Identification of the receptor tyrosine kinase AXL in breast cancer as a target for the human miR-34a microRNA

Cell lines

Breast cancer cell lines MCF7 (ER⁺/PR⁺/HER-2⁻), SK-BR-3 (ER⁻/PR⁻/HER-2⁺), MDA-MB-231 (ER⁻/PR⁻/HER-2⁻), BT549 (ER⁻/PR⁻/HER-2⁻), T47D (ER⁺/PR⁺/HER-2⁻), and Hs578T (ER⁻/PR⁻/HER-2⁻); colorectal cell lines SW480 and SW620; and the renal cell line SN12C were maintained in RPMI 1640 basal medium supplemented with 10% v/v heat-inactivated fetal bovine serum (Invitrogen; Carlsbad, CA). The immortal non-tumorigenic breast cell line MCF10A (ER⁻/PR⁻/HER-2⁻) was maintained in DMEM/F12 basal medium supplemented with 5% v/v horse serum (Invitrogen; Carlsbad, CA), 20 ng/ml final epidermal growth factor, 0.5 μg/ml final hydrocortisone, 100 ng/ml final cholera toxin, and 10 μg/ml final insulin (all from Sigma-Aldrich, St. Louis, MO). Antibiotics were excluded from all the media. Cells were grown at 37°C with 5% CO₂ and passaged every 3–4 days for up to 10 generations. MCF10A and MDA-MB-231 cells were obtained from ATCC (Manassas, VA; 2008). MCF7, SK-BR-3, SN12C, and HeLa cells were obtained from the NCI Developmental Therapeutics Program, NCI, and NIH (Bethesda, MD; 2008). Hs578T, BT549, and T47D cells were gifted by Dr. Stan Lipkowitz (NCI, NIH, Bethesda, MD; 2008). SW480 and SW620 cells were gifted by Dr. Thomas Reid (NCI, NIH, Bethesda, MD; 2009). All cell lines tested negative for mycoplasma contamination using the MycoAlert® Mycoplasma Detection Kit with a Lucetta™ Luminometer (Lonza Rockland, Inc.; Rockland, ME).

Tissue collection

Fresh-frozen tumor specimens were obtained at time of surgery from breast cancer patients, as described previously [28]. Patients were recruited, who signed letters of consent at hospitals in Baltimore, Maryland, USA, between 1993 and 2003. Clinical and pathological information was obtained from medical records and pathology reports. Disease staging was performed according to the tumor-node-metastasis (TNM) system of the American Joint Committee on Cancer/the Union Internationale Contre le Cancer (AJCC/UICC). The Nottingham system was used for determining the tumor grade. The protocol for collection of tumor specimens, survey data, and clinical as well as pathological information was reviewed and approved by the University of Maryland Institutional Review Board (UMD protocol #0298229). IRB approval of this protocol was then obtained for all participating institutions (Veterans Affairs Medical Center, Union Memorial Hospital, Mercy Medical Center, and Sinai Hospital). The research was also reviewed and approved by the NIH Office of Human Subjects Research.

Immunohistochemical characterization of basal-like breast cancer

Basal-like breast tumors were identified by immunohisto-chemistry for estrogen receptor (ER), HER-2 receptor, EGFR receptor, and cytokeratin 5/6 using the following antibodies: ready-to-use monoclonal (Clone 6F11) antibody (Ventana Medical Systems; Tucson, AZ) for estrogen receptor; 1:100 diluted rabbit polyclonal antibody (Dako North America, Inc.; Carpinteria, CA) for HER-2; ready-to-use mouse antibody for EGFR; and 1:250 diluted mouse monoclonal antibody for cytokeratin 5 (Lab Vision Corporation; Fremont, CA). Basal-like breast tumors were ER negative, HER-2 negative, and either EGFR positive or cytokeratin 5/6 positive, as described previously [29]. Aberrant nuclear p53 expression was determined using a 1:100 diluted monoclonal DO-7 antibody (Dako North America, Inc.; Carpinteria, CA). A tumor was scored p53-positive if more than 10% of the cancer cells showed nuclear positivity for p53 protein expression. Tumors were screened for p53 mutations as previously described [30].

MicroRNA expression profiling

The small RNA (sRNA) fractions from four separate passages (biological replicates) of MCF7, SK-BR-3, MDA-MB-231, and MCF10A cells were extracted using Qiagen's miRNeasy kit following the manufacturer's instructions (Qiagen; Germantown, MD). Five micrograms of sRNA was labeled with either Cy3 (MCF10A) or Cy5 (breast cancer cell) fluorescent label (GE Healthcare; Piscataway, NJ) using the mirVana miRNA labeling kit (Ambion; Austin, TX). The labeled products were hybridized to miRVana miRNA Bioarrays V2 (Ambion, Austin, TX) and washed following the manufacturer's protocol. Each array was composed of 328 human miRNA probes in addition to 114 mouse and 46 rat miRNAs. The comparative hybridization for each breast cancer cell line to the reference MCF10A cell line control was conducted in quadruplicate (n = 12 total arrays). Processed arrays were scanned for dual channel hybridization using a GenePix 4000B scanner (Molecular Devices; Sunnyvale, CA). Analyses were performed using BRB-Array Tools Version 3.8.1 (http://linus.nci.nih.gov/BRB-ArrayTools.html). Replicate spots for each miRNA were averaged within the array, background adjusted, and log₂ normalized. Normalization was performed using a per chip median normalization method. Only those miRNAs with at least a threefold change in expression from its median value in 4 of 12 total arrays were retained. Samples and genes were subjected to a hierarchical cluster analysis with average linkage under the Euclidian distance similarity metric using the Cluster 3.0 (http://rana.stanford.edu/software) and Java Tree View 1.1.4r3 [31] applications. MiRNA probes of mouse, rat, and the manufacturer's (Ambion) origins that passed the gene filter were excluded from future analyses. Microarray expression data are available at the Gene Expression Omnibus (GEO) (accession GSE21834) and are in accordance with MIAME guidelines.

MicroRNA-target predictions

In a PAM analysis of global gene expression array data conducted on 51 breast cancer cell lines, 305 genes were identified as classifiers for the luminal, basal A, and basal B subtypes based on a threshold scaling of 2.8 [6]. From this list, 20 genes (the top 10 highest and lowest scores for the basal B subtype, excluding multiple probes for the same gene) were used in three separate miRNA target prediction searches (TargetScanHuman 5.1, http://www.targetscan.org/; Microcosm Targets 5, http://www.ebi.ac.uk/enright-srv/microcosm/htdocs/targets/v5/; and MiRanda 2008, http://www.microrna.org/microrna/home.do) to identify putative miRNA regulators from the list of 34 differentially expressed miRNAs observed in our profiling experiment. For MDA-MB-231 cells, only those predicted miRNAs that matched with the 20 genes from the classifier list and exhibited an inverse relationship in expression to the classifier genes were considered.

Northern hybridization

Total RNA (10 μg) was resolved on a 15% TBE-Urea gel (1.0-mm thickness) (Invitrogen; Carlsbad, CA), electro blotted onto a positively charged nylon membrane (Roche Diagnostics, Indianapolis, IN), and UV cross linked at 120,000 μJoules using a UV Stratalinker 1800 (Agilent Technologies; Santa Clara, CA). Following prehybridization in DIG Easy Hyb Buffer (Roche Diagnostics; Indianapolis, IN) for 15 min at 56°C, human miR-34a was probed (56°C for 16 h) with a complementary 3′-DIG labeled miRCURY LNA™ detection probe ACAACCA GCTAAGACACTGCCA (Exiqon; Woburn, MA) (2nM final concentration of probe in 20 ml DIG Easy Hyb Buffer). The membrane was washed using a DIG Wash and Block Buffer (Roche Diagnostics; Indianapolis, IN) and RNA was detected by CSPD chemiluminescence on film following the manufacturer's protocol for the DIG Luminescent Detection Kit (Roche Diagnostics; Indianapolis, IN). The blot was stripped at 37°C in 0.2 M NaOH solution containing 0.1% SDS, washed with 10X SSC, and reprobed (2nM final probe concentration, 50°C hybridization for 16 h) with a complementary 3′-DIG labeled miRCURY LNA™ detection probe for human U6 (CACGAA TTTGCGTGTCATCCTT) (Exiqon; Woburn, MA).

Quantitative real-time PCR

Total RNA was prepared from cell lines using the miRN-easy extraction kit (Qiagen; Germantown, MD). Single-stranded cDNA was synthesized from 100 ng total RNA using the TaqMan® MicroRNA Reverse Transcription kit (Applied Biosystems; Foster City, CA). QRT-PCR was performed on an ABI 7900HT Sequence Detection System using a human miR-34a microRNA TaqMan® Expression Assay and TaqMan® Universal PCR Master Mix (Applied Biosystems; Foster City, CA). The small nucleolar RNA molecule RNU48 was used as an endogenous control for normalization, as it displayed the lowest variance in threshold cycle times (C_t) compared to seven other controls tested across all 10 cell lines. Relative miR-34a expression was determined by the comparative C_T method. Three biological replicates for each individual cell line were tested and PCR amplifications were performed in triplicate for each sample. The threshold cycle times were averaged across all technical and biological replicates within a given cell line. Fold difference in miR-34a expression for each cell line relative to the MCF10A cell line was calculated using the equation 2^−ΔΔCt. Error bars for the fold-difference calculations represent the range in ΔΔC_t values based on the incorporation of the standard deviations for this value.

For the analysis of patient tumor samples, total RNA from 26 histologically characterized triple-negative breast cancer samples were used in measuring miR-34a and AXL expression levels. In brief, total RNA from fresh-frozen breast tissue sample biopsies was isolated using TRIzol reagent according to the manufacturer's instructions (Invitrogen; Carlsbad, CA). The concentration of each RNA sample was measured using a NanoDrop spectrophotometer (NanoDrop Technologies; Wilmington, DE), and the RNA integrity was confirmed with an Agilent 2100 Bioanalyzer (Agilent Technologies; Palo Alto, CA). AXL expression was measured from cDNA synthesized from 1 μg total RNA using the High Capacity RNA-to-cDNA kit, Gene Expression Primer Assays, and TaqMan® Gene Expression Master Mix (Applied Biosystems; Foster City, CA). ACTB mRNA expression was used for normalization. Both miR-34a and RNU48 were measured from each sample as previously described for cell lines.

Construction of luciferase expression vectors

Complementary oligonucleotides (48 bp) spanning the putative miR-34a-binding site in the 3′UTR of AXL were synthesized with XhoI compatible ends. Wild type (WT) and mutant sequences representing changes in the seed sequence-binding region of the AXL 3′UTR at positions 2 (M2), 3 (M3), 2 and 4 (M2/4), or 3 and 5 (M3/5) relative to the 5′ end of the miRNA are listed in Online Resource Table 1. Following heat denaturation in HEPES buffer, the synthetic forward and reverse oligonucleotides were annealed at 37°C. The annealed inserts were phosphorylated at 37°C using T4 DNA Kinase and ATP. The pMIR-GLO Dual-Luciferase miRNA Target Expression Vector (Promega; Madison, WI) was prepared for ligation by XhoI digestion followed by BAP treatment of the ends. A threefold excess of insert to vector was used in the ligation reactions with T4 DNA Ligase at 15°C. Recombinant vector plus inserts were transfected into TOP10 cells (Invitrogen; Carlsbad, CA) and plated onto ampicillin selective LB agar plates. Positive clones containing inserts were identified as PCR products using pMIR-GLO vector-specific primers (Forward: CGCAAGATCCGCGAGA TTCT and Reverse: TAGTTATTGCTCAGCGGTGG). PCR positive clones were then subjected to sequencing to verify the integrity of the insert. Before the luciferase assays, clones were grown and plasmid DNA was purified using a Qiagen Plasmid Mini Kit (Qiagen; Germantown, MD) followed by endotoxin removal with MiraCLEAN® (Mirus Bio LLC; Madison, WI).

The miTarget™ microRNA 3′ UTR target sequence expression clone vector containing the full-length 3′ UTR of AXL insert (GeneCopoeia; Rockville, MD) and a control vector minus insert were transfected into One Shot® Stbl3™ cells (Invitrogen; Carlsbad, CA) and plated onto kanamycin selective LB agar plates. Plasmid DNA from positive clones was purified using a QIAfilter™ Plasmid Midi Kit with endotoxin removal (Qiagen; Germantown, MD).

Luciferase reporter assay

For assays involving the full-length AXL 3′ UTR, MDA-MB-231 cells (10,000 cells per well) were seeded in a white 96-well CulturPlate™ (PerkinElmer; Waltham, MA) and incubated for 24 h before transfection. The expression vectors (1 μg) were co-transfected with AllStar Negative Control siRNA (siNEG), hsa-miR-34a miScript mimic (UGGCAGUGUCUUAGCUGGUUGU) (34a-mimic), or hsa-miR-200c miScript mimic (UAAUACUGCCGGGU AAUGAUGGA) (200c-mimic) (Qiagen; Germantown, MD) at a 10 nM final concentration using Lipofectamine™ 2000 (Invitrogen; Carlsbad, CA). Both firefly and Renilla luciferase were measured 24-h post-transfection on a FLUOstar OPTIMA microplate reader (BMG Labtech; Cary, NC) using the Luc-Pair™ miR Luciferase Assay system (GeneCopoeia; Rockville, MD). For the wild-type and mutant seed sequence expression constructs (100 ng), both fast kinetic firefly and Renilla luciferase expression was measured 48-h post-transfection in MDA-MB-231 cells (25,000 cells per well) using the Dual-Luciferase Reporter Assay System (Promega; Madison, WI). The mean ± SEM is reported for each transfection condition (n = 3). Significance was determined by a two-tailed unpaired t test for means (siNEG vs. mimic or mimic for WT vs. mimic for mutant construct).

Branched DNA capture assay

MDA-MB-231 or Hs578T cells (20,000 cells per well; 96-well plate) were reverse transfected with siNEG, siRNA AXL (siAXL) (TCCAAGATTCTAATGATTAA) (Qiagen; Germantown, MD), or 34a-mimic at a 10 nM final concentration using Lipofectamine™ 2000. Relative AXL mRNA levels in cell lysates were determined 48-h post-transfection using an AXL-specific probeset with the Quantigene^® branched DNA reagent (1.0) and capture assay system (Panomics; Fremont, CA) (see Martin et al. for assay details [32]). All transfection experiments were performed in triplicate with a total of 36 measurements per transfection condition. AXL mRNA levels were normalized to corresponding PPIB mRNA levels measured from the same lysate followed by averaging of replicates across experiments. Significance was determined by a two-tailed unpaired t test for means (siNEG treated samples were used as the control for all comparisons).

Western blots

MDA-MB-231 or Hs578T cells (600,000 cells per well; 6-well plate) were reverse transfected with siNEG, siAXL, or 34a-mimic at a 10 nM final concentration using Lipofectamine™ 2000. Transfected cells were washed in PBS and directly lysed within the well by the addition of Cell Extraction Buffer (Invitrogen; Carlsbad, CA) containing 1X Halt™ Protease and Phosphatase Inhibitor Cocktail (Thermo Scientific; Rockford, IL) followed by gentle scraping. After a 30-min incubation on ice, protein lysates were centrifuged at 13,000 rpm for 10 min at 4°C to remove cellular debris. All the protein lysates were harvested 48-h post-transfection. Total protein concentration was measured by the BCA™ protein assay (Thermo Scientific; Rockford, IL). Total cell extracts (5 μg for AXL blots or 10 μg for all other protein blots) were reduced and resolved on a 4–12% Tris–glycine gradient gel (Invitrogen; Carlsbad, CA). Proteins were electroblotted (Tris-glycine transfer buffer + 10% v/v methanol; 3 h, 25 volts constant) to an Invitrolon™ PVDF membrane (Invitrogen; Carlsbad, CA) and blocked with 5% w/v milk in 1 × TBS + 0.1% v/v Tween-20. For immunodetection, membranes were incubated at 4°C overnight with primary antibodies specific for AXL, phospho-p44/42 ERK1/2 (Thr202/Tyr204), p44/42 ERK1/2, phospho-AKT (Ser473), phospho-AKT (Thr308), AKT (pan), and β-actin (Cell Signaling Technology; Danvers, MA). Beta-actin expression was examined as a loading control. Following washes with 1 × TBS + 0.1% v/v Tween-20, membranes were incubated 1 h with either HRP-linked anti-rabbit or anti-mouse secondary antibody (Cell Signaling Technology; Danvers, MA). Chemiluminescence was detected on film using SuperSignal^® West Pico Chemiluminescent Substrate (Thermo Scientific; Rockford, IL). Membranes were stripped with Restore™ Plus Western Blot Stripping Buffer (Thermo Scientific; Rockford, IL) in a 30-min incubation at room temperature. The blots were re-blocked in 5% w/v milk, 1 × TBS + 0.1% v/v Tween-20 for 1 h and re-probed. The signal density for each protein was normalized to the respective β-actin band in that sample and expressed as a fraction of the siNEG transfected sample on the blot using ImageJ 1.43 s (http://rsb.info.nih.gov/ij/) software [33].

For rescue experiments, MDA-MB-231 cells (150,000 cells per well) were seeded in six-well plates and incubated for 24 h before transfection. Using Lipofectamine™ 2000, cells were cotransfected with siNEG, siAXL, or 34a-mimic at a 10 nM final concentration along with either an OmicsLink™ Expression-Ready ORF-cloning vector containing an AXL cDNA (NM_021913) insert that lacks the 3′ UTR or a control vector with an EGFP insert (2 μg) (GeneCopoeia; Rockville, MD). Twenty-four hours post-transfection, cells were harvested and lysed as previously described.

Cell migration assay

MDA-MB-231 or Hs578T cells (20,000 cells per well) were reverse transfected on the surface of a microporous (8.0-um pore size) PET membrane (BD Falcon™ HTS Insert System, BD Biosciences; San Jose, CA) with siNEG, siAXL, siRNA MET (siMET) (ACCGAGGGAATCAT CATGAAA) (Qiagen; Germantown, MD), or 34a-mimic at a 10 nM final concentration using Lipofectamine™ 2000. Cells were starved 48-h post-transfection by aspirating the growth media and replacing it with serum-free media. After a 24-h starvation, 10% serum-containing media was added to the lower chamber (underside of membrane). Cells were allowed to migrate for 16 h before processing. In brief, cells on the transfection side of the membrane were swabbed and cells on the underside of the membrane were fixed in methanol. Following washes in PBS, cells on the underside of the membrane (migrated cells) were DAPI stained, imaged, and counted using a Discovery-1 Screening System (Molecular Devices; Sunnyvale, CA). All migration assays were performed in triplicate with a total of 10 measurements per transfection condition. Significance was determined by a one-way ANOVA with a Dunnett multiple comparison post-test using siNEG treated samples as the control.

Cell viability/proliferation assay

In parallel with the migration assay, MDA-MB-231 or Hs578T cells (20,000 cells per well; 96-well plate) were reverse transfected with the same siRNA/miRNA mimic molecules and either fluorescence (579_Ex/584_Em nm, CellTiter-Blue^®) or absorbance (490 nm, CellTiter 96^® Aqueous) (Promega; Madison, WI) was determined after the addition of the reagents 76-h post-transfection. All viability assays were performed in triplicate with a total of 30 measurements per transfection condition for each assay. Significance was determined by a two-tailed unpaired t test for means (siNEG treated samples were used as the control for all comparisons).

Cell-cycle analysis and apoptosis

MDA-MB-231 or Hs578T cells (600,000 cells per well; 6-well plate) were reverse transfected with siNEG, siAXL, siMET, or 34a-mimic at a 10 nM final concentration using Lipofectamine™ 2000. After 72 h, cells were washed in DPBS and fixed in 70% ethanol overnight at 4°C. DNA staining was performed with 50 μg/ml propidium iodide and 100U RNAse A for 30 min at 4°C. Data collection was performed on a FACscalibur flow cytometer (Becton-Dickinson; Franklin Lakes, NJ) with CellQuest Pro software. For each sample, 25,000 events were collected and cell-cycle modeling was performed using FlowJo 9.3 (Tree Star, Inc.; Ashland, OR) software. Aggregated cells were gated out, and the percentages of cells within the different phases of the cell cycle were determined. Flow cytometric measurements were made on transfections performed in triplicate for each condition. Significance was determined by a two-tailed unpaired t test for means (siNEG treated samples were used as the control for all comparisons).

For assessing the effects on cell viability and the possible induction of apoptosis following transfection with miR-34a, Annexin V-PE and PI flow cytometry was conducted. MDA-MB-231 or Hs578T cells were harvested 72 h post-transfection and stained following the manufacturer's instructions for the Annexin V-APC Apoptosis Kit (eBioscience, Inc.; San Diego, CA). Data collection was performed on a FACscalibur flow cytometer (Becton–Dickinson; Franklin Lakes, NJ) using CellQuest Pro software. For each sample, 10,000 events were collected, and apoptotic fractions were determined using FlowJo 9.3 (Tree Star, Inc.; Ashland, OR). Flow cytometric measurements were made on transfections performed in triplicate for each condition. Significance was determined by a two-tailed unpaired t test for means (siNEG treated samples were used as the control for all comparisons).

Statistical analyses

All statistical analyses were performed using GraphPad Prism v5.0c software (GraphPad Software, Inc.; La Jolla, CA).

Differentially expressed miRNAs among three breast cancer cell lines with different histotypes

To identify differentially expressed miRNAs among the breast cancer cell lines that differed in their hormone receptor histotypes, we profiled miRNA expression in cell line models representative of the luminal (MCF7 and SK-BR-3; ER⁺/PR⁺/HER-2⁻ and ER⁻/PR⁻/HER-2⁺, respectively) and basal B (MDA-MB-231; ER⁻/PR⁻/HER-2⁻) molecular subtypes, and normalized each profile to that of a common reference sample (MCF10A; normal breast, non-tumorigenic control). Thirty-four miRNAs from a total of 328 human miRNAs (10.4%) spotted on the array passed the threshold of having at least a threefold change in expression in either direction from its median value across all samples/arrays for at least 33% of all arrays tested (Fig. 1a). Of these 34 miRNAs, 21 (61.8%) were observed to meet this threshold in the MDA-MB-231 cell line alone (with either up or down expression relative to MCF10A cells consistent in at least two of 4 arrays (replicates) that were tested). Interestingly, across the three cell lines, only two of these 21 miRNAs (miR-200b and miR-375) had the same directionality in expression (up in both MDA-MB-231 and SK-BR-3 cell lines relative to MCF10A); indicating that the majority of these miRNAs in MDA-MB-231 cells were inversely expressed in the luminal cell lines. Some of our results are consistent with a previously reported miRNA microarray expression profiling study in breast cancer cell lines [34]. For example, the down regulated expression of miR-141 and 200c (clustered on human chromosome 12) and the up regulated expression of miR-221 and miR-222 (clustered on human chromosome X) in the MDA-MB-231 cell line relative to MCF10A were also observations shared in this study. The correlated up regulation of miR-221 and miR-222 has also recently been confirmed in triple-negative breast cancer patient samples [35].

Bioinformatic prediction of a miR-34a/AXL interaction is relevant to the basal B molecular subtype in breast cancer cell lines

Previous studies have shown that breast cancer cell lines, as tumor samples, can be grouped into distinct molecular subtypes based on their global gene expression profiles [4, 6]. In a study by Neve et al. [6], a Prediction Analysis of Microarrays (PAM) that was applied to gene expression data from 51 breast cancer cell lines identified 305 classifiers that would distinguish these cell lines as belonging to either the basal A, B, or luminal subtypes. From this list of classifiers, we focused on those genes specific to the basal B subtype that had the top ten positive and negative PAM scores when applying our target prediction analysis. We chose the basal B subtype because cell lines classified within this group (including MDA-MB-231) are often of the triple-negative histotype, with a characteristic mesen-chymal-like appearance, high proliferation rates, and high invasiveness [4, 6]. We used three separate miRNA target prediction algorithms (each of which with weight different parameters in binding and having different levels of predictive power, yet all are confined to the 3′UTR of the mRNA) to identify putative interactions between our 34 differentially expressed miRNAs with genes in the basal B gene classifier subset. Only those interactions in which there was an inverse correlation in expression between the miRNA and the mRNA target were considered (Online Resource Table 2) for possible follow-up studies. There were six miRNA/mRNA interactions that were identified by all the three-prediction programs: miR-34a/AXL, miR-34a/FOSL1, miR-191/TSPAN13, miR-200b/TSPAN13, miR-222/TSPAN13, and miR-200a/FOXA1. We excluded the miR-200a/FOXA1 prediction because the FOXA1 gene was more of a distinguishing classifier for the luminal subtype (both basal A and B subtypes exhibited low expression of FOXA1) and the free energy of binding between miR-200a, and its target site was not as strong as some of the other predicted interactions. Likewise, with regard to the miRNAs targeting TSPAN13, there were only poorly conserved binding sites for these miRNAs across several species, and the free energy of binding for each miRNA was comparatively weak. The putative interactions predicted for miR-34a with either AXL or FOSL1 were much more appealing, as both had highly conserved (across 11 species in the case for AXL) seed sequence matches of eight consecutive nucleotides beginning at the 5′ end of the miRNA, and strong binding energies (ΔG = −28.5 and DG = −22.8 or −24.3 kcal/mol for AXL and FOSL1, respectively). Although there are two putative binding sites for miR-34a in the 3′UTR of FOSL1 (one site is considered poorly conserved; Online Resource Fig. 1), there appears to be an additional stretch of base complementarity (7 nt) following the seed sequence region of miR-34a that is unique to the targeting site within AXL; making this binding interaction stronger with the possibility of a higher probability of occurrence (Fig. 1b). We focused our next series of experiments on characterizing the possible interaction between miR-34a and the receptor tyrosine kinase AXL, as AXL has been directly associated with a variety of different cancers including those of the breast [21, 36–38]. Moreover, AXL was shown to play an important role in mediating cell motility and invasiveness in breast cancer cells [38].

miR-34a expression is significantly downregulated in triple-negative breast cancer cell lines and is inversely related to AXL protein expression

To confirm the observed expression levels of miR-34a from our array results, we performed a Northern blot hybridization and qRT-PCR analysis with probes and primers specific for the mature form of miR-34a, respectively, using the same cell lines representative of the array as well as six additional cancer cell lines (Fig. 1c, d). Mature miR-34a expression was significantly down regulated in triple-negative breast cancer cell lines representative of the basal B subtype (MDA-MB-231, Hs578T, and BT549), but was expressed in the luminal cell lines MCF7, T47D, and HER-2 expressing SK-BR-3 cell line. MDA-MB-231 cells and other triple-negative metastatic breast cancer cell lines express higher than normal levels of the receptor tyrosine kinase AXL [38–40] (Online Resource Fig. 2); high AXL expression being a positive correlate of cell aggressiveness. Interestingly, AXL expression was also high in the renal cancer cell line SN12C and detectable in the colorectal cell lines SW480 and SW620, whereas miR-34a expression in these cell lines was either absent or weakly expressed, respectively. We examined whether decreased mature miR-34a expression in these cell lines could be the result of an error in the cellular processing of the precursor molecule to the mature form of the miRNA. Our qRT-PCR results indicate that the primary transcript for miR-34a is poorly expressed and that this would explain the low levels of both precursor and mature forms of miR-34a in these cells (Online Resource Fig. 3). Since miR-34a is known to be under the direct regulatory control of p53 [41–48], and mutations in this gene were previously shown to be present in the cell lines we used, miR-34a expression might be influenced by the inability of p53 to bind or activate transcription. Alternatively, a recent study has shown that the inactivation of miR-34a transcription could be the direct result of CpG methylation [49]. Given the correlation between low levels of miR-34a expression in cell lines exhibiting high levels of AXL, and the putative target site for miR-34a in the AXL 3′ UTR, we hypothesize that AXL is a direct molecular target of miR-34a.

miR-34a targets the 3′UTR of AXL to negatively affect luciferase reporter expression

To test whether AXL contains a putative target site for miR-34a, we used the full-length AXL 3′UTR as well as a more defined region of the UTR (48 bp representing nucleotides 18–65 from the start site of the 3′UTR; denoted as wild-type or WT) in a luciferase reporter expression system. We also created expression vectors containing base pair mutations within the putative seed sequence of this defined region to examine the specificity of this targeting interaction. In MDA-MB-231 cells, co-transfection of the full-length 3′UTR of AXL with a siRNA (target site is within the AXL 3′UTR) results in a 97% loss of luciferase reporter expression (Fig. 2a). On average, there was a 38% loss in luciferase expression with the 34a-mimic molecule (P < 0.0001) but no significant change in luciferase expression with a miR-200c mimic. The miR-200c microRNA does not have a predicted target site within the AXL 3′UTR and therefore served as a negative control.

In assays involving a portion of the AXL 3′UTR that contains the putative target site for miR-34a, co-transfection of the WT vector with the 34a-mimic also resulted in an approximate 40% decrease in luciferase expression (Fig. 2b), but this effect was nearly abolished when mutations in the putative seed sequence-binding regions were introduced. A single nucleotide alteration (M2 or M3; corresponding to nucleotide positions 2 and 3, respectively) in the seed sequence region was enough to produce the inhibitory effect.

miR-34a represses AXL at the transcriptional and translational level and has a downstream affect on AKT signaling in MDA-MB-231 cells

Using MDA-MB-231 cells that were previously shown to be deficient in miR-34a, we examined the effects of ectopic miR-34a expression (in the form of a mimic molecule) on AXL mRNA and protein levels. Normalized AXL mRNA levels (relative to PPIB) were determined for each transfection condition using a branched DNA capture assay. As expected, when compared with negative control (siNEG) transfected cells, a loss of ∼80% in relative AXL mRNA levels was observed with a siRNA targeting AXL (Fig. 3a). Surprisingly, transfections with the mimic molecule resulted in an approximate 60% decrease in relative AXL mRNA levels that was highly significant (P < 0.0001). Similar effects were observed in the triple-negative breast cancer cell line Hs578T, in which there was an 86% loss in AXL mRNA levels with the siRNA and a 77% loss in AXL mRNA levels with the miR-34a mimic (Fig. 3a). This decrease in AXL mRNA upon miR-34a over expression, however, is consistent with recent reports suggesting mRNA degradation as the major component of mammalian miRNA-mediated gene regulation [50, 51]. For Hs578T cells, the effects of miR-34a on AXL mRNA levels may have been more pronounced as a result of lower endogenous AXL mRNA levels on average in these cells when compared to MDA-MB-231 cells.

At the protein level, the miR-34a mimic was effective in decreasing AXL protein by 90% compared to siNEG treated MDA-MB-231 cells (Fig. 3b), whereas the receptor tyrosine kinase c-MET (a previously published target of miR-34a; see Refs. [22–27]) was decreased by only 30%. In comparison to negative control treated Hs578T cells, a 60% reduction in AXL protein levels was seen following transfection with the miR-34a mimic, but just a 30% decrease in c-MET levels (Fig. 3b). These data suggest that in both MDA-MB-231 and Hs578T breast cancer cell lines, the affinity of AXL as a target for miR-34a appears to be greater than that of c-MET.

We also examined the downstream effects on AXL signaling when this gene is targeted for regulation by miR-34a. One signaling network known to involve AXL is the PI3K/AKT effector pathway. Although there was not a significant change in PI3K phosphorylation (data not shown), there was an appreciable decrease in phosphor AKT (both Thr308 and Ser473 residues) but not total AKT levels in cells transfected with either siAXL or the 34a-mimic (Fig. 3c). We did not detect any noticeable changes in phospho-ERK1/2 (p42/p44 MAP kinase) and total ERK1/2 levels under these same transfections (data not shown).

More importantly, we conducted an AXL cDNA rescue experiment and observed the changes in phospho-AKT signaling (at the Thr308 residue). MDA-MB-231 cells co-transfected with an AXL cDNA expression vector lacking the 3′UTR sequence fully recovered AXL protein expression as the mRNA transcripts were not targeted by either the siRNA or the 34a-mimic (Fig. 3d). This resulted in a recovery of phospho-AKT (Thr308) levels and supports the targeting influence miR-34a has on AXL.

AXL targeting by miR-34a results in impaired cell migration but has no significant effect on proliferation or cell cycle regulation

In previous studies, AXL was shown to have an active role in cellular migration for a variety of cell types including triple-negative breast cancer [20, 36, 37, 52–55]. We examined the effects of miR-34a mediated knockdown of AXL protein levels on cell migration in both the MDA-MB-231 and Hs578T cell lines. MDA-MB-231 cells transfected with a miR-34a mimic resulted in a 37% decrease in migratory cells (P < 0.001), whereas no effect on cell migration was observed in cells treated with an siRNA targeting c-MET (Fig. 4a). The effect on cell migration was even more pronounced in Hs578T cells, as transfections with the miR-34a mimic resulted in a 53% decrease in migratory cells (Fig. 4b). Interestingly, the si-MET transfection resulted in an overall increase in the number of migratory Hs578T cells that was statistically significant (P < 0.001).

The reduced numbers in migratory cells upon transfection with the miR-34a mimic may be the result of impaired growth, proliferation, or even activation of cell death programming; and it may not necessarily reflect a single direct effect on decreased AXL levels within these cells. To determine whether the regulation of AXL by miR-34a principally alters the migration phenotype of MDA-MB-231 and Hs578T cells rather than alterations in cell growth and survival, we measured cell proliferation/viability and examined the cell cycle for growth arrest or induction of apoptosis in our transfected cells parallel to our migration assays. In MDA-MB-231 cells, we observed a modest decrease in cell viability (13%, P < 0.05) for the miR-34a mimic transfection when using the CellTiter-Blue® Assay (Fig. 4c). The observed results for the Hs578T cell line were very similar to that of MDA-MB-231. Transfection with the miR-34a mimic resulted in a 12% decrease in proliferative ability (P < 0.0001) whereas transfection with an siRNA targeting c-MET increased the viability of the cells by ∼ 15% (Fig. 4d).

When comparing the various transfections in regard to cell-cycle analysis, the miR-34a mimic molecule did result on average in a 2.9% increase in the number of MDA-MB-231 cells in the G0/G1 phase with a concomitant decrease in cells entering the S-phase of the cell cycle when compared to siNEG transfected cells (Fig. 5a). The Hs578T cell line also exhibited an increase in the number of cells that remained in the G0/G1 cell cycle (∼5%) with a decrease in cells at the S-phase (Fig. 5b). In a flow cytometric analysis with Annexin V staining as an indicator of apoptosis, there was a slight increase on average in early or pre-apoptotic MDA-MB-231 cells (compare lower right quadrants) for the siAXL, 34a-mimic, and siMET transfections (Fig. 6a). We did not, however, observe a significant increase in the number of apoptotic cells (upper right quadrant) for each transfection compared to siNEG. Similarly, no significant change in the number of pre-apoptotic or apoptotic Hs578T cells (Fig. 6b) was observed with the 34a-mimic transfection compared to siNEG.

As AXL may not be the only miR-34a target with influence on cell migration, we conducted a global transcriptome profiling experiment on MDA-MB-231 cells transfected with either siNEG or the 34a-mimic to determine other contributing transcripts that might be affected. Other than MET which also exhibited a −0.8 log₂ fold change, we did not observe downregulation of other computationally predicted targets of miR-34a with relevance to cell motility (e.g., RhoA, VCL, TBCA, TBCE, VASP, ANK2, and isoforms of MYO). Moreover, other than CCNE2, known miR-34a-targeted transcripts involved in cell-cycle regulation (e.g., CCND1, CDK6, and E2F3) were also not significantly affected.

Overall, our results suggest that transfections with a miR-34a mimic did not cause a gross inhibitory effect on the viability or proliferative ability of these breast cancer cell lines, and that the observed negative effect of miR-34a expression on cell motility is most likely related to a decreased AXL protein expression in these cells.

miR-34a expression is inversely correlated with AXL mRNA expression in basal-like breast cancer tumor patient samples

We examined whether the expression of both miR-34a and AXL mRNA were correlated with one another in a clinically relevant setting by profiling the expression of these RNAs in 26 tumor samples from patients with basal-like breast cancer (Online Resource Table 4). Application of a nonparametric (Spearman) correlation analysis to the expression data (AXL mRNA normalized to β-actin and miR-34a normalized to RNU48) generated for all the 26 samples shows that AXL and miR-34a expression are inversely correlated with one another (r = −0.52; two-tailed P < 0.01) with significance (Fig. 7).