Lymphangioleiomyomatosis Biomarkers Linked to Lung Metastatic Potential and Cell Stemness

Lymphangioleiomyomatosis (LAM) is a rare lung-metastasizing neoplasm caused by the proliferation of smooth muscle-like cells that commonly carry loss-of-function mutations in either the tuberous sclerosis complex 1 or 2 (TSC1 or TSC2) genes. While allosteric inhibition of the mechanistic target of rapamycin (mTOR) has shown substantial clinical benefit, complementary therapies are required to improve response and/or to treat specific patients. However, there is a lack of LAM biomarkers that could potentially be used to monitor the disease and to develop other targeted therapies. We hypothesized that the mediators of cancer metastasis to lung, particularly in breast cancer, also play a relevant role in LAM. Analyses across independent breast cancer datasets revealed associations between low TSC1/2 expression, altered mTOR complex 1 (mTORC1) pathway signaling, and metastasis to lung. Subsequently, immunohistochemical analyses of 23 LAM lesions revealed positivity in all cases for the lung metastasis mediators fascin 1 (FSCN1) and inhibitor of DNA binding 1 (ID1). Moreover, assessment of breast cancer stem or luminal progenitor cell biomarkers showed positivity in most LAM tissue for the aldehyde dehydrogenase 1 (ALDH1), integrin-ß3 (ITGB3/CD61), and/or the sex-determining region Y-box 9 (SOX9) proteins. The immunohistochemical analyses also provided evidence of heterogeneity between and within LAM cases. The analysis of Tsc2-deficient cells revealed relative over-expression of FSCN1 and ID1; however, Tsc2-deficient cells did not show higher sensitivity to ID1-based cancer inhibitors. Collectively, the results of this study reveal novel LAM biomarkers linked to breast cancer metastasis to lung and to cell stemness, which in turn might guide the assessment of additional or complementary therapeutic opportunities for LAM.

Intriguingly, diverse data indicate that LAM cells originate from a different organ to the lung [9]; for example, LAM cells can be found circulating in the blood and lymphatic systems [18,19], and LAM lesions can reappear after lung transplantation, although not derived from the tissue donor [20,21]. Thus, a specific cell type(s) may possess metastatic potential with lung tropism when, most commonly, a TSC1 or TSC2 mutation is acquired and mTORC1 is abnormally activated. Interestingly, however, there are recorded cases of sporadic LAM without mutations in TSC1 or TSC2 and, therefore, without abnormal activation of mTORC1 [22]. mTORC1 regulates a cancer invasion and metastatic transcriptional program [23]. Notably, in breast cancer, relative low expression of hamartin and tuberin (TSC1 and TSC2 products, respectively) is associated with poor clinical outcome [24], and depletion of tuberin promotes metastasis [25]. Together, these observations led us to hypothesize that, beyond the proposed role for chemokines [26], the mediators of breast cancer metastasis to lung could also play a role in LAM. To assess this hypothesis, we analyzed breast cancer gene expression data and subsequently evaluated the presence of defined metastasis mediators in LAM tissue.

Breast cancer gene expression analyses
Data for gene expression profiles of metastatic breast cancer were taken from the corresponding publication [27] and from the Gene Expression Omnibus (GEO) references GSE11078 [28] and GSE2034 [29]. The osteosarcoma dataset was downloaded from GSE14827 [30].
Pre-processed and normalized gene expression data were downloaded from the corresponding repository of The Cancer Genome Atlas (TCGA) (July 3, 2012; http://tcga-data.nci.nih.gov/ tcga/tcgaHome2.jsp). Expression profiles were clustered based on the PAM50 signature, which assigns tumors to the basal-like, HER2-enriched, luminal A or luminal B breast cancer subtypes. The MCF7 RNAs were extracted using TRIzol (Life Technologies) and, following quality control and labeling, hybridized on the Human Genome U133 Plus 2.0 Array microarray platform (Affymetrix). The RMA method implemented in BioConductor was applied for background adjustment and normalization of intensity values. The data have been deposited under the GEO reference GSE68324. The Gene Set Expression Analysis (GSEA) [31] tool was run using default values for all parameters. The GSEA analyses used pathway annotations from the Kyoto Encyclopedia of Genes and Genomes (KEGG) [32].

shRNA assays
The short-hairpin (sh) RNA sequence targeting the expression of TSC2 corresponded to a construct from the MISSION (Sigma-Aldrich) library and has been described previously [33,34]. The lentiviral packaging, envelope, control and GFP expression plasmids (psPAX2, pMD2.G, non-hairpin-pLKO.1, scrambled-pLKO.1, and pWPT-GFP, respectively) were purchased from Addgene. Production and collection of lentiviral particles followed a modified Addgene protocol. Initial viral titers > 5 x 10 5 /ml were confirmed by Lenti-X GoStix (Clontech) and supernatants were then concentrated by ultracentrifugation or Lenti-X Concentrator (Clontech) and stored at −80°C. Concentrated viral supernatants were titrated for optimal inhibition of the target.

LAM patients
LAM patients were recruited and lung tissue samples collected by the centers participating in this study and with the support of the Spanish LAM Association (AELAM). Part of this cohort has been described previously [35]. All patients provided written informed consent and the study was approved by the ethics committees of the Bellvitge Institute for Biomedical Research (IDIBELL) and the Instituto de Investigación Sanitaria La Princesa (SEPAR-2012). LAM clinical data included year of birth, year of first symptom, all symptoms presented (included pneumothorax and chylothorax), lung transplantation, presence of angiomyolipomas and/or uterine myomas, treatment, smoker status (previous and current), and comorbidity with other diseases (S1 Table).

Immunohistochemistry
Immunohistochemical assays were performed using standard protocols with the EnVision (Dako) method. Each tissue and biomarker was evaluated in at least two independent assays and no substantial differences were observed. Equivalent sections were processed to include incubation with immunoglobulin controls (Sigma-Aldrich), which did not reveal staining in any case. Secondary antibodies for immunofluorescence (Alexa) were obtained from Molecular Probes (Life Technologies). The immunohistochemistry results were scored independently and blindly (to molecular and clinical status) by two expert pathologists. The association between biomarkers was assessed by computing the Spearman's correlation coefficient (SCC).

Cell culture
The cell lines were cultured following standard protocols and cellular viability was evaluated by performing methylthiazol tetrazolium (MTT, Sigma-Aldrich) assays. Everolimus was purchased from Selleck Chemicals and the inhibitors of ID1 expression or stability (apigenin, C527, and cannabidiol) from Sigma-Aldrich. The results correspond to two independent assays for each compound and to triplicates for each data point. Given the half-maximal inhibitory concentration (IC 50 ) and the maximal response to everolimus in Tsc2 -/-/Tp53 -/murine embryonic fibroblasts (MEFs) and Tsc2-deficient Eker rat leiomyoma (ELT3) cells, these cells were exposed to 1 and 100 μM of the rapalog, respectively.

Results
Low TSC1/2 expression in breast tumors that metastasize to the lung The link between TSC1/2 expression, mTORC1 activity and breast cancer metastatic potential was primarily investigated by analyzing publicly available gene expression datasets. Relatively low TSC2 expression was found to be significantly associated with lung but not bone metastasis events in the analysis of a seminal breast cancer dataset [27] (Fig 1A). Consistent with clinical observations [36], lung metastatic events were preferentially linked to ERa-negative tumors ( Fig 1B) and the above association was also significant in this subtype (P = 0.029). Thus, low expression of TSC2 (and of TSC1, although the univariate association was not significant: P = 0.09) was evident in the tumors that caused lung metastases ( Fig 1B). Consequently, TSC1 and/or TSC2 were found to be significantly co-expressed (Pearson's correlation coefficient (PCC) P values < 0.05) in the expected direction of mediating lung metastasis with 10 of the 18 the genes that made up the seminal lung metastasis signature [27] (positively correlated: C10orf116, MAN1A1, and RARRES3; and negatively correlated: ANGPTL4, CXCL1, FSCN1, LTBP1, MMP1, PTGS2, and VCAM1; CXCR4 and LY6E were correlated in the opposite expected direction; Fig 1B).
Beyond the specific analysis of TSC1/2, a pathway-based analysis of the same breast cancer dataset suggested an association between mTORC1 activity and lung but not bone metastasis events ( Fig 1C). In this analysis, suppressors and activators of the mTORC1 pathway pointed to increased signaling linked to lung metastatic potential ( Fig 1C). Moreover, consistent with the role of mTOR as a nutrient sensor, significant associations were also found between metabolic pathways and lung but not bone metastasis ( Fig 1D).
Next, analysis of an independent dataset from primary breast tumors [28] also showed low TSC1 and TSC2 expression to be specifically associated with lung metastasis (S1A Fig). The same dataset also revealed an analogous association with differential expression of genes from the mTOR pathway (S1B Fig). Moreover, analysis of a third dataset [29] confirmed the association between relatively low TSC2 expression and poor survival of patients with ERα-negative In addition, analysis of the TCGA breast cancer dataset revealed significant under-expression of TSC1/2 in primary tumors that are expected to preferentially metastasize to lung (basal-like type [36]; two-tailed t-test P values < 0.001) and, consistently with this observation, TSC1/2 were found to be co-expressed with most of the genes that constitute the seminal lung metastasis signature [27]: 13 of the 18 signature genes showed significant PCCs in the expected direction of mediating lung metastasis (positively correlated: C10orf116 and RARRES3; and negatively correlated: ANGPTL4, CXCL1, FSCN1, ID1, KYNU, KRT81/ KRTHB1, LTBP1, MMP1, PTGS2, TNC, and VCAM1; Fig 1E). Lastly, in support of the above observations, depletion of tuberin expression by a short-hairpin RNA transduced into breast cancer cells with wild-type TSC2, MCF7, revealed a significant change in the expression of the lung metastasis signature (P = 0.002); thus, genes that when up-regulated mediate lung metastasis showed a trend to be up-regulated with tuberin depletion, and the opposite trend was observed for the down-regulated set ( Fig 1F).
Remarkably, although the lung is the most common metastatic site for other cancer types, in particular for osteosarcoma, analysis of an equivalent microarray dataset [30] did not reveal similar associations (S2 Fig). Collectively, these results suggest a specific link between loss of TSC1/2 expression-and probably, therefore, activated mTORC1 signaling-and lung metastatic potential in breast cancer.

Biomarkers of lung metastatic potential in LAM lesions
Having suggested an association between TSC1/2 expression and the lung metastatic potential of breast cancer, we performed immunohistochemistry studies of the lung metastasis mediators in paraffin-embedded lung tissue from 23  Next, several antibodies that may recognize proteins encoded by the breast cancer lung metastasis signature [27] were assessed by immunohistochemical assays. Positivity in LAM lesions was revealed for FSCN1 and ID1 (Fig 2A). FSCN1 is an endothelial cell biomarker that also identifies differentiated luminal and spindle-like cells in normal breast tissue (Fig 3), and which has been extensively linked to the metastatic potential of cancer cells [37]. Notably, the staining pattern of FSCN1 in LAM lesions was found to be more extensive than that of a normal endothelial biomarker, CD34 (Fig 2B). In addition, all cases showed positivity for both metastatic biomarkers (S1 Table) and their intensity scores were somewhat correlated (SCC = 0.38, one-tailed P = 0.041).
The subcellular localization of ID1 was prominently cytoplasmic in most cases (Fig 2A, bottom left panel), although major nuclear localization was also observed (Fig 2A, bottom right  panel). Notably, cytoplasmic-nuclear shuttling of ID proteins regulates their function and cytoplasmic ID1 has been implicated in active angiogenesis [38]. Examination of normal breast and lung tissue for ID1 expression revealed weak positivity (in luminal cells) and negativity, respectively (Fig 3). Together, this study reveals the presence of two biomarkers (defined as linked to a specific biological function) of lung metastatic potential in LAM lesions. TSC1/2 and genes from lung metastasis signature, in the TCGA dataset [70]. (F) GSEA results for the expression difference of the lung metastasis signature between MCF7 cells transduced with control or TSC2-target shRNAs. The left top panel shows the results for absolute expression differences, and the middle and bottom panels show the results for real differences of the up-regulated and down-regulated subsets of the signature, respectively. The right panels show the Western blot results for tuberin, pS6 and control loading, TUBA. Cell stemness biomarkers in LAM lesions mTORC1 regulates hematopoietic stem cell homeostasis and FSCN1 is a key mediator of this function [39]. In addition, the ID proteins regulate stem cell phenotypes [40] and, particularly, ID1 maintains embryonic stem cell self-renewal [41]. In breast cancer, over-expression of canonical stem and/or progenitor cell biomarkers, such as ALDH1 and CD61, has been associated with poor-prognosis [42,43]. CD61 defines a mammary cell population postulated to contribute to the origin of the tumor subtype that preferentially metastasizes to lung [44]. In this setting, SOX9 has a key role conferring lung metastasis-seeding properties and, intriguingly, has also been identified as a lung stem cell biomarker [45][46][47]. In normal breast tissue, ALDH1, CD61 and SOX9 (and FSCN1) also mark spindle-like cells within the loose specialized intra-lobular stroma and/or surrounding the terminal extra-lobular ducts (Fig 3), which is consistent with previous observations for other stem/progenitor cell markers of mammary development [48]. The above observations led us to assess the expression of ALDH1, CD61 and SOX9 in the LAM lesions. The results showed positivity in 90%, 63%, and 77% of the cases, respectively (Fig 4 and S1 Table). There were no significant correlations between the expression of these and/or the above biomarkers; nevertheless, CD61 showed a trend for positive correlations (SCCs > 0.20) with FSCN1, ID1 and SOX9, but a negative correlation (SCCs < -0.20) with ALDH1. Studies in larger tissue series may be required to corroborate these trends.
Analogous to the observations for ID1, SOX9 showed heterogeneity between cytoplasmic and nuclear sub-cellular staining (Fig 4, SOX9 left and right panels, respectively); however, normal differentiated lung was found to be mostly negative for this biomarker (Fig 3). Considering the potential influence of SOX9 in LAM pathogenesis, cytoplasmic positivity has been associated with invasive breast cancer and metastasis, but normal differentiated mammary epithelia generally exhibit nuclear localization (Fig 3) [45]. In addition, the staining of SOX9 was also found to be heterogeneous intra-LAM tissue (Fig 4, right panels). Notably, in normal breast tissue, pS6 and SMA staining identifies different cell populations, including the basal layer of ducts [49,50]. HMB-45 is generally negative in this tissue, but positivity has been observed in breast tumors with a myoepithelial or melanocytic phenotype [51,52].
The potential for differential expression correlations and the observed heterogeneity could indicate the existence of diverse cell status and/or phenotypes in LAM lesions. This observation has been raised previously [7,[53][54][55] and can be appreciated in the results for the canonical biomarkers; for example, by comparing the expressions of ERa, PR and SMA, the former being much less representative in a given lesion (Fig 5A). In this regard, pS6 staining also indicated potential intra-tissue heterogeneity for mTORC1 activity (Fig 5B, top panel), which in this case appeared to correlate with the positivity of ID1 but not of FSCN1 (Fig 5B, bottom panels). This complexity was further exposed by double immunostaining assays, which revealed partial colocalization for these biomarkers and SOX9; this observation was more evident between FSCN1 and SOX9, which has an unknown functional significance but could indicate protein co-expression ( Fig 5C). Moreover, in large LAM lesions, pS6 staining was principally apparent at the periphery of the lesion (Fig 5D), which would suggest a link to active cell proliferation. Collectively, the data reveal novel biomarkers of LAM that are associated with breast cancer stemness features and further support the existence of relevant heterogeneity among LAM cells and/or lesions.

Assessment of biomarkers in Tsc2-deficient cell models
To further evaluate the relevance of the novel LAM biomarkers, we analyzed their expression level in two cell models of tuberin deficiency: MEFs derived from littermate embryos with the Tsc2 -/-/Tp53 -/genotype and Tsc2-deficient ELT3 (V3) cells. These models were compared to their control counterparts: Tsc2 +/+ /Tp53 -/-MEFs and ELT3 reconstituted with human TSC2 (T3), respectively. Thus, Western blot analyses revealed over-expression of FSCN1 and ID1, particularly in the Tsc2-deficient MEFs (Fig 6A). Treatment with everolimus also revealed differences between the models; ELT3 cell lines showed a higher degree of down-regulation of FSCN1 and ID1 with exposure to the rapalog (Fig 6B). Cell type and/or signaling interplay and molecular specificities might explain the differential consequences of mTOR inhibition on the expression of the biomarkers.
Since all LAM lesions revealed ID1 positivity and compounds are available that target the expression or stability of this factor in cancer cells (apigenin, C527, and cannabidiol) [56][57][58][59], we next sought to evaluate the selective effect of these compounds in Tsc2-deficient cells; particularly in MEFs because this setting revealed a better dose-response relationship for everolimus ( Fig 6C). Yet, none of the three ID1 inhibitors revealed higher sensitivity in Tsc2-deficient cells (Fig 6D). Surprisingly, cell exposure to the inhibitors led to a relative increase in ID1 expression (Fig 6E). This effect, which is opposite to the effect previously described in cancer cells, might be due to the specificity of the models and/or to interactions with additional targets for those compounds. As novel ID1 inhibitors may be developed, higher specificities may be required for the evaluation in Tsc2-deficient cells.

Discussion
In this study we test the hypothesis that the metastatic properties of LAM cells could be further depicted using knowledge of breast cancer tropism to lung, and identify the expression of metastatic mediators and cancer cell stemness molecular determinants in LAM lesions. The hypothesis was led by the observation that primary breast tumors with relatively low TSC1/2 expression and predicted enhancement of mTORC1 signaling preferentially metastasize to lung. Consistently, significant co-expression (in the direction of promoting metastasis) was identified between TSC1/2 and several of the genes that made up a seminal lung metastasis signature. In addition, depletion of TSC2 in MCF7 breast cancer cells led to a significant gene expression activation of the signature. Collectively, these results are in agreement with previous studies demonstrating an association between low expression of hamartin or tuberin and poor breast cancer clinical outcome [24], between enhanced AKT-mTOR signaling and poor postrelapse breast cancer survival [60], and between depletion of tuberin in mice and breast cancer progression and metastasis [25]. While these observations could indicate a molecular and/or biological link between breast cancer metastatic potential and LAM pathogenesis, the precise role of the identified biomarkers downstream of mTOR signaling remains to be determined. There is growing evidence for the involvement of mTOR in the regulation of stem cell biology [39,61,62], thus the biomarkers might be linked to a down-stream transcriptional program characteristic of this setting.
The function of FSCN1 and ID1 in promoting normal angiogenesis and cancer metastasis is well known [27,37,63]. In addition, the ID proteins are members of a family of basic Helix-Loop-Helix (bHLH) transcription factors whose expression has been shown to be deregulated in different cancer types, being associated with poor-prognosis [63,64]. Over-expression of IDs promotes the maintenance of a stem cell phenotype [40], and the identification of ALDH1, CD61 and SOX9 further endorses the presence of a stem cell-like phenotype in LAM lesions [65]. Importantly, since all LAM lesions revealed ID1 positivity, compounds targeting this factor could potentially exhibit therapeutic benefit in LAM; however, the compounds tested in this study did not show specificity in Tsc2-deficient cells or the expected molecular down-regulation [56][57][58][59], which would discourage their use in combination with rapamycin or rapalogs. Additional studies may be warranted using specific compounds targeting ID1 function and/or based on the newly defined biomarkers. The potential of using this information to target LAM is further supported by the observation that cyclooxygenase-2, encoded by PTGS2, is overexpressed in TSC2/Tsc2-deficient cells and constitutes a therapeutic target for LAM [66]; importantly, PTGS2 is also included in the seminal breast cancer lung metastasis signature [27], being anti-correlated with TSC1/2 (this study).
The immunohistochemical results also showed inter-and intra-tissue heterogeneity, which further suggests the involvement of LAM cells of different types and/or different proliferation/ differentiation status. Although differences in proliferation are likely to exist based on the staining results for pS6 at the front of large LAM lesions, the observed molecular and phenotypic differences could also have been initiated in the tissue of origin and/or through metastatic spread [67]. In this regard, it has been shown that LAM cells proliferate extensively in the lymphatic system and line the thoracic duct, possibly making metastasis to the lung a simple mechanistic phenomenon. Nonetheless, identification in LAM lesions of the expression of specific mediators of lung metastasis could indicate that these molecules particularly facilitate the seeding and/or colonization of the lung. For different cancer types, it has been clearly shown that specificities exist for the preferred tissue/organ of metastasis depending on the molecular profiles of the tumors [68]. The associations observed here for breast cancer metastasis are in agreement with this concept. However, the functional role of the defined biomarkers cannot be clearly demonstrated unless in vivo depletion studies are performed and/or the tissue/organ of origin of LAM cells is defined in each patient. Since the cell origin(s) of LAM is unknown, the expression levels of biomarkers cannot be definitively categorized (relative to the cell origin).
Given the functional relevance of the sub-cellular localization of ID1 and SOX9 [38,45], the differences observed across LAM cases could potentially inform about disease diagnosis and progression. Intriguingly, common genetic variation in SOX9 has been associated with lung function in the general population [69], which further leads us to speculate that SOX9 function could act as a modifier of LAM pathophysiology. Moreover, SOX9 expression characterizes alveolar bipotential progenitors and surfactant-secreting cuboidal alveolar type 2 cells [47], which are altered in related lung diseases such as idiopathic pulmonary fibrosis. Additional studies on the cell origin and on related cancer types, such as smooth muscle sarcomas and metastasizing uterine leiomyoma, are also required to evaluate the functional relevance of these observations. Collectively, the results of our study suggest that further understanding of the molecular and cellular basis of breast cancer metastasis to lung may complementarily reveal fundamental insights for LAM.

Conclusions
Starting from the observation that breast tumors which preferentially metastasize to lung are characterized by low TSC1/2 expression and possibly by activated mTORC1 signaling, this study reveals five novel biomarkers for LAM lesions: two recognized lung metastasis mediators, FSCN1 and ID1, and three recognized stem and/or progenitor cell regulators, ALDH1, CD61, and SOX9. The examination of these biomarkers further reveals heterogeneity between and within LAM cases. Together, these results might help to further decipher the pathological features and/or tissue origin of LAM cells.
Supporting Information S1 Table. LAM patients and immunohistochemical results. Clinical data and immunohistochemical results for ALDH1, CD61, FSCN1, ID1, pS6, and SOX9. (XLSX) S1 Fig. Replication of the association between mTOR pathway gene expression and breast cancer metastasis to lung. (A) Unsupervised clustering of TSC1/2 (top panels) and ESR1 (bottom panels) expression profiles across primary breast tumors that produced lung metastases (left panels) and those that produced non-lung metastases (right panels). The data correspond to the GEO GSE11078. All TSC1/2 probes revealed differential expression between the tumor groups (two-tailed t-test P values < 0.05). (B) GSEA graphical output for the association analysis between the gene set corresponding to the mTOR pathway and the differential expression between tumors that developed lung versus non-lung metastases. The top panel corresponds to the gene rank according to the P value of the t-test for differential expression; the bottom panel corresponds to the gene rank according to the t-test statistic value. The top genes whose expression differentiate the two tumors types are shown. The GSEA enrichment score and the nominal P values are also shown. (C) Kaplan-Meier survival curves based on categorization of TSC2 expression in tertiles, using GEO GSE2034. The top and bottom panels correspond to ERa-positive and ERa-negative tumors, respectively. The left panels correspond to the analysis of the complete dataset and the right panels correspond to the analysis excluding the 25% of tumors with the lowest (in the ERa-positive set) or the highest (in the ERa-negative set) ESR1 expression values. The P values of the log-rank test when using the complete dataset are shown.