|Home | About | Journals | Submit | Contact Us | Français|
Metastasis from lung adenocarcinoma can occur swiftly to multiple organs within months of diagnosis. The mechanisms that confer this rapid metastatic capacity to lung tumors are unknown. Activation of the canonical WNT/TCF pathway is identified here as a determinant of metastasis to brain and bone during lung adenocarcinoma progression. Gene expression signatures denoting WNT/TCF activation are associated with relapse to multiple organs in primary lung adenocarcinoma. Metastatic subpopulations isolated from independent lymph node-derived lung adenocarcinoma cell lines harbor a hyperactive WNT/TCF pathway. Reduction of TCF activity in these cells attenuates their ability to form brain and bone metastases in mice, independently of effects on tumor growth in the lungs. The WNT/TCF target genes HOXB9 and LEF1 are identified as mediators of chemotactic invasion and colony outgrowth. Thus a distinct WNT/TCF signaling program through LEF1 and HOXB9 enhances the competence of lung adenocarcinoma cells to colonize the bones and the brain.
Metastasis is an ominous feature of malignant solid tumors and its natural course varies extensively depending on the type of disease. The tissue and cell of origin of the tumor markedly influence the sites and severity of distant metastasis. Depending on the cancer subtype, metastasis may emerge quickly in multiple organs or in particular organs after a protracted latency period. Breast tumors, for example, undergo dissemination from an early stage (Husemann et al., 2008; Klein et al., 1999), but metastasis to bones, lungs, brain, or liver may not manifest until years or decades after the diagnosis and removal of the primary tumor (Karrison et al., 1999; Schmidt-Kittler et al., 2003). This course suggests that disseminated breast cells normally require further malignant evolution to overtly colonize distant organs. This process gives rise to organ-specific metastatic cell populations, which has prompted the identification of organ-specific metastasis genes and functions (Nguyen et al., 2009).
Metastasis in other types of cancer follows a different course, with rapid spread to multiple organs. A case in point is lung adenocarcinoma, the most common subtype of lung cancer with a high mortality rate (Hoffman et al., 2000). Even if diagnosed at an early stage and surgically removed, lung adenocarcinomas can relapse within months, spreading to the lymph nodes, contra-lateral lung, adrenal glands, the bones, and the brain (Feld et al., 1984; Hess et al., 2006; Martini et al., 1995). Brain metastasis has particularly morbid physical and cognitive consequences (Gaspar, 2004; Gavrilovic and Posner, 2005; Thomas et al., 2000). Lung adenocarcinoma is a principal source of brain metastasis, due in part to the arterial circulation flowing directly from the lungs to the brain, which pulmonary cancer cells can more easily access (Gaspar, 2004; Gavrilovic and Posner, 2005; Thomas et al., 2000).
The genes and mechanisms that mediate lung adenocarcinoma metastasis are largely unknown. However, the natural course of this disease suggests the acquisition of mechanisms in lung adenocarcinoma cells that allow the aggressive infiltration and colonization of multiple organs soon after cancer cell dispersion from a primary tumor. This hypothesis led us to consider the involvement of developmental pathways as providers of these functions. By combining genome wide transcriptional analysis and experimental modeling, we identified the activation of the WNT/TCF pathway as a determinant of lung adenocarcinoma metastatic competence to multiple organ sites.
The WNT signaling pathway plays major roles in stem cell biology, organogenesis, tissue homeostasis, and cancer (Clevers, 2006; Klaus and Birchmeier, 2008; Logan and Nusse, 2004; Moon et al., 2004). Binding of secreted Wnt ligands to Frizzled and LRP5/6 transmembrane receptors prevents the phosphorylation-dependent polyubiquitination and degradation of the signaling pool of β-catenin in the cytoplasm. This degradation is initiated by a protein complex including APC (the adenomatous polyposis coli gene product), Axin, and glycogen synthase kinase-3β (GSK-3β) (Clevers, 2006; Logan and Nusse, 2004). Spared from degradation, β-catenin accumulates in the nucleus and binds to transcription factors of the LEF/TCF (lymphoid enhancer-binding factor and T cell factor) family to regulate gene expression (Arce et al., 2006).
Mutations that constitutively activate the WNT pathway are frequently involved in colorectal cancer initiation (Clevers, 2006; Moon et al., 2004; Morin et al., 1997; Polakis, 2007). However, such mutations are infrequent in lung adenocarcinomas (Ding et al., 2008). Prominent tumor-initiating events in lung adenocarcinoma include KRAS and EGFR mutations (Rodenhuis et al., 1988; Sharma et al., 2007). Surprisingly, we found that gene-expression signatures denoting WNT/TCF pathway activity in lung adenocarcinomas are associated with a high rate of relapse to distant sites. In vivo selection of metastatic cells from independent human lung adenocarcinoma lines yields derivatives that harbor a hyperactive WNT/TCF pathway, in association with an enhanced ability to infiltrate and colonize the brain and bones. Using these cells as a model, we provide functional evidence for the involvement of two WNT target genes, the transcription factors HOXB9 and LEF1, as mediators of chemotactic invasion and colony outgrowth.
Because of the propensity of lung adenocarcinomas to rapidly metastasize to multiple organs, and the lack of mutations associated with this transition thus far, we undertook the approach of analyzing a clinical cohort of lung adenocarcinomas using gene-expression signatures that denote activation of specific pathways. We utilized gene sets that were responsive to a K-Ras mutation in a mouse model (Sweet-Cordero et al., 2005), to a dominant-negative TCF4 (dnTCF4) in colon cancer cells (van de Wetering et al., 2002), to treatment of epithelial cells with TGF-β (Padua et al., 2008) or to overexpression of c-Myc, SRC, or E2F3 in human mammary epithelial cells (Bild et al., 2006). We queried gene-expression data on 107 annotated primary lung adenocarcinomas including 65 stage I tumors from patients who had relapsed and who remained disease free (MSKCC set 1, Supplementary Table 1). In an unsupervised analysis, the TCF4 gene signature identified a distinct sub-group (Figure 1A) with an elevated rate of recurrence to multiple organs (Figure 1B, p=0.006). The K-Ras, SRC, and E2F3 signatures were not predictive of increased overall metastasis (Figure 1C). The TGF-β signature, which is associated with lung relapse in estrogen receptor-negative breast tumors (Padua et al., 2008) was not specifically associated with lung adenocarcinoma metastasis, whereas the MYC signature was modestly predictive (Figure 1C, p=0.03). A β-catenin gene set (Bild et al., 2006) was not associated with relapse (data not shown). This set however largely consists of down-regulated genes in the absence of Wnt signal, potentially reflecting β-catenin functions that are independent of TCF transcription (Gottardi and Gumbiner, 2004). Following statistical correction for multiple comparisons, the TCF4 gene set was the only significant predictor of relapse (Bonferroni correction, p=0.042). Because TCF4 is a member of the LEF/TCF family of WNT pathway transcriptional effectors (Arce et al., 2006), we postulated that canonical WNT transcriptional activity may distinguish a subset of lung adenocarcinomas with high competence for metastasis.
To develop an experimental model for the analysis of lung adenocarcinoma metastasis we isolated metastatic subpopulations from two human lung cancer cell lines. The H2030 cell line was established from a lymph node of a stage-III lung adenocarcinoma patient and harbors a KRASG12C mutation (Phelps et al., 1996). The PC9 line was established from a lymph node of a lung adenocarcinoma patient with an EGFRΔexon19 mutation (Koizumi et al., 2005; Sasaki et al., 1991). As these two cell lines were derived from lymph nodes, they provided appropriate systems for the isolation of malignant cells that might harbor multi-organ metastatic competence before adaptation to grow in distant organs.
When placed in the arterial circulation of immunodeficient mice by intracardiac inoculation, both cell lines colonized the bones and brain. We extracted tumor cells from the brain and expanded them in culture for two more rounds of in vivo selection (Figure 2A). The resulting BrM3 cell populations consistently demonstrated enriched metastatic activity, as detected by bioluminescence imaging (BLI) and magnetic resonance of the brain or X-ray imaging of the bones (Figure 2B). H2030-BrM3 and PC9-BrM3 cells formed brain metastasis in 100% of animals compared to a 10%-13% efficiency of the parental lines (Figure 2C,D). Although isolated from the brain, the H2030-BrM3 cells and, to a lesser extent, the PC9-BrM3 cells showed an increased ability to colonize the vertebral and limb bones over their starting metastatic potential (Figure 2D). Thus the ability to metastasize to bone and brain are linked in the BrM3 derivatives.
In the bones, PC9-BrM3 and H2030-BrM3 formed osteolytic lesions (Figure 2E,F). PC9-BrM3 and H2030-BrM3 cells avidly colonized the cerebrum and cerebellum, forming nodules as well as smaller invasive lesions with occasional spread into the meninges (Figure 2G and Supplemental Figure 1A). Invasive metastases in the brain were not angiogenic, but displayed perivascular localization and proliferation (Figure 2H), with gliosis (Figure 2I) and astrocyte proliferation at the invasive edge of metastasis (Supplementary Figure 1B). Importantly, H2030-BrM3 cells implanted orthotopically in the lung could disseminate to the brain independently of lung and pleural tumor burden (Figure 2J).
BrM3 cells did not proliferate faster than their parental populations in monolayer culture, nor were they more resistant to growth factor depletion (Supplementary Figure 2A,B). When implanted with extracellular matrix Matrigel directly into the lungs, H2030-BrM3 cells had an initial growth advantage (Figure 2K, left), whereas PC9-BrM3 cells did not have altered tumorigenic rates compared to their parental counterparts (Figure 2K, right). Similar results were obtained from subcutaneously implanted tumors (data not show). The advantage of H2030-BrM3 cells is consistent with their enhanced colony-forming ability in Matrigel (see below). However, the rate of subsequent expansion of H2030-BrM3 tumors in the lung over 7 weeks was not significantly different from those of the parental line (Figure 2K, left). In sum, these two lung adenocarcinoma models recapitulate salient features of metastasis from lung adenocarcinoma, including the capacity to rapidly disseminate and colonize bone and brain.
We analyzed the transcriptome of the parental and BrM3 derivatives under basal cell culture conditions. The PC9-BrM3 transcriptome was significantly enriched for the TCF4 classifier and the H2030-BrM3 cells showed a detectable albeit not statistically significant enrichment (Figure 3A). To evaluate the activity of WNT/TCF pathway in these cell lines we performed transcriptional reporter assays using the TOP/FOP reporter system (Korinek et al., 1997). The BrM3 cells in both systems showed higher basal TCF transcriptional activity (2-fold and 2.8-fold, respectively) compared to the corresponding parental lines (Figure 3B). Furthermore, addition of the ligand Wnt3A caused a strong increase in TCF activity in the BrM3 sub-populations, which significantly surpassed the Wnt3a-induced increase in the parental lines (Figure 3B). These results revealed the presence of a deregulated TCF pathway in both metastatic cell populations.
β-catenin is distributed as an abundant membrane associated pool and a free labile pool available for WNT/TCF signaling. The stabilization and nuclear localization of this free pool are necessary for β-catenin as a co-activator of TCF-dependent transcription (Clevers, 2006; Klaus and Birchmeier, 2008; Logan and Nusse, 2004; Moon et al., 2004). Immunoblotting analysis revealed nuclear accumulation of β-catenin following Wnt3a treatment in all adenocarcinoma lines (Figure 3C). Notably, both the cytosolic and nuclear/signaling levels of β-catenin were higher in BrM3 cells compared to parental H2030 cells (Figure 3C). Phosphorylation of β-catenin at Ser33/37 and Thr41 marks the signaling pool for proteosomal degradation (Clevers, 2006). Upon Wnt3A stimulation, β-catenin was more rapidly dephosphorylated at Ser33/37 and Thr41 in H2030-BrM3 cells, indicating a greater rate of stabilization (Figure 3D). In the PC9 derivatives, β-catenin was abundant in both parental and BrM3 cells, suggesting the pre-existence of a hyperactive WNT pathway (Figure 3C). Dephosphoryation of Ser33/37 and Thr41 in response to Wnt3a was similar in the parental and BrM3 PC9 lines, suggesting that enhanced TCF activity in PC9-BrM3 cells may depend on additional alterations (Figure 3D).
Somatic mutations in APC or the β-catenin gene (CTNNB1) can increase β-catenin stability and activity in colorectal cancer (Clevers, 2006; Moon et al., 2004; Morin et al., 1997; Polakis, 2007). Mutations in these genes have been detected at low frequency in lung adenocarcinomas (Ding et al., 2008). In BrM3 and parental sets, we sequenced APC, CTNNB1 and 13 additional genes (APC2, AXIN1, AXIN2, GSK3B, STK11, LEF1, TCF7/TCF1, TCF7L1/TCF3, TCF7L2/TCF4, FRZB, SFRP1, SFRP2 and SFRP4) that encode proximal components of the WNT pathway. No mutations were found to be specifically associated with BrM3 populations, except for a silent mutation in SFRP4 (G to A, Chromosome 7: 37918251) in H2030-BrM3 cells. qRT-PCR and GeneChip array analysis of 84 WNT pathway related genes (including those sequenced) revealed a common up-regulation of LEF1 in highly metastatic cells (Figure 3E,F) but no consistent alterations in the expression of any of the other genes (data not shown). These results suggest that hyperactivation of WNT/TCF in these highly metastatic cells is not associated with mutations in core components of the WNT pathway but to other alterations that commonly converge on WNT transcriptional effectors.
Because the TCF4 signature was obtained from a colon cancer cell line (van de Wetering et al., 2002) we asked whether the metastatic potential of lung adenocarcinomas also correlates with a WNT gene-expression response derived specifically from lung adenocarcinoma cells. We generated transcriptomic gene sets from Wnt3a-treated versus untreated cells in both the H2030 and PC9 systems, combining the responses in both parental and metastatic derivatives. To obtain a more comprehensive transcriptional readout, we employed approaches based on a correlation coefficient method (Chang et al., 2005; Saal et al., 2007; Xu et al., 2008). With this method, the expression trends of many pathway-associated genes can be converted into a continuous score that avoids arbitrary expression level criteria and is applicable across multiple data sets.
When tested in our 107-tumor lung adenocarcinoma cohort, the resulting H2030 and PC9 WNT-responsive gene sets were independently associated with metastatic recurrence to multiple organ sites (Supplementary Figure 3A,B). PC9 and H2030 cells express WNT responsive genes which only partial overlap. These differences may reflect pathway activation in the context of different tumor developmental histories. To derive a broader WNT/TCF pathway classifier from these two gene sets, we combined probe sets that were regulated by Wnt3a in at least one experimental system with similar expression trends in the other and which were still associated with clinical outcome in MSKCC set 1 (refer to Experimental Procedures). This yielded a lung cancer WNT gene set (LWS), consisting of 91 probes sets including 81 known genes (Supplementary Table 2).
The LWS displayed improved prediction of metastatic relapse compared to the TCF4 classifier in the initial cohort of primary lung adenocarcinomas (MSKCC set 1, p=0.001; Figure 4A) and notably, in an independent lung adenocarcinoma data set (MSKCC set 2, p=9.6×10-5; Figure 4A) and among lung adenocarcinomas in a cohort of mixed lung tumors (SMC set, p=0.0025; Figure 4B). Importantly, the LWS was associated with recurrence in stage I lung adenocarcinomas (Figure 4C), and did not predict metastasis in squamous lung carcinomas (Figure 4B), or in breast carcinomas (Padua et al., 2008) (Figure 4D). Consistent with our model systems, the LWS predicted distant lung adenocarcinoma relapse to both brain and bone (Supplementary Figure 3C,D). Although the H2030-BrM3 or PC9-BrM3 were brain isolates, we were unable to obtain from these cells a common signature that would predict brain relapse using methods that previously revealed organ-specific metastasis signatures (Minn et al., 2005a). To determine the association of LWS status with overall patient survival we tested four independent stage I lung adenocarcinoma cohorts from a multi-institutional study (Shedden et al., 2008). In three out of the four cohorts tested, the LWS predicted poor survival in stage I adenocarcinomas (Supplementary Figure 4A-E). Altogether these results validate the association of WNT hyperactivation with distant multi-organ metastatic potential of lung adenocarcinomas.
By using the LWS in a Pearson correlation coefficient score, we compared brain samples from lung adenocarcinoma patients (n=28; Figure 4E) to either primary lung adenocarcinomas from patients who remained disease-free over 5 years (n=53) or primary lung adenocarcinomas from patients who recurred in any site (n=43). Lung adenocarcinoma metastases to bone are rarely resected, and were not available for study. The expression trends of LWS genes in brain metastases resembled more closely primary adenocarcinomas that recurred (Figure 4E). Thus, the activation of WNT/TCF appears to be maintained during brain metastatic progression in clinical samples.
To evaluate the functional role of WNT/TCF signaling in lung adenocarcinoma metastasis, we disrupted this pathway by expressing dominant negative TCFs (dnTCF1 and/or dnTCF4) that lack the n-terminal binding site for β-catenin (van de Wetering et al., 2002). Stable expression of dnTCF4 in H2030-BrM3 cells partially decreased the TCF activity (Supplementary Figure 5A,B) and metastatic activity of these cells (Figure 5A). Co-expression of dnTCF1 and dnTCF4 in PC9-BrM3 cells (Supplementary Figure 5C,D) abated the metastatic activity of these cells in different organs in the same mouse (Figure 5B). This reduction in TCF activity had little effect on BrM3 pulmonary growth (Figure 5C,D).
To identify mediators of WNT pro-metastatic functions, we focused on a subset of LWS genes whose expression level was strongly correlated with the metastatic potential of our cell derivatives and are implicated in developmental programs. Three genes that fulfilled these criteria were lymphoid enhancer-binding factor 1 (LEF1), homeobox B9 (HOXB9), and bone morphogenetic protein 4 (BMP4). HOXB9 is a TCF4 target (Hatzis et al., 2008) and belongs to the homeobox transcription factor gene family, which is critical for embryonic segmentation and limb patterning (Abate-Shen, 2002). LEF1 encodes a transcriptional effector of the canonical WNT pathway in addition to being a target of Wnt3a and TCF4. LEF1 upregulation by WNT may mediate signal amplification during malignancy (Hovanes et al., 2001). BMP4 belongs to the TGF-β superfamily, can be induced by WNT and regulates stem cell differentiation and multiple developmental processes (Shu et al., 2005; ten Dijke et al., 2003; Varga and Wrana, 2005).
The expression level of LEF1 and HOXB9 was marginal in parental H2030 cells, but was increased and potently induced by Wnt3a in the H2030-BrM3 cells (Figure 5E). In parental PC9 cells, Wnt3a addition augmented the expression level of LEF1 and HOXB9, whereas these genes were highly expressed and no further inducible by Wnt3a in PC9-BrM3 cells (Figure 5E). Expression of dnTCFs decreased the expression of LEF1 and HOXB9 to Wnt3a in BrM3 cells (Figure 5F). Thus the response patterns of LEF1 and HOXB9 are consistent with a constitutive hyperactivation of WNT/TCF signaling in PC9 cells (refer to Figure 3 C,D) and a hypersensitivity to Wnt stimulation in H2030 cells.
To achieve more specific inhibition of putative WNT/TCF pro-metastatic genes, we decreased LEF1 or HOXB9 by stable RNAi-mediated knockdown in BrM3 derivatives (Supplementary Figure 6A). Knockdown of LEF1 or HOXB9 significantly decreased the ability of H2030-BrM3 and PC9-BrM3 cells to form bone and brain metastases (Figure 6A and Supplementary Figure 6B). Knockdown of BMP4 had no effect on metastasis (data not shown). Reduction of LEF1 or HOXB9 did not affect the growth of PC9-BrM3 cells implanted in the lung (Supplementary Figure 7A). In H2030-BrM3 cells, HOXB9 knockdown did not affect lung and pleural growth rates whereas LEF1 knockdown had a partial inhibitory effect (Supplementary Figure 7B). These results are consistent with the tumor reinitiating advantage of H2030-BrM3 but not PC9-BrM3 cells over their parental counterparts in the lungs (refer to Supplementary Figure 2D). Moreover, overexpression of LEF1 and HOXB9 in the parental H2030 and PC9 cell lines increased their metastatic activity from circulation to bone and brain (Supplementary Figure 8 and Figure 6B-E). Collectively these results suggest that HOXB9 and LEF1 can enhance the competence of lung adenocarcinoma for brain and bone metastasis.
The biological properties of LEF1 and HOXB9 were further studied ex-vivo using the H2030-BrM3 population, which has a low threshold for exogenous WNT activation. Given the link between WNT/TCF and metastasis to multiple organs, we focused on the general functions of metastatic outgrowth and invasion. Knockdown of LEF1 and HOXB9 did not affect general cell viability in monolayer culture (Supplementary Figure 2A-B). However, when cultured in Matrigel, H2030-BrM3 cells formed compact spherical colonies that grew larger in size (Figure 7A) and in number compared to the parental H2030 cells (Figure 7B, black bars). Wnt3a addition further increased the colony outgrowth (Figure 7B, yellow bars), consistent with the tumor re-initiation advantage of these cells when implanted in the lung. Interestingly, knockdown of LEF1 expression decreased the size (Figure 7A) and number of the metastatic H2030-BrM3 colonies (Figure 7B), whereas decreasing HOXB9 expression had no effect.
To analyze metastatic invasion, we measured the infiltration of these cells through Matrigel in a modified Boyden chamber assay (Figure 7C). Parental H2030 and H2030-BrM3 cells showed a similar basal ability to infiltrate Matrigel (Figure 7D). However, given the strong astrocytic reaction elicited at the invasive edge of metastatic lesions in vivo (refer to Figure 2I) we repeated the invasion assays using astrocyte-conditioned media as a source of chemoattractant in the bottom chamber. Under these conditions, the invasive activity of H2030-BrM3 was significantly higher than that of parental cells (Figure 7D). Pre-incubation of H2030-BrM3 cells with Wnt3a increased their invasion towards stromal conditioned media (Figure 7D, blue bars). Wnt3A stimulated the invasion of the parental PC9 cells, while PC9-BrM3 cell displayed higher basal invasiveness (Supplementary Figure 9) consistent with their expression of LEF1 and HOXB9. Wnt3a did not act as a chemoatractant, since invasion was unaffected by the addition of Wnt3a to the bottom chamber (data not shown). Notably, shRNA-mediated knockdown of either LEF1 or HOXB9 inhibited invasion in H2030-BrM3 as well as PC9-BrM3 cells (Figure 7E). The metastatic potential of BrM3 cells and their Wnt3A responsiveness were not associated with changes in the TCF target c-myc, nor were these phenotypes linked to markers of epithelial-mesenchymal transition, such as loss of E-cadherin or gain of vimentin expression (Supplementary Figure 10).
The evidence presented here suggests that a hyperactive WNT/TCF pathway enhances the ability of human lung adenocarcinomas to develop brain and bone metastases. Two WNT target genes, LEF1 and HOXB9, are identified as promoters of lung adenocarcinoma metastasis and mediators of chemotactic invasion and colony outgrowth. These findings provide insights into the determinants of the typically rapid, multi-organ metastatic progression of this disease.
We have generated a lung cancer WNT gene response set –the LWS– that identifies lung adenocarcinomas with aggressive biological characteristics and poor clinical outcome. WNT/TCF pathway activation as detected by gene expression is associated with metastasis to brain and bones, two prominent sites of distant relapse in lung adenocarcinoma. The LWS can distinguish early stage I lung adenocarcinomas from independent cohorts of patients that are most likely to progress. However, it does not predict relapse in squamous lung carcinoma or breast carcinoma. The transcriptional output of WNT activation is context-dependent (Clevers, 2006; Logan and Nusse, 2004) and we do not exclude the possibility that other WNT targets or different pathway combinations may also contribute to lung cancer progression. Nevertheless, our evidence indicates a unique association between the WNT/TCF pathway and metastatic propensity in human lung adenocarcinomas.
Hyperactivation of WNT/TCF is a uniform property of highly metastatic subpopulations that we obtained from different human lung adenocarcinoma cell lines. These cells lines correspond to two distinct and defined subtypes of lung adenocarcinoma –the EGFR and KRAS molecular subtypes. KRAS and EGFR mutations are present in approximately 25% and 20% of human lung adenocarcinomas respectively (Rodenhuis et al., 1988; Sharma et al., 2007), and have been validated as oncogenes. However, lung adenocarcinomas initiated by these mutations require additional alterations in order to metastasize to distant sites (Ji et al., 2006; Johnson et al., 2001; Politi et al., 2006). The fact that these cell lines were derived from lymph node samples suggests that the acquisition of a hyperactive WNT/TCF pathway within a subset of malignant cells preceded their colonization of distant organs.
The hyperactivation of WNT/TCF in our model of metastatic lung adenocarcinomas does not correlate with a preponderance of typical somatic mutations in this pathway. Mutations in WNT pathway components are common tumor-initiating events in colorectal and gastric cancers but are less frequent in lung adenocarcinomas (Ding et al., 2008). Deregulation of this pathway may result from additional epigenetic alterations. Interestingly, APC methylation has been noted in metastatic non-small cell lung cancers (Brock et al., 2008). While the diverse mechanisms that regulate the WNT pathway in our models remain to be elucidated, an end result of this process is the elevated expression of at least two transcription factors, LEF1 and HOXB9 that encode different pro-metastatic functions, some of which are not required for tumor growth in the lung. The activation in primary tumors of pro-metastatic functions that provide no local growth advantage is not without precedent (Karnoub et al., 2007; Padua et al., 2008). The hyperactivity of the WNT/TCF pathway in lung adenocarcinomas observed here might reflect a presence of Wnt ligands in the tumor stroma, a feature of the tumor cell of origin, an epigenetic bystander alteration of WNT/TCF pathway components, or a combination of these mechanisms.
LEF1 and HOXB9 are direct transcriptional targets of the TCF4 transcription factor (Hatzis et al., 2008). HOXB9 is part of a homeobox family gene cluster whose expression is normally restricted to embryogenesis (Abate-Shen, 2002). HOXB9 expression has been reported in lung cancer cell lines (Calvo et al., 2000), but its function is unknown. LEF1 is a prototypical transcriptional mediator of WNT gene response, and its activation may amplify pathological WNT signaling (Hovanes et al., 2001). Moreover, LEF1 can interact with co-factors independently of β-catenin that cooperate in WNT transcriptional responses (Arce et al., 2006). High basal levels of LEF1 in metastatic cells may increase transcriptional responses without additional changes in β-catenin.
In our lung adenocarcinoma model, WNT/TCF hyperactivation can stimulate bone and brain metastasis independently of changes in intrinsic tumor cell proliferation or of a general growth advantage. We show that HOXB9 and LEF1 increase cancer cell invasion and LEF1 promotes tumor colony outgrowth in extracellular matrix. Interestingly, the phenotype of our metastatic derivatives resembles that described for bronchioalveolar stem cells (BASCs) with reduced differentiation and increased self-renewal properties (Kim et al., 2005). BASCs exhibit enhanced WNT activity following injury (Zhang et al., 2008). Thus HOXB9 and LEF1 may fulfill progenitor cell functions that are re-enacted during metastatic progression. The stem cell reprogramming gene c-MYC is a target of the LEF/TCF pathway in colon cancer cells (He et al., 1998) but not in the lund adenocarcinoma lines studied here. However, we do not exclude the possible involvement of other WNT target genes in lung adenocarcinoma metastasis.
The present findings illuminate a central concept in lung adenocarcinoma progression, namely, the role of a pro-metastatic signal that is acquired in primary tumors and enables their rapid metastatic spread. Unlike breast or prostate cancer metastasis, which may develop years to decades after the removal of a primary tumor, lung adenocarcinomas are typically competent to metastasize rapidly to multiple organs. The long latency of metastasis in breast or prostate carcinoma implies the need for further malignant evolution of the disseminated cancer cells in the infiltrated organs. In contrast to the protracted evolution of cancers which undergo long periods of remission before clinical metastasis, the short latency of relapse in lung adenocarcinomas predicts their acquisition of a pro-metastatic program early on and a lesser degree of organ speciation. Our identification of a mechanism that provides overall metastatic competence in early stage lung adenocarcinomas is in agreement with these predictions.
Intra-cranial relapses are a significant source of failure following treatments with chemotherapy (Chen et al., 2007) and targeted drugs (Omuro et al., 2005). While animal models of lung cancer represent a critical venue to address this problem, genetically modified mouse models of lung cancer reported to date yield metastatic tumors that occur in visceral organs but not in the brain (Jackson et al., 2005; Zheng et al., 2007). While not without limitations, our experimental systems recapitulate important phenotypic and molecular features of lung adenocarcinoma, including brain and bone metastasis, and a hyperactive WNT/TCF pathway in the background of relevant oncogenic mutations. This model may be useful for achieving a deeper understanding of early metastatic events and the development of improved treatments for lung adenocarcinoma patients at risk for metastasis.
Additional methods for cell culture, in vivo selection, in vitro assays, immunostaining procedures, retroviral shRNA targets, sequencing, and gene expression analysis are provided in a Supplemental Section.
Animal procedures were in accordance with the MSKCC Institutional Animal Care and Use Committee. Xenograft were performed on mice with SCID mutation in a nonobese diabetic background (NOD.SCID), or mice carrying the nude mutation crossed to the CBA/N X-chromosome-linked immune defect and beige mutations (bg-nu-Xid). Animals were aged-matched between 4-6 weeks. NSCLC cell lines were engineered to stably express a triple modality vector encoding GFP-luciferase fusion (Minn et al., 2005b). For experimental metastasis assays, between 104 and 105 cells were re-suspended in 0.1 mL PBS and injected into the right ventricle. Metastasis was detected by bioluminescence using an IVIS 200 Xenogen and confirmed by X-ray, MRI, and/or histology. Incidence of metastasis was quantified based on the luminescent signal in limbs and/or brain at a given time point and presented as Kaplan-Meir curves. For lung tumorigenesis, 104 cells were re-suspended in a 1:2 mixture of PBS and growth factor reduced Matrigel (BD Biosciences), and injected into the lung through the left rib cage of mice as previously described (Doki et al., 1999). Alternatively, cells were injected into the lateral tail vein. Growth rate in the lung/pleura was measured as a function of lung photon flux normalized to day 0 in live animals. Tumorigenic growth rates were similar when implanted subcutaneously. Metastasis phenotype was independent of animal sex.
The TOP-Flash and FOP-Flash promoters were subcloned into pGL4.82-Renilla luciferase constructs (Promega). Cells were transfected using Lipofectamine 2000 and selected for two days in puromicin. Transfected cell were then re-seeded into 24 well plates at 50,000 cells/well before treatment with recombinant Wnt3a (50 -100ng/ml; R&D Systems) for 12 hours. Reporter assays were performed according to the manufacturer's instructions (Promega). Promoter activity was calculated as the ratio of specific TOP-Flash over non-specific FOP-Flash relative renilla luciferase units (RLU).
Subconfluent parental and metastatic derivatives were serum starved overnight, and either mock treated or stimulated with recombinant Wnt3A (75-100ng/ml) for 3 hours. RNA was extracted from duplicate samples of each condition using the RNeasy mini kit (Qiagen). Labeling and hybridization of the cell line samples to the HG-U133A_2 gene expression chip (Affymetrix) was performed by the MSKCC Genomics Core Facility. Microarray data from a cohort of 107 lung adenocarcinoma tumors (MSKCC set 1), a second cohort of 129 adenocarcinomas (MSKCC set 2) including tumors from Chitale al. (Chitale et al., 2009) as well as 28 brain metastasis, were analyzed on Affymetrix HG-U133A and HG-U133A_2 platforms. A third data set containing a mix of lung adenocarcinomas and squamous carcinomas from the Samsung Medical Center (SMC set; HG-U133A Plus 2.0) was publicly available. For pathway and correlation coefficient analysis, raw CEL data were preprocessed using RMA algorithm implemented by the affy package of R statistical software (designated as R hereafter). Gene expression data of cell derivatives is deposited at GEO (GSE14107). Samsung Medical Center set was obtained from GEO (GSE8894). Raw data and clinical annotation for MSKCC sets 1, 2, and brain metastasis are available at http://cbio.mskcc.org/Public/lung_array_data/.
Genes regulated by KRAS and dnTCF4 were extracted from published results (Sweet-Cordero et al., 2005; van de Wetering et al., 2002) and matched to the corresponding Affymetrix probes (HG-U133A platform). The Affymetrix version of the TCF4 signature was used to cluster MSKCC set 1 in an unsupervised manner (by the “heatmap.2” function of “gplots” package of R). The cluster that overexpresses the majority of dnTCF4 genes was defined as TCF4+, and the other cluster was defined as TCF4-. The clinical outcomes of the two clusters were compared by log-rank test using the “survdiff” function of “survival” package of R. β-catenin, c-myc, SRC, E2F3, and TGF-β gene lists were obtained from previous results (Bild et al., 2006; Padua et al., 2008) and used for unsupervised analysis.
A combinatory approach based on correlation coefficient score and unsupervised clustering was used to evaluate the association between Wnt3A mediated transcriptional activation in lung adenocarincoma cell lines and relapse in MSKCC sets 1, 2, and SMC lung adenocarcinoma clinical cohorts. This was confirmed using pathway meta-analysis. A comprehensive description of these methods and the derivation of the Lung cancer WNT responsive gene set (LWS) are provided in supplementary material.
We would like to express deep gratitude to our late colleague William L. Gerald. We are indebted to H. Varmus, W. Pao, V. Rusch, and members of the Massagué laboratory for insightful discussions. We thank A. Heguy for supervising sequencing and mutational analysis. H. Clevers, E. Batlle, R. Moon, W. Pao, and H. Varmus generously provided constructs and cell lines. This work was supported by National Institutes of Health grant P01-129243 (M.G.K., M.L., W.L.G., and J.M.). DXN was the Bayer postdoctoral fellow of the Damon Runyon Cancer Research Foundation. ACC was supported by an ASCO CDA grant. JM is an Investigator of the Howard Hughes Medical Institute.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.