PDLIM5 inhibits STUB1-mediated degradation of SMAD3 and promotes the migration and invasion of lung cancer cells

Transforming growth factor  (TGF  ) signaling plays an important role in regulating tumor malignancy, including in non-small cell lung cancer (NSCLC). The major biological responses of TGF  signaling are determined by the effector proteins SMAD2 and SMAD3. However, the regulators of TGF  -SMADs signaling are not completely revealed yet. Here, we showed that the scaffolding protein PDZ and LIM domain protein 5 (PDLIM5, ENH) critically promotes TGF  signaling by maintaining SMAD3 stability in NSCLC. Firstly, PDLIM5 was highly expressed in NSCLC compared with that in adjacent normal tissues, and high PDLIM5 expression was associated with poor outcome. Knockdown of PDLIM5 in NSCLC cells decreased migration and invasion in vitro , and lung metastasis in vivo . In addition, TGF  signaling and TGF  -induced epithelial-mesenchymal transition (EMT) was repressed by PDLIM5 knockdown. Mechanistically, PDLIM5 knockdown resulted in a reduction of SMAD3 protein levels. Overexpression of SMAD3 reversed the TGF  -signaling-repressing and anti-migration effects induced by PDLIM5 knockdown. Notably, PDLIM5 interacted with SMAD3 but not SMAD2, and competitively suppressed the interaction between SMAD3 and its E3 ubiquitin ligase STUB1. Therefore, PDLIM5 protected SMAD3 from STUB1-mediated proteasome degradation. STUB1 knockdown restored SMAD3 protein levels, cell migration and invasion in PDLIM5-knockdown cells. Collectively, our findings indicate that PDLIM5 is a novel regulator of basal SMAD3 stability, with implications for controlling TGF  signaling and NSCLC progression.


INTRODUCTION
Transforming growth factor  (TGF) signaling plays essential and multifaceted roles in tumor malignancy (1). At the initial stage of tumor formation, TGF acts as a tumor suppressor by inhibiting cell proliferation; however, at the later stage, TGF plays pro-metastatic roles (2,3). The pro-metastatic effects of TGF signaling include accelerated tumor infiltration, cell adhesion, angiogenesis, cell epithelial-mesenchymal transition (EMT), and extracellular matrix degradation.
In previous decades, lung cancer constituted the leading cause of cancer mortality worldwide (4,5). Clinical data indicate that non-small cell lung cancer (NSCLC) accounts for most lung cancer cases.
Nevertheless, although numerous therapeutic approaches are available for treating NSCLC, its prognosis remains poor (6). As with most cancers, metastasis represents the dominant reason of NSCLC-related death (7). According to clinical studies, TGF signaling is frequently enhanced in NSCLC, promoting EMT and tumor metastasis (8,9). SMAD2 and SMAD3 are two key downstream regulators of TGFβ signaling and play critical roles in TGFβ-mediated biologic functions (10). SMAD3 inhibition diminishes TGFβ-induced EMT in cancer cells (8). In addition, the overexpression of SMAD3 accelerates TGFmediated metastasis in NSCLC cells (11). Notably, the posttranslational modification of SMAD3 protein is crucial for regulation of TGFβ signaling. The E3 ligase DEAR has been reported to interact with SMAD3 thus resulting in SMAD3 polyubiquitination and degradation (12). Moreover, phosphorylation at Thr66 in SMAD3, mediated by GSK3, primes the protein for phosphorylationdirected ubiquitination (13). Hence, it is important to fully understand SMAD3 regulatory mechanisms and how their regulation influences the TGFβ signaling pathway. 3 The scaffolding protein PDLIM5 (also termed ENH), a member of the PDZ-LIM domain protein family, was initially identified as a PKC kinase binding protein (14). PDLIM5 contains one PDZ and three LIM domains that mediate proteinprotein interactions (15). In addition to PKC, PDLIM5 binds multiple protein kinases, including PKD, AMPK, and PKA, making it a versatile scaffolding protein in signaling transduction pathways (16,17). PDLIM5 itself is a substrate of AMPK; its phosphorylation represses the migration of vascular smooth muscle cells (18). Furthermore, PDLIM5 functions as a cytoplasmic sequestration protein for transcription factors ID2 and YAP (19,20). PDLIM5 is upregulated in various cancers (21,22). It has been suggested that PDLIM5 promotes EMT and migration of prostate cancer cells (23). In addition, PDLIM5 regulates TGF3 signaling in vascular smooth muscle cells and hypoxia-induced pulmonary hypertension in mice (24). Moreover, silencing of PDLIM5 suppresses TGF signaling and SMAD3 expression in alveolar epithelial cells (25). However, the role of PDLIM5 in NSCLC and in regulation TGF signaling remains largely unknown.
In the current study, we found that PDLIM5 was upregulated in NSCLC tissues. PDLIM5 knockdown in NSCLC cells inhibited cellular migration and invasion, and TGF signaling. PDLIM5 interacted with SMAD3 and PDLIM5 knockdown enhanced the interaction between SMAD3 and its E3 ligase STUB1, resulting in SMAD3 degradation. Together, these results demonstrated that PDLIM5 is a novel SMAD3 regulator linking to SMAD3-mediated TGFβ signaling and lung cancer progression.

PDLIM5 is upregulated in NSCLC and associated with poor outcome.
To evaluate the importance of PDLIM5 in NSCLC, PDLIM5 expression was analyzed using datasets available in the Oncomine database (https://www.oncomine.org/). PDLIM5 was highly expressed in three different cohorts of NSCLC patients ( Fig. 1A-C). We confirmed the upregulation of PDLIM5 in NSCLC samples by western blot analysis (Fig. 1D). Moreover, PDLIM5 were significantly increased in NSCLC tissues compared to paired adjacent normal lung tissues as evaluated by immunohistochemistry assay in 40 NSCLC subjects ( Fig. 1E and F). Finally, survival analysis was performed to estimate the association between PDLIM5 expression and lung cancer patient survival using the Kaplan-Meier Plotter online bioinformatics resource (http://kmplot.com/analysis/), which revealed that elevated PDLIM5 expression in lung cancer patients was correlated with poor prognosis (Fig. 1G). Thus, PDLIM5 is upregulated in lung cancer and associated with poor outcome. 4 investigate the function of PDLIM5. PDLIM5 knockdown significantly inhibited the woundhealing capacity (Fig. 2D-E) and cellular migration ( Fig. 2F-G) of A549 and H1975 cells. Cell invasion was also greatly halted in PDLIM5-knockdown lung cancer cells (Fig. 2H-I). In addition, PDLIM5 was overexpressed in PC9 cells. PDLIM5 overexpression increased wound-healing capacity, cellular migration, and invasion ( Fig. S1A-D). To explore the role of PDLIM5 in lung metastasis in vivo, we intravenously injected Pdlim5knockdown LLC (Lewis lung carcinoma) cells (Fig.  2J) into BALB/c nude mice via tail vein. Lung metastatic nodules were significantly reduced in mice injected with Pdlim5-knockdown cells compared to those in mice receiving control LLC cells ( Fig. 2K-L). Together, these observations suggest that PDLIM5 promotes cell migration, invasion, and lung metastasis of lung cancer cells. Next, PDLIM5-knockdown in A549 and H1975 cells reduced cell spreading (Fig. S2A) and adhesion to vascular endothelial cells ( Fig. S2B-C). Using a Real-time Cell Analyzer (RTCA) to record cell adhesion (26), we found that PDLIM5 knockdown reduced cell adhesion to fibronectin (Fig. S2D). Thus, PDLIM5 is required for cell spreading, cell-ECM and cell-cell interactions. PDLIM5 knockdown in A549 and H1975 cells also significantly attenuated vasculogenic mimicry ( Fig.  S2E-F), a process associated with tumor malignancy whereby tumor cells mimic the vascular network (27). Moreover, PDLIM5knockdown did not affect cell proliferation, colony formation, and the expression of cell cycle markers PCNA and Cyclin D1 (Fig. S1F-H). Consistently, cell proliferation was not altered in PC9 cells with PDLIM5 overexpression (Fig. S1E). Furthermore, we isolated mouse embryonic fibroblasts from Pdlim5 knockout and wild-type mouse embryos to evaluate the universality of the findings. As in lung cancer cells, Pdlim5 deletion decreased wound-healing capacity without affecting cell proliferation ( Fig. S1I-K).

PDLIM5 knockdown attenuates TGF signaling.
Cell migration and invasion are regulated by TGF signaling and PDLIM5 has been reported to be related to this signaling (24,25). To examine the role of PDLIM5 in regulating TGF signaling in lung cancer, gene set enrichment analysis was used with the Cancer Genome Atlas database of lung cancer to analyze the potential association between PDLIM5 and TGF pathway. PDLIM5 was strongly co-expressed with genes belong to TGF pathway in lung cancer (Fig. 3A). TGFβ1 markedly induced SBE-Luc activity in A549 cells, which was significantly repressed by PDLIM5 knockdown (Fig. 3B). PAI1 and JUNB are two well-known down-stream targeting genes of TGF signaling, whose expression was notably decreased in PDLIM5-knockdown A549 and H1975 cells (Fig.  3C) and in embryonic fibroblasts isolated from Pdlim5-knockout mice (Fig. S3A). Consistent with this, forced expression of PDLIM5 in PC9 cells increased the expression of PAI1 and JUNB (Fig.  S3B). EMT is typically induced in lung cancer cells treated with TGF1 (9). Accordingly, upon TGF1 stimulation, A549 and H1975 cells exhibited a fibroblast-like morphology, whereas this morphological change was blocked by PDLIM5knockdown (Fig. 3D). Furthermore, reversetranscription polymerase chain reaction (RT-PCR) and western blot analysis revealed that PDLIM5knockdown in A549 and H1975 cells significantly decreased TGF1-responsive and mesenchymalrelated gene expression following TGF1 treatment ( Fig. 3E and F). Interestingly, SMAD3 but not SMAD2 expression was decreased in PDLIM5 knockdown cells ( Fig.  3F-H), and in Pdlim5-knockout mouse embryonic fibroblasts (Fig. S3C). p-Smad3/Smad3 was 5 unaltered in TGF1 treated PDLIM5-knockdown cells (Fig. 3G). PDLIM5-knockdown did not affect the expression of SMAD4, SMAD7 and TGFR1 (Fig. 3H). In addition, PC9 cells with PDLIM5 overexpression increased SMAD3 level (Fig. S3D). Together, these results indicated that PDLIM5 regulates SMAD3-mediated TGF signaling.

SMAD3 mediates PDLIM5 function.
To explore whether PDLIM5 function in lung cancer cells was mediated by SMAD3, we first explored the role of SMAD3 in cell migration and invasion. SMAD3 knockdown in A549 cells inhibited cell wound-healing capacity, cellular migration and invasion ( Fig. S1L-P). The expression of SMAD3 was reduced in Pdlim5 knockout mouse lungs (Fig. S3E). Pdlim5 deletion in mice resulted in alveolarization retardation, which was similar to Smad3 knockout mice (28) (Fig. S3F-G). Therefore, similar to PDLIM5, SMAD3 promotes cancer cell migration and invasion and lung development. Next, we restored SMAD3 in PDLIM5-knockdown A549 cells. The decreased expression of PAI1 and JUNB in PDLIM5-knockdown A549 cells was recovered upon SMAD3 over-expression (Fig. 4A). The repression of EMT induced by PDLIM5 knockdown was also attenuated by SMAD3 overexpression (Fig. 4B). Furthermore, SMAD3 ectopic expression rescued PDLIM5 knockdowninduced defects in cellular attachment to fibronectin (Fig. S2G) and vasculogenic mimicry (Fig. S2H). Finally, the decreased wound-healing capacity, cellular migration and invasion in PDLIM5-knockdown A549 cells were all restored by SMAD3 over-expression ( Fig. 4C-F). Taken together, these results clearly support our assessment that the defects in PDLIM5-knockdown cells were due to the decrease of SMAD3.

PDLIM5 suppresses proteasome-dependent degradation of SMAD3.
To explore the regulation between PDLIM5 and SMAD3, first, we analyzed PDLIM5 and SMAD3 mRNA levels in NSCLC cell lines available in the Expression Atlas database. No relationship between PDLIM5 and SMAD3 mRNA levels was suggested (r = 0.03, p = 0.77) (Fig. 5A). In contrast, PDLIM5 and SMAD3 protein levels were highly correlated in NSCLC cell lines and NSCLC tissues ( Fig. 5B and C). Further, RT-PCR analysis indicated that PDLIM5-knockdown did not alter SMAD3 mRNA level in A549 and H1975 cells (Fig.  5D). These observations indicated that in the lung cancer cells, PDLIM5 specifically regulates SMAD3 stability. Next, in order to evaluate SMAD3 turn-over we treated A549 cells with cycloheximide (CHX) to inhibit protein synthesis. PDLIM5 knockdown promoted SMAD3 turn-over (Fig. 5E), whereas this was markedly inhibited upon PDLIM5 overexpression in A549 cells (Fig. 5F). Since ubiquitination is an important mechanism for regulating SMAD3 stability, we co-expressed Flag-SMAD3 and Ubiquitin (Ub) constructs in A549 cells to evaluate whether PDLIM5 regulates SMAD3 ubiquitination. We found that SMAD3 ubiquitination was enhanced upon PDLIM5 knockdown (Fig. 5G), and repressed after PDLIM5 overexpression (Fig. 5G). Importantly, MG132 (an inhibitor of proteasome degradation) effectively inhibited SMAD3 decrease in PDLIM5knockdown cells (Fig. 5H). Collectively, these findings indicate that PDLIM5 protects SMAD3 from proteasome-mediated degradation.

PDLIM5 interacts with SMAD3
It has been reported that the serine/threonine kinase GSK3 is responsible for SMAD3 degradation 6 (13). However, GSK3 activity, as indicated by its phosphorylated state, was not altered in PDLIM5 knockdown A549 cells (Fig. S4A). Moreover, GSK3 knockdown (Fig. S4B) and a GSK3 inhibitor (LiCl) (Fig. S4C) all failed to restore SMAD3 levels in PDLIM5-knockdown A549 cells. Then, we questioned whether PDLIM5 stabilizes SMAD3 by direct binding. To test this, Flag-tagged SMAD3 and HA-tagged PDLIM5 were ectopically expressed in HEK293T cells. Protein pellets precipitated using the anti-Flag antibody contained the HA-tagged PDLIM5 protein. Specifically, no interaction between SMAD2 and PDLIM5 was detected (Fig. 6A). Immunoprecipitation of endogenous proteins in A549 cells confirmed the interaction between PDLIM5 and SMAD3 (Fig.  6B). We therefore proceeded to map the interacting domains in PDLIM5 and SMAD3. LIM domains, the PDZ domain, and PDLIM5 without LIM domains (ΔLIM) were individually expressed in HEK293T cells. SMAD3 was found to interact with the LIM but not the PDZ domain of PDLIM5 (Fig. 6C). In comparison, either the SMAD3 MH1 or MH2 domain interacted with PDLIM5 (Fig. 6D). Last, PDLIM5 co-localization with SMAD3 as revealed by immunofluorescence assay further confirmed this interaction (Fig. 6E). Therefore, PDLIM5 interacts with SMAD3 but not SMAD2. To answer whether the interaction between PDLIM5 and SMAD3 plays a role in stabilizing SMAD3 by PDLIM5, we first transfected truncated PDLIM5 without LIM domains (ΔLIM) in PC9 cells to detect whether it promotes SMAD3 expression. As expected, PDLIM5 mutation, destroying the PDLIM5-SMAD3 interaction, failed to upregulate SMAD3 in PC9 cells (Fig.  S3D). Similarly, PDLIM5 re-expression restored the wound-healing capacity, cellular migration and invasion in PDLIM5-knockdown cells. However, expression of PDLIM5 mutation (ΔLIM) did not restore the defects (Fig. S3H-L). Thus, the interaction is required for PDLIM5 to stabilize SMAD3.
PDLIM5 stabilizes SMAD3 by counteracting the interaction between SMAD3 and STUB1.
As several E3 ligases reportedly regulate SMAD3 degradation, we hypothesized that PDLIM5 may modulate SMAD3 degradation by regulating the activity of certain E3 ligase. An immunoprecipitation-mass spectrometry (IP-MS) experiment revealed that PDLIM5 bound multiple E3 ligases, among which STUB1 is a E3 ligase for SMAD3 (Fig. 7A). The interaction was confirmed in HEK293T cells expressing HA-PDLIM5 and Flag-STUB1 by a co-immunoprecipitation assay (Fig. S4D). Over-expression of STUB1 decreased SMAD3 without altering SMAD2 in A549 cells (Fig. S4E). Moreover, PAI1 and JUNB expression was reduced in STUB1 over-expressed cells (Fig.  S4F). Importantly, STUB1-knockdown restored SMAD3 level in PDLIM5 knockdown cells (Fig.  7B). Specifically, STUB1 interacted with the MH2 domain of SMAD3 ( Fig. S4G and Fig. 7C). Since both PDLIM5 and STUB1 interacted with the MH2 domain of SMAD3, we next tested whether PDLIM5 affected the interaction between STUB1 and SMAD3. Immunoprecipitation analysis revealed that reduced PDLIM5 expression enhanced the interaction between STUB1 and SMAD3, whereas PDLIM5 overexpression perturbed this interaction ( Fig. 7D and E). Next, acceptor bleach FRET (fluorescence resonance energy transfer) experiments were carried out with CFP-SMAD3 as the donor and YFP-STUB1 as the acceptor to verify the results. Photobleaching of YFP-STUB1 significantly increased the postbleaching fluorescence intensity of CFP-SMAD3, indicating significant FRET between these two protein. The FRET between CFP-SMAD3 and YFP-STUB1 was disrupted upon PDLIM5 7 overexpression, suggesting that PDLIM5 repressed STUB1 interacting with SMAD3 (Fig. 7F). Moreover, the decrease of SMAD3 in STUB1overexpression A549 cells was reversed by the expression of PDLIM5 but not LIM-domain deletion PDLIM5 mutation (Fig. 7G). Notably, STUB1 overexpression in A549 cells reduced cell migration, invasion, cell adhesion and vasculogenic mimicry ( All of these observations indicate that PDLIM5 and STUB1 competitively bind to SMAD3, and PDLIM5 stabilizes SMAD3 by protecting it from STUB1-mediated proteasome degradation.

DISCUSSION
TGF signaling has a prominent role in NSCLC tumorigenesis and metastases (3). SMAD3, as a key mediator in the TGF signaling pathway, plays an essential role in tumor invasiveness. In this study, we found that PDLIM5 is a new interacting protein to SMAD3 but not SMAD2. The interaction disrupts the binding between SMAD3 and its E3 ligase STUB1 to prevent SMAD3 degradation, thus promoting TGF signaling, and cancer cell migration, invasion, and NSCLC progression. Our findings therefore demonstrated a novel mechanism for PDLIM5 in regulating TGFβ-SMAD3 signaling and NSCLC progression.
PDLIM5 belongs to the PDZ-LIM protein family, featured as containing one PDZ domain and one or three LIM domains. The most studied PDZ-LIM domain proteins in cancers are PDLIM1, PDLIM2, and PDLIM4. PDLIM1/CLP36 is upregulated to mediate breast cancer metastasis by interacting with -actinin (29); In contrast, PDLIM1 is downregulated to inhibit cellular migration and invasion in highly metastatic colorectal cancer cells and metastatic hepatocellular cancer (30,31). PDLIM2 controls COP9 signalosome activity and promotes the degradation of multiple tumorinducers including NF-B and STAT3 in colorectal cancer (32,33). Similar anti-tumor role of PDLIM2 is recently reported in lung cancer (34). PDLIM4 is a tumor suppressor and reportedly downregulated in various cancers (35). Here, we showed that PDLIM5 is upregulated in NSCLC tissues, which is associated with poor outcome. Furthermore, our data revealed that PDLIM5 is required for NSCLC cell adhesion, migration, invasion, and lung metastasis in mice. Although it has been reported that PDLIM5 knockdown inhibits prostate cancer and thyroid carcinoma cell proliferation (23,36), we showed that PDLIM5 insufficiency in NSCLC cells and mouse embryonic fibroblasts does not inhibit cell proliferation, indicating that the function of PDLIM5 is context dependent. Therefore, the present results revealed a pro-tumor role of PDLIM5 in NSCLC.
Upon TGFβ stimulation, SMAD2 and SMAD3 are phosphorylated at their C-terminuses by the activated TGFβRI and shuttle the signal to nucleus, which is the main events in TGFβ signaling (37). Despite their similarity in sequence, SMAD2 and SMAD3 have different roles on tumor progression and are differentially regulated (38). In current study, we showed that PDLIM5 interacts with SMAD3 but not SMAD2 and stabilizes basal SMAD3. Several studies have demonstrated SMAD3 knockdown remarkably inhibits TGFinduced EMT and tumor metastasis in multiple cancers including NSCLC (11,39). In addition, the expression of SMAD3 target genes, including PAI1 and JUNB, are positively correlated with tumor progression (40,41). We found that both PAI1 and JUNB were downregulated in PDLIM5 8

knockdown NSCLC cells, and similar defects in cellular migration and invasion in both PDLIM5
and SMAD3 knockdown cells. Importantly, ectopic expression of SMAD3 prominently restored the deficiency in TGF signaling, migration and invasion caused by PDLIM5 loss. These observations demonstrated that PDLIM5 specifically regulates SMAD3-mediating TGF signaling.
TGF-SMADs pathway is regulated by ubiquitinmediated degradation (42). Various E3 ligases have been reported to mediate the degradation of multiple components in TGF signaling, including SMAD3 and SMAD2. PDLIM5 contains three LIM domains that mediate protein-protein interactions. Notably, LIM domain proteins have recently emerged as vital regulators that tightly control the activity of E3 ubiquitin ligase (43,44). Consistent with this, we found that PDLIM5 binds multiple E3 ligases, including STUB1. STUB1 ubiquitinates and degrades SMAD3 in a basal level, thus desensitizing the cells to TGF signaling (45). We showed that PDLIM5 stabilizes basal SMAD3 by repressing SMAD3 binding to STUB1. Whether other PDLIM5-interacting E3 ligases mediates the malignance of lung cancer requires more exploration.
The dichotomous roles of TGF in cancers complicate the development of TGF-targeted anticancer drugs. However, proteins that interact with SMAD3, such as PDLIM5, bring new light to this field. The reason is that targeting PDLIM5 does not promote the proliferation of NSCLC cells, while inhibits TGFβsignaling and tumor malignance.
In conclusion, we find that PDLIM5 specifically interacts with SMAD3 and prevents its degradation, and its expression is upregulated to promote TGFβ signaling and malignance of lung cancer. These findings pinpoint PDLIM5 as a potential prognostic biomarker and intervention target for NSCLC.

Bioinformatics analysis
The Oncomine (https://www.oncomine.org/) database was utilized to analyze the expression profiles of PDLIM5 in normal and NSCLC tissues. The survival Kaplan-Meier estimates of patients with NSCLC exhibiting different PDLIM5 expression levels was assessed using a Kaplan-Meier plotter (http://kmplot.com/analysis/).

Cell culture and transient transfection
Mouse lung cancer cell line LLC, human HEK293T cells and NSCLC cell lines A549, MSTO, H1975, H1299, H358, and PC9 were from the American Type Culture Collection (Manassas, VA, USA). Wild-type and Pdlim5 knockout mouse embryonic fibroblasts were isolated and cultured as described previously (46). Cells were grown in RPMI-1640 medium (Hyclone, Logan, UT, USA) supplemented with 10% fetal bovine serum.

Mice
Pdlim5 global knockout (Pdlim5 -/-) mice were bred from heterozygous mice, as described previously (47). The left lungs of Pdlim5 wild-type and knockout mice were embedded by paraffin. The tissues were cut into 5 m sections, and stained with hematoxylin and eosin for histological examination. 6-8-week-old nude mice were purchased from Shanghai Slack Laboratory Animal CO., Ltd (Shanghai, China) for in vivo metastasis assay. Mice were injected with LLC lung cancer cells suspended in phosphate-buffered saline (2.5  10 6 cells per mouse) via the tail vein. Twenty-eight days after the injection, the mice were euthanized by pentobarbital injection and the lungs were fixed in 4% paraformaldehyde (PFA) to prepare paraffin sections. The lung metastasis regions were counted under a light microscope. The total number of metastatic nodules in each mouse was determined. All animal experiments were reviewed and approved by the Animal Care and Use Committee of the Zhejiang University School of Medicine.

Lentivirus infection
To produce lentiviruses, HEK293T cells were transfected with plKO-puro shPDLIM5 (the target sequences are provided in Supplemental Table S1) or control shRNA (shScr) and packaging plasmids (pMD2.G and psPAX2, Sigma-Aldrich, St. Louis, MO, USA). Lentivirus-containing supernatants were collected 48 h after the transfection and filtered through a 0.22 m filter. Prior to infection, lentiviruses at same virus multiplicity of infection (MOI = 30) were incubated with polybrene (Beyotime, Shanghai, China) to increase the infection efficiency.

Transwell assay
For the migration assay, 2  10 4 cells suspended in 200 l of RPMI-1640 serum-free medium containing 0.5% bovine serum albumin were seeded into 8 m Boyden chambers (Corning, Armonk, NY, USA). For the invasion assay, 2  10 5 cells suspended in 200 l of RPMI-1640 serum-free medium containing 0.5% bovine serum albumin were plated in the upper chamber precoated with 50 l of 2 μg/ml matrigel (Corning, NY, USA). Then, 600 l of the complete medium was added to the lower chamber to encourage cell migration and invasion. The incubation time for migration was 24 h. The invasion required an additional 24 h. After incubation, the upper chambers were collected and fixed in methanol, and non-migrating or non-invading cells were removed. The migrating and invading cells were stained with crystal violet solution (Beyotime, Shanghai, China), and representative images were acquired using a light microscope (Olympus, Tokyo, Japan). The crystal violet-stained cells were then washed with 33% acetic acid, and absorbance was measured at 570 nm (Molecular Devices, Sunnyvale, CA, USA).

Cell adhesion assay
To detect cell adhesion to endothelial cells, H1975 and A549 cells were labeled with Calcein AM. The labeled cells (1  10 4 /100 l per well) were cocultured with an endothelial monolayer of human umbilical vein endothelial cells (isolated as described (48)) in 96-well plates for 1 h. After the incubation, detached cells were washed off with PBS, and the adherent cells were counted under a fluorescence microscope (Nikon Eclipse TE2000U-E, Tokyo, Japan). The xCELLigence RTCA instrument (Acea Biosciences, San Diego, California, USA) was used to evaluate the adhesion capacity of A549 and H1975 cells to fibronectin. For the experiment, 1  10 4 cells suspended in 200 l of RPMI-1640 medium containing 0.5% bovine serum albumin were seeded into 96-well E-plates pre-coated with 2.5 g fibronectin per well. The attachment was monitored by recording the cell index every minute for 2 h.

Vasculogenic mimicry assay
Vasculogenic mimicry assay was used to evaluate the vasculogenic mimicry of cancer cells in vitro. For the experiment, 10 l of matrigel (Corning, NY, USA) was placed in an u-Slide well (ibidi, Martinsried, Germany), and incubated at 37 C for 30 min to allow it to solidify. A549 and H1975 cells suspended in complete medium were seeded into the u-Slide at a density of 2  10 4 cells per well and incubated for 6 h. The formation of tubes was captured using an inverted light microscope at 100 magnification in five random fields. Tube length was also measured in five random fields, using the Image J software.

Immunofluorescence staining
Cells cultured on glass coverslips were fixed in 4% PFA and permeabilized using 0.5% Triton-X in PBS before incubation with an appropriate primary antibody (Table S2). The signal was visualized by incubating with goat anti-mouse antibodies conjugated with Alexa Fluor® 594 or goat antirabbit antibodies conjugated with Alexa Fluor®488 at room temperature for 1 h. The nucleus was labeled using 4,6-diamidino-2phenylindole (DAPI). Images were acquired by using the FV1000 confocal microscope (Olympus, Tokyo, Japan).

Western blot
All tissue and cellular proteins were extracted by RIPA lysis buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 0.1% SDS, 0.5% sodium deoxycholate, and 1% Nonidet P-40) following the manufacturer's instruction (Beyotime, Shanghai, China). The lysate was boiled at 100 C for 8 min, and the proteins were separated on 10% and 13% SDS-11 PAGE gels. They were then transferred onto nitrocellulose membrane (Pall, Port Washington, NY, USA), probed with primary antibody at 4 C overnight. The primary antibodies used herein were listed in Supplementary Table S2. The membranes were then washed three times with TBST (150 mM NaCl, 50 mM Tris, 0.05% Tween 20 [pH 7.6]) and probed with a secondary antibody (Table S2) at room temperature for 1 h. Finally, the target protein was visualized by using the Odyssey system (LI-COR Biosciences, Lincoln, NE, USA).

Co-immunoprecipitation
Cells were lysed in NP40 buffer (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 2 mM EDTA, and 1% NP40) freshly supplemented with a protease inhibitor (Roche, Basel, Switzerland) and a phosphatase inhibitor (Roche, Basel, Switzerland). Then, 30 l of magnetic beads (BioRad, Hercules, CA, USA) were incubated with 3 g of the appropriate antibody with agitation for 10 min at 37 °C. Cell lysates were incubated with antibody-conjugated magnetic beads at 4°C overnight. The immunoprecipitates were then eluted by boiling in SDS sample buffer (Beyotime, Shanghai, China) for 5 min. For the ubiquitination assay, cells transfected with Flag-SMAD3 and ubiquitin constructs were treated with 10 M MG132 for 8 h and then lysed for a subsequent coimmunoprecipitation assay.

IP-MS
A549 cells were transduced with HA-PDLIM5 plasmids for 24 h, then lysed and incubated with anti-HA antibody-conjugated beads. The immunoprecipitate was subjected to SDS-PAGE. After electrophoresis, the protein gel was stained with Coomassie brilliant blue and then analyzed by MS by Qinglian bio (Beijing, China).

Double luciferase reporter assay
The SBE reporter assay was performed as previously described (50). Reporter gene activity was detected according to the protocol for the Dual-Luciferase Reporter Gene Assay Kit (Promega, WI, USA) and measured by using SpectraMax M5 (Molecular Devices, Sunnyvale, CA).

Quantitative RT-PCR
Total RNA was isolated using TRIzol reagent (Invitrogen, Waltham, CA, USA). For cDNA synthesis, 1 g total RNA was transcribed using the ReverTraAce qPCR RT kit (Toyobo Inc, Osaka, Japan). The quantitative PCR reaction was performed using the LightCycler Roche480 (Roche Molecular Systems, Basel, Switzerland) according to the manufacturer's instructions of the SYBR Green dye (cwbio, Beijing, China). Gene expression was normalized to the endogenous control (18S RNA) and calculated using 2-ΔΔCt. All primers used for the analysis were listed in Supplementary Table S3.

Acceptor photobleaching FRET
CFP-SMAD3 (donor) and YFP-STUB1 (acceptor) together with HA-PDLIM5 or empty vector were co-transfected in A549 cell. After 24 h transfection, cells were fixed by 4% PFA. Acceptor photobleaching FRET were performed using the FV1000 confocal microscope (Olympus, Tokyo, Japan) as described previously (51). The CFP and YFP fluorophores were excited with an excitation wavelength of 458 nm and 514 nm respectively. After acquired the emission images, the cells marked by a region of interest and this region was by guest on November 4, 2020 http://www.jbc.org/ Downloaded from bleached by a high laser power (20 iterations, 100% laser power, 514 nm). The FRET efficiency was measured as the percentage increase of donor after photobleaching the acceptor. %FRET=100×(CFPpost-CFPpre) / CFPpost. The CFPpost and CFPpre were CFP emission after and before bleaching.

Statistical analysis
Statistical analysis was performed by GraphPad Prism 5. Data were presented as the mean ± SD. We performed unpaired t-tests to determine the significance of differences between two groups. For the comparison of more than two groups, one-way ANOVA was used. Wilcoxon matched-pairs signed-rank test was used to determine the expression difference between the cancer tissue and adjacent normal tissue. A p-value < 0.05 was considered statistically significant.

Data Availability
All the data were included in the manuscript.     Representative images of the wound-healing assay of PDLIM5 knockdown A549 cells with or without STUB1 knockdown. Images were captured at 0 and 24 h after scratching. Scale bar: 200 m. (H) The wound-healing rate was analyzed by the Image J software (n = 3). Data were shown as mean ± SD. Analysis was performed using two-tailed Student's t-test for C, D and one way ANOVA with Tukey post hoc test for F, H. * p < 0.05, ** p < 0.01, *** p < 0.001.