Oncogene (2014) 33, 5090–5099 & 2014 Macmillan Publishers Limited All rights reserved 0950-9232/14 www.nature.com/onc

ORIGINAL ARTICLE A signature of bone metastatic colonization sensitizes for tumor-induced osteolysis and predicts survival in lung cancer

D Luis-Ravelo1, I Anto´n1, C Zandueta1, K Valencia1, C Ormaza´bal1, S Martı´nez-Canarias1, E Guruceaga2, N Perurena1, S Vicent1,3, J De Las Rivas4 and F Lecanda1

Bone metastasis of lung adenocarcinoma (AC) is a frequent complication of advanced disease. The purpose of this study was to identify key mediators conferring robust prometastatic activity with clinical significance. We isolated highly metastatic subpopulations (HMS) using a previously described in vivo model of lung AC bone metastasis. We performed transcriptomic profiling of HMS and stringent bioinformatics filtering. Functional validation was assessed by overexpression and lentiviral silencing of single, double and triple combination in vivo and in vitro. We identified HDAC4, PITX1 and ROBO1 that decreased bone metastatic ability after their simultaneous abrogation. These effects were solely linked to defects in osseous colonization. The molecular mechanisms related to bone colonization were mediated by non-cell autonomous effects that include the following: (1) a marked decrease in osteoclastogenic activity in vitro and in vivo, an effect associated with reduced pro-osteoclastogenic cytokines IL-11 and PTHrP expression levels, as well as decreased in vitro expression of stromal in conditions mimicking tumor–stromal interactions; (2) an abrogated response to TGF-b signaling by decreased phosphorylation and levels of Smad2/3 in tumor cells and (3) an impaired metalloproteolytic activity in vitro. Interestingly, coexpression of HDAC4 and PITX1 conferred high prometastatic activity in vivo. Further, levels of both correlated with patients at higher risk of metastasis in a clinical lung AC data set and with a poorer clinical outcome. These findings provide functional and clinical evidence that this metastatic subset is an important determinant of osseous colonization. These data suggest novel therapeutic targets to effectively block lung AC bone metastasis.

Oncogene (2014) 33, 5090–5099; doi:10.1038/onc.2013.440; published online 28 October 2013 Keywords: HDAC4; PITX1; ROBO1; microenvironment; osteoclastogenesis; MMP; outcome

INTRODUCTION tumor secretion of MMP and other exacerbated at the Bone metastasis is a frequent complication of non-small cell lung tumor–stromal interphase.9 cancer associated with dismal prognosis.1–4 As many as 30–40% In lung cancer bone metastases, several critical mediators of the of lung cancer patients experience significant morbidity as a metastatic process required for one or several discrete steps have result of skeletal related events, including pain, spinal cord been characterized. For instance, receptor DDR1, associated compression associated with pathological fractures and other with poor prognosis in lung cancer patients, enhanced tumor cell metabolic disorders that entail a significant reduction in life survival and mediated cell engagement and colonization in the expectancy.4,5 bone.10 More recently, receptor of activated C (EPCR) Dissemination of tumor cells requires the acquisition of cell promoted tumor cell survival and was correlated with the clinical functions such as invasion and motility, survival in circulation and outcome in lung adenocarcinoma (AC) patients.11 Genetic silencing the ability to arrest in the osseous milieu and form secondary or antibody blockade of EPCR led to strong abrogation of bone outgrowths.6 The completion of each step involves the collective metastasis. Similarly, activation of signaling pathways in tumor cells activity of a complex multigenic program that also allows regulating large downstream gene networks has been shown to adaptation to the stringent conditions imposed in the target exert potent activity driving metastasis. Hyperactive WNT/TCF organ. Besides intrinsic tumor functions, this multifaceted process signaling mediates lung AC metastasis to the bone and brain.12 is heavily modulated by biological and mechanical factors derived Similarly, perturbation of critical pathways in the host bone milieu, from the selective pressure exerted by the microenvironment. At such as PDGFR/VEGFR axis, severely impairs bone metastatic the osseous site, colonization by malignant cells proceeds by the colonization.13 In all cases, inhibition of these molecular targets in tumor-induced secretion of osteoclastogenic factors such as IL-11, preclinical animal models leads to a substantial reduction of osseous IL-8, PTHrP that stimulate osteoclast-mediated bone resorption.7 metastasis. Yet, metastases proceed inexorably indicating that crucial In turn, this process liberates active TGF-b from the bone matrix, gene regulators at the crossroads of multiple critical signaling which fuels tumor cells to perpetuate the secretion of osteoclast- pathways might be targeted to attain more salient antimetastatic stimulating factors, thus instituting a self-feeding cycle implicated effects. Identification of critical components modulating large in destructive bone lesions.8 In addition, deleterious proteolytic regulatory effectors could be of great biological relevance and degradation of bone matrix also occurs through the concomitant potential therapeutic interest, but so far few have been identified.

1Division of Oncology, Adhesion and Metastasis Laboratory, University of Navarra, Pamplona, Spain; 2Bioinformatics and Proteomics Unit, Center for Applied Medical Research (CIMA), University of Navarra, Pamplona, Spain; 3Department of Histology and Pathology, School of Medicine, University of Navarra, Pamplona, Spain and 4Bioinformatics and Functional Genomics Research Group, Cancer Research Center, University of Salamanca (CSIC/USAL), Salamanca, Spain. Correspondence: Dr F Lecanda, Center for Applied Medical Research (CIMA), University of Navarra, 31080 Pamplona, Spain. E-mail: fl[email protected] Received 17 April 2013; revised 4 September 2013; accepted 9 September 2013; published online 28 October 2013 Antiosteolytic targets D Luis-Ravelo et al 5091 In this study, using comparative transcriptomic analysis in a High levels of ROBO1, HDAC4 and PITX1 characterize the model of lung cancer bone metastasis, we identified a gene phenotype of highly metastatic subpopulations lung AC cells signature of three genes that cooperatively confers high osseous To identify genes involved in lung AC metastasis to bone, we metastatic activity by inducing strong osteoclast and proteolytic performed a transcriptomic analysis comparison between highly activities. Moreover, this signature also sensitizes tumor cells to metastatic subpopulations (HMS) and parental cells. Robust TGF-b-induced response. Consistently, high levels of this triple bioinformatics filtering identified a gene signature of differentially gene subset were associated with poor survival in AC patients. overexpressed genes (Supplementary Table S1). On the basis of These findings provide experimental and clinical evidence of stringent bioinformatics tools9 and biological criteria, we selected critical targets involved in metastasis progression. a subset of genes that included ROBO1 (roundabout, axon guidance receptor, homolog 1), HDAC4 (histone deacetylase 4) and PITX1 (paired-like homeodomain1). HMS expressed high levels of these genes subset as assessed by western blot (Figure 2a) and RESULTS RT–qPCR analysis (Figure 2a and Supplementary Figure S1a). Lack Isolation of highly-metastatic subpopulations of an optimal anti-ROBO1 antibody precluded its analysis by We used a model of human AC A549 cells that displayed high western blot. proclivity to form bone metastasis. After intracardiac (i.c.) To test their functional relevance, we individually depleted the inoculation of luciferase-labeled A549 parental cells into athymic expression of each gene using three specific shRNAs. We selected nude mice, several subpopulations that metastasized to the hind the most efficient shRNA for further experiments. Western blot limbs were isolated, expanded in vitro, and their metastatic analysis revealed abrogation of PITX1 and HDAC4 protein levels, activity was assessed by subsequent i.c. re-inoculation. After whereas a dramatic downregulation of ROBO1 expression levels two in vivo passages, osseous metastatic activity of different was observed by RT–qPCR (Figure 2b). To assess their contribution subpopulations was evaluated using Kaplan–Meier curves. to bone prometastatic activity, we injected single-knockdown (KD) Compared with the parental control cells, several subpopulations cells into the left cardiac ventricle of athymic nude mice. Single KD named M1, M3 and M4 decreased animal survival (Figure 1a). did not affect the time of latency in the appearance of bone Consistent with these results, computerized X-ray image analysis metastatic lesions (Figure 2c). Bioluminescence imaging (BLI) revealed an increased area of bone metastasis in the hind limbs revealed a slight decrease, but not statistically significant, in induced by M1, M3 and M4 subpopulations (Figure 1b). To assess HDAC4 and PITX1 KD cells (Figure 2d). Similarly, a mild decrease in tumor burden, we isolated metastatic cells by bone marrow lesion area of hind limbs was observed, but differences did not ‘flushing’, and conspicuous single-cell-derived colonies from reach statistical significance (Figure 2d). Thus, abrogation of gene each animal were counted. Subpopulations M1, M3 and M4 expression for each single gene did not contribute significantly to induced a more abundant yield of single-cell-derived colonies the prometastatic activity in vivo. than did the parental cell line (Figure 1c). Parental and derived subpopulations displayed similar in vivo growth (Figure 1d), which indicated that the increased metastatic ability observed Double and triple KD decreases prometastatic activity in vivo was due to the acquisition of a bone metastatic Next, we posited that simultaneous inhibition of multiple genes phenotype. could be required to effectively target bone metastasis. Double

120 Parental 25 * ) M1 3 * 100 M3 20 *** M4 80 15 60 40 10 survival (%) 20 *** *** 5 Bone Metastasis -free 0 *** SCDC number (x10 0 5 10 15 20 25 30 35 0 Time (days) Parental M1 M3 M4

Parental M1 M3 M4

450 Parental ) 3 M1 M3 300 5 M4

* 4 * 150 * 3 Tumor volume (mm 0 2 0 10 20 30 40 Time (days) Lesion Area (%) 1

0 Parental M1 M3 M4 Figure 1. Isolation of HMS. (a) Assessment of bone metastasis-free survival in HMS derived from parental A549 cells (M1, M3 and M4, n ¼ 8, log test). HMS were selected by in vivo passage of the cells after i.c. inoculation of parental A549 cells into nude mice. (b) Indirect measurement of tumor burden in hind limbs of animals inoculated with different HMS by X-ray image analysis at day 22. Representative X-ray images of each group are shown on top. Arrows denote osteolytic lesions. (c) Direct measurement of tumor burden by counting ex vivo SCDCs (single-cell-derived colonies) obtained after bone marrow ‘flushing’ at day 30. (d) In vivo subcutaneous growth of different HMSs as compared with control cells (n ¼ 10 tumors each group). Data are presented as mean±s.e.m.

& 2014 Macmillan Publishers Limited Oncogene (2014) 5090 – 5099 Antiosteolytic targets D Luis-Ravelo et al 5092

HDAC4 100 ROBO1 80 10 ROBO1 -actin 60 Parental M1 M3 M4 8 40 HDAC4 6 20 4 PITX1 Fold change (%) 0

Fold change 2 PITX1 -actin 0 M1 M3 M4 -actin Control shROBOshROBO 1 shROBO 2 3

Parental

2.5×108 2.0×108 n.s. 1.5×108 100 1.0×108 flux (p/s) 80 5.0×107 Normalized Photon 0 60 Control HDAC4 PITX1 ROBO1 PITX1 controlHDAC4 ROBO1 shRNA 40 shRNA n.s. Control 20 shHDAC4 6 shPITX1 n.s.

Metastasis-free survival (%) shROBO1 0 4 0 204060 Time (days) 2 Lesion Area (%)

0 Control HDAC4 PITX1 ROBO1 ControlHDAC4 PITX1ROBO1 shRNA shRNA Figure 2. Identification and characterization of single HDAC4, PITX1 and ROBO1 gene subset as potential metastatic genes. (a) Left Panel: western blot analysis of PITX1 and HDAC4 in HMS derived from parental A549 cells. Right Panel: validation of ROBO1 in HMS by RT–qPCR. (b) Stable interference of genes by lentiviral-mediated delivery of three specific shRNAs targeting each gene in A549 cells. Control denotes the lentiviral transduction with a scramble sequence. Left Panel: western blot analysis for HDAC4 and PITX1 in cells lentivirally transduced with three different shRNAs targeting HDAC4 and PITX1, respectively. Right Panel: real-time PCR analysis of ROBO1 levels in cells lentivirally transduced with three different shRNAs targeting ROBO1. (c) Kaplan–Meier curve of metastatis-free survival after intracardiac inoculation of single shRNAs for each gene in athymic nude mice (n ¼ 10 each group) and metastatic activity was quantified. (d) Upper panels: quantification of bioluminescence imaging at day 35. Photonic flux was normalized by subtraction of the non-bearing tumor mice value. Representative photographs. Red and blue colors point out high and low expression, respectively. Bottom panels: X-Rays at the fifth week of the experiment were analyzed (right). Representative radiographic images are shown. Arrows indicate osteolytic lesions. Kruskal–Wallis multiple comparison test plus U de Mann–Whitney was performed. NS, not significant.

and triple KD was achieved by lentiviral shRNA transduction and bottom panel). These results were associated with an overt assessed by western blot analysis for HDAC4 and PITX1 and by decrease in bone colonization of hind limbs assessed by BLI in qPCR for ROBO1. We selected the most efficient shRNA for these mice inoculated with 3KD cells (Figure 3b). Consistently, X-ray experiments. Triple gene abrogation is shown in Supplementary image analysis showed a marked reduction in bone lesions in Figure S1b. We intracardially injected tumor cells with double KD mice injected with 3KD cells as compared with control groups and gene combinations into athymic nude mice. At day 28, BLI revealed preserved cortical and trabecular bone in the metaphy- revealed a decrease in tumor burden in all double gene seal region of these animals (Figure 3c). In contrast, overt combinations as compared with scramble control cells osteolytic lesions were detected in tibiae and femora of control (Supplementary Figure S1c). However, no significant differences mice (Figure 3c). Importantly, all these observations could not be were detected in overall survival (Supplementary Figure S1d). To explained by cell autonomous differential growth kinetics, as 3KD investigate whether triple KD gene combination had a greater cells showed similar in vitro cellular proliferation and clonogenic effect on survival, we intracardially injected triple KD (3KD) cells. ability as compared with control cells (Supplementary Figure S2a). Animals inoculated with 3KD cells showed a significantly longer Similarly, 3KD cells form tumors of similar size as control cells, after latency time for the appearance of metastasis assessed using subcutaneous injection (Supplementary Figure S2b). Kaplan–Meier curves (Figure 3a, upper panel). Notably, control To assess whether the observed low prometastatic activity was animals showed a marked tumor-induced weight loss, whereas caused by a decrease in the ability of cells to infiltrate and survive mice inoculated with 3KD cells showed slight increase in body at early stages of osseous homing, we killed the mice 7 days after weight observed at the end of the experimental period (Figure 3a, i.c. cell inoculation. No obvious differences in bone metastatic

Oncogene (2014) 5090 – 5099 & 2014 Macmillan Publishers Limited Antiosteolytic targets D Luis-Ravelo et al 5093 a b Weeks2 3 4 5

100 Photon flux (x10 ***

1.5×1009 Parental Parental 50 Ctrl Parental 3-shRNA 7.5×1008

Ctrl Ctrl 6 p/sec/cm

Survival (%) Survival 3-shRNA Normalized 07

Photon flux (p/s) flux Photon 5.0×10 ** 0 0 10 20 30 40

0 10 20 30 40 50 60 2 Time (days) /sr) Time (days) 3-shRNA

20 *** c Parental Ctrl 3-shRNA *** 7.5 ** 10 5.0 * 0 X-Rays 2.5 Lesion area (%) area Lesion

Weight gain (%) Weight -10 0.0

al CT Ctrl  -20 Parent ParentalCtrl 3-shRNA 3-shRNA

Figure 3. Metastatic activity of cells with triple stable KD (3-shRNA) for HDAC4, PITX1 and ROBO1 cells. (a) Upper panel: Kaplan–Meier curve of overall survival for parental, Ctrl (control, denotes the lentiviral transduction with 3 different scramble sequences ) and 3-shRNA (triple KD of HDAC4, PITX1 and ROBO1 genes) cells after i.c. inoculation in mice (n ¼ 8 each group) (Log-rank test, ***Po0.001). Bottom panel: weight loss of mice at the end of the experimental period referred as the percentage of gain (Student t-test, ***Po0.001). (b) Left panel: quantification of bioluminescence imaging. Photonic flux was normalized by subtraction of the non-bearing tumor mice value. Right panel: representative bioluminescence images. Data were analysed by Mann–Whitney test, **Po0.01. (c) Left panel: indirect measurement of tumor burden in hind limbs of animals inoculated with triple gene KD and control cells by X-ray image analysis at day 35. Right panel: representative X-ray images and micro-CT scans of each group also are shown. Arrows denote osteolytic lesions (Welch test, **Po0.01).

burden were detected between groups, as assessed by BLI to colonize the bone compartment, whereas striking osteolytic (Supplementary Figure S2c) and bone marrow flushing of hind lesions were detected in tibiae of control mice (Figure 4b). limbs (data not shown), indicating that 3KD did not significantly Osteoclastogenic activity fostered by tumor cells is a major contribute to bone homing metastatic activity. Taken together, mechanism of bone degradation in osseous metastases.4 Thus, we these data indicate that HDAC4, PITX1 and ROBO1 cooperatively studied whether the reduction of osteolytic lesions in our in vivo contribute to the metastatic activity of lung cancer cells by model was related to an impairment of prosteoclastogenic mechanisms independent from cell autonomous proliferation and activity. Consistent with previous findings, analysis of tibiae homing activity. harboring 3KD cells revealed a dramatic decrease in the number As complementary validation of these findings, we simulta- of TRAP þ cells at the tumor–bone interface (Figure 4c). neously overexpressed these pro-metastatic genes in parental To test whether differences in tumor burden were due to A549 cells (Supplementary Figure S3a). Limitations of the changes in cell apoptosis or proliferation within the bone maximum packaging limit of the retroviral system for ROBO1 microenvironment, immunohistochemical staining for Ki-67 and precluded the development of the triple overexpressor. In line cleaved caspase-3 were performed. Few apoptotic cells were with previous findings, i.c. inoculation of double overexpressor detected, and no differences were observed between bone PITX1 and HDAC4 led to a substantial increase in metastatic tumor metastatic tumors derived from 3KD and control cells (data not burden assessed by BLI and the formation of overt osteolytic shown). Interestingly, a low percentage of Ki-67-positive cells was lesions as compared with control mice (Supplementary Figures observed in 3KD tumors (Figure 4d). Taken together, these data S3b and c). Slight decrease in cell growth kinetics of HDAC4/PITX1 suggest that HDAC4, ROBO1 and PITX1 cooperatively contribute cells was detected in vitro (Supplementary Figure S3d). Thus, these to mediate osseous colonization in response to microenviron- findings further support the relevance of HDAC4 and PITX1 genes mental factors acting on tumor cells. as robust prometastatic drivers of bone metastasis in lung AC.

3KD cells regulate non-cell autonomous mechanisms critical for 3KD appeases osseous colonization bone degradation To assess whether prometastatic activity to the bone was To explore whether the mechanism of osteoclast activation was mediated by the ability of cells to colonize the osseous induced by soluble factors released in the tumor microenviron- compartment, tumor cells were injected directly into the tibial ment, we performed an in vitro osteoclastogenesis assay with bone. Three weeks post injection, BLI showed an overt decrease in bone murine macrophages in the presence of conditioned tumor burden of tibiae in animals injected with 3KD cells as medium (CM) of cells in coculture with the bone stromal cell line compared with control cells (Figure 4a). Similarly, assessment of ST-2. Incubation of bone marrow macrophages with CM derived X-ray imaging showed that 3KD cells displayed an impaired ability from 3KD cells in the presence of ST-2 cells led to a decrease in the

& 2014 Macmillan Publishers Limited Oncogene (2014) 5090 – 5099 Antiosteolytic targets D Luis-Ravelo et al 5094

20 150 15 *** *** 8

2.0×10 area

15 * + Ctrl 100 8 10 1.5×10 3-shRNA 10 1.0×108 %TRAP+ 50 5 flux (p/s) 5.0×107 5 Lesion Area (%)

** Tumor-bone Area % Nuclear Ki-67 Normalized photon 0 0 0 0 1 2 3 0 Ctrl 3-shRNA Ctrl 3-shRNA Time (weeks) Ctrl 3-shRNA

Weeks1 2 3 Photon flux(x106 p/sec/cm2/sr) T Ctrl Ctrl

Ctrl B Ctrl

T

B 3-shRNA 3-shRNA 3-shRNA 3-shRNA

Figure 4. Osseous colonization of 3KD cells. (a) Top panel: quantification of bioluminescence imaging of control and 3KD cells inoculated in bone marrow cavity by intratibial injection (n ¼ 10 per group). Photonic flux was normalized by subtraction of the non-bearing tumor mice value. Ctrl, denotes the lentiviral transduction with 3 different scramble sequences. Bottom panel: representative bioluminescence images. (b) Top panel: measurement of tumor burden in hind limbs of animals inoculated with triple-KD and control cells by X-ray image analysis at day 35. Bottom panel: representative images of X-rays, micro-CT and H&E staining. (c) Top panel: area covered by TRAP þ multinucleated cells was quantified by image analysis at tumor–bone interface (Mann-Whithney, *Po0.05). Bottom panel: representative images ( Â 20). Black arrows indicate TRAP þ cells; B, bone; T, tumor. (d) Quantification of Ki-67-positive cells in bone metastatic after immunostaining (Mann– Whithney, ***Po0.001).

size of multinucleated TRAP þ cells, as compared with CM derived shown that MMP-3/MMP-10 activities are rapidly induced upon from control cells (Figure 5a). Together, these findings suggest tumor–stromal interactions.15 Global MMP activity of 3KD cells that the three-gene signature tightly controls non-cell autono- cultured alone was unaffected (data not shown). However, in line mous mechanisms involving osteoclastogenic activity to promote with previous findings, increase in MMP activity using a osteolytic lesions. fluorogenic peptide for a MMP-3 and À 10 in the presence of Next, we investigated the molecular mechanisms associated ST-2 cells was lower in coculture incorporating 3KD cells with the decreased osteoclast formation. To this end, expression (Figure 5d). These data confirmed that triple combination of levels of critical prosteoclastogenic factors such as RANKL, IL11 genes contributes to tumor-stromal associated metalloproteolytic and PTHrP were studied. First, we observed that murine rankl was activity and suggest regulation of non-cell autonomous mechan- expressed by ST-2 stromal cells in coculture with tumor cells, isms by the three-gene signature. despite the lack of rankl expression by ST-2 cells individually cultured (data not shown). Interestingly, 3KD cells incubated with ST-2 cells resulted in a significant attenuation of rankl expression TGF-b signaling is altered in triple gene KD cells by ST-2 cells (Figure 5b, upper panel). Second, 3KD cells expressed As matrix-derived TGF-b has a pivotal role in tumor-induced lower levels of osteoclastogenic factors PTHrP and IL11 in both osteolytic lesions, we explored whether this gene signature was single (Figure 5b, middle and bottom panels) and coculture induced by TGF-b. Incubation of tumor cells with increasing doses conditions (data not shown). More importantly, upon TGF-b of TGF-b did not show significant changes in PITX1 and HDAC4 stimulation, expression levels of PTHrP were decreased in expression levels (Supplementary Figure S4). shROBO1 cells (Po0.001), whereas, in shPITX1cells, PTHrP levels Two of the genes identified chromatin modifier HDAC4 and the showed a stunted abrogation (Po0.001). This difference was transcriptional factor PITX1 could modulate large gene network; further diminished in triple KD cells as compared with their we explored their contribution to the TGF-b signaling pathway. To respective controls (Figure 5c) indicating an additive effect of this end, we analyzed the activation status of R-SMADs in silenced combined decreased levels in ROBO1 and PITX1 KD cells. tumor cells. Upon TGF-b treatment, no significant changes were Furthermore, downregulation of IL11 expression levels in stimu- observed in single PITX1 KD cells in SMAD 2 and 3 activation lated conditions was observed for all three single-KD cells, (Supplementary Figures S5a and b), whereas single ROBO1 KD whereas a more pronounced decrease was detected in triple-KD cells showed no changes in SMAD 2 activation and a fast cells (Figure 5c), suggesting that each gene differentially activation of Smad 3 at 15 min that diminished at 30 min contributes to PTHrP and IL11 levels after TFG-b stimulation. (Supplementary Figures S5a and c). In contrast, a robust SMAD 2 Interestingly, enhanced expression levels of IL11 were also and 3 activation was observed in HDAC4 KD cells 30 min after TGF- detected upon TGF-b stimulation in double overexpressor b stimulation (Figure 6a). This activation was maintained in double HDAC4/PITX1 (Supplementary Figure S3e). These data indicate HDAC4/PITX1 KD cells (Supplementary Figure S6b), but not in that this triple gene subset is required in tumor cells for proper other double combinations (Supplementary Figures S6a and c) PTHrP and IL11 expression under TGF-b-stimulated conditions. except for the HDAC4/ROBO1 KD where, similarly to single It has been well established that MMPs significantly contribute ROBO1KD cells, a strong activation was observed for SMAD3 at to bone matrix degradation in bone metastases.14 Although type I 15 min that further declined at 30 min (Supplementary Figure S6a). Collagen can be degraded by several MMPs, we have previously Interestingly, at the same time point, 3KD cells displayed an

Oncogene (2014) 5090 – 5099 & 2014 Macmillan Publishers Limited Antiosteolytic targets D Luis-Ravelo et al 5095 ST2 Parental PTHrP expression 60 10 ** *** *** 8 40 6 *** Ctrl 3-shRNA 20 4 per field (10x) Fold Increase 2 % Osteoclast surface 0 ST2 Parental Ctrl 3-shRNA 0  +ST2 TGF- -+-++ -+-+ -+- Controlsh-HDAC4 sh-PITX1 sh-ROBO1 Ctrl 3-shRNA

IL-11 expression 200 150 Stromal rankl * *** * *** 100 150

50 100

0 A 50 Relative expression (%) T2 trl T2 T2 ST2 C N

+S Fold Increase trl+S A 5 Parental C 3-shRN 4 arental+S P 3 3-shR 2 Tumor PTHrP 1 150 0 *** TGF- -+-++ -+-+ - + - 100 Controlsh-HDAC4 sh-PITX1 sh-ROBO1 Ctrl 3-shRNA

50 * 100 0

Relative expression (%) Parental Ctrl 3-shRNA 50

150 Tumor IL-11 RFU (%)  ** 100 0 Ctrl 3-shRNA

50

0

Relative expression (%) Parental Ctrl 3-shRNA Figure 5. Abrogation of triple gene signature alters osteoclastogenic and metalloproteolytic activities. (a) Left panel: conditioned media (CM) derived from controls and 3KD cells cocultured with ST2 cells were collected and incubated with murine bone marrow macrophages for 6 days in presence of M-CSF (20 ng/ml) and RANKL (15 ng/ml). The experiments were performed three times with similar results. Ctrl, denotes the lentiviral transduction with three different scramble sequences. Right panel: representative images of TRAP þ multinucleated cells. Nonparametric test of the median was used (*Po0.05). Statistical validation of experiment with coculture CM was carried out by two-tailed Student’s t test (**Po0.01). (b) Upper panel: stromal rankl in ST-2 stromal cells cocultured with 3KD and control cells was quantified by murine-specific real time PCR. Center and bottom panels: tumor PTHrP and IL-11 expression levels were quantified by specific real-time PCR (Welch test, ***Po0.001; Mann–Whithney test, **Po0.01). (c) Assessment of PTHrP (Upper panel) and IL11 (Bottom panel) expression levels in single and triple KD cells 24 h after TGF-b stimulation (7.5 ng/ml) by RT–qPCR. ‘Control’, denotes lentiviral transduction with a single scramble sequence, whereas ‘Ctrl’, denotes lentiviral transduction with three different scramble sequences. (d) Metalloproteolytic activity assessed by fluorogenic peptide recognizing MMP-3,-10. CM derived from 24 h incubation of control and 3KD cells in coculture with ST2 cells. Data are expressed as a percentage of increased activity (RFU: relative units) when tumor cells are cultured with ST2 (median ± interquartile range, n ¼ 4). Mann–Whitney test was used (*Po0.05). impaired activation of SMAD 2 and 3 and a marked decrease in Combined high HDAC4 and PITX1 expression predicts metastasis total levels of SMAD 2/3 (Figure 6b). Furthermore, evaluation of risk in lung AC patients SMAD 2/3 expression levels by RT–qPCR showed almost unaltered To investigate the clinical implications of the three-gene signature expression levels 30 min after TGF-b stimulation in triple KD as in lung AC metastasis, we used a previously published database12 compared with control cells (Figure 6c). Of note, a more sustained from a lung AC cohort that includes information on lung Smad 3 activation from 30 min was observed in double over- tumor progression to metastasis for 107 lung AC patients expressor HDAC4/PITX1 (Supplementary Figure S6d). Taken (Supplementary Table S2a). Normalized expression values of together, these data indicate that the triple gene combination is single probe sets were divided into high-expression and low- required for proper sensitivity of TGF-b signaling. These results expression groups of patient using median as a cutoff. A shorter also suggest that triple HDAC4, PITX1 and ROBO1 combination is time of metastasis-free survival was found in HDAC4 and PITX1 required for proper R-Smads activation and together contributes high-expression patients as compared with low expressors to unimpaired TGF-b response in tumor cells. (Figure 7a). Classification according to ROBO1 levels showed no

& 2014 Macmillan Publishers Limited Oncogene (2014) 5090 – 5099 Antiosteolytic targets D Luis-Ravelo et al 5096 pSmad2/Smad2 pSmad3/Smad3 shHDAC4 2500 600 Control shRNAControl shRNA 2000 Control shRNA TGF- ---+--- + + + 400 1500 pSmad2 1000 pSmad3 200 Smad2 500 Smad2/3 Relative Intensity Smad3 Relative Intensity 0 0 GAPDH TGF- -+-+ -+ -+ TGF- - +-+ -+-+ Control shRNAControl shRNA Control shRNAControl shRNA 0 15´ 30´ 15´ 30´ 15´ 30´

SMAD-2 expression 2.5 Ctrl 3-shRNACtrl 3-shRNA Ctrl 3-shRNA TGF- ---++++--- 2.0 1.5 pSmad2 1.0 pSmad3 0.5 Smad2 Smad2/3 0.0 Smad3 TGF- - - -+-+ -+-+

Ctrl 3-shRNACtrl 3-shRNA GAPDH Ctrl 15´ 30´

0 15´ 30´ 3-shRNA SMAD-3 expression pSmad2/Smad2 2.0 2000 4000 pSmad3/Smad3 1.5 1500 3000 1.0 1000 2000 0.5

500 1000 0.0 -- -+-+ -+-+

Relative Intensity 

Relative Intensity TGF- 0 0 Ctrl 3-shRNACtrl 3-shRNA

TGF- -+-+ -+-+ TGF- -+-+ -+-+ Ctrl 15´ 30´ Ctrl 3-shRNACtrl 3-shRNA Ctrl 3-shRNACtrl 3-shRNA 3-shRNA 15´ 30´ 15´ 30´ Figure 6. Activation of TGF-b signaling in single HDAC4 and triple KD cells. (a) Left panel: western blot analysis of phosphorylated and total Smad-2,-3 in TGF-b-treated or non-treated HDAC4 KD cells (7.5 ng/ml). GAPDH was used as loading control. Right panels: quantification by densitometric analysis. (b) Upper panel: western blot analysis of phosphorylated and total Smad-2,-3 in TGF-b-treated or non-treated triple KD cells. Bottom panel: quantification by densitometric analysis. (c) Evaluation of SMAD 2 (Upper panel) and SMAD 3 (Bottom panel) expression levels in single and triple KD cells 15 and 30 min upon TGF-b stimulation by RT–qPCR.

significant differences (data not shown). Interestingly, clinical the marked abrogation of osteolysis and tumor burden induced samples with higher levels of both HDAC4 and PITX1 (24 of 107 by 3KD cells and revealed the significance of these genes as lung AC patients) showed a stronger correlation with shorter regulators of the tumor-induced osteoclastogenic process. As a latency time in metastasis events (Figure 7b). Discrimination consequence, it is tempting to speculate that decreased bone shown by triple combination was significant, although it was not resorption could lead to the low release of bone matrix-derived as robust as the double combination (Figure 7c). Furthermore, TGF-b levels. Because TGF-b is an indispensable signal for RANKL- using an independent cohort of 442 lung cancer patients from induced ostoclastogenesis through the formation of a TRAF6/ Shedden16 (Supplementary Table S2b), Kaplan–Meier survival TAB1/TAK1 molecular complex and its interaction with SMAD 3,17 analysis showed similar results (Supplementary Figures S8a–c). this will further contribute to impaired osteoclastogenesis. In These data indicate that combined high HDAC4, PITX1 and ROBO1 addition, in triple 3KD tumor cells, the secretion of PTHrP was transcriptomic levels discriminate patients at higher risk of obliterated and IL11 expression severely impaired, that is, there metastasis development and poor survival. was a sharp downregulation of two major prosteoclastogenic factors. Moreover, the response to TGF-b was severely mitigated in triple KD, which indicated that this gene subset might be DISCUSSION cooperatively required for proper sensitivity to TGF-b signaling This study identified a subset of genes involved in the osseous response. Inhibition of these mediators desensitizes tumor cells to prometastatic activity of lung AC. Abrogation of this gene TGF-b response and unveils a novel strategy to obliterate this signature entailed decreased tumor-induced osteolytic lesions, vicious cycle driving osteolysis in lung cancer. an effect partially mediated by impaired osteoclastogenesis. One possible explanation for the decreased activity of SMAD 2 This finding was associated with downregulation of several key and SMAD 3 could be induced by concurrent changes in prosteoclastogenic factors including a cell autonomous decrease chromatin and transcriptional regulation of several components in PTHrP and IL11 production. Moreover, stromal induction of rankl involved in the pathway. Perturbed signaling of upstream mediated by tumor–stromal interactions was also severely protein kinases involved in SMAD 2/3 phosphorylation, including impaired in coculture with 3KD cells. These effects might explain MAPK, CDK2/4, and others could account for these effects.

Oncogene (2014) 5090 – 5099 & 2014 Macmillan Publishers Limited Antiosteolytic targets D Luis-Ravelo et al 5097 HDAC4 HDAC4-PITX1 100 100

80 80

60 60

40 p=0.0297 40 p<0.0001 20 High 20 High 1.60 (0.86-2.94) Low 7.817 (3.225- 18.94) Low 0 0 Metastasis-free survival (%) 0 20 40 60 80 100 Metastasis-free survival (%) 0 20 40 60 80 100 Time after first metastasis (months) Time after first metastasis (months)

PITX1 HDAC4/PITX1/ROBO1 100 100 80 80

60 60 p=0.023 40 p=0.0054 40 High 20 20 High Low 2.43 (1.30-4.55) 5.42 (1.26-23.27) Low 0 0 Metastasis-free survival (%) 0 20 40 60 80 100 Metastasis-free survival (%) 0 20 40 60 80 100 Time after first metastasis (months) Time after first metastasis (months) Figure 7. Assessment of risk of metastasis in lung AC patients. (a) Metastasis-free survival curves in patients displaying high or low expression in HDAC4 and PITX1 in a published cohort by Nguyen et al.6,12 (b) Metastasis-free survival curves in patients displaying simultaneous high expression of double HDAC4 and PITX1 combination according to median cutoff of singles genes. (c) Metastasis-free survival curves in patients displaying simultaneous high expression levels of triple gene combination HDAC4, PITX1 and ROBO1. In this analysis, 24 patients were double positive for both HDAC4 and PITX1 (vs 83 displaying any or only a single positive), whereas only nine patients were triple positive. Hazard ratio and 95% confidence intervals are shown for each analysis.

Complementary to this view, increased activity of phosphatases of colon tumor cells through the inhibition of p53.22 It also such as SCP1-3 involved in SMAD 2/3 dephosphorylation might participates in cisplatin-mediated activation of STAT1, contributing also be required for attenuating TGF-b response.18 Another to platinum chemoresistance in ovarian cancer.23 Recently, HDAC4 complementary explanation, based on our findings, could be has been associated in the mechanism elicited by the ‘metastasis- associated with an increased degradation of total SMAD 2/3 levels associated protein-1’, a master regulatory gene involved in in triple KD cells substantiated by the finding of its post- prometastatic activity.24 transductional decrease upon TGF-b incubation. Thus, either Similarly, PITX-1 belongs to a family of transcription factors that directly or indirectly, it is tempting to speculate that this triple are also involved in development25 and tumorigenesis. Silenced combination could regulate the baseline levels of SMAD activity PITX1 levels in deficient p16-RAS-mutated cells resulted in an expression and/or activity of kinases and/or phosphatases increase of tumorigenic properties.26 This finding could explain required for proper TGF-b signaling. the mild nonsignificant increase in tumor cell proliferation In addition to the effects derived from osteoclast activation, observed in single KD cells (data not shown). Similarly, tumor-induced MMPs significantly contribute to bone matrix decreased levels of PITX-1 have been detected in esophageal degradation.14 We have previously shown that MMP-3/ MMP-10 cancer27 and lung cancer progression.28 In contrast, the activities were rapidly induced upon tumor–stromal interactions.15 observation that PITX1 inhibition led to a slight decrease in Despite the fact that type I collagen can be degraded by several bone metastasis suggests a cell context-dependent function. In MMPs, it has been recently suggested that MMP-10 is required for this setting, it is possible that PITX-1 might be required for the lung cancer stem cell maintenance and metastatic potential.19 The secretion of tumor-derived prosteoclastogenic factors leading to fact that MMP3/10 activity was dramatically decreased in stromal tumor-induced osteolytic lesions. Moreover, the elucidation of coculture of 3KD cells indicated that this gene subset might also downstream targets induced by PITX-1 requires further studies. be involved in MMP signaling events driven by tumor–host ROBO1 and its ligand SLIT2 have been implicated as having interactions. Abrogation of MMP activity was only detected in dual roles in tumorigenesis and metastasis. Ligand-induced coculture conditions, whereas no changes were found in cells ROBO1 signaling causes decreased cell– and elicits cultured alone. These findings underscore the relevance of tumor– migratory pathways.29,30 In line with our findings, overexpression stromal heterotypic contacts in modulating the metastatic of ROBO1 has been detected in the metastatic signature of brain response. It should be noted that similar tumor growth was and lung metastases derived from breast cancer.31,32 Similarly, observed in subcutaneous tumors of triple KD despite sharp SLIT-ROBO signaling mediates tumor lymphatic metastasis in a differences in tumor burden observed in the osseous compart- model of pancreatic cancer.33 Interestingly, TGF-b and Wnt ment. Therefore, modulation of cell autonomous and host-derived signaling pathways12 critical in lung AC metastases also effects were involved in the osteolytic response and metastatic induce ROBO1 expression34 in agreement with our findings tumor growth observed in vivo. (Supplementary Figure S7). Thus, ROBO1 could most likely Of the three genes, HDAC4 belongs to a class II histone contribute to lung cancer bone metastasis. deacetylase family of key regulators in growth, development20,21 Intriguingly, genes that do not have an effect when singly and tumorigenesis. For instance, HDAC4 promotes growth abrogated become relevant when cooperatively depleted in HMS.

& 2014 Macmillan Publishers Limited Oncogene (2014) 5090 – 5099 Antiosteolytic targets D Luis-Ravelo et al 5098 This is line with previous studies supporting a collective RNA extraction and quantitative RT–PCR 35 contribution of multiple genes in the metastatic process. A RNA from single cultures and cocultures in 100 mm plates was isolated gene signature of TCF4, MCAM, PRKD3 y SUSD5 was involved in using Trizol reagent (Invitrogen, Life Technologies, Grand Island, NY, USA). metastatic colonization of lung cancer to the bone in a model of One microgram of RNA was reverse transcribed using SuperScript III large cell carcinoma.9 None of these genes was shared by our (Invitrogen). Then cDNA was used as a template for real-time PCR using gene signature, although both studies found identical bone SYBR dye (SYBR Green PCR Master Mix, Applied Biosystems, Life tropism and similarities in osteolytic lesions and were performed Technologies, Grand Island, NY, USA) in ABI PRISM7500 device (Applied Biosystems). Primer sequences were as follows: human IL-11, forward 50- using an identical methodology (Supplementary Materials and CTCTCTCCTGGCGGACACG-30, reverse 50-AGGGAATCCAGGTTGTGGTCC-30; methods). This divergence among studies suggests different human PTHrP, forward 50-CGTCGCTGGAGCTCGATT-30, reverse 50-AATCCTG activities of bone colonization mechanisms, which may reflect CAATATGTCCTTGGAA-30; human GAPDH, forward 50-CTGCTCCTCCTGTT different biological events and functions elicited by the subset of CGACAGT, reverse 50-CCATGGTGTCTGAGCGATGT-30; mouse rankl, forward identified genes. Alternatively, different histological subtypes of 50-GCACACCTCACCATCAATGC-30, reverse 50- CCTCGATCGTGGTACCAA- tumors (large cell carcinoma vs AC) might utilize distinct set of GAG-30; mouse gapdh, forward 50-GTTCCAGTATGACTCCACTCAC-30, reverse gene signatures to accomplish cell metastatic functions. 50- GGCCTCACCCCATTTG-30. Human or mouse gapdh was used as HDAC4 and PITX1 transcript levels could predict the risk of endogenous controls. The comparative Ct method (DDCt) was performed metastasis in a clinical cohort of lung AC patients. Their to calculate the fold induction or inhibition. cooperative contribution yields the most salient clinical outcome in two independent cohorts (Figure 7 and Supplementary Western blot analysis Figure S8), indicating that in primary tumors expression levels of Cells were lysed in protein extraction buffer (pH 7.4, 25 mM sodium these genes could be a prognostic factor for developing b-glycerophosphate, 1 mM Na2P3O4,1mM benzamidine, 5 mM EDTA, 5 mM metastases. In this setting, ROBO1 expression levels either alone EGTA, 0.05% sodium deoxycholate, 0.01% SDS, 1% Triton X-100, 10% or in combination with HDAC4 and PITX1 might not have a glycerol, 0.1% b-mercaptoethanol, 1 mM sodium orthovanadate and 25 mM NaF; all reagents from Sigma) plus a inhibitor cocktail (Roche, significant role in the prediction of the clinical course of lung Basel, Switzerland). Lysates were then centrifuged at 9000 r.p.m. using a cancer patients. However, in late steps of metastasis recapitulated tabletop centrifuge and protein-containing supernatants were collected. in our model system, high levels of PITX1 and HDAC4 markedly Protein concentration was determined using a Bradford assay (Bio-Rad, increase bone metastatic activity despite the most significant Hercules, CA, USA). For western blotting, 20–40 mg total protein was effects on metastasis occurred with the triple gene combination. electrophoresed in polyacrylamide gels (Bio-Rad). After electrophoresis, Indeed, ROBO1 contributed to promote aggressive bone lesions in protein was electroblotted onto nitrocellulose membranes (Bio-Rad). combination with HDAC4/PITX1, as triple gene abrogation Membranes were then blocked in 5% skimmed milk solution (buffered resulted in a significant decrease in osseous metastatic burden in TBS–0.05% Tween) and incubated overnight with primary antibodies and lesion area as compared with mice inoculated with double directed against phospho-Smad2 (1:2000), phospho-Smad3 (1:1000), Smad2/3 (1:2000), PITX1 (1:1000), HDAC4 (1:1100) and b-actin (1:5000). HDAC4/PITX1-silenced tumor cells (data not shown). The con- After overnight incubation, blots were washed three times with TBS– tribution of ROBO1 in the triple-silenced combination to the Tween and incubated with HRP-linked secondary antibodies against rabbit osteolytic lesions is also supported by the finding of a marked or mouse immunoglobulins (Amersham, Bucks, UK) for 1 h at room decrease in TGF-b response in these cells as compared with temperature. Membranes were washed four times with TBS–Tween and double HDAC4/PITX1-KD cells (Figure 6b and Supplementary treated for 5 min with the peroxidase substrate LumiLight Plus (Roche). Figure S6b). Blots were exposed to ECL films (Amersham) for 1 min and developed. Thus, the use of these markers could have a clinical benefit in the outcome of these patients, but rigorous assessment of their In silico analysis clinical relevance requires further studies in other independent Metastasis-free survival analysis was performed using expression and series. It is worth noting that triple stable inhibition was able to clinical data available at http://www.cbio.mskcc.org/Public/lung_array_- extend animal survival and preserve body weight in the data/ (Gerald experiment). The main features of this cohort are experimental setting preventing tumor-induced cachexia. This summarized in Supplementary Table S2. For single probe set studies, daunting clinical complication associated with metastasis normalized expression values from microarray experiments were divided in progression is frequently observed in different lung cancer cell high- and low-expression groups using median as cutoff. For multiple models.15,36 The finding that PTHrP inhibition allowed partial probe set analysis, tumor samples with high levels of transcripts of interest were compared with the rest of the clinical cohort. restoration of parameters associated with cachexia such as weight or serum calcium levels without affecting tumor weight37 could partially explain our results. Most likely, other pro-cachectic factors Statistical analysis might also be regulated by the triple gene subset. Log-rank test was used to calculate the statistical significance (P-value) of In summary, we have identified a set of genes mediating differences observed among Kaplan–Meier curves. To study differences in osseous colonization. Together with clinical observations, proliferation rates, tumor growth, differences in metastatic area, number of osteoclasts, number of single-cell-derived colonies and metalloproteolytic our data collectively support the role of these genes as combined activity data were analyzed by different comparison tests. The test used for factors with potential clinical value in lung AC. Future studies parametric analysis was ANOVA followed by Tukey Post hoc test in cases will address the impact of these findings to predict risk of with variance homogeneity and Brown-Forsythe, whereas T2 Tamhane was metastasis progression and will ultimately evaluate patient used when variances were heterogeneous. For nonparametric analysis, benefits. tests of Kruskal–Wallis, Mann–Whitney with Bonferroni adjustment were used. Values were expressed as mean±s.d. or median±interquartile range according to parametric or nonparametric analyses. Statistical significance was defined as *Po0.05, **Po0.01 and ***Po0.001. MATERIALS AND METHODS In vitro osteoclastogenic assay, MMP activity, and in vivo experiments Reagents have been previously described.38 TGF-b was purchased from Peprotech (Rocky Hill, NJ, USA). Antibodies against phospho-Smad2 (Millipore, Billerica, MA, USA), phospho-Smad3 ABBREVIATIONS and total Smad 2/3 (Cell Signaling, Danvers, MA, USA), PITX1 (Bethyl Laboratories, Montogomery, TX, USA), HDAC4 (Abcam, Cambridge, UK), AC, adenocarcinoma; BLI, bioluminescence imaging; CM, condi- Ki-67 (Thermo Scientific, Waltham, MA, USA), b-actin (Sigma, St Louis, MO, tioned medium; i.c., intracardiac; i.t., intratibial; KD, knockdown; USA) and GAPDH (Biolegend, San Diego, CA, USA) were used. Lentiviral MMP, matrix metalloprotease; SCDC, single-cell-derived colonies; shRNAs were obtained from Sigma Mission shRNA (Sigma). TRAP, Tartrate-resistant acid phosphatase

Oncogene (2014) 5090 – 5099 & 2014 Macmillan Publishers Limited Antiosteolytic targets D Luis-Ravelo et al 5099 CONFLICT OF INTEREST 18 Sapkota G, Knockaert M, Alarcon C, Montalvo E, Brivanlou AH, Massague The authors declare no conflict of interest. J. Dephosphorylation of the linker regions of Smad1 and Smad2/3 by small C-terminal domain phosphatases has distinct outcomes for bone morphogenetic protein and transforming growth factor-beta pathways. J Biol Chem 2006; 281: ACKNOWLEDGEMENTS 40412–40419. 19 Justilien V, Regala RP, Tseng IC, Walsh MP, Batra J, Radisky ES et al. Matrix We are grateful to the Core Histology Unit and Imaging Group facilities for their -10 is required for lung cancer stem cell maintenance, tumor outstanding technical assistance. We also thank the Proteomics, Genomics and initiation and metastatic potential. PLoS One 2012; 7: e35040. Bioinformatics Unit at the CIMA. This work was supported by ‘UTE project FIMA’ 20 Vega RB, Matsuda K, Oh J, Barbosa AC, Yang X, Meadows E et al. Histone dea- agreement, The Cancer Research Thematic Network of the Health Institute Carlos III cetylase 4 controls chondrocyte hypertrophy during skeletogenesis. Cell 2004; (RTICC RD06/0020/0066), PI042282, SAF-2009–11280, SAF2012-40056, grants 67/2005 119: 555–566. and 09/2009 from the Government of Navarra, and ‘La Caixa Foundation’ to FL. DL-R 21 Backs J, Song K, Bezprozvannaya S, Chang S, Olson EN. CaM kinase II selectively was supported by the FIMA and FPU. IA was funded by the Basque Government. JDLR signals to histone deacetylase 4 during cardiomyocyte hypertrophy. J Clin Invest was supported by FIS-ISCIII grant PI12/00624. SV is an investigator of the Ramon y 2006; 116: 1853–1864. Cajal Program (MICINN, RYC-2011-09042). 22 Wilson AJ, Byun DS, Nasser S, Murray LB, Ayyanar K, Arango D et al. HDAC4 promotes growth of colon cancer cells via repression of p21. Mol Biol Cell 2008; 19: 4062–4075. REFERENCES 23 Stronach EA, Alfraidi A, Rama N, Datler C, Studd JB, Agarwal R et al. HDAC4- 1 Feld R, Rubinstein LV, Weisenberger TH. Sites of recurrence in resected stage I regulated STAT1 activation mediates platinum resistance in ovarian cancer. non-small-cell lung cancer: a guide for future studies. J Clin Oncol 1984; 2: Cancer Res 2011; 71: 4412–4422. 1352–1358. 24 Reddy SD, Pakala SB, Molli PR, Sahni N, Karanam NK, Mudvari P et al. Metastasis- 2 Fry WA, Menck HR, Winchester DP. The National Cancer Data Base report on lung associated protein 1/histone deacetylase 4-nucleosome remodeling and deace- cancer. Cancer 1996; 77: 1947–1955. tylase complex regulates phosphatase and tensin homolog and 3 Goldstraw P, Ball D, Jett JR, Le Chevalier T, Lim E, Nicholson AG et al. Non-small- function. J Biol Chem 2012; 287: 27843–27850. cell lung cancer. Lancet 2011; 378: 1727–1740. 25 Marcil A, Dumontier E, Chamberland M, Camper SA, Drouin J. Pitx1 and Pitx2 are 4 Weilbaecher KN, Guise TA, McCauley LK. Cancer to bone: a fatal attraction. Nat Rev required for development of hindlimb buds. Development 2003; 130: 45–55. Cancer 2011; 11: 411–425. 26 Kolfschoten IG, van Leeuwen B, Berns K, Mullenders J, Beijersbergen RL, Bernards 5 Coleman RE. Skeletal complications of malignancy. Cancer 1997; 80: 1588–1594. R et al. A genetic screen identifies PITX1 as a suppressor of RAS activity and 6 Nguyen DX, Massague J. Genetic determinants of cancer metastasis. Nat Rev tumorigenicity. Cell 2005; 121: 849–858. Genet 2007; 8: 341–352. 27 Lord RV, Brabender J, Wickramasinghe K, DeMeester SR, Holscher A, Schneider PM 7 Guise TA, Yin JJ, Taylor SD, Kumagai Y, Dallas M, Boyce BF et al. Evidence for a et al. Increased CDX2 and decreased PITX1 homeobox gene expression in Bar- causal role of parathyroid hormone-related protein in the pathogenesis of human rett’s esophagus and Barrett’s-associated adenocarcinoma. Surgery 2005; 138: breast cancer-mediated osteolysis. J Clin Invest 1996; 98: 1544–1549. 924–931. 8 Yin JJ, Selander K, Chirgwin JM, Dallas M, Grubbs BG, Wieser R et al. TGF-beta 28 Chen Y, Knosel T, Ye F, Pacyna-Gengelbach M, Deutschmann N, Petersen I. signaling blockade inhibits PTHrP secretion by breast cancer cells and bone Decreased PITX1 homeobox gene expression in human lung cancer. Lung Cancer metastases development. J Clin Invest 1999; 103: 197–206. 2007; 55: 287–294. 9 Vicent S, Luis-Ravelo D, Anton I, Garcia-Tunon I, Borras-Cuesta F, Dotor J et al. 29 Rhee J, Mahfooz NS, Arregui C, Lilien J, Balsamo J, VanBerkum MF. Activation of A novel lung cancer signature mediates metastatic bone colonization by a dual the repulsive receptor Roundabout inhibits N-cadherin-mediated cell adhesion. mechanism. Cancer Res 2008; 68: 2275–2285. Nat Cell Biol 2002; 4: 798–805. 10 Valencia K, Ormazabal C, Zandueta C, Luis-Ravelo D, Anton I, Pajares MJ et al. 30 Rhee J, Buchan T, Zukerberg L, Lilien J, Balsamo J. Cables links Robo-bound Abl Inhibition of Collagen Receptor Discoidin Domain Receptor-1 (DDR1) reduces cell kinase to N-cadherin-bound beta-catenin to mediate Slit-induced modulation of survival, homing, and colonization in lung cancer bone metastasis. Clin Cancer Res adhesion and transcription. Nat Cell Biol 2007; 9: 883–892. 2012; 18: 969–980. 31 Minn AJ, Gupta GP, Siegel PM, Bos PD, Shu W, Giri DD et al. Genes that mediate 11 Anto´n I, Molina E, Luis-Ravelo D, Zandueta C, Valencia K, Ormazabal C et al. breast cancer metastasis to lung. Nature 2005; 436: 518–524. Receptor of activated protein c promotes metastasis and correlates with clinical 32 Bos PD, Zhang XH, Nadal C, Shu W, Gomis RR, Nguyen DX et al. Genes that outcome in lung adenocarcinoma. Am J Resp Crit Care Med 2012; 186: 96–105. mediate breast cancer metastasis to the brain. Nature 2009; 459: 1005–1009. 12 Nguyen DX, Chiang AC, Zhang XH, Kim JY, Kris MG, Ladanyi M et al. WNT/TCF 33 Yang XM, Han HX, Sui F, Dai YM, Chen M, Geng JG. Slit-Robo signaling mediates signaling through LEF1 and HOXB9 mediates lung adenocarcinoma metastasis. lymphangiogenesis and promotes tumor lymphatic metastasis. Biochem Biophys Cell 2009; 138: 51–62. Res Commun 2010; 396: 571–577. 13 Catena R, Luis-Ravelo D, Anton I, Zandueta C, Salazar-Colocho P, Larzabal L et al. 34 Labbe E, Lock L, Letamendia A, Gorska AE, Gryfe R, Gallinger S et al. PDGFR signaling blockade in marrow stroma impairs lung cancer bone metas- Transcriptional cooperation between the transforming growth factor-beta and tasis. Cancer Res 2011; 71: 164–174. Wnt pathways in mammary and intestinal tumorigenesis. Cancer Res 2007; 67: 14 Lu X, Wang Q, Hu G, Van Poznak C, Fleisher M, Reiss M et al. ADAMTS1 and MMP1 75–84. proteolytically engage EGF-like ligands in an osteolytic signaling cascade for bone 35 Kang Y, Siegel PM, Shu W, Drobnjak M, Kakonen SM, Cordon-Cardo C et al. metastasis. Genes Dev 2009; 23: 1882–1894. A multigenic program mediating breast cancer metastasis to bone. Cancer Cell 15 Luis-Ravelo D, Anton I, Vicent S, Hernandez I, Valencia K, Zandueta C et al. Tumor- 2003; 3: 537–549. stromal interactions of the bone microenvironment: in vitro findings and potential 36 Lecanda F. Tumor-stroma: tumor microenvironment as a target to combat in vivo relevance in metastatic lung cancer models. Clin Exp Metastasis 2011; 28: metastasis. Dis Model Mech 2011; 8: 87–93. 779–791. 37 Iguchi H, Onuma E, Sato K, Ogata E. Involvement of parathyroid hormone- 16 Shedden K, Taylor JM, Enkemann SA, Tsao MS, Yeatman TJ, Gerald WL et al. Gene related protein in experimental cachexia induced by a human lung cancer- expression-based survival prediction in lung adenocarcinoma: a multi-site, blin- derived cell line established from a bone metastasis specimen. Int J Cancer 2001; ded validation study. Nat Med 2008; 14: 822–827. 94: 24–27. 17 Yasui T, Kadono Y, Nakamura M, Oshima Y, Matsumoto T, Masuda H et al. Reg- 38 Hernandez I, Moreno JL, Zandueta C, Montuenga L, Lecanda F. Novel alternatively ulation of RANKL-induced osteoclastogenesis by TGF-beta through molecular spliced ADAM8 isoforms contribute to the aggressive bone metastatic phenotype interaction between Smad3 and Traf6. J Bone Miner Res 2011; 26: 1447–1456. of lung cancer. Oncogene 2010; 29: 3758–3769.

Supplementary Information accompanies this paper on the Oncogene website (http://www.nature.com/onc)

& 2014 Macmillan Publishers Limited Oncogene (2014) 5090 – 5099