or OVOL2 were treated for 5 days with DOX and later incubated for 24 hours with the following inhibitors: KU-60019, an ATM inhibitor (ATMi), AZ20, an ATRi, or NU7441, a DNA-PKi. Cell extracts were analyzed as in (A). (D) MDA-MB-231 cells overexpressing inducible alleles of OVOL1 and OVOL2 (OVOL1-iOE and OVOL2-iOE, respectively), along with the EV (control) cells, were treated with doxycycline (1 μg/ml) for 72 hours and then seeded on glass slides and incubated for 72 additional hours. At the end of the incubation, cells were fixed using paraformaldehyde (PFA; 4%) and immunostained using antibodies specific to the phosphorylated form of histone H2AX (gamma-H2AX). DAPI was used to visualize nuclei. Images were obtained with a Nikon CSU W1–02 spinning disk microscope (x63 magnification). Scale bar, 20 μm. Images are representative of 5 fields captured from each of 2 independent experiments. Blots in (A to C) are representative of 2 independent experiments. The dashed lines in the blots separate lanes to facilitate comparison between the different cell lines used in A-B, and they indicate the DOX induction in C.

Overexpression of OVOL1 and OVOL2 transcriptionally inhibits ATM and catalytically activates DNA-PKcs and p38-MAPK

To better comprehend the impact of OVOL1 and OVOL2 on the DDR, we first validated that their gain of function was associated with reduced ZEB1 and vimentin, as well as with increased levels of E-cadherin (Fig. 7B). Thereafter, we undertook two approaches: (i) analyzed the effects of OVOL1/2 (in the absence of chemotherapy) on activation of four kinases: the three PIKKs and p38-MAPK (Fig. 7B), which is sensitive to ROS and can phosphorylate H2AX (73), and (ii) determined the effects of the respective kinase-specific small molecule inhibitors (Fig. 7C). The first approach validated that overexpression of OVOL1/2 can downregulate ATM and ATR, as well as revealed concomitant activation (phosphorylation) of p38, DNA-PKcs and AKT. Because previous lines of evidence established an inhibitory effect of ATM toward both DNA-PK (81–83) and p38 (84, 85), the ability of OVOL1/2 to transcriptionally suppress ATM can explain how OVOLs activate DNA-PKcs and p38. Notably, in parallel to the activation of p38, we observed a moderate decrease in pERK. These reciprocal trends are consistent with early reports, which inferred that ERK is negatively regulated by p38. Hence high p38/pERK activity ratio might herald tumor dormancy (12, 72).

Our other approach used the following kinase inhibitors: KU-60019, an ATM inhibitor (ATMi), AZ20, an ATRi and NU7441, a DNA-PKi. In general, the results we obtained (Fig. 7C) further supported the ability of OVOL1/2 to stimulate a dual kinase switch: ATM-to-DNAPK and ERK-to-p38. For example, DNA-PKcs emerged as a major kinase that phosphorylates H2AX in OVOL1/2 overexpressing cells, in line with the observed OVOL1/2-induced decreased expression of ATM and ATR. In addition, using AZ20, an ATRi, we observed activation of DNA-PK and AKT, likely due to relief of ATR-mediated inhibition of these kinases. The function of ATR, however, remained unclear for the respective inhibitor enhanced rather than reduced pATR. Thus, on the one hand, OVOL-induced epithelialization elevates ROS, which activates p38, and on the other hand, OVOLs transcriptionally reduce ATM abundance, which activates DNA-PKcs. These biochemical events seem to alter the PIKK interplay and activate a dual kinase switch at DSBs: replacement of ATM by active DNA-PK, along with activating the stress kinase, p38, as a partial substitute for pERK.

OVOL-mediated epithelialization involves the adoption of a γH2AX pattern resembling nucleotide excision repair

To validate an OVOL-induced DNA damage, we analyzed the subnuclear localization of γH2AX. Phosphorylation of this histone variant post exposure to ionizing radiation generates foci serving as DSB markers, but UV irradiation and other stressors induce pan-nuclear staining, which typifies nucleotide excision repair (NER) (86, 87). Consistent with the possibility that OVOLs not only enhance γH2AX but also alter its distribution and the identity of the upstream kinases, we noted that control MDA-MB-231 and BT-549 cells exhibited faint and rare punctate signals of γH2AX, but DOX-treated cells displayed diffuse pan-nuclear staining (Figs. 7D and S7A). It is noteworthy that pan-nuclear patterns that are caused by localized DNA damage have previously been reported (86), but their transient nature and the inferred involvement of ATM and DNA-PK suggest that the underlying mechanism differs from the one activated by OVOL1/2. One potential mechanism could relate to differential degradation of the DDR kinases, as previously reported for ATM and ATR (88). Indeed, the application of a proteasome inhibitor and an autophagy blocker raised the possibility that degradation of DNA-PKcs and p38 is controlled by autophagy, whereas ATR and AKT are sorted for proteasomal degradation (Figs. S7B and S7C).

OVOL-mediated epithelialization involves the oxidation of DNA

NER and BER (base excision repair) constantly repair DNA damage caused by ROS and additional mutagens that frequently oxidize the guanine base to produce 8-oxoguanine (which pairs with adenine). Hence, when present in DNA, 8-oxoguanine induces deleterious G > T mutations (89). Congruent with the ability of OVOLs to increase ROS levels and alter the DDR, we detected 8-oxoguanine in the cytoplasm of MDA-MB-231 cells post-treatment with DOX (Fig. S7D). Notably, nuclear γH2AX and cytoplasmic 8-oxoguanine displayed a tendency to co-localize in the same cells. In conclusion, the epithelialization of mesenchymal tumor cells involves oxidative stress and the adoption of an altered mode of the DDR. This variant involves subnuclear re-distribution of DNA repair sites, re-division of tasks among the involved protein kinases, and possibly altering the way these kinases are controlled by transcription, autophagy and the 26S proteasome.

In summary, we assumed that introducing one of the most characteristic epithelial genes, either OVOL1 or OVOL2, in highly metastatic breast cancer cells will mirror developmental epithelialization and simulate mechanisms that suppress the progression of indolent metastases. Here we report that inducible expression of OVOL1/2 in highly metastatic cells preempted their many oncogenes, suppressed their proliferation, as well as inhibited the ability of growth factors to enhance their motility. Further analyses confirmed the association of the forced epithelial phenotype with a decreased abundance of EMT genes and an enhanced abundance of E-cadherin. To unravel the underlying mechanisms, we investigated one of the uncovered OVOL target genes, C1ORF116. The respective hormone-controlled adaptor emerged as a putative autophagy receptor that physically regulates two major redox proteins, TRX and GCLC. In line with this, we found that forcing mesenchymal mammary cancer cells to undergo epithelialization is associated with lowering glutathione levels and elevating ROS. Yet another unexpected feature emerged from experiments that exposed OVOL-overexpressing cells to carboplatin. Unlike chemotherapy, which activated all three DDR kinases, OVOL overexpression downregulated ATM and ATR but activated DNA-PKcs, p38-MAPK and AKT. Alongside, OVOL overexpression increased DNA oxidation and DSBs. Taken together, by harnessing C1ORF116-induced putative autophagy and by regulating the redox potential of mesenchymal cells, OVOLs might act as estrogen-inhibitable gatekeepers that prevent exit from breast cancer dormancy while permitting accumulation of deleterious DNA alterations. In aggregate, the observations we report uncovered a hitherto unknown signaling pathway that potentially sustains epithelial phenotypes in development and pathology (Figure 8).

Figure 8: Schematic representation of the inferred OVOL1-regulated signaling pathway that potentially sustains mammary tumor dormancy and epithelial cell identity.

Both steroid hormones (such as estrogen) and growth factors (such as EGF) control the expression of OVOL1, which represses mesenchymal genes like the ZEB family genes. In parallel, OVOL1 up-regulates epithelial genes that encode E-cadherin, ER, TROP2 and HER2. Another target is the C1ORF116 gene (chromosome 1, open reading frame 116), which is androgen-regulated and functions as a putative autophagy receptor. C1ORF116 physically binds with GABARAP/LC3 and likely inhibits two major antioxidants, GCLC (glutamate-cysteine ligase catalytic subunit) and thioredoxin (TRX). Thus, C1ORF116 increases the abundance of reactive oxygen species (ROS) and stimulates the p38 Mitogen Activated Protein Kinase (p38-MAPK; also known as Stress Activated Protein Kinase, SAPK). This induces chronic stress and growth arrest, but no accompanied apoptosis, probably because AKT/protein kinase B is also activated. Concurrent with the activation of p38 and AKT in the cytoplasm, transcripts encoding ATM, a nuclear kinase, decrease in OVOL1-expressing epithelial cells. Because ATM normally represses DNA-PKcs, this closely related member of the phosphoinositide 3-kinase-related kinase (PIKK) family undergoes stimulation, whereas the third member of the PIKK family, ATR, undergoes repression. As a result, DNA oxidation is elevated and the tail of histone variant H2AX undergoes phosphorylation at double strand DNA breaks (DSBs). These biochemical events likely herald DNA damage and might explain why awakened tumors commonly display heightened genetic heterogeneity and frequently remain hormone dependent.

Discussion

An important source of insights into differentiation of the mesenchymal lineage and, by inference, the metastasis cascade, emerged from in-depth understanding of the function of EMT-inducing transcription factors, such as ZEB1 (21, 90). With the exception of p63 (91), we only poorly understand the reciprocal group, MET-TFs, and their involvement in both mammary gland development and breast cancer progression (92). Hence, to better understand epithelial lineage differentiation and breast cancer dormancy, we considered several MET-inducing transcription factors and eventually selected two members of the OVOL group (17). During the development of the skin and the mammary epithelium, OVOL1 and OVOL2 are respectively required for the arrest of committed progenitor cells (93, 94). Similarly, our results raise the possibility that OVOLs regulate the universal re-epithelialization and growth arrest that DTCs undergo post arrival in distant organs. Mechanistically, OVOL1 directly represses several hub genes, including MYC and ZEB1 (93, 95). This explains several attributes of tumor dormancy, such as growth arrest and altered metabolism, along with suppression of EMT. Of note, although we manipulated only single OVOL genes, the engineered cells displayed altered morphological features, along with epithelial markers, such as increased expression of E-cadherin and amphiregulin. In line with these alterations, when using OVOL1/2 overexpressing cells we observed complete inhibition of EGF-induced cell migration, along with partial inhibition of both colony formation (in culture) and tumor growth (in vivo). Alongside these attributes, we unexpectedly observed increased H₂O₂ levels and decreased levels of glutathione, taurine and additional antioxidants. Consistent with these links between redox and tumor dormancy, a recent report attributed the scarcity of muscle colonization by DTCs to sustained oxidative stress and high H₂O₂ levels in muscle, as opposed to lung and other organs (28). In conclusion, in line with the ability of OVOLs to inhibit antioxidants, the volatile nature of ROS produced by either tumors or their stroma identified a growth-arresting metabolic bottleneck that might inhibit exit from dormancy.

Our studies of C1ORF116, a hitherto poorly investigated transcriptional target of OVOL1/2, highlighted yet another feature shared by dormant breast cancer cells: this hormone-inducible adaptor seems to act as an autophagy receptor that supports survival of epithelialized cells under stress conditions. Notably, C1ORF116 has neither recognizable structural domains nor family members. Nevertheless, it undergoes ubiquitination and likely binds with ubiquitin, in similarity to the well-studied autophagy receptor, p62/SQSTM1 (59). In line with our results, it has been shown that autophagy and an autophagy-related gene (ATG7) critically support the survival of disseminated dormant breast cancer cells (29). Likewise, autophagy is utilized as a survival pathway by chemotherapy-treated cells entering a reversible non-proliferative phase that resembles the dormancy state (96). In addition, it has been reported that continuous exposure of osteosarcoma cells to insulin can instigate a dormancy state characterized by enhanced autophagy (97). According to prevailing models, long-term growth arrest requires the elimination of both ROS and misfolded proteins, as well as selective autophagy of mitochondria (98). Hence, the herein reported elevation of ROS by the OVOL-C1ORF116 axis was unexpected. Presumably, this might be explained by the dual role played by ROS, namely toxic byproducts of aerobic metabolism, on the one hand, and versatile signaling molecules, on the other hand (99).

Apart from identifying OVOL1 and OVOL2 as potential epithelialization promoters and dormancy gatekeepers, which are capable of arresting cell growth and inactivating several antioxidants, our results might provide glimpses into the poorly understood process that allows arrested DTCs to exit from dormancy. Several mechanisms have previously been implicated in this process. They include aging, remodeling the dormant cell’s niche, post-menopausal obesity, and increased neovascularization (6, 47, 100, 101). Our observations raise yet an additional possibility: by means of downregulating OVOLs, steroid hormones like estrogen and progesterone may abrogate the state of dormancy. Similarly important, our study raises the intriguing possibility that the OVOL-C1ORF-GCLC/TRX signaling pathway functions in conjunction with the phosphorylation of histone 2AX and an altered interplay linking p38 and the three major kinases involved in DNA repair (Fig. 8). Conceivably, breast cancer cells in which these pathways are active might slowly accumulate DNA damage and new mutations, until they break a threshold required to escape dormancy. We propose that unchecked DNA damage that occurs during mass dormancy due to elevated ROS, DNA oxidation, and DSBs might increase the genetic heterogeneity of DTCs. This model is consistent with several lines of evidence suggesting that chromosomal instability contributes to the evolution of metastatic disease (102). Furthermore, this view is supported by the results of genomic analysis of metastases in autopsies from patients with therapy-resistant breast cancer (31), in which the authors concluded that metastases evolve as communities of clones that keep accumulating new mutations, such that their recurrences are more aggressive and more chemo-resistant than the parent tumors. Future studies might refine our model, as well as identify mutational signatures that may underlie escape from the state of dormancy.

Materials and methods

Plasmids and antibodies

The human ORFeome cDNA sequences of OVOL1 (pDONR223 OVOL1-no stop codon, Clone ID: 10848), human OVOL2 (pDONR223 OVOL2-no stop codon, Clone ID: 5098) and C1ORF116 (pDONR223 C1orf116 no stop codon, Clone ID: 5377) were obtained from CCSB Human ORFeome and cloned into the pLenti6.3/TO/V5-DEST vector to obtain the V5-tagged inducible overexpression vectors. psPAX2 and pMD2.G plasmids were used to produce lentiviral particles from all vectors, along with an empty-vector control. Promoter reporter clones were obtained from GeneCopoeia. The following antibodies were used: p44/42 MAPK (Erk1/2; 137F5, Cell Signaling Technology #4695), phospho-p44/42 MAPK (Erk1/2, Thr²⁰²/Tyr²⁰⁴; Cell Signaling Technology #9101), Akt1 (C73H10, Cell Signaling Technology #2938), phospho-Akt Ser⁴⁷³ (Cell Signaling Technology #4060), caspase-3 (Cell Signaling Technology #9662), cleaved caspase-3 Asp¹⁷⁵ (Cell Signaling Technology #9661), Bim (C34C5, Cell Signaling Technology #2933), phospho-histone H2A.X Ser¹³⁹ (Cell Signaling Technology #2577), histone H2A.X (Cell Signaling Technology #2595), cyclin B1 (D5C10, Cell Signaling Technology #12231) XP^® Ki-67 (8D5, Cell Signaling Technology #9449), OVOL1 (ProteinTech #14082), OVOL2 (Invitrogen #41620), E-cadherin (24E10, Cell Signaling Technology #3195), ZEB1 (H-102, Santa Cruz Biotechnology #25388), KLF4 (GKLF/EKLF/LKLF/KLF4/1/2; F-8, Santa Cruz Biotechnology #166238), ubiquitin (P4D1, Santa Cruz Biotechnology #8017), p62/SQSTM1 (ProteinTech #18420), DNA-PKcs (3H6, Cell Signaling Technology #12311), phospho-DNA-PKcs Ser²⁰⁵⁶ (E9J4G, Cell Signaling Technology #68716), ATM (D2E2, Cell Signaling Technology #2873), p-ATM Ser¹⁹⁸¹ (D6H9, Cell Signaling Technology #5883), ATR (Cell Signaling Technology #2790), phospho-ATR Ser⁴²⁸ Cell Signaling Technology #2853),α-tubulin (Invitrogen #PA5–58711), glyceraldehyde-3-phosphate dehydrogenase (Millipore #MAB374), estrogen receptor (Invitrogen, #MA1–310), HER2 (#2165), total and phosphorylated (Ser³⁰) Beclin-1 (Cell Signaling Technology #3738 and #54101, respectively), p63 (Santa Cruz Biotechnology, #sc-62686), and vinculin (Sigma-Aldrich, #V9264). For immunofluorescence, we used antibodies to vimentin (D21H3 XP^®, Cell Signaling Technology #5741), GABARAP+GABARAPL1+GABARAPL2 (EPR4805, Abcam #109364), thioredoxin (ProteinTech #14999), GCLC (ProteinTech #12601), C1ORF116 (SARG, ProteinTech #14888), NRF2 (Abcam #137550), N-cadherin (D4R1H XP^®, Cell Signaling Technology #13116), V5 (SV5-Pk1, Invitrogen #V9264), E-cadherin (BD Transduction Laboratories #610404), β-catenin (D10A8 XP^®, Cell Signaling Technology #8480), anti-mouse Alexa Fluor 555-conjugated secondary antibody (Thermo Fisher Scientific #A-31570), anti-rabbit Alexa Fluor 488-conjugated secondary antibody (Thermo Fisher Scientific #A-11008), and fluorescent gelatin (Thermo Fisher #G13186).

Cell lines and cell cultures

MCF-10A, MDA-MB-231 and BT-549 cells were purchased from the American Type Culture Collection. MCF-10A breast cells were cultivated with Dulbecco’s Modified Eagle Medium Nutrient Mixture F-12 (DMEM/F12, Biological Industries Israel Beit Haemek) supplemented with 5% horse serum (Biological Industries Israel Beit Haemek), 1% glutamine, 10 ng/ml cholera toxin, 0.5 μg/ml hydrocortisone, 10 μg/ml insulin, 10 μg/ml hydrocortisone and 10 ng/ml human EGF. For the time-course experiments, cells were starved overnight in DMEM/F12 without serum or EGF (starvation medium). MDA-MB-231 and BT-549 cells were cultured in RPMI medium (Gibco, Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (FCS; Gibco, Thermo Fisher Scientific).

Generation of inducible overexpression cell lines

Inducible overexpression cells were prepared using the pLenti6.3/TO/V5/DEST Gateway destination vector and the pLenti6/TR repressor vector from the ViraPower HiPerform Gateway system (Thermo Fisher Scientific). Clones for the cDNA without stop codon for OVOL1 (BC059408), OVOL2 (BC006148), and C1ORF116 (BC000765) were obtained from the CCSB Human ORFeome (hORFeome) collection in Gateway donor plasmids and cloned into the destination vector using LR Clonase II Enzyme Mix (Gateway, Thermo Fisher Scientific). High-efficiency NEB stable competent E. coli bacteria (New England Biolabs) were transformed with the resulting vectors. Generated pLenti6.3 plasmids and repressor pLenti6/TR plasmid were used to produce lentiviruses using the second-generation packaging systems pMD2.G and psPAX2. Cells were transduced with equal volumes of the overexpression and repressor lentivirus produced in HEK293 cells over 48 hours. Polybrene (10 μg/ml) was used to increase the efficiency of infection. Post-transduction antibiotic selection was performed with blasticidin (5 μg/ml) and geneticin (1 mg/ml) for MDA-MB-231 cells and geneticin (0.5 mg/ml) for BT-549 cells.

Transgenic and xenograft mouse models

Animal studies were approved by the ethics committees of the Weizmann Institute and the Veterinary Medicine University of Vienna and they were carried out in accordance with guidelines for animal care and protection and protocols approved by the respective authorities. MMTV^PyMT/+ mice were described previously (44). Ovol1^{tm1a(KOMP)Wtsi} mice were purchased from MMRRC (Davis, CA, USA) and they were described previously (43). Mice were maintained under specific pathogen-free conditions and had access to water and standard rodent diet (V1534, Ssniff, Soest, Germany) ad libitum. Beta-galactosidase staining was performed as described previously (103). Briefly, the organs of Ovol1^{tm1a(KOMP)Wtsi} mice, Ovol1^{tm1a(KOMP)Wtsi} + MMTV^PyMT/+ mice, and control littermates were fixed in phosphate buffered saline (PBS) containing 0.02% NP-40, 1% formaldehyde, and 0.2% glutaraldehyde at 4 °C for 2 hours, washed twice with PBS for 20 minutes at room temperature (RT), and incubated overnight at RT under shaking in staining solution containing X-gal (1 mg/ml) in PBS. After staining, samples were washed in PBS as above, postfixed in PBS containing 4% paraformaldehyde and embedded in paraffin blocks. Histological sections were stained with eosin. For tumor growth and metastasis assays, inducible OVOL1 overexpression ZsGreen-labelled MDA-MB-231 cells were incubated with or without doxycycline (1 μg/ml) for 72 hours. For all groups, 1×10⁹ cells were injected into the mammary fat pad of female NOG mice. The indicated mice received DOX (0.5 mg/ml) in the drinking water on the day of cell injection or when a palpable tumor appeared. All groups of animals were monitored for tumor size and well-being. When the tumors approached a size of 600 mm³ the experiment was taken to an end and the mice were sacrificed. The tumors were measured and weighed, and the lungs and livers were analyzed for metastatic growths.

Immunofluorescence staining

8-chamber glass slides were coated with 50μl Cultrex^®. MDA-MB-231 cells (5×10³/well) were re-suspended in 400μl assay media. The next day, cells were treated with DOX (0.25μg/ml) or left untreated. After 48 hours, cells were treated for 5 min with paraformaldehyde (PFA: 4%) containing sucrose (5%, as fixative) in Triton X-100 (0.2%), and fixed for an additional 25 minutes with the above fixative. Next, cells were washed for 10-min with saline and additional 15-min with saline containing Tween 20 (0.05%). The cells were blocked in IF buffer (130 mM NaCl, 7mM Na₂HPO4, 3.5 mM NaH₂PO₄, 7.7 mM NaN₃, 0.1% albumin, 0.2% Triton X-100, 0.05% Tween20) containing 5% donkey serum and 5% albumin for 1hr and incubated overnight at 4°C with V5-Tag (D3H8Q) rabbit antibody diluted in the above blocking buffer (1:500; Cell Signaling Technology), for OVOL1/OVOL2 detection, and with Alexa Fluor 488 phalloidin (1:40; Molecular Probes) for F-actin detection. The cells were washed thrice with IF buffer for 15-min and incubated for 1-hour with donkey anti-rabbit antibody conjugated to Alexa Fluor 647 (1:200; Invitrogen) and mounted with VECTASHIELD mounting medium with DAPI (Vector Laboratories). For F-actin staining only, cells were stimulated with DOX for 6 days before staining was carried out (cells were re-fed after 3 days). Images were captured using a Nikon A1-R confocal laser scanning microscope. To detect NRF2, cover slips were incubated overnight at 4°C with the primary antibody (anti-NRF2; Abcam, ab137550) diluted in blocking buffer. Cells were washed thrice with saline and subsequently incubated with Alexa Fluor 488-conjugated secondary antibodies, for 60-min in the dark. Cells were washed thrice with saline and incubated for 2-min with Hoechst 33342, washed again and mounted on microscopic slides using mounting media (10 mM phosphate buffer, pH 8.0, 16.6% w/v Mowiol 4–88 and 33% glycerol). A confocal laser-scanning microscope (LSM 800; Carl Zeiss) equipped with an M27 objective lens (Plan Apochromat; Carl Zeiss) was used. Images were captured using a Zeiss Spinning disk confocal microscope and processed using the Zeiss ZEN 3.1 software.

Cell migration assays

MDA-MB-231 cells were mixed with collagen and then cultured for 48 hours. The left parts of chemotaxis chambers were filled with control media and the right sides with EGF-containing media. Live imaging was performed for 16 hours. The respective rose plots were processed using a dedicated software.

Cell viability assays

Cell viability was assessed using MTT (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide). Cells were seeded in 96-well plates. On the next day, they were treated for 72 hours with the indicated drugs. Afterwards, cells were incubated for 3 hours at 37°C with the MTT solution (0.5 mg/ml). The formazan crystals formed by metabolically active cells were dissolved in DMSO and the absorbance was determined at 570 nm.

Colony formation assays

Cells were seeded in 6-well plates at a density of 1000 cells per well. Media with were refreshed once every 3 days. Following 14 days of incubation, cells were fixed for 20 minutes in ice-cold methanol, followed by staining for 15 minutes at RT with 2% crystal violet. Full-well photos were captured using the EPSON PERFECTION 4870 Photo Scanner (Long Beach, CA, USA). For signal quantification, images corresponding to 5 non-overlapping fields were captured using a light microscope (Olympus Corporation) and quantified using Image J.

Cell proliferation assays

Cells were seeded in 96-well plates at a density of 1000 cells per well. At the indicated time points, cells were fixed in ice-cold methanol for 20 -min at room temperature, followed by staining for 15 min with 2% crystal violet. Cell growth was quantified by dissolving the cells in a detergent solution and determining light absorbance (590 nm) using a microplate reader. Proliferation assays in 3-D BME made use of Cultrex^® growth factor-reduced basement membrane extract (from R&D). MDA-MB-231 cells (2.5×10³/well) were cultured in 96-well plates pre-coated with 50μl Cultrex^®. Cells were re-suspended in 0.1 ml/well of assay media (RPMI 1640 medium containing Pen-strep, supplemented with 2% FBS and 2% Cultrex^®). The next day, cells were treated with Doxycycline (DOX, at a concentration of 0.25 μg/ml) or they were left untreated. Cells were re-fed once every 4 days with assay media with or without DOX. Cell Titer 96 AqueousOne Solution cell proliferation assay kit (Promega) was added to the wells at the indicated time points for 2 hours (33). Proliferation was assayed by recording light absorbance at 490nm using an Elisa Plate Reader (Epoch, BioTek).

Invadopodia formation and gelatin degradation assays

The assay was performed as previously described (104). Briefly, 13-mm coverslips were treated with 1 N HCl and coated with poly-L-lysine (50 μg/ml). A gelatin solution (0.2%) was prepared in saline and a 1:10 mixture of Oregon green–labelled gelatin (Thermo Fisher)/unlabeled gelatin was warmed to 37°C before addition to the poly-L-lysine–coated plates. Gelatin was crosslinked with 0.01% glutaraldehyde. MDA–MB-231 cells (40,000) were plated on gelatin Alexa Fluor 488–labeled glass coverslips, incubated overnight and fixed in 3.7% paraformaldehyde. Cells were permeabilized with 0.15% Triton X-100, blocked with 1% foetal bovine serum, 1% albumin in saline, and then labelled with rhodamine-phalloidin (Sigma #P1951). Images were acquired using an inverted fluorescence microscope (Olympus IX83 60X PlanApo 1,4 NA lens) equipped with an ORCA-Flash 4.0 V2 digital CMOS camera (Hamamatsu Photonics). Invadopodia were identified as actin-rich punctate that colocalized with gelatin degradation areas. Matrix degradation was analyzed by quantifying the mean degraded area in pixels per field using ImageJ and normalized to the number of cells per field.

Real-time PCR assays, RNA isolation and sequencing

RNA was isolated using the TRIzol reagent (Life Technologies; Thermo Fisher Scientific, Inc., Waltham MA) and cDNA was generated using the qScript cDNA synthesis kit (Quantabio, Beverly, MA). Real-time qPCR relative quantification was performed using the Fast SYBR Green Master Mix (Applied Biosystems). Primer sequences were obtained from the Harvard PrimerBank. Beta-2-microglobulin (B2M) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) were used as reference genes. Isolated RNA was tested for purity and integrity using NanoDrop quantification and TapeStation analysis, respectively. Pooled libraries were prepared for the selected samples using a modified version of the MARS-Seq protocol for bulk-RNA sequencing. Briefly, the protocol involved the barcoding of samples with oligo dT primers by reverse transcription, pooling the samples for linear amplification and preparation for sequencing. We prepared 4 biological replicates for the cells in culture with 4 consecutive passages. Each sample was sequenced with a depth of at least 5 million reads. Samples were sequenced using the Nextseq 500 platform (75 cycles; Illumina, San Diego CA).

Evolutionary conservation analysis and LC3 interacting region motif prediction

The analysis of the C1ORF116 sequence conservation was performed using the ConSurf server (https://consurf.tau.ac.il) and 73 homologous sequences from different species. The percentage of residue variety for each amino acid for each of the residues in C1ORF116 was retrieved and projected in a prediction for the 3-D structure of C1ORF116 obtained from Alphafold (https://alphafold.ebi.ac.uk/). For the identification of functional LIRs, we used the LC3 Interacting Region Motifs (https://ilir.warwick.ac.uk/index.php). The complete protein sequence corresponding to C1ORF116 (Uniprot: Q9BW04) and a PSSM score of at least 7 were used.

Gene expression and pathway analyses

Gene expression data in breast cancers was obtained from the UCSC Toil RNAseq Recompute Compendium Hub and METABRIC. We retrieved the UCSC Toil data using the R package UCSCXenaTools: R API for UCSC Xena Hubs. This compendium comprises data from The Cancer Genome Atlas (TCGA), the Genotype-Tissue Expression (GTEx), and the Therapeutically Applicable Research To Generate Effective Treatments (TARGET) dataset. Differential expression analysis was performed using the “DESeq2” R package (105). The tool EnrichR was used to perform pathway enrichment analysis. Selected genes were used for overrepresentation analysis in the gene expression analysis. Genes with a fold change of at least 1, or −1, and an adjusted p-value smaller than 0.05 were analyzed. Pathway databases used were Elsevier’s Pathway Studio, NCATS BioPlanet, the Molecular Signatures Database and Human Cyc.

Yeast two-hybrid (Y2H) screens

The screens, which were outsourced to Hybrigenics Services (www.hybrigenics-services.com/), used the coding sequence for human C1ORF116 (amino acids 36 to 1034). This was PCR-amplified from the CCSB Human ORFeome C1orf116 Clone Without Stop Codon (clone 5377) and cloned into the pB27 (N-LexA-bait-C fusion) vector. The construct was used as a bait to screen a random-primed human breast epithelial cancer cDNA library. Overall, 93 million interactions were screened, and the prey fragments of the positive clones were amplified and sequenced at their 5′ and 3′ junctions. The resulting sequences were used to identify the corresponding interacting proteins in the GenBank database (National Center for Biotechnology Information). A confidence score was attributed to each interaction and the results were analyzed using R.

Promoter reporter assays

Promoter containing plasmids were purchased from Genecopoeia (GeneCopoeia, Foster City CA). The assay system used secreted Gaussia Luciferase (GLuc) as the reporter and SEAP (secreted alkaline phosphatase) as the internal control for signal normalization. The reporter system was transfected into HEK293 cells that were treated for 48 hours with the indicated reagents. The medium was collected and processed for measurement of the luminescent signal.

Immunoblotting analyses

Protein extracts were prepared either from cell lines or from tumors that were excised from mice. Cells were washed in saline and then extracted in RIPA buffer. Proteins were separated using gel electrophoresis and transferred to nitrocellulose membranes. After blocking, membranes were incubated overnight with the indicated primary antibodies, followed by incubation with horseradish peroxidase-conjugated secondary antibodies (1 hour), and treatment with Clarity^™ Western ECL Blotting Substrates (Bio-Rad). ECL signals were detected using the ChemiDoc^™ Imaging System (Bio-Rad) and images were acquired using the ImageLab Software.

Coimmunoprecipitation assays

Transfected HEK293T cells were transiently transfected with the C1ORF116-V5 construct using JetPEI (Polyplus-transfection, Illkirch France). Twenty-four hours later, cells were harvested and extracted in RIPA buffer containing protease and phosphatase inhibitors. The extracts (0.5 mg protein) were subjected to coimmunoprecipitation using Pierce Protein A/G Magnetic Beads (Thermo Fisher Scientific) according to the manufacturer’s instructions. Antibodies against V5 and control immunoglobulin G (IgG) were used. Proteins were eluted into SDS gel loading buffer before being subjected to electrophoresis and immunoblotting. For the ubiquitin binding assays, cells were harvested, and cell extracts were incubated with ubiquitin-conjugated agarose beads (Enzo Life Sciences, Farmingdale, New York) and naked agarose beads. Electrophoresis and immunoblotting were performed on the pulldown sample.

ROS production assays

Cells were seeded in 6-well plates (70% confluency) and drugs were added and incubated for 8 hours on the following day. Hydrogen peroxide was determined using 2ʹ,7ʹ-dichlorofluorescin diacetate (DCFDA) at a final concentration of 10 mM (diluted in Krebs-Ringer phosphate buffer). Cells were incubated for 30 minutes in the dark at 37ºC and 5% CO₂. After 30 minutes, the cells were washed twice in a fresh Krebs-Ringer phosphate buffer. Finally, the cellular fluorescence signal was recorded using epifluorescence microscopy (Olympus Corporation, Tokyo, Japan) at a wavelength of 500 nm (excitation), and 580 nm emission. Signals were quantified using Image J.

Mass spectrometry

Cells were rinsed with ice-cold PBS and lysed in RIPA buffer (Thermo Fisher Scientific) supplemented with 1x complete EDTA-free protease inhibitor, 1x PhosSTOP phosphatase inhibitor (both from Roche, Basel, Switzerland), 10 mM NaF, 1 mM Na3VO4, 250 U/ml Benzonase and 10 U/mL RNase-Free DNase (Qiagen). Lysates were incubated on ice for 30 min and cleared by centrifugation. Protein concentrations were determined using the Pierce^™ BCA Protein Assay Kit (Thermo Fisher Scientific). For mass-spectromemtry analysis, protein clean-up was performed with 55 μg as input following the automated single-pot solid phase enhanced sample preparation (SP3) workflow adapted from (106). Proteins were digested for 16 hours at 37°C with Trypsin in a protease:protein ratio of 1:25. Peptides were vacuum centrifuged to dryness and stored at −20°C. Peptides from 1 μg protein were dissolved in ULC/MS grade water containing 0.1% trifluoracetic acid (TFA) and 2.5% 1,1,1,3,3,3-Hexafluoro-2-propanol (HFIP), followed by sonication for 5 minutes. Peptides were separated using liquid chromatography (Ultimate 3000, Thermo Fisher Scientific) for 100 minutes with a gradient (4 to 30%) of acetonitrile. The LC system was operated at a flow of 300 nl/min and directly coupled to a MS system (Orbitrap Exploris 480, Thermo Fisher Scientific) via electrospray ionization. MS analysis was performed using data-independent acquisition (DIA). MS1 scans were acquired at a resolution of 120 K covering the range from 350–1400 m/z. Maximum injection time was 45 ms and the automated gain control (AGC) target was set to 3e6. MS2 acquisition was performed using 47 precursor isolation windows of variable width and 1 m/z overlap that covered the range from 400–1000 m/z. Fragment spectra were acquired at a resolution of 30 K and a normalized collision energy of 28% was applied. Maximum injection time was 54 ms and the AGC target was set to 1e6. Raw data were searched against the human proteome with isoforms (downloaded from Uniprot on March 15th, 2022; 79,052 entries) using Spectronaut (v. 17, Diagenode) in directDIA+ mode. The full proteome intensities were median-normalized and Empirical Bayes Statistics for differential enrichment analysis were performed using the limma R package (v. 3.50.3, R version 4.1.3). Missing value imputation was performed using K-Nearest Neighbors (KNN).”

Statistical analysis

The GraphPad Prism (version 8.0.2) and R programs were used to perform statistical analyses. Sample numbers and other information (mean ± SEM or SD, number of replicates and specific statistical tests) are indicated in the respective figure legends. Differences were considered statistically significant if p<0.05. The Image J, Cell Profiler, R, Image Lab, ReViSP and the IncuCyte S3 software packages were used to perform data analysis.

Data and materials availability:

Proteomics data has been submitted to ProteomeExhange via MassIVE (https://massive.ucsd.edu) with the identifier PXD058378. Transcriptomics data have been submitted to Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/gds) under the code GSE283235. All materials that are not commercially available will be made available upon request to the corresponding author (Yosef.yarden@weizmmann.ac.il).