The LPA1/ZEB1/miR-21-activation pathway regulates metastasis in basal breast cancer.

Lysophosphatidic acid (LPA) is a bioactive lipid promoting cancer metastasis. LPA activates a series of six G protein-coupled receptors (LPA1-6). While blockage of LPA1in vivo inhibits breast carcinoma metastasis, down-stream genes mediating LPA-induced metastasis have not been yet identified. Herein we showed by analyzing publicly available expression data from 1488 human primary breast tumors that the gene encoding the transcription factor ZEB1 was the most correlated with LPAR1 encoding LPA1. This correlation was most prominent in basal primary breast carcinomas and restricted to cell lines of basal subtypes. Functional experiments in three different basal cell lines revealed that LPA-induced ZEB1 expression was regulated by the LPA1/Phosphatidylinositol-3-Kinase (Pi3K) axis. DNA microarray and real-time PCR analyses further demonstrated that LPA up-regulated the oncomiR miR-21 through an LPA1/Pi3K/ZEB1-dependent mechanism. Strikingly, treatment with a mirVana miR-21 inhibitor, or silencing LPA1 or ZEB1 completely blocked LPA-induced cell migration in vitro, invasion and tumor cell bone colonization in vivo, which can be restored with a mirVana miR-21 mimic. Finally, high LPAR1 expression in basal breast tumors predicted worse lung-metastasis-free survival. Collectively, our results elucidate a new molecular pathway driving LPA-induced metastasis, thus underscoring the therapeutic potential of targeting LPA1 in patients with basal breast carcinomas.


INTRODUCTION
Lysophosphatidic acid (LPA) is a naturally occurring bioactive lipid promoting cancer progression and metastasis [1,2]. LPA exhibits growth-factor-like activities by mobilizing multiple intracellular signaling pathways down-stream a series of six G protein-coupled receptors (LPA [1][2][3][4][5][6] ) that mediate proliferation, motility, and survival of both normal and cancer cells [1,2]. Pharmacological blockage of LPA receptors with the pan-LPA antagonist BrP-LPA inhibits human breast MDA-MB-231 cancer cell motility in vitro and promotes tumor regression associated with a decrease in blood vessel density surrounding the tumors in a mouse xenograft model, suggesting that LPA signaling may be a potential therapeutic target for patients with breast cancers [3]. Among the LPA receptors, LPA 1 is up regulated in many types of primary tumors, and plays an important role in regulating cancer malignancy due to pro-oncogenic and pro-metastatic properties [4]. Studies showed that induction of LPA 1 expression induces metastasis in breast and ovarian cancer cells [5,6] and stimulates in the motility of human pancreatic cancer cells [7]. Using immune compromised mice we demonstrated that expression of this receptor confers a high propensity to human breast cancer cells to induce bone metastasis [8]. Stable knockdown of LPA 1 expression (using small-hairpin RNAs or treatments with the LPA 1-3 antagonist Ki16425) inhibited the progression of osteolytic bone metastases by impairing breast cancer cell proliferation, motility and cytokine secretions [8]. Pharmacological inhibition of LPA 1 with Debio-0719 inhibits spontaneous metastatic dissemination of breast cancer cells independently of www.impactjournals.com/oncotarget primary tumor proliferation and angiogenesis [9]. In a later stage of cancer progression Debio-0719 induces breast metastasis dormancy in a murine model [10]. Therefore, LPA 1 is an attractive target. However the underlying mechanism and the genes involved in LPA 1 -mediated breast cancer metastatic dissemination remains to be determined.
Breast carcinomas are currently classified in four molecular subtypes (luminal A, luminal B, HER2-enriched, basal-like) based on gene expression patterns [11]. The basal-type subgroup also refers as triple negative breast cancers (TNBC). Kennecke and colleagues in their seminal work determined that TNBC metastasize predominantly to the brain and lungs [12]. TNBC constitute 10%-20% of all breast cancers [13] and 70% of women with metastatic TNBC have a survival rate of less than 5 years [14]. Absence of estrogen receptor, progesterone receptor, and HER2/neu makes TNBC insensitive to some of the most effective therapies available for breast cancer treatment including HER2-directed therapy and endocrine therapies urging the need for identification of new molecular targets in TNBC.
In this study we investigated the activated genes involved in LPA-mediated metastatic dissemination of breast cancers. Using both genetic and pharmacological approaches in vitro and in vivo we demonstrate that the pro-invasive activity of LPA in TNBC cells depends on the LPA 1 /Pi3K/ZEB1/miR-21 activation cascade. These findings identify LPA 1 as a potential therapeutic target for patients with triple negative breast cancers.

LPAR1 expression correlates with ZEB1 in basal breast cancers
In order to identify the genes correlated with LPAR1, we screened nine publicly available breast tumor databases (GSE2109; GSE5460; GSE1456; GSE2034; GSE12276; GSE3494; GSE2603; GSE7390; GSE16391) comprising of a total of 1488 patient samples. We found ZEB1 to be the most correlated gene to LPAR1 (Table 1). To further explore whether the correlation between LPAR1 and ZEB1 was specific to a particular subtype of breast cancer, we sub-grouped patient samples from 3 publicly available databases of primary breast tumor (GSE20685; GSE21653; GSE1456) into basal (n = 75) and non basal subtypes (n = 138) based on the ER, PR and HER2 receptor expression status. We found that the correlation between LPAR1 and ZEB1 was stronger in the basal subtype (R Spearma n = 0.59) of human primary breast tumors than the non-basal subtype (R Spearma n = 0.40) (Figures  1a and 1b). We then extracted the LPAR1 and ZEB1 expression data from 51 human breast cancer cell lines from the publically available database GSE12777, classified into basal (n = 24) and non basal subtypes (n = 27). We observed that 54.16% of the basal breast cancer cell lines expressed LPAR1 and 37.5% expressed ZEB1 above the overall means. In contrast, only 11.1% of the non basal breast cancer cell lines expressed LPAR1 and none expressed ZEB1 above the corresponding overall means (Figures 1c and 1d). Scatter plot analysis (Figures 1e and 1f) for LPAR1 and ZEB1 correlation in the non basal and basal breast cancer cell lines highlights the fact that LPAR1 and ZEB1 expression correlates significantly in the basal but not in the non-basal subtypes. We did not observe any positive correlation between LPAR2 and LPAR3 to ZEB1 in either subtype (Supplementary Figures 1a, 1b, 1c, 1d, 1e and 1f).

Functional blockade of LPA 1 inhibits ZEB1 expression in vitro and in vivo
We chose the human basal breast cancer cell lines MDA-MB-231, MDA-BO2 and Hs578T to analyze the role of LPA and LPA 1 on ZEB1 expression in vitro and in vivo. Stimulation with LPA caused significant up-regulation of ZEB1 mRNA and protein levels in all three cell lines (Figures 2a). LPA receptor expression screening showed that LPA 1 was predominantly expressed in all the three cell lines (Figure 2b). While LPA 2 was also expressed in MDA-MB-231 cells, all three cell lines had very low or undetectable levels of LPA 3-6 ( Figure 2b). To further validate the relationship between LPA 1 and ZEB1 we used MDA-MB-231 cells, as they co-express both LPA 1 and LPA 2 . Blocking LPA 1 expression by transfecting these cells with siRNA specifically directed against LPA 1 (Supplementary Figure 2a) as well as by treatment with the pharmacological antagonist of LPA 1/3 , Ki16425 [15] inhibited the LPA-induced ZEB1 expression at both mRNA and protein levels (Figures 2c and 2d).
To assess the effect of antagonizing LPA 1 on ZEB1 expression in vivo, we used an animal model of preestablished bone metastases caused by MDA-MB/B02.Luc cells. A short treatment period (daily administration for 5 days) of Ki16425 from day 18 post-tumor cell injection (Figure 2f), had no effect on the extent of bone metastasis lesions detected by BLI ( Figure 2g) and X-ray (Figure 2h) analyses. Immunohistochemistry analyses performed on the bone sections from tumor bearing hind limbs revealed no significant difference of Ki67 staining in Ki16425-treated group compared to vehicle-treated group, indicating that the agent had no effect on cell proliferation. In contrast, the signal intensity of ZEB1 staining in the Ki16425 group was markedly lower than that of the vehicle group (Figure 2i), suggesting that inhibition of LPA 1 activity in vivo inhibits ZEB1 expression at the site of bone metastases.

LPA up regulates microRNA miR-21 through a LPA 1 -dependent mechanism
In order to further identify the genes involved in the metastatic properties of LPA, we carried out a microarray profiling of miRNAs in MDA-MB-231 cells in response to LPA stimulation (GSE64100). Micro-RNA miR-21 was found to be one of the most up-regulated miRNAs in these cells following LPA stimulation ( Figure 3a). MicroRNA miR-21 is known to act as an oncomiR, promoting metastasis in a large number of cancer types [16,17]. Using the TaqMan RT-QPCR system, we confirmed the upregulation of miR-21 by LPA in MDA-MB-231 cells and determined that LPA induced a dose-dependent increase in miR-21 expression by three to four folds after 45 min exposure (EC50 = 0.1 μM) that is sustained for at least 24 h (Figure 3b). In subsequent in vitro studies, cells were treated with LPA (10 μM) for 45min to reflect early exposure times of cells to LPA during the metastatic process. First, we observed that up regulation of miR-21 by LPA was a common feature between all the three basal human breast cancer cell lines, MDA-MB-231, MDA-B02 and Hs578T, (  (Figure 3e). LPA is a constituent of the serum [18] that mediates tumor cells proliferation and invasion [8]. We found that Ki16425 treatment also inhibited miR-21 expression induced by serum in MDA-MB-231 cells (Figure 3f). To identify the signaling pathways regulating the LPAR1/miR-21 axis we used the same strategy as previously by treating the MDA-MB-231 cells with wortmannin or PD98059. Again, only wortmannin was able to inhibit LPA-induced miR-21 expression (Figure 3g), suggesting that LPA up-regulated miR-21 through an LPA 1 /PI3K-dependent mechanism.

LPA-induced ZEB1 expression up regulates miR-21 in human primary breast tumors
As LPA mediated up regulation of both ZEB1 and miR-21 through a common LPA 1 /PI3K axis, and because the strong correlation between ZEB1 and LPAR1, we examined the correlation between LPAR1, ZEB1 and miR-21 in human primary breast tumors. LPAR1/miR-21, ZEB1/miR-21 and LPAR1/ZEB1 pairwise correlations were determined using data from in three publically available human breast cancer databases containing both mRNA and microRNA data (GSE5460; GSE16391; GSE12276). We found that ZEB1/miR-21, LPAR1/ZEB1 and LPAR1/miR-21 correlated significantly in all three databases (Figures 4a-4c).
To examine the direct link between LPA, ZEB1 and miR-21 expression, MDA-MB-231 cells were transfected with synthetic siRNAs against ZEB1 ( Figure 4d). Silencing ZEB1 caused complete inhibition of The table shows the top 50 highest positively correlated genes to LPAR1 (204036_at) across 9 publically available breast tumor databases (GSE2109; GSE5460; GSE1456; GSE2034; GSE12276; GSE3494; GSE2603; GSE7390; GSE16391) comprising a total of 1488 samples. The list of correlated genes to LPAR1 was extracted using R2 genomics analysis and visualization platform.
LPA-induced miR-21 expression (Figure 4e) resulting in a significant increase in the mRNA levels of known miR-21 target genes PTEN, PDCD4 and SPRY2 (Figure 4f), suggesting that LPA might functionally impact on miR-21 activity through ZEB1. LPA is known for activating different transcription factors, including cFos and STAT3, in different cell lines [19,20], and this was confirmed in MDA-MB-231 and MDA-B02 cells (Figure 5a and 5b). Because these factors also are regulators of miR-21 expression we evaluated whether cFos and STAT3 could mediate LPA-induced miR-21 expression. Silencing cFOS or STAT3 using specific siRNAs in MDA-MB-231 cells (Figures 5c and 5d) did not alter LPA-induced miR-21 expression (Figures 5e and 5f).
Altogether these results indicate that LPA induces miR-21 expression through an LPA 1 /PI3K/ZEB1dependent pathway. gene correlation were constructed using data combined from three publically available databases (GSE20685, GSE21653, and GSE1456) of human primary breast tumors of a. non basal subtype (n = 138; r spearma n = 0.40) and b. basal subtype (n = 75, r spearma n = 0.59). c. LPAR1 and d) ZEB1 expression data from 51 breast cancer cell was extracted using BIOGPS online tool from GSE12777 data set, subclassified into non basal (n = 27) and basal (n = 24) subtypes. The scatter plots are also shown for the correlation between LPAR1 and ZEB1 for both the e) non basal (n = 27; r spearman = −0.06) and f. basal (n = 24; r spearma n = 0.43) subtypes of human breast cancer cell lines.

LPA 1 and ZEB1 mediate LPA-induced basal breast cancer migration and invasion in a miR-21-dependent manner
To examine the functional impact of the LPA 1 / ZEB1/miR-21 pathway on the metastatic activity of LPA on basal breast cancer cells we carried out wound healing migration assays and matrigel invasion assays. Transfecting MDA-MB-231 and Hs578T cells with antimiR-21, SiLPA1 or SiZEB1 alone impaired LPAinduced cell migration (Figure 6a, 6c, 6e) and invasion (Figure 6b, 6d, 6f) as compared to the cells transfected with the Negative Control (antimiR control) or BlockiT (SiRNA control). Remarkably, in all cases, impaired motility and invasion was rescued when cells were cotransfected with a mirVana miR-21 mimic. This suggested that the role of LPA in these metastatic processes was mediated by miR-21 through the expression of ZEB1, down-stream activation of LPA 1 .

Basal breast cancer cell bone colonization depends on miR-21 downstream ZEB1 activation
Micrometastasis formation is a well-recognized poor prognostic marker in breast cancers [21]. Patients with tripe negative metastasize to bone and with a higher incidence to lungs and brain [12]. To examine the role of miR-21 during the early phase of breast cancer cell metastasis we used our mouse model of tumor cell colonization to bone (TCB) in which MDA-BO2.luc cells transfected with AntimiR-21 or Negative control were injected intravenously BALB/c nude mice. Cancer cells that colonized the bone marrow cavity were collected seven days post injection and expanded in vitro for two weeks in the presence of puromycin that kills non-tumoral cells. Counting of tumor clones revealed that silencing miR-21 decreased TCB clone number by 70% compared to control (p < 0.001). This result indicated that miR-21 drives basal breast cancer cell metastasis to bone (Figure 7a). We next addressed the functional relationship between miR-21 and ZEB1 in vivo using the TCB mouse model. Mice were injected with MDA-BO2. luc cells transfected with BlockiT (control SiRNA) or SiZEB1+BlockiT or SiZEB1+miR-21 mimic. Silencing ZEB1 significantly impaired the TCB, as judged by a significantly decrease by 88% (p < 0.001) in the number of TCB clones compared to the control group. Interestingly, the inhibition in the TCB induced by silencing ZEB1 was abolished when the cells were previously co-transfected with miR-21 mimic. The number of TCB clones in the groups of mice injected with breast cancer cells transfected with SiZEB1+miR-21 mimic was not significantly different to those obtained with cells transfected with BlockiT ( Figure 7b). These results indicated that ZEB1 regulates the TCB in a miR-21-dependent manner.
Finally we analyzed whether LPAR1 levels might have an impact on metastasis recurrence in basal breast cancers. Among several publically available databases with representative breast cancer patient populations subclassified in basal and non-basal tumors we used the GEO database GSE2603 that additionally provided clinical recording of different tissue metastases. As opposed to records of recurrence data to the lungs the number of bone metastasis events in these populations was too low for developing a comprehensive statistical analysis. However, using the median value of LPAR1 level as cut-off in each population, the predictive value for lung metastasisfree survival of LPAR1 levels was significantly different comparing the group of patients with basal tumors versus those with non-basal tumors. Patients with basal tumors but not with non-basal tumors expressing high LPAR1 levels had a worse lung-metastasis-free survival (P < 0.05) (Figure 8a). These results suggested that LPAR1 level could be considered as a new prognostic factor for patients with basal breast cancers.

DISCUSSION
LPA exhibits complex mode of actions, due to multigene programs activated downstream of multiple LPA receptor subtypes that are frequently co-expressed in cells and tissues and having potential synergistic and opposite functions [1]. Integrating genetic and pharmacological approaches in vitro and in vivo we demonstrate that the proinvasive activity of LPA on triple negative breast cancer cells depends on an LPA 1 /PI3K/ZEB1/miR-21 activation cascade.
Overexpression of other LPA receptors, LPA 2 and LPA 3 , has been previously linked to the progression of breast cancers [4]. However, activated genes driving the oncogenic potential of these receptors have not been defined. As opposed to LPAR1, we did not observe significant correlations between ZEB1 and LPAR2 or LPAR3 in publicly available breast cancer cell line databases. The strong correlation between ZEB1 and LPAR1 was most prominent in primary tumors of patients with basal breast cancers. Moreover, in addition to a higher prevalence of LPA 1 expression in basal than in non-basal breast cancer cell lines, there was a significant correlation between ZEB1 and LPAR1 in cell line of basal subtypes but not of non-basal subtype. This suggests that controlling ZEB1 expression through LPA 1 activation might have a functional impact on basal breast cancer behaviors. ZEB1 is a known driver of the epithelial-to-mesenchymal transition (EMT), conferring metastasis and drug resistance properties to epithelial cells and contributing to the poor clinical outcomes [22]. Therefore, mediators of ZEB1 might represent potential therapeutic targets for metastasis suppression. In vitro, inhibition of LPA 1 expression using SiLPA 1 , or blocking its activity with the LPA 1/3 antagonist Ki16425, significantly inhibited LPA-induced expression of ZEB1. Functional involvement of LPA 1 in ZEB1 expression in vivo was supported by the effect of a short treatment with Ki16425 (5 days) of animals with pre-established skeletal metastases that markedly reduced ZEB1 expression in the tumor metastasis.
LPA 1 is known to be responsible for LPA-induced cell mobility in both neoplastic and non-neoplastic cells [23]. Here, we demonstrate that specific silencing of LPA 1 and ZEB1 inhibited the LPA-induced migration and invasion of basal breast cancer cell lines, supporting that ZEB1 is a downstream activated transcription factor of the LPA/ LPA 1 axis stimulating cell motility. Silencing ZEB1 leads to inhibition of bladder cancer cell migration and invasion [24]. In bladder cancer cells ZEB1 regulates vimentin, MMP2 and cytokeratins [24]. Interestingly, MMP2 gene expression correlates to LPAR1 in human primary breast tumors (hit 23 in Table 1). Moreover, we recently identified a set of 74 early genes including VIM (the gene encoding vimentin) that are up-regulated in a LPA 1 -dependent manner in breast and prostate cancer cells stimulated with LPA [25] (GEO Series accession number GSE56265). In addition, LPA-induced ZEB1 expression in basal breast cancer cells was mediated through a PI3K-dependent but not of a MEK1-dependent signaling pathway. In bladder cancer cells LY294002, a PI3K inhibitor, inhibits phosphorylation of AKT and GSK3β, and ZEB1 expression due to inhibition of β-catenin/transcription factor 4 (TCF4) complex binding and transcriptional activity on ZEB1 promoter [26]. Intriguingly, our analysis of the nine publically available databases of human primary breast tumors revealed that after ZEB1, TCF4 had the highest Spearman correlation coefficient to LPAR1 (hit 2 in Table 1). Altogether, these data suggest that activation of ZEB1 through the PI3K/AKT pathway might represent a general mechanism in LPA-induced cancer cell invasion and metastasis. In addition, phosphorylation of GSK3β might play a pivotal role in coordinating LPA signals dependent on Pi3K/AKT activation that mediate β-catenin/TCF4-induced ZEB1 expression.
Breast cancer dissemination involves multi-gene programs coordinated by transcription factors. MicroRNAs are post-transcriptional gene expression regulators that are also involved in breast cancer metastasis [27][28][29]. LPA triggers a series of 74 early genes directly through the activation of LPA 1 in breast and prostate cancer cells [25].
Here we showed that exposure to LPA for only 45 min up-regulates 36 microRNAs in MDA-MB-231 cells. We focused our attention on miR-21 because it is known to act as an oncomiR. High miR-21 expression is reported to promote invasion and proliferation of multiple cancer cell types, including breast [30,31] [32][33][34][35]. Using a mouse model we confirmed that miR-21 controls breast cancer cell migration and invasion in vitro, and metastasis using tumor cell colonization of bone as an endpoint in vivo. Moreover, it has been determined that ZEB1 is an upstream regulator of miR-21 [36]. In this study we demonstrated that in human breast tumors there was a significant positive correlation between LPAR1, ZEB1 and miR-21 and that in basal breast cancer cell lines ZEB1 regulates LPA-induced miR-21 expression through an LPA 1 /PI3K dependent mechanism.
We showed previously that blocking LPA 1 activity inhibits early stage of breast cancer cell metastasis to lungs and bone, independently of cell proliferation and angiogenesis but through inhibition of cell motility and invasion [9]. Inhibiting in vitro miR-21 activity with the MirVana miR-21 inhibitor, or silencing LPA 1 or ZEB1 independently resulted in a complete block of LPA-induced cell migration and invasion. In all cases, basal breast cancer cell functions were rescued by cotransfecting the cells with a miR-21 mimic molecule, indicating that LPA-induced basal breast cancer cell migration and invasion was regulated in a miR-21dependent manner, down-stream of LPA 1 and ZEB1 activations. This pathway might have a functional impact on the establishment of metastases as shown by the inhibition of micromedullar metastasis formation with silenced expression of ZEB1 that was rescued following introduction of mirVana miR-21 mimic. Beside a strong significant correlation between high levels of LPAR1 mRNA and the lymph node status from an unselected breast cancer population, we previously found that LPAR1 levels did not predict metastasis recurrence [9]. Recent studies reported that growth of TNBC cells relies upon the pro-inflammatory cytokines IL-6 and IL-8, whose secretion is regulated by the LPA/NF-κB signaling cascade [37]. LPA 1 is a wellknown inducer of IL-6 and IL-8 secretions in ovarian and breast cancer cells [38,39]. Here, based on the subclassification of human breast tumors from publically available databases, we found for the first time that as opposed to non-basal breast cancer patients, expression levels of LPAR1 predicted lung metastasis-free survival in TNBC patients.
Overall our results demonstrate functional relationships between ZEB1 and miR-21 driving LPAdependent metastasis through LPA 1 in basal breast cancers ( Figure 8c). Our findings may have a major impact by identifying a potential target for development of new adjuvant therapies of triple-negative breast cancer patients to prevent metastatic recurrences.

RNA extraction, reverse transcription, and real time RT-PCR
Total RNAs including miRNA were extracted using miRNeasy kit (QIAGEN). TaqMan microRNA assays (Applied Biosystems, Life Technologies) were used to quantify miR-21 and RNU-48 levels according to the manufacturer's recommendations. For gene expression analysis, cDNAs were synthesized using iScript cDNA Synthesis kit (Biorad) according to the manufacturer's protocol and real-time quantitative was performed using the SYBR Green PCR kit (Life technologies) using specific PCR primers (Table 2). Each PCR cycle consisted of 10s of denaturation at 95°C, 15s of annealing at 67°C, and 10s of extension at 72°C. miRNA microarray analysis miRNA from cells were isolated using miRNeasy kit (QIAGEN). Total RNA (100 ng) was directly labeled with Cyanine-3-labeled and array hybridized for 20 h using miRNA complete labeling and hybridization kit (Agilent). The arrays were washed with gene expression wash buffer kit (Agilent) and read with High-resolution microarray scanner (Agilent). The data was extracted using Agilent feature extraction software (Agilent v11.5.1.1). The data were normalized by spike labelling. The median fluorescent intensities were obtained after subtracting background. To identify differential miRNA expression between samples, the median fluorescent intensities were normalized using the median expression values within the array and log2 values analyzed. Data was deposited on Gene expression omnibus, the accession code is GSE64100.

Wound healing and invasion assays
Confluent cell culture monolayers were cells scratched using a 10ul tip. Cells were washed with PBS to remove all the debris and incubated for 24 h (MDA-MB-231 cells) or 48 h (Hs578T cells) with serum-free media supplemented with LPA (1 μM). Images were then captured using NIKON Coolpix 990 camera attached to a microscope and the area of cells migrated were tabulated using Morpho-expert software (Explora-nova). Cancer cell invasion across a Matrigel layer was performed as previously described [41]

Immunohistochemistry
The tumor bearing hind limbs were fixed and embedded in paraffin. Five μm sections were deparaffinized in methylcyclohexan, hydrated through a graded series of ethanol, then immersed in a peroxidase blocking reagent (DakoCytomation) 5 min. Sections were incubated with normal goat serum for 30 min and incubated overnight at 4°C in humid chambers with primary antibody to Ki67 (dilution1: 25) or to human ZEB1 (Abnova, PAB19268, dilution 1:300). The slides were incubated with biotinylated polyclonal rabbit anti-rat immunoglobulin (DakoCytomation) for 45 min. After washing, the slides were treated with peroxidase-conjugated streptavidin (DakoCytomation) for 45 min and developed by addition of a solution of 3, 39-diaminobenzidine tetrahydrochloride (DakoCytomation). Light counterstaining was performed with Mayer's hematoxylin.

Animal experiments
Mice used in experimental procedures at the Université Claude Bernard Lyon1 (Lyon, France) were handled according to the rules of Décret N° 87-848 du 19/10/1987, Paris. The experimental protocol have been reviewed and approved by the Institutional Animal Care  France). BALB/C nude mice purchased from Janvier Labs (Le Genest-Saint-Isle, France), 4 weeks of age, were housed under barrier conditions in laminar flow isolated hoods. Autoclaved water and mouse chow were provided ad libitum. Animals were carefully monitored for established signs of distress and discomfort and were euthanized when these were confirmed. The institutional committee approving the experiments, including details regarding animal welfare are indicated in Supplementary Information 1. For the breast cancer cell colonization to bone (TCB) experiments, transfected MDA-B02.Luc cells were injected intravenously to BALB/c nude mice as previously described [42]. Sevendays post injection, mice were sacrificed and hind limbs were collected, dissected. Bones were minced and treated with 300μL of gentle collagenase/hyaluronidase (Stem cell technologies) for 2 h at 37°C. The cell suspension was washed with phosphate buffer saline (PBS), plated in wells of a 6 well plate in complete culture media supplemented with puromycin (1 μg/mL). After 2 weeks clones of tumor cells that had colonized bone (TCB) were fixed and stained in 50% (v/v) methanol containing 0.05% (w/v) crystal violet and enumerated. For experiments on established osteolytic lesions, animals were injected with MDA-B02.Luc cells as previously described [40] and evidences of metastasis formation were determined by bioluminescence imaging (Nightowl, Berthold, Germany) and X-ray analyses (MX-20; Faxitron X-ray corporation). Metastatic animals were then treated once daily for 5 days with 20 mg/kg Ki16425 or the vehicle.

Microarray correlation analysis
To identify genes correlated with LPAR1 we calculated the coefficient of correlation values of LPAR1 for all genes covered by the given microarray platform across 9 publically available breast tumor datasets (Gene Expression Omnibus (GEO) accession numbers: GSE2109, GSE5460, GSE1456, GSE2034, GSE12276, GSE3494, GSE2603, GSE7390, GSE16391) using R2 genomics analysis and visualization platform. The list of genes in the descending order of coefficient of correlation was generated after averaging the values across all 9 databases. Publicly available gene expression data obtained from three independent studies comprising in total 138 luminal and 75 basal breast carcinoma samples were retrieved from the GEO (accession numbers: GSE20685, GSE21653, and GSE1456). Microarray data were generated on Affymetrix HG-U133Plus2.0 or Affymetrix HGU133A arrays, respectively, and were simultaneously normalized by robust multi-array average (RMA) using custom brainarray CDF files (v17 ENTREZG) yielding one optimized probeset per gene as previously described [43]. LPAR1 and ZEB1 expression data from 51 breast cancer cell lines was extracted using BIOGPS online tool from GSE12777 data set. The cell lines were classified as luminal (n = 27) and basal (n = 24) subtypes.

Statistical analysis
Analysis of correlation and computation of linear regression of the data were performed using GraphPad Prism v5.0c software. Differences between groups were determined by 1-way ANOVA followed by Bonferroni post-test. Single comparisons were carried out using nonparametric Mann-Whitney test. Kaplan-Meier analysis for 5 year lung metastases-free survival of 25 patients with basal (triple negative) breast cancer was performed using the GSE2603 dataset. Microarray transcriptomic data was normalized by robust multi-array average (RMA) using R-package yielding one optimized probeset per gene. Patients were divided in 2 groups based on the median of the expression of LPAR1. Gehan-Breslow-Wilcoxon test was performed to calculate the p value. p < 0.05 was considered statistically significant.