SATB1 expression was shown previously to promote metastasis in established but non-metastatic breast cancer cells 
. To address whether SATB1 is able also to induce malignancy in non-tumorigenic cells, we examined SATB1 levels in eight non-tumorigenic and tumorigenic breast epithelial cell lines. The non-tumorigenic cell line analyzed, MCF10A, was established after spontaneous immortalization of primary human mammary epithelial cells derived from reduction mammoplasty 
. We examined two different sources of these cells, designated MCF10A-1 and MCF10A-2 and two other immortalized non-malignant cell lines, 184A1 and 184B5 that were derived from reduction mammoplasty of a normal human breast epithelial cells (specimen#184) after treatment with benzo(a)pyrene 
. In addition, we analyzed the MCF10A cancer progression series: premalignant neoT cells (also called MCF10AT; 
) expressing mutant HRAS
and malignant Ca1d (undifferentiated carcinoma with low metastatic potential) 
. Quantitative RT-PCR and immunoblot analyses showed that endogenous SATB1 levels were very low to undetectable in all immortalized non-tumorigenic cell lines tested in contrast to its easily detectable expression in metastatic human breast cancer cell lines, MDA-MB-231 and BT549 (). Endogenous SATB1 expression levels (for both mRNA and protein) in cells from the MCF10A progression series were detectable, but significantly lower- 1/6th
- than those in aggressive cancer cell lines ().
SATB1 is endogenously expressed in aggressive breast cancer cell lines and its ectopic expression induces malignant phenotype in non-malignant MCF10A-1, but not in MCF10A-2 cells.
To determine whether ectopic SATB1 expression induces a malignant phenotype in non-tumorigenic cells, we transduced MCF10A-1 and MCF10A-2 cells with a retroviral vector to overexpress SATB1 (pLXSN-SATB1) (). We also overexpressed SATB1 in neoT and Ca1d of the MCF10A cancer progression series (Fig. S1A) and examined the colony morphology in three dimensional (3D) cultures grown on top of Matrigel ( and Fig. S1B). Upon SATB1 overexpression, neoT and Ca1d cells underwent major morphological changes from spherical to spindle-shaped network structures (Fig. S1B). These altered colony morphologies resembled the structures formed by aggressive breast cancer cell lines, MDA-MB-231 and BT549, which express a high level of endogenous SATB1 (; Fig. S1B, right). shRNA-mediated SATB1 knockdown in MDA-MB-231 and BT549 cells reversed the colony morphologies from stellate to spherical structures as reported previously (; Fig. S1B, right) 
, suggesting a direct role of SATB1 in the stellate morphology in these cells. Interestingly, SATB1 overexpression also altered the colony morphology of MCF10A-1, from uniformly spherical to a mixture of spherical (white arrow) and spindle-shaped structures (blue arrow), representing 85% and 15% of the cell population, respectively (). On the other hand, SATB1 overexpression in another MCF10A-2, only led to the formation of slightly enlarged spheres (). This result suggested that these two MCF10A cultures possess different properties responsible for their differential responsiveness to SATB1 overexpression.
The abnormal 3D morphology of SATB1 expressing MCF10A-1 cells led us to ask whether ectopic SATB1 expression induces invasive properties in these cells. Invasion assays demonstrated that SATB1 overexpression elevated the invasive potential of multiple cell lines derived from MCF10A cells compared to the parental cell lines. For MCF10A-1 and neoT cells, the invasiveness increased by approximately 400-fold (). Ca1d cells showed the highest increase (more than 800-fold), while MCF10A-2 cells showed the least increase (). To examine whether the altered 3D morphology upon SATB1 expression reflected changes in cell polarity, we immunostained MCF10A-1 cells expressing control vector or SATB1 (pLSXN-SATB1) for filamentous actin (F-actin), ß-catenin and α6-integrin (). These proteins are markers for properly formed mammary acini, which are glandular-like structures with a hollow lumen surrounded by polarized epithelial cells 
. In 3D cultures, MCF10A cells form acinar structures that have basal polarity with a hollow center resembling a lumen 
, as evidenced by staining of control MCF10A-1 cells (, top row). Immunostaining of SATB1-expressing MCF10A-1 cells revealed that the spherical colonies had lost cell basal polarity, as indicated by reduced ß-catenin expression and disorganized F-actin and α6- integrin, and did not posses hollow lumen (, center row). Spindle-shaped colonies exhibited even more aberrant localization of these markers (, blue arrow; , bottom row). To examine the level of SATB1 expression in the spherical and spindle colonies, we immunostained SATB1-expressing MCF10A-1 cells for SATB1 on the 6th
day of 3D culturing (, far right). The results showed that SATB1 expression was found in both the depolarized spheroids and spindle- shaped colonies, with SATB1 detected in approximately 50% of the cells in a single colony. In comparison to the spherical colonies, the spindle colonies appeared to have a greater signal intensity for SATB1.
To validate the observed changes caused by SATB1 overexpression in MCF10A-1 cells, we isolated two single-cell-derived clones, clone-1 and clone-2, which expressed SATB1 at high levels from pooled population (Fig. S1C). Compared to the parental and vector control populations, clone-1 and clone-2 cells proliferated at significantly higher rates, whether on plastic in 2D or on top 3D gels (). Morphological analysis of both clones in 3D cultures showed predominantly spindle colonies with a few spherical colonies (Fig. S1C). In soft agar, both SATB1-overexpressing clones formed colonies with larger size and approximately 10-fold higher numbers over those formed by the vector control ().
SATB1 overexpression induces EMT in MCF10A-1 cells.
The SATB1-induced changes in colony morphology; invasive potential and anchorage-independent growth all support its role in driving epithelial-to-mesenchyme transition (EMT). EMT entails loss of epithelial cell polarity, cell-to-cell contact and cytoskeletal organization, and presages the development of aggressive phenotypes during cancer progression 
. EMT is characterized by upregulation of mesenchymal markers, fibronectin and vimentin, and loss of an epithelial marker, E-cadherin. To test if SATB1 overexpression indeed alters the expression of these EMT markers in MCF10A1 cells, we examined the levels of these proteins in MCF10A-1 derivatives including parental cells, control vector, clone-1 and clone-2 by immunoblot analysis. As positive control cell lines, we used MDA-MB231 cells to express each of two SATB1 shRNAs (shRNA1, shRNA2) 
(). SATB1 is known to directly regulate ERBB2
, an important regulator of breast cancer progression 
, in MDA-MB-231 cells 
. Therefore in this assay, we used ERBB2 as a positive control: SATB1 expression in MCF10A-1 cells increased ERBB2 expression, whereas depletion in MDA-MB-231 cells greatly decreased ERBB2 expression. SATB1 overexpression in MCF10A-1 cells also led to an increase in fibronectin and vimentin. Conversely, SATB1 depletion in aggressive MDA-MB-231 cells led to a loss or reduction of these proteins (). In particular, fibronectin, at the protein level, appeared to be entirely dependent on SATB1 expression; while at the transcript level, it was shown to be reduced only 2-fold in MDA-MB-231 upon SATB1 knockdown 
(). In contrast to fibronectin, ß-catenin levels were greatly reduced and E-cadherin levels were slightly decreased upon SATB1 overexpression in MCF10A-1 cells (). In conjunction with the immunoblot analysis, immunostaining experiments in clone-1 confirmed the SATB1-dependent regulation of fibronectin and ß-catenin. (Fig. S1D). These results suggest that SATB1 expression induces EMT and causes a gain of cancer phenotypes in non-malignant cells.
To verify the observations made in cell cultures, we tested if SATB1-expressing MCF10A-1 and MCF10A-2 cells could form tumors in nude mice. We injected SATB1-overexpressing MCF10A-1 cells (pooled pLXSN-SATB1 and single cell-derived clone-1) or vector control cells into the fat pad of the fourth mammary gland and monitored tumor formation. Both types of SATB1-overexpressing MCF10A-1 cells formed large tumors in all the injected mice, whereas control cells did not (, and Fig. S2A). Furthermore, to test whether SATB1-overexpressing MCF10A-1 cells exhibit elevated migration and invasion properties in vivo
, we injected clone-1 or control cells into the tail veins of nude mice and scored metastatic nodules in the lungs (, lower panel). A large number of metastatic nodules (average
135±89.9/lung) were formed from clone-1 cells (4/5 mice injected) while none were formed from control (, lower panel). These data demonstrated that SATB1 overexpression in MCF10A-1 induces cells to adopt a metastatic cancer phenotype in vivo
. In contrast to MCF10A-1, in vivo
analysis showed that MCF10A-2 cells, which did not exhibit a pronounced change in their 3D morphology () and invasiveness () upon SATB1 overexpression, failed to form tumors in nude mice () and did not produce an appreciable number of metastatic nodules (, lower panel).
Ectopic expression of SATB1 in MCF10A-1 cells induces tumor growth and lung colonization.
We examined whether the maintenance of the aggressive phenotype in MCF10A-derived tumor cells requires sustained SATB1 expression. As illustrated in , we isolated clone-1-TUM cells from tumors derived from xenografts of SATB1-expressing MCF10A-1 clone-1 (). We knocked down SATB1 expression from clone-1-TUM cells by shRNA [clone-1-TUM (SATB1-shRNA)] and reduced the overall SATB1 transcript level by ~70% compared to control cells (). The residual SATB1 expression after knockdown could possibly be due to the varied knockdown efficacies of SATB1-shRNA for different cells within the clone-1-TUM population. We then injected clone-1-TUM or clone-1-TUM (SATB1-shRNA) cells into the tail veins of mice (, upper panel). Of 7 mice injected with clone-1-TUM cells, 3 mice had >100 nodules and the remaining 4 mice had 8–31 nodules each, with an average of 105 nodules per lung. In contrast, clone-1-TUM (SATB1-shRNA) cells, which still retain 30% SATB1 expression, produced far fewer metastases. Of 6 mice, all had 25 or fewer nodules, with an average of 9 nodules per lung (, upper panel). These data indicate that the aggressive phenotype of clone-1-TUM is reversible by reducing SATB1 levels.
We observed varying degrees of lung metastases by clone-1-TUM cells, which may be attributable to different SATB1 expression levels. To address this possibility, we isolated the metastatic lung tissues and performed quantitative RT-PCR. The levels of human SATB1 transcripts, normalized to human actin levels, showed a positive, although not directly linear, correlation between the levels of SATB1 mRNA and the number of metastatic nodules per lung (, lower panel). These results support the contention that high SATB1 levels drive MCF10A-1 cells to form lung metastases. In contrast to lung metastasis, when we examined tumor formation by injecting clone-1-TUM or clone-1-TUM (SATB1-shRNA) into the mammary fat pads of nude mice, we found virtually no difference in the ability to form tumors by these two cell lines (), except for a slight delay in onset of tumor growth by the latter during the first 2 weeks (Fig. S2B). These in vivo data indicate that the SATB1 expression level is critical for tumor progression; whereas low levels of SATB1 expression are sufficient for tumor formation, higher levels of SATB1 are necessary for lung metastasis of MCF10A-1 cells.
The contrasting behaviors of MCF10A-1 and MCF10A-2 upon SATB1 expression highlighted the possibility that fundamental differences in the genomes of the two cell lines may be responsible for the disparate responsiveness to SATB1 activity. To determine whether MCF10A-1 and MCF10A-2 exhibit a gross genotypic difference, we performed genome-wide copy number analysis to compare genomes (). The genotypes of the MCF10A-1 and MCF10A-2 were 98.83% identical, providing strong evidence that they were derived from the same predecessor. Furthermore, analyses of their copy number profiles revealed that they were very similar as determined by a Pearson correlation of 0.96 (p<2.2e-16), providing additional evidence that these two lines are genomically identical. It has been reported that the copy number profiles of non-malignant MCF10A and premalignant neoT only differ in chromosome 9.13 and 9.20 
. We found no differences in the copy number profiles from these regions for MCF10A-1 and MCF10A-2 cells (Fig. S3A). In addition, p53 was similarly activated in MCF10A-1 and MCF10A-2 cells 
, as determined by phosphorylation of serine 15 and by induction of downstream genes, p21 and MDM2, upon exposure to ionizing radiation (12 Gy X-ray) (Fig. S3B), suggesting that p53 is intact in both cell lines.
Differential response to SATB1 overexpression by MCF10A-1 and MCF10A-2 cells is attributed to their disparate gene expression patterns involved in cell cycle regulation.
To identify the molecular bases for the contrasting phenotypes of MCF10A-1 and MCF10A-2 cells upon SATB1 overexpression, we analyzed gene expression patterns. Quantitative RT-PCR using the Cancer Pathway Superarray (89 genes) revealed differential expression (>1.5 fold up- or down-modulation with p<0.05) of genes involved in cell cycle control/DNA repair, adhesion, angiogenesis and invasion/metastasis. Notably, ATM
mRNA levels were 10-fold lower in MCF10A-1 compared to MCF10A-2 cells (). Additional analyses using the Cell Cycle Pathway and p53 Pathway arrays (89 genes each) confirmed that the expression of cell cycle and checkpoint genes significantly differed between the two cell lines (; Fig. S3C). Among the cell cycle/checkpoint-related genes, ATM
M and R
ad3-related) were those that were the most significantly downregulated in MCF10A-1 cells compared to MCF10A-2 cells. ATM and ATR are members of the PI3-kinase family that sense and transduce responses to DNA double strand breaks and regulate cell cycle checkpoints, DNA repair, apoptosis and senescence 
. Both ATM and ATR share many biochemical and structural similarities, but differ in certain cellular activities. For example, ATM is critical for the mitotic checkpoint 
and its mutation is associated with human cancer 
, whereas ATR is not. We then decided to focus on ATM and explore whether the difference in the ATM level between the MCF10A-1 and MCF10A-2 accounts for their contrasting responsiveness to SATB1 overexpression.
We tested whether MCF10A-1 cells, which have reduced ATM expression, were impaired for G2/M cell cycle checkpoints. To this end, we arrested cells at the G1/S boundary of the cell cycle using hydroxyurea (HU), released cells from HU, and immediately added paclitaxel, a drug which inhibits mitosis by stabilizing microtubule polymerization and causes G2/M cell cycle blockage 
(Fig. S4). After 18 h of incubation with paclitaxel, MCF10A-1 and MCF10A-2 cells were mostly arrested in G2/M phase. However, after 22 h of paclitaxel incubation, there was a slight increase [~8% (p
0.0078)] in MCF10A-1 cells that had bypassed the G2/M checkpoint and entered G1 phase compared to MCF10A-2 cells (Fig. S4). The escape from the G2/M arrest was followed by apoptosis, which occurred at a later time point (28 h) for both MCF10A-1 and MCF10A-2. These results suggest the G2/M checkpoint for MCF10A-1 cells is slightly impaired compared to MCF10A-2.
Next, we asked whether ATM mRNA levels are consistently reduced in SATB1-responsive cells, such as MCF10A-1 and MCF10-neoT, which both could be induced to become malignant by SATB1 overexpression, compared to SATB1-resistant cells, such as MCF10A-2. Indeed, ATM mRNA was lower in MCF10A-1 and in several other SATB1-responsive breast cancer cell lines as compared to MCF10A-2 cells (Fig. S5A), suggesting that the level of ATM negatively correlates with the responsiveness to SATB1 overexpression. We found that the ATM protein and transcript levels were comparable for MCF10A-1 cells and MCF10A-2 cells at early passage, but declined after continuous passage in culture for 90 days. This observation indicates that during passage ATM expression shifts from high to low in MCF10A cells, leading to a phenotypic drift from SATB1-resistant to SATB1-responsive types (Fig. S5A and S5B). The result also suggests that MCF10A-1 cells could be derived from MCF10A-2 cells and that they are not independent sublines. Prolonged continuous culture also induced SATB1 expression at a low, but detectable, level (Fig. S5C). These results suggest that MCF10A-1 is not a rare variant of MCF10A cells and that MCF10A cells are prone to downregulate ATM upon prolonged culture and acquire phenotypes exhibited by MCF10A-1 cells. Given that early passage MCF10A-1 express ATM at levels comparable to MCF10A-2 (Fig. S5A), we tested whether early passage MCF10A-1 behave similarly to MCF10A-2 upon SATB1 expression. We overexpressed SATB1 in early passage MCF10A-1 and analyzed 3D morphology (Fig. S5D). After 6 days culture on top 3D gels, SATB1 overexpression in early passage MCF10A-1 and in MCF10A-2 both formed spherical structures and no detectable spindle colonies (Fig. S5D). In vitro invasion assays also demonstrated that early passage MCF10A-1 exhibited similar levels of invasive potential to MCF10A-2 upon SATB1 overexpression (Fig. S5E).
We then tested whether ATM depletion is sufficient to convert MCF10A-2 cells from SATB1-resistant to SATB1-responsive types. We generated MCF10A-2 clones that stably express a shRNA against ATM (shATM) 
together with a SATB1-overexpressing construct (). Expression of the control vector or shATM alone did not alter MCF10A-2 phenotypes. However, co-expression of shATM and SATB1 induced a malignant cell morphology () and invasive activity (). Thus, ATM depletion rendered MCF10A-2 cells responsive to SATB1-mediated malignant induction, suggesting a role of ATM in suppressing tumorigenic progression driven by SATB1.
The combination of ectopic SATB1 expression and ATM depletion induces aggressive phenotype in non-malignant mammary epithelial cells.
To determine whether the oncogenic activity of SATB1 could be induced in other non-malignant cells lines besides MCF10A, we analyzed two additional immortalized, non-malignant human mammary epithelial cell lines, 184A1 and 184B5 
. 184A1 cells are stable and 184B5 cells are semi-stable during passage in culture, in contrast to MCF10A cells that are unstable and susceptible to phenotypic drifts depending on culture conditions 
. ATM levels in the two 184 cell lines were comparable to MCF10A-2 (Fig. S5A), and neither cell line adopted malignant phenotypes upon ectopic SATB1 expression. However, after stable ATM knockdown (), both 184A1 and 184B5 cells became SATB1-responsive and exhibited malignant phenotypes upon SATB1-overexpression. In 3D culture, ATM knockdown and SATB1 overexpression led 184A1 cells to exhibit a spindle-like morphology and led 184B5 cells to form large irregular aggregates (). Furthermore, ATM knockdown and SATB1 overexpression promoted the invasive activities of 184A1 and 184B5 cells as well as MCF10A cells (). These observations further suggest that ATM serves to protect cells from transformation upon SATB1 overexpression.