LincIN, a novel NF90-binding long non-coding RNA, is overexpressed in advanced breast tumors and involved in metastasis

Primary human mammary epithelial cells (HMECs) from breast tumors and matched adjacent non-tumor tissues were isolated and cultured as previously described [30]. For evaluating the expression of LincIN in clinical specimens by in situ analysis, a breast cancer tissue microarray (TMA) was prepared by the Biosample Core Facility of Fox Chase Cancer Center (FCCC). In addition, RNASeq reads per kilobase million (RPKM) values at the LincIN locus (reads falling into: chr10:3360887-3361048) as well as clinical and follow-up information were downloaded from The Cancer Genome Atlas (TCGA) Data Portal (https://​tcga-data.​nci.​nih.​gov) [31].

Genomic deoxyribonucleic acid (gDNA), RNAs and double-stranded cDNAs (ds-cDNA) from paired normal and tumor primary HEMCs were prepared as previously described [30]. gDNA (quantified by PicoGreen assay) and ds-cDNA samples were subjected to whole genome application and fragmentation prior to Illumina HumanOmni5-quad BeadChip hybridization (Additional file 1: Figure S1). gDNAs and ds-cDNAs from seven paired normal-tumor samples plus two technical replicates were analyzed in a total of 32 arrays. The data from five HMEC pairs were included for final analysis after two pairs were excluded by Illumina quality control. Data were analyzed using the Linear Models for Microarray (LIMMA) data package from R with modification (detailed in “Statistics”).

Quantitative PCR (qPCR) was performed using the ABI 7900HT system (Applied Biosystems, Foster City, CA, USA). TaqMan assays for p21 ^WAF1 , LincIN, and GAPDH were designed and purchased from Applied Biosystems. In addition, the qPCR amplicon for each gene was cloned into the pCR4-TOPO vector (Invitrogen, Carlsbad, CA, USA). Linearized plasmids carrying respective gene amplicons were diluted and used for constructing standard curves for each gene. All cDNA samples calculated from 20 ng of total RNA per reaction were assayed in quadruplicate in 384 microwell plates.

Total RNA was isolated using TRIzol reagent (Invitrogen) and the quality of total RNA was assessed by an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). A total of 250 ng of total RNA sample was labeled and hybridized to the Affymetrix Human Gene 2.0 ST Array according to the manufacturer’s instructions (Affymetrix, Santa Clara, CA, USA). Scanned microarray images were analyzed using the Affymetrix Gene Expression Console with the RMA (Robust Multi-array Average) normalization algorithm. Further statistical analyses were performed using BRB-ArrayTools [32].

5′ and 3′ RLM-RACE analysis was performed using the FirstChoice^® RLM-RACE Kit (Life Technologies, Carlsbad, CA, USA) as previously described [30]. For the 5′ end, first internal and nested primer sequences were as follows, ATAAAAAGGATAGATATTTATTTCTCTCACAC and AGAACTCCTGCCCCTCCCCTGT. For the 3′ end, internal and nested primer sequences were as follows, AGCAAAACCTGAAGCCCCAAAGAG and AATTCCCATGGAGGAAAGAG.

RNA ISH was performed with the RNAscope® 2.0 HD formalin-fixed, paraffin-embedded (FFPE) Assay Kits [Advanced Cell Diagnostics (ACD), Newark, CA, USA] following the manufacturer’s user manual. LincIN- specific RNA probes were custom designed by ACD. RNA staining of individual cells was scored semi-quantitatively according to the manufacturer’s instructions with minor modifications. Briefly, the expression levels of LincIN, dapB (negative control) and POLR2 (positive control) were scored manually by two independent observers using the slightly modified guidelines recommended by ACD: 0 (no staining or <1 dot/10 cell), 1 (1–10 dots/cell and few dot clusters) and 2 (>10 dots/cell and >10% of dots are in clusters).

Non-tumorigenic mammary epithelial cell lines, MCF-10A and -10F, and human breast cancer cell lines, BT-20, HCC-1937, MCF-7, MDA-MB-231, SK-BR-3, T-47D, and ZR-75-1, were purchased from American Type Culture Collection (ATCC). Cell lines were maintained according to ATCC recommended medium at 37 °C in the presence of 5% CO₂. MDA-MB-231luc was gifted by Dr. Jose Russo (FCCC). MCF10ADCIS and SUM225 cells were gifts from Dr. Fariba Behbod (University of Kansas Medical Center) and were maintained as previously described [33].

The following primary antibodies were used: NF90 rabbit polyclonal antibody (1:10000, Cat# ab131004, Abcam, Cambridge, MA, USA), NF45 rabbit monoclonal antibody (1:1000, Cat# ab131004, Abcam), p21^WAF1 mouse monoclonal antibody (1:200, Cat# OP64, Calbiochem, La Jolla, CA, USA), Keratin mouse monoclonal antibody (1:1000, Cat# ab8068, Abcam), vimentin mouse monoclonal antibody (1:1000, Cat# V5255, Sigma-Aldrich, St. Louis, MO, USA), and β-actin mouse monoclonal antibody (1:5000, Cat#A5316, Sigma-Aldrich). For secondary antibodies, ECL^™ HRP-conjugated anti-mouse or anti-rabbit IgG (GE Healthcare, Chicago, IL, USA) were used. For pull-down western experiments, HRP-conjugated anti-rabbit heavy chain (Jackson ImmunoResearch Labs, West Grove, PA, USA) was used.

For LincIN short hairpin RNAs (shRNAs), sense and antisense oligos were designed using the Whitehead Institute online tool (http://​sirna.​wi.​mit.​edu/​) and synthesized by Integrated DNA Technologies (IDT) (Coralville, IA, USA). The oligos were then annealed prior to cloning into the Age I and EcoR I restriction enzyme sites of pMKO.1-GFP, which was purchased from Addgene (Cambridge, MA, USA). After screening for LincIN knockdown efficacy, two of the most efficient shRNA constructs were chosen from a total of seven shRNA designs (sequences listed in Additional file 1: Table S1). In addition, the full length of LincIN was synthesized by Genewiz, Inc. (Cambridge, MA, USA), and was cloned into the pRetroX-IRES-ZsGreen vector (Clontech, Mountain View, CA, USA) at the BamH I and Not I restriction enzyme sites.

Dicer-substrate small interfering RNAs (siRNAs) for LincIN were designed and synthesized from IDT. Sequences for siLincIN.A and siLincIN.B were as follows: GGACAUUAUGCAAGGAGAUGGCATC (sense), GAUGCCAUCUCCUUGCAUAAUGUCCUU (antisense); and CACCCUGCCAGAUGUGUCUUGUUCC (sense) and GGAACAAGACACAUCUGGCAGGGUGUC (antisense). Predesigned NF90 siRNAs and scrambled controls (SC) were purchased from IDT. siRNAs were transfected into cells using Oligofectamine^™ or Lipofectamine 3000 reagents (Life Technologies) according to the manufacturer’s instructions.

After breast cancer cells were cultured in a monolayer and reached 90–95% confluence, a scratch wound was carried out by creating a linear cell-free region using sterilized pipette tips as described previously [34]. The progress of cell migration into the scratch was photographed every 24 hours using an inverted fluorescence microscope. The images are further analyzed quantitatively using National Institutes of Health ImageJ software. Invasion assay was performed using the BD BioCoat^™ Matrigel^™ Invasion Chamber (BD Biosciences, Franklin Lakes, NJ, USA) following the manufacturer’s instructions. The invaded cells were stained with 1% crystal violet and examined under a light microscope. Cell cycle assay was done using the Nuclear-ID^® Red Cell Cycle Kit (Enzo Life Sciences, Farmingdale, NY, USA).

All the animal protocols were approved by the Institutional Animal Care and Use Committee (IACUC) at FCCC. Stable lines expressing empty green fluorescent protein (GFP) vector or LincIN shRNAs were generated using MDA-MB-231-Luc-D3H1 cells. To establish a lung metastasis model, 2 × 10⁶ MDA-MB-231-Luc cells were injected intravenously (tail vein) into female SCID mice of 6–7 weeks old. Images were taken at multiple time points (days 1, 7, 14, 21, 28, and 43) after injections of tumor cells. Bioluminescence imaging of animals was performed using standard display methods (exposure time, 1-60s; binning 8; field of view 4; f/stop 1; open filter) on the IVIS® Spectrum system (PerkinElmer, Waltham, MA, USA). Total photon plux (photons/sec) was determined from the region-of-interest (ROI) using Living-Image (Xenogen, Hopkinton, MA, USA) analysis software.

Full-length sense and antisense of LincIN or LincIN RNA fragments (1-320, 332-554, and 538-1031) were in vitro transcribed with the TranscriptAid T7 High Yield transcription Kit (Thermo Fisher Scientific, Waltham, MA, USA) and labeled with biotin using the Pierce™ RNA 3′End Desthiobiotinylation Kit (Thermo Fisher Scientific). The whole lysates from MDA-MB-231 cells were freshly prepared with the RNasin® Ribonuclease Inhibitor (Promega, Madison, WI, USA) and protease/phosphatase inhibitor cocktail (Roche, Basel, Switzerland). Pull-down experiments were performed using the Pierce™ Magnetic RNA-Protein Pull-Down Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. LincIN-associated proteins were eluted and resolved by gel electrophoresis followed by staining with the SilverQuest™ Silver Staining Kit (Life Technologies) or directly used for Western blotting. Protein bands of interest were excised, de-stained, and digested prior to analysis by LC-MS/MS using reverse phase capillary high-performance liquid chromatography (HPLC) with a Thermo Electron LTQ OrbiTrap XL mass spectrometer at The Wistar Institute.

RIP experiments were performed using the Magna RIP^™ RNA-Binding Protein Immunoprecipitation Kit (EMD Millipore, Billerica, MA, USA) according to the manufacturer’s protocol. Briefly, fresh lysates from MDA-MB-231 cells were prepared using RIP lysis buffer containing a protease inhibitor cocktail and RNase inhibitor. Five μg of NF90 or control antibodies (anti-SNRNP70 and normal rabbit IgG) were captured by magnetic beads and incubated with 100 μL cell lysate for 8 h at 4 °C. The co-precipitated RNAs were extracted using protease K and phenol/chloroform precipitation. Precipitated RNAs and total RNAs (input controls) were treated with TURBO^™ DNase (Life Technologies) prior to the reverse transcription with the iScript^™ cDNA Synthesis Kit (Bio-Rad Laboratories, Hercules, CA, USA). Analysis of LincIN RIP signals were performed using a custom-designed TaqMan assay (Applied BioSystems) via real-time PCR. IP was performed using the Magnetic Dynabeads Kit (Thermo Fisher Scientific) according to the manufacturer’s manual.

The LIMMA (Linear Models for Microarray Data) methodology [35, 36] was used to identify differentially expressed intergenic lncRNAs between paired tumor and normal samples. The Benjamini-Hochberg method was used to adjust for multiple testing and to calculate the false discovery rate (FDR) for each lncRNA probe [37]. Computations were performed using the R statistical language and environment [38]. The Wilcoxon signed-rank test was used to compare differences between paired tumor and normal samples for RNA-ISH and TCGA data; and Kaplan-Meier survival curves were compared using log-rank tests. Kruskal-Wallis and Fisher’s exact tests were used to test the association between clinical variables and LincIN expression levels. In vitro data was analyzed using one-way analysis of variance (ANOVA) and Dunnett’s test to account for multiple post hoc comparisons or the two-sample t test. All tests were two-sided and used a type I error of 5%. TCGA data was analyzed using SAS software, version 9.4 (SAS Institute Inc., Cary, NC, USA). If not mentioned specifically, all experiments were repeated in triplicate.

To identify novel breast cancer-associated lncRNA candidates at intergenic regions, we employed a high-density SNP array-based approach by specifically probing intergenic regions to uncover lncRNA genes, whose expression was altered in breast tumors (Fig. 1a). To minimize the confounding effect from admixed stromal cells, we enriched the epithelial cell population and investigated their differential expression patterns between invasive ductal carcinoma (IDC) tissues and adjacent normal breast tissues (Additional file 1: Figure S1A).

As a result, we identified 26 intergenic lncRNA transcripts that are dysregulated in tumors [P < 0.005, FDR < 0.15, |Log₂ (fold change)| >1]. Among them, 14 lncRNAs (targeted by 22 SNP markers) were upregulated, and 12 lncRNAs (targeted by 16 SNP markers) were downregulated (Fig. 1b and Additional file 1: Table S2). One of the dysregulated intergenic lncRNAs, termed LincIN (GenBank access number: KX352723), is located at Ch10p11-12 and between two coding genes, I TGB1 and N RP1 (Fig. 1c). Notably, data from five lncRNA exon-targeting probes (kgp29833987, kgp21612648, kgp5242998, kgp21865691, and kgp391420) showed that LincIN is significantly upregulated in breast tumors in comparison to the normal components (Log₂FC = 1.3–2.3 and P < 0.005) (Fig. 1b and Additional file 1: Table S2). Further gene expression analysis by RT-qPCR demonstrated that LincIN is significantly upregulated in the majority of tumor HMEC lines in comparison to paired normal lines (8 out of 10, P < 0.05) (Fig. 1d).

To characterize the full-length transcript of LincIN, we performed 5′ and 3′ rapid amplification of cDNA ends (RACE) and identified two spliced LincIN RNA variants (Fig. 1e). Sequence analysis of LincIN transcripts revealed that LincIN RNA utilizes two poly(A) sites to generate a short transcript (837 bp) and a longer transcript (1031 bp) (Fig. 1e). The expression of the longer LincIN variant appears to be more abundant and thus is the focus of our functional analysis. In addition, cellular fractionation analysis revealed that LincIN is distributed in both the nucleus and the cytoplasm (Additional file 1: Figure S2). Lastly, the 3′ end of LincIN appears to be polyadenylated since LincIN is clearly detected in poly(A)- enriched RNA fractions (Additional file 1: Figure S2).

Based on a BLASTX analysis of all possible reading frames identified by the open reading frame (ORF) finder from the NCBI and ATGpr (http://​atgpr.​dbcls.​jp/​), LincIN lacks the potential to encode any recognizable protein domains. As described previously [19], we further examined the coding potential of LincIN using a bioinformatics tool, PhyloCSF, a comparative genomics method for distinguishing protein-coding and non-coding regions based on their evolutionary signatures characteristic to alignments of conserved coding regions [39]. This analysis confirmed the low coding potential of LincIN, which receives a maximum smoothed codon substitution frequency (CSF) value similar to other well-characterized lncRNAs (Additional file 1: Figure S3). These findings established LincIN as a new lncRNA, and we next investigated its role in breast cancer development.

Previously, one genome-wide association study (GWAS) identified a breast cancer survival variant, rs11591508 (P = 3.27 × 10^-6, hazard ratio = 2.41–3.29), which is located at the LincIN locus [40] (Fig. 1c). This finding suggested that LincIN may be a biomarker for breast cancer prognosis. Therefore, the expression of LincIN in breast tissues was evaluated initially by RNA in situ hybridization (RNAscope^®) utilizing an in-house TMA, which contained a panel of breast normal (36) and tumor (103) specimens. After inspecting RNA quality with the POLR2 and dapB staining, 20 samples with RNA degradation were excluded, which left the cohort with 88 tumors and 31 normal specimens including 27 tumor/normal pairs for final analysis. Figure 2a shows the representative RNAscope^® ISH images with different LincIN staining levels. Among all the specimens, LincIN exerts positive staining in approximately 72% of the breast tumor tissues (63 out of 88, score ranging from 1–2) while only approximately 29% of the normal tissues show positive LincIN expression (9 out of 31, score 1–2; P < 0.001) (Fig. 2b). Furthermore, LincIN levels are significantly higher in tumors comparing to those in matched adjacent normal tissues (P < 0.001) by Wilcoxon test (Fig. 2c). To further examine the potential value of LincIN as a breast cancer prognostic biomarker, we used a larger breast cohort retrieved from the TCGA Data Portal. After initial exclusions for replication and probe values of zero, there were expression data for 752 tumor specimens and 98 normal specimens. Among them, 88 normal samples were collected from matched breast tumor. Consistently, quantification of LincIN levels demonstrated significantly higher expression levels of LincIN in tumors versus matched adjacent normal tissues (n = 88) (4.28 vs.1.96, P = 0.018) (Fig. 2d). Notably, the levels of LincIN in breast tumors increase significantly with the pathologic stages defined by the American Joint Committee on Cancer (AJCC) or the tumor size (Table 1). For overall survival analysis, patients with follow-up days = 0 and pathologic tumor stage stated as “not available” were excluded, and 738 patients were included in the final survival analyses. Because of the skewed distribution of LincIN, we used categories based on LincIN levels rather than percentiles, with < =1, >1–5, >5–10, and >10–83 as the initial groups. Based on similarities in survival in the higher LincIN groups, the upper categories were combined into LincIN > 1. The survival analysis showed that patients expressing high levels of LincIN (>1.0) have worse survival outcomes (P = 0.044 and P = 0.011 after adjust for age) (Fig. 2e).

As LincIN is frequently overexpressed in advanced breast tumors, we sought to explore the functional roles of LincIN in breast cancer progression-metastasis. Examining LincIN expression levels in a panel of 12 breast cell lines showed that LincIN is overexpressed up to 40-fold in highly metastatic MDA-MB-231 cells (ER-/PR-/HER2-), versus immortalized but non-transformed MCF10A cells (P < 0.001) (Additional file 1: Figure S4). We next evaluated the impact of LincIN loss on cell migration and invasion in MDA-MB-231 cells. Suppression of LincIN by RNA interference (RNAi) leads to an approximately 40–50% reduction of cell invasion compared to the control group (P < 0.05) (Fig. 3a). This inhibitory effect is even more profound in HCC1937 cells (high expression of LincIN, ER-/PR-/HER2-), where about 50–80% reduction is observed as the result of silencing LincIN (P < 0.01) (Fig. 3b). Knockdown of LincIN in MDA-MB-231 cells also significantly decreases cell migration about approximately 30% compared to those transfected with vector in the wound closure assay (Additional file 1: Figure S5A). In contrast, overexpression of LincIN in MCF10ADCIS cells (non-invasive breast cancer cells and low expression of LincIN) tended to accelerate cell migration (P < 0.05) (Additional file 1: Figure S5B). We also examined the role of LincIN in cell proliferation. As shown in Additional file 1: Figure S5C, overexpression of LincIN in MCF-10A and MCF10ADCIS cells increases cell proliferation moderately (P < 0.05). However, downregulation of LincIN in MDA-MB-231-luc cells has no significant effects on cell proliferation. Taken together, these data suggest that LincIN plays a role in breast tumor cell migration and invasion in vitro, while the effects of LincIN on cell proliferation may be cell type-specific.

Next, we injected MDA-MB-231luc cells via the tail vein into severely combined immunodeficiency (SCID) mice to test a possible role of LincIN in cancer cell metastasis. Strikingly, we found that knockdown of LincIN significantly decreases lung metastases compared to those in the vector control, as evaluated by bioluminescent signals (P < 0.005–0.05) (Fig. 3c and d). Furthermore, quantification of scanned images of whole lung tissues showed a significant decrease (approximately 60%) in the metastatic area in mice receiving shRNA-treated cells as compared to those of the vector control (P < 0.05) (Fig. 3e). In addition, we observed variations of the inhibitory effects on lung metastasis between shRNA1 and shRNA2 treatments. To exclude potential off-target effects, we performed transcriptome analysis in MDA-MB-231 cells treated with vector control, shRNA1 or shRNA2 in duplicate. As a result, we found 173 and 321 differentially expressed genes for shRNA1 and shRNA2 treatments, respectively, in comparison to the vector control (P < 0.001, Additional file 1: Table S3). Among them, 122 genes are overlapped (Additional file 1: Figure S6A), and the values of Log2 (FC) of overlapping “hits” are significantly correlated between the two shRNA groups (R2 = 0.9596, P < 0.0001 by Spearman’s rho test, Additional file 1: Figure S6B). Furthermore, biological process analysis of differentially expressed genes by Ingenuity Pathway Analysis (IPA) showed that over 97% of the identified pathways are overlapped (74 out of 75 or 76) for shRNA1 or shRNA2 treatments, respectively (P < 0.01, Additional file 1: Figure S6C and Additional file 1: Table S4). These results suggest that the variations from different shRNA effects are not accounted for by off-target effects. Notably, the very top cellular functions targeted by both shRNAs was cellular movement (P < 1.05 × 10^-6 approximately 5.27 × 10^-3), providing molecular insight for the role of LincIN in tumor cell invasion through the regulation of gene transcription (Fig. 3f). Collectively, our results suggest that LincIN is involved in breast cancer cell invasion in vivo and knockdown of LincIN in breast cancer cells may effectively inhibit the metastatic processes.

To identify LincIN-interacting protein partner(s) that may contribute to its function in breast cancer development, we used in vitro transcribed biotin-labeled full-length LincIN for pull-down experiments followed with protein identification by mass spectrometry (MS) (Fig. 4a). The spectra counts were used to determine which proteins were unique/more abundant in a particular condition/gel band. Among the proteins identified by MS (Fig. 4b), NF90/ILF3 was the most abundant LincIN-binding partner. NF45/ILF2, which dimerizes with NF90 to form a functional complex, was also identified by MS as a LincIN-interacting protein. The LincIN-NF90 interaction was then validated by RNA pull-down Western blotting, and NF90 or NF45 were enriched in LincIN-pull-down cell lysates (Fig. 4c). To identify the critical regions of LincIN RNA required for NF90 binding, in vitro transcribed biotin-labeled truncated LincIN transcripts were also prepared for RNA pull-down experiments (Fig. 4d, left panel). Transcripts that contained the 5′ region of LincIN exhibited the strongest binding to NF90, while was weaker binding was observed in the 3′ and center regions (Fig. 4d, right panel). We further confirmed the interaction between LincIN and NF90 by performing RNA immunoprecipitation (RIP) with an antibody against NF90. As shown in Fig. 4e, LincIN RNA was enriched by approximately 40-fold in NF90 antibody precipitates. Altogether, RNA-pull-down and RIP experiments provided reciprocal evidence that LincIN directly interacts with NF90.

Recent studies showed that NF90 is multifunctional in cells, including repressing p21 protein expression in cancer cells [28]. We next examined if LincIN plays a role in the NF90-mediated p21 pathway. Overexpression of LincIN in MCF10A cells diminishes p21 protein expression by approximately 40% (left panel, Fig. 5a,). Reciprocally, the level of p21 protein is consistently induced (approximately twofold) in LincIN knockdown groups compared to scrambled control (right panel, Fig. 5a). In contrast with altered protein expression, static p21 mRNA levels remain unchanged in both LincIN overexpression and knockdown experiments (Fig. 5b), suggesting that LincIN regulates p21 at the translational level. Moreover, loss of LincIN results in increased G1/G0 arrest (P < 0.05) (Fig. 5c and d), which is a typical phenotype associated with the elevated expression of p21. In addition, we have evaluated the p21 level in the mouse lung metastasis colonies by immunohistochemistry (IHC), and the nuclear p21 levels are higher in the metastasis colonies from the LincIN shRNA knockdown group compared to those tissues from shRNA control group (P < 0.05, Additional file 1: Figure S7 and Additional file 1: Table S5). This finding suggests that upregulation of nuclear p21 by LincIN knockdown may be associated with less aggressive phenotype in our metastasis model.

Since recent studies showed that NF90 may regulate p21 protein expression at the translational level via binding its 3′ untranslated region (UTR) [29], we sought to determine if LincIN mediates p21 protein expression via NF90. We knocked down LincIN, NF90, or both using siRNAs and measured p21 expression by Western blotting. As expected, knockdown of NF90 upregulates p21 protein levels (approximately twofold) (right panel, Fig. 5e). Importantly, under the condition of NF90 knockdown, the effects of siLincIN on elevating p21 expression (evaluated by the ratio of siLincIN + siNF90/siNF90) in HeLa cells is decreased in comparison to the cells treated with siLincIN alone (the ratio of siLincIN/SC) (P < 0.05) (right panel, Fig. 5e). Furthermore, LincIN overexpression-induced p21 inhibition is also partially diminished by NF90 silencing in MCF-10A cells (Fig. 5f). Knockdown or overexpression of LincIN had no significant effects on the expression of NF90 (Fig. 5e and f). Taken together, these results suggest that LincIN may cooperate with NF90 to inhibit p21 translation.