DEAR1 Is a Chromosome 1p35 Tumor Suppressor and Master Regulator of TGF-β–Driven Epithelial–Mesenchymal Transition

Chromosome 1p35 has long been known to harbor an important tumor suppressor(s) involved in multiple tumor types, including a 1.5cM major consensus interval for pancreatic cancer and a region of high-frequency LOH in both early and late colon cancers (1, 2). LOH in chromosome 1p35-36 has also been observed in other tumors, including stomach, colon and rectum, lung, breast, endometrium, testis, kidney, and thyroid cancers, and sarcomas (3). Furthermore, deletions in chromosome 1p35-36 correlate with poor survival in colon and breast cancers (4) and shorter disease-free survival in colon cancer (5). We previously identified DEAR1 as a novel gene mapping into this genomic interval and encoding a member of the TRIM protein family intimately associated with differentiation control as well as oncogenesis (6, 7). DEAR1 was shown to be both mutated and homozygously deleted in breast cancer, and its expression was observed to be downregulated/lost in 56% of early-onset breast cancers as well as in ductal carcinoma in situ (DCIS), one of the earliest preinvasive forms of breast cancer. DEAR1 expression was also found to predict local recurrence-free survival in early-onset breast cancer, suggesting that elucidation of the function of DEAR1 could aid in stratification of breast cancers for targeted therapies (6). Importantly, introduction of DEAR1 to complement a mutation in a breast cancer cell line restored acinar morphogenesis in three-dimensional (3D) culture, whereas stable knockdown (KD) in immortal human mammary epithelial cells (HMECs) recapitulated the phenotype of mutant DEAR1 cells with loss of apical–basal polarity, diffuse apoptosis, and failure of lumen formation, indicating that DEAR1 regulates polarity and tissue architecture. Therefore, we hypothesized that DEAR1 is a novel tumor suppressor, the loss of function of which might be critical in the loss of polarity associated with epithelial–mesenchymal transition (EMT; refs. 6, 8).

EMT is a complex and highly specialized developmental process in which tightly joined and polarized epithelia lose apical–basal polarity and tissue architecture and become disassociated, spindle-shaped, mesenchymal cells capable of migration (9). The preponderance of evidence also indicates that inappropriate activation of EMT in cancer results in the loss of epithelial polarity and a restructuring of tissue architecture as well as a remodeling of the extracellular matrix and actin cytoskeleton driving cancer cell migration, invasion, and ultimately metastasis (9).

The most potent inducer of EMT in epithelial cancers is the cytokine TGF-β (9). TGF-β elicits its effects through activation of its receptor and subsequent phosphorylation of the major effector SMADs (SMAD2/SMAD3), which, in complex with SMAD4, translocate to the nucleus to transcriptionally activate context-dependent gene sets that repress the proliferative response or activate EMT (10). Although much is known about the downstream molecular networks that signal EMT through TGF-β, the master regulatory controls on TGF-β's oncogenic axis are not understood as well as the underlying mechanisms governing cell polarity and tissue architecture that are lost in the initial stages of EMT (11). More importantly, elucidation of the regulatory controls on TGF-β's function is critical for design of novel therapies targeting the oncogenic arm of the pathway (12). We therefore hypothesized that loss of function of DEAR1 could play a causal role in the initiation of EMT in malignancies associated with 1p35 LOH or deletion. Experiments described herein address this hypothesis and identify DEAR1 as a tumor suppressor intimately linked to the regulation of TGF-β–driven EMT.

Because DEAR1 was previously found to be mutated or homozygously deleted and downregulated in breast cancer (6), we asked whether DEAR1 was a bona fide chromosome 1p35 tumor suppressor by conducting targeted disruption of Dear1 in the mouse (Fig. 1A; Supplementary Methods; Supplementary Fig. S1A). Absence of DEAR1 expression was observed in Dear1^−/− mouse tissues by immunohistochemistry (IHC) using an affinity-purified N-terminal antibody previously published (ref. 6; Fig. 1B). Quantitative real-time PCR (qRT-PCR) also confirmed the loss of DEAR1 expression in Dear1^+/− mice (Supplementary Fig. S1B). Genotypes of Dear1^+/− and Dear1^−/− mice showed expected Mendelian ratios and were similar phenotypically to wild-type (WT) littermates and had similar survival frequencies (averaging 20–23 months for Dear1^+/− and Dear1^−/− animals and 21 months for wild-type controls; Supplementary Fig. S1C and S1D). However, Dear1^−/− and Dear1^+/− mice formed late-onset tumors in 12.9% (8 of 62) and 17.7% (17 of 96) of animals, respectively, compared with a frequency of 4% (2 of 50) in WT littermates. Significantly, the tumor spectrum of Dear1^−/− and Dear1^+/− mice contained adenocarcinomas from multiple organs, including mammary, pancreatic, lung, and liver, as well as both low- and high-grade sarcoma, and lymphomas involving multiple organ sites (Fig. 1C and Supplementary Table S1). Mice with multiple epithelial tumors as well as metastatic adenocarcinomas were also observed (Supplementary Table S1). Thus, results indicate that Dear1 is a novel tumor suppressor and, significantly, that Dear1^+/− mice gave rise to tumors with a similar frequency as those observed in Dear1^−/− animals, suggesting that DEAR1 might function as a haploinsufficient tumor suppressor. To address this possibility, representative tumor tissue sections from liver, lung, and mammary gland adenocarcinomas, lymphoma, and sarcoma tumors were analyzed for DEAR1 expression by IHC in Dear1^+/− mice. DEAR1 expression was absent in eight of 11 of Dear1^+/− adenocarcinomas examined (Supplementary Fig. S1E). However, DEAR1 expression was positive in six of eight lymphoma samples, ranging from 30% positive-staining cells in Hodgkin lymphoma to 80% positive-staining in ureter lymphoma (Supplementary Table S2). Allele-specific PCR of Dear1^+/− lymphoma samples confirmed loss of the WT allele in the two cases of lymphoma that also showed loss of expression by IHC (Supplementary Fig. S1F). Thus, loss of the WT allele was observed in the majority of epithelial carcinomas, consistent with Dear1 behaving as a classical tumor suppressor, but clearly loss was not observed in all Dear1^+/− tumors, suggesting that haploinsufficiency could play a more predominant role in certain tumors, such as lymphoma. Tissue-specific haploinsufficiency has been suggested as a viable mechanism for inactivation of tumor suppressors in a context- or tissue-specific manner, depending on the cellular milleu in the particular tissue and in the context of other genetic hits in late-onset tumors (13).

The novel mouse model developed a variety of tumors that were strikingly similar to human tumors with chromosome 1p35 LOH. We therefore examined the human tumor spectrum for genetic alteration in DEAR1. Because DEAR1 maps into a tumor suppressor consensus region in pancreatic cancer (1), we conducted sequencing on 55 pancreatic adenocarcinoma samples, which indicated that two of 55 tumors (3.6%) contained novel missense mutations (R223H and R254Q), which were not seen in the Single Nucleotide Polymorphism database (dbSNP), 1000 Genomes project, or 192 normal allele controls, and were predicted to be deleterious by PolyPhen2. In addition, we conducted database analyses of The Cancer Genome Atlas (TCGA), cBio (14), and the International Cancer Genome Consortium (15) to determine whether DEAR1 was also mutated in other chromosome 1p35-related tumors. Results in Fig. 1D and Supplementary Table S3 indicate that DEAR1 undergoes rare mutation in multiple tumor types associated with chromosome 1p35 LOH, including lung squamous, endometrial, and renal carcinomas as well as glioblastoma multiforme and upperaerodigestive tract tumors. In colorectal carcinoma, four of 221 tumors contained missense mutations in DEAR1 (R190H, R223C, R307C, and D421G). Both codon 223 and 307 had previously been found to be mutated in pancreatic and lung adenocarcinoma, respectively, by our laboratory and by Rudin and colleagues (16), correspondingly.

Significantly, DEAR1 also undergoes copy number alterations in a variety of different cancer cell lines and primary tumors associated with 1p35 LOH. Screening of the CONAN database indicated that copy number losses affecting the DEAR1 locus were frequent events in multiple epithelial cell lines and hematopoietic cancers (Supplementary Fig. S2A and S2B; ref. 17). In addition, using the cBio database (14), we analyzed the frequency of copy number losses within the DEAR1 locus across multiple tumor types in data reported by TCGA. The results indicated that DEAR1 undergoes heterozygous loss in multiple tumor types observed in the Dear1 mouse model (Supplementary Table S4). Moreover, a putative homozygous deletion, encompassing the DEAR1 locus in glioblastoma multiforme, was discovered in the provisional TCGA cohort (14). Loss or downregulation of expression of DEAR1 was also associated with tumor types undergoing copy number losses using data from the TCGA cohorts (Fig. 1E; Supplementary Table S5; ref. 14). DEAR1 IHC was also conducted on a pancreatic adenocarcinoma tissue microarray, which revealed a significant number of pancreatic tumor samples showing either no expression (9 of 32; 28%), or barely detectable levels of DEAR1 staining (11 of 32; 34%), indicating that 62% were downregulated for DEAR1 expression (Supplementary Fig. S3A). In addition, DEAR1 expression was downregulated or undetectable in three of four pancreatic cancer cell lines as well as in human immortalized pancreatic ductal epithelial cells (HPDE) transformed with KRAS compared with HPDE cells that express high levels of DEAR1 (Supplementary Fig. S3B).

Our data identify DEAR1′s role as a tumor suppressor based on loss/downregulation of expression, mutation, and copy number loss in multiple human tumors, as well as the tumor spectrum recapitulated in the Dear1 mouse model. Loss of function of DEAR1 in the mouse also resulted in a subset of metastatic lesions, indicating that DEAR1 loss underlies not only tumorigenesis, but also metastasis. Loss of a single tumor suppressor rarely results in both primary adenocarcinomas and metastasis. However, it has recently been shown that well-characterized tumor suppressors involved in the regulation of cell cycle and cell proliferation, such as p53, could also function in the regulation of EMT and the initiation of metastasis (18, 19). Because our previous data indicated that DEAR1 is an important regulator of morphogenesis and cell polarity in 3D culture (6) and disruption of cell polarity is a key event associated with EMT, we further analyzed the correlation between DEAR1 mutation and clinical status. Results indicated that of the human tumors with DEAR1 mutation for which clinical information was available, eight of 10 were invasive/metastatic (Supplementary Table S6). On the basis of these compelling data, we hypothesized that DEAR1 might be important in the regulation of epithelial plasticity and critical in preventing the onset of EMT.

Because loss of cell polarity and tissue architecture is an important feature of epithelial cells undergoing EMT, we investigated the role of DEAR1 in the regulation of acinar morphogenesis in 3D culture in the presence or absence of TGF-β, a critical regulator of epithelial plasticity (20). Three stable DEAR1 knockdown (DEAR1-KD) clones in the immortal HMEC cell line 76N-E6 and two short hairpin RNA (shRNA) control clones generated previously (Supplementary Fig. S4; ref. 6) were plated in 3D culture in the presence or absence of TGF-β. Results indicated that, in the presence or absence of TGF-β, 76N-E6 cells as well as control-shRNA (CshR) clones initiated acinar morphogenesis and differentiated into uniform polar acini in the presence of TGF-β (Fig. 2A and Supplementary Fig. S5). Stable knockdown of DEAR1, however, resulted in irregularly shaped acini (Fig. 2A), as we reported previously, resulting from lack of apical basal polarity and proper lumen formation (6). Strikingly, three of three DEAR1-KD (DshR) clones propagated in 3D culture in the presence of TGF-β (2 ng/mL) failed to initiate acinar morphogenesis and showed a complete inability to differentiate (Fig. 2A and Supplementary Fig. S5). Because acini formation was not disrupted in control clones in the presence of TGF-β at the same concentration, results in DEAR1-KD clones were not due to toxicity effects from the addition of TGF-β to the cultures. To determine whether the failure of acinar morphogenesis in DEAR1-KD cells in the presence of TGF-β was due to apoptosis, acini collected from 3D culture were examined for caspase-3 expression. Results indicated that caspase-3 expression did not vary significantly between DEAR1-KD clones and control clones in the presence of TGF-β (Fig. 2B), suggesting that failure of acini formation in DEAR1-KD clones treated with TGF-β was not a result of apoptosis. However, because TGF-β plays a critical role as a potent inducer of EMT and because mesenchymal-like cells have been reported to be unable to form acini in 3D culture (21), we examined acini for expression of vimentin. Results indicated that vimentin expression was significantly induced in TGF-β–treated DEAR1-KD clones compared with control clones (Fig. 2B), suggesting that in the presence of TGF-β, DEAR1-KD clones acquired mesenchymal features which prevented acinar morphogenesis. These data indicate that the presence or absence of DEAR1 dictated the formation of acini in 3D culture in the presence of the extracellular signals from TGF-β. Thus, we asked whether DEAR1 loss of function might play a causal role in the initiation of TGF-β–driven EMT.

Morphologic changes in two-dimensional (2D) culture were clearly evident in DEAR1-KD clones, consistent with a partial mesenchymal phenotype with less compact epithelial patterning with scattering, as compared with control clones which grew as compact, cobblestone-like epithelial cells (Supplementary Fig. S6A). After treatment with TGF-β for 5 days, control clones showed no obvious difference in morphology, whereas DEAR1-KD clones showed a marked increase in scattering as well as cells displaying a prominent mesenchymal morphology (Supplementary Fig. S6A). DEAR1-KD and control clones were also observed in 3D culture in the presence of TGF-β. Strikingly, by day 5, clear traces of cell migration through the Matrigel were visible in some cells from DEAR1-KD clones in the presence of TGF-β, as opposed to control clones that showed no evidence of traces, indicating that the DEAR1-KD clones exhibited cell motility in the presence of TGF-β (Fig. 2C). To better quantify cell migration, we conducted time-lapse microscopy to observe migration following plating of DEAR1-KD and control clones with and without TGF-β in Matrigel 3D cultures. Results demonstrated that the average distance of cell migration in both DEAR1-KD (40.28 ± 12.36 μm) and CshR clones (42 ± 14.7 μm) did not differ significantly in the absence of TGF-β, indicating that without TGF-β, loss of function of DEAR1 does not affect cell migration through the matrix (Fig. 2D, and Supplementary Fig. S6B and S6C). Moreover, time-lapse video clearly documented that DEAR1-KD cells exhibited extensive migration on top of the Matrigel and invasion through the Matrigel in the presence of TGF-β (Supplementary Movies S1–S4). The average migration distance of DEAR1 WT clones following TGF-β treatment was 16.02 ± 6.52 μm (P = 0.0057 vs. no treatment), compared with a much enhanced migration distance in DEAR1-KD clones following TGF-β treatment (173.2 ± 15.19 μm, P < 0.00001 vs. no treatment; Supplementary Fig. S6C). The longest migration distance recorded for DEAR1-KD cells was 1,400 μm. Like control clones, DEAR1-KD cells showed some evidence of migration without TGF-β treatment; however, the migration halted after the first 15 to 20 hours of experiments, whereas the migration in TGF-β–treated DEAR1-KD cells continued for the entire experimental period (72 hours; Fig. 2D). Furthermore, both control clones (with or without TGF-β treatment) and DEAR1-KD clones (without TGF-β treatment) formed acini in 3D culture once their movement halted, perhaps reflecting an evolutionary adaptation of the cells due to the environmental changes (from 2D culture to 3D culture; Fig. 2A). However, the DEAR1-KD cells failed to form acini in 3D culture following TGF-β treatment (Fig. 2A and Supplementary Fig. S5), which suggested that this type of migration might represent an EMT-related migration. Therefore, this evidence clearly indicates that loss of function of DEAR1 results in TGF-β–induced migration and invasion through the Matrigel matrix in these immortal HMEC lines. Because 3D culture more closely approximates growth in an in vivo environment, results also suggest the potential importance of loss of DEAR1 expression or mutation to the initiation of TGF-β–induced EMT involved in breast cancer progression.

Next, wound-healing assays were conducted to assess cell migration in 2D culture in DEAR1-KD cells versus controls. In CshR clones, TGF-β treatment failed to accelerate cell migration across the wound, whereas in DEAR1-KD clones, TGF-β treatment resulted in rapid wound closure compared with untreated cells (Fig. 2E, left). We then generated stable shRNA DEAR1 knockdown clones in the breast epithelial line MCF10A (10A–DEAR1-KD) and conducted the same experiments. Results were consistent with experiments in 76N-E6 cells, in that migration was accelerated in the presence of TGF-β in 10A–DEAR1-KD cells compared with CshR cells (Fig. 2E, right and Supplementary Fig. S6D). To further confirm the specificity of DEAR1-KD, we generated additional DshR stable clones in 76N-E6 cells (76N-E6–DEAR1-KD2) targeting a different region of DEAR1 that did not overlap with the original targeted region and carried out the same wound-healing assays. Similar results were observed as in the initial shRNA clones, providing additional support for the specificity of the cellular knockdown for DEAR1 (Supplementary Fig. S6E). In addition, we transiently transfected vector controls or DEAR1 expression constructs into U2OS cells, which have no detectable DEAR1 expression (data not shown), and conducted wound-healing assays. Results indicated that in the vector control, TGF-β treatment enhanced cell motility and accelerated wound closure (Supplementary Fig. S6F). However, in U2OS cells ectopically expressing DEAR1, TGF-β treatment failed to significantly enhance migration across the wound, indicating that DEAR1 expression inhibited TGF-β–driven migration (Supplementary Fig. S6F). Thus, cumulative data from DEAR1-KD and ectopic expression experiments using both a 3D culture model and 2D migration models show that DEAR1 is a dominant inhibitor of TGF-β–induced cell migration and that loss of DEAR1 expression is requisite for TGF-β–driven migration and invasion in immortal HMECs.

We then examined the expression of EMT markers in both 76N-E6 cells and MCF10A cells. Although the basal level of β1-Integrin, a major effector of TGF-β–driven EMT (22), varied between the two breast control lines, its expression was downregulated in both cell lines following TGF-β treatment. However, in DEAR1-KD clones in both 76N-E6 and MCF10A cells, β1-Integrin expression was upregulated in the presence of TGF-β (Fig. 2F and Supplementary Fig. S7A). Expression levels of E-cadherin were slightly downregulated in DEAR1-KD clones compared with controls. However, both 76N-E6 and MCF10A knockdown clones showed increased expression of the EMT marker N-cadherin in the presence of TGF-β. We also analyzed the expression of the mesenchymal marker vimentin. Importantly, compared with the control clones, DEAR1-KD clones in 76N-E6 and MCF10A cells showed increased basal expression of vimentin (Fig. 2F and Supplementary Fig. S7A) which was further increased upon treatment with TGF-β. Upregulation of vimentin was also confirmed by immunostaining, which indicated that 13% of cells in control clones were positive for vimentin staining, whereas up to 65–80% of cells in stable 76N-E6–DEAR1-KD clones showed strong vimentin expression even without TGF-β treatment (Supplementary Fig. S7B). In 76N-E6–DEAR1-KD2 cells, EMT markers were upregulated similarly as reported for the initial shDEAR1 clones (Supplementary Fig. S7C). mRNA expression analysis of β1-Integrin and vimentin by qRT-PCR confirmed Western blot analyses. TGF-β-target PAI-1 was dramatically increased in DEAR1-KD clones after TGF-β treatment (Supplementary Fig. S7D). Because upregulation of vimentin and β1-Integrin have been shown to correlate with the induction of EMT as immediate early events in the EMT process in HMECs (23), our data suggest that DEAR1 loss of function results in immediate early induction of EMT in the presence of TGF-β.

Anoikis is a process triggered by inadequate or inappropriate cell–matrix contact in which epithelial cells, separated from the epithelial layer, die by a mechanism similar to apoptosis. Resistance to anoikis confers a selective advantage to precancerous epithelial cells, and also increases survival of cells detached from the matrix. Therefore, anoikis resistance is thought to be an important marker of EMT and characteristic of invasive and metastatic cells (20, 24). To examine whether DEAR1 functions in the regulation of anoikis, we determined DEAR1′s ability to affect the growth of cells plated on ultralow attachment (ULA) dishes in the presence or absence of TGF-β. After 2 to 4 days of incubation, massive cell death occurred in control 76N-E6 clones with or without TGF-β, whereas in DEAR1-KD clones, cellular aggregates were clearly observed and greatly enhanced in the presence of TGF-β (Fig. 3A). A marked increase in colony formation was observed in DEAR1-KD clones when suspension cells were propagated in regular culture dishes, especially in the presence of TGF-β (Fig. 3B), indicative of anoikis resistance. Because anoikis is thought to be controlled in part by regulators of programmed cell death, such as caspase-3 (25), we analyzed control and knockdown clones for active caspase-3 expression following plating under ULA conditions without TGF-β treatment. Active caspase-3 expression was observed in control clones for the entire experimental period, up to 48 hours, whereas in DEAR1-KD clones, downregulation of active caspase-3 was apparent at 24 hours (Fig. 3C). These data indicate that DEAR1 is an important regulator of anoikis in HMEC-76N-E6 cells, and that loss of function of DEAR1 drives TGF-β–mediated anoikis resistance in immortal HMECs.

We next investigated the role of DEAR1 in regulating TGF-β–induced signal transduction. As shown in Fig. 3D, overexpression of DEAR1 inhibited TGF-β–induced TGF-β–SMAD response element–binding site (CAGA)12-driven luciferase reporter activity (26) in HEK293T cells, which have undetectable DEAR1 expression (6). Similar results were obtained with a PAI-1 luciferase reporter, a transcriptional target of TGF-β/SMAD3 signaling, in HEK293T cells (26). Furthermore, DEAR1 or vector control was cotransfected along with CAGA12 reporter with either DshR or CshR as well as DEAR1-rescue (DR1mut) containing mutations in the shRNA-targeting region to determine whether the inhibition of TGF-β signaling was specifically mediated through DEAR1. Knockdown of DEAR1 reversed the inhibitory effect of DEAR1 on the CAGA12 reporter. Furthermore, cotransfection of a mutation abolishing the shRNA-targeting region in DEAR1 strongly rescued DEAR1-mediated inhibition of TGF-β signaling, indicating that the repression of TGF-β signaling was specifically mediated by DEAR1 (Fig. 3E). TGF-β–induced signal transduction experiments were also conducted in 76N-E6–DEAR1-KD cells. Results indicated that TGF-β–induced transcriptional activity was much enhanced in DEAR1-KD clones compared with control clones (Fig. 3F). We also investigated this effect of DEAR1 in the breast cancer cell line MDA-MB-231 and the cervical cancer cell line HeLa, in which similar results were obtained (Supplementary Fig. S8A), indicating that DEAR1 inhibits TGF-β signaling in both breast and cervical carcinoma lines. Cumulative data indicate that DEAR1 is an important inhibitor of TGF-β–induced signal transduction. To determine whether any loss-of-function mutations could be identified that target TGF-β signaling, DEAR1 constructs with the tumor-derived missense mutations D106V and R254Q (Supplementary Table S3) were introduced with a CAGA12 reporter into HEK293T cells. Results indicated that missense mutation R254Q did not rescue TGF-β–induced luciferase activity; however, the D106V mutation resulted in partial (∼45%) restoration of TGF-β–induced luciferase activity compared with vector controls and similar to results obtained from deletion of the entire exon 1 (Fig. 3G). These results provide evidence that DEAR1 undergoes loss-of-function mutation that, in the case of the D106V mutation, restores TGF-β signaling.

Because SMAD3 is a major effector of TGF-β–driven EMT (23) and both CAGA12 and PAI-1 reporters have been shown to be regulated by SMAD3 (26), we investigated whether DEAR1′s inhibition of TGF-β signaling was mediated by repression of signaling through SMAD3. Results indicated that SMAD3 transfection alone stimulated reporter luciferase activity driven by CAGA12 or PAI-1 in HEK293T cells, and this effect was greatly enhanced after TGF-β treatment. However, cotransfection of HA-tagged DEAR1 (HA/DEAR1) resulted in a dramatic inhibition of SMAD3-mediated signal transduction using either CAGA12 or PAI-1 reporters (Supplementary Fig. S8B). In addition, the inhibitory effect of DEAR1 on TGF-β–SMAD3-induced signaling increased with increasing amounts of DEAR1 transfected, indicating a dose-dependent effect (Supplementary Fig. S8C). To confirm that the inhibitory effect was specifically mediated by DEAR1, we cotransfected DshR in the system and found that DshR absolutely reversed the inhibition by DEAR1 compared with the CshR (Fig. 3H). Thus, these data show that DEAR1 inhibition of TGF-β signal transduction is mediated, at least in part, by repression of SMAD3-dependent signaling.

Because DEAR1 is a member of the TRIM family of proteins, which has been associated with the formation and architecture of large protein complexes, we asked whether DEAR1 might bind to SMAD3 and affect SMAD3 protein stability. Endogenous SMAD protein levels were determined in 76N-E6 cells by Western blot analysis, which indicated that SMAD2 levels did not vary between control and knockdown clones (Fig. 4A), whereas DEAR1-KD clones showed a dramatic increase (two- to threefold of control clones) in SMAD3 protein levels before TGF-β treatment, and that level was not elevated in the presence of TGF-β (Fig. 4A), suggesting that DEAR1 attenuates SMAD3 protein levels independently of TGF-β treatment. Similar results were also observed using 76N-E6–DEAR1-KD2 cells, as well as confirmed in MCF10A–DEAR1-KD cells, which expressed increased SMAD3 levels relative to shRNA controls (Supplementary Figs. S9A and S11A). To determine the mechanism underlying the increase in SMAD3 expression following DEAR1-KD, 76N-E6–DEAR1-KD and CshR clones were treated with the translational inhibitor cycloheximide. SMAD3 protein levels significantly decreased within 3 hours in the control cells (Fig. 4B and C), suggesting that DEAR1 expression reduced SMAD3 stability. Moreover, SMAD3 levels were stabilized following treatment with the 26S proteasome inhibitor MG132 in control clones in which SMAD3 levels accumulated within 3 hours following treatment with MG132, whereas there was almost no effect in DEAR1-KD clones in the presence of MG132, suggesting that DEAR1 mediates proteasomal degradation of SMAD3 (Fig. 4B and C).

To investigate the role of DEAR1 in mediating the proteasomal degradation of SMAD3, we first determined whether DEAR1 binds to SMAD3. Results in Fig. 4D show that DEAR1 coimmunoprecipitated with SMAD3, but not SMAD2 or SMAD4. We next conducted immunostaining following cotransfection of HA/DEAR1 with Myc/SMAD2 or Myc/SMAD3. Results in Fig. 4E clearly show strong colocalization of DEAR1 with SMAD3 (yellow arrows in Fig. 4E, ii), whereas colocalization with SMAD2 was not observed (Fig. 4E, i). Furthermore, we mapped the domain of SMAD3 required to interact with DEAR1. By both coimmunoprecipitation and colocalization experiments using immunofluorescence staining, the only domains of SMAD3 involved in interacting with DEAR1 were the linker and MH2 domains of SMAD3 (Supplementary Fig. S9B and S9C). The endogenous interaction between SMAD3 and DEAR1 was confirmed by anti-SMAD3 pull down in HeLa cells (Fig. 4F). Glutathione S-transferase (GST) pull-down assays were then conducted to determine whether DEAR1 directly interacts with SMAD3. GST or GST–DEAR1 fusion proteins were incubated with cell lysate from HEK293T cells containing transiently transfected Flag-SMAD3. Specific interaction of GST–DEAR1, but not GST alone, with Flag-SMAD3 was observed (Fig. 4G), indicating that DEAR1 physically interacts with SMAD3. Because ubiquitination of SMAD proteins has been found to be an important tool in regulation of protein stability (27), we investigated whether DEAR1 promotes ubiquitination of SMAD3 by cotransfecting Flag-SMAD3, HA-ubiquitin, and DEAR1 into HEK293T cells. SMAD3 was then pulled down by anti-Flag, and ubiquitination was detected using anti-HA-tag antibodies. Results indicated that no ubiquitination was observed in vector controls, whereas, as indicated in Fig. 4H, DEAR1 markedly increased polyubiquitination of SMAD3, indicating that DEAR1 promotes the ubiquitination of SMAD3.

To determine whether the inverse relationship between loss of function of DEAR1 and increased stability of SMAD3 observed in in vitro experiments could be recapitulated in human tumors showing loss of expression of DEAR1, we examined DEAR1 and SMAD3 expression in early-onset human breast tumors compared with adjacent normal tissues. Results were striking in that five of seven tumor/normal pairs showed an inverse correlation between DEAR1 expression and SMAD3 expression (Supplementary Fig. S9D), whereas SMAD2 expression did not show an inverse relationship with DEAR1 expression. In addition, DEAR1 and SMAD3 expression were determined by IHC in 10 human DCIS samples with adjacent normal ductal epithelium as well as invasive carcinoma. Results showed that four of 10 tumors showed an inverse correlation in expression. In one case, a uniform staining pattern for both DEAR1 and SMAD3 was observed throughout the tumor, in which DEAR1 staining was downregulated or completely absent in DCIS lesions and adjacent invasive disease was contrasted with SMAD3 expression, which was upregulated in DCIS and invasive breast cancers (Fig. 5A). In the remaining cases (Supplementary Figs. S10A–S10F and S11A–S11F), tumor heterogeneity was observed, with the identical region within the tumor showing loss/downregulation of DEAR1 expression, correlating with upregulation of SMAD3 expression relative to staining in adjacent normal epithelium. Interestingly, the four cases showing an inverse expression were all derived from early-onset cases (ages ranging from 37 to 50 years), consistent with Western blot analysis data derived from early-onset tumor and adjacent normal samples (Supplementary Fig. S9D). These data are also consistent with our previous finding of DEAR1 mutation as well as loss of expression correlating with local recurrence in early-onset breast cancers (6). Furthermore, to confirm that the in vivo human expression studies were also recapitulated in the Dear1 mouse model, we examined the correlation between Dear1 and Smad3 expression in a heterozygous Dear1 mouse with lung adenocarcinoma for which normal bronchial epithelium, well-differentiated adenocarcinoma, as well as foci of poorly differentiated adenocarcinoma with high-grade atypia and an invasive pattern were present in the same section (Fig. 5B). Dear1 expression was detected as pale cytoplasmic staining in the normal bronchial epithelium, as would be expected in a heterozygous genotype, but not expressed in either the lung adenocarcinoma or associated aggressive foci (Fig. 5B, d–f). However, Smad3 expression was detected in the normal bronchial epithelium as well as in the well-differentiated tumor, but strikingly upregulated in foci that were morphologically poorly differentiated with characteristics of an aggressive, invasive tumor (Fig. 5B, g–i). Cumulative data provide compelling evidence, by detailed in vitro mechanistic studies as well as by expression analyses in vivo, that DEAR1 is an important regulator of SMAD3 and that loss of DEAR1 results in upregulation of SMAD3 and unbridles TGF-β–mediated EMT.

Nuclear translocation of phosphorylated SMAD3 (p-SMAD3) after TGF-β stimulation is requisite for SMAD3′s transcriptional activation function. Because ubiquitination of SMAD3 by DEAR1 results in lower levels of SMAD3 as well as p-SMAD3 in control clones (Fig. 4A), we examined localization of active p-SMAD3 by immunostaining, which indicated that nuclear p-SMAD3 staining was slightly elevated in DEAR1-KD cells even without TGF-β compared with control cells, but TGF-β treatment resulted in a much increased accumulation of p-SMAD3 in the nuclei of knockdown cells (Fig. 5C and Supplementary Fig. S12A), indicating that DEAR1 loss of function increases the amount of p-SMAD3 in the nucleus and therefore available to transcriptionally activate EMT-related genes.

To examine whether introduction of DEAR1 could attenuate SMAD3 levels in breast cancer cells, we conducted transient transfection of HA/DEAR1 into MCF7 cells, which have no detectable DEAR1 protein (6). Transient transfection allowed the visualization of cells with and without DEAR1 expression, along with localization and expression of SMAD3. Results in Fig. 5D show that, without TGF-β treatment, endogenous, primarily cytoplasmic SMAD3 levels are lower in DEAR1-overexpressing cells (green arrows) than in adjacent untransfected cells (red arrow; without DEAR1 expression), indicating that DEAR1 expression indeed degrades SMAD3, and, significantly, this regulation of SMAD3 levels by DEAR1 is independent of TGF-β. In the presence of TGF-β, the majority of SMAD3 moved into nuclei in DEAR1-untransfected cells (red arrow), whereas in DEAR1-transfected cells, SMAD3 staining was less apparent and observed throughout the cell (green arrow). These data further confirm our previous observation that DEAR1 promotes the degradation of SMAD3 and thus results in less SMAD3 entering into nuclei after TGF-β stimulation in MCF7 cells. Because signal intensity of DEAR1-transfected cells precluded visualization of SMAD3 binding, we further analyzed the colocalization of DEAR1 and SMAD3 using the Imaris image analysis system that allowed 3D reconstruction of stacked images to determine colocalization of the two proteins (28). Using Imaris, more than 60% of individual HA/DEAR1 signals colocalized with SMAD3 (Supplementary Fig. S12B and S12C). Thus, cumulative data show that DEAR1 binds to SMAD3 independently of TGF-β and regulates SMAD3 protein levels by inducing ubquitination of SMAD3; therefore, DEAR1 controls the amount of SMAD3 available for phosphorylation and nuclear translocation by TGF-β.

To confirm that DEAR1 mediates TGF-β–induced EMT through degradation of SMAD3, we asked whether SMAD3 loss of function would rescue DEAR1 loss of function to prevent TGF-β–mediated EMT. First, we conducted SMAD3-KD using SMAD3-shRNA in DEAR1-KD clones of both 76N-E6 and MCF10A lines and showed that in both DEAR1-KD 76N-E6-DshR cells and MCF10A-DshR cells, SMAD3-KD strongly inhibited TGF-β–induced EMT markers (Fig. 6A) as well as greatly retarded TGF-β–induced cell migration in wound-healing assays (Fig. 6B and C). These data clearly indicate that SMAD3-KD in DEAR1-KD cells rescued DEAR1′s loss-of-function phenotype. We next investigated the effect of a newly designed SMAD3 inhibitor, SIS3 (29), on TGF-β–induced EMT in DEAR1-KD clones. Results in Fig. 6D show that with increasing concentrations of SIS3, N-cadherin and vimentin expression induced by TGF-β were markedly attenuated. Similar results were obtained in MCF10A–DEAR1-KD cells (Supplementary Fig. S13A). We also examined the role of SIS3 in blocking TGF-β–induced cell migration in wound-healing assays. Results in Fig. 6E show that, after TGF-β treatment, cell migration across the wound increased almost three- to fourfold in DEAR1-KD clones compared with control clones. Remarkably, the addition of SIS3 absolutely abolished TGF-β–driven cell migration in DEAR1-KD HMECs (P < 0.001) as well as in MCF10A–DEAR1-KD cells (Fig. 6E and F and Supplementary Fig. S13B). These data provide strong evidence that degradation of SMAD3 by DEAR1 is the primary mechanism by which DEAR1 inhibits TGF-β–induced EMT, and also give novel insight into the possibility of designing clinical inhibitors to interfere with the tumor promotion axis of the TGF-β–SMAD3 pathway.

Although multiple signaling pathways initiate EMT, these pathways use common sets of downstream effectors as master transcriptional regulators of EMT which include SNAIL1/2, ZEB1/2, and TWIST1/2 (30, 31). Because SNAIL1 (SNAI1) and SNAIL2 (SNAI2) have been shown to be targets of TGF-β–SMAD3 signaling (32, 33), we examined mRNA levels of SNAIL1/2 as well as TWIST1/2 to determine whether these master transcription factors were deregulated with loss of function of DEAR1. qRT-PCR results indicated that the basal level of SNAIL1 and SNAIL2 expression in the 76N-E6–DEAR1-KD clone increased (twofold) compared with control clones even without TGF-β treatment; further, TGF-β stimulation resulted in higher induction of SNAIL1/2 expression (fourfold; Fig. 7A). However, no difference in expression of TWIST1/2 was observed following DEAR1-KD in the presence or absence of TGF-β (Fig. 7A). Upregulation of SNAIL2 was confirmed at the protein level by Western blot analysis, which showed an increase in both basal levels as well as TGF-β–inducible levels of SNAIL2 (Fig. 7B). These data indicate that DEAR1 blocks a major consensus pathway downstream of TGF-β, and loss of function of DEAR1 selectively upregulates SNAI but not TWIST. Because SNAI1 and SNAI2 expression have been reported to be directly regulated by TGF-β–SMAD3 (32, 33), our data provide additional evidence that DEAR1′s functional role in degrading SMAD3 is a major mechanism to block TGF-β–induced EMT. To determine the clinical significance of the relationship between DEAR1 and SNAIL1/2 in invasive breast cancer, we analyzed the relationship between DEAR1 loss of function including copy number variation in DEAR1, increased mRNA expression and increased protein levels, and amplification of either SNAI1 or SNAI2. Heterozygous loss of DEAR1 occurred in 33% of cases in this cohort and typically involved regions larger than the DEAR1 locus, and in some cases involved the whole short arm. Results indicated that in 889 cases of invasive breast adenocarcinoma from the TCGA cohort, accessed via cBio (14), heterozygous loss of DEAR1 trended toward significance in this large cohort (P = 0.095), but by itself it was not significant as a clinical indicator of poor prognosis. Likewise, SNAI1 or SNAI2 alterations alone did not significantly affect survival (P values of 0.212 and 0.508, respectively). The combination of DEAR1 heterozygous loss with SNAI1 alteration was nearly significant in predicting poor overall survival (P = 0.056). However, strikingly, the combination of DEAR1 heterozygous loss with SNAI2 alteration significantly predicted shortened overall survival (P = 0.023). Furthermore, in accordance with the qRT-PCR data, TWIST1 and TWIST2 did not significantly associate with DEAR1 heterozygous loss (P values of 0.265 and 0.175, respectively; Fig. 7C). The data strongly support our conclusion that DEAR1 loss of function results in increased levels of SMAD3, leading to increased expression of its downstream effectors, SNAIL1 and SNAIL2, and that genetic alteration of this pathway has a significant effect on patient survival.

Identification of the genetic drivers of tumorigenesis and metastasis is essential for the design of targeted therapies aimed at the underlying pathways regulated by these genes. Our combined data from the Dear1 mouse model, compilation of human tumor data, and mechanistic model are consistent with DEAR1 being a critical tumor suppressor involved in many different cancers. Results presented herein from the Dear1 knockout model are striking in that loss of Dear1 in the 129/C57/BL6 background resulted in late-onset tumors with a tumor spectrum similar to human tumors with LOH at chromosome 1p35 and consisting of diverse epithelial adenocarcinomas, lymphoma, and sarcoma. Epithelial tumors are rarely observed in mouse models; yet, malignant epithelial adenocarcinomas are frequent in this model. Thus, these data indicate the potential importance of loss of DEAR1 to the initiation or progression of these tumor types. Because DEAR1 expression was absent in the majority of tumors analyzed, we conclude that Dear1 seems to behave as a classic tumor suppressor. However, a detailed investigation of its loss of function is warranted to determine in which circumstances haploinsufficiency might drive tumorigenesis (34, 35). Retention of the WT allele has been observed in tumors derived from the p53 mouse models as well as other important tumor suppressors, such as PTEN. Interestingly, haploinsufficiency in a subset of tumors has been observed in p53 models in which the frequency of retention of the WT Trp53 allele increases with increasing age of mice, and animals with tumor incidence of more than 18 months had 85% of tumors retaining the WT allele (35). Loss of the WT allele could also occur very late in tumor development and could be a rate-limiting step in tumor progression (36). Thus, the novel Dear1 mouse model provides a striking example of a classic tumor suppressor that could also behave as a haploinsufficient tumor suppressor in certain tissues or contexts (35). Because chromosome 1p cytogenetic deletion is one of the most frequent deletions observed in human tumors, and because important genes involved in human tumorigenesis map distal to DEAR1 within chromosome 1p36, including the genes encoding the CHD5 and p73 tumor suppressors (37, 38), evolution of epithelial tumors may involve not only an interstitial deletion of important loci within chromosome 1p, but also the loss of the entire short arm, in which case the loss of DEAR1 expression could also be pivotal to the initiation or progression of these tumor types to invasive and metastatic disease. In that regard, heterozygous loss of Dear1 in conjunction with oncogenic activation has been shown to drive tumorigenesis, with dramatic differences in survival compared with controls or oncogene-activated tumors (Quintas-Cardama and colleagues, manuscript in preparation).

When activated in cancer, EMT is thought to drive invasion and metastasis in breast cancer and other epithelial cancers and potentiate the generation of cells with stem cell–like characteristics resistant to current chemotherapies (39). Results in 3D culture of DEAR1-KD HMECs grown in the presence of TGF-β are consistent with induction of a mesenchymal transdifferentiation program associated with DEAR1 downregulation, resulting in an inability of these mesenchymal cells to form acinar structures in 3D culture (21). Our data also indicate that loss of DEAR1 in immortalized HMECs results in a partial EMT morphology with upregulation of vimentin and N-cadherin, well-known markers of early EMT induction. DEAR1 is also downregulated in a majority of DCIS lesions as well as in invasive carcinoma immediately adjacent to normal ductal structures that highly expressed DEAR1 (6). Herein, we show as well that DEAR1 undergoes a reduction in copy number and mRNA expression in the TCGA cohorts and that this loss of copy number is a predictor (in combination with alteration in SNAI1/2) of overall poor survival in invasive breast cancer. Thus, DEAR1 loss of function could play an early role in the initiation of EMT and could be an important biomarker of aggressive disease. In addition, cumulative results in both 2D and 3D culture indicate that DEAR1 is an important negative regulator of TGF-β–SMAD3 signaling and that loss of function in the presence of TGF-β results in the activation of anoikis resistance, migration, and invasion, key characteristics of EMT and essential steps in tumor progression and the metastatic cascade.

Our results also document that the interaction of DEAR1 with SMAD3 is an underlying mechanism by which DEAR1 regulates TGF-β–SMAD3-induced EMT. Although SMAD3 is highly homologous to SMAD2, recent reports found that SMAD2 and SMAD3 can play different roles in mediating TGF-β signaling. Our data indicate that DEAR1 interacted with the linker and MH2 domains of SMAD3, but did not seem to significantly affect SMAD2, which is consistent with reports that AKT binds to the linker and MH2 domains of SMAD3 but does not interact with SMAD2 (40, 41). Because DEAR1 binds to SMAD3 and induces SMAD3 ubiquitination, the rapid turnover of nonactivated SMAD3 by DEAR1 may reflect the necessity for cells to more stringently control SMAD3 levels and thus the activity of this protein (42). TGF-β–induced EMT has also been shown to be associated with SMAD3-dependent and -independent signaling pathways (43). However, studies using Smad3 knockout mice have indicated that TGF-β signaling through the SMAD3-dependent pathway is required for EMT (44). SMAD3 induces vimentin transcription required for the intermediate filament system of the cytoskeleton to change to a vimentin-based cytoskeleton during EMT (23). Our data in both 2D and 3D culture indicate that loss of DEAR1 results in dramatic upregulation of vimentin, β1-integrin, and N-cadherin in the presence of exogenous TGF-β. These studies, together with our findings that both a SMAD3 inhibitor and SMAD3 knockdown blocked TGF-β–induced cell migration, suggest that SMAD3 is a critical mediator of TGF-β–induced EMT and that DEAR1 targets SMAD3 for degradation. Therefore, regulation of SMAD3 protein levels is a critical step in preventing EMT.

TGF-β–SMAD pathways have been shown to play roles in both tumor suppression and EMT that have always been explained as context-dependent outcomes in human tumors (10). Thus, although the TGF-β–SMAD axis is a major tumor suppressor pathway in pancreatic cancer development, accumulating data have shown that this pathway also underlies EMT in pancreatic cancer as well as many other epithelial carcinomas (9–11). We have now clearly identified the first mechanism of action for DEAR1 as a novel and important negative regulator of TGF-β–SMAD3-driven EMT. Future studies will determine whether DEAR1 plays a role in the regulation of SMAD3′s tumor suppressor functions in response to TGF-β or is a critical selective switch that dictates the context-dependent fate of TGF-β/SMAD3. Future studies will also elucidate DEAR1′s role in the regulation of other key tumor suppressor pathways that could underlie its broader role in tumor suppression and unravel how DEAR1 loss of function could result in such a wide spectrum of epithelial adenocarcinomas as well as lymphoma and sarcoma.

In summary, our combined data provide critical and compelling evidence that DEAR1 is a chromosome 1p35 tumor suppressor involved in multiple human cancers as well as provide a novel paradigm for its regulation of both SMAD3 protein levels and TGF-β–induced EMT. Overall, we propose a model for DEAR1 function in the presence of TGF-β (Fig. 7D). In the presence of WT DEAR1, SMAD3 is degraded and TGF-β's oncogenic EMT arm is restrained, consistent with epithelial cells undergoing differentiation and acinar morphognesis in 3D culture. Loss of function of DEAR1 results in elevated SMAD3 available for phosphorylation and nuclear translation induced by TGF-β to activate EMT, thus consistent with epithelial cells undergoing transdifferentiation to mesenchymal cells and a failure in acinar morphogenesis (Fig. 7D). Because TGF-β inhibitors are being used currently in clinical trials with partial success (45), our data suggest that DEAR1 could be a novel and important biomarker to stratify those patients with breast cancer, and potentially other tumors showing DEAR1 mutation/deletion or loss of function, for more effective treatment with specific targeted therapies aimed at the SMAD3 pathway. In addition, there is a critical need to identify novel prognostic biomarkers of aggressive disease for which increased surveillance and treatment might be necessary. Our results suggest further studies to examine the potential clinical utility of DEAR1 and SMAD3/SNAI2 as prognostic biomarkers for aggressive and invasive disease.

Immortalized HMEC cells, 76N-E6, were propagated in D-Medium (6); HEK293T, U2OS, and HeLa cell lines (American Type Culture Collection) were propagated in Dulbecco's Modified Eagle Medium (DMEM) with 10% FBS. MCF-7 cells were grown in DMEM/F12 medium with 10% FBS. MCF10A cells were cultured in medium as previously described (46). TGF-β1 and MG132 were from Calbiochem Co., and cycloheximide was from Sigma. The Luciferase kit was from Promega (see Supplementary Methods).

DshR (989-CGCCAAAGCGCTTCGATGT) and CshR constructs for transient transfection were generated by inserting the synthesized oligonucleotides (Invitrogen) into the pSuper-shRNA vector, which was a generous gift of Dr. Bryan R. Cullen (Duke University Medical Center, Durham, NC; ref. 47). pcDNA3Myc/SMAD2, pcDNA3Myc/SMAD3, and pcDNA3Myc/SMAD4 vectors were gifts from Dr. Michelle Barton (The MD Anderson Cancer Center; ref. 48). HA-ubiquitin vector was kindly provided by Dr. Edward Yeh. pcDNA3/SMAD3 (CAGA)12-MLP-Luc reporter construct (pCAGA12-luc) and pPAI-1-Luc were generous gifts of Dr. P. ten Dijke (Leiden University Medical Center, The Netherlands; ref. 26). Flag-SMAD3 N and LC (linker and MH2 domains) constructs were generous gifts from Dr. Hyunjung Ha (Chungbuk National University, Republic of Korea; ref. 49). pRetroSuper-GFP SMAD3 shRNA plasmids were from Addgene (ref. 50; The Addgene plasmid 15723; see Supplementary Methods).

Dear1 null mice were generated by deletion of the 5′ end of exon 1, containing the first ATG, via replacement with a neomycin selection cassette. Following germline transmission, mice were intercrossed to maintain an identical background across genotypes. All mice used in this study were littermates (see Supplementary Methods).

3D culture assays for acini formation were conducted as described by Debnath and colleagues (46). For time lapse, cells were plated on the top of Matrigel (Invitrogen). After 6 hours of incubation, cells were monitored under the inverted phase microscope. Images were captured at 30-minute intervals for 72 hours in five to seven different fields of each sample simultaneously. Recorded images were stacked as movies and average migration distances determined for at least 30 cells per field. Cell migration distance was counted and measured using the MetaMorph Program.

GST and GST–DEAR1 fusion protein were expressed in BL21 cells and purified according to the manufacturer's instructions (GE Healthcare Life Science). For the GST pull-down assay, HEK293T cells were transfected with Flag-SMAD3 for 24 hours and then lysed in radioimmunoprecipitation assay lysis buffer. The lysates in GST-binding buffer (20 mmol/L Tris–HCl pH 8.0, 0.1 mol/L NaCl, and 0.2 mmol/L EDTA) were precleared by incubating with control GST and glutathione-sepharose beads for 2 hours at 4°C. Equal amounts of precleared lysates were then incubated with either GST–DEAR1 or control GST for 2 hours at 4°C. The beads were washed five times with GST-binding buffer and then eluted in 30 μL of 2× SDS sample buffer, and then detected by immunoblotting.

For in vivo ubiquitination assay, HEK293T cells were transfected with the indicated plasmids. At 20 hours after transfection, cells were treated with MG132 20 μmol/L for 4 hours and harvested. Cells were lysed in SDS-lysis buffer by boiling for 10 minutes and then diluted 10-fold in 0.5% NP-40 buffer and used for further analysis.

Exponentially growing cells on glass were fixed with 4% formaldehyde and permeabilized with 0.3% Triton X-100 in PBS before incubation with blocking buffer and antibodies. Stained cells were mounted with cover slips by antifade fluorescent mounting medium (ProLong Gold) and observed and recorded by deconvolution microscopy. Colocalization experiments using confocal microscopy and the Imaris imaging system were carried out as described previously (28).

Cells from DEAR1-KD and its control clones (2 × 10⁵/well) were seeded in 6-well plates. Cells were incubated for 24 hours until confluent. The wounds were made by scraping with 200-μL plastic tip across the cell monolayer. The wounded cells were treated or untreated with 2 ng/mL TGF-β. Phase contrast images were recorded after 1 to 2 days of culture.

Freshly trypsinized cells (4 × 10⁵/dish) were seeded into ULA 60-mm cell culture dishes (CoStar) with a covalently bound hydrogel layer that effectively inhibits cellular attachment. After 4 days of culture, suspension cells were collected and plated to a regular 100-mm dish for colony formation. Plates were stained by crystal violet and scanned with an Epson Perfection 4990 photo scanner 4 days later. Also, the suspension cells were collected and lysed by 1× SDS sample buffer for Western blotting.

See Supplementary Methods.

All data are expressed as mean with SDs. The statistical significance of differences between two groups was analyzed by two-tailed Student t tests. Differences with P values of less than 0.05 were considered statistically significant.

D.S. Chandler, S.T. Lott, and A.M. Killary have ownership interest (including patents) in US Patent UTSPN 6943245 and US Patent UTSPN 7169384. No potential conflicts of interest were disclosed by the other authors.

Conception and design: N. Chen, S. Balasenthil, D.S. Chandler, G. Lozano, M.L. Frazier, S.T. Lott, A. McNeill Killary

Development of methodology: N. Chen, S. Balasenthil, Y. Wang, D.S. Chandler, A. Rashid, I.I. Wistuba, S.T. Lott

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Chen, J. Reuther, A. Frayna, Y. Wang, D.S. Chandler, A. Rashid, G. Lozano, S.T. Lott

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Chen, J. Reuther, Y. Wang, D.S. Chandler, L.V. Abruzzo, J. Rodriguez, J. Chen, M.L. Frazier, A.A. Sahin, I.I. Wistuba, S.T. Lott, A. McNeill Killary

Writing, review, and/or revision of the manuscript: N. Chen, S. Balasenthil, J. Reuther, D.S. Chandler, M.L. Frazier, A.A. Sahin, S. Sen, S.T. Lott, A. McNeill Killary

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Chen, J. Reuther, J. Rodriguez, Y. Cao, E. Lokken, A.A. Sahin, S.T. Lott, A. McNeill Killary

Study supervision: N. Chen, A. McNeill Killary

The authors thank Dr. John Parant for assistance with targeting construct design, and G. Lozano for initial support for electroporation of construct [National Cancer Institute (NCI) grant to G. Lozano]. The authors also thank Dr. Richard Behringer for assistance with the immunohistochemical analysis of tumors from the mouse model as well as Dr. Mien-Chie Hung (MD Anderson Cancer Center) for critical reading of the article, and Hank Adams for microscopy technical support. The authors acknowledge the Genetically Engineered Mouse Facility for help with the electroporation of constructs as well as the DNA extraction core and the DNA Analysis Core for genotyping of knockout mice and qRT-PCR support, and the Histology core for immunohistochemical staining; and the TCGA project groups involved in the compilation of the TCGA data associated with breast carcinoma, colon and rectal carcinoma, squamous cell lung carcinoma, endometrial carcinoma, clear cell renal carcinoma, and glioblastoma.

The research described herein was supported by the NCI Early Detection Research Network grant (5U01CA111302-08; to A. McNeill Killary, S. Sen, and M.L. Frazier), the Department of Defense grant (BC111524; to N. Chen and A. McNeill Killary), the Metastasis Research Center, The University of Texas MD Anderson Cancer Center (to A. McNeill Killary), and the Rosalie B. Hite Fellowship, the American Legion Auxiliary scholarship, and the PEO (Philanthrophic Education Organization) Scholars award (to J. Reuther).