LSD1 modulates the non-canonical integrin β3 signaling pathway in non-small cell lung carcinoma cells

The epigenetic writer lysine-specific demethylase 1 (LSD1) is aberrantly upregulated in many cancer types and its overexpression correlates with poor survival and tumor progression. In this study, we analysed LSD1 function in non-small cell lung cancer adenocarcinomas. Expression profiling of 182 cases of lung adenocarcinoma proved a significant correlation of LSD1 overexpression with lung adenocarcinoma progression and metastasis. KRAS-mutated lung cancer cell clones were stably silenced for LSD1 expression. RNA-seq and comprehensive pathway analysis revealed, that genes related to a recently described non-canonical integrin β3 pathway, were significantly downregulated by LSD1 silencing. Hence, invasion and self-renewal capabilities were strongly decreased. Notably, this novel defined LSD1/integrin β3 axis, was also detected in human lung adenocarcinoma specimens. Furthermore, the linkage of LSD1 to an altered expression pattern of lung-lineage specific transcription factors and genes, which are involved in alveolar epithelial differentiation, was demonstrated. Thus, our findings point to a LSD1-integrin β3 axis, conferring attributes of invasiveness and tumor progression to lung adenocarcinoma.

Epigenetic dysregulation has been demonstrated to be involved in the initiation and progression of many cancer types 8,9 . In NSCLC, histone modifications such as histone acetylation and methylation are crucial prognostic indicators for tumor progression and malignancy 9 . The lysine specific demethylase 1, LSD1, also known as KDM1A, is a key epigenetic writer and specifically demethylates mono-and di-methylated histone H3K4 and H3K9 10,11 . LSD1 is aberrantly upregulated in many human cancer types, such as prostate, breast and lung carcinoma as well as neuroblastoma and leukemia [12][13][14][15][16] . Importantly, cancer-related LSD1 overexpression is associated with increased cell proliferation, invasion and migration underscoring the therapeutic potential of LSD1 inhibition 13,16,17 . In a proliferation screen of cell lines representing various tumor types, acute myeloid leukemia (AML) and SCLC cell lines have been shown to be particularly sensitive to pharmacological LSD1 inhibitors, resulting in cytostatic growth inhibition and advanced differentiation 15,18 .
In the present study, we demonstrate that LSD1 is specifically upregulated in poorly differentiated and metastasized lung AC. Surprisingly, LSD1 knockdown induced a broad transcriptome change as demonstrated by comprehensive RNA-seq gene expression profiling. In particular, LSD1 knockdown resulted in downregulation of genes, which are involved in the previously described non-canonical integrin β3 affecting the stem-like phenotype with invasion and self-renewal capacities 7 . Mechanistically, LSD1 was critical for integrin β3 and galectin-3 expression. Furthermore, overexpression of LSD1 significantly correlated with integrin β3 expression in lung AC. Therefore, our findings provide a rationale to overcome cancer stemness and metastasis by targeting LSD1 in NSCLC in future therapeutic strategies.

Results
High LSD1 expression indicates lung cancer malignancy. We first determined LSD1 expression in the previously published transcriptome sequencing data comprising 198 lung cancer specimens ( Fig. 1A and Supplementary Table S1) 3,[19][20][21] . Interestingly, the highest LSD1 expression levels were detected in SCLC, which is associated with the worst prognosis 22 (Fig. 1A). Albeit LSD1 expression was lower in AC than in other lung cancer types, immunohistology on lung AC compared with non-tumorous lung tissues implies that LSD1 is also overexpressed in AC (Fig. 1B). To determine the implication of LSD1 expression in NSCLC AC, LSD1 expression levels were assessed by comprehensive immunohistochemistry analyses of 182 lung AC specimens ( Fig. 1B and Table 1). All ACs were derived from patients who did not receive any preoperative or postoperative treatment and were classified according to the tumor grade, lymph node metastasis, and the KRAS and EGFR mutational status (Table 1). A markedly more intense LSD1 staining was observed in poorly differentiated, high grade AC ( Fig. 1B and C) 23 . Moreover, LSD1 expression was significantly higher in primary AC with lymph node metastasis than the ones which had not developed metastasis (Fig. 1D). In addition, in NSCLC specimens with high LSD1 expression, we observed a slightly less KRAS mutation frequency. However, no statistical significance was found for an inverse correlation between the KRAS mutation status and LSD1 expression (Fig. 1E). Further analyses of LSD1 protein expression on various NSCLC cell lines also suggested a possible inverse link between LSD1 expression and constitutive KRAS activation by mutation (Fig. 1F).

LSD1 is critical for an invasive phenotype of A549 cells by regulating the non-canonical integrin pathway.
To investigate the function of LSD1 in NSCLC, stable LSD1 knockdown (KD) cells from A549 lung AC cell line were generated using retroviral transduction of a shLSD1 construct primarily tested to be highly efficient. Two monoclonal cell populations (KD9 and KD15) ( Fig. 2A) were established and used for RNA-seq  Table S3). Ingenuity pathway analysis (IPA) showed that pathways such as clathrin-or caveolar-mediated endocytosis, Gα12/13, RhoA and integrin signaling, which are all associated with invasive migration, were affected by LSD1 silencing (Fig. 2C). The influence of LSD1 on altered expression profiles of invasion and migration related pathways, were confirmed by transwell migration assays demonstrating impaired invasive capacities of LSD1 silenced A549 cells (Fig. 2D). In order to validate these findings, we established additional LSD1 KD clones from the KRAS Q61H mutated H460 and H1975 carrying the EGFR L858R, T790M mutation ( Fig. 2A).
Similarly to LSD1 silencing of A549 cells, LSD1 knockdown decreased invasion capacity of H1975 cells (Fig. 2D). Surprisingly, LSD1 silencing had only a moderate effect on cell growth and no signs of apoptosis were observed in A549 cells ( Supplementary Fig. S1A).
LSD1 contributes to self-renewal by modulating integrin β3-KRAS-NF-κB pathway. IPA Upstream Regulator analysis suggested that transcriptional target genes of the EGF/EGFR and KRAS signaling pathway were negatively regulated in LSD1 silenced A549 cells (Fig. 3A). In particular, signaling components related to the integrin β3-KRAS-NF-κB signaling axis, previously described by Seguin et al. 7 , were affected. Thus, galectin-3 (LGALS3), RalA binding protein (RALBP1) and integrin β3 (ITGB3) were significantly downregulated Original data are provided in Supplementary Table S1. Mann-Whitney U test was used to calculate the statistical significance. ***P < 0.001. (B) Representative images of immunostainings with LSD1 antibodies on TMAs including 182 NSCLC AC and 11 non-tumor lung tissues. Immunostaining showed only moderate LSD1 expression in normal lung tissues, but elevated LSD1 expression in NSCLC-AC. The LSD1 expression in NSCLC-AC was classified using a low, medium and high immunostaining score as shown exemplarily for low and high LSD1 expression. Scale bars indicate 100 µm. (C-E) LSD1 expression analysed by immunohistochemistry on 182 lung adenocarcinomas was correlated with tumor grades (C), lymph node metastasis (D) and KRAS mutation (E). Chi-Square tests were used to calculate the statistical significance for linear-by-linear association. *P < 0.05, ***P < 0.001 and n.s. = not significant. Original data for C-E can be found in Supplementary Table S2. (F) LSD1 expression in non-small cell lung cancer cell lines with various mutational background determined by immunoblot. by LSD1 knockdown of A549 cells (Fig. 3B). In order to validate downregulation of members involved in the non-canonical integrin β3 signaling, we generated LSD1 overexpressing A549 cells by stable transfection with a Flag-LSD1 construct ( Supplementary Fig. S1B) and analyzed the expression profiles by quantitative RT-PCR. Whereas A549 LSD1 knockdown cells showed downregulation of the integrin β3 signaling member genes, we observed an upregulation of these genes in A549 cells stably expressing the flag-LSD1 construct (Fig. 3B). The effect of LSD1 on gene expression of galectin-3, RALBP1, and integrin β3 could be further confirmed on H460 NSCLC LSD1-knockdown cells (Fig. 3B). Previous work showed that the non-canonical integrin β3 pathway is involved in stem cell like properties and erlotinib resistance 7 . In order to test self-renewal capacity, we analyzed tumor cell growth in 3-dimension (3D) cell culture conditions. Knockdown of LSD1 impaired tumorsphere formation, whereas stable overexpression of LSD1 clearly enhanced 3D colony formation (Fig. 3C). Since it has been shown that an integrin β3 positive stem-like tumor cell population is selectively resistant to receptor tyrosine kinase inhibitors such as erlotinib, we exposed A549 cells to erlotinib 7 . In agreement with these previous findings, we observed that knockdown of LSD1 sensitized A549 cells to erlotinib treatment, decreasing the 3D cell growth capability ( Supplementary Fig. S1C).
Subsequently, signaling components related to the integrin β3-KRAS-NFκB signaling axis were analyzed in LSD1 silenced A549 monolayers and 3D tumorspheres, respectively (Fig. 3D). The integrin β3 protein level was significantly diminished upon LSD1 knockdown in A549 cells as well as in other integrin β3 positive cells such as H460 or H1975 (Fig. 3E). Galectin-3 and RalBP1 of which basal levels were higher in A549 spheroids than in the monolayer culture, were prominently downregulated in the sphere condition of LSD1 silenced A549 cells (Fig. 3D). Knockdown of LSD1, however, failed to abrogate KRAS downstream MEK1/2 signaling in the sphere cultures, suggesting that KRAS-MEK1/2 signaling is not essential for LSD1-mediated 3D colony formation (Fig. 3D).
Next, we investigated whether LSD1 affects the NF-κB mediated transcriptional response using a luciferase reporter expression system containing NF-κB responsive elements. Knockdown of LSD1 significantly decreased basal NF-κB transcriptional activity (Fig. 3F). In addition, transcription of NF-κB target genes was greatly inhibited upon LSD1 knockdown ( Supplementary Fig. S1D). Galectin-3, the non-canonical ligand of integrin β3, has been reported to facilitate the integrin β3-KRAS interaction leading to NF-κB activation 7 . Addition of recombinant galectin-3 slightly enhanced NF-κB activity, however, did not recover the LSD1 knockdown-mediated blockage of NF-κB pathway possibly due to the strong downregulation of integrin β3 in LSD1 silenced cells (Fig. 3F).
The LSD1-integrin β3 axis in NSCLC-AC. After having demonstrated the impact of LSD1 on integrin β3 expression, we studied whether LSD1 expression might influence the integrin β3 level in clinical specimens from lung AC patients. Indeed, the association between LSD1 and integrin β3 was shown to be highly significant. Negative integrin β3 staining was observed prominently in low LSD1 expression group but very rarely in high LSD1 expression group ( Fig. 4A and B), which supports a functional link between LSD1 and integrin β3 expression in vivo. LSD1 knockdown impacts differentiation of A549 cells. As detailed above, LSD1 knockdown induced significant changes in gene expression profiles of key mediators of cell morphogenesis in differentiation ( Fig. 2C and Supplementary Fig. S2A). Further, LSD1 conferred tumor cells a stem-like phenotype by modulating the integrin β3-KRAS-NF-κB pathway. We thus hypothesize that LSD1 might be a critical factor affecting the differentiation state of lung AC A549 cells. Targets of the lung lineage transcription factors FOXA1/2 and NKX2-1 were highly affected by LSD1 silencing (Fig. 5A) 24,25 . In addition, the transcriptional responsive genes of dexamethasone, a synthetic glucocorticoid and C/EBPα/β, both of which are involved in late alveolar maturation, were greatly regulated by LSD1 knockdown (Fig. 5A) 24 .
Furthermore, a survey of lung epithelial marker genes revealed that many hallmarks of alveolar type 2 (AT2) and bronchial clara cell markers were altered reflecting a change in cell differentiation state upon LSD1 knockdown (Fig. 5B). A549 cells primarily originated from AT2 cells 26 , appear to have distorted molecular signatures such as the loss of AT2 marker genes and the aberrant gain of clara cell marker genes. The expression of the AT2 cell marker genes, e.g. NKX2-1 and SFTPB is silenced in A549 cells indicating that the transformed AT2 cells lost their cell identity and are not capable of producing surfactant proteins like SFTPC (Fig. 5B,C) 26 . Inhibition of LSD1 partially reactivated AT2 cell marker gene expression while on the other hand it decreased expression of genes responsible for the clara cell phenotype (Fig. 5B,C and Supplementary Fig. S2B). Finally, some of AT2 and clara cell marker genes were suggested to be directly targeted by LSD1, as gene regulation upon LSD1 knockdown was reversed by overexpression of LSD1 in A549 (Fig. 5D).

Discussion
In our study, we found that LSD1 expression level varied considerably among the different subtypes of lung cancer. RNA-seq analysis of 198 lung cancer specimens showed highest LSD1 mRNA levels in SCLC, which might explain the marked effect of the LSD1 inhibitor GSK2879552 in SCLC cell lines 18 . In comparison to SCLC, AC presented with lower LSD1 mRNA levels. However, analysis of LSD1 expression in 182 AC specimens showed that high LSD1 expression correlated with enhanced lung tumor malignancy. Strong LSD1 expression co-occurred with higher tumor grade and lymphatic invasion, suggesting a crucial role of LSD1 in cellular dedifferentiation and metastasis in lung AC.
in the pathways are shown in a table (right panel). (D) Representative images of invasion assay of A549 cells expressing GFP shRNA control or LSD1 shRNA (left panel). Bar graphs showing the quantification of Crystal Violet staining of Boyden Chamber transwell filters (right panel). The bar graphs represents the mean ± SEM for n = 4 (A549) and n = 5 (H1975). *P < 0.05 (Student's t-test). Scale bars, 50 µm.  17 . In our study we observed only a slightly reduced LSD1 expression in AC with an activating KRAS mutation compared to AC, which did not harbor a KRAS mutation. For our further studies we established stable LSD1 knockdown cells from NSCLC cell types. Although LSD1 knockdown-mediated influence seems to have less effect on the cellular proliferation in early passages, cell growth was inhibited in late passages. In A549 and H1975 KD cells, we show a pronounced impact on the oncogenic transformational capacities such as invasion and self-renewal by LSD1 knockdown. Accordingly, the constitutive activation of MEK1/2-ERK pathways, known to be crucial for cell proliferation, was not blocked by LSD1 knockdown in A549 cells. Instead, galectin-3 and integrin β3 expression was severely downregulated by LSD1 silencing. This suggests that LSD1 is rather implicated in cancer stem cell function by regulating galectin-3-integrin β3-KRAS signaling 7 . Interestingly, galectin-3 has been identified as the marker gene for alveolar type 1 (AT1) cells 27 and damage in AT1 cells can induce the activation of KRAS signal axis invoking a stemness phenotype in lung alveolar progenitor cells 28 . Similarly to galectin-3, galectin-1 has recently been identified to promote lung cancer metastasis by potentiating integrin α6β4 and Notch1/Jagged2 signaling 29 . This indicates an important role of galectins as signal modulators that activate various oncogenic signal pathways during tumor progression.
Whereas the contribution of LSD1 to proliferation is not clear in KRAS mutated AC, there is accumulating evidence that LSD1 is significantly associated with cell invasion, migration and metastasis in NSCLC 16,17 . Previous studies showed that LSD1 is required for upregulation of epithelial-mesenchymal transition (EMT) markers including twist family bHLH transcription factor 1 (TWIST1), E-cadherin and N-cadherin and downregulation of metallopeptidase inhibitor TIMP3, thereby contributing to cellular invasion and migration 16,17 . Here, we performed a comprehensive expression profiling using RNA-seq showing that LSD1 knockdown in A549 cells induced a global transcriptome change. EMT signaling pathways and EMT markers such as Zinc finger E-box binding homeobox 2 (ZEB2), KIT proto-oncogene receptor tyrosine kinase (KIT), and cadherins (CDHs) were differentially regulated in agreement with the previous studies (Fig. 2C) 16,17 . Furthermore, pathways associated with invasive phenotypes such as endocytosis, RhoA and integrin signaling were affected by LSD1 knockdown.
Most importantly, LSD1 knockdown interfered with integrin β3-KRAS signaling axis by a pronounced integrin β3 repression. Recently, a number of integrin subunits, particularly β3 and β4 have been highlighted as markers and functional regulators of normal adult stem and progenitor cells, as well as cancer stem cells 30 . Integrin β3 is enriched in mammary luminal progenitors, whereas differentiated luminal cells lack integrin β3 31 . In breast and pancreatic tumors, expression of integrin αvβ3 is necessary to drive tumor cell anchorage-independence and metastasis by recruiting and activating Src kinase 32 . In the Kras G12D ;Trp53 fl/fl lung cancer model, integrin β4 has been characterized as a new cancer stem cell marker involved in self-renewal, tumor propagation and cisplatin resistance 33 . Integrin α6β4 positive cell subpopulation from mouse lung could expand clonally as progenitors but also differentiate into mature airway and alveolar cells 34,35 .
In light of this data, blocking LSD1 appears to be critical to impair integrin β3-KRAS mediated stemness along with invasive phenotype by downregulating integrin β3. In agreement, LSD1 expression was higher in poorly differentiated and metastasized adenocarcinomas and correlated with the expression of integrin β3 in vivo. In agreement to the previous studies of Seguin et al. 7 , the integrin β3-KRAS signal axis is also important for the invasive and self-renewal ability of H1975 cells due to their activation by the EGFR/KRAS pathway. Since we show that LSD1 affects integrin β3 expression, as well as self-renewal and invasion capacities of H1975 cells, the LSD1 influence on this non-canonical integrin β3 signaling seems to be also valid for the EGFR mutant cell context. While potentiating stemness, LSD1 suppresses pro-differentiation processes as indicated by the altered expression of genes involved in cell morphogenesis in differentiation and late alveolar maturation upon LSD1 knockdown. A549 cells which are derived from AT2 cells, have distorted molecular signatures such as the aberrant gain of airway cell marker genes. This suggests that transformed A549 AC cells mimic a cell type with partially bronchioalveolar progenitor-like gene signature expressing either bronchial or alveolar marker gene expression 4 . LSD1 inhibition could partially reactivate expression of silenced AT2 marker genes, whereas decrease expression of bronchial marker genes, altering cell differentiation state in A549 cells, a mechanism similar to the observed pro-differentiation effect in SCLC and leukemia upon loss of LSD1 15,18 .
Taken together, our data show that LSD1 confers lung adenocarcinoma cells with invasive and dedifferentiated attributes by modulating a non-canonical integrin β3-KRAS signaling pathway. Pharmacological LSD1 inhibition (A549-Flag-LSD1) and the LSD1 knockdown H460 cells vs. shGFP H460 cells (H460-KD21). Changes in transcript levels were calculated by ΔΔCt values and expressed as Log 2 fold change (FC). (C) Phase contrast microscope images of self-renewal tumorspheres of A549 expressing non-silencing GFP shRNA or specific LSD1 shRNA (left panel). Effect of LSD1 on tumorsphere formation in A549 and H1975 cells overexpressing or lacking LSD1 measured by quantifying the number of primary tumorspheres (right panel). The bar graph represents the mean ± SEM for n = 3 independent experiments. Scale bar, 200 µm. (D) Effect of LSD1 knockdown on galectin-3-integrin β3-RalBP1 pathway in A549 cells. Immunoblot analysis of whole cell lysates from monolayer and 3D tumorsphere culture was done. Data are representative of at least two independent experiments. (E) Effect of stable LSD1 knockdown on integrin β3 protein expression in H460 (KD21, KD22) and H1975 (KD2) cells in comparison to the corresponding shGFP control cell line. Immunoblot analysis of whole cell lysates was done. Data are representative of at least three independent experiments. (F) Effect of LSD1 knockdown on NF-κB transcriptional activity measured by luciferase assay. Data are expressed in relative luciferase units (RLU). The bar graph represents the mean ± SEM for n = 3. The P-value was estimated using Student's t-test in D and F. *P < 0.05, **P < 0.01, ***P < 0.001.
in combination with anti-proliferative cancer drugs might be a rational strategy for the treatment of lung adenocarcinoma to suppress tumor proliferation while preventing metastatic phenotype.

Methods
Patients' tissue specimens and TMA preparation. Lung adenocarcinoma tissue specimens were collected from the archive of the Institute of Pathology, University Hospital Cologne, Germany. Formalin-fixed, paraffin-embedded tissues of 182 lung adenocarcinomas were used to prepare tissue microarrays as previously described 36 . Clinicopathological parameters were obtained from patient records. In addition, the KRAS and EGFR mutation status of tumor specimens was determined by routine diagnostic processing using next generation sequencing as described previously 37 . The characteristics of the cohort are summarized in Table 1. The study was approved by the ethics committee (KEK Nr. 200/2014) and informed consent was obtained from all patients. All methods were performed in accordance with the relevant guidelines and regulations of the institution.
Genetic knockdown and LSD1 overexpression. A549, H460, H1975, lung cancer cell lines were transfected with GFP shRNA control (shGFP) or LSD1 shRNA using a pSUPER retroviral system. Using the puromycin resistance, cells were screened for recombination and from single cells monoclonal cell populations were established. After checking the LSD1 knockdown, A549, H460, and H1975 monoclonal cell cultures expressing shLSD1 or shGFP were used only in passages 5 to 8 to ensure stability of the cell culture system. Therefore, not all KD clones could be used in the study, but from A549 cell line the KD clones 9 and 15, from the H460 the clones KD21 and 22, and from H1975 clone KD2 were applied to LSD1 functional analysis.
RNA sequencing and data analysis. Expression levels of LSD1 were determined by referring to the RNA-seq data of lung ACs (n = 40) 19,20 , SQs (n = 9) 19 , pulmonary carcinoids (n = 69) 21 , and SCLC (n = 81) 3 .  Expression values were determined with Cufflinks and represented as fragments per kilobase of exon per million fragments mapped (FPKM).
For transcriptome sequencing, total RNA from three biological replicates of each A549 shLSD1 KD15 cells and A549 shGFP cells was isolated using the RNeasyMini kit (Qiagen Inc, Hilden, Germany). The Illumina TruSeq RNA Sample Prep Kit (Cat#FC-122-1001, Illumina, San Diego, USA) was used with 1 µg of total RNA for the construction of sequencing libraries. RNA libraries were prepared for sequencing using standard Illumina protocols. Fastq files were obtained after demultiplexing using Illuminas CASAVA v1.8.2 pipeline with default parameters. Reads were mapped against the human genome GRCh37/hg19 using BWA v0.5.9-r16 with default parameters. Exon read coverages were obtained with coverageBed (Bedtools-v2.17.0) using exon coordinates from Ensembl database v73. Subsequent analyses were done using the edgeR package in R 38 . Expression changes were calculated as log2 ratios and p-values were adjusted for multiple testing by controlling the Benjamini-Hochberg false discovery rate at 5%. Ingenuity (Qiagen Inc) pathway analysis was used to find enriched canonical pathways, and Upstream Regulators. Furthermore, the database for annotation, visualization and integrated discovery (DAVID) was used to identify the enriched gene ontology terms 39 . Proliferation and transwell invasion assay. Cell proliferation assays were carried out as described previously 13 . Briefly, cells were seeded at a density of 2,500 cells per well in 96 well microplates and cultured in standard medium. Viable cell numbers were determined 0, 1, 2, 3 and 6 days after cell plating using CellTiter 96 Aqueous Cell Proliferation assay kit (Promega, Madison, USA). For invasion assay, Boyden Chambers (Cell Biolabs, San Diego, USA) were layered with 2 mg/ml extracellular matrix gel from Engelbreth-Holm-Swarm mouse sarcoma (Sigma-Aldrich, St. Louis, USA) overnight at 37 °C. Cells were seeded at a density of 100,000 per well. Wells were then filled with standard medium. After 48 h, Boyden chambers were washed twice with PBS, followed with fixation of the cells using 4% formaldehyde (Sigma Aldrich). Subsequently, cells were stained with 1% Crystal Violet (Sigma Aldrich), photographed and counted.
Tumor-sphere cell culture. Tumorsphere assays were carried out as described by Seguin et al. 7 . Single cells were seeded on ultralow attachment plates (Corning, New York, USA) at a concentration of 2,000 cells/ml in DMEM/F12 medium supplemented with insulin-transferrin-selenium, 50 ng/ml EGF and 20 ng/ml bFGF (Invitrogen). Tumorspheres larger than 50 μm in diameter were counted 7 days after seeding. To analyze erotinib resistance in A549 cells, cells were seeded at a density of 3,000 cells per well into ultralow adherent 96-well plates (Corning). The cells were then treated with 1 µL DMSO and erlotinib (1 µM-20 µM) (Sigma-Aldrich) for three days. Viable cell numbers were determined using the CellTiter 96 Aqueous Cell Proliferation assay kit (Promega).
Luciferase assay. Cells were transfected with NF-kB luciferase reporter vector (pNF-kB-luc, Clontech, Mountain View, USA) together with pRL Renilla luciferase control reporter vector (pRL-null, Promega) using lipofectamine (Invitrogen) according to the recommendation of the manufacturer. After 24 h incubation, cells were transferred into the ultra-low adherent well plates (Corning). Next day, cells were treated with 2.5 µg/ml human recombinant galectin-3 (PepproTech, Rocky Hill, USA) for 4 hours. Luciferase assays were then carried out using the Dual Luciferase Reporter assay kit (Promega) according to the manufacturer's protocol. Each assay consisted of three or four replicates and each experiment was repeated at least three times. Data were presented as relative luciferase units (RLU) normalized to the Renilla luciferase signal.
Quantitative real time-PCR. Complementary DNA was synthesized using the AB Reverse Transcription kit using random primers (Applied Biosystems, Foster City, USA) and quantitative PCR was carried out on real-time PCR machine (BioRad, Hercules, USA) with SYBR reagent (Promega). Expression values were normalized to the mean of HPRT1 and the 2 −ΔΔCT method was applied to calculate relative gene expression levels. A list of primers used for qRT-PCR validation is available in Supplementary Table S4. Statistical analysis. Statistical analysis for immunohistochemistry was performed using SPSS 17.0 program (SPSS) and Chi-Square tests were used to calculate the statistical significance for linear-by-linear association. Mann-Whitney U test or Student's T-tests were used to calculate statistical significance using GraphPad Prism software. A P-value less than 0.05 was considered to be significant. *P < 0.05, **P < 0.01, ***P < 0.001. Accession numbers. The RNA-seq data of A549 cells has been deposited in Gene Expression Omnibus under accession code GSE86874.