Induction of the RNA Regulator LIN28A Is Required for the Growth and Pathogenesis of RESTless Breast Tumors

We recently described a highly aggressive subset of human breast tumors in which loss of RE1 silencing transcription factor (REST) is associated with decreased disease-free survival (1). During an unbiased screening for tumor suppressors, REST was identified as a factor whose loss confers anchorage-independent growth upon human mammary epithelial cells (2). We identified a group of patients with RESTless breast cancers and showed that these patients have a poor prognosis; however, the causative role of REST loss and the downstream mechanism responsible for the aggression of these tumors has not been described.

REST is a transcriptional repressor that functions by binding to the repressor element 1 (RE1) found in the regulatory regions (3–5) of its approximately 2,000 target genes (4, 6) and recruiting chromatin-modifying enzymes including histone deacetylases (HDAC) and histone methyltransferases (HMT; refs. 7–11).

REST regulates the expression of the RNA-binding protein LIN28A in embryonic stem and neural progenitor cells (12). LIN28A, which is highly expressed during normal development, maintains the self-proliferative capacity of progenitor cells (12–14). During cellular differentiation, LIN28A expression is decreased, resulting in the repression of genes involved in self-renewal and subsequent lineage commitment.

Aberrant expression of LIN28A in differentiated cells promotes transformation. Overexpression of murine Lin28a drives soft agar colony formation and tumor growth in nude mice (15). LIN28A expression also promotes the metastasis of MDA-MB-231 cells in a mouse model of breast cancer (16) and is associated with a variety of aggressive human cancers (15). However, little is known about how LIN28A is upregulated in these cancers. In this report, we show that REST directly regulates LIN28A expression in breast cancer cell lines and that REST loss upregulates LIN28A expression, which drives breast tumor growth and local invasion in vivo and increased clonogenicity and soft agar colony formation in vitro. We find that LIN28A is overexpressed in RESTless human breast tumors, supporting a clinically relevant role for the REST-LIN28A axis in breast cancer.

Detailed materials and methods are available in Supplementary Material.

All cell lines were purchased and authenticated from American Type Culture Collection and grown as previously described (1). MDA-MB-231 low-serum growth was conducted as previously described (17). REST knockdown was conducted as previously described (1). Stable knockdown of LIN28A (LIN28A^low) was achieved via lentiviral delivery of an anti-LIN28A shRNA (TRCN0000102579) in pLKO.1 vector (Open Biosystems). Plasmid 1864 (Addgene) was used as a control (LIN28A^norm); lentiviral particles were generated and used as previously described (18).

Antibodies used for Western blotting were REST (Upstate, 05-579), LIN28A (Abcam, ab46020), and β-actin (MP Biomedicals).

Chromatin immunoprecipitation (ChIP) was carried out as previously described (10); 2 μg of antibody (REST H-290, Santa Cruz Biotechnology; G9a 07-551, Upstate; rabbit IgG, Sigma-Aldrich) was added to 300 μg total protein. All washes were conducted with TSE with 500 mmol/L NaCl (10).

Total RNA was harvested from cultured cells (in biologic triplicate) or xenograft tumors using TRIzol (Invitrogen) according to the manufacturer's protocol, resuspended in sodium citrate (1 mmol/L, pH 6.4) containing RNASecure (Ambion), quantitated via NanoDrop (Thermo Scientific), and reverse-transcribed using SuperScript III (Invitrogen) according to manufacturer's instructions. Real-time PCR (RT-PCR) was carried out using SYBR Premix Ex Taq (Takara Bio Inc.).

Approximately 2 kb of the LIN28A promoter region including (+RE1) or excluding (−RE1) the REST-binding (RE1) site was amplified from HEK-293T genomic DNA and cloned into pGL3-Basic (Promega). Cells were transfected with pGL3-Promoter, pGL3-Basic, pGL3-Basic +RE1, or pGL3-Basic −RE1, and RL-TK Renilla (Promega) using calcium phosphate. Forty-eight hours after transfection, luciferase activity was measured using the Dual-Luciferase Reporter Assay System (Promega) according to manufacturer's instructions and luciferase signal was normalized to Renilla.

Paraffin-embedded cells were deparaffinized, rehydrated, and stained according to antibody manufacturer's protocol (Abcam). Antibodies used were LIN28A (Abcam ab46020) and Alexa Fluor goat anti-rabbit 594 (Invitrogen); Vectashield Mounting Medium with 4′,6-diamidino-2-phenylindole (DAPI; Vector Laboratories). Microscopic imaging was conducted with a digital camera (Spot II; Diagnostic Instruments) on a Nikon E600 Eclipse epifluorescent microscope with ×20 plan apochromatic objective and a standard TRITC filter cube. Images were acquired at an initial 36-bit tone scale and saved as 16-bit files. Minimal image manipulation was conducted using Adobe Photoshop according to the AACR guidelines.

Clonogenic assays were conducted as previously described (19).

All procedures were conducted with the approval of the University of Wisconsin-Madison (Madison, WI) School of Medicine and Public Health Institutional Animal Care and Use Committee and according to national guidelines and policies. Adult intact female athymic nude-Foxn1^nu mice (Harlan Laboratories) were used and MCF7 cells were treated as previously described (20).

Recently, we showed that REST is lost in 20% of human breast cancers, and that these RESTless tumors are highly aggressive (1). Here, we show that REST loss drives tumor aggression. We used lentiviral delivery of short hairpin RNAs (shRNA) to generate stable control (REST^norm) or REST knockdown cell lines (REST^low cells; the term RESTless refers only to human tumors). We knocked down REST in a nontransformed breast cell line (MCF10A), and in estrogen receptor–positive (ERα⁺) and triple-negative (ER⁻/PR⁻/Her2⁻) breast cancer cell lines (MCF7 and MDA-MB-231, respectively). REST knockdown was confirmed by Western blotting (Fig. 1A and B and previously shown in ref. 1).

We first tested the tumorigenicity of REST^low and REST^norm breast cells using a clonogenic assay (19), which estimates the ability of individual cells to form colonies. In all 3 cell lines tested, REST knockdown significantly increased plating efficiency (Fig. 1C–E).

Using an ERα⁺ (MCF7) and an ER-negative (ER⁻; MCF10A) cell line, we repeated the clonogenicity assay using a soft agar protocol. In both cell lines, REST knockdown significantly increased (>3-fold) the number of colonies observed (Fig. 1F and G).

We carried out xenograft experiments to test the in vivo tumorigenicity of REST^norm and REST^low MCF7 cells. Work that will be described elsewhere suggests that REST^low MCF7 cells would show enhanced growth in vivo in the absence of estrogen supplementation. REST^norm or REST^low MCF7 cells were injected s.c. into the flanks or mammary fat pads of intact female athymic Foxn1^nu mice, and tumor growth was monitored. Two hundred days after injection, 30% (8 of 28) of REST^low mammary fat pad injection sites developed tumors (1,980 mm³ = tumor volume), compared with 0% (0 of 28, 0 mm³ = tumor volume) of REST^norm injections (Fig. 2A and B). The tumor take rate was also significantly greater for REST^low versus REST^norm cells injected s.c. into the flanks of the mice, with 34.4% (11 of 32) of REST^low injection sites giving rise to tumors, compared with 12.5% (4 of 32) of REST^norm sites (Fig. 2C). The total tumor burden for flank tumors was significantly greater for REST^low than REST^norm tumors, at 4,314 mm³ and 1,017 mm³, respectively (Fig. 2D). REST knockdown therefore results in a significant increase in tumorigenicity of MCF7 cells at both orthotopic and nonorthotopic sites.

Histopathologic examination of REST^low tumors showed that they were highly anaplastic, with enlarged nuclei, prominent nucleoli, many convoluted nuclear envelopes, and a high mitotic rate (Fig. 2E). REST^low tumors contain approximately 1.5-fold more cells undergoing mitosis than REST^norm tumors; however, because of the small number of REST^norm tumors available for analysis (n = 3), this difference did not reach statistical significance (P = 0.1). About 62.5% (5 of 8) of REST^low flank tumors examined show localized invasion into adjacent muscle (Fig. 2F) and one showed lymphovascular invasion (Fig. 2G).

We hypothesized that one or more of the genes that become simultaneously derepressed upon REST knockdown (1) must contribute to RESTless breast tumor aggression. One such gene was LIN28A, given that LIN28A was upregulated in RESTless cells in vitro and in vivo, was a REST target (4, 12), and was previously implicated in breast cancer progression (15, 16), we sought to determine whether LIN28A was an effector of the RESTless phenotype. We confirmed previously published microarray data showing that LIN28A mRNA was elevated in the breast cancer cell line T47D upon REST knockdown (1) by quantitative RT-PCR (Fig. 3A). This increase in mRNA correlated with increased LIN28A protein (Fig. 3B). LIN28A mRNA and protein are similarly upregulated upon REST knockdown in MCF7 cells (Fig. 3C and D). REST has been shown to regulate LIN28 in embryonic stem and neural progenitor cells (12), and the LIN28A promoter region contains an evolutionarily conserved REST-binding site. To determine whether LIN28A is a direct target of transcriptional regulation by REST in our tumor model (MCF7 cells), we conducted ChIP with an anti-REST antibody and interrogated for the LIN28A RE1 sequence (Fig. 3E); a region of DNA that does not contain an RE1 site serves as a negative control (NC). To confirm that REST binding to the putative RE1 site was ablated by REST knockdown, this ChIP was conducted in both REST^norm and REST^low MCF7 cells. We find robust enrichment of REST at the LIN28A RE1 site in REST^norm MCF7 cells. Indeed, REST enrichment at the LIN28A RE1 is twice that at the brain-derived neurotrophic factor (BDNF) RE1, a well-characterized REST-binding site (21–23). As expected, upon REST knockdown, REST-binding enrichment is lost at both the BDNF and LIN28 RE1 sites. To show the specificity of REST binding and corepressor recruitment to the LIN28 RE1 site, we conducted ChIP using an anti-REST antibody or anti-G9a (a REST corepressor; ref. 10) antibody, and sham IgG; the ChIP DNA was analyzed via qRT-PCR using primers spaced every 2 kb from the LIN28A transcriptional start site (TSS) to 12 kb upstream, and primers 8 kb downstream of the TSS. REST and G9a enrichment is found exclusively at the LIN28A RE1 site and not elsewhere in the vicinity of the LIN28A gene (Fig. 3F).

To determine whether REST directly represses transcription by binding to the LIN28A RE1 site, we transfected a luciferase reporter containing approximately 2 kb of the LIN28A promoter region including (+RE1) or excluding (−RE1) the REST-binding site into REST^norm and REST^low MCF7 and HEK cells. As predicted, luciferase activity was decreased by greater than 50% in the presence (+RE1) than in the absence (−RE1) of the RE1 site in REST^norm cells; this repression was lost in REST^low cells (Fig. 3H and I). This repression therefore requires both REST and the REST-binding site. These data indicate that REST directly represses transcription from the LIN28A promoter region by binding to the RE1 site.

To test whether the regulation of LIN28A by REST observed in vitro was also present in vivo, we measured LIN28A RNA levels in REST^norm and REST^low MCF7 tumor tissue. Both LIN28A mRNA, as measured by 2 different primer pairs (pair 1 and pair 2) and LIN28A immature RNA (pre-mRNA), measured using primers in the second intron, are significantly increased in REST^low versus REST^norm MCF7 tumors (Fig. 3J). None of the housekeeping genes analyzed shows significant differences in expression between REST^norm and REST^low tumors.

To evaluate the contribution of LIN28A to the REST^low phenotype, REST^norm and REST^low MCF7 cells were transduced with a lentiviral construct expressing an anti-LIN28A shRNA (LIN28A^low) or control shRNA (LIN28A^norm), and LIN28A knockdown was verified (Fig. 4A). The anti-LIN28A shRNA used does not knockdown LIN28B (data not shown). We then conducted a soft agar colony formation assay and found that the increase in colony formation observed upon REST knockdown is ablated by concurrent knockdown of LIN28A (Fig. 4B).

We then examined the effects of REST and LIN28A on the growth rate of estrogen receptor–negative MDA-MB-231 cells. When cells were grown in high serum (10% FBS), there was no significant difference in growth rate between REST^low and REST^norm cells; however, LIN28A knockdown significantly retarded the growth of REST^low but not REST^norm cells (Fig. 4C), suggesting that the REST^low cells have acquired a dependence upon LIN28A that is not present in REST^norm cells. Because MDA-MB-231 cells proliferate at near-maximal rate in the presence of 10% FBS, we then conducted the growth assay using low-serum (2.5%) conditions previously shown to unmask growth phenotypes in MDA-MB-231 cells (17). Under low-serum conditions, REST^low cells show increased proliferation relative to REST^norm cells but lose their growth advantage upon LIN28A knockdown (Fig. 4D).

To determine whether LIN28A overexpression is a functional effector of the REST^low phenotypes, LIN28A was stably overexpressed in MCF7 cells (Fig. 4E). LIN28A overexpression caused approximately a 2× increase in plating efficiency and colonies grown in soft agar (Fig. 4F and G). We therefore conclude that in MCF7 cells, LIN28A overexpression is sufficient to induce the cell culture phenotype of REST^low cells.

To determine whether the upregulation of LIN28A observed in REST^low cells contributes to the tumorigenicity observed in vivo, we compared the tumorigenicity of REST^low cells with and without LIN28A expression. REST^lowLIN28A^norm or REST^lowLIN28A^low MCF7 cells were injected s.c. into the mammary fat pads of athymic nude mice as in Fig. 1. After 100 days, 50% (6 of 12) of REST^lowLIN28A^norm mammary fat pad injections had given rise to tumors, compared with only 8.3% (1 of 12) of fat pads injected with REST^lowLIN28A^low cells (Fig. 5A). The tumor burden in the mammary fat pads was also significantly decreased when LIN28A was knocked down, with a total tumor volume of 345 mm³ for REST^lowLIN28A^norm compared with only 56 mm³ for REST^lowLIN28A^low tumors (Fig. 5B).

Given the dependence of tumor growth on LIN28A in REST^low cells in xenograft assays, we assessed LIN28A levels in human RESTless tumors. We analyzed publicly available cDNA microarray data from 2 breast tumor cohorts, GSE2034 (24, 25) and GSE2990 (26) and classified ER⁺ tumors as RESTless or REST-expressing (RESTfl) using the 24-gene signature method as previously described (ref. 1; Supplementary Fig. S1A and S1B). In these 2 independent breast tumor cohorts, LIN28A expression was significantly higher in RESTless versus RESTfl tumors (Fig. 6A and B). The list of tumors in each data set defined as RESTless or RESTfl is provided in Supplementary Table S1. Our results strongly argue that LIN28A overexpression contributes to the previously described aggression of RESTless tumors (1).

We show that REST is a tumor suppressor; the loss of which promotes tumor growth in xenografts and increases clonogenicity, soft agar colony formation, and growth rate in vitro. We find that REST is a direct transcriptional repressor of the tumor promoter LIN28A in breast epithelial cells, and we present in vitro and in vivo data showing that LIN28A expression is required for the increased tumorigenicity of cells lacking REST. Finally, we show that LIN28A expression is heightened in human RESTless breast tumors, highlighting the clinical relevance of the REST-LIN28A axis.

REST was first identified as a tumor suppressor in a screen for genes whose loss confers anchorage independence upon human mammary epithelial cells (2). In 2009, Reddy and colleagues (27) examined 19 normal or cancerous breast biopsy samples and claimed a statistically significant, inverse correlation between tumor grade and REST expression. However, in examining publicly available microarray data from approximately 1,000 breast tumor samples, our group was unable to detect any significant decrease in REST mRNA in tumor samples compared with healthy tissue and in fact found that REST mRNA was significantly increased in tumors relative to normal tissue in multiple independent data sets (1). We further saw no significant difference in REST mRNA level across tumor stage or grade. The reason for this discrepancy is unclear and may be due to the small sample size used by Reddy and colleagues.

Lv and colleagues (28) used immunohistochemical staining for REST in normal and breast tumor tissues and found that REST was detectable in all normal and benign samples tested (n = 10 and 22, respectively) but was absent from 30.9% (21 of 68) of tumor samples. Consistent with their findings, when we examined REST expression in primary tumors via immunohistochemistry, we found that REST staining was absent in approximately 20% (37 of 182) of tumor samples. We further found that patients with RESTless tumors had a significantly poorer outcome than patients whose tumors stained positive for REST protein (1).

Here, we show that REST loss is not only a marker but a driver of tumor aggression. Using REST knockdown breast cell lines, we examined the effects of REST loss in in vitro tumorigenicity assays. We found that loss of REST increases clonogenicity in MCF7, MCF10A, and MDA-MB-231 cell lines (Fig. 1C–E). We confirmed the transformed phenotype of REST^low cells in a soft agar colony formation assay, which measures anchorage-independent growth. Because REST loss is observed in ER⁺ and ER⁻ negative human tumors, we tested the effect of REST loss on both MCF7 (ER⁺) and MCF10A (ER⁻) cells in this assay. In both cell lines, REST knockdown significantly increased the number of colonies formed in soft agar, a hallmark of transformation (Fig. 1F and G). This is consistent with previous data showing that REST knockdown promotes the growth of MCF7 cells in soft agar (28).

We next confirmed the transformative capacity of REST loss in vivo; this is the first study to examine the effects of REST loss on tumorigenicity in a xenograft model. Xenograft experiments using MCF7 cells are often conducted in estrogen-supplemented mice, as estrogen supplementation of the host mice increases the tumor take rate of MCF7 cells (29). Consistent with previous reports, REST^norm MCF7 cells generate tumors at low frequency in either mammary fat pads or flanks in the absence of estrogen supplementation. However, REST^low MCF7 cells gave rise to tumors in both orthotopic and nonorthotopic sites in the absence of estrogen supplementation, further showing the enhanced tumorigencity of REST^low cells. This finding suggests that REST loss could confer a growth advantage under conditions of low estrogen availability. REST loss could be one mechanism by which tumors escape the growth restriction normally conferred by low estrogen availability.

We previously found that RESTless tumors show increased lymph node metastasis relative to their REST-containing counterparts (1). Consistent with this, REST^low MCF7 tumors in mice show local invasion into adjacent muscle tissue, as shown by the presence of muscle fibers within the tumors (Fig. 2F). REST^low tumors also show lymphovascular infiltration (Fig. 2G), consistent with their human counterparts showing increased lymph node metastasis.

Given the established role of LIN28 in cancers, including breast cancer, and our findings that LIN28A is regulated by REST in vitro and is elevated in RESTless tumors, we hypothesized that LIN28A contributes to the aggressive phenotype of RESTless breast cancer and find support for our hypothesis both in vitro and in vivo. REST knockdown cells show enhanced soft agar colony formation in vitro, and this phenotype is dependent upon aberrant LIN28A expression (Fig. 4B). Ectopic LIN28A expression in MCF7 cells increases growth in clonogenicity assays, thus recapitulating the REST^low phenotype (Fig. 4E and F). In MDA-MB-231 cells, REST knockdown has no effect on cellular growth rate, but concomitant knockdown of LIN28A severely retards the growth of REST^low, but not REST^norm, cells (Fig. 4C), suggesting a synthetic lethal interaction between REST and LIN28A. Exploring this interaction further exposed a LIN28A-dependent growth phenotype in low-serum conditions (Fig. 4D).

We further find that LIN28A expression contributes to the tumorigenicity of REST knockdown cells. The ability of REST^low cells to form tumors in mice is significantly diminished by knockdown of LIN28A (Fig. 5). Importantly, we found that human RESTless breast tumors also have higher levels of LIN28A mRNA (Fig. 6), supporting a clinically relevant role for the REST-LIN28A axis. Given the critical role of LIN28A in mouse models of breast cancer metastasis (16), the higher levels of lymph node metastasis in RESTless breast cancer (1) and the aberrant expression of LIN28A in other aggressive cancers, we propose that LIN28A contributes to the aggressive nature of RESTless breast cancer.

Recently, it has been shown that LIN28A is a tumor promoter. However, little is known about how it becomes expressed in tumors. Myc has been shown to induce LIN28A expression by direct binding to an E-box proximal to the LIN28A promoter (16). Here, we describe the upregulation of LIN28A upon loss of the transcriptional repressor REST.

LIN28A has a REST-binding site approximately 2 kb upstream of the promoter and we chromatin immunoprecipitated REST at this site in MCF7 cells (Fig. 3C and D), consistent with the findings of a genome-wide ChIPSeq screen that found robust binding of REST to the LIN28A RE1 in Jurkat T cells (4). Reporter assays confirmed the necessity of the RE1 site for REST-mediated regulation of LIN28A in mammary epithelial cells (Fig. 3H and I). REST has previously been shown to regulate LIN28A expression in embryonic stem and neural progenitor cells (12, 30), but this is the first report to show regulation of LIN28A by REST in epithelial cancer cells.

The best-characterized mechanism by which LIN28A promotes tumor progression and metastasis is by blocking the maturation of the let-7 family of tumor-suppressing miRNAs (13, 14, 31–33), which in turn reduce the expression of multiple breast cancer oncogenes. However, we did not observe a change in the expression of the let-7 family members let-7a, let-7b, let-7e, let-7g, let-7i, and miR-98 upon REST knockdown in MCF7 cells by either miRNA array or quantitative PCR. We saw no change in let-7 expression (as measured by quantitative PCR) in MDA-MB-231 or T47D cells upon REST knockdown, nor did we observe a change in let-7 expression in REST^low versus REST^norm xenograft tumors (via quantitative PCR), despite measuring a robust increase in LIN28A expression upon REST knockdown in all cell lines and tumors (data not shown). These findings suggest a let-7 processing-independent role for LIN28A in RESTless tumor growth. It is well established that LIN28A is not limited to regulating the maturation of miRNAs. During gliogenesis, LIN28A alters cell fate independently of let-7 (34) and the first reports of the molecular mechanism of LIN28A focused on its function as an mRNA-binding protein. LIN28A binds to RNA helicase A (RHA; ref. 35) and regulates the translation and stability of mRNAs including Oct4 (35–37) and several cell-cycle regulators (38). Our observation that LIN28A mediates tumor aggression in a let-7–independent manner in our model suggests the need for further investigation of the mechanistic role of LIN28A in cancers.

LIN28B, which is encoded by a separate gene but is functionally similar to LIN28A, is also overexpressed in advanced human malignancies and transforms cells in a manner similar to LIN28A (15). LIN28B is a REST target in other systems (4). The shRNA used to knock down LIN28A expression in our studies does not target LIN28B (data not shown) and our data show that the focus formation, soft agar colony formation, and xenograft tumor growth of REST^low cells are LIN28A-dependent, and that ectopic LIN28A expression is sufficient to recapitulate the REST^low phenotypes observed in vitro. Thus, although our data show that LIN28A is necessary for our REST^low in vivo phenotypes and is sufficient for a subset of REST^low in vitro phenotypes, how LIN28B integrates into this mechanism is not yet clear.

The work presented here shows that REST is a breast cancer tumor suppressor that works, at least in part, by suppressing aberrant expression of LIN28A.

No potential conflicts of interest were disclosed.

Conception and design: K.T.W. Gunsalus, M.P. Wagoner, C.M. Alexander, A. Roopra

Development of methodology: K.T.W. Gunsalus, M.P. Wagoner, B. Schoenike, S. Kim, C.M. Alexander, A. Roopra

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.T.W. Gunsalus, M.P. Wagoner, K. Meyer, W.B. Potter, B. Schoenike, S. Kim, C.M. Alexander, A. Friedl, A. Roopra

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.T.W. Gunsalus, M.P. Wagoner, K. Meyer, W.B. Potter, A. Friedl, A. Roopra

Writing, review, and/or revision of the manuscript: K.T.W. Gunsalus, K. Meyer, C.M. Alexander, A. Friedl, A. Roopra

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): W.B. Potter, B. Schoenike, A. Roopra

Study supervision: A. Roopra

This work is dedicated in the memory of Shawn Bates who fought breast cancer and was an inspiration for this work.

The authors thank John Newton for funding and support. This work was supported by grants NIH RO1 NS065067 (A. Roopra) and NIH RO1 CA107012 (A. Friedl), Era of Hope Scholars Award to C.M. Alexander (DOD CDMRP-W81XWH), and Dual Sports Riders of Wisconsin gift to A. Roopra.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.