IMP3 Promotes Stem-like Properties in Triple-Negative Breast Cancer by Regulating SLUG

IMP3 expression is elevated in breast CSCs and contributes to self-renewal and tumor initiation

Analysis of a published gene expression profile²⁹ revealed that IMP3 expression is significantly higher in the tumor initiating CD44⁺CD24⁻ESA⁺ population¹ isolated from human breast tumor tissues compared to the bulk population of tumor cells (Fig. 1A). There was no significant difference in the expression of IMP1 and IMP2 (two other members of the IGF2 mRNA binding protein family) between these populations (Fig. 1A). Based on this observation and the report that IMP3 is preferentially expressed in TNBCs⁴⁸, we assessed the contribution of IMP3 to the genesis and function of the CD44⁺CD24⁻ESA⁺ population in TNBC. Depletion of IMP3 in TNBC cells (SUM1315 and MDA435) and in cells isolated from a human breast tumor decreased the frequency of CD44⁺CD24⁻ESA⁺ cells (Figs. 1B, S1B). The schematic for FACS analysis of the CD44⁺CD24⁻ESA⁺ population is presented in Fig. S1A.

Given that mammosphere culture can increase the frequency of breast CSCs¹⁸, we characterized mammosphere-derived cells for their expression of stem cell markers and frequency of CD44⁺CD24⁻ESA⁺ cells. For this purpose, we used SUM1315 cells and patient-derived xenografts (PDX) of TNBC⁴⁶, which express IMP3 (Fig. S1C). As shown in Figs. 1C, D & S1D, expression of stem cell genes (SOX2, OCT4, NANOG and ALDH1A) along with IMP3, as well as the frequency of CD44⁺CD24⁻ESA⁺ cells, are significantly elevated in mammospheres generated from these cells compared to adherent cells. Importantly, collagen-1 induced differentiation of mammosphere-derived PDX cells resulted in a decrease of IMP3 expression and in the frequency of CD44⁺CD24⁻ESA⁺ cells (Fig. S1D). This observation is strengthened by our analysis of SUM159 and T47D cells, which do not express IMP3 when grown as adherent cultures. Interestingly, IMP3 expression is induced significantly in these cells when grown as mammospheres, and collagen-1 induced differentiation of these mammospheres¹³ resulted in a dramatic loss of IMP3 expression (Fig. 1E). Moreover, IMP3 loss also decreased the ability of TNBC cells to retain the lipophilic dye PKH26, a measure of the quiescent nature of CSCs³⁴ (Fig. 1F).

Elevated expression of IMP3 in mammospheres and its reduction upon collagen induced differentiation suggests that IMP3 may regulate the mammosphere forming ability of TNBC cells. Indeed, depletion of IMP3 expression in SUM1315 cells and cells isolated from a human breast tumor decreased their ability to form mammospheres (Figs. 2A, S1E). The ability of IMP3-depleted cells to form mammospheres was restored upon transfection of IMP3-depleted SUM1315 cells with an IMP3 expressing construct that is resistant to shIMP3-1 (IMP3-mut), demonstrating the specificity of IMP3 (Fig. 2B). Moreover, IMP3 depletion specifically in the CD44⁺CD24⁻ESA⁺ population isolated from SUM1315 cells also decreased mammosphere formation (Fig. 2C). Given that self-renewal is a defining feature of stem cells, we assessed the role of IMP3 in self-renewal by quantifying the ability of mammospheres to be passaged serially. As shown in Fig. 2D, depletion of IMP3 significantly reduced the number of mammospheres with increasing passage. The ability of IMP3 to regulate self-renewal and the frequency of CD44⁺CD24⁻ESA⁺ cells suggested that it contributes to tumor initiation. Indeed, transplantation of IMP3-depleted SUM1315 cells in the mammary fat pads of non-obese diabetic (NOD)-Cg-Prkdc^scidIL2rg^tm1Wjl (NSG) mice significantly increased tumor-free survival compared to control cells (Fig. 2E, left panel). Tumor initiating ability was restored by transfecting IMP3-depleted cells with the IMP3-expressing construct that is resistant to the shRNA (IMP3-mut) (right panel).

To validate our findings, we employed another model for studying breast CSCs. Inducible Src transformation of MCF10A cells increases the frequency of CD44⁺CD24^−/low cells²¹. Recently, we reported that this CD44⁺CD24^−/low population is comprised of distinct epithelial and mesenchymal populations, which differ markedly in their tumorigenic properties¹⁷. Specifically, the mesenchymal population is highly tumorigenic and enriched for breast CSCs compared to the epithelial cells. We observed that the expression of IMP3, but not IMP1 and IMP2, is significantly higher in the mesenchymal population of CD44⁺CD24^−/low cells compared to the epithelial population and that depletion of IMP3 in the mesenchymal cells reduced their self-renewal ability (Fig. 2F & G). Taken together, our data demonstrate that IMP3 promotes self-renewal and tumor initiation.

IMP3 regulation of SOX2 underlies its contribution to mammosphere formation

To investigate the mechanism by which IMP3 contributes to mammosphere formation and tumor initiation, we focused initially on SOX2 for several reasons. Its expression is enhanced in mammospheres (Fig. 1C) and it has been implicated in the genesis of TNBC³⁷. Moreover, SOX2 expression is associated with basal-like breast cancers¹², the majority of which are triple-negative and harbor a high frequency of tumorigenic cells²⁰. These observations suggest a causal link between IMP3 and SOX2. Indeed, we found that SOX2 expression is elevated in CD44⁺CD24⁻ESA⁺ cells and in the mesenchymal population isolated from SRC-transformed MCF10A cells (Fig. 3A). Moreover, SOX2 expression increases in mammosphere culture and decreases upon differentiation of these cells (Fig. 3B). The observed similarities between IMP3 and SOX2 prompted us to investigate whether IMP3 can regulate SOX2 expression. Depletion of IMP3 in TNBC cell lines using shRNAs resulted in a decrease in SOX2 mRNA and protein expression (Fig. 3C). Similar results were observed when IMP3 was depleted using smart-pool siRNAs (Fig. S2A).

To extend our results on IMP3 and SOX2 to populations enriched for CSCs, we depleted IMP3 in mammosphere cultures of SUM1315 cells (Fig. S2B), the CD44⁺CD24⁻ESA⁺ population isolated from SUM1315 cells and the mesenchymal population of CD44⁺CD24^-/low cells isolated from SRC- transformed MCF10A cells (Fig. 3D). In all cases, diminishing IMP3 decreased SOX2 expression. The same effect was seen in tumor cells isolated from a human breast tumor tissue (Fig. S2C). Also, exogenous expression of IMP3 in luminal SKBR3 cells increased SOX2 expression (Fig. 3E). Furthermore, exogenous expression of SOX2 in IMP3-depleted SUM1315 cells rescued mammosphere formation (Fig. 3F), indicating that the ability of IMP3 to promote mammosphere formation is SOX2-dependent. Moreover, depletion of SOX2 in SUM1315 cells significantly reduced mammosphere formation compared to control cells (Fig. 3G). Importantly, transfection of IMP3-depleted SUM1315 cells with the shRNA-resistant IMP3 construct increased SOX2 expression and mammosphere formation (Fig. 3H), providing specificity for IMP3 regulation of SOX2. Interestingly, BMI1, which has been implicated in self-renewal³³, is not regulated by IMP3 (Fig. S2D). Since IMP3 was originally identified as IGF2 binding protein and that IGF2 promotes tumor formation⁴⁵, we explored the possible contribution of IGF2 to SOX2 expression. Depletion of IGF2 did not alter SOX2 mRNA expression even though IMP3 regulates the expression of IGF2 (Fig. S2E). Collectively, our data demonstrate that IMP3 contributes to mammosphere formation by regulating SOX2.

In pursuit of the mechanism by which IMP3 regulates SOX2, we performed an RNA- immunoprecipitation (RIP) assay to determine whether IMP3 can bind to SOX2 mRNA directly. As shown in Fig. 3I, we detected modest binding of IMP3 to SOX2 mRNA compared to IGF2 mRNA. Given that IMP3 usually binds to 3′ or 5′ UTR of its target mRNAs, we generated reporter constructs in which luciferase mRNA was fused with either the 3′UTR or 5′UTR of SOX2. Transfection of these constructs into control (empty vector) or IMP3-expressing (pIMP3) SUM159 cells resulted in no significant changes in luciferase activity, diminishing the possibility of a direct interaction between IMP3 protein and SOX2 mRNA (Fig. S2F).

SLUG phenocopies the functions of IMP3

SLUG (SNAI2) is a zinc-finger transcription factor, expressed preferentially in poorly differentiated breast cancers and it has been implicated in the function of breast CSCs and the initiation of basal-like breast cancer^{26, 36, 43}. Based on these observations, we hypothesized that SLUG and IMP3 function in concert. We observed elevated expression of SLUG in CD44⁺CD24⁻ESA⁺ cells isolated from human breast tumors compared to the bulk population (Fig. 4A) in the same dataset that we used to analyze IMP3²⁹. SLUG expression is also elevated in CD44⁺CD24⁻ESA⁺ cells isolated from TNBC cell lines (Fig. 4B). Moreover, SLUG expression is enhanced by mammosphere culture and repressed by collagen-I-induced differentiation of these cultures (Fig. 4C). These similarities between IMP3 and SLUG coupled with the previous report that SLUG can induce SOX2 expression in hepatocellular carcinoma cells prompted us to investigate whether SLUG can also regulate SOX2 expression in TNBC⁵⁰. In fact, we found that depletion of SLUG in TNBC and human breast tumor cells decreased SOX2 expression and mammosphere formation (Figs. 4D, S2G). SLUG depletion in SUM1315, MDA435 cells and in cells isolated from a human breast tumor, also reduced the frequency of CD44⁺CD24⁻ESA⁺ cells (Figs. 4E, S2G). Moreover, exogenous expression of SLUG in luminal SKBR3 cells increased SOX2 expression (Fig. 4F). We observed similar results with the CD44⁺CD24⁻ESA⁺ population isolated from SUM1315 cells and mesenchymal cells isolated from SRC-transformed MCF10A cells (Fig. 4G).

The parallels between IMP3 and SLUG expression and function prompted us to investigate the possibility that IMP3 regulates SLUG expression and that SLUG mediates the contribution of IMP3 to tumor initiation. In fact, IMP3 was shown recently to regulate SLUG expression by binding to its mRNA and SLUG rescue in IMP3-depleted TNBC cells restored their migration and invasive ability. These data implicate SLUG as an important functional target of IMP3⁴⁴. We confirmed this finding and demonstrated that IMP3 binds to the 5′UTR of SLUG mRNA (Fig. S3A–C & Fig. 5A). Moreover, we found that IMP3 can regulate SLUG expression in CD44⁺CD24⁻ESA⁺ cells (Fig. 5B). IMP2, however, does not appear to impact SLUG expression (Fig. 5C) and no correlation between IMP2 and SLUG was detected in a gene expression dataset (Fig. D)^{8, 9, 15}. SLUG depletion did not affect IMP3 expression, indicating that SLUG functions downstream of IMP3 (Fig. S3D & E). Collectively, our data demonstrate that IMP3 can regulate SLUG expression and that SLUG functions downstream of IMP3.

The above findings prompted us to interrogate whether SLUG mediates IMP3 regulation of SOX2 and tumor initiation. Expression of SLUG in IMP3-depleted SUM1315 cells, which express low levels of SLUG, rescued SOX2 expression (Fig. 5E) and increased the CD44⁺CD24⁻ESA⁺ population (Fig. 5F). Moreover, IMP3 expression in SKBR3 cells that had been pre-transfected with SLUG-specific shRNA, failed to induce SOX2 expression (Fig. 5G). These findings demonstrate that SLUG mediates the IMP3 regulation of SOX2 expression. This conclusion was substantiated in vivo by showing that expression of SLUG in IMP3-depleted SUM1315 cells increased their tumor initiating ability (Fig. 5H). The role of the IMP3/SLUG axis in breast CSCs is also supported by our analysis of a published gene expression profile of doxorubicin-resistant subline derived from MCF7 cells, a luminal breast cancer cell line. These drug-resistant cells are highly enriched for the tumor initiating CD44⁺CD24^−/low population and are much more tumorigenic than parental MCF7 cells⁷. Interestingly, these drug-resistant cells were found to be slow-cycling and express much higher levels of both IMP3 and SLUG compared to parental MCF7 cells (Fig. S3F). Finally, we observed that the expression of IMP3, SLUG and SOX2 correlates in a breast cancer database³⁰ (Fig. 6A).

Although most studies on SLUG in breast cancer have focused on its ability to repress transcription^{5, 19}, our data suggest that it may contribute to the activation of SOX2 transcription. We detected several E-boxes (CAnnTG)¹⁰ within the proximal region of the SOX2 promoter (Fig. 6B) and chromatin immunoprecipitation analysis detected a significant enrichment of SLUG in the region containing E-box-3 (E3) at −2093 bp. In contrast, we did not detect any SLUG binding in the E-boxes proximal or distal to E3 (Fig. 6B).

Cells and reagents

The human breast cancer cell lines SUM1315 and SUM159 were obtained from Dr. Stephen Ethier (Kramanos Institute, MI, USA). MCF10A, MDA435, MDA468, SKBR3, BT549, T47D and HEK293T cell lines were obtained from American Type Culture Collection (ATCC). SUM1315 cells were maintained in F-12 medium supplemented with 5% fetal bovine serum (FBS), insulin (5μg/mL), epidermal growth factor (10ng/mL) and 1% penicillin-streptomycin. SUM159 cells were maintained in F-12 medium supplemented with 5 % FBS, insulin (5μg/mL), hydrocortisone (1 μg/mL) and 1% penicillin-streptomycin. The MDA468, MDA435, BT549, T47D and SKBR3 cell lines were maintained in RPMI supplemented with 10% FBS and 1% penicillin-streptomycin. HEK293T cells were cultured in DMEM (high glucose) supplemented with 10% FBS, non-essential amino acids (1×, Gibco), HEPES (pH 7.4, 1 mM, Gibco), sodium pyruvate (1mM, Gibco) and 1% penicillin-streptomycin. ER-SRC- transformed MCF10A cells was provided by Dr. Kevin Struhl (Harvard Medical School) and maintained as described¹⁷.

shRNAs specific for GFP (RHS4459), IMP3 (TRCN0000074675, TRCN0000074677), SLUG (TRCN0000015389, TRCN0000015390), SOX2 (TRCN0000003252, TRCN0000003250) and IGF2 (TRCN0000062430, TRCN0000062432) were obtained from Open Biosystems (Rockford, IL, USA). Smartpool siRNA specific for IMP3 and IMP2 were purchased from Thermo Scientific (Lafayette, CO, USA). Antibody for IMP3 was purchased from DAKO (CA, USA). SLUG, SOX2 and BMI1 antibodies were purchased from Cell Signaling technology (Danvers, MA, USA). GAPDH and p84 antibodies were obtained from GenTex (CA, USA). The luciferase reporter constructs PsiCheck-2 and pLightSwitch_5UTR were obtained from Dr. Sean Ryder (UMass Medical School, Worcester, MA, USA) and SwitchGear Genomics (CA, USA), respectively. Lipofectamine-2000 and Fugene-6 were procured from Invitrogen (CA, USA) and Promega Corporation (WI, USA), respectively. Collagen-1 was purchased from BD Biosciences (San Jose, CA, USA).

Antibodies used for FACS are: CD44-PE, CD24-APC (BD Biosciences); mouse-ESA (AbCam) and anti-mouse-488 (AbCam).

IMP3, SLUG and SOX2-depleted cell lines were generated by infecting them with pLKO.1 based lentiviruses expressing the corresponding shRNAs and subsequent selection in puromycin (2μg/mL). Stable cell lines were maintained regularly in puromycin (1μg/mL). IMP3 and SLUG expression constructs were generated by inserting their cDNAs into a lentivirus vector (pCDH-CMV- MCS-EF1-copGFP) into EcoRI/NotI and EcoRI/BamHI sites respectively. Primers used are listed in Supplementary Table 1.

Human breast tumor tissue

Human breast tumor tissue was obtained in compliance with the Institutional Review Board of the University of Massachusetts Medical School. Tumor cells were isolated as described previously¹⁴. Briefly, the tissue was minced and digested for 6h with a mixture of collagenase-A (Roche, Indianapolis, IN, USA) and hyaluronidase (MP Biomedicals, Solon, OH, USA). The digested cells were spun down, washed twice with serum-free DMEM/F12 medium and resuspended cells were plated in complete medium (DMEM/F12, 10% FBS, 1% penicillin-streptomycin and 5μg/mL insulin) for 2h to deplete mammary fibroblasts. The collected organoids were dissociated into a single cell suspension by trypsinization and filtered through 40 μm filter (BD Biosciences) to remove residual clustered cells and plated for subsequent experiments.

Patient derived xenograft (PDX)

Breast tumor tissue (triple-negative) samples were obtained from breast cancer surgery patients, through Brigham and Women’s Hospital and Massachusetts General Hospital. These breast tumors samples were cut into small fragments, which were then surgically implanted into the fourth mammary fat pad of NSG mice. Tumor xenografts were expanded into multiple recipient NSG mice for several passages. Tumors were harvested and analyzed as described in the text.

Flow cytometry

To quantify CD44⁺CD24⁻ESA⁺ population, cells were harvested by trypsinization, washed with phosphate buffered saline (PBS) and incubated with the desired antibodies as follows: ESA (1:200, 30′ at 4°C); anti-mouse 488 (1:500, 30′ at 4°C); CD44-PE (1:1000) and CD24-APC (1:100) for 30′ at 4°C. Cells were washed twice with PBS following antibody incubation. For sorting CD44⁺CD24⁻ESA⁺ population, cells were incubated with the same antibodies, re-suspended in serum free DMEM-F12 (1:1) with 1mM EDTA and collected in same medium with 10% FBS. All analyses were performed using BD LSRII flow cytometer and data analysis were performed using FlowJo software.

Mammosphere, self-renewal and tumor initiation experiment

Mammospheres were cultured as described previously¹⁶. For self-renewal assays, mammospheres were collected by centrifugation, washed with PBS, trypsinized for 5 minutes and single cell suspension was prepared. This single cell suspension was counted and seeded for subsequent mammosphere formation. Mammospheres were counted by taking bright field images of 15-20 different areas per well. Differentiation was induced by culturing the mammosphere derived cells on collagen-1 coated dish with serum containing medium¹³. Trypsin was inactivated by soybean-trypsin inhibitor (1mg/mL, Sigma).

For tumor initiation experiment, cells were mixed with matrigel (1:1,v/v) and 50 μL of the cell-matrigel mix was injected into the mammary fat pads of NSG mice (Jackson Laboratory, Bar Harbor, ME) as described previously¹⁶. Tumor onset was determined by palpation and visual observation.

PKH26 labeling

For PKH26 staining, control and IMP3-depleted SUM1315 cells were stained with PKH26 Red Fluorescent Cell Linker Kit (Sigma, MO, USA). Labeled cells were maintained for three weeks and PKH26⁺ cells were quantified using FACS. 7-Aminoactinomycin-D (7-AAD) was used to discriminate dead cells. The same procedure was followed to sort PKH26⁺ population from MDA435 cells.

Biochemical assays

For immunoblotting, cell extracts were prepared using RIPA buffer containing EDTA and EGTA (Boston Bioproducts, MA, USA). Cytoplasmic and nuclear extracts were prepared using NE-PER Nuclear and Cytoplasmic Extraction Reagents (Thermo Scientific, Lafayette, CO, USA). A protease and phosphatase inhibitor cocktail was added separately (Roche Applied Biosciences, IN, USA). Protein extracts were blotted with the appropriate primary Abs and then incubated with either mouse or rabbit IgG horseradish peroxidase-conjugated secondary antibody. An ECL kit (Thermo Scientific, IL, USA) was used to develop the blots. qPCR was performed by isolating total RNA from adherent cells or cells grown as mammospheres using Trizol reagent (Invitrogen) following the manufacturer’s protocol. cDNA was synthesized using superscript-II reverse transcriptase (Invitrogen). mRNAs were quantified by real time PCR analysis (ABI Prism, Applied Biosystems) using Power Syber Green PCR master mix (Applied Biosystems). Quantification was performed using ΔΔCt method and GAPDH was used as reference gene. Primers used for qPCR are listed in Supplementary Table 2.

Transfections and reporter assays

Transient transfection of IMP3 and SLUG expression constructs was performed using Fugene-6 as the transfection agent, and total mRNA and protein extracts were prepared 48 h post-transfection. siRNAs specific to IMP3 and IMP2 (25 nM) was transfected using Dharmafect-4 (Thermo Scientific) as the transfection agent and protein extracts were prepared 72 h post-transfection. For siRNA experiments, a pool of non-targeting siRNA (Thermo Scientific) was used as control (25 mM).

To generate SOX2 and SLUG 3′/5′ UTR reporter constructs, human SOX2 or SLUG 3′ UTR and 5′ UTR were amplified using cDNA prepared from SUM1315 cells and cloned into the psiCheck-2 and pLightSwitch-5′UTR vectors, respectively. The 3′UTR was cloned into XhoI/NotI sites and the 5′ UTR was cloned into BglII/NcoI sites. Same restriction sites were used for both SLUG and SOX2 UTRs. PCR amplified UTR fragments were confirmed by restriction mapping and cloned reporter constructs were confirmed by sequence analysis. Primers used are listed in Supplementary Table 3. Parental or IMP3 expressing SUM159 cells were transfected using Fugene-6 with either the control or UTR- constructs. Firefly reporter construct was used as transfection control for both 3′ and 5′ UTR constructs which express renilla luciferase. The Relative Light Units (RLU) value was calculated as the ratio of renilla luciferase to firefly luciferase activity (normalized luciferase activity). The protocol used for transfection and measurement of luciferase activity has been described previously³¹. Luciferase activity was measured in a Beckman Coultier, DTX 880 luminometer.

Binding assays

The interaction between IMP3 protein and associated mRNAs was determined using RNA-immunoprecipitation assay (RIP) followed by qPCR analysis. The assay was performed using RiboCluster Profiler RIP-Assay kit (MBL International, MA, USA). RIP grade antibody for IMP3 was purchased from the same company. Isotype matched IgG was used as control for immunoprecipitation. The chromatin immunoprecipitation (ChIP) assay was performed using ChIP-IT Express kit (Active motif, CA, USA). Chromatin was prepared by sonication using MISONIX Sonicator-3000 (power level 2, 15 second pulse X 4). DNA fragments were amplified by qPCR. Primers are listed in Supplementary Table 4.

Statistical analysis

The data are shown as the ± S. E. p values (*) were determined using student’s t test and p ≤ 0.05 was considered as significant. For in vivo serial dilution experiment, p values were calculated using Extended Mantel-Haenszel Stratified Test of Association. For correlation studies, statistical significance was calculated by Pearson’s correlation.