**CD99–PTPN12 Axis Suppresses Actin Cytoskeleton-Mediated Dimerization of Epidermal Growth Factor Receptor**

**Kyoung-Jin Lee 1 , Yuri Kim <sup>1</sup> , Min Seo Kim <sup>1</sup> , Hyun-Mi Ju 1 , Boyoung Choi 1 , Hansoo Lee 2 , Dooil Jeoung 3 , Ki-Won Moon 4 , Dongmin Kang 5 , Jiwon Choi 6,7 , Jong In Yook 6,7 and Jang-Hee Hahn 1, \***


Received: 4 September 2020; Accepted: 5 October 2020; Published: 9 October 2020

**Simple Summary:** The epidermal growth factor receptor (EGFR) is activated through growth factor-dependent dimerization accompanied by functional reorganization of the actin cytoskeleton. Lee et al. demonstrate that CD99 activation by agonist ligands inhibits epidermal growth factor (EGF)-induced EGFR dimerization through impairment of cytoskeletal reorganization by protein tyrosine phosphatase non-receptor type 12 (PTPN12)-dependent c-Src/focal adhesion kinase (FAK) inactivation, thereby suppressing breast cancer growth.

**Abstract:** The epidermal growth factor receptor (EGFR), a member of ErbB receptor tyrosine kinase (RTK) family, is activated through growth factor-induced reorganization of the actin cytoskeleton and subsequent dimerization. We herein explored the molecular mechanism underlying the suppression of ligand-induced EGFR dimerization by CD99 agonists and its relevance to tumor growth in vivo. Epidermal growth factor (EGF) activated the formation of c-Src/focal adhesion kinase (FAK)-mediated intracellular complex and subsequently induced RhoA-and Rac1-mediated actin remodeling, resulting in EGFR dimerization and endocytosis. In contrast, CD99 agonist facilitated FAK dephosphorylation through the HRAS/ERK/PTPN12 signaling pathway, leading to inhibition of actin cytoskeletal reorganization via inactivation of the RhoA and Rac1 signaling pathways. Moreover, CD99 agonist significantly suppressed tumor growth in a BALB/c mouse model injected with MDA-MB-231 human breast cancer cells. Taken together, these results indicate that CD99-derived agonist ligand inhibits epidermal growth factor (EGF)-induced EGFR dimerization through impairment of cytoskeletal reorganization by PTPN12-dependent c-Src/FAK inactivation, thereby suppressing breast cancer growth.

**Keywords:** actin cytoskeletal reorganization; breast cancer; CD99 agonist; EGFR dimerization; endocytosis; FAK dephosphorylation; PTPN12; Rac1; RhoA; tripeptide

#### **1. Introduction**

Many studies have focused on uncovering the molecular basis of epidermal growth factor receptor (EGFR) activation and its implication in tumor development and progression. EGFR is activated by ligand binding, which induces sequentially their conformational change, auto- and trans-phosphorylation, dimerization, and internalization [1–3]. Structural study demonstrated that EGFR can be multimerized through a specific region of subdomain IV of the extracellular domain [4]. On the other hand, an increasing number of studies have suggested different aspects of EGFR activation. An inactive pre-formed dimer of EGFR without ligand was identified at the cell surface, which undergoes conformational changes during the activation process by stimulated ligand binding [5–7]. In spite of various aspects of the regulation of EGFR activation, dimerization of EGFR is a common feature required for its activation and transmission of downstream signals in tumorigenesis.

Non-receptor type 12 protein tyrosine phosphatase (PTPN12) acts as a core regulator in actin cytoskeleton-mediated modulation of growth factor receptor dynamics [8]. PTPN12 controls Rho GTPase activity by suppressing the interaction of p120 catenin with guanine nucleotide exchange factor VAV2 [9–12]. More recently, it was demonstrated that ephrin receptor (Eph) signaling depends on cytoskeletal reorganization to form polymeric assembly of the receptors and that PTPN12 contrarily downregulates EphA3 activity by inhibiting actin cytoskeletal remodeling. Interestingly, PTPN12 acts as a tumor suppressor which regulates the activities of multiple oncogenic tyrosine kinases, including EGFR, human epidermal growth factor receptor 2 (HER2), and platelet-derived growth factor receptor-β (PDGFRβ), and its deficiency is identified in several carcinomas [13,14]. Loss of PTPN12 promotes in vivo tumor progression of the implanted breast cancer cells, which express a constitutively active form of ErbB2 (CA-ErbB2), and correlates with the impaired feedback regulation of RTKs, thereby resulting in their aberrant activation [15,16]. In this regard, the expression levels of PTPN12 correlate inversely with poor prognosis of hepatocellular carcinoma [17]. Therefore, these results suggest that PTPN12 may play a role in regulation of growth factor receptor dimerization through actin cytoskeleton remodeling.

CD99 is a 32-kDa heavily O-glycosylated type I transmembrane protein [18]. CD99 is expressed on the surface of nearly all normal cell types including thymocytes, peripheral T cells, hematopoietic cells, and also several tumors including Ewing's sarcoma [19,20]. It has been known that CD99 is implicated in various cellular processes including differentiation, apoptosis, homotypic aggregation, and proliferation of lymphocytes, extravasation of leukocytes, transport of several transmembrane proteins, and apoptosis of tumor cells [21–23]. We previously reported that CD99CRIII3, a CD99 agonistic peptide ligand, activated PKA-SHP2-HRAS-ERK1/2 signal transduction pathway, which led to upregulation of PTPN12 expression and interaction with its downstream targets, FAK and PIN1, resulting in dephosphorylation of FAK at Y397 [24]. Consistent with our study, CD99 regulates contact strength and motility of osteosarcoma cells through inhibition of the expression of coiled-coil containing protein kinase 2 (ROCK2), which is a crucial mediator of actin cytoskeleton remodeling [25]. Inhibition of ROCK2 expression leads to a significant decrease in expression and phosphorylation of Ezrin, thereby collapsing the crosslinks between the plasma membrane and cytoskeleton. These results prompted us to hypothesize that CD99 activation can regulate actin cytoskeleton dynamics through the PTPN12/FAK/Rho/Rac axis, thereby suppressing EGFR dimerization and activation.

In this study, we examined the molecular mechanism through which CD99 agonist ligand suppresses ligand-induced dimerization and internalization of EGFR in breast carcinoma cells. Our study suggests that CD99-derived agonist ligands inhibit EGF-induced EGFR dimerization

through impairing RhoA-Rac1 signaling-mediated reorganization of the actin cytoskeleton, thereby contributing to the suppression of breast cancer growth.

#### **2. Results**

#### *2.1. Epidermal Growth Factor Stimulates Dimerization and Activation of the Epidermal Growth Factor Receptor through c-Src*/*FAK-Mediated Actin Cytoskeleton Remodeling*

Previous studies showed that lovastatin, a statin medication, inhibits EGF-induced EGFR dimerization, activation, and downstream signaling through inhibition of RhoA-mediated actin polymerization and that ligand-induced remodeling of the actin cytoskeleton is required for clustering of transmembrane receptors, thereby resulting in endocytosis and signal transduction [26–28]. We determined whether impairing actin polymerization could disrupt ligand-induced dimerization of receptor tyrosine kinases in two human breast cancer cell lines, low EGFR-expressing MCF-7 and high EGFR-expressing MDA-MB-231. Recombinant human EGF induced EGFR dimerization in a time-dependent manner in MDA-MB-231 cells, but not in actin filament-disrupted cells treated with cytochalasin D (Figure 1A). EGF treatment could induce actin polymerization and stimulate EGFR dimerization and phosphorylation at tyrosine (Y) 1068 residue in a dose-dependent manner (Figure 1B,F and Figure S1A,B,E). In contrast, disruption of actin filaments by cytochalasin D significantly interfered with EGF-induced phosphorylation and dimerization of EGFR. Dose-dependent inhibitory effect of cytochalasin D on EGFR dimerization was confirmed by in situ proximity ligation assay (PLA) (Figure S1B). Furthermore, we confirmed that actin polymerization is critical for not only EGFR/EGFR homo-dimerization but also EGFR/HER2 hetero-dimerization. EGFR/HER2 hetero-dimerization was observed from a very early time point after treatment of MCF-7 cells with EGF and increased in a time-dependent manner, whereas cotreatment with cytochalasin D significantly reduced hetero-dimerization (Figure S1C). These results indicate that actin polymerization is necessarily required for ligand-induced receptor dimerization.

**Figure 1.** *Cont.*

**Figure 1.** EGF induces EGFR dimerization and endocytosis through c-Src/FAK-mediated cytoskeleton reorganization. (**A**,**D**) The dimerization level of EGFR was assessed by in situ proximity ligation assay (PLA). (**B**,**C**) For dimerization assay, human breast carcinoma cells were treated with increasing concentrations of EGF in the presence or absence of cytochalasin D for 1 h on ice, to allow for EGF-induced EGFR dimerization but not endocytosis. Cells were subjected to BS<sup>3</sup> chemical-mediated crosslinking as described in Materials and Methods. To determine the phosphorylation level of EGFR at Y1068, cells were incubated in serum-free medium (SFM) with EGF for 15 min at 37 ◦C. Cell extracts were assessed by Western blot analysis with the indicated antibodies. β-actin was used as a loading control. \*\* *p* < 0.01; \*\*\* *p* < 0.001; \*\*\*\* *p* < 0.0001. (**E**,**F**) EGFR endocytosis and actin cytoskeleton organization were determined by immunofluorescent assay (IFA). (**A**,**D**,**E**,**F**) Original magnification of representative images, 600×. Scale bars = 10 µm.

Recruitment and activation of c-Src and FAK have been implicated in cell adhesion and motility by regulating actin cytoskeleton rearrangement and focal adhesion dynamics via activation of RhoA or Rac1/Cdc42 GTPases [29–31]. We determined whether inhibition of FAK function affects EGFR dimerization in the breast carcinoma cells. It was observed that EGF dose-dependently induced FAK phosphorylation at residue Y397 (Figure S1A). FAK knockdown revealed a markedly decreased rate of EGFR dimerization upon EGF binding (Figure 1C). To further investigate the functional relationship between c-Src/FAK-mediated actin rearrangement and EGFR dimerization and endocytosis, we carried out in situ PLA and immunofluorescent assay (IFA) after treatment with FAK small interfering RNA (siRNA), cytochalasin D, and dominant negative c-Src plasmid. Impairing actin polymerization with cytochalasin D or inhibiting c-Src/FAK signaling using dominant negative c-Src (DN-c-Src) or siRNA against FAK or c-Src inhibited EGF-induced EGFR receptor–receptor interaction, endocytosis, as well as actin polymerization (Figure 1D–F and Figure S1D,E). These results suggest that c-Src/FAK-mediated actin cytoskeleton rearrangement plays an important role in ligand-induced EGFR dimerization and activation.

#### *2.2. EGF Induces EGFR Dimerization and Endocytosis through FAK-Mediated RhoA and Rac1 Signaling*

Actin cytoskeletal reorganization is regulated by the Rho family of GTPases, including Rho, Rac, and CDC42 [32–35]. We found that although MCF-7 has low expression level of EGFR, EGF treatment dose-dependently stimulates upregulation of the activity of GTPases, Rac1 and RhoA, which is consistent with the results in Figure 1F and Figure S1E showing the pattern of increase in F-actin polymerization (Figure 2A). To determine the role of FAK in activating small GTPase signaling, we transiently transduced constitutively active FAK mutant (CA-FAK), dominant-negative FAK mutant (FAK Y397F) or FAK siRNA. Interaction of FAK with both the GTP-binding proteins and their GTPase activities were upregulated by overexpressing CA-FAK or treating with EGF (Figure 2B,C and Figure S2A). Contrarily, the increased interaction of GTPases with FAK and their upregulated GTPase activities were suppressed by overexpression of kinase-dead FAK Y397 mutant or by knockdown of FAK using siRNA. In addition, knockdown of FAK resulted in inhibition of EGF-induced EGFR endocytosis (Figure 2G). Furthermore, interactions among signaling molecules downstream of GTPases, including Wiskott-Aldrich syndrome protein (WASp) family Verprolin-homologous protein-2 (WAVE2), Actin-related protein-2 (ARP2), ROCK2, and Ezrin, showed patterns similar to those of FAK with RhoA and Rac1 (Figure 2D and Figure S2B). These results show that FAK contributes as a key regulator of RhoA and Rac1, leading to activation of GTPase signaling.

Next, we investigated the effects of activating and inhibiting RhoA and Rac1 GTPases on dimerization and endocytosis of EGFR. Transiently transfected MCF-7 cells expressing CA-Rac1 or CA-RhoA showed significantly enhanced GTPase activity upon EGF treatment (Figure 2E). However, the CA-GTPases influenced neither the dimerization of EGFR nor its endocytosis, even though they induced actin cytoskeleton polymerization (Figure 2F,G, Figure 3F and Figure S2C). On the other hand, DN-Rac1 or DN-RhoA specifically inhibited EGF-stimulated activation of these GTPases (Figure 2E). Contrary to the effect of CA-GTPases, DN-GTPases efficiently suppressed both EGFR dimerization and endocytosis, which were induced by EGF (Figure 2F,G and Figure S2C). We further confirmed the effect of GTPase signaling activity on EGFR dimerization and endocytosis. Consistent with the results in Figure 2E, transfection with DN-Rac1 specifically inhibited the interaction between WAVE2 and ARP2, while transfection with DN-RhoA inhibited only ROCK2–Ezrin interaction, but not WAVE2–ARP2 interaction (Figure 3A). However, EGF-induced dimerization and phosphorylation at Y1068 of EGFR in MDA-MB-231 cells were significantly reduced, even by a single knockdown of ARP2 or Ezrin (Figure 3B). In ARP2 knockdown cells, EGF-induced EGFR endocytosis as well as actin filament branching was significantly inhibited (Figure 3C,D). Knockdown of Ezrin disrupted actin filament polymerization and also suppressed endocytosis of EGFR. In addition, simultaneous knockdown of ARP2 and Ezrin also inhibited actin polymerization. Consistent with the results of DN-GTPases, transfection with CA-Rac1 stimulated the interaction between WAVE2 and ARP2, whereas transfection

with CA-RhoA stimulated the interaction of ROCK2 with Ezrin (Figure 3E). Constitutively active forms of Rac1, RhoA, or FAK induced actin filament polymerization in the presence or absence of EGF (Figure 3H). However, although they induced actin filament polymerization as efficiently as EGF, CA-Rac1, CA-RhoA, or CA-FAK failed to stimulate EGFR dimerization and endocytosis without EGF treatment (Figure 3F,G). In addition, EGF binding to EGFR is necessary to initiate the phosphorylation of EGFR, regardless of actin polymerization. These results suggest that GTPase-driven actin polymerization is necessary, but not sufficient for EGFR dimerization and endocytosis.

**Figure 2.** FAK functions as a critical mediator in EGF-induced activation of Rac1 and RhoA GTPases during EGFR signaling. (**A**,**C**) MCF-7 cells stimulated by binding of ligand to its receptor were analyzed for activation of small GTPases. Activated GTP-bound Rac1 or RhoA in the cell lysates were determined by immunoblotting with anti-Rac1 or anti-RhoA antibodies. β-actin was used as a loading control. (**B**,**D**) MDA-MB-231 cells were transfected with CA-FAK or FAK Y397F plasmids and incubated in the presence or absence of 25 ng/mL EGF at 37 ◦C, 5% CO<sup>2</sup> for 15 min. The interactions between the pairs of molecules indicated were assessed by in situ PLA. \*\*\* *p* < 0.001. (**E**) Activation of small GTPases in MCF-7 cells was determined by immunoblotting. (**F**) EGFR dimerization in MDA-MB-231 cells was assessed by in situ PLA and the experiments were duplicated. (**G**) EGFR endocytosis in MCF-7 cells was determined by IFA as described above. Original magnification of representative images, 600×. Scale bars = 10 µm.

**Figure 3.** *Cont.*

**Figure 3.** Modulation of actin polymerization by Rac1/RhoA GTPases is essential for EGF-induced dimerization and endocytosis of EGFR. (**A**,**E**) The changes in the activation of Rac1/RhoA-mediated signaling were observed in MCF-7 cells. The interactions between the pairs of molecules indicated were assessed by in situ PLA. \*\*\* *p* < 0.001. (**B**,**F**) To determine EGFR dimerization, MDA-MB-231 cells were subjected to BS<sup>3</sup> chemical-mediated crosslinking, as described above and in the Materials and Methods. Cell extracts were assessed via Western blotting to determine the dimerization and phosphorylation levels of EGFR and the expression levels of indicated proteins. β-actin was used as a loading control. EGFR endocytosis (**C**,**G**) and actin cytoskeleton organization (**D**,**H**) in MCF-7 cells transfected with siRNAs specific for ARP2 and Ezrin or plasmids encoding CA-GTPases or CA-FAK. Original magnification of representative images, 600×. Scale bars = 10 µm.

#### *2.3. CD99 Activation Attenuates EGF-Induced Dimerization and Activation of EGFR via the PKA*/*SHP2*/*HRAS*/*PTPN12*/*FAK Signaling Pathway*

To determine whether CD99 activation can disrupt EGF-induced dimerization and activation of EGFR, two breast cancer cell lines, MDA-MB-231 and MCF-7, were treated with CD99 agonist ligands, CD99CRIII3 or CD99-Fc. We previously demonstrated that CD99CRIII3, a CD99-derived peptide, can function as a CD99 agonist as efficiently as CD99 protein derivatives or anti-CD99 agonist monoclonal antibody [24]. Western blotting and in situ PLA showed that those two CD99 agonists significantly inhibited EGFR dimerization and phosphorylation at Y1068, which were induced by EGF (Figure 4A and Figure S4A). Moreover, neither CD99-Fc nor CD99CRIII3 had any effect on the EGF-induced dimerization and phosphorylation of EGFR at Y1068 in CD99-knockdown cells. EGFR dimerization and phosphorylation at Y1068 were significantly reduced by CD99CRIII3 in a dose-dependent manner (Figure 4B and Figure S4B). CD99CRIII3 inhibited EGF-induced phosphorylation of FAK at Y397. Consistent with this, CD99-Fc and CD99CRIII3 coordinately inhibited EGF-induced interactions of FAK with Rac1 and RhoA in MCF-cells, but not in CD99-knockdown MCF-7 cells (Figure S4C,D). In addition, both molecules suppressed the EGF-induced actin organization in a dose-dependent manner (Figure S4E). These results demonstrate that CD99 agonists inhibited EGF-induced dimerization and activation of EGFR by disrupting FAK-mediated actin organization.

**Figure 4.** CD99 agonistic ligands inhibit EGFR dimerization and endocytosis via the PKA/SHP2/HRAS/ERK/PTPN12 signaling pathway. (**A**,**B**) For dimerization assay, wild type or CD99 shRNA stable-expressing MDA-MB-231 cells were treated with CD99CRIII3 or CD99-Fc in the presence of EGF (25 ng/mL) for 1 h on ice. Cells were subjected to BS<sup>3</sup> chemical-mediated crosslinking. Cell extracts were subjected to SDS-PAGE to assess EGFR dimerization and phosphorylation levels of EGFR at Y1068 and FAK at Y397 and the expression levels of indicated proteins. β-actin was used as a loading control. (**C**) In MCF-7 cells, dimerization of EGFR was assessed by in situ PLA. \*\* *p* < 0.01; \*\*\* *p* < 0.001; \*\*\*\* *p* < 0.0001. (**D**) Whole cell lysates extracted from MDA-MB-231 cells treated with EGF with or without CD99CRIII3 were subjected to SDS-PAGE to examine the phosphorylation levels of EGFR and expression levels of each target protein. (**E**) EGFR endocytosis in MCF-7 cells treated with EGF or CD99CRIII3 either alone or combined. (**C**,**E**) Original magnification of representative images, 600×. Scale bars = 10 µm.

Since our previous results showed that CD99CRIII3 dephosphorylated FAK at Y397 through the PKA/SHP2/HRAS/PTPN12 signaling pathway [24], we next determined whether CD99CRIII3 regulates EGF-induced dimerization and activation of EGFR via the PKA/SHP2/HRAS/PTPN12 signaling pathway. Knockdown of PKA, SHP2, HRAS, or PTPN12 restored EGFR dimerization, which had been inhibited by treatment with CD99CRIII3 (Figure 4C). CD99CRIII3-mediated dephosphorylation of EGFR at Y1068 was also inhibited by knockdown of each of the intracellular signaling molecules (Figure 4D). In particular, knockdown of PTPN12 abrogated the inhibitory effect of CD99CRIII3 on EGFR endocytosis induced by EGF (Figure 4E). Consistent with our previous results, CD99CRIII3 efficiently inhibited EGF-induced actin rearrangement and EGFR dimerization by dephosphorylating FAK at Y397 via the PKA/SHP2/HRAS/PTPN12 signaling pathway.

To further validate the function of CD99CRIII3 in regulating FAK activity, the cells were treated with a selective FAK inhibitor 14 or transfected with CA-FAK. Inhibition of FAK activity by FAK inhibitor 14 attenuated EGF-induced EGFR dimerization, phosphorylation, and endocytosis by a similar degree to that attenuated by CD99CRIII3 treatment (Figure 5A,B). In contrast, CA-FAK partially recovered the functional activities of EGFR, which had been suppressed by CD99CRIII3, suggesting that persistent activation of FAK partially resists the inhibitory effects of CD99CRIII3 on EGFR dimerization, phosphorylation, and endocytosis. However, CA-FAK alone did not have any effect on dimerization and endocytosis of EGFR. EGFR regulates various cellular signals related to cell growth, proliferation, differentiation, and tumorigenesis [36–38]. As expected, EGF-induced activation of EGFR significantly increased proliferation of MCF-7 cells, whereas CD99CRIII3- or FAK inhibitor 14-induced inhibition of EGFR resulted in a reduced proliferation rate (Figure 5C). The cell proliferation rate, which was suppressed by CD99CRIII3 treatment, was completely restored in both types of cells when transfected with CA-FAK. These results demonstrate that CD99 agonist ligands suppress EGF-induced activation of EGFR through the PKA/SHP2/HRAS/ERK/PTPN12/FAK signaling pathway.

**Figure 5.** FAK plays an important role in breast cancer cell proliferation induced by EGFR activation. (**A**) MDA-MB-231 cells transfected with CA-FAK plasmid were subjected to BS<sup>3</sup> chemical-mediated crosslinking as described above. Cell extracts were assessed by Western blot analysis to determine the dimerization and phosphorylation levels of EGFR or the expression and phosphorylation levels of FAK. β-actin was used as a loading control. (**B**) EGFR endocytosis in MCF-7 cells was assessed by IFA as described above. Original magnification of representative images, 600×. Scale bars = 10 µm. (**C**) MCF-7 cells were stained with crystal violet. Images were captured using a Nikon Eclipse TE2000-U and the representative images are shown. Lines indicate statistical comparisons, and significant differences between treatments are shown by asterisks as follows: \*\*\* *p* < 0.001.

#### *2.4. CD99CRIII3-Activated PTPN12 Regulates the Activation of EGFR Signaling through the PTPN12*/*FAK*/ *Rho*/*Rac Axis*

As described above, PTPN12 acts as a negative regulator of multiple RTKs implicated in tumor progression [8,13,15,39]. We hypothesized that PTPN12 may restrain the activation of several cytoplasmic adaptor and kinase proteins that are recruited to EGFR following ligand binding, resulting in attenuation of the activated intracellular signals. Here, using in situ PLA assay, we evaluated the effect of CD99CRIII3-induced PTPN12 on EGFR signaling cascade. Consistent with our hypothesis, EGF treatment facilitated the interactions of EGFR with c-Src, Shc1, Grb2, Gab1, and FAK (Figure 6A). In particular, EGFR showed strongest interaction with c-Src and Shc1 after 5 min of treatment with EGF, whereas the highest degree of interaction between Grb2, Gab1, FAK, and EGFR was observed after 15 min of treatment. Surprisingly, CD99CRIII3 completely attenuated all interactions induced by EGF binding, whereas knockdown of PTPN12 caused a restoration of EGF-induced interactions between EGFR and other intracellular proteins. Consistent with these findings, co-immunoprecipitation revealed that EGF-induced interactions of EGFR with the intracellular molecules were attenuated by co-treating with CD99CRIII3. However, CD99CRIII3 lost its inhibitory effect on EGF-induced EGFR activation by knockdown of PTPN12 (Figure 6B and Figure S5A). To further characterize the kinetics of EGFR-PTPN12 interaction, we evaluated the time-dependent pattern of EGFR-PTPN12 interaction following stimulation with EGF alone or combined stimulation with EGF and CD99CRIII3. PTPN12 was found to co-precipitate with EGFR after treatment with EGF. After 10 min of treatment it showed the highest degree of interaction with EGFR (Figure 6C and Figure S5B). In contrast, the interaction between both molecules in cells treated with EGF plus CD99CRIII3 occurred much earlier than that in cells treated with EGF only and was continued until 10 min after treatment. These results show that PTPN12 activated by CD99CRIII3 plays a critical role in the disruption of the intracellular adapter/kinase complex involved in the EGFR signaling cascade.

It is certain that PTPN12 is involved in regulating cellular motility and morphology, since the phosphatase acts as a central regulator of actin cytoskeleton reorganization [8,9,40]. We performed co-immunoprecipitation to further verify the effects of CD99CRIII3-activated PTPN12 on Rac1/RhoA GTPase signaling pathways. CD99CRIII3 strongly inhibited WAVE2–ARP2 and ROCK2–Ezrin interactions, which were stimulated by EGF treatment in MDA-MB-231 cells (Figure 6D). Contrarily, knockdown of PTPN12 was able to neutralize the effects of CD99CRIII3 and maintain EGF-induced interactions between WAVE2 and ARP2 as well as ROCK2 and Ezrin. Consistent with these observations, while CD99CRIII3 suppressed the EGF-induced activation of Rac1 and RhoA GTPases, its inhibitory effect was neutralized by siRNA-mediated knockdown of PTPN12 in MCF-7 cells (Figure 6E). In addition, transfection with plasmids encoding CA-Rac1, wt-WAVE2, or wt-ARP2 reinstated the Rac1-mediated interaction of WAVE2 with ARP2, which had been inhibited by CD99CRIII3. Similar results were obtained in RhoA-mediated signaling cascade by expression of CA-RhoA, wt-ROCK2, or Ezrin. These results were demonstrated by in situ PLA and immunocytochemical assay for monitoring the localization of each of the Rac1/RhoA GTPase signaling-related proteins and the interactions between them (Figure S6A,B). CD99CRIII3 inhibited colocalization and physical proximity of WAVE2 and ARP2, ROCK2 and Ezrin at the cell membrane region, which were induced by treatment with EGF. Consistent with the above results, knockdown of PTPN12 abrogated the inhibitory effect of CD99CRIII3 on the interactions of actin polymerization-regulating proteins. Moreover, transfection with CA-Rac1 or overexpression of either of WAVE2 or ARP2 maintained the interaction between WAVE2 and ARP2. We also identified similar patterns of results showing RhoA-dependent localization of ROCK2 and Ezrin. Taken together, we found that CD99CRIII3 inhibits EGF-induced dimerization and endocytosis of EGFR, as well as actin polymerization in a PTPN12-dependent manner at a level equivalent to that exhibited by cytochalasin D treatment (Figure 6F–H and Figure S6C). These observations suggest that PTPN12 functions as a key regulator in CD99CRIII3-induced inhibition of EGFR dimerization and endocytosis via suppression of actin polymerization.

**Figure 6.** CD99CRIII3 activates PTPN12 to facilitate inhibition of EGFR signaling. (**A**) In situ PLA performed to assess the interactions between the pairs of molecules indicated in MCF-7 cells. \* *p* < 0.05; \*\* *p* < 0.01; \*\*\* *p* < 0.001; \*\*\*\* *p* < 0.0001. (**B**) MDA-MB-231 cells were transfected with PTPN12 siRNA, followed by treatment with EGF (25 ng/mL) with or without CD99CRIII3 (40 µM) for 15 min. (**C**) MDA-MB-231 cells were treated with EGF and/or CD99CRIII3 in a time-dependent manner. (**D**) MDA-MB-231 cells were transiently transfected with PTPN12 siRNA or expression plasmids encoding CA-GTPases or WAVE2, ARP2, ROCK2, and Ezrin. Cell lysates were immunoprecipitated with the antibodies indicated. The immunoprecipitates were analyzed by Western blot with the antibodies indicated. (**E**) The cells stimulated by binding of ligand to its receptor were assayed for activation of small GTPases. β-actin was used as a loading control. (**F**) For dimerization assay, MDA-MB-231 cells were subjected to BS<sup>3</sup> chemical-mediated crosslinking as described above. Cell extracts were assessed by Western blot analysis to determine the dimerization and phosphorylation levels of EGFR. β-actin was used as a loading control. Actin cytoskeleton organization in MDA-MB-231 cells (**G**) and EGFR endocytosis in MCF-7 cells (**H**) were determined by IFA as described above. Original magnification of representative images, 600×. Scale bars = 10 µm.

#### *2.5. CD99CRIII3 Dose-Dependently Inhibited TNBC Progression In Vivo through PTPN12-Mediated Suppression of Breast Cancer Cell Proliferation*

To determine the anti-tumorigenic effect of CD99CRIII3 andtheimportance of PTPN12in suppressing tumor progression, we generated a stable MDA-MB-231 cell line (shPTPN12-MDA-MB-231) with 75% reduction in PTPN12 protein using the shRNA system (Figure S7A). shPTPN12-MDA-MB-231 cells exhibited no changes in the expression level of EGFR and CD99. In addition, the dimerization and phosphorylation on Y1068 of EGFR and Y397 of FAK, which had been suppressed by CD99CRIII3 in wt-MDA-MB-231 cells, were not affected in shPTPN12-MDA-MB-231 cells (Figure 7A and Figure S7B). CD99CRIII3 inhibited the proliferation of wt-MDA-MB-231 cells, which was increased by treatment with EGF. Contrarily, the cell proliferation rate, which was enhanced by EGF, was not suppressed by CD99CRIII3 in the PTPN12 knockdown cell line (Figure 7B and Figure S7C). We confirmed the physiological characteristics of shPTPN12-MDA-MB-231 using in situ PLA (Figure S7D). EGF treatment induced the interactions of EGFR with Grb2, c-Src, Shc1, FAK, Gab1, and itself. However, the shPTPN12-MDA-MB-231 cells did not inhibit these interactions by treatment with CD99CRIII3. These results suggest that CD99CRIII3 inhibits EGF-induced EGFR activation via PTPN12-mediated signaling.

**Figure 7.** *Cont.*

**Figure 7.** Evaluation of the in vivo efficacy of CD99CRIII3 in the xenograft model of human breast cancer. (**A**) MDA-MB-231 cells were transiently transfected with PTPN12 shRNA plasmid. Cell extracts were assessed by Western blot analysis to determine the phosphorylation levels of EGFR at Y1068 or FAK at Y397. β-actin was used as a loading control. The uncropped Western blots have been shown in Figure S8. (**B**) Comparison of cell growth rate between wt-MDA-MB-231 and shPTPN12-MDA-MB-231 cells. (**C**) Images of tumor xenografts. Scale bar represents 10 mm. (**D**,**E**) Graph shows the mean difference in tumor volume and weight between wild-type and PTPN12 knockdown cells. Lines indicate statistical comparisons, and significant differences between treatments are shown by asterisks as follows: \* *p* < 0.05; \*\* *p* < 0.01. (**F**) Hematoxylin and Eosin (H&E) staining of tumor xenografts. Scale bar represents 10 µm for 400× magnification. The expression levels of EGFR, PTPN12, and Ki67 by IHC in xenografts. (**G**) In situ PLA analysis was performed to determine the dimerization pattern of EGFR within xenograft tumor tissues. Confocal images were taken from four tumors of each treatment group and displayed in compressed z-stack form. Numerical values are the mean intensities (±SD) of red spots in three randomly selected fields per tumor section. Significant differences between treatments are shown by asterisks as follows: \*\* *p* < 0.01; \*\*\* *p* < 0.001. Scale bar, 10 µm (600×). wt, wild type; sh, short hairpin. (**H**) Schematic model for the inhibitory effect of CD99 agonistic ligand on EGF-induced activation of EGFR.

Finally, we carried out a tumor xenograft assay in the BALB/c nude mouse model with wtand shPTPN12-MDA-MB-231 human breast carcinoma cells. The daily injection of CD99CRIII3 led to significantly decreased tumor volume and weight in the wt-MDA-MB-231-inoculated mice, compared with PBS-treated mice (Figure 7C–E). However, there were no differences in tumor size and weight in the CD99CRIII3-treated mice injected with shPTPN12-MDA-MB-231 cells, indicating that CD99CRIII3 exerts its anti-tumorigenic activity via the CD99–PTPN12 axis. After 15 days of CD99CRIII3 administration, tumors were collected and sliced into small pieces. Serial sections of the tumor specimens were stained with hematoxylin and eosin (H&E) and antibodies to measure the expression of Ki67, EGFR, and PTPN12 (Figure 7F). Histological analysis of the specimens revealed that CD99CRIII3 led to reduced Ki-67 expression in a group of wt-MDA-MB-231-inoculated mice, but not in the shPTPN12-MDA-MB-231-inoculated mice group. The majority of tumor mass was identified to be positive for EGFR and Ki-67. The number of Ki-67-positive cells was reduced in the CD99CRIII3-treated mice, showing correlation with the reduced tumor size (Figure S7E). In addition, CD99CRIII3 dose-dependently inhibited EGFR dimerization only in wt-MDA-MB-231-originated tumor tissues (Figure 7G). These results indicate that CD99 agonist ligand could significantly suppress the proliferation of MDA-MB-231 human breast cancer cells through PTPN12-mediated inactivation of EGFR.

#### **3. Discussion**

In this study, we clearly showed that actin polymerization plays an important role in EGFR receptor dimerization and activation, which was inhibited by the CD99/PTPN12/FAK/Rho/Rac axis. Our novel findings provide an insight into the role of CD99 in the inactivation of several oncogenic tyrosine kinases including EGFR [13,14].

As the actin cytoskeleton plays an important role in controlling the movement of intracellular organelles as well as cell surface receptors [41–43], we elucidated the importance of cytoskeleton reorganization in ligand-induced dimerization and subsequent activation of EGFR. EGFR directly associates with the actin filament via its C-terminal actin-binding domain [44,45]. We found that impairment of actin cytoskeleton organization by cytochalasin D inhibits the phosphorylation, dimerization, and internalization of EGFR, consistent with previous results showing that disruption of actin polymerization inhibits ligand-induced EGFR dimerization, activation, and downstream signaling [26,28]. Besides the role for EGFR dimerization, the blocking of F-actin polymerization inhibits the CXCL12-mediated dimerization of CXCR4 [46]. The actin cytoskeleton intimately interacts with plasma membrane integral proteins and regulates intricate membrane events, such as the formation of focal adhesions as well as the internalization, recycling, compartmentalization, dynamics, clustering, and diffusion of membrane receptor proteins [27,41,47]. The assembly and disassembly of cytoskeletal actin filaments (F-actin) are regulated by c-Src and FAK [31,48,49]. Functional impairment of c-Src or FAK inhibited actin polymerization, leading to the suppression of dimerization and internalization of EGFR. In contrast, transduction with CA-FAK facilitated the FAK-mediated activation of Rac1/RhoA signaling pathways and actin polymerization. Importantly, it looks like dimerization of EGFR is not sufficient to activate its kinase activity. Although CA-FAK could induce the formation of actin filaments without EGF ligand, it failed to proceed to the next step, EGFR dimerization and endocytosis. CA-Rac1 and CA-RhoA also showed similar results. The binding of growth factor ligands to EGFR may be critical for conformational changes of the receptor or its dynamics, leading to dimerization or oligomerization of EGFR and subsequent activation of the downstream signaling pathway. In other words, EGFR activation may require both ligand binding to EGFR and subsequent dimerization. Therefore, disruption of ligand-induced EGFR dimerization as well as ligand binding would be a promising therapeutic strategy for the treatment of breast cancer patients with aberrant expression or activation of EGFR.

Breast cancer can be classified into several subtypes according to the expression level of various surface marker proteins, including estrogen receptor (ER), progesterone (PR), and HER2. On the other hand, EGFR is expressed in a wide range of breast cancer cell lines at different levels. EGFR has long been in spotlight as a reasonable target molecule for developing antitumor strategies, since its aberrant activation by increased expression of a constitutively activated truncated variant EGFRvIII or itself is implicated in the development and progression of a broad range of solid cancer diseases including breast cancer [50]. We adopted two breast cancer cell lines, MDA-MB-231 and MCF-7. MDA-MB-231 cells lack the expression of ER, PR, and HER2, while they show high expression of multiple RTKs. On the other hand, adenocarcinoma MCF-7 cells express ER, PR, and glucocorticoid receptors. We found that MDA-MB-231 cells express EGFR at high levels, whereas MCF-7 cells express very low levels of this receptor. The expression levels of CD99, in contrast, are similar in both cell lines. Despite different levels of EGFR expression, these two breast carcinomas were similarly affected by EGFR ligands and CD99 agonists. These results suggest that EGFR might play a dominant role in cellular and physiological systems of breast cancer cells, so that it can be a valuable target for the development of a broadly applicable anti-cancer drug.

PTPN12 is a tumor suppressor which regulates cellular transformation from normal to malignant cells via the inhibition of multiple oncogenic tyrosine kinases [13,14]. Consistent with this, our data showed that stable knockdown of PTPN12 increased tumor progression in vivo. Although several studies imply the functional significance of PTPN12 in controlling tumor progression, the activator of PTPN12 has not been identified yet. One novel finding of this study is that the CD99–PTPN12 axis participates in the regulation of ligand-induced activation of EGFR by suppressing the reorganization of the actin cytoskeleton. This observation is consistent with a previous study, which demonstrated that PTPN12 controls EphA3 activation by regulating actin cytoskeletal organization during Eph clustering [8]. Furthermore, we previously reported the molecular mechanism by which CD99 induces β1 integrin inactivation via PTPN12 activation [24]. Likewise, CD99CRIII3 showed significant inhibitory effects on EGF-induced EGFR dimerization and internalization via activation of PTPN12. When EGFR is activated with its ligand, PTPN12 is recruited to the activated EGFR to return to an inactive state within 15 min [39]. On the other hand, CD99CRIII3 induced very early recruitment of PTPN12 to the EGF-induced EGFR signaling complex, which led to the inhibition of EGFR dimerization and activation. The remarkable inhibitory effect of CD99CRIII3 on EGFR activation was suppressed when CD99 or PTPN12 expression was downregulated, indicating that CD99 activation by CD99CRIII3 stimulated PTPN12 to inhibit the early stage of the EGFR signaling pathway.

PTPN12 exhibited relatively low expression levels in triple-negative breast cancer (TNBC) cells [13]. However, ectopic restoration of PTPN12 in TNBC resulted in the suppression of anchorage-independent proliferation and metastatic ability. Consistent with this finding, we observed that CD99CRIII3 significantly suppressed the EGF-induced proliferation of MDA-MB-231 and MCF-7 breast cancer cells, which was not observed in PTPN12-knockdown cells. Furthermore, shPTPN12-MDA-MB-231 cells allowed us to examine whether CD99CRIII3 affects in vivo tumorigenesis via the PTPN12-dependent negative feedback loop. CD99CRIII3 dose-dependently suppressed the growth of MDA-MB-231 human breast cancer cells implanted in nude mice, while it failed to suppress the growth of shPTPN12-MDA-MB-231 cells implanted in nude mice, suggesting that PTPN12 serves as a key executor of the CD99 signaling pathway. Consistently, CD99CRIII3 inhibited EGFR dimerization in wt-MDA-MB-231-originated tumor tissues, but not in shPTPN12-MDA-MB-231-originated tumor tissues. Here we pay attention to recent reports showing that CD99 activates p53 tumor suppressor by inducing degradation of Mdm2, an E3 ubiquitin ligase, resulting in the death of Ewing sarcoma (EWS) [20,51]. Collectively, these results suggest that CD99 might play a key role in modulating the activities of intracellular tumor suppressors, PTPN12 and p53, whose interrelationship still remains elusive.

The growth of various tumors is promoted by tumorigenic growth factor receptors, such as FGFR, TGF-βR, IGF-1R, InsR, and PDGF [52–56]. Given that their kinase activity is induced via dimerization and activation according to ligand binding, and actin cytoskeleton is implicated in controlling receptor compartmentalization [27,57,58], it is important to determine whether CD99CRIII3 can regulate the activity of those RTKs. Additionally, protein tyrosine phosphatases (PTPs) act as inhibitors, regulating tumor-inducing activity [53]. Thus, our results suggest that PTPN12 activated by CD99CRIII3 may suppress the activity of other abnormal RTKs as well as EGFR and that their dimerization and activation processes are regulated by actin cytoskeleton-controlled clustering. However, the underlying mechanism by which CD99CRIII3 inhibits the dimerization and activation of other tumorigenic RTKs needs further investigation in a broad range of tumors.

#### **4. Materials and Methods**

#### *4.1. Reagents and Antibodies*

All cultureware and reagents were purchased from Invitrogen (Carlsbad, CA, USA). Immun-Blot polyvinylidene fluoride (PVDF) membranes for protein blotting were purchased from Bio-Rad Laboratories (Hercules, CA, USA). The WEST-ZOL plus Western blot detection kit was obtained from iNtRON Biotechnology, Inc. (Seongnam, Korea). Lipofectamine LTX/PLUS and RNAiMAX reagents were purchased from Invitrogen (Life Technologies, Grand Island, NY, USA). Protein A/G agarose beads were purchased from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA, USA). Cytochalasin D, FAK inhibitor 14, recombinant human epidermal growth factor (EGF), FITC (Fluorescein isothiocyanate)-conjugated Phalloidin (1:200 for IFA), and purified mouse IgG (1:200 for IFA) were purchased from Sigma-Aldrich Co. (St. Louis, MO, USA). Rhodamine-conjugated anti-mouse IgG antibody, rhodamine-conjugated anti-rabbit IgG antibody, FITC-conjugated anti-mouse IgG antibody, FITC-conjugated anti-rabbit IgG antibody (1:200 for IFA), and Horseradish peroxidase (HRP)-conjugated anti-mouse IgG antibody (1:10,000 for Western blot) were purchased from DiNonA (Seoul, Korea). HRP-conjugated goat anti-rabbit IgG antibody (1:10,000 for Western blot) was

purchased from Chemicon (Temecula, CA, USA). Mouse monoclonal anti-human epidermal growth factor receptor (EGFR) antibody (1:150 for IFA), rabbit polyclonal anti-human EGFR antibody (1:150 for IFA), and rabbit polyclonal or mouse monoclonal anti-EEA1 antibody (1:150 for IFA) were purchased from Abcam (Cambridge, UK). Antibodies against HRAS, SHP2, PKA, Ezrin, WAVE2, Arp2, ROCK2, Grb2, Shc1, Rac1, RhoA, EGFR, β-actin, and HRP-conjugated donkey anti-goat IgG antibody (1:10,000 for Western blot) were obtained from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA, USA). Antibodies against phospho-FAK (Tyr397), FAK, phospho-EGFR (Tyr1068), PTPN12, c-Src, HER2, and Gab1 were purchased from Cell Signaling Technology (Danvers, MA, USA).

#### *4.2. Cell Culture*

Human breast adenocarcinoma cell line MCF-7 was obtained from American Type Culture Collection (ATCC). Triple-negative breast carcinoma cell line MDA-MB-231 was kindly provided by Dr. Hyung Geun Song (DiNonA Inc., Seoul, Korea). MCF-7 cells were cultured in Dulbecco's modified Eagle's medium (DMEM) containing 10% (*v*/*v*) fetal bovine serum (FBS), 100 unit/mL penicillin, and 100 µg/mL streptomycin (Gibco-BRL, Grand Island, NY, USA). MDA-MB-231 cells were cultured in Roswell Park memorial Institute (RPMI) 1640 (10% FBS, 100 unit/mL penicillin, 100 µg/mL streptomycin, and 25 mM HEPES). All cells were maintained at 37 ◦C in a humidified 5% CO<sup>2</sup> incubator.

#### *4.3. Synthesis of Polypeptides*

CD99 agonist polypeptide CD99CRIII3 was synthesized using an automatic peptide synthesizer (PeptrEx-R48, Peptron, Daejeon, Korea) according to the 9-fluorenylmethoxycarbonyl (Fmoc) solid-phase method. The synthesized polypeptides were purified and analyzed using reverse-phase high-performance liquid chromatography (Prominence LC-20AB, Shimadzu, Japan) equipped with a C18 analytical RP column (Capcell Pak column, Shiseido Co., Ltd., Japan). The mass was analyzed using a mass spectrometer (HP1100 Series LC/MSD, Hewlett-Packard, Roseville, CA, USA). The analytical results are described in Figure S3.

#### *4.4. Plasmids and RNA Interference*

The coding sequences of human WAVE2 and Arp2 were obtained by PCR with respective pairs of primers. Sense primer 5′ -GGGGTACCGCCACCATGCCGTTAGTAACGAGGAAC-3′ and antisense primer 5′ -GCTCTAGAGAGTTAATCGGACCAGTCGTC-3′ for the cDNA of WAVE2, sense primer 5′ -GGGGTACCGCCACCATGGACAGCCAGGGCAGG-3′ and antisense primer 5 ′ -GCTCTAGATTATCGAACAGTCACACCAAG-3′ for the cDNA of Arp2. Full length human WAVE2 and Arp2 cDNAs were subcloned into KpnI and XbaI sites of the pcDNA3 vector. The sequences of the constructs were confirmed by DNA sequencing. The expression vectors pEXV/constitutively active Rac1, pEXV/dominant negative Rac1, pEXV/constitutively active RhoA, pEXV/dominant negative RhoA, and pcDNA3.1/dominant negative c-Src were kindly donated by Dr. Hansoo Lee. The expression vector kinase-dead, non-phosphorylatable dominant negative FAK Y397F (pcDNA3/FAK Y397F) was kindly provided by Dr. Soo-Chul Park (Sookmyung Women's University, Seoul, S. Korea). Constitutively active FAK plasmid (pCDM8/CD2-FAK) was kindly provided by Dr. Andrey V. Cybulsky (McGill University, Montreal, QC, Canada). The pCS2/ROCK2 vector was kindly provided by Dr. Anming Meng (Tsinghua University, Beijing, China). The pCB6/rsr-G-Tag/Ezrin vector was kindly provided by Dr. Janet Allopenna (Stony Brook Medicine, NY, USA). For gene knockdown experiments, the small interfering RNAs (siRNAs) against FAK, Shc1, c-Src, Arp2, Ezrin, PKA-α, SHP2, HRAS, PTPN12, and shRNA targeting PTPN12 were purchased from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA, USA). Plasmids or siRNA duplexes were transfected into cells using Lipofectamine LTX/PLUS or RNAiMAX (Invitrogen Life Technologies, Grand Island, NY, USA). The PTPN12 knockdown MDA-MB-231 resistant clone was established by selecting with 0.4 mg/mL puromycin. After transfection, knockdown of each of the target molecules or expression of dominant negative or constitutively active DNA was confirmed by Western blotting.

#### *4.5. In Situ Proximity Ligation Assay (PLA)*

The in situ PLA analysis was performed using Duolink® in situ reagents (O-LINK® Bioscience, Uppsala, Sweden) according to the manufacturer's instructions. Cells were transfected with plasmids or siRNAs and then seeded on glass coverslips in 24-well cell culture plates (1 × 10<sup>5</sup> cells/well). After 24 h growth under standard conditions, cells were treated with EGF (25 ng/mL) in the presence or absence of peptides and each reagent for 15 min in a CO<sup>2</sup> incubator at 37 ◦C, and then washed twice with 1X PBS. Cells were fixed with 2% formaldehyde in PBS for 10 min at room temperature (RT), and subsequently washed twice with 1X PBS, permeabilized with 0.1% Triton X-100 in PBS, and then washed twice with wash buffer A. Cells were incubated with blocking solution at 37 ◦C for 30 min, and then washed twice with wash buffer A. Cells were stained with specific antibodies (1:100 for in situ PLA) as indicated. Protein–protein interactions were analyzed using a confocal laser scanning microscope Olympus FluoView FV1000 (Olympus, Tokyo, Japan). PLA signals in cell populations (*n* = 4) were quantified by NIS-Elements analysis, and four or two independent experiments were performed. The average number of rolling-circle products (RCPs) per cell ± standard error is shown.

#### *4.6. Dimerization Assay*

BS<sup>3</sup> [bis(sulfosuccinimidyl) suberate] was obtained from Thermo Scientific (Waltham, MA, USA) and used according to the manufacturer's instruction and reference [59]. Cells were serum-starved in DMEM or RPMI containing 0.1% bovine serum albumin (BSA) and incubated in serum-free medium supplemented with human EGF in the presence or absence of CD99CRIII3 for 1 h at 4 ◦C (incubation on ice during ligand stimulation allows for ligand-induced receptor dimerization but inhibits receptor endocytosis). Cells were washed three times with ice-cold 1X Ca2+-, Mg2+-free PBS, then incubated with BS<sup>3</sup> (2 mM) at 4 ◦C for 30 min and an additional 20 min at RT, followed by quenching with 1M of Tris (pH 7.5). Cells were lysed with 1% NP40 buffer containing 10 mM of β-mercaptoethanol. Cell lysates were subjected to SDS-PAGE and analyzed by Western blotting.

#### *4.7. Active GTPase Detection*

Active GTPase assay was performed using an active Rac1 or RhoA detection kit (Cell Signaling Technology, Danvers, MA, USA) according to the manufacturer's instructions. Cells were rinsed with 1× ice-cold PBS, then lysed with 1× lysis/binding/wash buffer plus 1 mM PMSF. Whole cell lysate was harvested and incubated on ice for 5 min, and then centrifuged at 16,000× *g* at 4 ◦C for 15 min. The clear supernatant was transferred to a new tube and subjected to active GTPase assay. Glutathione resin slurry (50%, 100 µL) was added to a spin cup with a collection tube and the tube was centrifuged at 6000× *g* for 1 min. After washing the resin with 400 µL of 1× lysis/binding/wash buffer, 20 µg of GST-PAK1-p21 binding domain (PBD) (for GTP-bound Rac1) or GST-Rhotekin-Rho binding domain (RBD) (for GTP-bound RhoA) was added to the spin cup containing glutathione resin. The cell lysate was immediately transferred to the spin cup and vortexed. The reaction mixture was incubated at 4 ◦C for 1 h with gentle rocking. The active GTPase-bound GST resin was washed with 1X lysis/binding/wash buffer containing 1 mM PMSF. GTP-bound Rac1 or RhoA was eluted with 2X SDS reducing buffer containing 200 mM 1,4-dithiothreitol (DTT), followed by Western blotting with mouse anti-Rac1 mAb or rabbit anti-RhoA pAb.

#### *4.8. Western Blot Analysis and Immunoprecipitation*

Western blotting and immunoprecipitation were carried out as described previously [24]. Serum-starved breast carcinoma cells were treated with the appropriate reagents for 15 min at 37 ◦C, 5% CO2. Cells were harvested and lysed with 1% NP40 lysis buffer (1% Nonidet P40, 150 mM NaCl, 50 mM Tris-HCl (pH 8.0), 5 mM EDTA) containing 10 mM phenylmethylsulfonyl fluoride, 1 µg/mL pepstatin A, 10 µg/mL leupeptin, 1 µg/mL aprotinin, and 1 mM sodium orthovanadate. Cell extracts

were subjected to SDS-PAGE and subsequently transferred to PVDF membranes. Immunoblotting was carried out with the indicated antibodies (1:1000 for Western blot) to detect the target proteins.

For immunoprecipitation, cells were treated for the indicated time with each set of reagents at 37 ◦C, 5% CO2. Cells were harvested and lysed with PRO-PREPTM protein extraction solution (iNtRON Biotechnology, Inc., Seongnam, Korea). After centrifugation at 13,000 rpm and 4 ◦C for 15 min, supernatants were collected, and then incubated overnight with the appropriate antibodies at 4 ◦C on a nutator. The immunoprecipitates were incubated with Protein A/G PLUS-Agarose (Santa Cruz Biotechnology, Inc., Santa Cruz, CA, USA) beads for 3 h at 4 ◦C on a rotator and washed with PRO-PREPTM solution. The precipitates were eluted with 1× sample buffer (50 mM Tris-HCl (pH 6.8), 100 mM DTT, 2% SDS, 0.1% bromophenol blue, and 10% glycerol). Western blot analysis was carried out with the indicated antibodies to detect the target proteins.

#### *4.9. Immunofluorescence Assay (IFA)*

The EGFR distribution, cytoskeletal organization, and localization of the related proteins were detected by immunofluorescence assay. To determine the changes in RTK distribution, cytoskeletal organization, and localization of the related proteins, cells were seeded on round-shaped glass coverslips as described above. The following day, cells were treated with the appropriate reagents, then washed with ice-cold PBS and fixed with 4% paraformaldehyde (PFA) in 1× PBS for 10 min at RT. Subsequently, cells were washed twice with PBS, permeabilized for 5 min with 0.1% Triton X-100 in PBS. After washing with PBS, the cells were incubated with appropriate primary and secondary antibodies to detect the target proteins. Alternatively, cells were stained with 0.2 µM of FITC-conjugated phalloidin to detect the fibrous actin filaments. The stained cells were mounted onto slides with an aqueous mounting medium. Fluorescence images were acquired using a confocal microscope Olympus FluoView FV1000 (Olympus, Tokyo, Japan).

#### *4.10. Proliferation Assay*

Cell proliferation was assessed by the Cell Counting Kit-8 (CCK-8) assay or crystal violet staining method. Wild-type or CA-FAK-transfected cells were seeded at a density of 5 × 10<sup>3</sup> cells per well in 96-well culture plates. After 24 h incubation, cells were treated with EGF (25 ng/mL) in the presence or absence of CD99CRIII3 (40 µM) or FAK inhibitor (25 µM) in 100 µL of serum-free medium (SFM) for 24 h, 48 h, and 72 h. At the indicated time points, cell proliferation was assessed using Cell Counting Kit-8 (CCK8, Dojindo Molecular Technologies, Inc., Rockville, MD, USA)-based assay or crystal violet staining method. CCK-8 colorimetric reactions were assessed by measuring the absorbance at 450 nm using a microplate reader (Versa Max, NY, USA). The images of crystal violet-stained cells were captured with a Nikon Eclipse TE2000-U (Nikon Instruments Inc., Melville, NY, USA).

#### *4.11. Tumor Xenograft and Immunohistochemistry (IHC)*

All animal experiments were performed in accordance with the Institutional Guidelines of the Animal Care and Use Committees (IACUC) of Kangwon National University. This research has been approved by IACUC of Kangwon National University on 26 October 2016 (KW-161020-1). MDA-MB-231 cells were cultured in complete Roswell Park Memorial Institute (RPMI) 1640 (10% FBS, 25 mM HEPES, 100 units/mL penicillin, and 100 µg/mL streptomycin) medium. Wild-type or PTPN12 shRNA-transfected MDA-MB-231 cells (5 × 10<sup>6</sup> cells in 50 µL SFM/mouse) were mixed 1:1 (*v*/*v*) with Matrigel and subcutaneously injected into the right flanks of 6-week-old female BALB/c nude mice. One week after tumor cell inoculation, mice were divided into three groups of five mice each when the tumor size exceeded 5 mm in diameter (10 mg/kg or 20 mg/kg of body weight of CD99CRIII3 and PBS control). CD99CRIII3 was intraperitoneally administered to mice every day for 14 days. The tumor size was measured using a caliper every other day for 15 days. After 21 days of tumor cell inoculation, mice were sacrificed and the tumor masses were removed and weighed. After measuring the volume and weight, four tumor masses from each treatment group were paraffinized, sectioned

with a microtome and stained with hematoxylin and eosin (H&E) following internal procedures. The effect of CD99CRIII3 on EGFR dimerization in tumor xenograft was assessed by in situ PLA assay according to the manufacturer's instructions. Tumor sections were stained with anti-human EGFR and anti-phospho-EGFR (Tyr1068) antibodies. The z-stacks were generated from images taken at 0.2–0.4 µm intervals. Z-stack images were collected from three randomly selected fields per tumor. EGFR–EGFR interaction was quantified and analyzed by mean red intensity of automated measurements. Six tumors, one of each treatment, were frozen using liquid nitrogen, subjected to cryosectioning and stained with primary antibodies specific for EGFR, PTPN12, and ki67, followed by incubation with respective fluorescent secondary antibodies. Fluorescence images were analyzed using a confocal laser scanning microscope Olympus FluoView FV1000 and H&E staining images were captured using an Olympus BX50 microscope (Olympus, Tokyo, Japan).

#### *4.12. Statistical Analysis*

Values are given as mean ± standard deviation (SD). Statistical significance was determined by the Student's *t*-test using the statistical analysis software GraphPad Prism (version 8.0; San Diego, CA, USA) and *p* < 0.05 was considered statistically significant. All experiments were conducted twice or more to minimize experimental error. The representative data are shown in the figures.

#### **5. Conclusions**

We demonstrated that CD99 activation regulates actin cytoskeleton dynamics through PTPN12/FAK/Rho/Rac axis, thereby suppressing EGFR activation and relevant tumor growth. We propose a schematic model illustrating the possible mechanism for the CD99 agonist ligand-induced suppression of EGFR activation (Figure 7H). CD99 acts as an upstream regulator of the PTPN12-mediated negative feedback loop for regulating ligand-induced dimerization or oligomerization of plasma membrane protein kinases, which are involved in tumor development and progression. Taken together, we propose that CD99 agonist ligands have potential as novel therapeutic drug candidates to suppress human breast carcinoma via inhibition of EGF-mediated EGFR signaling.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2072-6694/12/10/2895/s1, Figure S1: c-Src/FAK plays an important role in actin cytoskeleton reorganization, resulting in EGFR dimerization and internalization, Figure S2: Binding of ligand to its receptor is essential to induce Rac1/RhoA-mediated EGFR dimerization, Figure S3: HPLC/MS chromatogram of CD99-derived agonistic peptide (CD99CRIII3), Figure S4: Functional validation of the equivalence of CD99 agonist ligands, Figure S5: PTPN12 plays a critical role in the dissociation of the EGFR-associated signaling complex induced by CD99CRIII3, Figure S6: CD99CRIII3 suppresses EGF-induced Rac1/RhoA GTPase signaling cascades via PTPN12, Figure S7: The physiological characteristics of shPTPN12-MDA-MB-231 cells, Figure S8: Uncropped western blot figures.

**Author Contributions:** J.-H.H. and K.-J.L. proposed the concept, conceived the entire study, and wrote the paper; K.-J.L., Y.K., M.S.K., B.C., and H.-M.J. performed most of the experiments; H.L., D.J., D.K., K.-W.M., J.C., and J.I.Y. provided discussion and advice. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the Basic science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2018R1D1A3B07043170) and by the 2017 Research Grant from Kangwon National University (No. 520170543).

**Acknowledgments:** We highly appreciate Andrey V. Cybulsky (Department of Medicine, McGill University, Montreal, Canada) for providing pCDM8/CD2-FAK plasmid; Hyung Geun Song (DiNonA Inc., Seoul, S. Korea) for the triple-negative breast cancer MDA-MB-231 cell line.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Genomic Mapping of Splicing-Related Genes Identify Amplifications in** *LSM1***,** *CLNS1A,* **and** *ILF2* **in Luminal Breast Cancer**

**María del Mar Noblejas-López 1,2,† , Igor López-Cade 3,† , Jesús Fuentes-Antrás 4,5 , Gonzalo Fernández-Hinojal 4,5 , Ada Esteban-Sánchez 3 , Aránzazu Manzano 4,5 , José Ángel García-Sáenz 6 , Pedro Pérez-Segura 6 , Miguel De La Hoya 3 , Atanasio Pandiella 7 , Balázs Gy ˝orffy 8,9,10 , Vanesa García-Barberán 3, \* and Alberto Ocaña 1,2,4,5, \***


**Simple Summary:** The alternative splicing (AS) process is highly relevant, affecting most of the hallmarks of cancer, such as proliferation, angiogenesis, and metastasis. Our study evaluated alterations in 304 splicing-related genes and their prognosis value in breast cancer patients. Amplifications in *CLNS1A*, *LSM1*, and *ILF2* genes in luminal patients were significantly associated with poor outcome. Downregulation of these genes in luminal cell lines showed an antiproliferative effect. Pharmacological modulation of transcription and RNA regulation is key for the optimal development of therapeutic strategies against key proteins. Administration of a BET inhibitor and BET-PROTAC reduced the expression of these identified genes and displayed a significant antiproliferative effect on these cell models. In conclusion, we describe novel splicing genes amplified in luminal breast tumors that are associated with detrimental prognosis and can be modulated pharmacologically. It opens the door for further studies confirming the effect of these genes in patients treated with BET inhibitors.

**Abstract:** Alternative splicing is an essential biological process, which increases the diversity and complexity of the human transcriptome. In our study, 304 splicing pathway-related genes were evaluated in tumors from breast cancer patients (TCGA dataset). A high number of alterations were detected, including mutations and copy number alterations (CNAs), although mutations were less frequently present compared with CNAs. In the four molecular subtypes, 14 common splice genes showed high level amplification in >5% of patients. Certain genes were only amplified in specific breast cancer subtypes. Most altered genes in each molecular subtype clustered to a few chromosomal regions. In the Luminal subtype, amplifications of *LSM1*, *CLNS1A*, and *ILF2* showed a

**Citation:** Noblejas-López, M.d.M.; López-Cade, I.; Fuentes-Antrás, J.; Fernández-Hinojal, G.; Esteban-Sánchez, A.; Manzano, A.; García-Sáenz, J.Á.; Pérez-Segura, P.; La Hoya, M.D.; Pandiella, A.; et al. Genomic Mapping of Splicing-Related Genes Identify Amplifications in *LSM1*, *CLNS1A,* and *ILF2* in Luminal Breast Cancer. *Cancers* **2021**, *13*, 4118. https:// doi.org/10.3390/cancers13164118

Academic Editors: Ion Cristóbal and Marta Rodríguez

Received: 19 July 2021 Accepted: 13 August 2021 Published: 16 August 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

strong significant association with prognosis. An even more robust association with OS and RFS was observed when expression of these three genes was combined. Inhibition of *LSM1*, *CLNS1A*, and *ILF2*, using siRNA in MCF7 and T47D cells, showed a decrease in cell proliferation. The mRNA expression of these genes was reduced by treatment with BET inhibitors, a family of epigenetic modulators. We map the presence of splicing-related genes in breast cancer, describing three novel genes, *LSM1*, *CLNS1A*, and *ILF2*, that have an oncogenic role and can be modulated with BET inhibitors.

**Keywords:** splicing pathway; luminal breast cancer; BET inhibitors

#### **1. Introduction**

The RNA splicing process regulates gene expression in eukaryotic cells through a complex process in which introns are removed from precursor RNAs (pre-mRNAs) and consecutive exons are precisely joined together to produce mature mRNAs, with the final goal of maintaining mature transcripts to guarantee a successful translation process [1]. The alternative splicing process (AS) is the way in which exons or portions of exons or non-coding regions within a pre-mRNA transcript are differentially joined or skipped, resulting in multiple protein isoforms being encoded by a single gene [1]. Alternative splicing (AS) contributes to transcriptome (and proteome) diversity in development- and tissue-regulated pathways, as well as in response to multiple physiological signals [2]. Remarkably, large-scale proteomic studies suggest that many predicted alternative transcripts are not translated into proteins, so the exact contribution of AS to protein diversity is currently under dispute [3,4]. On top of that, some authors have suggested a role for AS in buffering mutational consequences [5], and mounting evidence indicates that AS coupled to nonsense-mediated decay is a major post-transcriptional regulator of gene expression [6,7]. Five major types of AS have been described: exon skipping, mutually exclusive exons, intron retention, and alternative 5′or 3′ splice site [8]. The AS process is carried out by the spliceosome and consists of four stages: assembly, activation, catalysis or splicing, and disassembly. In each specific stage of a splicing cycle, different spliceosome subcomplexes are involved (pre-B, B, Bact, B\*, C, C\*, P, and ILS complexes), which are composed of small nuclear ribonucleoproteins (snRNPs) and non-snRNPs splicing factors [9]. AS is a highly regulated process, with five snRNPs and over 300 non-snRNP proteins identified as recruited to the spliceosome at these specific stages [10].

Changes due to AS can affect the translation rate and the functional role of the mRNA. AS can act on different cellular and biological processes or be involved in tissue specificity, developmental states, or disease conditions, such as cancer [11]. It has a relevant role in cancer development and maintenance, affecting most of the hallmarks of cancer [12,13]. In addition, it can be involved in cancer relapse or resistance to different treatment modalities [12]. Thus, specific isoforms have been identified promoting and supporting neoplastic transformation and tumor development. In a variety of tumor types, AS has been linked to up-regulation of oncogenes, participating in different processes of tumor development, including angiogenesis, cell division, altered metabolism, proliferation, or metastasis [10,14]. In addition, they can also contribute to the deregulation of several non-oncogenic vulnerabilities that are also relevant in the initiation and maintenance of the oncogenic transformation [12].

Alterations in the AS machinery have been identified in different human tumors, and they can affect a network of downstream splicing targets. Using high throughput methodologies, Seiler et al. have described somatic mutations in 119 splicing factors in 33 tumor types, bladder carcinoma and uveal melanoma being those with higher frequencies [15]. Moreover, mutations in splicing factors appear to be mutually exclusive within a tumor, which might indicate that co-occurrence of these mutations may be lethal [15].

Specifically in breast cancer, AS affects major breast cancer-related proteins, such as the estrogen receptor (ER), BRCA1, and BRCA2, among others [16]. Thus, disequilibrium

between ERα66 and ERα36 induce abnormal proliferation, and high levels of ERα36 can cause resistance to Tamoxifen [16]. Alterations in components of the regulatory splicing machinery have been described in breast cancer. For example, SF3B1 is involved in the 3 ′ -SS recognition and is one of the most commonly mutated genes with a higher frequency in the metastatic setting [17]. Mutant SF3B1 produces aberrant splicing, inducing metabolic reprogramming [18]. In addition, AS has been described to have a role in drug resistance. For instance, it has been described that, in carriers of BRCA1 exon 11 premature termination codon variants, tumors upregulate exon 11 skipping to produce isoforms that retain residual activity, contributing to PARPi resistance [19]. Overexpression of SF3B1 and SF3B3 are associated with tamoxifen and fulvestrant resistance, and inhibition of another splicing factors, such as ZRANB2 and SYF2, reduces resistance to doxorubicin in breast cancer cells [20,21]. On the other hand, SRSF4 induces apoptosis in cancer cells, in combination with platinum agents [22].

In our study, alterations in 304 splicing factors were evaluated in breast cancer patients using several large datasets. We found high frequency of amplification in *CLNS1A*, *LSM1*, and *ILF2* in Luminal tumors, with a significant association with poor prognosis. Despite the limited information about these genes, they have been associated with oncogenic processes and resistance to treatments. IFL2 deregulation has been related to an aberrant RNA splicing pattern, mainly deregulated skipped exons in genes involved in DNA repair [23]. LSM1 is included in the heteroheptameric complex LSM1-7, which initiates mRNA decay [24]. CLNS1A acts as a Sm chaperone, recruiting Sm proteins to the PRMT5 complex [25]. In our study, genomic regulation of these genes, with a reduction of their expression, decreased proliferation of luminal tumor cells. In addition, treatment with epigenetic modulators, such as the Bromodomain and extraterminal (BET) family of inhibitors, reduced the expression level of these genes, leading to cell growth reduction.

#### **2. Materials and Methods**

#### *2.1. Data Collection and Processing*

Processed TCGA (The Cancer Genome Atlas) PanCancer dataset was downloaded through cBioportal (www.cbioportal.org; accessed on 4 December 2019). This dataset contains whole exome sequencing and RNA-Seq data from patients with breast invasive carcinoma, consisting of 696 Luminal, 78 HER2 positive, and 171 Basal tumors and their matched normal tissues. WES data was used to explore CNAs and mutations in 304 splicing factor genes. Splicing related genes were collected from four sources: HUGO Gene Nomenclature Committee and the studies of Hegele et al. [26], Wan et al. [9], and Koedoot et al. [20]. Only somatic non-silent mutations in splicing factor genes were considered (missense, premature termination codon, and IVS+-1,2). Somatic non-silent mutations in splicing factor genes were only considered. In the PanCancer dataset, identification of somatic single nucleotide variations was performed using Mutect. CNAs were assessed as deviations in the tumor sample from the paired normal tissue sample using GISTIC 2.0. GISTIC 2.0 identifies regions significantly amplified or deleted and lists genes found in each "wide peak" region [27]. Value +/− 2 indicates high-level thresholds for amplifications/deletions, respectively, and those with +/− 1 exceed the low-level thresholds but not the high-level thresholds. In addition, the Metabric dataset (www.cbioportal.org; accessed on 4 December 2019) was used to validate results of identified genes with a high level of amplification in >5% of tumors.

#### *2.2. Outcome Analysis*

The relationship between gene expression levels and patient clinical prognosis in terms of relapse-free survival (RFS) and overall survival (OS) was evaluated using the Kaplan–Meier Plotter platform, as described previously [28,29] (accessed on 6 June 2020). This tool used gene expression and survival data from 7830 breast tumors (sources include GEO, EGA and TCGA). Samples were split into two groups using the best threshold as the cutoff (auto select best cutoff). When testing multiple genes, the analysis was performed

using the mean expression. Patients above the threshold were labelled as "high" expressing, while patients below the threshold were labelled as "low" expressing. The two groups were compared using Cox survival analysis. The prognostic value of the identified signature (containing *LSM1*, *CLSN1A,* and *ILF2*) was validated using the TCGA project.

The correlation between CNAs and patient clinical outcome was analyzed using the Genotype-2-Outcome platform (accessed on 8 January 2021) [30]. This tool links genotype to clinical outcome by utilizing next generation sequencing and gene chip data of 6697 breast cancer patients. It allows the association with prognosis of a specific transcriptomic signature linked to an altered gene, by classifying patients according to the average expression of significant genes designated as a surrogate metagene of its alteration status. The median expression values for different transcripts are used as a cut-off to discriminate "high" and "low" expression cohorts, which are compared using a Cox survival analysis. To identify factors independently associated with OS and RFS, a multivariate analysis (Cox proportional risk regression model) was performed.

#### *2.3. Cell Culture and Compounds*

MCF7 and T47D cells (American Type Culture Collection, Manassas, VA, USA) were cultured in DMEM (Sigma-Aldrich, Saint Louis, MO, USA) containing 10% fetal bovine serum with 100 U/mL penicillin, 100 µg/mL streptomycin, and 2 mM L-glutamine, and cells were maintained at 37 ◦C in a 5% CO<sup>2</sup> atmosphere. The BET inhibitor JQ1 was purchased from Tocris Bioscience, and BET-PROTAC MZ1 was purchased from Selleckchem (Houston, TX, USA).

#### *2.4. Small Interfering RNA*

siRNA oligonucleotides (Sigma-Aldrich, Saint Louis, MO, USA) were transfected into cells using Lipofectamine RNAiMax protocol (Thermo Fisher Scientific, Rockford, IL, USA) at a final concentration of 20 nM. References: siLSM1(EHU121391), siCLNS1A (EHU147241), and siILF2 (EHU084311). siGFP (EHUEGFP) was used as the negative control of transfection. Briefly, cells were transfected (~80% of confluency), and after 24 h, cells were reseeded for validation experiments.

#### *2.5. Quantitative Reverse-Transcription PCR*

Total RNA was collected from cells using the RNeasy Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. Determination of concentration and purity were measured using a NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific, Rockford, IL, USA), and then 1 µg of total RNA was reverse transcribed using the RevertAid H Minus first-strand cDNA synthesis kit (Thermo Fisher Scientific) in a thermal cycler (Bio-Rad, Hercules, CA, USA) under the following reaction conditions: 65 ◦C for 5 min, 42 ◦C for 60 min, and 70 ◦C for 10 min. cDNAs were then subjected to real-time PCR analysis using Fast SYBR Green master mix on the StepOnePlus Real-Time PCR system (Applied Biosystems) according to the manufacturer's instructions. Primer sequences used were as follows: h-LSM1 F: TTCCTCGAGGGATTTTTGTG, h-LSM1 R: TTCTCTGCTTCCAGCTTGGT, h-CLNS1A F: TCGGCACTGGTACCCTTTAC, h-CLNS1A R: AATGGTGGGGTATTCCAGTG, h-ILF2 F: GCTCCAGGGACATTTGAAGT, h-ILF2 R: CAGCCACATTGTGTCCTGTAG, h-18S F: GAGGATGAGGTGGAACGTGT, h-18S R: TCTTCAGTCGCTCCAGGTCT. An initial step was performed at 95 ◦C for 5 min, followed by 40 cycles of 95 ◦C for 15 s, and finished at 60 ◦C for 1 min. Each sample was analyzed in triplicates, and cycle threshold (Ct) values of transcripts were determined using StepOne Software v.2.1. Ct values were calculated using the 18Sas reference. Untreated control cells were used as the control to calculate the Ct value and determine the X-fold mRNA expression.

#### *2.6. Proliferation MTT Assays*

Cell proliferation was measured using MTT reagent (3-(4, 5-dimethylthiazol-2-yl)- 2, 5 diphenyltetrazolium bromide) (Sigma-Aldrich). MTT reduction by mitochondria of living cells generate formazan accumulates.

For evaluated siRNAs antiproliferative effect, MCF7 and T47D cells (5000/well, 48 multiwell plates) were seeded after siRNAs transfection during 24, 48, and 120 h.

For antiproliferative drugs validation, MCF7 and T47D cells (5000/well, 48-multiwell plates) were seeded. After, they were treated with increased doses of JQ1 and MZ1 for 72 h. Later medium was replaced with red-phenol free DMEM containing MTT (0.5 mg/mL) and incubated for 45 min at 37 ◦C. After, medium was removed and dimethylsulfoxide (DMSO) (Thermo Fisher Scientific) was used for dissolved formazan accumulates. Absorbance (A555 nm–A690 nm) was recorded in a multiwell plate reader (BMG labtech, Ortenberg, Germany).

#### *2.7. Growth Studies*

To compare the growth between siRNAs-transfected cells and siGFP-transfected cells (control), proliferation rate was studied by cell count. Cell lines were cultured at a density of 50,000 cells in 6-well. At the times of 24 and 48 h, cells were trypsinized and counted. Images was performed at 48 h using inverted microscope (10×).

#### *2.8. Cell Cycle Assay*

siRNAs-transfected cells (MCF7 and T47D) were collected and fixed in ethanol (70%, cold) for 30 min. Cell pellets were washed in PBS+2% BSA and incubated in the dark for 1 h at 4 ◦C with Propidium iodide/RNAse staining solution (Immunostep).

#### *2.9. Statistical Analysis*

We used the student's *t*-test unpaired for independent samples. The level of significance was considered 95%; therefore, *p* values lower than 0.05 were considered statistically significant: \* *p* < 0.05; \*\* *p* < 0.01, and \*\*\* *p* < 0.001. Statistics and representations were made with statistical software GraphPad Prism 7.0. All results (unless indicated) are presented as the mean ± SEM of three independent experiments, each of them performed in triplicate.

#### **3. Results**

#### *3.1. Mutations in Splicing-Related Genes*

Alterations in 304 splicing-related genes (Supplementary Table S1) were analyzed in 945 breast cancer patients (499 Luminal A, 197 Luminal B, 171 Basal, and 78 HER2+ samples) using the Breast Invasive PanCancer Atlas Dataset, as described in the materials and methods section. Non-synonymous mutations were detected in 278 genes, with 525 tumors showing at least one altered splicing-related gene (Supplementary Table S2, Supplementary Figure S1). When patients were classified based on molecular subtypes, several differences were observed in the distribution of the identified genes. Regarding the 278 genes with presence of mutations, the number of altered genes were 231 (83.1%) for Luminal A, 157 (56.5%) for Luminal B, 175 (62.9%) for Basal, and 150 (54%) for HER2+ subtype (Figure 1A). Tumors with modifications in any of these genes were observed in a higher percentage in HER2+ (70.5%) and Basal (66.1%) compared with the Luminal A and B subtypes (47.1% and 61.9%, respectively) (Figure 1B). When the frequency of tumors with alterations in each gene were evaluated, HER2+ and the basal subtype population showed splicing-related genes with a higher percentage of alterations (Figure 1C; Supplementary Table S2). In the four molecular subtypes, no gene was detected to be mutated in more than 6% of tumors. Splicing genes with mutations in >3% of tumors are displayed in Figure 1C and mainly belonged to the HER2+ and Basal subgroups (Figure 1D).

**Figure 1.** Percentage of splicing-related genes showing mutations in each molecular subtype (**A**). Percentage of tumors with at least one mutated splicing-related gene (**B**). Frequency of non-synonymous mutations in splicing-related genes in each molecular subtype (**C**). List of splicing-related genes mutated in >3% of tumors (**D**). \*: *p* < 0.05; \*\*\*: *p* < 0.001.

#### *3.2. Copy Number Alterations (CNAs) in Splicing-Related Genes*

The TCGA PanCancer series also includes putative copy-number data [31]. Thus, we evaluated the following changes in splicing-related genes: homozygous or hemizygous deletions, no change, gain, and high level of amplification. In this large series of breast cancer patients, we found information about 301 genes from those identified. High level of amplification (GISTIC thresholded CN value of +2) in any of these genes were detected in a high percentage of tumors (HER2+: 87.2%; Basal: 81.3%; Luminal A: 51.1%; and Luminal B: 67.5%). Considering only those genes in regions with homozygous deletion or a high level of amplification in >5% of patients, we found 61 altered splicing-related genes (58 amplified and 3 loss). Regarding the molecular subtypes, 33 (10.9%) splicing-related genes were altered in the Basal subtype (30 amplified and 3 loss), 41 (13.6%) amplified in HER2+, 30 (9.9%) amplified in Luminal A, and 28 (9.3%) altered in Luminal B (26 amplified and 2 loss) (Figure 2A–D, respectively). Therefore, a large number of genes showed higher frequencies of CNAs versus mutations (only 6 genes with mutations in >5% of patients, Figure 1D). In the four molecular subtypes, 14 common splicing-related genes showed a high level of amplification in >5% of tumors (Figure 2E). On the other hand, certain genes were amplified only in specific subtypes (Figure 2F). A complete list of genes is displayed in the Supplementary Table S3.

**Figure 2.** Copy number alteration frequencies in splicing-related genes: list of genes with high amplification in >5% of tumors for Basal (**A**), HER2+ (**B**), Luminal A (**C**), and Luminal B (**D**) molecular subtypes. In total, 14 common splicingrelated genes showed high level amplification in >5% of tumors, shown in bold font. Percentage of tumors with presence of amplifications in >5% of tumors in splicing-related genes both common in all molecular subtypes (**E**) and specific in each subtype (**F**).

With this information, we next aimed to explore the chromosome location of splicingrelated genes with a high level of amplification or homozygous deletion. Interestingly, most altered splicing-related genes in each molecular subtype were distributed in a few chromosome regions: 1q, 8q, and 17q, as shown in Figure 3A. In total, 12 out of 14 common amplified genes were located in 1q and 8q. Different altered regions were specific for each subtype: (a) 10p, 12p 13q, 15q, and 19q for Basal; (b) 6q, 17q, and 3q for HER2+; (c) all genes in the 16p region for Luminal A; and (d) no one for Luminal B (Figure 3A; Supplementary Table S3). Copy-number gain in these regions is represented in Figure 3B.

**Figure 3.** Number of splicing-related genes with high amplification in >5% of patients by chromosome location (**A**) (Created with BioRender.com (accessed on 4 December 2019)). Tumors with chromosomal gain (red) in each molecular subtype (**B**).

#### *3.3. Associations of Splicing-Related Genes with Clinical Outcome in Patients*

To identify which of the identified genes could have a role in cancer progression, we intended to link the described data with patient clinical outcome. To do so, we used published transcriptomic microarray data, as described elsewhere [32]. The prognostic value of the high amplified genes (with a cutoff of >5% of tumors) were analyzed in a dataset of 6234 breast cancer patients (Figure 4A). CNA frequencies in identified genes were validated in an additional dataset (Supplementary Figure S2). Frequencies of high level of amplified genes were correlated between both datasets. Alterations in splicing-related genes were most frequently observed in the HER2+ and Basal-like subtypes. However, as there were few patients in these subtypes, it was not possible to establish the prognostic value for most amplified genes. Despite the low number of patients in this subgroup, several genes showed association with RFS and OS (Figure 4B and Supplementary Table S4). We focused on those genes, with a clear impact on survival by using an arbitrary selection based on statistical outcome relevance and low false discovery rate (FDR) (*p* < 0.002, Hazard Ratio (HR) > 1.5; FDR < 5). For Luminal A, high expression of *ESRP1*, *LSM1*, *CLNS1A*, *ILF2,* and *PPP1CA* showed a clear association with detrimental OS and RFS (Figure 4B). In the same way, high levels of these first four genes were associated with a poor prognostic in the Luminal B subtype. In the Luminal series, CNAs were significant associated with expression levels in these genes (Supplementary Figure S3).

**Figure 4.** Prognosis value for splicing-related genes with CNAs. Splicing-related genes (only showed those genes with high amplification in >5% of patients) with higher prognostic value based on hazard ratio and *p* values (**A**). List of genes showing significant association between expression levels and detrimental prognosis in RFS (KM Plotter software was used) (**B**). Prognostic value (OS) of selected genes (based on: *p* < 0.002, HR > 1.5; FDR < 5 from KM Plotter) was confirmed using genotype-2-outcome web-server (**C**). Summary of outcome results obtained to *LSM1*, *CLNS1A,* and *ILF2* in the Luminal A subtype (**D**). Kaplan-Meier plots (OS and RFS) for the combination of these three genes using KM Plotter software (**E**) and their validation in another Luminal A cohort (TCGA project) (**F**).

Next, to confirm the prognostic role of the alterations described before, we used a transcriptomic fingerprint of the amplified splicing-related genes by using the genotype-2-outcome (Figure 4C). With this approach, we can obtain the clinical outcome of a gene signature associated with a specific genomic alteration, including gene amplification, as described in the materials and methods section. Thus, the transcriptomic fingerprint associated with the amplification of *LSM1*, *CLNS1A,* and *ILF2* showed a strong association with survival (Supplementary Figure S4). The transcripts included in the signatures associated with the CNA gain of *LSM1*, *CLNS1A,* and *ILF2* are displayed in Supplementary Table S5.

In Figure 4D, we summarized the prognostic value (RFS and OS), percentage of amplification, subtype, and chromosome location of the identified genes. In addition, a more robust association with OS and RFS was observed when expression of these three genes was combined together (Figure 4E). Finally, the prognostic role of the identified signature in the Luminal A subtype was confirmed in a validation cohort, confirming the reproducibility of the findings described before (TCGA dataset; Figure 4F). Univariate and multivariate COX regression analysis showed that the combination of LSM1, CLNS1A, and ILF2 was a clear, independent prognostic factor, mainly with OS (Supplementary Table S6).

#### *3.4. Genomic Down-Regulation of LSM1, CLNS1A, and ILF2 Reduces Cell Proliferation*

To validate *LSM1*, *CLNS1A,* and *ILF2* dependency in Luminal breast cancer cells lines, mRNA expression of these genes was downregulated by using small interfering RNA (siRNA). *LSM1*, *CLNS1A,* and *ILF2* downregulation in MCF7 and T47D efficiently reduced gene expression, as shown in Figure 5A. Cell growth (Figure 5B,C) and cell proliferation, evaluated as MTT metabolization (Figure 5D), was reduced after siRNA knockdown of the mentioned splicing genes. Growth reduction was observed clearer with gene interfering

of *LSM1* in MCF7 cells and *CLNS1A* in T47D. The antiproliferative effect produced after gene inhibition evaluated as a metabolization of MTT was significantly observed after 120 h, with no differences at a shorter time. To explore how the mechanism for genomic down-regulation of *LSM1*, *CLNS1A,* and *ILF2* inhibits cell proliferation, we performed cell cycle analysis using propidium iodure. No major changes were observed in cell cycle phases, only *CLNS1A* down-regulation showed a G0/G1 arrest in T47D cells in accordance with previous findings (Figure 5E).

μ **Figure 5.** Splicing-related genes genomic inhibition by siRNA and pharmacological inhibition by BET inhibitor and PROTAC. (**A**). *LSM1*, *CLNS1A,* and *ILF2* mRNA expression in MCF7 and T47D luminal A breasts cancer cells after transfection with siRNAs. Cells were transfected using lipofectamine reagent and 24 h later were reseeded. After 24 h, (48 h post-transfection), cells were collected, RNA was extracted, and qPCR was performed. siGFP was used as the control of transfection. (**B**). Transfected cells were seeded (50,000 cells 6-well plate and were counted after 24, 48 and 120 h). (**C**). Images obtained by inverted microscope of transfected cells after 48 h. (**D**). Antiproliferative effect of siRNA evaluated by MTT assays after 24, 48, and 120 h. (**E**). Changes in cell cycle phases after genomic inhibition (representative plot of two independent experiments is shown). Scale bar = 100 µm. \* *p* < 0.05; \*\* *p* < 0.01; \*\*\* *p* < 0.001.

#### *3.5. BET Inhibitors Reduce the Expression of LSM1, CLNS1A, and ILF2*

Epigenetic agents can modulate the expression of genes that have a role in transcription and maturation [33]. With this in mind, we explored the effect of Bromo and Extra terminal domain (BET) inhibitors and BET derivatives, such as BET-Proteolysis targeting chimeras (PROTAC), on the expression of the identified genes. Administration of the BET inhibitor (JQ1) and BET-PROTAC (MZ1) produced a reduction of the gene expressions of *LSM1*, *CLNS1A,* and *ILF2*. In MCF7 cells, *ILF2* was downregulated with MZ1 treatment after 12 and 24 h of administration. Moreover, *LSM1* was downregulated with MZ1 after 12 h (Figure 6A). In T47D cells, after 12 h of treatment, these three genes were downregulated by both JQ1 and MZ1. This effect was maintained at 24 h of treatment for MZ1, but not for JQ1, suggesting that the PROTAC had a more prolonged effect (Figure 6B). Following these findings, we explored their effect on cell growth. We observed that JQ1 and MZ1 displayed an antiproliferative effect in Luminal cells lines (Figure 6C). EC50 values showed that MZ1 PROTAC was more potent than the inhibitor JQ1 (Figure 6D). In summary, these findings confirm the modulation of the expression of these three genes by JQ1 and MZ1 and the pharmacological effect of these agents on cell proliferation.

**Figure 6.** Pharmacological inhibition of splicing-related genes by BET inhibitor and PROTAC. *LSM1*, *CLNS1A*, and *ILF2* mRNA expression in MCF7 (**A**) and T47D (**B**) luminal A breast cancer cell lines after 12 h and 24 h JQ1 and MZ1 exposure. Cell viability evaluated by MTT assays for MCF7 (left) and T47D (right) cells treated with increasing doses of JQ1 and MZ1 (**C**). JQ1, MZ1, and EC50 doses in luminal A cell lines (**D**). \* *p* < 0.05; \*\* *p* < 0.01; \*\*\* *p* < 0.001.

#### **4. Discussion**

In the present article, we characterize the presence and role of genomic alterations in splicing genes in breast cancer. Splicing is a biological process that permits transcriptional diversity and redundancy of molecular functions, allowing the integrity of key cellular activities [34]. Transcriptional regulation by splicing has been involved in the control of different biological tasks from DNA damage, to cell survival, or stemness, among others [13]. In this context, several genes implicated in splicing have been described in cancer, leading to the promotion of different oncogenic properties. For instance, some known

factors, such as SRSF1, have been described as overexpressed in cancer, leading to malignant transformation by an alternative splicing of genes involved in proliferation and apoptosis [35]. Other examples include RBM39 in Acute Myeloid Leukemia or RBM11 in glioblastoma cells, among others [36,37]. In breast cancer, mutations in *SF3B1* are more frequently observed in the metastatic setting, and its potential role in the regulation of protein degradation or metabolism is known [17,18]. In addition, overexpression of *SF3B1* and *SF3B3* has been associated with resistance to hormone therapy, and others, such as *ZRANB2* and *SYF2*, to chemotherapy, particularly for doxorubicin [20,21]. Taking this background into consideration, the identification of deregulated genes involved in splicing and the understanding of their role in cancer is a main objective, with the final goal of designing or implementing therapeutic strategies to reduce their presence.

In our study, we analyzed a set of genes involved in splicing in breast cancer. Genomic modifications of splicing proteins were highly presented in breast cancer, the HER2 subtype being the most common tumor (70.5%), with the less frequency presence observed in the Luminal A subtype (47.1%). Although mutations in splicing genes have been widely reported [15], in our study, no specific gene was mutated in more than 6% of the tumors. On the other hand, when CNAs in our splicing-related gene lists were evaluated, 61 of them were altered in >5% of patients. In a similar way to mutations, the HER2 subtype showed a higher number of altered genes compared with the other groups. This really demonstrates that mutations are less frequently observed than other structural alterations, and the splicing pathway is mainly altered in the HER2 subtype compared with the other breast subtypes. Nevertheless, 14 common splicing-related genes showed high-level amplification in >5% of patients in the four molecular groups, most of them located in 1q and 8q chromosome regions.

The next step in our study was to select those altered splicing-related genes with a role in patient clinical outcome. The results were not conclusive for the HER2 subtype due to the small number of patients. Regarding the Luminal subtype, we identified three genes with clear association with poor prognosis: *ILF2*, *LSM1,* and *CLNS1A*. Although prognosis value cannot be evaluated in HER2 and Basal subtypes, these three genes were also detected as amplified in tumors of these molecular subtypes. Moreover, when the presence of CNAs in our selected genes was analyzed in different tumor types (GDC Data Portal; 67 primary sites), breast cancer was one of most frequently amplified for *LSM1*, *CLNS1A,* and *ILF2* (Supplementary Figure S5). IFL2 has been described as implicated in the RNA splicing regulation of crucial effectors involved in DNA damage response [23]. In addition, overexpression of this gene mediated resistance to DNA damaging agents [23]. Of note, *IFL2* is located at the 8p chromosome, where other genes with a particular oncogenic role in breast cancer, such as FGFR1, has been described as amplified [38]. LSM1 is involved in pre-RNA splicing by acting on the removal of the 5′ cap structure [24,39,40]. *LSM1* has been studied in other tumor types, such as pancreatic cancer, observing a role in cancer progression, metastasis, and resistance to chemotherapies [41]. CLNS1A is involved in both the assembly of spliceosomal snRNPs and the methylation of Sm proteins [25,42]. CLNS1A cooperates with the protein PRMT5 and functions as an epigenetic activator of AR transcription in castration resistance prostate cancer [43]. *CLNS1A* has also been described in malignant gliomas [44], but data for breast cancer is very limited.

An interesting observation is the fact that the overexpression and amplification of these three genes was associated with detrimental prognosis in two large datasets, particularly in the luminal breast cancer subtype. Furthermore, the genomic knockdown of these transcripts reduced cell proliferation, suggesting an effect on cell growth.

Pharmacological modulation of transcription and RNA regulation is key for the optimal development of therapeutic strategies against key proteins. Spliceosome inhibitors have been developed, particularly those that bind to the HEAT repeats domain of some proteins, such as SF3B1 [45]. A comprehensive description has been recently reviewed elsewhere and beyond the scope of this article [12]. However, another approach to target this family of genes is the use of epigenetic modulators, such as BET inhibitors, to modulate

transcriptional regulators or genes involved in RNA maturation. Examples have been provided with BET inhibitors, such as MK-8628 or ZEN003694 [12]. In this context, we observed that administration of the BET inhibitor JQ1 and the BET-PROTAC MZ1 reduced the expression of the three identified genes at different levels in two characteristic estrogen receptor breast cancer cell lines, MCF7 and T47D. In addition, these agents displayed a significant antiproliferative effect on these cell models. Although we agree that the antiproliferative effect of the compound could be multifactorial and the participation of these genes is a part and not a whole, we demonstrate in breast cancer that BETi can modulate the expression of splicing-related agents.

#### **5. Conclusions**

In conclusion, we describe novel splicing genes amplified in luminal breast tumors that are associated with detrimental prognosis and can be modulated pharmacologically. This data opens the door for further studies, confirming the effect of these genes in patients treated with BET inhibitors.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/cancers13164118/s1, Figure S1: Percentage of tumors with non-silent mutations for each splicing-related gene; Figure S2: Relation between expression level and CNAs for *ILF2*, *CLNS1A*, *LSM1,* and *ESRP1* genes; Figure S3: Associations of the transcriptomic fingerprint associated with amplification of *LSM1*, *CLNS1A,* and *ILF2* genes (genotype-2-outcome) and clinical outcome (OS and RFS); Figure S4: Presence of CNAs in our selected genes (*LSM1, CLNS1A,* and *ILF2*) in different tumor types (GDC Data Portal; 67 primary sites; Gain: red and Loss: blue); Table S1: List of splicing related genes evaluated in our study; Table S2: Percentage of tumors with mutation for each gene; Table S3: Common and specific splicing-related genes depending on molecular subtypes. Table shows the percentage of tumors with a high level of amplification for each gene (only included those with high amplification in >5%); Table S4: Prognostic value (RFS and OS) of the amplified genes (with a cutoff of >5% of tumors) in the HER2+ and Basal-like subtypes; Table S5: List of the transcripts included in the signatures associated with the CNA gain of *LSM1*, *CLNS1A,* and *ILF2*. Table S6: Univariate and Multivariate COX regression analysis to assess the potential prognostic value of CLNS1A, ILF2, and LSM1 expression in Luminal breast cancer patients.

**Author Contributions:** Conceptualization, M.D.L.H., A.P., V.G.-B. and A.O.; data curation, G.F.- H. and B.G.; formal analysis, M.d.M.N.-L., I.L.-C., G.F.-H., A.P., B.G., V.G.-B. and A.O.; funding acquisition, V.G.-B. and A.O.; investigation, A.M., J.Á.G.-S., P.P.-S. and M.D.L.H.; methodology, M.d.M.N.-L., I.L.-C., J.F.-A. and A.E.-S.; project administration, V.G.-B. and A.O.; resources, A.M., J.Á.G.-S., P.P.-S., M.D.L.H. and V.G.-B.; software, M.d.M.N.-L., I.L.-C., G.F.-H. and B.G.; supervision, V.G.-B. and A.O.; validation, I.L.-C., J.F.-A. and A.E.-S.; visualization, M.d.M.N.-L.; writing—original draft, M.d.M.N.-L., I.L.-C., V.G.-B. and A.O.; writing—review and editing, A.M., J.Á.G.-S., P.P.-S., M.D.L.H., A.P. and B.G. All authors have read and agreed to the published version of the manuscript.

**Funding:** A.O.'s lab is supported by the Instituto de Salud Carlos III (ISCIII, PI19/00808); ACEPAIN; CRIS Cancer Foundation and Diputación de Albacete. This research is also supported by PI18/01020 from the Instituto de Salud Carlos III and co-financed by the European Development Regional Fund (FEDER) "A way to achieve Europe" (ERDF); N.L. MDM was supported by the Spanish Ministry of Education (FPU grant; Ref.: FPU18/01319). B.G. was financed by the 2018-2.1.17-TET-KR-00001, 2020-1.1.6-JÖVO-2021-00013, and 2018-1.3.1-VKE-2018-00032 grants and by the Higher ˝ Education Institutional Excellence Programme (2020-4.1.1.-TKP2020) of the Ministry for Innovation and Technology in Hungary.

**Institutional Review Board Statement:** Not applicable. All the data corresponding to the breast cancer patient series used in this study are available in the public functional genomics data repository.

**Informed Consent Statement:** Not applicable. All the data corresponding to the breast cancer patient series used in this study are available in the public functional genomics data repository.

**Data Availability Statement:** All the data used in this study are available in public functional genomics data repositories (GEO, EGA, cBioportal, and TCGA).

**Acknowledgments:** We are grateful to Francisco J. Cimas for support in platforms about mechanism of action of drugs, and to Miguel Burgos for support in cell cycle experiments.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


#### *Article*

### **Expression of Phosphorylated BRD4 Is Markedly Associated with the Activation Status of the PP2A Pathway and Shows AStrong Prognostic Value in Triple Negative Breast Cancer Patients**

**Marta Sanz-Álvarez 1,† , Ion Cristóbal 2, \* ,† , Melani Luque 1 , Andrea Santos 2 , Sandra Zazo 1 , Juan Madoz-Gúrpide 1 , Cristina Caramés 2 , Cheng-Ming Chiang 3 , Jesús García-Foncillas 4 , Pilar Eroles 5 , Joan Albanell <sup>6</sup> and Federico Rojo 1, \***


**Simple Summary:** The use of BRD4 inhibitors has emerged as a novel therapeutic approach in a wide variety of tumors including the triple negative breast cancer. Moreover, PP2A has been proposed as the phosphatase involved in regulating BRD4 phosphorylation and stabilization. Our aim was to evaluate for the first time the clinical impact of BRD4 phosphorylation in triple negative breast cancer patients and as well as its potential linking with the PP2A activation status in this disease. Our findings are special relevant since they suggest the prognostic value of BRD4 phosphorylation levels, and the potential clinical usefulness of PP2A inhibition markers to anticipate response to BRD4 inhibitors.

**Abstract:** The bromodomain-containing protein 4 (BRD4), a member of the bromodomain and extra-terminal domain (BET) protein family, has emerged in the last years as a promising molecular target in many tumors including breast cancer. The triple negative breast cancer (TNBC) represents the molecular subtype with the worst prognosis and a current therapeutic challenge, and TNBC cells have been reported to show a preferential sensitivity to BET inhibitors. Interestingly, BRD4 phosphorylation (pBRD4) was found as an alteration that confers resistance to BET inhibition and PP2A proposed as the phosphatase responsible to regulate pBRD4 levels. However, the potential clinical significance of pBRD4, as well as its potential correlation with the PP2A pathway in TNBC, remains to be investigated. Here, we evaluated the expression levels of pBRD4 in a series of 132 TNBC patients. We found high pBRD4 levels in 34.1% of cases (45/132), and this alteration was found to be associated with the development of patient recurrences (*p* = 0.007). Interestingly, BRD4 hyperphosphorylation predicted significantly shorter overall (*p* < 0.001) and event-free survival (*p* < 0.001). Moreover, multivariate analyses were performed to confirm its independent prognostic impact in our cohort. In conclusion, our findings show that BRD4 hyperphosphorylation is an alteration associated with PP2A inhibition that defines a subgroup of TNBC patients with unfavorable prognosis, suggesting the potential clinical and therapeutic usefulness of the PP2A/BRD4 axis as a novel molecular target to overcome resistance to treatments based on BRD4 inhibition.

**Citation:** Sanz-Álvarez, M.; Cristóbal, I.; Luque, M.; Santos, A.; Zazo, S.; Madoz-Gúrpide, J.; Caramés, C.; Chiang, C.-M.; García-Foncillas, J.; Eroles, P.; et al. Expression of Phosphorylated BRD4 Is Markedly Associated with the Activation Status of the PP2A Pathway and Shows AStrong Prognostic Value in Triple Negative Breast Cancer Patients. *Cancers* **2021**, *13*, 1246. https:// doi.org/10.3390/cancers13061246

Academic Editor: Sara A. Hurvitz

Received: 9 February 2021 Accepted: 9 March 2021 Published: 12 March 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

**Keywords:** pBRD4; SET; PP2A; prognosis; triple negative breast cancer

#### **1. Introduction**

Breast cancer has the highest prevalence in cancer diagnosis and represents the second leading cause of female cancer-related deaths [1]. Breast cancer is a very heterogeneous disease, with different molecular subtypes including luminal A, luminal B, HER2+, basal and normal-like tumors [2,3]. The triple negative breast cancer (TNBC) is molecularly characterized by the lack of hormonal receptors expression (estrogen (ER) and progesterone receptors (PR)), and by an absence of expression of the HER2 receptor [3]. TNBC represents 15–20% of all breast carcinomas [4] and shows more aggressive features such as emergence at a younger age, higher tumor size and grade, and greater proportion of positive lymph node metastases. TNBC has been largely described as the breast cancer subtype with the worst overall and progression-free survival rates, and represents a major challenge for current clinical management due to the lack of established and effective therapeutic strategies [5,6]. TNBC cells have very aggressive behavior that leads to a shorter time of disease progression. In fact, TNBCs show the highest recurrence rates, with brain and visceral organs as the main metastatic niches [7]. Triple negative tumors are heterogeneous at the molecular level, and *TP53*, *PIK3CA*, *PTEN*, *RB1*, *EGFR* and *MYC* have been reported as the most commonly mutated genes [8,9]. However, it remains urgent to improve our understanding about the molecular alterations that govern TNBC progression in order to develop novel therapeutic strategies for this disease.

Bromodomain-containing protein 4 (BRD4) is a member of the bromodomain and extra-terminal domain (BET) protein family, along with BRD2, BRD3, and BRDT. BRD4 is structurally composed of two N-terminal bromodomain domains (BD1 and BD2), and a C-terminal extra-terminal domain. BD1 and BD2 allow for the formation of a hydrophobic pocket that binds to acetylated lysine residues of histones or transcription factors [10,11], ultimately regulating a wide variety of cell functions. Specifically, BRD4 is involved in chromatin decompaction, the recruitment of components of the transcriptional complex, as well as in the stages of initiation, release pause and elongation of transcription due to its interaction with PTEF-b that phosphorylates RNA Pol II [10]. Due to its role in important cellular processes, BRD4 dysfunction can lead to the appearance of various human diseases, including inflammation, cardiovascular diseases and cancer [10–12]. BRD4 has been found to play oncogenic roles in many hematological and solid tumors, including melanoma, prostate and breast cancer among others, and has been proposed as a druggable promising target in human cancer [12–16]. BRD4 has been shown to regulate the expression of different set of oncogenic drivers, such as c-MYC [13], NF-κB [16] or Jagged1 [17]. In breast cancer, several BRD4 alterations involved in the different molecular subtypes have been reported to date. Thus, BRD4 activity has been found to be required for proliferation and ERα function in ER+ breast cancer cells [18], and promotes the migration and invasion of triple negative tumors through controlling Jagged1 expression [17]. Regarding its post-translational modifications, CK2-mediated BRD4 hyperphosphorylation has been associated with greater stability and nuclear localization of the BRD4 protein [19], with important functional and therapeutic implications in TNBC [20,21]. In fact, the therapeutic value of BRD4 inhibition in TNBC has been previously reported by Shu and co-workers [21], analyzing a set of BRD4 inhibitors across a panel of cell lines with different breast cancer subtypes, observing that these drugs showed the strongest antitumor effects in the triple negative subtype. These results were confirmed in vivo using primary human TNBC xenografts. After an exhaustive analysis of potential mechanisms of drug resistance, BRD4 was identified as a novel PP2A target and its hyperphosphorylation as an alteration that promotes resistance to BRD4 inhibition in TNBC cells.

In the last years, several studies have evaluated distinct therapeutic approaches related totargeting BRD4 in TNBC. It has been reported promising antitumor properties using cell-penetrating peptides including EGFR and BRD4 siRNAs in TNBC cells [22], or a dual-target small-molecule inhibitor co-targeting PARP1 and BRD4 [23]. Moreover, it has been described that BRD4 regulates PD-L1 expression in TNBC cells, which could have interesting implication for immunotherapy-based treatments [24], or the therapeutic usefulness of strategies based on BRD4 inhibition, due to its role as regulator of the oncogenic c-MYC pathway in this disease [25,26].

Altogether, the different studies in the literature regarding BRD4 in TNBC highlight its promising therapeutic value. However, little is known about its clinical impact as well as the functional and therapeutic significance of pBRD4 in this disease. Moreover, the relevance of the PP2A pathway as a potential regulator of pBRD4 remains to be investigated and confirmed in TNBC patient cohorts.

#### **2. Experimental Section**

#### *2.1. Patient Samples*

A total number of 132 surgical resection specimens from patients diagnosed withprimary breast cancer were included in this study. Formalin-fixed paraffin-embedded breast tumor specimens from this patient cohort were retrospectively selected from Fundación Jiménez Díaz Biobank (Madrid, Spain) following these criteria: infiltrating carcinomas, operable, enough available tissue, molecular and/or clinical follow-up data and triple negative subtype. Clinical data were collected from medical clinical records by oncologists. Samples were taken anonymously. TNM (tumor–node–metastasis) staging classification was performed using the American Joint Committee on Cancer (AJCC) staging system. The Scarff–Bloom–Richardson modified by Elston criteria [27] was used to define the histological grade. Two independent pathologists who were blinded to patient outcomes evaluated tumor tissue staining.

#### *2.2. Determination of the Molecular Subtype*

We evaluated the expression of hormonal receptors as well as HER2 to define the molecular subtype and confirm that all patients included in this study have triple negative breast tumors. The expression of both estrogen receptor (ER) and progesterone receptor (PR) were determined by immunohistochemistry (IHC) (SP1 and PgR636 clones, respectively; Dako, Carpinteria, CA, USA), establishing positivity criteria in >1% of nuclear tumor staining [28]. Determination of HER2 amplification was carried out by FISH (Pathvysion; Abbott Laboratories, Green Oaks, IL, USA) [29].

#### *2.3. Ethics Approval and Consent to Participate*

This study was conducted in full accordance with the guidelines for Good Clinical Practice and the Declaration of Helsinki. All participants gave written informed consent for tissue storage and analysis at Fundación Jiménez Díaz biobank, Madrid (Spain). The ethical committee institutional review board of Fundación Jiménez Díaz University Hospital reviewed and approved the project (ref. PIC 13-2016).

#### *2.4. Immunohistochemistry*

Representative areas of each tumor were carefully selected, and three tissue cores (1mm diameter) were obtained using a tissue microarray (TMA) workstation (T1000 Chemicon). Immunostainings were performed on tissue sections (3 µM) obtained from FFPE tumors, as previously described [30]. Expression levels of Ki-67 were studied by IHC using the MIB1 clone (Dako, Carpinteria, CA, USA) [31]. High proliferation in our breast cancer patient cohort based on Ki-67 labelling by IHC has been defined following the 13th St Gallen International Breast Cancer Conference (2013) criteria based on a threshold ≥ 20% of proliferation [32]. Other antibodies used were: pBRD4 (developed and kindly provided by Prof. Chiang's laboratory) [19,21], rabbit polyclonal anti-SET (ab1183) (Abcam, Cambridge, UK) and rabbit monoclonal anti-PP2AY307 (1155-1) (Abcam, Cambridge, UK). Antibody dilutions were as follows: pBRD4 (1:100), SET (1:5000), and phospho-

PPP2CA (pPPP2CA) (1:2000). pBRD4, SET and pPPP2CAexpression blinded to clinical data was evaluated by two pathologists (F.R. and S.Z.). The specific phosphorylation sites recognized by the antibodies were Y307 for PPP2CA and S484/488 for BRD4. A semiquantitative histoscore (Hscore) was calculated by estimation of the percentage of tumor cells positively stained with low, medium, or high staining intensity. The final score was determined after applying a weighting factor to each estimate. The formula used was Hscore = (low%) × 1 + (medium%) × 2 + (high%) × 3, and the results ranged from 0 to 300.

#### *2.5. Statistical Analysis*

Statistical analyses were performed using SPSS20 for windows (SPSS Inc, Chicago IL, USA). We applied the χ 2 test (Fisher exact test) based on bimodal distribution of data to evaluate the significance of potential associations between BRD4 phosphorylation and the molecular and clinical characteristics of the tumor specimens included in this study.

Overall survival (OS) was defined as the time from diagnosis to the date of death from any cause or last follow-up. Event-free survival(EFS)was defined as the time from the date of diagnosis until relapse at any location, death or last follow-up. Kaplan–Meier plots and survival comparisons were carried out using the log-rank test if the proportional hazard assumption was fulfilled, and Breslow otherwise. The Cox proportional hazards model was adjusted taking into consideration significant parameters in the univariate analysis. A receiver operating characteristic (ROC) curve was used to determine the optimal cutoff point based on progression end point for pBRD4 as previously described [33,34]. *p*-Value less than 0.05 was considered statistically significant. This work was carried out in accordance with Reporting Recommendations for Tumor Marker Prognostic Studies(REMARK) guidelines [35].

#### **3. Results**

#### *3.1. Prevalence of BRD4 Hyperphosphorylation in Triple Negative Breast Cancer Patients and Its Association with Molecular and Clinical Parameters*

To investigate the prevalence and potential clinical impact of pBRD4 in TNBC, we analyzed the expression of pBRD4 by immunohistochemistry in a cohort of 132 patients with early breast cancer and triple negative subtype, observing high pBRD4 levels in 45 of 132 of cases (34.1%). Patient characteristics are presented in Table S1. We next correlated pBRD4 expression with molecular and clinical features of our patient cohort. Interestingly, high pBRD4 levels were found to be strongly associated with the subgroup of patients who relapsed (*p* = 0.007). Associations between pBRD4 status and clinical and molecular characteristics are shown in Table 1.

#### *3.2. Clinical Impact of pBRD4 in Triple Negative Breast Cancer*

We analyzed the clinical significance of pBRD4 in the same cohort of 132 TNBC patients. Clinical follow-up data were available in all cases. The median of age was 57 years (with an age range of 31 to 90 years). Interestingly, we found that the subgroup of high pBRD4 expressing patients had a markedly shorter OS (*p* < 0.001) (Figure 1A). Moreover, we observed that pBRD4 also had predictive value for EFS in our patient cohort (*p* < 0.001) (Figure 1B).

Interestingly, multivariate Cox analysis showed that high pBRD4 expression is an unfavorable independent factor associated with patient outcome in our cohort (Hazard ratio (HR) = 5.342; 95% confidence interval (CI), 2.286–12.482; *p* < 0.001) (Table 2).


**Table 1.** Association of bromodomain-containing protein 4 (BRD4) phosphorylation levels with molecular and clinical parameters in a cohort of 132 triple negative breast cancer (TNBC) patients.

IDC <sup>1</sup> = invasive ductal carcinoma; ILC <sup>2</sup> = invasive lobular carcinoma; T <sup>3</sup> = tumor size; N <sup>4</sup> = lymph node metastases. *Cancers* **2021**, *13*, x 6 of 12

**Figure 1.** Clinical significance of pBRD4 in TNBC. (**A**) Immunohistochemical images showing pBRD4 positive and negative staining. The line shows 25 μm. Original magnification ×400, (**B**) Kaplan–Meieranalysesof overall survival(OS) and event-free survival(EFS) in a cohort of 132 TNBC patients. Interestingly, multivariate Cox analysis showed that high pBRD4 expression is an **Figure 1.** Clinical significance of pBRD4 in TNBC. (**A**) Immunohistochemical images showing pBRD4 positive and negative staining. The line shows 25 µm. Original magnification ×400, (**B**) Kaplan–Meieranalysesof overall survival(OS) and event-free survival(EFS) in a cohort of 132 TNBC patients.

unfavorable independent factor associated with patient outcome in our cohort (Hazard

*p* **HR** 

0.063 -

0.014 0.100

Stage 0.049 0.195

pBRD4 <0.001 <0.001

*<sup>p</sup>***Lower Upper Lower Upper**

**95% CI** 

ratio (HR) = 5.342; 95% confidence interval (CI), 2.286–12.482; *p* < 0.001) (Table 2).

**95% CI <sup>2</sup>**

2–3 2.280 0.957 to 5.433 - -

High 1.366 0.586 to 3.182 - -

 + 2.286 1.180 to 4.429 1.983 0.877 to 4.484 Grade 0.470 -

III 2.935 1.006 to 8.564 2.174 0.672 to 7.033

High 5.016 2.155 to 11.676 5.342 2.286 to 12.482


I-II 1.000 1.000

Low 1.000 1.000

Ki-67 0.864

High 1.091 0.402 to 2.962

**HR <sup>3</sup>**

**Parameters** 

T <sup>4</sup>

N <sup>5</sup>

0–1 1.000

L/M <sup>6</sup> 1.000

Low 1.000


**Table 2.** Univariate and multivariate Cox analyses in the cohort of 132 TNBC patients.

OS <sup>1</sup> : overall survival; CI <sup>2</sup> : confidence interval; HR <sup>3</sup> : Hazard ratio; T <sup>4</sup> = tumor size; N <sup>5</sup> = lymph node metastases; L/M <sup>6</sup> : low/moderate.

> To further evaluate the prognostic value of pBRD4 in TNBC, we stratified our patient cohort by stage. Of note, we observed that relevance of high pBRD4 expression levels as a biomarker predictor of poor outcome was retained in all cases for both OS and EFS, but the significance was particularly marked in the subgroup of TNBC patients with stage III (*p* < 0.001 for OS, and *p* = 0.001 for EFS), compared to those with stages I-II (*p* = 0.005 for OS, and *p* = 0.017 for EFS) (Supplementary Materials Figure S1).

#### *3.3. BRD4 Phosphorylation Is Associated with the Activation Status of the PP2A Pathway*

We next studied the molecular mechanisms that could be involved in BRD4 hyperphosphorylation. Due to BRD4 having been previously proposed as a direct target of the tumor suppressor protein phosphatase 2A (PP2A) in TNBC, the activation status of this phosphatase was evaluated in our patient series. The phosphorylation of the PP2A catalytic subunit in its tyrosine 307, as well as the overexpression of endogenous inhibitors such as SET, have been reported as major contributing alterations to inhibit PP2A in human cancer. Thus, we analyzed both pPPP2CAand SET levels in 128 TNBC cases from our cohort with enough material available. High pPPP2CAexpression was found in 31 out of 128 cases (24.2%), whereas 17 out of 128 cases (13.3%) showed SET overexpression. Interestingly, we found that high pBRD4 expression was strongly associated with both PP2A hyperphosphorylation (*p* < 0.001) and SET overexpression (*p* < 0.001) (Table 3), which highlights that PP2A inhibition could be a key molecular mechanism to maintain BRD4 phosphorylation in TNBC.

Since pPPP2CA and SET have been described to be associated alterations, we analyzed how many patients had a concomitant PP2A hyperphosphorylation and SET overexpression. As expected, we observed a significant correlation between both alterations (*p* < 0.001), which were found in 12 cases from our series (Table S2). Moreover, we also analyzed the prognostic value of pPPP2CA in our patient cohort. As expected, those patients with high pPPP2CA expression levels showed a significantly worse OS (*p* < 0.001) and EFS (*p* < 0.001) (Figure S2).


**Table 3.** Association between pBRD4 expression and PP2A activation status in TNBC patients.

#### **4. Discussion**

The TNBC subtype has been previously reported to be particularly sensitive to the treatment with bromodomain inhibitors. In addition, BRD4 hyperphosphorylation has been defined as a molecular alteration that promotes resistance to BRD4 inhibitors, and the tumor suppressor PP2A as the major regulator of BRD4 dephosphorylation. However, the potential clinical impact of this pBRD4 together with the validation of its linking with the PP2A activation status remain to be fully clarified in TNBC patients. It has been recently reported that BRD4 expression is significantly higher in breast cancer tissues than in normal controls, and defines poor prognosis in breast cancer patients [36]. These results would further strengthen our findings in the present study, especially considering that BRD4 phosphorylation has been described as an alteration involved in BRD4 protein stabilization [21]. Moreover, we observed that the prognostic impact of pBRD4 was particularly evident in stage III TNBC patients (Figure S1). This observation, together with the fact that this alteration is associated with recurrence (Table 1), would suggest that BRD4 hyperphosphorylation could be an event with functional relevance in TNBC progression. Thus, its evaluation in a TNBC cohort with metastatic disease would be of high interest in forthcoming studies.

The fact that decreased PP2A activity has been described to induce in vitro BRD4 hyperphosphorylation and resistance to BRD4 inhibition [21] prompted us to analyze the PP2A activation status in our patient cohort. PP2A is a key tumor suppressor commonly deregulated in human cancer [37]. PP2A hyperphosphorylation, as well as upregulation of the endogenous PP2A inhibitors such as SET, has been reported as main molecular mechanisms of PP2A inhibition in many tumors including breast cancer. These alterations have progressively emerged as promising therapeutic targets in this disease [38–44]. Although it has been recently reported that PP2A inhibition is a frequent alteration in breast cancer related with poor outcome and therapy resistance, such studies have been carried out in cohorts including cases with different molecular subtypes [40,45,46]. Therefore, the evaluation of the precise PP2A status in a cohort of TNBC patients as well as its clinical impact in this breast cancer subtype remains still to be performed. Previous works have shown that the PP2A inhibitor CIP2A confers poor outcome in TNBC cells, which has been recently confirmed in the work by Tawab Osman and co-workers [47–49]. These findings would suggest that PP2A inhibition could be of relevance in this breast cancer subtype. In fact, we found in this work that high pPPP2CA were predictor of poor outcome in our TNBC patient cohort (Figure S2). We observed PP2A hyperphosphorylation in 24.2% of cases (31/128) and SET overexpression in 13.3% of cases (17/128). Both alterations were present in 12 patients from our cohort, indicating that 5 patients had SET overexpression without high pPPP2CAexpression, and 19 cases only showed high pPPP2CAlevels. Thus, 82.9% of cases (34/41) with BRD4 hyperphosphorylated had at least one of the PP2A inhibitory markers altered. Therefore, our results suggest that both PP2A hyperphosphorylation and SET overexpression could be molecular contributing alterations to enhance BRD4 phosphorylation levels in TNBC, but it remains to be experimentally confirmed. Only 2 out of 31 cases with high pPPP2CA had low pBRD4 expression. However, the observation that 7 pBRD4 overexpressing patients without any PP2A inhibitory alteration detected would also indicate the potential existence of alternative PP2A inhibitory alterations or molecular mechanisms distinct that PP2A inhibition that deregulate pBRD4 in this disease. Altogether, these results are in concordance with the conclusions reported by Shu and co-workers [21] identifying PP2A as the phosphatase responsible of dephosphorylating BRD4. However, they did not observe prognostic value for pBRD4 and discrepancies in clinical impact may be due to sample size and the fact that those authors stratified their cohort by pBRD4 expression using a median split of pBRD4 intensity.

Furthermore, these findings are of therapeutic relevance, since the use of PP2A activators could serve to overcome a foreseeable development of resistance to BRD4 inhibitors in TNBC patients with high pBRD4 levels. In fact, Shu and co-workers showed that the combination of the PP2A activator perphenazine with JQ1 served to overcome resistance to BRD4 inhibitors in TNBC cells [21]. In this line of thinking, FTY720 is an FDA-approved immunosuppressant used to treat multiple sclerosis, which has shown potent antitumor effects in many tumor types [50]. Moreover, FTY720 has been described as a PP2A activating drug through targeting pPPP2CA and SET, which are the PP2A inhibitory alterations reported in this work. Another relevant issue is the fact that BRD4 is expressed in two major isoforms, short and long, that have been reported to play opposite functions as regulators of gene transcription and tumor progression [51]. The antibody used in our work recognizes phosphorylation on S484/488, which is a region present in both BRD4 isoforms. Therefore, we analyzed here by IHC the total levels of pBRD4 expression, corresponding to the contribution of the long and short BRD4 isoforms. However, it would be of high interest to investigate the potential functional and clinical implications derived from the phosphorylation of each BRD4 isoform separately. Altogether, our results show that high pBRD4 levels define a subgroup of TNBC cases with very poor outcomes. Moreover, our findings are consistent with PP2A inhibition as a key molecular mechanism to induce BRD4 hyperphosphorylation in TNBC patients, which could benefit from a future inclusion of PP2A activators and BRD4 inhibitors in clinical protocols. Moreover, it would be of high interest to study the potential benefit derived from the clinical use of PP2A activators to anticipate and overcome the development of resistance to BRD4 inhibition in TNBC.

#### **5. Conclusions**

In conclusion, BRD4 hyperphosphorylation is a frequent alteration that associates with patient recurrence and independently predicts shorter OS and EFS in TNBC patients. Moreover, we observe a molecular background based on PP2A inhibition as the potential molecular mechanism that contributes to enhanced pBRD4 levels. Altogether, our findings highlight the clinical impact of pBRD4, as well as the PP2A/pBRD4 signaling axis as a novel therapeutic target in TNBC, which needs to be fully confirmed in forthcoming studies.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/2072-669 4/13/6/1246/s1, Figure S1: Clinical impact of pBRD4 in the cohort of 132 TNBC patients stratified by stage in (A) OS and (B) EFS, Figure S2: Clinical impact of pPPP2CA in the cohort of 132 TNBC patients in (A) OS and (B) EFS, Table S1: Clinical and molecular characteristics in a series of 132 TNBC patients, Table S2: Association between SET and pPPP2CA expression levels in TNBC.

**Author Contributions:** Conceptualization, I.C., J.G.-F. and F.R.; methodology, M.S.-Á., I.C., A.S., S.Z. and M.L.; software, M.S.-Á., I.C. and C.C.; formal analysis, M.S.-Á., I.C. and M.L.; investigation, M.S.-Á., I.C., A.S., M.L.; C.-M.C., writing—original draft preparation, M.S.-Á., I.C. and C.C.; writing review and editing, J.M.-G., P.E., J.A. and F.R.; funding acquisition, F.R. and J.G.-F. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by PI18/00382 and PI16/01468 grants from "Instituto de Salud Carlos III FEDER". M.S-A. is supported by "Fundación Conchita Rábago de Jiménez Díaz".

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board of Fundación Jiménez Díaz University Hospital (ref. PIC 13-2016).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** Data sharing is not applicable for this article.

**Acknowledgments:** We especially thank the Fundación Jiménez Díaz Biobank for their exceptional work in sample collection and organization.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Article* **The Novel Oral mTORC1/2 Inhibitor TAK-228 Reverses Trastuzumab Resistance in HER2-Positive Breast Cancer Models**

**Marta Sanz-Álvarez 1,† , Ester Martín-Aparicio 1,† , Melani Luque 1 , Sandra Zazo 1 , Javier Martínez-Useros 2 , Pilar Eroles 3,4, Ana Rovira 5,6 , Joan Albanell 5,6,7 , Juan Madoz-Gúrpide 1, \* and Federico Rojo 1, \***


**Simple Summary:** Hyperactivation of the PI3K/AKT/mTOR cell signalling pathway is an important and well-described mechanism of trastuzumab resistance in HER2-positive breast cancer. In cell-line models of acquired trastuzumab resistance generated in our laboratory, we demonstrate this type of activation, which is independent of HER2-mediated regulation. We investigate whether the use of specific mTOR inhibitors, a PI3K/AKT/mTOR pathway effector, could lead to decreased activity of the pathway, influencing trastuzumab resistance. We demonstrate that TAK-228, a mTORC1 and mTORC2 inhibitor, can reverse resistance and increasing response to trastuzumab in models of primary and acquired resistance.

**Abstract:** The use of anti-HER2 therapies has significantly improved clinical outcome in patients with HER2-positive breast cancer, yet a substantial proportion of patients acquire resistance after a period of treatment. The PI3K/AKT/mTOR pathway is a good target for drug development, due to its involvement in HER2-mediated signalling and in the emergence of resistance to anti-HER2 therapies, such as trastuzumab. This study evaluates the activity of three different PI3K/AKT/mTOR inhibitors, i.e., BEZ235, everolimus and TAK-228 in vitro, in a panel of HER2-positive breast cancer cell lines with primary and acquired resistance to trastuzumab. We assess the antiproliferative effect and PI3K/AKT/mTOR inhibitory capability of BEZ235, everolimus and TAK-228 alone, and in combination with trastuzumab. Dual blockade with trastuzumab and TAK-228 was superior in reversing the acquired resistance in all the cell lines. Subsequently, we analyse the effects of TAK-228 in combination with trastuzumab on the cell cycle and found a significant increase in G0/G1 arrest in most cell lines. Likewise, the combination of both drugs induced a significant increase in apoptosis. Collectively, these experiments support the combination of trastuzumab with PI3K/AKT/mTOR inhibitors as a potential strategy for inhibiting the proliferation of HER2-positive breast cancer cell lines that show resistance to trastuzumab.

**Keywords:** breast cancer; resistance; anti-receptor therapy; trastuzumab; PI3K; mTOR; TAK-228

**Citation:** Sanz-Álvarez, M.; Martín-Aparicio, E.; Luque, M.; Zazo, S.; Martínez-Useros, J.; Eroles, P.; Rovira, A.; Albanell, J.; Madoz-Gúrpide, J.; Rojo, F. The Novel Oral mTORC1/2 Inhibitor TAK-228 Reverses Trastuzumab Resistance in HER2-Positive Breast Cancer Models. *Cancers* **2021**, *13*, 2778. https:// doi.org/10.3390/cancers13112778

Academic Editor: Giuseppe Curigliano

Received: 26 April 2021 Accepted: 29 May 2021 Published: 3 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

#### **1. Introduction**

Despite ongoing advances in understanding diagnosis and treatment, breast cancer continues to place an enormous burden on healthcare systems worldwide and poses a risk to the lives of many patients. Breast cancer is the second leading cause of cancer deaths among women worldwide, representing 30% of all new cancer diagnoses: More than 2.25 million new cases and around 700,000 deaths were estimated in 2020 [1]. Breast cancer is a heterogeneous disease comprising four major subtypes, each with distinct pathological features and clinical implications [2]. Among those subgroups, HER2-positive breast cancer accounts for 25% of all cases and is associated with high relapse rates and poor prognosis [3,4]. This subtype is characterised by amplifying the *ERBB2/neu* oncogene and/or overexpression of its associated HER2 tyrosine kinase receptor [5]. Despite the absence of a ligand for this transmembrane receptor, HER2 forms homodimers or heterodimers with other HER family members, activating different downstream signalling pathways, including MAPK and PI3K/AKT/mTOR, which ultimately regulate processes, such as cell survival, proliferation, motility and metabolism [6,7]. In 1998, the advent of trastuzumab, the first targeted anti-HER2 therapy and humanised monoclonal antibody against HER2, brought about considerable improvement in the prognosis of metastatic and early-stage HER2-positive breast cancer patients [8,9]. In spite of the efficacy demonstrated by trastuzumab, both alone and in combination with chemotherapy as first-line treatment, primary or acquired resistance emerges within a few months after the start of treatment, and resistance remains one of the main problems in managing these patients [8,10]. Several mechanisms of resistance to trastuzumab have been described in recent decades, such as the expression of splicing variants like p95HER2 [11], heterodimerisation with other RTKs [12–14], Src activation [15] and aberrant activation of the PI3K signalling pathway, most commonly through mutations in PIK3CA and loss of PTEN [16,17]. The intertwining of HER2-mediated signalling and the PI3K pathway takes the form, at the molecular level, that signalling by the HER family is primarily mediated through the PI3K and MAPK cascades [18,19]. As a result, the PI3K/AKT/mTOR signalling pathway has been implicated in the anti-HER2 response [17,20,21], and targeting the PI3K/AKT/mTOR pathway has proven to be a valuable strategy to overcome resistance to HER2-directed therapy [22].

Due to the involvement of the PI3K pathway in both HER2-mediated signalling and in the emergence of resistance to trastuzumab, this network becomes a good target for drug development. Because inhibition of the PI3K/AKT/mTOR axis results in enhanced HER2 signalling in HER2-overexpressing breast cancer, especially increased expression of HER2 and HER3 [23], targeting both pathways could prevent the development of resistance. However, the clonal evolution of cancer itself causes genetic and molecular diversity in patients' tumours that manifests as long-recognised functional and phenotypic heterogeneity. It is, therefore, unclear whether, in a HER2-positive breast cancer subtype scheme, such a therapeutic combination will be effective in different scenarios characterised by small molecular variations, this despite previously published reports in the scientific literature. As reported elsewhere [24], our laboratory generated and characterised several cellular models of trastuzumab-resistant HER2-positive breast cancer lines, covering, albeit to a limited extent, a range of genetic heterogeneity. Moreover, several drugs that are effective against different nodes of the PI3K/AKT/mTOR signalling pathway are available, namely, BEZ235, everolimus, and TAK-228. Different preclinical studies have demonstrated the efficacy of combining trastuzumab with different PI3K/AKT/mTOR inhibitors. For instance, BEZ235, a dual pan-class I PI3K and mTOR kinase inhibitor, has shown antitumor activity in vitro and in vivo in breast cancer models that harbour PI3KCA mutations [25] or are resistant to anti-HER2 therapies [26]. In murine models of HER2-positive mammary tumours, combined therapy with trastuzumab and everolimus, an allosteric mTORC1 inhibitor, obtained better results than either agent alone [27]. Furthermore, in a resistance model generated by the loss of PTEN, trastuzumab combined with everolimus restored sensitivity to trastuzumab and showed greater efficacy than either agent independently [28]. TAK-228 is an ATP-competitive inhibitor that targets both mTORC1 and mTORC2. TAK-

228 has shown efficacy in different preclinical models of breast cancer [29,30]. The aim of our study was to evaluate the efficacy of three different mTOR inhibitors in in vitro models of trastuzumab-resistant breast cancer cells to assess their potential use in both primary resistance and the development of acquired resistance. We show that trastuzumab, in combination with mTOR inhibitors, exerts an antiproliferative effect by inducing alterations in the PI3K/AKT/mTOR and ERK pathways, as well as through the induction of apoptosis and cell cycle arrest in different models of trastuzumab resistance. Our data suggest a potential benefit of using mTOR inhibitors in combination with trastuzumab in acquired resistance.

#### **2. Materials and Methods**

#### *2.1. Cell Lines*

The effects of trastuzumab on cell growth were studied in a panel of eleven HER2 amplified breast cancer cell lines, including four trastuzumab-conditioned cell lines selected for long-term outgrowth in trastuzumab-containing medium. BT-474 (HTB-20) ductal carcinoma, SK-BR-3 (HTB-30) and AU-565 (CRL-2351) adenocarcinoma, as well as HCC1419 (CRL-2326) and HCC1954 (CRL-2338) ductal carcinoma cell lines, were obtained from the American Type Culture Collection. EFM-192A (ACC-258) and JIMT-1 (ACC-589) ductal carcinoma cells were obtained from the German Tissue Repository DSMZ. Trastuzumabresistant BT-474.rT3, SK-BR-3.rT1, AU-565.rT2 and EFM-192A.rT1 cell lines were generated as previously described [24]. BT-474, SK-BR-3 and JIMT-1 cells were maintained in DMEM-F12 supplemented with 10% heat-inactivated foetal bovine serum (FBS), 2 mmol/L glutamine, and 1% penicillin G-streptomycin. AU-565, HCC1419 and HCC1954 cells were cultured in RPMI 1640 supplemented with 10% heat-inactivated FBS, 2 mmol/L glutamine, and 1% PSF. EFM-192A cells were grown in RPMI 1640 medium supplemented with 20% heat-inactivated FBS, 2 mmol/L glutamine, and 1% PSF. Cells were maintained at 37 ◦C with 5% CO2. All cell lines were checked for authentication every 6 months, either by using the Cell Line Authentication service at LGC Standards (UK) (tracking no: 710259498; 710274855; 710281607; 710272355), or by running a home-made mutational profiling assay.

#### *2.2. Reagents*

The recombinant humanised monoclonal HER2 antibody trastuzumab (a concentration of 15 µg/mL was selected as indicated elsewhere [24]) (Herceptin, Genentech, San Francisco, CA, United States) was supplied by the pharmacy of our hospital; BEZ235 (S1009), everolimus (S1120) and TAK-228 (S2811) were obtained from Selleckchem (Selleckchem Spain, Madrid, Spain).

#### *2.3. Determination of the Resistance Rate*

Establishment of drug resistance was confirmed by cell proliferation assay, as determined in P100 plates containing 5 × 10<sup>5</sup> cells for each condition (sensitive and resistant), grown both in the absence and in the presence of trastuzumab for 7 days. The results were processed using the algorithm described by O'Brien, which correlates the rate of growth between the treated and nontreated cells, reflecting the doubling time of the cells [31]. Once resistance was confirmed, cells were maintained in the absence of treatment for 30 days. After this pause, resistance was reconfirmed using the same protocol. Resistant cell lines populations were maintained with 15 µg/mL of trastuzumab in the medium for months. Periodically, vials of both the sensitive (parental) and resistant cell populations (pools and clones) were stored in liquid nitrogen to keep a stock of young cells.

#### *2.4. Cell Proliferation Assays*

Cells were seeded in triplicate in P100 plates at a density of 5 × 10<sup>5</sup> cells per plate and allowed to adhere and enter the growth phase before being treated with or without 15 µg/mL trastuzumab for 7 days in the appropriate culture medium. Cells were then harvested by trypsinisation and counted with trypan blue using the TC20 Automated Cell

Counter (BioRad, Hercules, CA, USA). The appropriate culture media and trastuzumab were replaced every 3 days. All experiments were repeated three times with readings at least in triplicate for each concentration.

#### *2.5. Determination of IC50*

To determine the IC50 of the mTOR inhibitors, a panel of HER2-positive breast cancer cell lines was treated with escalating concentrations of BEZ235, everolimus and TAK-228. Proliferation was measured by counting after 7 days of treatment. Viable cells were counted by trypan blue exclusion. IC50 (half the maximal inhibitory concentration) was calculated using SigmaPlot software. Values are mean IC50 from three independent experiments.

#### *2.6. Protein Extraction and Quantification*

Cells were washed with 3 mL PBS at RT. Next, cells were scraped in the presence of 150 µL lysis buffer (RIPA, peptidase inhibitor, phosphatase inhibitor) at 4 ◦C and transferred to a 1.5-mL tube. Cells were incubated in lysis buffer for 20 min at 4 ◦C and sonicated afterwards. Then the cell lysate was spun at 13,000× *g* for 10 min at 4 ◦C, and the supernatant was retained and stored. Protein extracts were quantified using the Pierce BCA protein assay kit (Thermo Fisher Scientific, Whaltman, MA, USA), following the manufacturer's instructions.

#### *2.7. Western Blotting (WB)*

Protein aliquots were prepared at 1 µg/µL in 4× Laemmli loading buffer and boiled at 95 ◦C for 6 min. Twenty µL of protein extract was loaded in a 10% polyacrylamide gel (SDS-PAGE). Next, proteins were transferred to a nitrocellulose membrane for 1 h at 100 V and 4 ◦C. The membrane was blocked (5% milk in TBST 1×) for 1 h, washed 3 times for 10 min and then incubated with the primary antibody at RT overnight under agitation. The concentrations used were as follows: HER3 (1:500; Thermo Scientific), p-HER3 Tyr1197 (1:1000), HER2 (1:500), p-HER2 Tyr1221/1222 (1:1000), AKT (1:1000), p-AKT Thr308 (1:300), p-AKT Ser473 (1:500), p44/42 MAPK (ERK1/2) (1:1000), p-p44/42 MAPK (ERK1/2) Thr202/Tyr204 (1:1000), 4E-BP1 (1:500); p-4E-BP1 Thr37/46 (1:500); p-4E-BP1 Thr70 (1:500); S6 ribosomal protein (S6) (1:500); p-S6 ribosomal protein (p-S6) Ser235/236 (1:1000) (Cell Signaling, Danvers, MA, USA) and GAPDH (1:5000; Sigma-Aldrich, St. Louis, MO, USA). All primary antibodies were rabbit, except the anti-HER3, which was mouse; all were monoclonal. Then the membranes were washed 3 × 10 min in TBST and incubated with a secondary antibody (diluted in 2.5% BSA in TBS 1×) at RT for 1 h. ECL-anti-mouse and ECL-anti-rabbit secondary antibodies attached to peroxidase (HRP; GE Healthcare, Chicago, IL, USA) were used at a concentration of 1:5000. The membranes were washed 3 × 10 min again, and immeserd in the detection reagent (ECL or ECL Prime, if applicable; Amersham, GE Healthcare) for 1 min, prior to developing on a photographic film. Densitometry and quantification of proteins were carried out using ImageJ software.

#### *2.8. Flow Cytometric Determination of Cell Cycle Arrest and Apoptosis*

Before carrying out cell cycle detection and apoptosis, cell lines were synchronised by serum starvation for 24 h. Cell cycle and apoptosis were analysed after treatment with either vehicle (i.e., trastuzumab 15 µg/mL, TAK-228 0.5 µM or both) for 24 and 72 h, respectively. For cell cycle arrest analysis, cells were collected after treatment, washed with PBS and fixed with 70% cold ethanol at –20 ◦C for at least 2 h. Cells were incubated with 0.5 mg/mL RNase (Sigma-Aldrich) at 37 ◦C for 30 min, and finally stained with propidium iodide (BD Biosciences, Franklin Lakes, NJ, USA) for 10 min. Apoptosis was assessed with the Annexin-V-FITC Apoptosis Detection Kit (BD Biosciences) according to the manufacturer's instructions. Flow cytometry was performed on a FACS Canto II (BD Biosciences), and data were analysed with FACS Diva software (BD Biosciences).

#### *2.9. Statistical Analysis*

All data are expressed as means ± standard deviations for at least three replicates (unless otherwise indicated). Statistical significance was analysed by a two-tailed Student's *t*-test (\*: *p* < 0.05, \*\*: *p* < 0.01, \*\*\*: *p* < 0.001). This work was performed in accordance with the Reporting Recommendations for Tumour Marker Prognostic Studies (REMARK) guidelines [32].

#### **3. Results**

#### *3.1. Development and Characterisation of a Panel of Breast Cancer Cell-Line Models of Acquired Trastuzumab Resistance*

To test the efficacy of a combination of HER2 blockade with mTOR inhibition as a potential therapeutic strategy to overcome resistance to trastuzumab in HER2-positive breast cancer cell line (BCCL) models, we first developed four different cellular models with acquired resistance to trastuzumab [24]. Briefly, we used prolonged exposure to moderate doses of the drug to generate novel BCCLs with acquired resistance to trastuzumab, authenticated them based on their molecular profile and their resistance rate was determined. We selected clones for each of the BCCLs and screened them for trastuzumab sensitivity after seven days of treatment (Figure 1). We observed that in all cases, resistant cells showed a higher growth rate in the presence of the drug than the parental sensitive cells. The biochemical analysis of the status of kinase receptors and effectors from different cellular pathways actionable by HER2 signalling revealed differences in phosphorylation levels for several targets between sensitive and resistant lines (Figure S1), as we reported previously [24]. After treatment with trastuzumab, changes occurred in the phosphorylation levels of HER2, AKT (Thr308 and Ser473), ERK1/2, and S6, with more relevant changes between sensitive and resistant populations in the BT474 and AU565 cell lines. This finding was consistent with patterns of molecular alterations commonly described in breast cancer [25]. De novo trastuzumab-resistant cell lines HCC1419, HCC1954 and JIMT-1 were also examined for biochemical changes in the HER2 and PI3K/AKT/mTOR pathways (Figure S2). The most notable signal was the abundant expression of 4E-BP1 in both cell lines, which does not appear to translate into strong activation in either case. Phosphorylation levels of S6 were not elevated either. On the other hand, we observed a slight decrease in AKT phosphorylation levels in the JIMT-1 cell line compared to HCC1954. Overall, the two lines do not show phosphorylation activation signals for either of the two pathways studied.

≥ ≤ **Figure 1.** Characterisation of a panel of cell-line models of acquired trastuzumab resistance. Effect of trastuzumab treatment on sensitive and resistant cells. Proliferation was measured after seven days of treatment by trypan blue exclusion. T: Trastuzumab 15 µg/mL. Data are expressed as mean ± SD from ≥ three independent experiments. \*\*\* denotes *p* ≤ 0.001.

#### *3.2. Effect of Anti-HER2 and MTORC1/2 Treatments on HER2-Positive Breast Cancer Cell Lines (Determination of IC50)*

To determine the effects of BEZ235, everolimus and TAK-228 on the inhibition of the PI3K/AKT/mTOR proliferation axis in HER2-positive cells, the panel of eleven cell lines with varying sensitivity to trastuzumab was treated with increasing inhibitor concentrations. After seven days of treatment, cellular proliferation was measured to determine the IC50 for each drug and cell line (Figure S3). In general, a similar sensitivity was observed in all cell lines for every drug, so when treated with any of the three mTOR inhibitors, the proliferation of the eleven cell lines was significantly inhibited at low nanomolar ranges. The determination of sensitivity to BEZ235 showed that all cell lines behaved very similarly when exposed to the treatment, and only the SK-BR-3.rT1 line was more sensitive to this drug than its parental line. The everolimus sensitivity study showed that all the lines were sensitive to treatment at high concentrations. In addition, JIMT-1 was very sensitive to this drug, decreasing its cell proliferation by more than 50% at 1 nM everolimus, and AU-565.rT2 was also found to be more sensitive to treatment than its sensitive parental line. Finally, treatment with TAK-228 showed highly similar sensitivity to treatment in all lines, both trastuzumab-sensitive and trastuzumab-resistant. Based on these results, the IC50 was calculated for each of the lines and for each drug (Table 1). Notably, the IC50 value of everolimus was more heterogeneous between cell lines than the IC50 values of the other two drugs. In addition, the IC50 values of BEZ235 and TAK-228 between the resistant lines and their parents were very similar, though this was not the case for everolimus in AU-565.rT2 and EFM-192A.rT1, which had a significantly lower IC50 value than their respective parental cell lines. The exceptions were the effect of everolimus in HCC1419 and particularly in JIMT-1, which showed at least a 10× increased sensitivity with respect to the other cells. This is probably because different mutations in nodes of the PI3K/AKT/mTOR pathway make some cell lines more sensitive to everolimus than others, which turn out to be more resistant [33].


**Table 1.** Inhibitory concentrations of mTOR inhibitors as a measure of proliferation inhibition in a panel of breast cancer cell lines.

Note: A panel of HER2-positive breast cancer cell lines was treated with escalating concentrations of BEZ235, everolimus and TAK-228. Proliferation was measured by counting cells after seven days of treatment. Viable cells were counted by trypan blue exclusion. IC50 (half-maximal effective concentration) was calculated using the SigmaPlot software. Values are mean IC50 from three independent experiments. BT-474: BT-474 trastuzumabsensitive cells. BT-474.rT3: BT-474 trastuzumab-resistant cells. SKBR3: SKBR3 trastuzumab-sensitive cells. SK-BR-3.rT1: SK-BR-3 trastuzumab-resistant cells. AU-565: AU-565 trastuzumab-sensitive cells. AU-565.rT2: AU-565 trastuzumab-resistant cells. EFM-192A: EFM-192A trastuzumab-sensitive cells. EFM-192A.rT1: EFM192A trastuzumab-resistant cells.

In view of these results, we considered that combining anti-HER2 therapy with each of these mTOR inhibitors might show a greater antiproliferative effect. For therapeutic studies, the concentration and time of treatments were based on previous reports, and administered as follows: Trastuzumab (15 µg/mL) [15]; BEZ235 (1 nM, 5 nM and 20 nM) [34], everolimus (0.5 nM and 1 nM) [35] and TAK-228 (1 nM and 5 nM) [30].

#### *3.3. Combined Treatment of Trastuzumab and MTORC1/C2 Inhibitor TAK-228 in HER2-Positive Breast Cancer Cell Lines with Acquired Resistance to Trastuzumab*

In order to assess the potential synergistic effects of trastuzumab in combination with mTOR inhibitors, we performed viability assays in the four sensitive cell lines, as well as their correspondent resistant models. Overall, the combination of trastuzumab with BEZ235 or everolimus influenced the therapeutic response to a lesser degree than the combination treatment of trastuzumab with TAK-228 because, although it causes a reduction of mTOR activation in the cell lines, cell viability was not affected. In contrast, the combination of trastuzumab with TAK-228 significantly increased the therapeutic response in all cases, suggesting that the decreased mTOR activation status by TAK-228 affects sensitivity to trastuzumab.

The treatment effect of the TAK-228 inhibitor was evaluated using two treatment concentrations (1 nM and 5 nM), as monotherapy and in combination with trastuzumab (Figure 2). A single treatment with TAK-228 showed no effect on cell proliferation in any of the cell lines for either of the two concentrations used. Combination treatment with trastuzumab and TAK-228 5 nM resulted in the reversal of acquired resistance in all lines. BT-474.rT3 cells showed a highly significant decrease in proliferation in the trastuzumab and TAK-228 condition (52%) compared to trastuzumab (84%, *p*-value < 0.01) and TAK-228 (67%, *p*-value < 0.01). In addition, a significant decrease in proliferation was also observed in trastuzumab with TAK-228 1 nM combination therapy (77% vs. 84% for trastuzumab and vs. 102% for TAK-228 1 nM, *p*-value < 0.05). In the SK-BR-3.rT1 line, the combination of trastuzumab and TAK-228 5 nM decreased growth very significantly (44%) compared to treatment with trastuzumab (96%) and TAK-228 (77%, *p*-value < 0.001). The same effect was observed in the AU-565.rT2 line, with reduced proliferation in combination therapy (64% vs. 100% for trastuzumab, and 84% for TAK-228, *p*-value < 0.001). Finally, in EFM-192A.rT1, a significant decrease in proliferation was identified trastuzumab plus TAK-228 5 nM combined therapy compared to individual treatments (65%, *p*-value < 0.01).

≥ ≤ ≤ ≤ **Figure 2.** Decrease in mTOR activation status by TAK-228 affects trastuzumab sensitivity. Sensitive and trastuzumabresistant cells were treated for seven days with DMSO, 15 µg/mL trastuzumab (T), 1 or 5 nM TAK-228 (TAK), or a combination of 15 µg/mL trastuzumab plus 1 or 5 nM TAK-228. Viable cells were then counted by trypan blue exclusion. Viability is presented as a percentage of the DMSO-treated control vector group. Error bars represent standard deviation between replicates (*n* ≥ 3). \* denotes *p* ≤ 0.05, \*\* denotes *p* ≤ 0.01 and \*\*\* denotes *p* ≤ 0.001. (**A**) BT-474 sensitive (BT474) and trastuzumab-resistant (BT-474.rT3) cells. (**B**) SK-BR-3 sensitive (SK-BR-3) and trastuzumab-resistant (SK-BR-3.rT1) cells. (**C**) AU-565 sensitive (AU565) and trastuzumab-resistant (AU-565.rT2) cells. (**D**) EFM-192A sensitive (EFM-192A) and trastuzumab-resistant (EFM-192A.rT1) cells.

To test the effect of BEZ235 in combination with trastuzumab on cell proliferation, three concentrations of the drug (1 nM, 5 nM and 20 nM) were selected, all below the IC50 value for all lines. The effect on cell proliferation was assessed in the four trastuzumabsensitive and trastuzumab-acquired resistance lines (Figure 3). Using a BEZ235 concentration of 20 nM, a significant decrease in proliferation was observed in BT-474.rT3 (19%, *p*-value < 0.001) and EFM-192A.rT1 (30%, *p*-value < 0.001) compared to control and trastuzumab treatment conditions. Furthermore, in BT-474.rT3, the combined treatment of BEZ235 plus trastuzumab significantly reversed trastuzumab resistance compared to the trastuzumab treatment condition (45%, *p*-value < 0.001). In sensitive cell lines, trastuzumab combined with BEZ235 20 nM potentiated the effect of trastuzumab individually, with no significant effect.

≥ ≤ ≤ ≤ **Figure 3.** Effect of blocking mTOR activation by BEZ235 on trastuzumab sensitivity in trastuzumab-sensitive and trastuzumab-acquired resistance cell lines. Sensitive and trastuzumab-resistant cells were treated for seven days with DMSO, 15 µg/mL trastuzumab (T), 1, 5 or 20 nM BEZ235 (**B**), or a combination of 15 µg/mL trastuzumab plus 1, 5 or 20 nM BEZ235. Viable cells were then counted by trypan blue exclusion. Viability is presented as a percentage of the DMSO-treated control vector group. Error bars represent standard deviation between replicates (*n* ≥ 2). \* denotes *p* ≤ 0.05, \*\* denotes *p* ≤ 0.01 and \*\*\* denotes *p* ≤ 0.001. (**A**) BT-474 sensitive (BT-474) and trastuzumab-resistant (BT-474.rT3) cells. (**B**) SK-BR-3 sensitive (SK-BR-3) and trastuzumab-resistant (SK-BR-3.rT1) cells. (**C**) AU-565 sensitive (AU-565) and trastuzumab-resistant (AU-565.rT2) cells. (**D**) EFM-192A sensitive (EFM-192A) and trastuzumab-resistant (EFM-192A.rT1) cells.

Two concentrations of everolimus (0.5 nM and 1 nM) were selected below the IC50 value in all cell lines (Figure 4). Treatment with either concentration of the drug alone had no effect on cell proliferation in any of the sensitive or acquired-resistant lines. Combination therapy of trastuzumab with 0.5 nM everolimus showed only slightly stronger effects than trastuzumab alone on proliferation in most cell lines, both sensitive and resistant. However, in the combined condition consisting of trastuzumab and everolimus 1 nM, a reversal of trastuzumab resistance was observed, very significantly decreasing proliferation in the BT-474.rT3 (20%, *p*-value = 0.003) and EFM-192A.rT1 (42%, *p*-value = 0.005) lines, compared to trastuzumab-alone treatment. Furthermore, this treatment combination enhanced the effect of trastuzumab in the four sensitive lines (i.e., BT-474 (13%), SK-BR-3 (26%), AU-565 (35%) and EFM-192 A (31%)), decreasing their proliferation compared to the single-treatment conditions, without being statistically significant (Figure 4).

≥ ≤ ≤ ≤ **Figure 4.** Effect of blocking mTOR activation by everolimus on trastuzumab sensitivity in trastuzumab-sensitive and trastuzumab-acquired resistance cell lines. Sensitive and trastuzumab-resistant cells were treated for seven days with DMSO, 15 µg/mL trastuzumab (T), 0.5 or 1 nM everolimus (E), or a combination of 15 µg/mL trastuzumab plus 0.5 or 1 nM everolimus. Viable cells were then counted by trypan blue exclusion. Viability is presented as a percentage of the DMSOtreated control vector group. Error bars represent standard deviation between replicates (*n* ≥ 2). \* denotes *p* ≤ 0.05, \*\* denotes *p* ≤ 0.01 and \*\*\* denotes *p* ≤ 0.001. (**A**) BT-474 sensitive (BT474) and trastuzumab-resistant (BT-474.rT3) cells. (**B**) SK-BR-3 sensitive (SK-BR3) and trastuzumab-resistant (SK-BR-3.rT1) cells. (**C**) AU-565 sensitive (AU-565) and trastuzumab-resistant (AU-565.rT2) cells. (**D**) EFM-192A sensitive (EFM192A) and trastuzumab-resistant (EFM-192A.rT1) cells.

#### *3.4. Potentiation Effect between Trastuzumab and mTORC1/2 Inhibitor TAK-228 in Breast Cancer Cell Lines with Primary Trastuzumab Resistance*

The effects of drug combinations on BCCLs with primary resistance to trastuzumab were markedly dependent on each particular cell line (but less so on the nature of the inhibitor, Figure S4). In HCC1419, the combination of trastuzumab with any of the inhibitors had a greater effect than treatment with the inhibitor alone but was generally not effective with respect to treatment with trastuzumab, possibly because at baseline these cells are somewhat sensitive to trastuzumab. In the case of HCC1954, a significant effect was observed in the combination of trastuzumab with any inhibitor, both with respect to trastuzumab and the inhibitor alone. However, JIMT-1 cells showed minimal response to the different treatments, except for a small decrease in cell proliferation, due to the effect of the combination of trastuzumab with TAK-228.

Treatment with BEZ235 at any of the three concentrations tested in combination with trastuzumab resulted in a significant decrease in proliferation in the HCC1954 line. A 65% decrease in proliferation was observed in the BEZ235 1 nM plus trastuzumab condition compared to the single-treatment conditions (*p*-value < 0.001). In the BEZ235 5 nM plus trastuzumab combination, 16% proliferation was identified compared to BEZ235 5 nM treatment (29%) and trastuzumab treatment (87%), with a significant reduction in proliferation (*p*-value < 0.01). Finally, 8% proliferation was observed in the BEZ235 20 nM plus trastuzumab combination, compared to the single treatments (*p*-value < 0.01). JIMT1 cell proliferation was not affected by any of the treatment conditions.

For everolimus, two concentrations (0.5 nM and 1 nM) were selected below the IC50 value in the cell lines, except in JIMT-1. Its effect on cell proliferation was evaluated in the untreated condition, treatment with trastuzumab 15 µg/mL, everolimus 0.5 nM or 1 nM and the combination of both treatments at the two selected everolimus concentrations (Figure S4). Treatment with everolimus at 0.5 nM demonstrated a significant effect on cell proliferation in the HCC1954 line, in combined treatment with trastuzumab (59%, compared to 84%, *p*-value < 0.01), reversing trastuzumab resistance. In addition, treatment with everolimus 1 nM significantly reduced the proliferation of this line (12%, *p*-value < 0.001). In the JIMT-1 cell line, treatment with everolimus 0.5 nM, both alone and in combination with trastuzumab, showed no effect on cell proliferation, while treatment with everolimus 1 nM resulted in a significant reduction in cell proliferation (44%, *p*-value < 0.01).

The treatment effect of the TAK-228 inhibitor was evaluated using two treatment concentrations, 1 nM and 5 nM, in monotherapy and in combination with trastuzumab. This resistance reversal effect was also observed in the primary resistant line HCC1954. Combination therapy with trastuzumab and 5 nM TAK-228 significantly reduced cell proliferation compared to trastuzumab (57% vs. 85%, *p*-value < 0.001) and TAK-228 (57% vs. 81%, *p*-value < 0.001). Cell proliferation of the JIMT-1 line was not modified by any of the treatment conditions tested.

#### *3.5. Downregulation of PI3K/AKT/mTOR and MAPK Signalling by the Combination of Trastuzumab with TAK-228 in HER2-Positive Breast Cancer Cell Lines*

Since treatment with the inhibitor TAK-228 was shown to reverse trastuzumab resistance in the four cell lines with acquired resistance in combination with trastuzumab, the effect of the combination of both treatments on inhibition of the PI3K/AKT/mTOR pathway was evaluated. The molecular effect of the treatment was assessed by analysing the phosphorylation of the effector proteins of the two mTOR complexes: p-S6 (Ser235/236), p-4E-BP1 (Thr37/46) and p-4E-BP1 (Thr70) of the mTORC1 complex; and p-AKT (Ser473) of the mTORC2 complex, as well as their total forms; in addition, the analysis of the phosphorylated form of ERK was included. Protein expression profiling was performed after 24 h of treatment with trastuzumab 15 µg/mL, or treatment with TAK-228 5 nM, with TAK-228 50 nM, or the combination of trastuzumab plus TAK-228 at the two concentrations above, as well as the control condition.

Combination treatment of trastuzumab with TAK-228 (at either of the two concentrations tested) resulted in a decrease in AKT phosphorylation levels (Ser473) in the BT-474 line, but not in the BT-474.rT3 line (Figure 5A). In both lines, combined treatment with TAK-228 5 nM plus trastuzumab resulted in a significant reduction in p-S6 (Ser235/236) compared to the monotherapy condition, although this reduction was not observed in the two phosphorylated forms of 4E-BP1. In the 50 nM TAK-228 treatment condition, combination with trastuzumab induced disappearance of p-S6 (Ser235/236) and a significant reduction of p-4E-BP1 (Thr37/46 and Thr70) levels in both sensitive and resistant cells. In addition, only in the sensitive line did we observe that TAK-228 combined with trastuzumab resulted in a decrease in the phosphorylated form p-ERK1/2 (Thr202/Tyr204) compared to the levels detected in the single-treatment conditions. Furthermore, the combination of trastuzumab plus TAK-228 5 nM in the sensitive cell line induced a decrease in HER2 phosphorylation levels, while in the BT-474.rT3 line, it was necessary to increase the concentration of the inhibitor to 50 nM (in combination with trastuzumab) to observe the same effect in reduced p-HER2 levels. In both lines, TAK-228 5 nM increased p-HER3, as previously described, and combined treatment with both concentrations of TAK-228 reduced phosphorylation only in the resistant line.

In SK-BR-3 and SK-BR-3.rT1 lines, combined treatment consisting of TAK-228 50 nM and trastuzumab reduced p-AKT levels (Ser473) compared to baseline and trastuzumab treatment, with no change in total form expression (Figure 5B). In both lines, treatment with TAK-228 plus trastuzumab was also found to decrease S6 (Ser235/236) phosphorylation compared to levels detected in the treatment conditions alone. In addition, the 50 nM TAK-228 treatment condition and the trastuzumab combination condition resulted in a highly significant decrease in S6 (Ser235/236) activation, as did the phosphorylated forms of 4E-BP1 (Thr37/46 and Thr70). It is also noteworthy that the total forms of S6 and 4E-BP1 were affected by treatment with TAK-228 50 nM and the combination with trastuzumab. Treatment of both sensitive and resistant cells with TAK-228 alone or in combination with trastuzumab induced an increment in HER2 and HER3 phosphorylation.



**Figure 5.** *Cont.*


**Figure 5.** (**A**) Inhibition of p-S6 (Ser235/236) in trastuzumab-sensitive and -resistant BT-474 cells treated with a combination of trastuzumab and TAK-228. Sensitive and trastuzumab-resistant cells

were treated for 24 h with DMSO, 15 µg/mL trastuzumab (T), 5 and 50 nM TAK-228 (TAK), or a combination of 15 µg/mL trastuzumab plus 5 or 50 nM TAK-228. Whole-cell protein extracts were analysed with the indicated antibodies. Images are representative of three independent experiments. (**B**) Inhibition of p-S6 (Ser235/236) in trastuzumab-sensitive and -resistant SK-BR-3 cells treated with a combination of trastuzumab and TAK-228. Sensitive and trastuzumab-resistant cells were treated for 24 h with DMSO, 15 µg/mL trastuzumab (T), 5 and 50 nM TAK-228 (TAK), or a combination of 15 µg/mL trastuzumab plus 5 or 50 nM TAK-228. Whole-cell protein extracts were analysed with the indicated antibodies. Images are representative of three independent experiments. (**C**) Inhibition of p-S6 (Ser235/236) in trastuzumab-sensitive and -resistant AU-565 cells treated with a combination of trastuzumab and TAK-228. Sensitive and trastuzumab-resistant cells were treated for 24 h with DMSO, 15 µg/mL trastuzumab (T), 5 and 50 nM TAK-228 (TAK), or a combination of 15 µg/mL trastuzumab plus 5 or 50 nM TAK-228. Whole-cell protein extracts were analysed with the indicated antibodies. Images are representative of three independent experiments. (**D**) Inhibition of p-S6 (Ser235/236) in trastuzumab-sensitive and -resistant EFM-192A cells treated with a combination of trastuzumab and TAK-228. Sensitive and trastuzumab-resistant cells were treated for 24 h with DMSO, 15 µg/mL trastuzumab (T), 5 and 50 nM TAK-228 (TAK), or a combination of 15 µg/mL trastuzumab plus 5 or 50 nM TAK-228. Whole-cell protein extracts were analysed with the indicated antibodies. Images are representative of three independent experiments.

In AU-565 and AU-565.rT2 lines, treatment with TAK-228 in combination with trastuzumab resulted in decreased phosphorylation of AKT (Ser473) and S6 (Ser235/236) (Figure 5C). In addition, single TAK-228 treatment lowered the level of p-S6 (Ser235/236) compared to baseline. As in the sensitive and resistant SK-BR-3 lines, the total form of 4E-BP1 decreased in the presence of TAK-228 treatment at either of the two concentrations tested and in combination with trastuzumab, as did the phosphorylated form of 4E-BP1 (Thr37/46). In these lines, the phosphorylation levels of 4E-BP1 (Thr70) are almost undetectable, and no differences between treatment conditions were in evidence. We observed an increase in p-HER2 levels in AU-565 cells when treated with TAK-228 at 5 or 50 nM in combination with trastuzumab. However, TAK-228 50 nM plus trastuzumab in the resistant cell line induced a reduction in phosphorylation levels. Regarding the levels of HER3 phosphorylation, we did not observe a decrease with the different combinatorial treatments in either cell line.

Similarly, in the EFM-192A and EFM-192A.rT1 lines, treatment with TAK-228 at 5 nM and 50 nM and combination with trastuzumab resulted in inhibition of the PI3K/AKT/mTOR pathway (Figure 5D). In both lines, p-AKT (Ser473) levels were found to decrease in the presence of trastuzumab with TAK-228 (at both concentrations) compared to single treatments. In the EFM-192A line, a decrease in p-AKT (Ser473) levels was also observed in the presence of TAK-228 50 nM. In both lines, the combined treatment with TAK-228 50 nM caused a disappearance of p-S6, as well as a decrease in total protein levels. Finally, the EFM19-2A.rT1 line under baseline conditions showed significant activation of p-4E-BP1 (Thr70) compared to its parental line, with very similar levels of total 4E-BP1. Combination treatment with trastuzumab plus TAK-228 50 nM resulted in inhibition of this p-4E-BP1 (Thr70) activation to levels below those of trastuzumab or TAK-228 monotherapy. In addition, as observed in the other cell lines, the levels of the total 4E-BP1 form decreased in the presence of TAK-228 compared to baseline. In the EFM-192A line, as in BT474, combined treatment of trastuzumab with TAK-228 at both concentrations resulted in decreased levels of ERK1/2 (Thr202/Tyr204) phosphorylation compared to levels observed in the single-treatment conditions. In the EFM-192A and EFM-192A.rT1 cells, single or combined treatments did not induce significant changes in HER2 phosphorylation levels. Additionally, we observed an increase in HER3 phosphorylation with the single TAK-228 treatment, though the addition of trastuzumab did not produce a decrease in those levels.

The molecular effect of TAK-228 on the two lines with primary resistance to trastuzumab (i.e., HCC1954 and JIMT-1) was also studied under the treatment conditions mentioned above. In the presence of combined treatment at both concentrations, the HCC1954 line showed a slight decrease in AKT phosphorylation (Ser473) (Figure S5). It was also observed that p-S6 (Ser235/236) was significantly decreased by treatment with TAK-228 at both concentrations, independent of trastuzumab. The same was true for the full form of 4E-BP1 and its phosphorylated form, p-4EBP1 (Thr37/46). In this line, phosphorylation levels of p-4EBP1 (Thr70) were almost undetectable, so no differences between treatments could be assessed. In the JIMT-1 line, only treatment with TAK-228 at either concentration resulted in a trastuzumab-independent decrease in p-S6 (Ser235/236). No changes were observed in 4E-BP1 or its phosphorylated forms, nor in AKT and its phosphorylated form. The original WB images can be found as Supplementary Material (Figure S6).

In summary, the combined treatment decreased the phosphorylation levels of HER2/HER3, diminished PI3K/AKT/mTOR signalling and limited ERK phosphorylation, as a direct consequence of the TAK-228 mechanism of action.

#### *3.6. Cell-Cycle and Apoptosis Analysis in Trastuzumab-Resistant Breast Cancer Cell Lines Treated with Trastuzumab and TAK-228*

The results of resistance reversal obtained in cell proliferation assays with the combination of trastuzumab plus TAK-228 led us to investigate whether the treatment would also have an impact on cell cycle control, as well as on apoptosis induction. We firstly checked cell viability at shorter times, after treatment with trastuzumab in combination with different concentrations of TAK-228, to discard a deleterious effect. Cell cycle arrest was analysed after treatment with trastuzumab, TAK-228 and the combination of both for 24 h, in the cell lines SK-BR-3, AU-565 and EFM-192A, as well as in their corresponding resistant lines, SK-BR-3.rT1, AU-565.rT2 and EFM-192A.rT1 (Figure 6A). We observed a significant increase in the G0/G1 phase signal in SK-BR-3 and SK-BR-3.rT1 cells treated with the mTOR inhibitor alone (*p* = 0.004, *p* = 0.009, respectively), and the combination with trastuzumab improved the cell cycle delay (*p* = 0.004, *p* = 0.006, respectively). AU-565 and EFM-192A lines showed an increase in G0/G1 arrest with trastuzumab (*p* = 0.04, *p* = 0.007, respectively) and TAK-228 (*p* = 0.006, *p* = 0.005, respectively) treatment alone, but the effect was enhanced with the combined treatment (*p* = 0.004, *p* = 0.001, respectively). However, in the corresponding resistant lines AU-565.rT2 and EFM-192A.rT1, only TAK-228 (*p* = 0.001, *p* = 0.03, respectively) and both treatments (*p* = 0.0005, *p* = 0.01, respectively) were able to significantly induce cell cycle arrest. No significant changes were detected in the cell lines BT-474 and BT-474.rT3.

Apoptosis was determined by positive staining with annexin V by flow cytometry, including both early and late apoptosis. We analysed the apoptotic effect of each treatment as a single agent and in combination in the BT-474, BT-474.rT3, SK-BR-3 and SK-BR-3.rT1 cell lines (Figure 6B). In BT474 and BT-474.rT3 we observed a significant increase in cell death with TAK-228 alone (*p* = 0.02, *p* = 0.0001, respectively), but the combination of both drugs (*p* = 0.012, *p* = 0.0001) showed a greater rise in cell death. Furthermore, treatment of BT-474.rT3 cells with trastuzumab alone induced a significant increase in the percentage of apoptotic cells (*p* = 0.03). In SK-BR-3 and SK-BR-3.rT1 cell lines trastuzumab did not significantly affect the percentage of apoptotic cells, though the treatment with TAK-228 (*p* = 0.013, *p* = 0.021) or the combination of the two led to a significant increase in cell death.

≤ ≤ ≤ **Figure 6.** (**A**) Cell cycle arrest induced by trastuzumab and TAK-228 in trastuzumab-sensitive and -resistant cell lines. Cell lines were treated with 15 µg/mL trastuzumab (T), 0.5 µM TAK-228 (TAK) or a combination (T+TAK). Cell cycle arrest was analysed by flow cytometry after 24 h. (**B**) Apoptosis induced by trastuzumab and TAK-228 in trastuzumab sensitive and resistant cell lines. Cell lines were treated with 15 µg/mL trastuzumab (T), 0.5 µM TAK-228 (TAK) or the combination (T+TAK). Apoptosis was measured after 72 h by Annexin V positive staining by flow cytometry. Data are expressed as mean ± SD from three independent experiments. \* denotes *p* ≤ 0.05, \*\* denotes *p* ≤ 0.01 and \*\*\* denotes *p* ≤ 0.001.

#### **4. Discussion**

The development of anti-HER2 targeted therapies to treat patients with HER2-positive breast cancer has proved to be effective in survival in both early and advanced settings. For this reason, trastuzumab has been the standard treatment for HER2-positive breast cancer for more than two decades. Despite this advance, almost all patients eventually experience disease progression on trastuzumab-based therapy, due to de novo or acquired resistance. Aside from alterations in the receptor itself, one mechanism that trastuzumab interferes with HER2 signalling is inhibition of the PI3K/AKT/mTOR signalling pathway [36]. As a logical consequence, among the many causes that have been associated with resistance to anti-HER2 therapies in breast cancer, dysregulations in the signalling of the PI3K/AKT/mTOR pathway seem to play an important role [17,21], as we confirmed in our cellular models of acquired resistance (Figure 1 and Figure S1). As we can see in Figure S1 and as previously reported by our group [24], the acquisition of resistance to trastuzumab in these four HER2-positive breast cancer cell lines was associated with an increase in the amounts of p-ERK, p-AKT and p-S6, suggesting a higher level of activation of their PI3K and MAPK pathways and a plausible association with mechanisms of resistance generation in these cell line models. This finding is consistent with previous reports of a correlation between increased activation of the PI3K/AKT pathway and resistance to trastuzumab [31]. Mechanistically, PI3K activation, followed by AKT activation, triggers the release of mTOR from the mTORC1 complex, which in turn activates the S61 and 4E-BP1 proteins. In addition, the complex itself has a negative feedback mechanism, which inactivates AKT [37]. The mTOR protein also localises to the mTORC2 complex, exhibiting direct AKT-activation capability at the Ser473 residue, leading to AKT and BAD activation [38]. Unlike the mTORC1 complex, the activation of this complex appears to be AKT-independent and controlled by RAS/MAPKs [37,38]. At the same time, it has been previously described that PI3K/AKT/mTOR pathway inhibition may result in the activation of compensatory pathways that could reduce the antiproliferative activity of these inhibitors [23,39–41]. From a clinical point of view, due to the involvement of this pathway in both HER2-mediated signalling and in the emergence of resistance to HER2 targeted therapies, such as trastuzumab, it would therefore be very interesting to consider inhibiting or modulating this pathway. Because inhibition of the PI3K/AKT/mTOR axis results in enhanced HER2 signalling in HER2-overexpressing breast cancer, especially in increased expression of HER2 and HER3 [23], targeting both pathways could prevent the development of resistance.

However, given the importance of this network in the cellular processes of proliferation, differentiation and apoptosis, its inhibition can be expected to be compensated by hyperactivation of alternative molecular pathways, which would offer the tumour cells escape routes to continue oncogenesis and would eventually lead to the therapy failure. Therefore, it seems logical to test different inhibitors of the pathway, from PI3K to AKT to mTOR (both the mTORC1 and mTORC2 complexes) together with trastuzumab, to see which combination is most effective in controlling tumorigenesis and preventing the development of resistance. We decided to test three different inhibitors covering a broad spectrum of effectors in the pathway, from PI3K to the two mTOR complexes, to ensure the effective blockade of the pathway. One strategy has focused on inhibiting the HER2 signalling pathway more effectively with dual blockade approach. The combined use of trastuzumab and mTOR inhibitors has been shown to be more effective to treat HER2-positive breast cancer than single agents [27]. In addition, receptor tyrosine kinasedependent ERK1 and ERK2 activation following PI3K/AKT/mTOR inhibition have also been described in preclinical models of HER2-positive breast tumours [23,42]. In these cases, the combination of PI3K/AKT/mTOR inhibitors with an anti-HER2 drug or a MEK inhibitor was more effective than single treatments.

The availability of four cellular models of acquired resistance to trastuzumab over an extended period of time (as well as models of primary resistance), in which we had observed hyperactivation of PI3K/AKT/mTOR pathway markers, led us to explore whether

combined suppression of HER2 and PI3K/AKT/mTOR signalling was necessary to achieve optimal therapeutic efficacy, given that there are few such studies in the literature. BEZ235 is an inhibitor of the PI3K/AKT/mTOR pathway with a dual inhibitory capacity of PI3K and mTOR due to the high similarity of the tyrosine kinase domains of both proteins. The combination of trastuzumab plus BEZ235 targets those cells with alterations in the PI3K/AKT/mTOR signalling pathway, due to loss of PTEN or activating mutations in PI3K, while maintaining therapeutic pressure on other cells in the same heterogeneous population that are still sensitive to HER2-targeted drugs [43]. Our results confirm that the addition of BEZ235 overcame resistance to the trastuzumab-only regimen in the sensitive cell lines, some acquired-resistant cells, and in some cells with primary resistance (Figure 3), probably due to its inactivation effect on AKT, S6 and 4E-BP1 phosphorylation [25]. Similarly, several in vivo and in vitro models have shown the efficacy of this combination in restoring sensitivity to HER2-targeted therapy [44,45]. Our results demonstrate that the combination with trastuzumab and BEZ235 significantly results in the reversal of trastuzumab resistance in the primary resistant line HCC1954 (Figure S4). This cell line has an activating H1047R mutation in PI3K, which likely makes it significantly susceptible to BEZ235 treatment [46], and consequently, the combination of BEZ235 with trastuzumab can reverse trastuzumab resistance. However, this did not occur in the JIMT-1 line. This line not only showed loss of PTEN, but also overexpression of mucin 4, which has been described as a mechanism of trastuzumab resistance in breast cancer [47]. The limited effect of trastuzumab plus BEZ235 combination therapy in reversing trastuzumab resistance in the acquired resistance cell lines may be because this dual inhibitor only blocks the action of the mTORC1 complex and not the mTORC2 complex. This results in activation of AKT (Ser473) by the mTORC2 complex and overactivation of the pathway, which may not be affected by PI3K inhibition [39]. In addition, dual PI3K/mTOR inhibition by this drug has been reported to produce compensatory ERK activation, due to activation of receptor tyrosine kinases, such as IGF-1R [23,41]. Despite encouraging results in in vitro and preclinical animal models, few clinical trials with BEZ235 in combination with trastuzumab have been conducted, mainly due to the toxicity of the inhibitor, which causes frequent adverse effects in patients, and high variability in responses to the high doses at which treatment is required.

Everolimus is a rapamycin derivative with mTORC1 complex inhibitory capacity, approved by the FDA to treat postmenopausal patients with ER-positive and HER2-negative metastatic breast cancer. Early phase I trials demonstrated that this drug, in combination with trastuzumab, resulted in decreased cell proliferation in trastuzumab-sensitive cell lines [48]. These results were not confirmed in patient cohorts, such as the phase III BOLERO-1 trial [49], but were confirmed in other trials, such as BOLERO-3 [50]. Given that the patient safety profile of everolimus is superior to that of BEZ235, our results at the cellular level are of interest, although its antiproliferative effects were not as pronounced (Figure 4). This difference between everolimus and BEZ235 in terms of cell growth reflects the different mechanisms of action of the drugs in cell lines with different mutational profiles, as reported previously [25]. Over a decade ago, it was proven that the combination of trastuzumab with everolimus can rescue cancer cells from trastuzumab resistance caused by alterations in the PI3K/AKT/mTOR signalling pathway, with greater efficacy than either agent alone [28]. This is achieved by blocking 4E-BP1 and S6 activation, as well as suppressing AKT activation (which everolimus itself phosphorylates and activates in a feedback loop). Our results showed that the combined treatment of trastuzumab and everolimus in trastuzumab-sensitive lines potentiates, although not significantly, the inhibitory effect of trastuzumab on cell proliferation. However, the combination showed no effect in lines with acquired trastuzumab resistance. Notably, our results demonstrate that in the primary resistant line HCC1954, both combination therapy and individual treatment with everolimus had an impact on cell viability, statistically significantly reversing primary trastuzumab resistance (Figure S4). This may be because the HCC1954 line has the PI3K activating mutation H1047R [31]. But this reversal did not occur in the JIMT-1 line, which

has loss of PTEN. In other preclinical models of trastuzumab resistance, trastuzumab and everolimus (or rapamycin) combined therapy obtained better results than either agent alone [27]. Today, combining everolimus with anti-HER2 drugs to decrease tumour activity in HER-2-overexpressing patients with resistance to trastuzumab-based therapy for metastatic breast cancer has proven to be a useful clinical strategy, which has been confirmed in numerous clinical trials [48,51,52]. The limited effect of everolimus observed in our results could be because this dual inhibitor, like BEZ235, is only capable of inhibiting the mTORC1 complex. In addition, inhibition of mTORC1 causes a reactivation loop in the PI3K/AKT/mTOR signalling cascade, due to inhibition of S6, which negatively regulates PI3K activation [53].

Our most conclusive results in cellular models, however, were obtained with the combination of trastuzumab plus TAK-228. TAK-228 is a competitive inhibitor of the ATP domain of mTOR that can simultaneously block the activity of the mTORC1 and mTORC2 complexes. In the three primary trastuzumab-resistant lines and the four lines with acquired resistance, dual blockade of the HER2 and PI3K pathways significantly increased the therapeutic response. In sensitive lines, the association of TAK-228 with trastuzumab significantly decreased cell proliferation and demonstrated, at the molecular level, an ability to block both mTOR complexes, decreasing phosphorylation of all the effectors analysed. Therefore, TAK-228 potentiates the inhibitory effect of trastuzumab on the PI3K/AKT/mTOR pathway (Figure 2). Our results demonstrate that treatment with trastuzumab in combination with TAK-228 results in a statistically significant decrease in cell proliferation in all lines with acquired resistance, and reverses resistance to trastuzumab. Furthermore, at the molecular level, trastuzumab plus TAK-228 combination treatment proves superior to individual treatments, decreasing the activation of PI3K/AKT/mTOR pathway effectors that trastuzumab alone was unable to inhibit. In the primary trastuzumab-resistant cell line, HCC1954, treatment with trastuzumab plus TAK-228 also significantly reversed trastuzumab resistance (Figure S4). The effect at the molecular level shows that TAK-228 can block mTORC1, decreasing phosphorylation of S6 and 4E-BP1, but not the mTORC2 complex, because it does not decrease AKT (Ser473) activation. This effect is different from that reported in the literature for TAK-228 treatment in combination with lapatinib, which causes complete inhibition of S6, 4E-BP1 and AKT (Ser473) phosphorylation in the HCC1954 line [30]. The JIMT-1 line, however, is not affected by any TAK-228 plus trastuzumab treatment condition, which supports the data presented above indicating that this line, in addition to the loss of PTEN, could present mutations in MUC4 that stabilise the HER2/HER3 heterodimer, thus making inhibition with this type of drug useless for reversing resistance [47]. Furthermore, no molecular modification of its phosphorylation pattern was observed with treatment, suggesting that this cell line exhibits a PI3K/AKT/mTOR-independent mechanism of resistance to trastuzumab that results in activation of the pathway even in the presence of specific inhibitors. TAK-228 has shown efficacy in preclinical models of resistant breast cancer when combined with different anti-HER2 therapies [29,30]. In a preclinical model with HER2-positive breast cancer patient-derived xenografts, TAK-228 sensitised tumours to trastuzumab, so that the combination of both drugs strongly suppressed tumour growth [54]. Given that this and other preclinical trials have shown that combination treatment of the dual mTOR inhibitor TAK-228 with trastuzumab is more potent in treating HER2-positive breast cancers than either agent alone, it is hoped that in the coming years, we will see clinical trials that comprehensively translate the biology of these cancers and subsequently explore targeted therapy strategies. Clinical trials combining TAK-228 with other drugs (such as letrozole, alisertib or paclitaxel) are still under way in solid tumours, including breast cancer.

Sensitivity to trastuzumab is related to activating alterations of the PI3K/AKT/mTOR pathway (either by PIK3CA mutations [55], low/loss of expression of PTEN [56] or both). As described above, biochemical analysis of HER2 and PI3K/AKT/mTOR pathway targets confirmed that trastuzumab treatment partially suppressed pathway signalling in sensitive lines, which lack activating alterations in the PI3K/AKT/mTOR pathway (Figure S1).

HCC1954 and JIMT-1 cell lines were found to harbour activating alterations of the PI3K pathway, whereas the sensitive cell lines were not (BT-474 presents a nonactivating K111N PIK3CA mutation [55]). In the case of primary resistant lines, which do have activating alterations in the pathway, less phosphorylation is reported to be affected. Treatment with TAK-228 (alone or in combination with trastuzumab) resulted in even greater inhibition of these signals in most cell lines. However, in HER2-positive cell lines with primary resistance to trastuzumab and PI3K mutations, treatment with TAK-228 was shown not to affect cell proliferation, in contrast to treatment with BEZ235. These data suggest that in the presence of PI3K activating point mutations, treatment with BEZ235 in combination with trastuzumab may be superior to combination treatment with TAK-228 plus anti-HER2 therapy [25,44,46]. The mutational status of PI3K and expression of PTEN of the cell lines have been previously described (Cosmic Database) [31].

Here, we demonstrate that dual blockade of HER2 and PI3K/AKT/mTOR signalling is effective in improving the therapeutic response in HER2-positive breast cell lines with long-term induced resistance to trastuzumab. The combination of trastuzumab with TAK-228 significantly increased the therapeutic response in all the cases (Figure 2A), suggesting that a decrease in mTOR activation status by TAK-228, as determined by the reduction in phosphorylation levels of S6 and 4E-BP1 (Figure 5), affects trastuzumab sensitivity. One limitation to our study is that we have considered resistance in single trastuzumab treatment models, when the current therapeutic protocol establishes firstline treatment with trastuzumab in combination with pertuzumab (a second monoclonal antibody) for HER2-positive breast cancer. To address this limitation, we have generated four de novo models of HER2-positive cell lines with acquired resistance to trastuzumab plus pertuzumab combination therapy. It will be interesting to see whether some of these models also exhibit the PI3K/AKT/mTOR pathway hyperactivation characteristic of the models presented here, and if so, whether ablation of this signal by dual treatment with inhibitors, such as TAK-228 (or others) are effective in treating this refractory cancer.

#### **5. Conclusions**

In summary, our results obtained in models of sensitive breast cancer cell lines, lines with acquired resistance, and lines with primary resistance to trastuzumab, exposed to combination therapy with specific inhibitors of the PI3K/AKT/mTOR signalling pathway plus trastuzumab, suggest that this combination therapy favours the reversal of trastuzumab resistance. Inhibition of the PI3K/AKT/mTOR pathway using the mTORC1 and mTORC2 inhibitor, TAK-228, can reverse acquired resistance to trastuzumab in all models generated and in some primary resistant lines. When combined with trastuzumab, treatment with the inhibitor TAK-228 has been shown to be superior to the other two inhibitors tested, BEZ235 and everolimus, in reversing acquired trastuzumab resistance. However, in the presence of PI3K activating mutations, single and combined treatment with BEZ235 has been shown to be superior to treatment with TAK-228 and everolimus.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/cancers13112778/s1: Figure S1. Immunoblotting analysis of trastuzumab-sensitive and -resistant cells. Cell lines were treated with 15 µg/mL trastuzumab for 24 h. Whole-cell protein extracts were analysed with the indicated antibodies. Images are representative of three independent experiments. C, control culture medium; T, trastuzumab 15 µg/mL. Figure S2. Characterisation of a panel of cell line models of de novo trastuzumab resistance. (A) Effect of trastuzumab treatment on primary resistant cells. Proliferation was measured after seven days of treatment by trypan blue exclusion. Data are expressed as mean +/− SD from ≥ three independent experiments. \* denotes *p* ≤ 0.05, \*\* denotes *p* ≤ 0.01 and \*\*\* denotes *p* ≤ 0.001. (B) Immunoblotting analysis of primary resistant cell lines. Whole-cell protein extracts were analysed with the indicated antibodies. Images are representative of three independent experiments. Figure S3. Effect of increasing concentration of PI3K/AKT/mTOR inhibitors over seven days of treatment on cell lines BT-474, BT-474.rT3, SK-BR-3, SK-BR-3.rT1, AU-565, AU-565.rT2, EFM-192A, EFM-192A.rT1, HCC1954 and JIMT-1. Figure S4. Effects of decrease in mTOR activation on trastuzumab sensitivity in primary resistant cell lines. Trastuzumab resistant cells were treated for seven days with DMSO, 15 µg/mL trastuzumab (T), 1 or 5 nM BEZ235 (B), 0.5 or 1 nM everolimus (E), 1 or 5 nM TAK-228 (I), or a combination of 15 µg/mL trastuzumab plus each mTOR inhibitor. Viable cells were then counted by trypan blue exclusion. Viability is presented as a percentage of the DMSO-treated control vector group. Error bars represent standard deviation between replicates (*n* ≥ 3). \* denotes *p* ≤ 0.05, \*\* denotes *p* ≤ 0.01 and \*\*\* denotes *p* ≤ 0.001. (A) HCC1419 trastuzumab-sensitive/resistant cells. (B) HCC1954 trastuzumab-resistant cells. (C) JIMT-1 trastuzumab-resistant cells. Figure S5. Biochemical analyses of primary trastuzumabresistant cells treated with trastuzumab and TAK-228. HCC1954 and JIMT-1 cells were treated for 24 h with DMSO, 15 µg/mL trastuzumab (T), 5 and 50 nM TAK-228 (I), or a combination of 15 µg/mL trastuzumab plus 5 or 50 nM TAK-228. Whole-cell protein extracts were analysed with the indicated antibodies. Images are representative of three independent experiments. Figure S6: Uncropped Western blot images.

**Author Contributions:** Conception and design: J.M.-G., F.R. Development of methodology: M.S.-Á., E.M.-A., S.Z., J.M.-G., F.R. Acquisition of data: M.S.-Á., E.M.-A., M.L., S.Z., J.M.-U. Analysis and interpretation of data: M.S.-Á., E.M.-A., M.L., S.Z., J.M.-U., J.M.-G., F.R. Writing, review, and/or revision of the manuscript: M.S.-Á., P.E., A.R., J.A., J.M.-G., F.R. Administrative, technical, or material support: M.S.-Á., E.M.-A., M.L., S.Z. Study supervision: J.M.-G., F.R. All authors have read and agreed to the published version of the manuscript.

**Funding:** The present work was supported by grants from the Spanish Ministry of Economy and Competitiveness (MINECO) with European Regional Development Fund (ERDF) funding through the Institute of Health Carlos III (AES Program, grants PI18/00382, PI18/00006 and PI18/01219; CIBERONC, Biomedical Research Networking Centre for Cancer). M.S.-Á. was supported by a Jiménez Díaz predoctoral research grant funded by the Fundación Conchita Rábago de Jiménez Díaz.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data sharing is not applicable to this article.

**Acknowledgments:** We thank Oliver Shaw for language editing.

**Conflicts of Interest:** The authors declare no competing financial interests.

#### **References**


### *Article* **Downregulation of Snail by DUSP1 Impairs Cell Migration and Invasion through the Inactivation of JNK and ERK and Is Useful as a Predictive Factor in the Prognosis of Prostate Cancer**

**Desirée Martínez-Martínez 1 , María-Val Toledo Lobo 2,3 , Pablo Baquero 4 , Santiago Ropero 4 , Javier C. Angulo 5 , Antonio Chiloeches <sup>4</sup> and Marina Lasa 1, \***


**Simple Summary:** The role of dual specificity phosphatase 1 (DUSP1) in metastasis-associated processes in prostate cancer and its impact on patient outcome remains to be elucidated. Our results reveal that this phosphatase reduces Snail expression and impairs cell migration and invasion in prostate cancer cells through a mechanism involving the inhibition of DUSP1 molecular targets, c-Jun N-terminal kinase (JNK) and extracellular-signal-regulated kinase (ERK). In clinical samples, we evidence an inverse correlation between DUSP1 expression and Snail levels, which are further associated with JNK and ERK activation. Importantly, patients with the pattern DUSP1high/activated JNKlow/activated ERKlow/Snail low exhibit a longer time to progression and a better outcome than those with the opposite pattern. All these findings highlight new opportunities to improve current therapeutic strategies for the diagnosis and treatment of prostate cancer.

**Abstract:** Dual specificity phosphatase 1 (DUSP1) is crucial in prostate cancer (PC), since its expression is downregulated in advanced carcinomas. Here, we investigated DUSP1 effects on the expression of mesenchymal marker Snail, cell migration and invasion, analyzing the underlying mechanisms mediated by mitogen-activated protein kinases (MAPKs) inhibition. To this purpose, we used different PC cells overexpressing or lacking DUSP1 or incubated with MAPKs inhibitors. Moreover, we addressed the correlation of DUSP1 expression with Snail and activated MAPKs levels in samples from patients diagnosed with benign hyperplasia or prostate carcinoma, studying its implication in tumor prognosis and survival. We found that DUSP1 downregulates Snail expression and impairs migration and invasion in PC cells. Similar results were obtained following the inhibition of c-Jun N-terminal kinase (JNK) and extracellular-signal-regulated kinase (ERK). In clinical samples, we evidenced an inverse correlation between DUSP1 expression and Snail levels, which are further associated with JNK and ERK activation. Consequently, the pattern DUSP1high/activated JNKlow/activated ERKlow/Snail low is associated with an overall extended survival of PC patients. In summary, the ratio between DUSP1 and Snail expression, with additional JNK and ERK activity measurement, may serve as a potential biomarker to predict the clinical outcome of PC patients. Furthermore, DUSP1 induction or inhibition of JNK and ERK pathways could be useful to treat PC.

**Keywords:** DUSP1; MAPK; Snail; prostate cancer; migration and invasion; patient survival; biomarkers

**Citation:** Martínez-Martínez, D.; Toledo Lobo, M.-V.; Baquero, P.; Ropero, S.; Angulo, J.C.; Chiloeches, A.; Lasa, M. Downregulation of Snail by DUSP1 Impairs Cell Migration and Invasion through the Inactivation of JNK and ERK and Is Useful as a Predictive Factor in the Prognosis of Prostate Cancer. *Cancers* **2021**, *13*, 1158. https://doi.org/10.3390/ cancers13051158

Academic Editors: Ion Cristóbal and Marta Rodríguez

Received: 1 February 2021 Accepted: 3 March 2021 Published: 8 March 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

#### **1. Introduction**

Prostate cancer is one of the most frequently diagnosed cancers in men worldwide and is the second leading cause of cancer-related deaths among males [1]. The majority of the deaths associated with this type of tumors are related to metastasis, in which the so-called epithelial–mesenchymal transition (EMT) is one of the most important events involved [2]. EMT is a cell plasticity program that plays very important roles during embryonic development and can be reactivated in adult physiological situations to maintain epithelial homeostasis in order to guarantee tissue integrity and organ function [3,4]. Moreover, EMT also has important roles in pathological processes such as cancer metastasis. This process is defined by a loss of epithelial cell-specific characteristics, such as polarity and cohesiveness, and by an acquisition of a mesenchymal-like morphology with increased motility [5]. The abnormal activation of EMT in cancer disrupts the intercellular junctions, causing the dissociation of surrounding cells and the acquisition of migratory phenotype. Thus, EMT is often associated with the invasion and metastatic ability of tumor cells. In agreement with this, a large amount of evidence have shown that metastatic cells display a decreased expression of epithelial markers and an increased expression of mesenchymal markers both in vitro and in vivo [4]. One of the hallmarks of the EMT is the overexpression of Snail, which is a transcription factor that downregulates the expression of epithelial genes and upregulates the expression of mesenchymal genes, ultimately leading to increased migration and invasion [6]. Thus, Snail overexpression has been found in the invasive fronts of several human tumors derived from epithelial cells, including hepatocellular, breast, or thyroid carcinomas, among others [7–11]. Accordingly, Snail is widely associated with invasiveness, metastasis, tumor recurrence, and poor prognosis [7–9]. In particular, metastatic prostate cancer cells display typical features of EMT, and Snail plays an important role in the regulation of cell polarity, the expression of epithelial and mesenchymal markers, as well as migration and invasion [2,12]. Consistently, Snail expression increases with prostate cancer progression from benign to bone metastatic tumors [13–15]. From a molecular point of view, several studies in different tumor contexts have demonstrated that the expression and activity of Snail can be regulated by multiple molecular mechanisms, including transcriptional regulation and post-translational modifications. In this sense, one of the most important mechanisms that affects Snail stability involves its export from the nucleus and its subsequent degradation by the proteasome in the cytosol [16]. Furthermore, it has been demonstrated that mitogen-activated protein kinase (MAPK) activation results in an increase of Snail protein levels, which in turn regulate the expression of EMT-associated genes [16].

Dual specificity phosphatase 1 (DUSP1) acts as a tumor suppressor by negatively regulating MAPK activity in different tumors, including prostate cancer. Thus, we and others have previously demonstrated that the expression of this phosphatase decreases with prostate tumor progression. Whereas DUSP1 levels are high in benign prostatic hyperplasia (BPH) and hormone-sensitive prostatic adenocarcinoma (HS-PC), the expression of this phosphatase is almost absent in hormone-refractory prostatic adenocarcinoma (HR-PC) [17,18]. Consistently, DUSP1 overexpression in androgen-independent prostate cancer cells promotes apoptosis through inhibition of the p38 mitogen-activated protein kinase (p38MAPK)/nuclear factor-kappaB (NF-kB) signaling pathway [17]. Moreover, DUSP1 is also involved in the pro-apoptotic effects of the chemopreventive molecule resveratrol in prostate cancer cells [19]. In addition, it has been reported that DUSP1 inhibits cell migration, invasion, and metastasis in other cancer types [20–24]. However, despite all these studies showing DUSP1 as an apoptosis inducer in prostate cancer, the role of this phosphatase in cell migration and invasion in these kind of tumors remains largely unknown. Therefore, in this work, we aimed to investigate whether DUSP1 is involved in the motility of prostate cancer cells and whether this protein regulates the signaling pathways that control these processes. In brief, our results demonstrate that DUSP1 decreases Snail expression as well as cell migration and invasion in prostate tumor cells. Moreover, our data also support that DUSP1 regulates both processes, together with Snail expression, through

the inactivation of c-Jun N-terminal kinase (JNK) and extracellular-signal-regulated kinase (ERK). Importantly, we also elucidate a new molecular pattern, which might be useful as a prognosis biomarker for prostate cancer monitoring. This molecular signature is characterized by an inverse correlation between DUSP1 and Snail levels with an additional activation of JNK and ERK pathways. Finally, our results show that expression of DUSP1 and Snail, as well as levels of active ERK and JNK correlate with time of progression and with exitus rate. In line with this, those patients with high DUSP1 expression, low JNK and ERK activities, and low Snail expression exhibit a longer time until they reach metastatic disease, a better outcome, and a lower exitus rate than those with the opposite expression pattern (DUSP1low/activated JNKhigh/activated ERKhigh/Snailhigh). Importantly, we consider that our findings suggest new opportunities to improve current strategies for the diagnosis and treatment of prostate cancer.

#### **2. Materials and Methods**

#### *2.1. Cell Lines, Inhibitors, Plasmids, Cell Transfection and Luciferase Assay*

DU145 and PC3 androgen-independent prostate cancer cells were purchased from the American Tissue Culture Collection (Manassas, UA, USA) and were cultured as recommended. The inhibitors were U0126 (Promega Biotech Ibérica, Madrid, Spain), SB203580, SP600125, and MG132 (Calbiochem, Merck Chemicals, Barcelona, Spain). The pCMV-DUSP1 and the Snail-Luc reporter plasmids were previously described [25,26]. For overexpression and siRNA experiments, cells were transiently transfected as previously described [19]. Luciferase assays were performed as in [27], being the luciferase levels normalized to those of renilla, and expressed as the induction over the controls.

#### *2.2. Western Blot Analyses and Immunofluorescence Staining*

Western blot analyses were performed as described in [27]. The antibodies were anti-DUSP1, anti-p38MAPK, anti-JNK1, and anti-ERK2 (Santa Cruz Biotechnology, Heidelberg, Germany); anti-phospho-p38MAPK (pp38MAPK), anti-phospho-ERK (pERK), and anti-Snail (Cell Signalling Technology, Izasa S.A., Barcelona, Spain); anti-phospho-pJNK (pJNK) (Promega Biotech Ibérica, Madrid, Spain); anti-Tubulin (Sigma Aldrich, Madrid, Spain); peroxidase-conjugated secondary antibodies (GE Healthcare Europe GMBH, Barcelona, Spain). Tubulin was utilized as a loading control for Western blotting analysis. Relative protein levels compared to tubulin were analyzed by Image J software and plotted.

Immunofluorescence staining was performed as previously described [28]. Briefly, cells cultured on coverslips were fixed, permeabilized, blocked and, after several washes, stained for Snail with the specific antibody, followed by the anti-rabbit Alexa Fluor® 488 secondary antibody (BD Biosciences, Franklin Lakes, NJ, USA). Samples were mounted using ProLong® Gold Antifade Mountant with DAPI (Invitrogen, Life Technologies, Carlsbad, CA, USA), and fluorescence visualization was performed by ICTS "NANBIOSIS", more specifically by the Confocal Microscopy Service (Ciber in Bioengineering, Biomaterials & Nanomedicine (CIBER-BNN)) at the Alcalá University.

#### *2.3. Cell Migration and Invasion Assays*

Cell migration was examined by wound-healing assays. After transfection/treatment of cells, scratches were made using sterile 200 µL-pipette tips, and bright-field microphotographs were taken at different times. The percentages of cell migration were quantitated, by the ImageJ software, measuring the width of the cell-free zone immediately after making the scratch, and at different times after scratching. Migration velocities represented the average velocities at which the cells moved into the gap.

Cell invasion was examined in Matrigel-coated transwells (BD Biosciences, Franklin Lakes, NJ, USA) as previously described [29]. The number of cells loaded onto the surface of each Matrigel-coated transwell was 100,000 in DUSP1 overexpression and MAPK inhibitors experiments, and 50,000 in DUSP1 silencing experiments. Invaded cells were stained with crystal violet, and three different cell fields of each well were photographed under a phase contrast microscope (Nikon TS100). Changes in cell invasion were expressed as percentages of the corresponded controls.

#### *2.4. Experimental Subjects and Immunohistochemistry of Prostate Tissues*

Paraffin-embedded samples from patients diagnosed with BPH (*n* = 9) or PC (*n* = 35) were used (Table 1). Five-micron thick sections from samples were incubated overnight at room temperature with each primary antibody (anti-DUSP1 and anti-Snail1, clone G7 (Santa Cruz Biotechnology, Heidelberg, Germany); anti-pJNK (Promega, Promega Biotech Ibérica, Madrid, Spain); anti-pERK (Cell Signalling Technology, Izasa S.A., Barcelona, Spain)). Afterwards, samples were washed and sequentially incubated with the biotin free, peroxidase-detection system (polymer-based detection kit, MasVisionTM, Master Diagnostica, Spain). Nuclei were stained with Caracci's hematoxylin. Samples were dehydrated and mounted with DePex. The intensity of the immunostaining was evaluated by two independent observers who were blinded to patient clinical information through a system of subjective gradation. Immunostaining scores were ranged into four categories based on the staining pattern of the majority of tumor cells in the whole section, which were grouped into two main categories for statistical purposes (0–1: negative/low staining; 2–3: moderate/high staining).


**Table 1.** Clinical data of prostate cancer patients (*n* = 35).

#### *2.5. Statistical Analyses*

In the experiments with cell lines, all data were expressed as means ± SEM. Student's *t* test was performed using the SSC-Stat software (V2.18, University of Reading, UK). In the immunohistochemistry assays, GraphPad Prisma 3.0 software was used for statistical purposes. Immunostaining score and clinical data were analyzed using one-way ANOVA and either the Bonferroni's or Dunnet´s multiple comparison tests. The correlation among markers was analyzed using the Pearson´s test (95% confidence interval). Log-rank test and survival curves were used to determine the relationship among markers and time to clinical progression. The statistical significance of difference between groups was expressed by asterisks (\* 0.01 < *p* < 0.05; \*\* 0.001 < *p* < 0.01; \*\*\* *p* < 0.001).

#### **3. Results**

#### *3.1. DUSP1 Downregulates Snail Expression and Impairs Cell Migration and Invasion in Prostate Cancer Cells*

To study the role of DUSP1 in the migration and invasion of prostate cancer cells, we first analyzed the effect of DUSP1 knockdown on Snail expression in DU145 cells. DUSP1 silencing efficiency was tested by measuring its protein levels, observing a significant decrease in DUSP1-deficient cells (Figure 1a). The results showed an increase in Snail levels both at a transcriptional (Figure 1b) and at a protein level (Figure 1c). Consistently, DUSP1-deficient cells significantly displayed an enhanced capacity of both cell migration (Figure 1d–f) and invasion (Figure 1g,h). Conversely, cells overexpressing DUSP1 showed a significant increase in protein levels (Figure 1i), significantly reduced Snail expression levels (Figure 1j,k), were less migratory (Figure 1l–n), and displayed limited cell invasion (Figure 1o,p). Similar results were obtained from experiments performed in PC3 cells, thus ruling out the cell-type specific effects of this phosphatase (Figure S1 in Supplementary Materials). All these results indicate that DUSP1 downregulates Snail expression, which in turn results in a further decrease in migration and invasion of prostate cancer cells.

#### *3.2. The Inhibition of JNK and ERK Downregulates Snail Expression, Cell Migration and Invasion*

Given that DUSP1 is able to dephosphorylate and inhibit different MAPK signaling pathways, we next investigated which of them were involved in the effects of this phosphatase on Snail expression, cell migration, and invasion in DU145 cells. Our results confirmed that p38MAPK, JNK, and ERK were targets of this phosphatase, since the abrogation of its expression activated these three MAPKs (Figure 2a). In addition, the inhibitory effect of DUSP1 on MAPK's activities was confirmed by monitoring the levels of their phosphorylated forms in cells overexpressing this phosphatase (data not shown).

Further analysis of Snail expression after inactivation of these MAPKs was performed upon treatment of cells with specific inhibitors. The efficiency of selective inhibition of MAPK activity by SB203580 (p38MAPK inhibitor), SP600125 (JNK inhibitor), or U0126 (MEK inhibitor) was confirmed by measuring MAPK phosphorylation levels in cells incubated with these compounds (Figure S2 in Supplementary Materials). Moreover, the inhibition of these MAPKs differently affected cell proliferation and survival [17] (unpublished results). Regarding Snail expression, the inhibition of p38MAPK with SB203580 did not affect Snail expression (Figure 2b,c). In contrast, treatment with either SP600125 or U0126 achieved a significant reduction in Snail levels (Figure 2b), although only ERK inhibition exerted its effects at a transcriptional level (Figure 2c). Moreover, the effect of JNK and ERK inhibition on Snail proteasomal degradation was assessed, and the analysis of these data revealed that the reduction in Snail levels achieved by SP600125 or U0126 was reversed by the inhibitor MG132 (Figure 2d), suggesting that Snail regulation by JNK or ERK pathways is proteasome dependent.

**Figure 1.** DUSP1 downregulates Snail expression and impairs cell migration and invasion in DU145 cells. (**a**) Cells were transfected for 48 h with the control siRNA (siControl) or the DUSP1 siRNA (siDUSP1) and expression levels of DUSP1 and Tubulin were determined by western blotting. (**b**) Cells were transfected for 48 h with the siControl or the siDUSP1 together with the Snail-Luc plasmid and luciferase activity was measured in cell extracts. (**c**) Cells were transfected as in *a* and expression levels of Snail and Tubulin were determined by western blotting. (**d**–**f**) Wound healing assay and measurement of wound closure area and velocity in cells transfected as in *a*. (**g**,**h**) Invasion capacity using transwell assays in cells transfected as in *a*. (**i**) Cells were transfected with a control vector (Control) or a vector encoding DUSP1 (DUSP1) and expression levels of DUSP1 and Tubulin were determined by western blotting. (**j**) Cells were transfected for 48 h with the Control or the DUSP1 vectors together with the Snail-Luc plasmid and luciferase activity was measured in cell extracts. (**k**) Cells were transfected with the Control or the DUSP1 vectors and expression levels of Snail and Tubulin were determined by western blotting. (**l**–**n**) Wound healing assay and measurement of wound closure area and velocity in cells transfected as in *i*. (**o**,**p**) Invasion capacity using transwell assays in cells transfected as in *i*. For all the results, data are shown as the mean ± SEM of at least three independent experiments. For migration and invasion assays, pictures are from one representative experiment of three with similar results. Student's *t* test: \* 0.01 < *p* < 0.05; \*\* 0.001 < *p* < 0.01; \*\*\* *p* < 0.001.

**Figure 2.** The inhibition of JNK and ERK downregulates Snail expression in DU145 cells. (**a**) Cells were transfected for 48 h with the siControl or the siDUSP1 and expression levels of DUSP1, phosphorylated MAPKs (pp38, pJNK, pERK), total MAPKs and Tubulin were determined by western blotting. (**b**) Cells were incubated at different times in the absence or presence of 1 µM SB203580 (SB), 10 µM SP600125 (SP) or 20 µM U0126 (U0), and expression levels of Snail and Tubulin were determined by western blotting. (**c**) Cells were transfected with the Snail-Luc plasmid, incubated for 48 h as in *b* and luciferase activity was assayed in cell extracts. (**d**) Cells were incubated for 48 h with 10 µM SP600125 or 20 µM U0126, treated in the absence or presence of 10 µM MG132 for the last 4 h and expression levels of Snail and Tubulin were determined by western blotting. For all the results, data are shown as the mean ± SEM of at least three independent experiments. Student's t test: \* 0.01 < *p* < 0.05; \*\* 0.001 < *p* < 0.01; \*\*\* *p* < 0.001.

Additionally, both JNK and ERK inhibition reduced cell migration (Figure 3a–f) and invasion (Figure 3g–j), mimicking the results obtained following DUSP1 overexpression (Figure 1j–n). In contrast, p38MAPK inhibition did not affect cell migration (Figure S3 in Supplementary Materials), suggesting that this kinase is supporting other processes in prostate cancer progression. All these results, together with those showed in Figure 1, demonstrate that both pharmacological inhibition of JNK or ERK and DUSP1 overexpression exert similar effects on Snail expression, cell migration, and invasion, suggesting that this phosphatase regulates these processes by specifically targeting these two pathways.

#### *3.3. Snail Subcellular Location Is Regulated by the Phosphatase DUSP1 and JNK and ERK Signaling Pathways*

One of the most common molecular mechanisms by which Snail expression is downregulated involves its nuclear export to the cytoplasm and its subsequent proteasomal degradation. Since we demonstrated that JNK and ERK inhibition decreased Snail expression by affecting its proteasomal degradation (Figure 2d), we next analyzed Snail location upon treatment with the specific MAPKs inhibitors. As expected, our results showed that SP600125 and U0126 induced a more diffuse location of Snail with an increase in the cytosolic compartment (Figure 4).

**Figure 3.** The inhibition of JNK and ERK decreases migration and invasion in DU145 cells. (**a**–**f**) Wound healing assay and measurement of wound closure area and velocity in cells incubated for 48 h with 10 µM SP600125 (**a**–**c**) or 20 µM U0126 (**d**–**f**). (**g**–**j**) Invasion capacity using transwell assays in cells incubated as above. For all the results, data are shown as the mean ± SEM of at least three independent experiments. Pictures are from one representative experiment of three with similar results. Student's *t* test: \*\* 0.001 < *p* < 0.01; \*\*\* *p* < 0.001.

**Figure 4.** Snail subcellular location is regulated by the JNK and ERK signaling pathways. DU145 cells were incubated for 48 h with 10 µM SP600125 or 20 µM U0126 and Snail subcellular location was determined by immunofluorescence as described in Material and methods. DAPI was used to identify the nuclei. Pictures are from one representative experiment of three with similar results.

Consistently, DUSP1 overexpression also induced a predominantly cytosolic location of Snail, while DUSP1 knockdown maintained this transcription factor in the nucleus (Figure 5). These results reveal that both DUSP1 overexpression and JNK or ERK inhibition induce the export of Snail from the nucleus to the cytoplasm; hence, these data strengthen our hypothesis that this phosphatase exerts its effects on Snail subcellular location through the downregulation of these MAPKs.

#### *3.4. JNK and ERK Cooperatively Regulate Snail Expression, Cell Migration and Invasion*

Given that DUSP1 impaired the activity of JNK and ERK (Figure 2a), and that the individual inhibition of these MAPKs downregulated Snail expression (Figure 2b), as well as cell migration and invasion (Figure 3), we further studied whether these MAPKs cooperated in the regulation of these events in our prostate cancer cells. Interestingly, the combination of SP600125 and U0126 significantly achieved a higher reduction in Snail expression than the single treatments in DU145 cells (Figure 6a).

Notably, cells treated with SP600125 plus U0126 were even less migratory (Figure 6b–d) and displayed less invasion capacity (Figure 6e,f) compared to cells treated with the single agents. To further strengthen these results, we extended our study, performing similar experiments in PC3 cells. As expected, our results showed that JNK and ERK cooperatively regulated Snail expression and cell migration also in these cells (Figure S4 in Supplementary Materials). All these results indicate that the dual inhibition of JNK and ERK pathways in prostate cancer cells is more effective in decreasing Snail expression, cell migration, and invasion than blocking each pathway independently. Altogether, these results suggest

once again that DUSP1 regulates these events through a dual inhibition of both JNK and ERK pathways.

**Figure 5.** Snail subcellular location is regulated by the phosphatase DUSP1. (**a**) DU145 cells were transfected for 48 h with the Control or the DUSP1 vectors. (**b**) Cells were transfected for 48 h with the siControl or the siDUSP1. In both set of experiments, Snail subcellular location was determined by immunofluorescence as described in Material and methods. DAPI was used to identify the nuclei. Pictures are from one representative experiment of three with similar results.

**Figure 6.** JNK and ERK cooperatively regulate Snail expression, cell migration and invasion in DU145 cells. Cells were incubated in the absence (C) or presence of 10 µM SP600125 (SP, 24 h) and 20 µM U0126 (U0, 48 h). (**a**) Expression levels of Snail and Tubulin were determined by western blotting. (**b**–**d**) Wound healing assay and measurement of wound closure area and velocity. (**e**,**f**) Invasion capacity using transwell assays. For all the results, data are shown as the mean ± SEM of at least three independent experiments. For migration and invasion assays, pictures are from one representative experiment of three with similar results. Student's *t* test: \* 0.01 < *p* < 0.05; \*\* 0.001 < *p* < 0.01; \*\*\* *p* < 0.001.

#### *3.5. DUSP1 Expression Inversely Correlates with Snail Levels and Activated JNK and ERK in Human Prostate Samples*

To investigate whether our results obtained from the experiments performed with the cell lines were clinically relevant, we next analyzed the expression levels of DUSP1 and Snail in a series of samples from patients with BPH, HS-PC, and HR-PC (Table 1). Prostatic

glands from BPH samples showed a high expression of DUSP1 (Figure 7a-I) and a weak expression of Snail (Figure 7a-X). In prostate cancer samples, DUSP1 expression was high in HS-PC (Figure 7a-II), whereas low or no signal for Snail was detected (Figure 7a-XI). Conversely, HR-PC samples showed a weak or even undetectable DUSP1 expression (Figure 7a-III) but a moderate to strong signal for Snail (Figure 7a-XII). Consequently, the immunohistochemical analyses demonstrated an inverse correlation between DUSP1 and Snail, with a DUSP1high/Snaillow pattern in both BPH and HS-PC samples, and a DUSP1low/Snailhigh pattern in HR-PC samples. Importantly, results from the Pearson´s Test confirmed the inverse correlation between DUSP1 and Snail expression (Figure 7b).

**Figure 7.** DUSP1 expression inversely correlates with Snail levels and activated JNK and ERK in human prostate samples. (**a**) Immunohistochemical analysis of expression levels of DUSP1 (I–III), phosphorylated JNK (pJNK, IV–VI), phosphorylated ERK (pERK, VII–IX) and Snail (X–XII) from human prostate cancer samples. Micrographs were taken at 200× magnification and show serial sections from the same gland stained with each one of the four used antibodies. (**b**) Immunohistochemical score for DUSP1, pJNK, pERK and Snail in samples from HS-PC and HR-PC. The statistical analysis was performed with One-way ANOVA and Dunnet´s multiple comparison test, and asterisks show the statistical significance of differences between the groups (*a:* comparison with DUSP1 from HS-PC samples; *b:* comparison with DUSP1 from HR-PC samples; *c:* HS-PC vs HR-PC for each marker), \* 0.01 < *p* < 0.05; \*\* 0.001 < *p* < 0.01.

Since our data in prostate cancer cells revealed that DUSP1 inhibits JNK and ERK (Figure 2a) and these MAPKs negatively regulated Snail expression (Figure 2b–d), we also analyzed the levels of activated JNK and ERK (pJNK and pERK) in patient samples. Accordingly, our results indicated that the levels of active JNK and ERK were low in BPH samples (Figure 7a-IV,VII). Moreover, an inverse correlation was also detected for PC samples, with a DUSP1high/pJNKlow /pERKlow pattern in samples from HS-PC patients (Figure 7a-II,V,VIII) and a DUSP1low/pJNKhigh /pERKhigh pattern in HR-PC samples (Figure 7a-III,VI,IX). As in previous results, the Pearson´s Test confirmed these inverse correlations (Figure 7b).

In all cases, subcellular localization for DUSP1 and pERK was mainly cytosolic, while Snail was located in the cell nucleus. Regarding pJNK subcellular expression, it was predominantly nuclear, although a mild-to-moderate signal for this marker was also observed in cytosol (Figure S5 in Supplementary Materials). Moreover, a compilation of different IHC images for each marker can be observed in Figure S6 in Supplementary Materials.

#### *3.6. The Relationship of DUSP1 and Snail Levels and JNK and ERK Activities Are Associated with Disease Progression and Clinical Outcome in Patients with Prostate Cancer*

Since we observed a differential expression of DUSP1, Snail, and the active forms of JNK and ERK in samples from prostate cancer patients at different stages, we next studied the interrelation between the levels of these proteins and some of the most important clinical parameters. Firstly, we analyzed the correlation of expression patterns of DUSP1, Snail, and activated JNK and ERK with either Gleason score (Figure 8a) or American Joint Committee on Cancer (AJCC) group staging at diagnosis [30] (Figure 8b), and no correlation was observed in any of these cases. In contrast, we did observe a significant correlation when we compared the levels of DUSP1, Snail, and activated JNK and ERK with both the disease progression and the clinical outcome (Figure 8c–e). Thus, shorter intervals to clinical progression were related with lower DUSP1 expression and higher levels of activated JNK (*log-rank*, *p* = 0.0237) and ERK (*log-rank*, *p* = 0.0005) (Figure 8c), although we did not observe correlation of time to clinical progression with lower DUSP1 expression and higher levels of Snail (Figure 8c). Despite this, the combined pattern DUSP1low/pJNKhigh/pERKhigh/Snailhigh was strongly related with overall time to clinical progression (*log-rank*, *p* = 0.0002) (Figure 8d). More importantly, our data also evidenced a significant relationship between the expression pattern of these proteins and exitus (Figure 8e). Indeed, the median overall survival of patients with the combined pattern DUSP1low/pJNKhigh/pERKhigh/Snailhigh was 29 months, compared to 79 months in patients with DUSP1high/pJNKlow/pERKlow/Snaillow.

Collectively, all the results in human prostate samples reveal the existence of an inverse correlation between DUSP1 expression and the levels of Snail and activated JNK and ERK (negative correlation at Pearson´s test, *p* < 0.001), supporting our experiments in prostate cancer cells which demonstrate that DUSP1 downregulates Snail expression. In addition, our results indicate that low levels of DUSP1 and high levels of pJNK (*p* < 0.02) and pERK (*p* < 0.0005), but not Snail (*p* > 0.05), are related to shorter intervals to clinical progression. Finally, and more interestingly, we evidence that the levels of all proteins tested are related to clinical outcome, suggesting that the ratio between the expression of DUSP1, Snail, and activated JNK and ERK is an important marker for diagnostic purposes in prostate cancer.

**Figure 8.** The relationship of DUSP1 and Snail levels and JNK and ERK activities are associated with disease progression and clinical outcome in patients with prostate cancer. (**a**,**b**) Immunohistochemical score for DUSP1, phosphorylated JNK and ERK (pJNK and pERK) and Snail in samples ranged into three categories based on their Gleason Score (**a**) or AJCC group staging at diagnosis (**b**). (**c**) Progression-free survival of patients showing immunohistochemical score for DUSP1/pJNK, DUSP1/pERK or DUSP1/Snail. Samples were ranged into two categories based on the staining pattern of the majority of tumor cells in the whole section (negative/low (ng/lo); moderate/high (md/hi)). (**d**) Progression-free survival of patients showing immunohistochemical score for DUSP1/pJNK/pERK/Snail. Samples were ranged into two categories as described in *c*. (**e**) Immunohistochemical score for DUSP1, pJNK, pERK and Snail in samples from patients either alive or dead. The statistical analysis was performed with One-way ANOVA and Dunnet´s multiple comparison test, and asterisks show the statistical significance of differences between the groups (*a:* comparison with DUSP1 from HS-PC samples; *b:* comparison with DUSP1 from HR-PC samples; *c:* HS-PC vs HR-PC for each marker). *TCP, Time to clinical progression*, \* 0.01 < *p* < 0.05; \*\* 0.001 < *p* < 0.01; \*\*\* *p* < 0.001.

#### **4. Discussion**

DUSP1 expression has been previously related to different stages of human prostate carcinomas. In line with this, the expression of this phosphatase is high in BPH and HS-PC, but it is lost in later stages, such as HR-PC [17]. Furthermore, DUSP1 overexpression in androgen-independent prostate cancer cells induces apoptosis through both p38MAPK and NF-kB dependent mechanisms [17]. Here, we show for the first time that this phosphatase plays an additional anti-tumorigenic role in prostate cancer cells, since it decreases the expression levels of the EMT master regulator, Snail, and inhibits cell migration and invasion through the inactivation of JNK and ERK. Interestingly, we also demonstrate a correlation between the expression levels of DUSP1 and Snail and the activity of JNK and

ERK in samples from prostate cancer patients, discovering a novel approach to predict the prognosis and outcome of this disease.

Previous studies have shown that the overexpression of Snail in prostate cancer cells is associated with an increased cell migration and invasion, while its silencing induces a decrease in these processes [31]. In agreement with this, here, we demonstrate that DUSP1 downregulates Snail expression and inhibits migration and invasion in prostate cancer cells. Our data are similar to those observed in different types of tumors, in which DUSP1 suppresses cell migration, cell invasion, metastasis, and/or angiogenesis by inhibiting either ERK [21,23], JNK [22,24], or p38MAPK [20]. Consistently with DUSP1 effects on MAPK activity, the ERK pathway is one of the major oncogenic signals in human cancers because its activation leads to an increase in proliferation, invasion, and metastasis [32]. Particularly in prostate cancer, the ERK pathway is often hyperactivated [33], acts as an inducer of cell migration and invasion [34,35] through a Snail-mediated mechanism [36], and is involved in the effects of different molecules on these processes [37–39]. In addition, the JNK pathway has also been described to be important as a pro-tumorigenic signal through Snail regulation in different tumors [40–42]. Regarding prostate cancer, it has been previously described that JNK activity is related to elevated cell migration and invasion [43] and controls tumor growth in DU145 prostate carcinoma xenografts [44], although the involvement of Snail in these processes is still unknown. Our results are in agreement with all these data, since we demonstrate that the effects of DUSP1 on Snail levels, cell migration, and cell invasion are similar to those observed upon specific inhibition of the ERK and JNK pathways. By contrast, our findings evidence that p38MAPK is not involved in the regulation of these processes by DUSP1. Although several reports have showed that this kinase promotes cancer by enhancing migration in tumor cells [45], we demonstrate that the pro-tumorigenic role of p38MAPK in prostate cancer is more related to its effects on cell apoptosis [17] than to those involved in cell migration and invasion. Overall, all these data suggest that the role that DUSP1 plays as a tumor suppressor in prostate cancer is complex and depends on the specific inactivation of one or the other MAPK, which ultimately controls either cell apoptosis, or cell migration and invasion.

The regulatory mechanisms that control the cellular levels of Snail are very complex and involve changes at the transcriptional level or post-translational modifications, which affect its location in the cell nucleus and/or cytosol, as well as its susceptibility to degradation [16]. Here, we show for the first time that DUSP1 expression regulates the transcription of Snail. Moreover, only the concomitant ERK inhibition affects Snail expression at this level, while JNK controls it exclusively at protein level. Similar data in other cancer cell contexts have shown that the activation of Snail transcription requires an active ERK pathway [46], whereas no data on JNK involvement in this process have been reported. Regarding the regulation of Snail at a protein level, several mechanisms control the migration and invasion of prostate cancer cells by modulating the location and stability of this transcription factor. In this regard, one of the most common regulatory mechanisms is the phosphorylation of Snail by glycogen synthase kinase 3 beta (GSK-3β), which induces its nuclear export to cytosol and marks this protein for degradation in prostate cancer [47–49]. Interestingly, active ERK phosphorylates and inhibits GSK-3β, maintaining Snail in an active non-phosphorylated state and located at the cell nucleus [50]. Thus, the location of Snail in the cytosol promoted by DUSP1-dependent ERK inactivation is a possible mechanism that explains the decrease of Snail levels following DUSP1 overexpression. However, other regulatory mechanisms of Snail expression, independent of GSK-3β, have been previously identified in different tumors. For example, in hepatocarcinoma and breast cancer cells, the JNK pathway upregulates the lysil oxidase-2 (LOXL-2) [51], which oxidizes Snail, preventing its phosphorylation by GSK-3β [52]. In prostate cancer cells, elevated levels of LOXL-2 have been detected [53], supporting the possible involvement of this protein in the effects of the JNK pathway on the prostatic carcinogenesis. Alternatively, our group has previously shown that Snail expression is regulated by ERK and an autocrine loop involving transforming growth factor beta (TGFβ)/Src/focal adhesion kinase (FAK)

complex in thyroid cancer cells [28]. Similarly, other authors have demonstrated that FAK activation induces Snail expression and enhances mesothelial cell migration, promoting peritoneal metastasis from ovarian cancer [54]. Moreover, the JNK pathway activates migration by inducing the phosphorylation of paxillin, which is an adaptor protein related to FAK activation in different cancer cells [55,56]. In this regard, DUSP22, a member of the DUSP1 family which reduces JNK activation, negatively regulates cell migration through FAK dephosphorylation and inactivation in lung cancer cells [57]. Given that FAK and paxillin expression is elevated in prostate cancer and both proteins are associated with tumor progression, lymph node metastasis, and/or shortened survival [58,59], it is also plausible that in our cancer model, the paxillin/FAK pathway could contribute to the regulation of Snail expression by ERK and JNK. However, due to the difference between ERK- and JNK-dependent mechanisms, further research is required to investigate the molecular mechanisms underlying Snail regulation by these kinases.

Interestingly, we also demonstrate in this work the existence of an inverse correlation between DUSP1 and Snail expression levels in patients with different stages of prostate cancer. Importantly, in BPH and HS-PC samples, high levels of this phosphatase and low or none Snail expression were detected, while in HR-PC samples, either low or no DUSP1 expression and high Snail levels were observed. In agreement with our results, an increase in Snail expression has been related to disease progression, since there are higher levels of this protein in bone metastasis from prostate cancer compared to BPH samples [13–15]. Furthermore, other studies indicate that 66% of patients with prostatic adenocarcinoma show elevated Snail levels [60]. Here, we add new related information, demonstrating for the first time that Snail expression in patient samples is inversely correlated with DUSP1 levels and directly correlated with activated ERK and JNK pathways. In addition, the increase of active ERK in samples of HR-PC compared to those of HS-PC or BPH observed in our study is coincident with previous works. Accordingly, higher levels of phosphorylated ERK are found in samples obtained from tumors in advanced or metastatic phase, with respect to more localized tumors or BPH samples [61,62]. However, to our knowledge, this is the first study showing that the level of activated JNK is increased in prostate tumors with a more invasive phenotype, as previously seen in breast and urothelial carcinomas [63,64]. All these data obtained from the experiments carried out with patient samples confirm the results derived from our experimental cell line models and suggest that DUSP1 regulates prostate tumor progression by controlling Snail expression through ERK and JNK inactivation.

The presence of Snail has been strongly associated in prostate tumors with a high Gleason score [13,60] but not with other parameters such as the risk of recurrence or the Stage T [13]. In fact, no significant differences have been previously found in Snail expression in non-metastatic, non-recurrent cancer, recurrent cancer, or metastatic cancer at the time of diagnosis, suggesting that increased Snail expression is a relatively early event in the progress of the disease [13]. Most of the samples we analyzed in this study were locally advanced cancers. In fact, just one of our samples was graded as Gleason 6. Intermediate-risk Gleason grade 7 is usually considered as an individual group between grade 6 or lower and grade 8 or higher. Previous studies focused on the differences among the lower and the higher grades, but usually, no significant differences among grade 7 and higher grades were reported. When we correlated the expression of DUSP1, Snail, and activated ERK and JNK to clinical information, we found that their expression patterns did not correlate with either Gleason score or AJCC group staging at diagnosis. However, our results demonstrate that the pattern DUSP1low/pJNKhigh/pERKhigh/Snailhigh is closely related with a worse survival. This observation is in agreement with previous data showing that DUSP1 expression correlates with better prognosis in glioblastoma [22] and with other studies where the association of Snail expression with a worse prognosis in prostate cancer was reported [13]. Therefore, since low DUSP1 expression and high levels of Snail and activated JNK and ERK are positively associated with final outcome (death), we can conclude that besides the overall immunohistochemical profile, high levels of Snail might

be considered an independent indicator of bad prognosis that is predictive for worst outcome independently of time to progression. Moreover, since the expression pattern DUSP1high/pJNKlow/pERKlow/Snaillow is associated with an overall extended survival of patients and decreased cell migration and invasion, our results suggest that therapies based on DUSP1 induction combined with ERK and/or JNK inhibition may be promising in the treatment of metastatic prostate cancer.

#### **5. Conclusions**

Our study provides new insights about the molecular mechanisms underlying the effects of the phosphatase DUSP1 on metastasis-associated events in prostate cancer (Figure 9). In summary, our experiments show that the overexpression of this phosphatase downregulates Snail levels and decreases cell migration and invasion, whereas DUSP1 silencing shows opposite effects. Moreover, we demonstrate that DUSP1 inactivates JNK and ERK pathways. Interestingly, the inhibition of these two kinases leads to similar effects on Snail expression, cell migration, and invasion to those observed following the overexpression of this phosphatase. In addition, JNK and ERK cooperate to regulate Snail levels, cell migration, and invasion through different mechanisms. Strikingly, we also demonstrate in human prostate tissue samples an inverse correlation between DUSP1 levels and both active JNK and ERK, as well as Snail expression. Thus, we show that the expression pattern DUSP1high/pJNKlow/pERKlow/Snaillow is associated with the overall extended survival of patients. Based on all these data, we conclude that the ratio between the expression levels of DUSP1 and Snail could be an important biomarker for diagnostic purposes in prostate cancer, as they may serve for identifying patients at risk for an unfavorable clinical outcome. In addition, our results strongly suggest that the induction of DUSP1 or the inhibition of ERK and JNK pathways could be useful as a therapeutic approach to treat prostate cancer.

**Figure 9.** The phosphatase DUSP1 regulates metastasis-associated events in prostate cancer. This study demonstrate that DUSP1 overexpression downregulates Snail levels and decreases cell migration and invasion. Moreover, DUSP1 inactivates ERK and JNK pathways, whose inhibition exert similar effects on Snail expression, cell migration and invasion than overexpression of the phosphatase. In addition, JNK and ERK cooperate to regulate Snail expression, cell migration and invasion through different mechanisms. Finally, in clinical samples, the expression pattern DUSP1high/activeJNKlow/activeERKlow/Snaillow is associated with overall extended survival of patients and may serve as potential biomarker for identifying patients with favorable clinical outcome.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/2072-669 4/13/5/1158/s1, Figure S1: The phosphatase DUSP1 regulates Snail expression and migration in PC3 cells; Figure S2: The MAPKs selective inhibitors reduce MAPK activation; Figure S3: The inhibition of p38MAPK does not affect cell migration; Figure S4: JNK and ERK cooperatively regulate Snail expression and migration in PC3 cells; Figure S5: Immunohistochemical analysis showing details of subcellular localization of DUSP1, pERK, pJNK and Snail levels in samples from BPH, HS-PC and HR-PC selected from Figure 7; Figure S6: Immunohistochemical analysis of expression levels of DUSP1, pERK, pJNK and Snail in samples from three different patients diagnosed with BPH, HS-PC or HR-PC.

**Author Contributions:** Conceptualization, M.L. and A.C.; methodology, M.L., A.C., P.B., D.M.-M. and M.-V.T.L.; formal analysis, M.L., A.C., D.M.-M., M.-V.T.L. and S.R.; investigation, M.L., D.M.- M. and M.-V.T.L.; resources, M.L., A.C., P.B., M.-V.T.L., S.R., and J.C.A.; writing—original draft preparation, M.L., A.C. and M.-V.T.L.; writing—review and editing, M.L., A.C., P.B., M.-V.T.L., S.R. and J.C.A.; funding acquisition, A.C., P.B. and M.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** D.M.-M. was recipient of grants from UAM ("Post-Master Program of Dpt. Biochemistry) and from Comunidad de Madrid ("Ayudas para la contratación de investigadores predoctorales y postdoctorales, ref. PEJD-2018-PRE/BMD-8987). P.B. was recipient of a grant from Comunidad de Madrid ("Atracción al Talento Investigador", ref. 2017-T1/BMD-5704).

**Institutional Review Board Statement:** The study involving human specimens was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Ethics Committee of Hospital Universitario de Getafe (A17-11 of 10/27/2011).

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** No new data were created or analyzed in this study. Data sharing is not applicable to this article.

**Acknowledgments:** We are grateful to J. Renart (Instituto de Investigaciones Biomédicas "Alberto Sols", Madrid, Spain), and Clark (University of Birmingham, UK) for providing Snail-Luc, and pCMVDUSP1 plasmids, respectively. We are grateful to Larriba and Ferrer for their help with cell invasion assays. We thank I. Trabado (Universidad de Alcalá, Spain) for technical help.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Article* **NEK1 Phosphorylation of YAP Promotes Its Stabilization and Transcriptional Output**

**Md Imtiaz Khalil 1 , Ishita Ghosh 1 , Vibha Singh 1 , Jing Chen 2 , Haining Zhu <sup>2</sup> and Arrigo De Benedetti 1, \***


Received: 3 November 2020; Accepted: 4 December 2020; Published: 7 December 2020

**Simple Summary:** We earlier described the involvement of the TLK1>NEK1>ATR>Chk1 axis as a key determinant of cell cycle arrest in androgen-dependent prostate cancer (PCa) cells after androgen deprivation. We now report that the TLK1>NEK1 axis is also involved in stabilization of yes-associated protein 1 (YAP1), the transcriptional co-activator in the Hippo pathway, presumably facilitating reprogramming of the cells toward castration-resistant PCa (CRPC). NEK1 interacts with YAP1 physically resulting in its phosphorylation of 6 residues, which enhance its stability and activity. Analyses of cancer Protein Atlas and TCGA expression panels revealed a link between activated NEK1 and YAP1 expression and several YAP transcription targets.

**Abstract:** Most prostate cancer (PCa) deaths result from progressive failure in standard androgen deprivation therapy (ADT), leading to metastatic castration-resistant PCa (mCRPC); however, the mechanism and key players leading to this are not fully understood. While studying the role of tousled-like kinase 1 (TLK1) and never in mitosis gene A (NIMA)-related kinase 1 (NEK1) in a DNA damage response (DDR)-mediated cell cycle arrest in LNCaP cells treated with bicalutamide, we uncovered that overexpression of wt-NEK1 resulted in a rapid conversion to androgen-independent (AI) growth, analogous to what has been observed when YAP1 is overexpressed. We now report that overexpression of wt-NEK1 results in accumulation of YAP1, suggesting the existence of a TLK1>NEK1>YAP1 axis that leads to adaptation to AI growth. Further, YAP1 is co-immunoprecipitated with NEK1. Importantly, NEK1 was able to phosphorylate YAP1 on six residues in vitro, which we believe are important for stabilization of the protein, possibly by increasing its interaction with transcriptional partners. In fact, knockout (KO) of NEK1 in NT1 PCa cells resulted in a parallel decrease of YAP1 level and reduced expression of typical YAP-regulated target genes. In terms of cancer potential implications, the expression of NEK1 and YAP1 proteins was found to be increased and correlated in several cancers. These include PCa stages according to Gleason score, head and neck squamous cell carcinoma, and glioblastoma, suggesting that this co-regulation is imparted by increased YAP1 stability when NEK1 is overexpressed or activated by TLK1, and not through transcriptional co-expression. We propose that the TLK1>NEK1>YAP1 axis is a key determinant for cancer progression, particularly during the process of androgen-sensitive to -independent conversion during progression to mCRPC.

**Keywords:** tousled-like kinase (TLK); NIMA-related kinase 1 (NEK1); yes-associated protein 1 (YAP1); thioridazine (THD); MS-determined phosphopeptides

#### **1. Introduction**

The founding member of the NIMA (never in mitosis gene A) family of protein kinases was originally identified in *Aspergillus nidulans* as a protein kinase essential for mitosis [1], and expression of a dominant-negative mutant of NIMA results in G2 arrest in vertebrate cells [2]. NIMA-related kinases (NEKs) have adapted to a variety of cellular functions in addition to mitosis [3]. In human cells, 11 NEKs were identified that are involved in several functions. For example, NEK2 is critical for centrosome duplication [3], whereas NEK6, 7, and 9 are regulators of the mitotic spindle and cytokinesis [4]. NEK1, NEK4, NEK8, NEK10, and NEK11 have been linked to the DNA damage response (DDR) and DNA repair pathways as well as ciliogenesis [3]. NEK1 mediates Chk1 activation likely by modulating the ATRIP/ATR interaction and activity [5], although this may be controversial [6]. NEK1 activity and relocalization to nuclei were reported to increase upon a variety of genotoxic stresses [5,7]. A defect in DNA repair in NEK1-deficient cells is suggested by the persistence of Double Strand Breaks (DSBs) after low-dose ionizing radiation (IR). NEK1-deficient cells fail to activate the checkpoint kinases Chk1 and Chk2, and fail to arrest properly at G1/S- or G2/M-phase checkpoints after DNA damage [8]. NEK1-deficient cells suffer major errors in mitotic chromosome segregation and cytokinesis, and become aneuploid [9]. Genomic instability is also manifested in NEK1+/<sup>−</sup> mice, which later in life develop lymphomas with a higher incidence than wild type littermates [9]. NEK1 is also known to negatively regulate apoptosis by phosphorylating VDAC1, regulating the closure of the anion channel of the mitochondrial membrane, which promotes survival of renal cell carcinoma [10–12]. Loss of function mutation of NEK1 leads to DNA damage accumulation in the motorneurons that may lead to several neurodegenerative diseases such as amyotrophic lateral sclerosis (ALS) [13,14]. NEK1 is associated with primary cilia and centrosomes [15,16], which was reported to be implicated in the development of polycystic kidney disease (PKD) when there is a NEK1 deficiency [17]. However, the precise mechanism leading to PKD due to NEK1 insufficiency is not clear, but a clue came from the discovery that NEK1 interacts with and phosphorylates TAZ, involved in the E3 ligase complex, which regulates the stability of polycystin 2 [18]. TAZ is also a paralog of yes-associated protein (YAP), a transcriptional coactivator that mediates many functions in normal development and in disease pathology, such as cancer progression, including prostate cancer [19–22].

We recently uncovered a new DDR axis involving the protein kinase tousled-like kinase (TLK)1 as an early mediator of the DDR. TLK1 serves as an upstream activator of NEK1>ATR>Chk1 [6,23], which has important implications during the early stages of prostate cancer (PCa) progression to androgen independence (AI) [24,25]. We found that overexpression of wt-NEK1 (but not the T141A kinase-hypoactive mutant that cannot be phosphorylated by TLK1) hastens the progression of LNCaP cells to androgen-independent growth [24]. The protective cell cycle arrest mediated by the TLK1>NEK1 DDR pathway seems insufficient to explain the rapid growth recovery observed in bicalutamide-treated cells when NEK1 is overexpressed, and suggests that NEK1 may have additional functions. We suspected that it may regulate the Hippo pathway, as it was reported that ectopic expression of YAP is sufficient to convert LNCaP cells from androgen-sensitive (AS) to AI in vitro [19]. NEK1 was also found to phosphorylate TAZ specifically at S309 [18], and this was related to increased CTGF expression (one of TAZ/YAP transcriptional targets). TLKs may regulate the Hippo pathway through their activity on NEK1 upstream of YAP/TAZ. YAP/TAZ (60% identical) are the main effectors of the Hippo signaling pathway. This pathway is involved in regulating organ size through controlling multiple cellular functions including cell proliferation and apoptosis [26]. The Hippo pathway responds to a variety of signals, including cell–cell contact, mechano-transduction [21], and apico–basal polarity [20,26]. When the Hippo pathway is activated, kinases MST1/2 and LATS1/2 phosphorylate and inactivate YAP and TAZ. YAP and TAZ are transcriptional co-activators but lack DNA binding activity. Upon phosphorylation by MST and LATS kinases, they are sequestered in the cytoplasm, ubiquitylated by the β-TrCP ubiquitin ligase, and marked for proteasomal degradation (reviewed in [20]). YAP/TAZ are usually inhibited by cell–cell contact in normal tissues [26], while over-activation of YAP/TAZ through aberrant regulation of the Hippo pathway has been noted in many types of tumors. This is associated

with the acquisition of malignant traits, including resistance to anticancer therapies; maintenance of cancer stem cells; distant metastasis [26]; and, in prostate, adenocarcinoma progression [27,28]. When the Hippo core kinases are "off", YAP/TAZ translocate into the nucleus, binds to TEAD1–4, and activates the transcription of TEAD downstream target genes, leading to multiple oncogenic activities, including loss of contact inhibition, cell proliferation, epithelial–mesenchymal transition, and resistance to apoptosis. In PCa, YAP has been identified as an Androgen Receptor-binding partner that colocalizes with AR in both androgen-dependent and androgen-independent manners in castration-resistant PCa (CRPC) patients [27]. YAP is also found to be upregulated in AI-LNCaP-C4-2 cells and, when expressed ectopically in LNCaP cells, it activates AR signaling and confers castration resistance. Knockdown of YAP greatly reduces the rates of migration and invasion of LNCaP, and YAP-activated androgen receptor signaling is sufficient to promote LNCaP cells from an AS to an AI state in vitro, while YAP conferred castration resistance in vivo [19]. It was also recently determined that ERG (and the common *TMPRSS2*–*ERG* fusion) activates the transcriptional program regulated by YAP1, and that prostate-specific activation of either ERG or YAP1 in mice induces similar transcriptional changes and results in age-related prostate tumors [29]. However, it has remained unclear as to what the upstream activators of the Hippo pathway are in PCa, and we show in this report that TLKs have a role in this process via activation and induced stabilization of YAP from elevated phosphorylation by NEK1.

#### **2. Materials and Methods**

#### *2.1. Plasmids and Antibodies*

Wild type human full length NEK1 mammalian expression plasmid was purchased from Origene (MR216282). NEK1 T141A variant was generated by site-directed mutagenesis, as previously described [23]. Generation of His-tagged N-terminal NEK1 (aa 1–480) bacterial expression plasmid was conducted as previously described. Human full length MK5 bacterial expression plasmid was purchased from Vector Builder. The following antibodies were used in this study: mouse anti-YAP (Santa Cruz Biotechnology, SCBT, Dallas, TX, USA, cat# sc101199), rabbit anti-phospho-YAP (Cell Signaling Technology, CST, Dallas, TX, USA, cat# 13008), mouse anti-NEK1 (SCBT, cat# sc 398813, Dallas, TX, USA), rabbit anti-phospho-NEK1 pT141 (lab-generated), rabbit anti-phospho-tyrosine (CST, cat# 8954S, Dallas, TX, USA), HRP-conjugated anti-β-tubulin (SCBT, Dallas, TX, USA, cat# sc-23949), mouse IgG (SCBT, Dallas, TX, USA, cat# sc-2025), and rabbit anti-actin (Abcam, Cambridge, MA, USA, cat# ab1801).

#### *2.2. Cell Culture*

Human embryonic kidney HEK293 and HeLa cells were cultured in Dulbecco Modified Eagle Medium (DMEM) supplemented with 10% Fetal Bovine Serum (FBS) and 1% penicillin/streptomycin. HEK293T cells were cultured in D10 medium containing 10% FBS, 0.25% penicillin/streptomycin, and 1% glutamine in DMEM media. LNCaP cells were cultured in Roswell Park Memorial Institute (RPMI) 1640 supplemented with 10% FBS and 1% penicillin/streptomycin. NT1 cells were a kind gift from Dr. Xiuping Yu (Department of Biochemistry, Louisiana State University Health Sciences Center Shreveport) and cultured according to the published literature [30]. All other cells were purchased from American Type Culture Collection (ATCC). All the cells were maintained in a humidified incubator at <sup>37</sup> ◦C with 5% CO2.

#### *2.3. Cell Treatment*

LNCaP or HeLa or NT1 cells were plated as 5 × 10<sup>5</sup> cells per well in a 6-well plate and grown until 70–80% confluency. Cells were treated with either 10 µM of either bicalutamide (Selleckchem, Houston, TX, USA, cat# S1190), thioridazine (THD; Sigma Aldrich, St. Louis, MO, USA, cat# T9025 or J54 [31], or in combination with both bicalutamide and THD for 24 h. After the treatment, cells were harvested for Western blotting (WB) analysis or qPCR analysis.

#### *2.4. Cell Transfection*

LNCaP cells were transfected with either wild type mouse full-length NEK1 or NEK1 T141A variant, as previously described [23]. TLK1 shRNA (ATTACTTCATCTGCTTGGTAGAGGTGGCT) was obtained from origene (Rockville, MD, USA, cat# TR320623). HeLa cells were plated as 10<sup>5</sup> cells per well in a 6-well plate 24 h before shRNA transfection. Transfection was conducted using 140 nM and 280 nM of TLK1 shRNA by lipofectamine 3000 (Thermo Scientific, Waltham, MA, USA, cat# L3000-015) reagent for 24 h, following the manufacturer's protocol, and subsequently selected the cells with 1 µg/mL of puromycin for 7 days. Puromycin-selected cells were harvested and knockdown efficiency was determined by WB.

#### *2.5. Co-immunoprecipitation (co-IP)*

Cells were lysed by sonication in 1X RIPA lysis buffer (SCBT, Dallas, TX, USA, cat# 24948). A total of 50 µL of equilibrated protein A/G agarose (SCBT, Dallas, TX, USA, cat# sc-2003) was incubated with either mouse anti-NEK1 antibody or mouse IgG antibody at 4 ◦C for 4 h with rotation. A total of 500 µg of protein lysate was added to the reaction and incubated overnight at 4 ◦C. Beads were washed thrice and eluted with 25 µL of 2X SDS-Laemmli buffer, and the entire volume was loaded into SDS-PAGE gel for WB analysis.

#### *2.6. Generation of NT1 NEK1 Knockout (KO) Cells Lines*

NT1 NEK1 KO clones were generated by lentiviral infection using NEK1 CRISPR gRNA (AAGGAGAGAAGTTGCTGTAT) cloned into pLentiCRISPR V2 vector backbone from Genscript (Piscataway, NJ, USA). Lentivirus containing NEK1 CRISPR gRNA was packaged using HEK293T cells. NT1 cells were infected with lentivirus using polybrene transfection reagent following standard protocol. After 72 h of infection, cells were supplemented with fresh media and selected with 1–2 µg/mL of puromycin for 10 days. To generate a single clonal population of NEK1 KO cells, we seeded 1–2 cells per well in a 96-well plate and grew them until confluency, and then transferred them to a bigger dish for expansion. KO efficiency was measured by Western blotting (WB) using anti-NEK1 mouse antibody from Santa Cruz Biotechnology (Dallas, TX, USA, cat# sc-398813).

#### *2.7. Protein Purification*

Recombinant His-tagged full-length MK5 and His-tagged NEK1 N-terminal-truncated proteins (NEK1∆CT) were purified by affinity chromatography. Both MK5 and NEK1∆CT were transformed into Rosetta2 DE3 strain [23]. Expression of His-MK5 was induced with 1mM Isopropyl βd-1-thiogalactopyranoside (IPTG) at 37 ◦C for 3–4 h, and His-NEK1∆CT expression was induced with 0.5 mM IPTG overnight at 25 ◦C. Bacteria were pelleted down; dissolved in buffer containing 50 mM sodium phosphate (Na2HPO<sup>4</sup> + NaH2PO4) of pH 8.0, 300 mM NaCl, 20 mM imidazole, and 1mM phenylmethylsulfonyl fluoride (PMSF); and lysed by sonication. Supernatants were incubated with Ni-NTA agarose (Qiagen, cat# 30210), and protein was eluted in buffer containing 50 mM sodium phosphate (Na2HPO<sup>4</sup> + NaH2PO4) of pH 8.0, 300 mM NaCl, and 250 mM imidazole. Eluted proteins were dialyzed overnight at 4 ◦C using dialysis buffer containing 20 mM sodium phosphate (Na2HPO<sup>4</sup> + NaH2PO4) of pH 7.7, 1 M NaCl, 10 mM β-mercaptoethanol, 0.5 mM ethylenediaminetetraacetic acid (EDTA) of pH 8.0, and 5% glycerol. After the dialysis, protein samples were run in SDS-PAGE gel to check their purity and correct molecular weight.

#### *2.8. ADP Hunter Assay*

ADP hunter assays were conducted to determine the catalytic activity of the purified kinases by the fluorescence detection of ATP to ADP conversion using an ADP Hunter Plus Assay kit (Eurofins, DeSoto, TX, USA, cat# 90-0083). Increasing amount of purified recombinant NEK1 or MK5 were incubated with either dephosphorylated α-casein (substrate for NEK1, source: Sigma-Aldrich, St. Louis, Missouri, USA, cat# C8032) or purified recombinant HSP27 (substrate for MK5, source: Abcam, Cambridge, MA, USA, cat# ab48740). The manufacturer provided kinase buffer, and 50 µM of ATP was added to the reaction, incubating the reaction at 30 ◦C for 30 min. Afterwards, reagent A and B were added sequentially, incubating the reaction at room temperature for 30 min. Stop solution was added and fluorescence intensity signal was measured at 530/590 nm excitation/emission wavelength. ADP concentration was determined by the standard curve through the serial dilutions of the ADP standards provided with the kit.

#### *2.9. In Vitro Kinase Assay*

In vitro kinase (IVK) assays were performed using purified recombinant proteins, kinase buffer, ATP, and/or [γ-<sup>32</sup>P] ATP. Purified recombinant GST-tagged YAP1 (Novus Biologicals, cat# Centennial, CO, USA, H00010413-P01) was incubated with either purified recombinant His-NEK1∆CT or purified recombinant His-tagged MK5. Kinase buffer (10X) contains 10 mM Tris-Cl of pH 7.5, 10 mM MgCl2, 10 mM dithiothreitol (DTT), and 10 mM ATP. For radioactive IVK assays, we added 10µCi of radiolabeled [γ-<sup>32</sup>P] ATP purchased from Perkin Elmer (cat# BLU002H250UC). The reactions were incubated for 30 min at 30 ◦C and subsequently were separated by SDS-PAGE, stained with Coomassie Brilliant Blue, and exposed to X-ray film for 72 h. For mass spectrometric (MS) analysis, YAP1 bands were excised after Coomassie staining and sent to the Kentucky MS facility.

#### *2.10. Identification of YAP1 Phosphorylation by Mass Spectrometry*

The band corresponding to YAP1 was excised and subjected to dithiothreitol reduction, iodoacetamide alkylation, and in-gel chymotrypsin digestion. Peptides were extracted, concentrated, and subjected to LC–MS/MS analysis at the University of Kentucky Proteomics Core Facility, as previously reported [32]. Briefly, LC–MS/MS analysis was performed using an LTQ-Orbitrap mass spectrometer (Thermo Fisher Scientific, Waltham, MA) coupled with an Eksigent Nanoflex cHiPLC system (Eksigent, Dublin, CA, USA) through a nano-electrospray ionization source. The peptide samples were separated with a reversed-phase cHiPLC column (75 µm × 15 cm) at a flow rate of 300 nL/min. Mobile phase A was water with 0.1% (*v*/*v*) formic acid, while B was acetonitrile with 0.1% (*v*/*v*) formic acid. The data-dependent acquisition method consisted of an Orbitrap MS scan (250–1800 m/z) with 60,000 resolution for parent ions, followed by MS/MS for fragmentation of the 10 most intense multiple charged ions. The LC–MS/MS data were submitted to a local Mascot server for MS/MS protein identification via Proteome Discoverer (version 1.3, Thermo Fisher Scientific, Waltham, MA, USA). Typical parameters used in the Mascot MS/MS ion search were chymotrypsin digestion with a maximum of two miscleavages; 10 ppm precursor ion and 0.8 Da fragment ion mass tolerances; and dynamic modifications, including cysteine carbamidomethylation, methionine oxidation, and serine/threonine/tyrosine phosphorylation. The identified phosphorylation sites were illustrated with relevant b and/or y ions labeled.

#### *2.11. Western Blotting*

Cells were collected and lysed by sonication in 1X RIPA lysis buffer. Protein concentration was determined using a Pierce BCA protein assay kit (Thermo Scientific, cat# 23225, Waltham, MA, USA). Samples from the lysate or co-IP or IVK assays were separated by SDS-PAGE gels and transferred to polyvinylidene fluoride (PVDF) membrane. The membrane was blocked in 5% non-fat dry milk for 1 h at room temperature and incubated with primary antibodies overnight at 4 ◦C. Afterwards, HRP-conjugated secondary antibodies were used to incubate the blots for 1 h at room temperature, and finally the specific proteins were detected by chemiluminescence using ECL substrates (Thermo Scientific, Waltham, MA, USA, cat# 32106) or by colorimetry using Opti-4CN substrate kit (Biorad, cat# 1708235, Waltham, MA, USA). The membrane was visualized by Biorad chemidoc imaging system (Biorad, Hercules, CA, USA, cat# 12003154). Densitometric quantifications of each blot in arbitrary units relative to the loading control are shown in Figure S5.

#### *2.12. Real-time Quantitative PCR (RT-qPCR)*

Total RNA was isolated using a RNeasy RNA isolation minikit (Qiagen, cat# 74104, Germantown, MD, USA) according to the manufacturer's instructions. Complementary DNA (cDNA) was synthesized using 1µg of RNA/reaction using ProtoScript First Strand RNA synthesis reverse transcriptase and oligo (dT) primers (New England Biolab, cat# E6300L, Ipswich, MA, USA). qPCR was conducted using iQ SYBR green supermix (Biorad, cat# 1708880, Des Plaines, IL, USA) and Bio-Rad CFX96 Fast Real-Time PCR Systems. Gene expression changes were determined by ∆∆Ct relative quantification method. GAPDH mRNA was used as an internal control. All values are presented as mean ± standard error mean (SEM).

#### *2.13. Bioinformatics Analysis*

mRNA expression analyses of TCGA patient datasets were conducted using the UALCAN online platform [33]. Oncoprints of the NEK1 and YAP1 protein level of at least more than 0.5-fold increase was generated using Cbioportal [34] from The Cancer Genome Atlas (TCGA-firehose legacy) datasets. Proteomic level of NEK1 and YAP1 based on the immunohistochemistry (IHC) analysis in different cancers were determined using the Human Protein Atlas [35] database. Representative IHC images of high-grade prostate adenocarcinoma (PRAD) and metastatic head and neck squamous cell (HNSC) carcinoma were also obtained from the Human Protein Atlas database. Volcano plot of gene enrichment correlated with NEK1 upregulation of TCGA (firehose legacy) head and neck cancer study was generated using the cBIOPORTAL web tool.

#### **3. Results**

#### *3.1. NEK1 Regulated the Stability of YAP*

We have previously reported that androgen deprivation in LNCaP cells results in a strong increase in expression of TLK1B. This increase is mTOR-dependent and suppressible with rapamycin [24]. Similar results were obtained with TRAMP-C2 cells [24], and more recently in a AR+/PDX adenocarcinoma model (NSG-TM00298 [25]). This is apparently a critical survival mechanism of AS-PCa cells that implement a DDR in order to arrest in G1 upon androgen deprivation-like treatment with bicalutamide (BIC) [36]. We have recently attributed the probable mechanisms causing this DDR activation to the role played by the AR as a replication licensing factor [37] in combination with the increased expression of TLK1B, and resulting activation of the NEK1>ATR>Chk1 axis [24], which is a key target of TLK1 [23]. Additional work from our lab suggested that this may be a conserved nexus in other cellular models, in the TRAMP mice, and probably in many patients, since the specific activating phosphorylation of NEK1 by TLK1 correlates with the Gleason score [25]. While the significance of the cell cycle arrest upon unfavorable growth conditions (androgen deprivation therapy, ADT) seems clear in order to avoid mitotic catastrophe, it is still unclear how AS-PCa cells eventually adapt to ADT and reprogram to become AI (CRPC progression). Interestingly, we have previously noticed that when LNCaP cells were stably transfected with a wt-NEK1 expression vector, they rapidly (less than 1 week) became tolerant to BIC and resumed growth to form AI colonies [24]. However, this did not happen when we expressed the hypoactive T141A-NEK1 variant [23] that cannot be phosphorylated/activated by TLK1, while these cells also remained AS when injected as xenografts [24]. The rapid resumption of growth of LNCaP-NEK1 cells in the presence of anti-androgen (BIC) could not be readily explained by the implementation of the pro-survival DDR checkpoint, suggesting that NEK1 also promotes the AI conversion. On the basis of a review of the literature (see the Introduction), we suspected that NEK1 may affect the Hippo pathway, and thus we carried out a Western blotting (WB) analysis of YAP expression in LNCaP cells overexpressing wt-NEK1 or NEK1-T141A variant. The cells were also treated or not treated with BIC and thioridazine (THD), which is a rather specific inhibitor of TLKs [38]. In Figure 1A (quantitation in Figure S5), we show that overexpression of the NEK1-T141A variant results in reduced levels of YAP (lane 1 vs. 5), along with evidence of an elevated cleaved

product (Cl-YAP). Decreased YAP levels and evidence of Cl-YAP were also seen in parental LNCaP cells treated with THD (+/− BIC, lane 3 and 4 vs. 1). In contrast, LNCaP cells that overexpress wt-NEK1 showed elevated expression of YAP and no evidence of Cl-YAP (lanes 9–12), where a possible mechanism is that the phosphorylation of YAP by elevated wt-NEK1 mediates a process of stabilization to counteract its degradation when TLK activity (upstream of NEK1) is suppressed with THD (lane 11 vs. 3). Furthermore, the expression of typical YAP/TEAD-dependent transcripts such as CTGF, CDH2 (N-cadherin), Twist1, and TP53AIP1 were decreased in LNCaP cells treated with THD, while in contrast, the expression of CDH1 (E-cadherin) that drives MET was slightly increased (Figure 1D). −

**Figure 1.** (**A**) The expression of yes-associated protein (YAP) was regulated by never in mitosis gene A (NIMA)-related kinases (NEK) activity and its upstream kinase tousled-like kinase (TLK). Overexpression of wt-NEK1 resulted in elevated YAP expression and conversely in its degradation in LNCaP cells overexpressing the dominant negative mutant NEK1-T141A. Thioridazine (THD) led to degradation of YAP in parental LNCaP cells, even after treatment with bicalutamide (BIC), which led to overexpression of TLK1B. (**B**) YAP interacted with NEK1 and was enriched upon co-immunoprecipitation. TLK1 inhibition with 10 µM THD did not affect NEK1 interaction with YAP, and thus the state of NEK1 kinase activity did not affect YAP binding. (**C**) The expression of YAP was decreased in NT1 cells treated with two different inhibitors of TLK (THD and J54), with a corresponding increase in CL-YAP products. (**D**) Expression of several typical YAP target genes in LNCaP cells treated with THD.

We also show that NEK1 interacted with YAP, as it was enriched by co-IP, and their association was not altered by THD (Figure 1B, top panel; quantitation in Figure S5), indicating that the NEK1 kinase activity was independent of its ability to interact with YAP. As we previously reported [23], the same co-IP also brought down TLK1, and THD did not affect their interaction (Figure 1B, bottom panel; quantitation in Figure S5). There is a possibility that in cells, NEK1, TLK1, and YAP are in a complex, or that NEK1 interacts independently with TLK1 and YAP. In either case, TLK1 was not found to interact directly with YAP [23].

To confirm the effect of inhibition of the TLK1>NEK1 axis on the expression of YAP in a different PCa cell line, we treated Neo-TAg1 (NT1) with two different inhibitors of TLK1: THD or J54. This resulted in a reduction of YAP level and appearance of a set of cleavage products (Figure 1C; quantitation in Figure S5).

### *3.2. NEK1 KO in NeoTag1 Cells Resulted in Reduced YAP Levels and Expression of Several of Its Target Genes*

Consistent with our initial observations that NEK1 activity is critical for YAP stabilization, we found that YAP expression was concomitantly reduced in CRISPR-mediated KO of NEK1 in the PCa line NT1 (Figure 2A; quantitation in Figure S5). Likewise, the expression of several YAP target genes (e.g., CTGF, Zeb1, Twist1) that drive Epithelial to Mesenchymal Transition (EMT) and invasiveness of these cells was suppressed in all the positive NEK1 KO clones (Figure 2B). Conversely, inhibition of TLK1 with THD, which we showed leads to reduced NEK1 activity [23], can inhibit cell migration via suppression of EMT-related genes such as Claudin1, E-cadherin, N-cadherin, Twist1, Snail3, Slug, FOXC2, MMP3, and MMP9 in Hepato Cellular Carcinoma (HCC) cells [39]. We now suggest this observation derives from reduction of YAP expression concomitant with loss of NEK1 (activity) due to inhibition of TLK. In fact, we showed in Figure 2C (quantitation in Figure S5) that YAP expression was reduced in LNCaP cells treated with THD, while conversely, pYAP(S127), which is a phospho-degron leading to its proteasomal degradation, was elevated.

**Figure 2.** (**A**) CRISPR/Cas9-mediated loss of NEK1 resulted in reduced levels of YAP protein, possibly due to instabilization (EV = empty vector). (**B**) Expression of several typical YAP target genes is reduced in NEK1 KO clones. GAPDH mRNA was used as an internal control. (**C**) Treatment of LNCaP cells with THD, a specific inhibitor of TLKs, resulted in reduced YAP protein level and conversely in its S127 hyperphosphorylation. (**D**) Reduction of TLK1 expression via (short hairpin) shRNA transfection led to loss of pNEK1-T141.

In order to confirm with a genetic approach that the inhibition of TLK1 results in suppression of the pathway that leads to activation of NEK1 and subsequent stabilization of YAP, we knocked down TLK1 with shRNA in HeLa cells. Effective knockdown of TLK1 was achieved in a dose-dependent manner with the shRNA (Figure 2D; quantitation in Figure S5), and importantly, activated NEK1 levels, i.e., pNEK1(T141) were similarly suppressed. This suggests that at least in these cells, TLK1 is the principal kinase responsible for the phosphorylation and activation of NEK1—note that the T141 residue resides in the kinase domain of the protein adjacent to the activation loop [40] that we have previously shown to be important for NEK1 kinase activity [23].

#### *3.3. NEK1 Phosphorylated YAP In Vitro on Several Residues*

In order to determine if NEK1 could phosphorylate YAP in vitro, we first purified a recombinant His-tagged NEK1-NT fragment spanning nearly half of the entire protein (total NEK1 protein = 1258 AA, Singh et al. (2017) [ref 23]) following standard protocol and determined its catalytic activity using dephosphorylated α-casein by ADP Hunter assay (Figure 3A,B, see the Section 2). ADP hunter assay revealed that our lab-purified truncated NEK1 is catalytically active, as the incubation of increasing amounts of NEK1 resulted in corresponding ATP to ADP conversion (Figure 3B). Afterwards, we carried out a preliminary in vitro kinase (IVK) reaction by incubating purified recombinant His-tagged NEK1 with purified recombinant GST-YAP (Novus Biologicals) and [γ-<sup>32</sup>P] ATP. For comparison, we also carried out the IVK reaction using recombinant MK5, which was recently reported to be a novel YAP1 kinase [41]. The purity of all recombinant proteins is shown in the Coomassie Blue-stained SDS/PAGE, and the autoradiography of the gel is shown above it (Figure 3C; quantitation in Figure S5). Notably, NEK1 was capable of strongly phosphorylating YAP, even when small amounts were used (see stained gel). In contrast, MK5 (even in high amount) was a very weak kinase for YAP, if at all, although it was clearly highly active since it was capable of auto-phosphorylation (see autoradiogram) and when tested with ADP Hunter reagent. α

∆ ∆ γ **Figure 3.** (**A**) Expression and purification of His-NEK1 kinase domain (NEK1∆CT). (**B**) NEK1∆CT was catalytically active and ATP/ADP conversion (kinase activity) was linear with the enzyme amount. (**C**) In vitro phosphorylation reactions of YAP using His-NEK1 and MK5 kinases in presence of [γ-<sup>32</sup>P] ATP. (**D**) In vitro phosphorylation of YAP using His-NEK1 and MK5 kinases for preparative isolation for MS determination of phosphopeptides. (**E**) His-NEK1 also phosphorylated YAP on Tyr, as demonstrated by immunoreactivity with pY antibody.

The IVK reactions were repeated with greater amounts of proteins for preparative isolation for MS analysis for assignment of the phosphorylated residues (Figure 3D). The bands corresponding to YAP incubated with NEK1, MK5, or mock were excised. Determination of the phosphorylated peptides and assignment of the phospho-amino acids were carried out at the University of Kentucky Proteomics facility. The YAP bands were digested with chymotrypsin and analyzed with an LTQ-Orbitrap mass spectrometer. MS datasets were searched with MASCOT against a custom database containing only human YAP1 and NEK1. A synopsis of the results is that (1) when searched against YAP1 and NEK1, only YAP was detected in these samples (well separated on the gel), with 43–49% peptide coverage and protein scores of 2573-3321; (2) potential phosphorylation sites S163/S164 were detected in all three samples (including the YAP1 no kinase sample), which can be explained as a basal phosphorylated

residue of recombinant YAP isolated from wheat germ; and (3) six unique phosphorylation sites were detected in the YAP\_NEK1 sample: T83, T361, S366, S388, S406, Y407, or T493 (Figure 4 and Figures S1–S4; Table S1). However, no unique phosphorylation site was detected in the YAP\_MK5 sample, which we now suggest is not an authentic YAP kinase. Interestingly, in the paper that purported MK5 as an important YAP kinase, the authors did not report whether they attempted to verify that MK5 can phosphorylate YAP in vitro, nor did they identify the phosphorylation target in vivo [41]. In Figure 4, we present an example of data identifying Y407 and T493, which we currently assume are the most interesting.

**Figure 4.** MS/MS spectra demonstrating the phosphorylation sites at T493 (**A**) and S406/Y407 (**B**) as examples of LC–MS/MS determinations.

All of the phosphorylated residues listed in Table S1 have been reported in MS studies in cells, according to the report of Phosphosite Plus, except for S406 (putative) and T493, which, as such, are the first report of phosphorylation of these residues specifically by NEK1. The phosphorylation of Y407 (putative) should not be surprising, since NEK1 is a dual specificity kinase that was originally identified as a tyrosine kinase [42]. Note that although the MS/MS spectrum could not distinguish the exact

phosphorylation site at S406 or Y407, a phospho-Tyr Western blot (Figure 3E; quantitation in Figure S5) supported the conclusion that Y407 (the only identified pTyr in the MS analysis) was phosphorylated. It is also noteworthy that the NEK1 protein was also phosphorylated on Tyr (Figure 3E), as we previously reported that it is in fact auto-phosphorylated on Y315 [23], confirming the specificity of the antiserum.

#### *3.4. Bioinformatic Studies Suggest NEK1 Mediated Stabilization of YAP1 in Di*ff*erent Cancers*

We analyzed mRNA expression of both NEK1 and YAP1 in prostate adenocarcinoma (PRAD) and head and neck squamous cell (HNSC) carcinoma patients from TCGA datasets using the UALCAN online platform. In PRAD, no significant alteration in mRNA expression of NEK1 was observed (Figure 5A), while YAP1 mRNA level was consistently downregulated with respect to the tumor Gleason score (Figure 5B). However, reverse phase protein array (RPPA)-based protein profiling of NEK1 and YAP1 in PRAD patients from TCGA datasets revealed upregulation of YAP1 level (Figure 5D), but no change in NEK1 protein level (Figure 5C). In addition, proteomic analysis based on immunohistochemistry (IHC) data from the Human Protein Atlas web server revealed a higher protein level of NEK1 and YAP1 in high-grade PRAD patients (Figure 5G,H). Representative IHC analysis revealed intense staining of both NEK1 and YAP1 in high-grade PRAD compared to normal prostate tissue (Figure 5I). This supports our hypothesis of NEK1 implication in YAP1 protein stabilization/accumulation in advanced PCa, despite YAP1 transcript downregulation.

**Figure 5.** Gene expression of (**A**) NEK1 and (**B**) YAP1 of prostate adenocarcinoma (PRAD) patients on the basis of the Gleason score extracted from The Cancer Genome Atlas (TCGA) datasets using the UALCAN web tool. OncoPrint representation of the protein level alteration of (**C**) NEK1 and (**D**) YAP1 of PRAD patients by reverse phase protein array (RPPA) extracted from TCGA (firehose legacy) datasets using cBIOPORTAL online platform. Gene expression of (**E**) NEK1 and (**F**) YAP1 in head and neck squamous cell (HNSC) patients on the basis of the tumor grade extracted from TCGA datasets using the UALCAN web tool. Percentage of patients of different types of cancer with higher level of (**G**) NEK1 and (**H**) YAP1 on the basis of immunohistochemistry (IHC) staining generated using the Human Protein Atlas database. Staining intensity correlated with the color code. Deeper color represents high staining intensity. (**I**) Representative IHC images of NEK1 and YAP1 of high-grade PRAD (top panel) and metastatic HNSC samples (bottom panel). (**J**) Volcano plot of gene enrichment analysis based on NEK1 overexpression in HNSC patients extracted from TCGA (firehose legacy) datasets using cBIOPORTAL online platform. \* represents *p* < 0.05, \*\* represents *p* < 0.005, and \*\*\* represents *p* < 0.0005. All comparisons were with the normal tissue.

Similarly, in head and neck squamous cell carcinoma (Figure 5F), glioblastoma, and other cancers (data not shown), there was no significant upregulation of YAP1 mRNA expression; nonetheless, YAP1 protein level was elevated in high-grade metastatic tumors (Figure 5). Moreover, gene set enrichment analysis significantly correlated NEK1 expression with several YAP1 target genes such as Zeb1, BirC2, BirC6, Ankrd11, and ARID1B (Figure 5J; Table 1). Overall, these data suggest NEK1 increases YAP1 level by reducing YAP1 protein turnover rate in different cancers.



#### **4. Discussion**

During studies aimed at elucidating the process of ADT adaptation of AS PCa cell (initially in LNCaP), which proceeds through a process of activating the DDR and increased activity of the kinases TLK1B and NEK1 [11,24,25], we made the observation that overexpression of wt-NEK1, but not the hypoactive NEK1-T141A variant that cannot be activated by TLK, resulted in a rapid adaptation to bicalutamide and formation of AI colonies. From a review of the literature on the process of AI conversion of LNCaP and other studies of CRPC progression, we suspected the involvement of Hippo pathway deregulation and, in particular, YAP-driven gene expression (for a recent review, see [43]). Moreover, Yim et al. reported that NEK1 can phosphorylate TAZ and regulates its turnover rate [18]. Since YAP1 and TAZ are two highly homologous proteins that possess several conserved phospho-residues, we set out to investigate the protein level of YAP in LNCaP overexpressing wt-NEK1 and the T141A mutant in conjunction with a TLK inhibitor (THD) to suppress the activating phosphorylation of NEK1. Interestingly, we observed an increased degradation of YAP in cells overexpressing NEK1-T141A mutant or parental LNCaP treated with THD, in contrast to elevated level of YAP (and no degradation) in cells that overexpress wt-NEK1 (Figure 1). Furthermore, treatment of LNCaP cells with THD resulted in downregulated expression of several YAP-dependent transcripts (Figure 1D). As an indication that this is in fact a general phenomenon in PCa, increased degradation of YAP1 after inhibition of the TLK1>NEK1 axis with THD or J54 was independently verified in mouse NT1 cells (Figure 1C). In addition, genetic depletion of NEK1 resulted in YAP1 loss and YAP1 target gene downregulation in NT1 cells (Figure 2). It should be noted that YAP is a generally unstable protein whose turnover rate is strongly regulated by multiple stabilizing [44] or de-stabilizing phosphorylation events controlled by multiple kinases (see [19,20,26] for some reviews). Large tumor suppressor 1 and 2 (LATS1/2), the core kinases of the Hippo signaling pathway, can phosphorylate YAP1 on Ser127 residue, which creates a binding site for 14-3-3 proteins. The 14-3-3 binding of YAP leads to the cytoplasmic sequestration of YAP [45,46]. Sequential phosphorylation by LATS1/2 on YAP Ser397 primes it for further phosphorylation by Casein Kinase CK1δ/ε on Ser400 and Ser403, which creates a phosphodegron motif for (Skp Cullin F box) β-TrCP/SCF E3 ubiquitin ligase-mediated proteasomal

degradation [47]. Recent findings also identify factors such as NR4A1 (nuclear receptor superfamily) that regulate the 14-3-3 interaction with YAP1 and promote its ubiquitination and degradation [48]. Several other kinases independent of the Hippo pathway can regulate the stability of YAP1 protein. For instance, nuclear Dbf2-related kinase (NDR1/2) can also phosphorylate YAP on Ser127 residue and can promote its cytoplasmic retention, thereby negatively regulating YAP stability [49]. Evidence suggests that the protein kinase B/AKT can also phosphorylate YAP on Ser127 residue, leading to binding of 14-3-3 and cytoplasmic retention [45]. In contrast, several members of the Src family of kinases such as Src, Yes, and c-Abl can positively regulate YAP stability. c-Abl/Src/Yes are known to phosphorylate YAP on Tyr357 residue, which results in the nuclear translocation and, hence, stabilization of YAP [44,50,51]. Moreover, Ras-associated factor isoform 1C (RASSF1C) is known to promote tyrosine phosphorylation of YAP1 (Tyr357) through activated Src (pTyr416) and cause nuclear localization of YAP1 [52]. Similarly, mitogen-activated protein kinases such as c-Jun-N-terminal kinases (JNK1/2) are also reported to be YAP kinases that phosphorylate YAP on Ser317 and Thr362, promoting YAP nuclear translocation and stabilization [53]. Thus, post-translational modifications such as phosphorylation determine YAP turnover rate and activity.

Therefore, we propose the phosphorylation of Y407 as one potential mechanism of YAP stabilization and increased transcriptional output, although the other 5-phosphorylation sites could be equally important (Figure 4 and Figures S1–S4; Table S1). There are examples in YAP and TAZ where phosphorylation of some residues impairs ubiquitination and subsequent proteasomal degradation, as in one example, phosphorylation of S128 by NLK competed for the destabilizing LATS1-dependent S127 phosphorylation [54]. However, we currently favor a pY407-related mechanism based on the equivalent pY316 of TAZ, where it was shown that the phosphorylation of that residue, reportedly by c-Abl, was necessary to mediate its interaction with the transcription factor NFAT5 [55]. This was implicated in an inhibitory pathway of NFAT5—a major osmoregulatory transcription factor—during hyperosmotic stress. Similarly, JNK1/2-mediated phosphorylation of YAP1 on Ser317 and Thr362 promotes YAP's ability to bind and stabilize both pro-apoptotic p73 and pro-proliferative ∆Np63α in different cell types [53,56]. We think that, likewise, pY407 promotes the interaction of YAP with some of its transcriptional partners, and hence promotes its nuclear translocation, function, and stabilization, away from cytoplasmic degradation. Importantly, while the phosphorylation of Y407 was identified in proteomic studies [57], to our knowledge, the kinase responsible for it has not been reported.

Resistance to androgen deprivation therapy (ADT) promotes androgen-independent growth and proliferation of PCa cells, which requires efficient DNA damage response (DDR) and repair mechanisms, activation of compensatory signaling pathways, transcription factors, and co-factors to drive castration resistance. Findings from our lab and others suggest that ADT activates the TLK1-NEK1 signaling pathway that promotes PCa progression by activating the DDR [11,24]. Hyper-activation of NEK1 may also lengthen G2/M checkpoints, which provides the cells sufficient time to repair their damaged DNA after ADT or radiation therapy [7,58]. However, DDR alone may not be able to induce androgen-insensitive growth of PCa cells. Thus, we hypothesize that TLK1-NEK1 may be implicated in some other signaling pathway, leading to AI growth. YAP1 is a major oncoprotein that drives many different types of malignancies, including PCa [59], head and neck cancer [59], gastric cancer [60], colon cancer [60], thyroid cancer [61], lung cancer [62], ovarian cancer [63], and liver cancer [64]. NEK1-mediated phosphorylation of YAP1 (most probably on Tyr407 and/or Thr493) may induce a conformational change that counteracts the sequential phosphorylation by LATS1/2 and CK1δ/ε and subsequently protects YAP from proteasomal degradation. Moreover, Tyr407 lies on the transcriptional activation domain of YAP1, which may increase its interaction affinity to its assigned transcriptional factors [65]. Ectopic YAP expression was reported to drive LNCaP cells from androgen-sensitive to androgen-insensitive states [19]. Reducing the turnover rate will increase cellular accumulation of YAP, which can enable its oncogenic properties to drive castration resistance by several mechanisms. Previous studies reported that YAP can mediate PI (3)K-mTOR signaling and activate AKT [66–68]. Activation of mTOR will lead to enhanced translation of TLK1B that can, in turn, increase YAP1

phosphorylation through TLK1-NEK1 nexus. This suggests a positive feed-forward mechanism for YAP accumulation. Elevated YAP can also activate ERK that will promote cell proliferation in absence of AR signaling. Kuser-Abali et al. reported that AR and YAP can interact, and this interaction contributes to the switch from androgen-dependent to castration-resistant phenotype [27]. Overexpression of YAP can also regulate the expression of AR target genes, including PSA, NKX3.1, PGC-1, and KLK2, which suggests that YAP may control AR activity. YAP Tyr407 phosphorylation could increase the binding affinity of AR and AR ligand-insensitive variant AR-V7, thus contributing to androgen refractory growth of PCa cells. Therapy-induced YAP overexpression may also induce EMT activation by upregulating EMT-specific genes. Increasing the stemness of PCa cells can be another mechanism by which stabilized YAP can promote castration-resistant growth of PCa cells, which will further contribute to chemo-resistance of cancer cells [69]. Our bioinformatics analyses also suggested a link between NEK1 and YAP1 in different cancers (Figure 5). YAP1 protein level is abundant in high-grade PCa tumors, despite the progressive downregulation of YAP1 mRNA expression. Other groups also reported that YAP protein is positively correlated with the Gleason score, consistent with the findings of our bioinformatics analysis [70]. We propose that the signaling of TLK1>NEK1-mediated YAP phosphorylation and stabilization contributes not only to PCa progression, but also many other cancers. Importantly, we found a correlation between increased phosphorylated NEK1(T141) in relation to the Gleason score [25] and YAP1 protein expression, whereas the mRNA for YAP1 actually decreased (Figure 5), consistent with our model of post-transcriptional protein stabilization.

#### **5. Conclusions**

YAP's transcriptional activity and degradation is mainly regulated by phosphorylation through several kinases dependent and independent of the Hippo pathway. Using small molecule inhibitors against YAP cannot completely abolish YAP transcriptional activity and is not very effective in treating YAP-driven cancers. Inhibitors such as verteporfin that can disrupt the YAP–TEAD interaction, but still cannot result in complete inhibition, as YAP can bind with other transcription factors such as TEF, SMADs, or TBX5. The majority of YAP kinases negatively regulate YAP by promoting its nuclear egress or degradation; however, NEK1 is found to stabilize YAP protein by phosphorylating it on several residues. Thus, targeting NEK1 or the TLK1–NEK1 axis can bring about therapeutic benefits in the clinical management of YAP-driven malignancies.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2072-6694/12/12/3666/s1: Figure S1: MS/MS spectrum of the phosphor-peptide N70AVMNPKTANVPQTVPMRL<sup>88</sup> to determine pT83, Figure S2. MS/MS spectrum of the phosphor-peptide R161QSSFEIPDDVPLPAGW<sup>177</sup> to determine pS163/pS164, Figure S3. MS/MS spectrum of the phosphor-peptide A347LRSQLPTLEQDGGTQNPVSSPGmSQEL<sup>374</sup> to determine pT361/pS366, Figure S4. MS/MS spectrum of the phosphor-peptide R375TMTTNSSDPFLNSGTY<sup>391</sup> to determine pS388, Figure S5. Densitometry of the original uncropped blots and gel images with their respective numbers from the main figures, Table S1. Assigned phosphorylated sites.

**Author Contributions:** Conception and design: M.I.K., V.S., I.G., A.D.B. Development of methodology: M.I.K., V.S., I.G., J.C., H.Z., A.D.B. Acquisition of data: M.I.K., V.S., I.G., J.C., H.Z., A.D.B. Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.I.K., J.C., H.Z., A.D.B. Writing, review, and/or revision of the manuscript: M.I.K., I.G., J.C., H.Z., A.D.B. Administrative, technical, or material support: M.I.K., I.G., J.C., H.Z., A.D.B. Study supervision: A.D.B., H.Z. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported primarily by DoD-PCRP grant W81XWH-17-1-0417 to A.D.B. LC–MS/MS equipment was acquired using a National Center for Research Resources High-End Instrumentation grant (1S10RR029127 to H.Z.).

**Acknowledgments:** We thank the Research Core Facility Genomics Core at LSU Health Shreveport for the help with the qPCR analysis. We acknowledge the University of Kentucky Markey Cancer Center's Redox Metabolism Shared Resource Facility partially supported by a National Cancer Institute Center Core support grant (P30 CA177558).

**Conflicts of Interest:** No potential conflicts of interest are disclosed.

#### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*

## **Two Secreted Proteoglycans, Activators of Urothelial Cell–Cell Adhesion, Negatively Contribute to Bladder Cancer Initiation and Progression**

**Vasiliki Papadaki 1,2,**† **, Ken Asada 3,4,**† **, Julie K. Watson 5,6,7,**† **, Toshiya Tamura 1,8 , Alex Leung <sup>1</sup> , Jack Hopkins <sup>1</sup> , Margaret Dellett 1,9, Noriaki Sasai 1,10 , Hongorzul Davaapil 1,11, Serena Nik-Zainal 12,13, Rebecca Longbottom 1,14 , Makoto Nakakido 1,15 , Ryo Torii <sup>16</sup> , Abhi Veerakumarasivam 5,17 , Syuzo Kaneko <sup>3</sup> , Mandeep S. Sagoo 1,18,19,20 , Gillian Murphy 5,6 , Akihisa Mitani <sup>21</sup>, Kohei Tsumoto <sup>15</sup> , John D. Kelly 5,22, Ryuji Hamamoto 3,4,5,15,\* and Shin-ichi Ohnuma 1,5,\***


Received: 23 September 2020; Accepted: 3 November 2020; Published: 13 November 2020

**Simple Summary:** Epithelial–mesenchymal transition (EMT) is associated with cancer progression. Here, we found that two secreted proteins of osteomodulin (OMD) and proline/arginine-rich end leucine repeat protein (PRELP) were selectively expressed in bladder umbrella epithelial cells, and they were suppressed in bladder cancer. We revealed that *OMD*−/<sup>−</sup> or *PRELP* <sup>−</sup>/<sup>−</sup> knockout mice caused a breakdown of the umbrella cell layer through weakening cell–cell integrity and the activation of partial EMT, which resulted in the formation of early bladder cancer-like structures, while OMD or PRELP application to bladder cancer cells inhibited cancer progression through reversing EMT, which was mediated by the inhibition of TGF-β and EGF. Our result indicates that OMD and PRELP function as tumor-suppressing proteins through inhibiting EMT. OMD and PRELP may be potential therapeutic targets in bladder cancer.

**Abstract:** Osteomodulin (OMD) and proline/arginine-rich end leucine repeat protein (PRELP) are secreted extracellular matrix proteins belonging to the small leucine-rich proteoglycans family. We found that OMD and PRELP were specifically expressed in umbrella cells in bladder epithelia, and their expression levels were dramatically downregulated in all bladder cancers from very early stages and various epithelial cancers. Our in vitro studies including gene expression profiling using bladder cancer cell lines revealed that OMD or PRELP application suppressed the cancer progression by inhibiting TGF-β and EGF pathways, which reversed epithelial–mesenchymal transition (EMT), activated cell–cell adhesion, and inhibited various oncogenic pathways. Furthermore, the overexpression of OMD in bladder cancer cells strongly inhibited the anchorage-independent growth and tumorigenicity in mouse xenograft studies. On the other hand, we found that in the bladder epithelia, the knockout mice of OMD and/or PRELP gene caused partial EMT and a loss of tight junctions of the umbrella cells and resulted in formation of a bladder carcinoma in situ-like structure by spontaneous breakdowns of the umbrella cell layer. Furthermore, the ontological analysis of the expression profiling of an OMD knockout mouse bladder demonstrated very high similarity with those obtained from human bladder cancers. Our data indicate that OMD and PRELP are endogenous inhibitors of cancer initiation and progression by controlling EMT. OMD and/or PRELP may have potential for the treatment of bladder cancer.

**Keywords:** OMD; PRELP; tumor suppression gene; bladder cancer initiation; tight junction; partial EMT

#### **1. Introduction**

Small leucine-rich proteoglycans (SLRPs) are a family of 17 secreted extracellular matrix (ECM) proteoglycans [1]. SLRP members function not only as modifiers of ECM organization but also as regulators of ligand-induced signaling pathways [1–4]. For example, Tsukushi regulates the Notch, Wnt, FGF, BMP4, and Nodal pathways through interactions with extracellular components in a context-dependent manner [5–8]. The expression of SLRPs is often altered in tumors. Biglycan, lumican, and fibromodulin are overexpressed in various types of cancer, whilst decorin is overexpressed in some types of cancer and suppressed in others [9]. High expression levels of lumican are associated with a

poorer survival in colorectal tumors, and they are also presented with increased metastasis in lung cancers [10]. Conversely, the overexpression of lumican in melanoma cells inhibited tumor formation in an animal model [11], whereas low expression levels of lumican and decorin are associated with a poorer patient survival in breast tumors and spindle cell carcinomas, respectively [12]. Thirty percent of decorin knockout mice develop intestinal tumors [13], and decorin/p53 double knockout mice demonstrate an enhanced susceptibility to thymic lymphoma [14]. Decorin suppressed squamous cell carcinoma in vitro by binding to EGFR to regulate downstream signaling pathways, while it also inhibited tumor formation and metastasis in a xenograft model [1,15,16]. However, no mutations or deletions of these genes have been reported so far in human cancers. Thus, their relevance to human carcinogenesis remains unclear.

With the development of epithelial malignancies, major changes occur in the organization of ECM, which normally provides the microenvironment for the maintenance of epithelial cell integrity. Many oncogenes cannot initiate a tumor if the extracellular microenvironment is normally maintained [17]. Moreover, in some cases, breakdown of the extracellular microenvironment by itself can trigger tumorigenesis [18]. These studies further demonstrate the importance of ECM proteins in cancer development.

Bladder cancer is one of the most common cancers worldwide, with 549,400 new cases and 200,000 deaths annually [19]. Our study shows that the two SLRPs or secreted ECM, osteomodulin (OMD) and proline/arginine-rich end leucine repeat protein (PRELP) are expressed in bladder and critical regulators of bladder cancer initiation and progression via altering cell–cell adhesion, probably through the regulation of epithelial–mesenchymal transition (EMT). Our findings can explain the mechanism of cancer initiation and can contribute to new therapeutic applications.

#### **2. Results**

#### *2.1. OMD and PRELP Expression and the Association with the Early Stages of Bladder Cancer*

We analyzed the expression levels of SLRP members in various epithelial cancers including bladder cancer using two independent microarray-based expression-profiling databases drawn from a worldwide population: Oncomine (Figure 1a,b; Figure S1) and Gene Logic Inc (Figure S2). Interestingly, the expression levels of *OMD* and *PRELP* are strongly suppressed in the majority of epithelial cancer types.

Next, we performed a detailed expression analysis of 126 bladder cancer samples and 31 normal control samples (Figure 1c–f; Table S1). The expression of both *OMD* and *PRELP* in tumors was drastically lower compared to normal tissues (Figure 1d,f) and declined progressively with cancer stage (Figure 1c,e; Table S1). No associations were found with gender or recurrence status, nor with age or tumor size (Table S1). *OMD* and *PRELP* were also downregulated in bladder cancer cell lines compared to normal bladder tissue (Figure S3a,b). Moderate *OMD* expression was seen only in the non-invasive bladder cell lines RT4 and LHT1376.

**Figure 1.** Expression of osteomodulin (OMD) and proline/arginine-rich end leucine repeat protein (PRELP) in cancer (**a**,**b**). Microarray analysis of *OMD* (**a**) and *PRELP* (**b**) expression in human bladder cancer samples and normal bladder tissues. (**c**) Quantitative analysis of *OMD* expression in bladder cancer at different stages by qPCR. (**d**) Box–whisker plot (median 50% boxed) of (**c**). Cutoff value (dash line) was determined as described in Materials and Methods. (**e**) Quantitative analysis of *PRELP* expression in bladder cancer at different stages by qPCR. (**f**) Box–whisker plot of (**e**). (**g**) Expression analysis of *OMD* and *PRELP* in bladder cell lines. Published expression profiling data of MIBC cell lines (GSE97768) are re-examined to elucidate the relative expression of *OMD* and *PRELP* in comparison

with known overexpressing genes in bladder cancer; *APP*, *CHEK1*, *EGFR*, *ERBB2*, and *TP53* and with housekeeping genes of *TUBA1C*, *TUBB*, and *TUBD1*. (**h**) Expression analysis of *OMD* and *PRELP* in bladder tissues samples from patients. Published expression profiling data of non- Non-muscle-invasive bladder cancer (NMIBC)(E-MTAB-4321) are re-examined to elucidate the relative expression of *OMD* and *PRELP* in comparison with *APP*, *CHEK1*, *EGFR*, *ERBB2*, and *TP53* and with housekeeping genes of *TUBA1C* and *TUBD1*. Details of both (**g**,**h**) analyses are in Materials and Methods. (**i**) Somatic mutations in human cancer samples that are predicted to generate a loss of function of *OMD*. Detail of cancers is described in Materials and Methods in the section of OMD and PRELP Expression Analysis in muscle-invasive bladder cancer (MIBC) cell lines and NMIBC Patient Samples. \*\* indicates *p* < 0.01.

To assess the potential role of OMD and PRELP as diagnostic markers, we set cutoff values to distinguish tumor samples from normal tissues through calculation of the interquartile range. The expression levels of *OMD* and *PRELP* in almost all normal bladder tissues were above the cutoff value (specificity: 83.9% (*OMD*) and 90.3% (*PRELP*)), while expression in the vast majority of tumor tissues was below the cutoff (sensitivity: 88.9% (*OMD*) and 90.5% (*PRELP*), Table S2). Expression levels of *OMD* and *PRELP* in the Ta (early) stage of almost all tumor tissues were below the cutoff value (sensitivity: 88.9% (OMD) and 88.9% (PRELP), Table S2). When we combined the data for *OMD* and *PRELP,* the expression of both genes below the cutoff value was found only in tumor samples and in none of the normal tissues (specificity 100%). These results show that the expression levels of *OMD* and *PRELP* genes are powerful markers for the prediction of the presence of urothelial carcinomas. The suppression of *OMD* and *PRELP* was also observed when we analyzed previously published expression profiling data for muscle-invasive bladder cancer (MIBC) and non-muscle-invasive bladder cancer (NMIBC) (Figure 1g,h) [20,21]. An examination of mutation analysis using The Cancer Genome Atlas (TCGA) (Figure 1i) found a total of 3,142,246 somatic substitutions/indels were interrogated from 33,096 primary human cancers, and the somatic mutations predicted to generate a loss-of-function effect in OMD are summarized in Figure 1i. However, relatively few mutations were observed (95 for OMD and 158 for PRELP) (unpublished data).

#### *2.2. Cell–Cell Adhesion and Cancer Signaling Regulated by OMD and PRELP*

To further assess the role of OMD and PRELP in cancer, we overexpressed or underexpressed the two proteins in cultured cells and performed gene expression analysis using microarrays (Affimetrix GeneChip® System). The T-Rex-293T system was used to express the genes at a near-physiological level without causing adverse effects due to their insertion site. To ablate gene expression, 5637 bladder cancer cells, expressing *OMD* and *PRELP* at a low level, were transfected with siRNA constructs for *OMD* or *PRELP*. After validating the altered expression of *OMD* and *PRELP* by RT-PCR, gene expression profiling was performed.

Figure 2a,b show the numbers of genes that are negatively and positively transcriptionally regulated by OMD and/or PRELP, respectively. The genes affected by OMD and PRELP include many oncogenes and tumor-suppressor genes such as *NF*-kB, *Ras*, and *c-Fos*. For example, 107 genes were activated by both OMD and PRELP overexpression, while 139 genes were suppressed by the double-depletion (Figure 2b). These observations indicate that OMD and PRELP have a functional redundancy while they also regulate various distinct target genes.

Next, to elucidate the affected signaling pathways, biological events, and mechanisms, the gene expression profiling data were analyzed with a data mining program (Ingenuity Pathway Analysis, IPA, Qiagen, (https://digitalinsights.qiagen.com/products-overview/discovery-insights-portfolio/anal ysis-and-visualization/qiagen-ipa/?cmpid=QDI\_GA\_IPA&gclid=CjwKCAiAtK79BRAIEiwA4Osk BpDKfEsg5CJdSERKm3IEd\_0gZRXNEGfgu7XJjKoC9hVggrFtzQnvxBoCY\_wQAvD\_BwE). Using the Functional Analysis mode, "molecular mechanism of cancer" was identified as one of the most significantly affected biological functions and/or diseases in all four conditions of OMD overexpression, OMD depletion, PRELP overexpression, and PRELP depletion (Figure 2c; Figure S3c–f). In total, 304 and 388 genes related to the "cancer" category are significantly affected by the altered expression of

OMD and PRELP, including members of the p53 pathway, the NF-kB pathway, the Ras pathway, the RB1 pathway, the Jun/Fos pathway, and the Myc pathway (Figure 2d).

Our analysis also revealed that both OMD and PRELP strongly influence cell–cell adhesion mediated by tight junctions (Figure 2e). Tight junctions are a type of cell–cell junction that binds the apical sides of epithelial cells. The breakdown of tight junctions has been proposed as a critical step in cancer initiation [22,23]. Tight junction components such as Zonula occlugens-1 (ZO-1) and Nectin were transcriptionally activated by OMD or PRELP overexpression, while they were suppressed in OMD or PRELP depletion, suggesting that OMD and PRELP have the ability to positively regulate tight junctions (Figure 2e).

**Figure 2.** Gene expression profiling in OMD/PRELP overexpressing or deleted cells. Gene expression profiling was performed under seven conditions; OMD overexpression in T-Rex-293T cells, PRELP overexpression in T-Rex-293T cells, control T-Rex-293T cells, OMD depletion in the 5637 bladder cancer cells, PRELP depletion in the 5637 cells, two controls of the 5637 cells. Details are in Materials and Methods. Then, genes with statistical significant changes of mRNA levels have been identified. Data were analyzed, and the following figures were made through the use of Ingenuity Pathway Analysis (IPA) (QIAGEN Inc., https://www.qiagenbioinformatics.com/products/ingenuitypathway-analysis). (**a**) Gene numbers significantly inhibited by OMD overexpression or PRELP overexpression but activated by OMD depletion or PRELP depletion. (**b**) Gene numbers activated by OMD overexpression or PRELP overexpression but suppressed by OMD depletion or PRELP depletion. (**c**) Heat map of signaling pathways significantly affected by OMD overexpression. This heat map was created using IPA software. Similar heat maps were observed in other three conditions of OMD depletion, PRELP overexpression, and PRELP depletion. (**d**,**e**) Schematic drawing of the most strongly influenced biological events regulated by OMD overexpression. "Molecular Mechanism of Cancer" (**d**) and "Tight junction signaling" (**e**) category. Both images of (**d**,**e**) were created by Ingenuity Pathway Analysis according to their rule. This pathway is one of the most strongly influenced ones by any of four conditions (OMD overexpression, OMD depletion, PRELP overexpression, PRELP depletion).

#### *2.3. OMD or PRELP Overexpression in EJ28 Bladder Cancer Cells*

To investigate the roles of OMD and PRELP at the molecular level, we constructed stable cell lines that overexpressed OMD, OMD-myc, PRELP, and PRELP-myc using the EJ28 bladder cancer cell line, as their endogenous expression is strongly suppressed.

Under standard cell culture using non-coated culture dish with non-confluent conditions, control EJ28 cells had a flattened fibroblast-like shape. In contrast, many OMD overexpressing cells had a markedly different round shape with many pin-like extensions (Figure 3a). PRELP overexpression also resulted in a change of cell morphology to round cells similar to OMD overexpressors together with elongated cells with protruding stress fiber-like filamentous extensions (Figure 3a). To evaluate relevant changes in the cytoskeletal structure, we stained for actin and tubulin. We found that there are many pin-like actin structures on the surface of the round OMD-expressing cells, similar to the phenotype induced by cdc42 activation [24]. On the other hand, PRELP overexpression resulted in both round cells with pin-like structures and elongated cells with long clear actin fibers (Figure 3b). These abnormal morphological changes were also observed with tubulin staining (Figure 3c).

We next analyzed the effect of OMD/PRELP overexpression on cell proliferation and survival. First, expression levels of *OMD* and *PRELP* in EJ28 cells were analyzed by qRT-PCR (S4a and b) and by Western blotting using the myc antibody for myc-tag protein expression (S8n and S8p). OMD and OMD-myc cells exhibited reduced proliferation, both in standard proliferation and BrdU incorporation assays (Figure S4c,d), while cell cycle analysis by flow cytometry revealed an enhanced G1 phase transition (Figure S4e). Finally, OMD and OMD-myc cells presented a small but significant increase in apoptosis, as assayed by annexin staining (Figure S4f). The overexpression of OMD, OMD-myc, and PRELP resulted in a slight but significant suppression of cell growth with modulation of the cell cycle phase distribution. In addition, OMD and PRELP overexpression slightly increased apoptosis, although the majority of cells remained non-apoptotic. Overall, we conclude that the overexpression of OMD and PRELP proteins results in a subtle but significant suppression of cell growth with the modulation of cell cycle phase distribution.

Anchorage-independent growth is a well-established property of transformed cancer cells. Therefore, we examined the effect of OMD or PRELP overexpression on anchorage-independent growth (Figure 3d,e). OMD or PRELP-myc overexpression completely abolished colony formation. These results indicate that OMD and PRELP suppression might be important for the transition from normal epithelial cells to mesenchymal-like cancer cells. Additionally, we tested the cell growth in a 3D environment using Matrigel to investigate growth under partial anchorage conditions. Control EJ28 cells grew well and showed a "spread-like" morphology (Figure 3f), as observed in standard cell culture dishes. However, OMD or PRELP overexpressing cells tended to make cell aggregates, suggesting that OMD and PRELP may influence cell migration. To address this, we performed the Boyden chamber assay with Matrigel-coated transwells. The assay clearly demonstrated that the overexpression of OMD or PRELP strongly suppressed cell migration and invasion (Figure 3g,h). The effect of OMD and PRELP overexpression on cell migration was also tested in standard 2D conditions with the scratch wound assay, where a small inhibition of the wound recovery was observed (Figure S4g). Collectively, these results suggest that the two proteins affect colony formation, migration, and invasion capabilities of cancer cells in a substrate-dependent manner.

As OMD and PRELP are secreted proteins, to confirm that the observed effects are mediated by the extracellular forms, we performed a co-culture assay, in which EJ28 cells (Cell A) overexpressing OMD or PRELP were cultured in the chamber above tester EJ28 cells (Cell B) (Figure 3i). OMD and PRELP significantly suppressed the growth of the lower layer of EJ28 tester cells (Figure 3j), as we expected.

**Figure 3.** Effect of OMD or PRELP overexpression in bladder cancer cell lines. (**a**) Cell morphology of EJ28 bladder cancer cells transfected with OMD, OMD-myc, PRELP, or PRELP-myc constructs, observed by differential interference contrast (DIC) microscope. Round cells are indicated as arrowheads. (**b**) Phalloidin staining of the transfected EJ28 cells. Phalloidin (red) and DAPI (blue). Pin-like structures of OMD overexpressing cells are phalloidin-positive. PRELP overexpression results in clear long actin fiber formation. (**c**) Anti-tubulin antibody staining. Tubulin (red) and 4′ ,6-diamidlino-2-phenylindole (DAPI) (blue). (**d,e**) Anchorage-independent growth using the soft agar. Photos of control, OMD, and PRELP overexpressing colonies formed in the top agar layer (**d**). Quantification of the cell percentage that formed colonies (**e**). (**f**) Cell growth in the Matrigel. (**g**,**h**) Cell migration and invasion assay using the Boyden chamber. Photos of cells that invaded to the bottom side of membrane after the addition of fetal bovine serum (FBS) as a chemoattractant (**g**). Quantification of cell migration/invasion in (**g**,**h**). (**i**,**j**) Transwell co-culture assay to evaluate the effect of secreted OMD/PRELP on non-contacting cells. Schematic drawing of the assay system and photos of EJ28 cells cultured at the button chambers (**i**). Quantification of viable cell density in the bottom well by trypan blue staining (**j**). \*, \*\*, \*\*\* indicate *p* < 0.01, *p* < 0.005, *p* < 0.001, respectively.

#### *2.4. The Relation between OMD or PRELP and Tight Junction Formation*

We examined the status of tight junctions of EJ28 cells using antibodies against occludin (Figure 4a–i), ZO-1 (Figure 4j–l), and cingulin (Figure 4m–o). In confluent monolayers of the control EJ28 cells, we observed partial staining at cell–cell interfaces, covering around 40% of the total cell–cell surface for occludin, (40% of total cell–cell surface), ZO-1 (46%), and cingulin (30%) (Figure 4p–r). This appearance of partial junction staining is found in cancer cell lines (personal communication, Karl Matter). Interestingly, the overexpression of OMD resulted in enhanced and continuous junctional staining of all three markers, covering almost the whole cell periphery (Figure 4b,h,k,n). PRELP overexpression had a similar effect, where tight junction formation was also markedly increased compared to the control cells. This enhanced junctional staining was accompanied by a reduction of the cytoplasmic staining of the corresponding markers. To further confirm the formation of tight junctions, the control EJ28 cells and OMD overexpressing cells were examined by electron microscopy.

A large number of tight junctions were observed in OMD overexpressing cells (Figure 4s–u). However, we failed to detect any tight junctions in the control EJ28 cells (Figure 4v,w).

**Figure 4.** Effect of OMD or PRELP overexpression on tight junction in EJ28 cells. (**a**–**i**) Occludin antibody staining of OMD or PRELP expressing EJ28 cells; low magnification (**a**–**c**), overlaid with DAPI (**d**–**f**), enlarged (**g**–**i**). **(j**–**l**) ZO-1 staining. (**m**–**o**) Cingulin staining. Scale bar represents 100 µm (**a**–**o**). (**p**) Quantification of occluding staining. (**q**) Quantification of ZO-1 staining. (**r**) Quantification of cingulin staining. (**s**–**w**) Electron microscope (EM) analysis of cell-cell junction. OMD expressing EJ28 cells (**s**–**u**) and wild type (WT) EJ28 cells (**v**,**w**). Tight junctions are indicated by arrows. Scale bar represents 1 µm (**t**,**u**,**w**) and 0.5 µm (**s**,**v**). OMD overexpression strongly activates tight junction formation. (**x**–ff) Antibody staining of confluent monolayer; β-catenin (**x**–**z**), E-cadherin (**aa**–**cc**), and vimentin (**dd**–ff). Scale bar represents 100 µm (**x**–ff). \*\*\* indicates *p* < 0.001.

Subsequently, to determine the effect on adherens junctions, we examined the expression of β-catenin, E-cadherin, and vimentin. Figure 4x shows that in the control group, many cells have a weak β-catenin localization in the nuclei. On the other hand, in OMD or PRELP overexpressing cells, β-catenin was almost exclusively localized at the plasma membrane, and the strength of the staining was much higher than the control (Figure 4x–z). E-cadherin staining was slightly enhanced (Figure 4aa–cc), indicating that OMD and PRELP activate adherens junctions. To test how OMD and PRELP regulated cell–cell adhesion, we examined the expression of vimentin, an EMT marker. The major characteristics of epithelial cells are cell polarity, strong cell–cell integrity, and anchorage-dependent growth. Cancer initiation in epithelia is always associated with EMT [25,26]. After conversion to mesenchymal cells, these cells can grow in an anchorage-independent manner, as observed in almost all cancer cells. Figure 4dd–ff shows that vimentin was more localized around or in the nucleus, while OMD or PRELP-expressing cells showed a diffuse expression of vimentin in the cytosol. This suggests that OMD and PRELP may regulate cell–cell adhesion through EMT.

#### *2.5. Signal Pathways Regulated by OMD and PRELP*

The gene expression profiling experiment revealed that OMD and PRELP were involved in the regulation of various components of several ligand-induced signaling pathways, including the IGF-1, Wnt, EGF, and TGF-β pathways. We aimed to determine the molecular mechanisms of OMD/PRELP activity using EJ28 stable cell lines that overexpress the two proteins. In the expression profiling data (Figure 2d), the Akt level was significantly affected by both OMD and PRELP. We found that OMD and PRELP overexpression downregulated the phosphorylation of Akt (Figure 5a), and OMD overexpression downregulated the phosphorylation of ERK1/2 (Figure 5b). Akt phosphorylation is known to be regulated by the EGF and IGF pathways [27,28]. Figure 5a and b show that upon EGF treatment (10 ng/mL), Akt phosphorylation was decreased in the OMD overexpressing cells compared to the control. EGF induced the phosphorylation of tyrosine-1068 of the EGFR, and this phosphorylation was suppressed by OMD expression (Figure 5b). ERK1/2 phosphorylation was elevated by exogenous EGF, and this phosphorylation was also suppressed by OMD (Figure 5b). Co-immunoprecipitation assays revealed that OMD was bound to the EGFR (Figure 5c). Total EGFR protein was reduced in OMD transfected cells (Figure 5d). Inhibition of the EGF pathway is known to lead β-catenin localization to the cell membrane [29], which we observed in OMD/PRELP activation (Figure 4x).

IGF activated Akt through the phosphorylation of the IGF-1R; however, OMD overexpression did not inhibit the IGF-mediated phosphorylation of Akt (Figure 5e) in our assays. In addition, we did not detect any direct interaction of OMD with the IGF receptor (Figure 5f). All the SLRP family members previously studied directly interact with TGF-β family members and regulate transcription of their targets via the phosphorylation of Smad2 [2]. Indeed, OMD and PRELP directly bound to TGF-β protein (Figure 5g) and resulted in Smad2 phosphorylation suppression, particularly in OMD (Figure 5h). The effect of OMD and PRELP on EGFR, β-catenin, and Smad2 were quantitated and the results are shown in Figure S5.

**Figure 5.** Mechanism of OMD or PRELP-mediated regulation of tight junction. Various effects of OMD and PRELP were examined in vitro using OMD or PRELP stably overexpressing EJ28 bladder cancer cell lines. OMD1 and OMD2 indicate different stable clones. (**a**) Effect of OMD or PRELP overexpression on Akt phosphorylation. (**b**) Effects of OMD overexpression and EGF application on EGF receptor, Akt, and ERK phosphorylation. (**c**) Interaction between OMD and EGF receptor. (**d**) Effect of OMD or PRELP on the total amount of EGF receptor. (**e**) Effects of OMD overexpression and IGF-1 application on phosphorylation of the IGF receptor, Akt, and ERK. (**f**) Interaction between the OMD and IGF receptors. (**g**) Binding of OMD or PRELP with TGF-β. (**h**) Effect of OMD or PRELP on Smad2 phosphorylation. (**i**) Effect of OMD and PRELP on the total levels of β-catenin protein expression. (**j**) Effect of OMD or PRELP on phosphorylation of p38. All original Western blotting data

are shown in Figure S8. (**k**) Effect of OMD on cdc42 activity. (**l**–**p**) Effect of EGF, IGF-1, and TGF-β 1 application on tight junction formation of confluent OMD overexpressing EJ28 cell monolayers. Occludin staining of normal EJ28 cells (**l**) and OMD expressing EJ28 cells (**m**). Effect of 10 ng/mL EGF (**n**), 100 ng/mL IGF-1 (**o**), or 10 ng/mL TGF-β 1 (**p**) on occludin staining of EJ28 cells overexpressing OMD. Scale bar represents 100 µm. (**q**) Quantification of occludin-positive cell–cell junctions. (**r**) TGF-β, EGF, and Wnt pathways are affected in OMD−/<sup>−</sup> mouse bladder. Ontological analysis of the expression profiling data obtained in Figure 8. (**s**) Schematic model of OMD/PRELP function. The uncropped Western Blot figure in Supplementary Figure S8. \*\*, \*\*\*\* indicate *p* < 0.01, *p* < 0.0001, respectively.

We found that OMD overexpression significantly increased the total amount of β-catenin (Figure 5i). However, we could not detect a change of Wnt-mediated transcription activity by the TOPFLASH assay (unpublished data). Taken together with our finding that OMD causes the translocation of β-catenin to the plasma membrane (Figure 4x–z), this suggests that the increased β-catenin mainly contributes to its adherens junction-related function.

The downstream segments of ligand-induced signaling pathways are remarkably interconnected with each other in context-dependent manners. Thus, we examined two common downstream components of the EGF and TGF-β pathways, p38 and cdc42, as the OMD or PRELP mediated in vitro phenotypes reported in this paper are similar to those caused by p38 or cdc42 modulation [30–32]. Moreover, our expression profiling analysis indicated the importance of the cdc42 and p38 pathways in this context (Figure 2d). We found that OMD and PRELP overexpression increased the phosphorylation of p38 (Figure 5j), and OMD activated cdc42 (Figure 5k).

Finally, we examined the contribution of OMD-mediated inhibition of pathways to the regulation of tight junctions. TGF-β, IGF, and EGF pathways are well known as major pathways to regulate EMT and mesenchymal–epithelial transition (MET). OMD overexpressing EJ28 cells were treated with either EGF (10 ng/mL), TGF-β (10 ng/mL), or IGF-1 (100 ng/mL) protein, and their effects on tight junction formation were assessed. Cellular response was confirmed by analysis of phosphorylation of ERK1/2, AKT, and Smad2. Figure 5l–q shows that EGF and TGF-β strongly inhibited OMD-induced tight junction formation, while IGF-1 had no effect, suggesting that the OMD-mediated regulation of both EGF and TGF-β pathways is important for the regulation of tight junctions. In addition, OMD overexpression induced the translocation of β-catenin to the plasma membrane (Figure 4x), which was accompanied by an increase in the total expression levels of β-catenin (Figure 5i). Such effects were previously reported as phenotypes caused by EGF pathway inhibition [29]. Later, we will show another gene expression profiling using bladder tissues isolated from OMD−/<sup>−</sup> or PRELP−/<sup>−</sup> mice (Figure 8). The ontological analysis shows that indeed, OMD/PRELP regulate EGF and TGF-β pathways (Figure 5r)

Our results demonstrate that the OMD-mediated simultaneous regulation of TGF-β and EGF pathways is important for the maintenance of cell–cell adhesion (Figure 5s).

#### *2.6. Tumor Progression in a Mouse Xenograft Model*

In order to examine the in vivo effects of OMD overexpression in cancer development, we performed mouse xenograft experiments using stably transformed EJ28 cells. When EJ28 cells were grafted in nude mice, the control EJ28 cancer cells grew well, while OMD-expressing EJ28 cells did not grow at all (Figure 6a). These observations are in accordance with the decreased anchorage-independent growth we observed in vitro (Figure 3d–e). Haemotoxylin and Eosin (H&E) staining of tumor sections showed that the density of nuclei was reduced and the nuclear–cytoplasmic ratio was increased in OMD-overexpressing samples (Figure 6b–e). Moreover, occludin staining revealed that OMD-expressing EJ28 cells have a more organized structure and stronger tight junctions (Figure 6f–h). Next, we analyzed the ultrastructure of the xenografted cells by electron microscopy. This analysis showed that adjacent cells of the control samples intercellular spaces between neighboring cells are always visible, and almost no tight junctions can be observed (Figure 6i,j), while the OMD-expressing

xenografts are in close contact and form multiple tight junctions (Figure 6k,l). These results confirmed that OMD/PRELP overexpression enhances cell–cell adhesion and suppresses cancer development in vivo.

**Figure 6.** Mouse xenograft model overexpression of OMD. (**a**) Xenograft of EJ28 cells stably expressing OMD. Tumor volume progression graph. EJ28-*WT* (*n* = 5) and EJ28-OMD (*n* = 5). (**b**–**e**) Histology of xenografted tissues; H&E staining of control EJ28 cells (**b**) and EJ28 cells overexpressing OMD (**c**), comparison of the number of nuclei in 100 µm<sup>2</sup> of sections (**d**), comparison of the ratio of nucleus vs cytosol. (**e**) Scale bar represents 100 µm (**b**,**c**). (**f**,**g**) Occludin staining of control EJ28 tumor (**f**) and OMD overexpressing EJ28 tumor (**g**). (**h**) Quantification of occludin staining. The stained percentage of cell surfaces was measured. (**i**) EM of control EJ28 cells. (**j**) Enlarged from (**i**). (k) EM of OMD-overexpressing EJ28 tumor cells. Scale bar represents 1 µm (**i,k**). (l) Enlarged from (**k**). Tight junctions are shown with arrows. \*\*\* indicates *p* < 0.001.

### *2.7. OMD*−/<sup>−</sup> *or PRELP*−/<sup>−</sup> *Mice and Tight Junctions between Umbrella Cells*

Next, we established constitutive *OMD*−/−, *PRELP*−/−, and *OMD*−/−/*PRELP*−/<sup>−</sup> double knockout mice (Figure S6a–d). The knockouts were designed to target exons 2 and 3, resulting in the complete removal of protein coding sequences while knocking in the β-galactosidase gene under the *OMD* and *PRELP* promoters, respectively. The mice were viable and fertile, and no severe developmental defects were observed. *OMD* and *PRELP* expression in mice were analyzed by qRT-PCR (Figure S6e,f).

*OMD* and *PRELP* were expressed in all organs tested in various levels (Figure S6g,h for mouse, Figure S6i,j for human). To characterize the expression in the bladder, we assayed β-galactosidase activity in heterozygous *OMD*+/−(LacZ) and *PRELP*+/−(LacZ) mice. We observed β-gal-positive cells only in the epithelial layer (Figure S6k,l). A similar pattern was found by the in situ hybridization with the *OMD* or *PRELP* gene probe (Figure S6m,n). The bladder epithelium contains three cell types: basal cells, intermediate cells, and superficial umbrella cells [33]. To identify which cell types express OMD or PRELP, bladder sections were co-stained with β-gal and uroplakin-III (umbrella), CK18 (umbrella), CK5 (basal), or laminin (basement membrane of epithelium) antibody. In *OMD*+/<sup>−</sup> mice, β-gal positive cells were always co-localized with a subpopulation of the uroplakin-III and CK18 positive cells, but not with CK5 or laminin (Figure S6o–r). We also stained with Ki67 (proliferative) markers (Figure S6s), but there was no overlap staining. PRELP showed an expression pattern similar to that of OMD (Figure S6t–x). These results indicate that at any one time, the active transcription of *OMD* and *PRELP* is occurring in a subpopulation of umbrella cells.

Umbrella cells are connected to each other strongly by tight and adherens junctions [33]. We examined the effect of OMD or PRELP deficiency on umbrella cell junctions. Electron microscopy images indicated that the apical–lateral interfaces between *WT* bladder umbrella cells were tightly sealed by dense tight junctions (Figure 7a,b). However, strong tight junctions were markedly reduced in OMD−/−, PRELP−/−, or the double knockout mice (Figure 7c–e). The reduction at the lateral surface was confirmed by immunostaining with the tight junction marker ZO-1. In the *WT*, ZO-1 staining was located at the lateral cell surface (Figure 7f). In *OMD*−/<sup>−</sup> or *PRELP*−/<sup>−</sup> bladder tissues, the ZO-1 signal at the lateral cell surface was significantly reduced (Figure 7g–j). Adherens junctions are localized in the lateral cell–cell surface between umbrella cells, below the tight junction level. In *WT* mice, the adherens junctions were visible in the basolateral surface of umbrella cells, as marked by E-cadherin staining (Figure 7k), while in *OMD*−/−, *PRELP*−/−, and the double knockout mice, E-cadherin was localized in the whole cell surface (Figure 7l–n). This demonstrates that the disruption of tight junctions enables E-cadherin to migrate to the apical side of the cell membrane. These observations indicate that OMD or PRELP depletion results in the induction of a partial EMT state, which is characterized by the loss of tight junctions but not adherens junctions (Figure 7o).

One of the major functions of tight junctions in the bladder is to form the blood–urine barrier to block the leakage of fluids into the bladder [34]. In accordance with this function, deletion of the *PRELP* gene resulted in the formation of clots containing fibrin/fibrinogen in the bladder lumen (Figure 7p,q) and the leakage of proteins into the urine (Figure 7r).

**Figure 7.** *OMD* or *PRELP* knockout resulted in a loss of tight junctions between bladder umbrella cells. (**a**–**e**) EM analysis of *WT* (**a,b**), *OMD*−/<sup>−</sup> (**c**), *PRELP*−/<sup>−</sup> (**d**), and their double (**e**) knockout bladders at 3 months old. A wide view of *WT* bladder epithelia, which includes two umbrella cells and an intermediate cell. The apical side of the cell–cell surface of umbrella cells (black arrowhead) are strongly sealed by dense tight junctions (white arrowheads) (**a**). Apical side of umbrella cell–cell interfaces. Black arrowheads; cell–cell interfaces. White arrowheads; tight junctions (**b**–**e**). Scale bar represents 200 nm. (**f**–**j**) Analysis of ZO-1 staining of a 3-month-old bladder. ZO-1 staining between umbrella cells is indicated by white arrowheads (**f**). Quantification of ZO-1 staining (**j**). (**k**–**n**) E-cadherin staining of a 3-month-old bladder. (**o**) Model of cell–cell adhesion in bladder epithelial cells. (**p**) Phosphotungstic acid hematoxylin (PATH) staining of 3-month-old *PRELP*−/<sup>−</sup> bladder. PATH staining stains fibrin and erythrocytes. (**q**) Fibrin antibody staining of 3-month-old *PRELP*−/<sup>−</sup> bladder. (**r**) Analysis of urinary fibrin. Urine samples were collected from *WT* and *PRELP*−/<sup>−</sup> mice at the morning and were tested using Multistix (SIEMENS). \*, \*\*, \*\*\* indicate *p* < 0.05, *p* < 0.01, *p* < 0.001, respectively.

### *2.8. Expression Profiling of OMD*−/−*, PRELP*−/<sup>−</sup> *Bladder Epithelia*

To consolidate our hypothesis that OMD and PRELP contribute to the maintenance of cell–cell adhesion and the inhibition of EMT, we performed gene expression profiling by RNA-seq using isolated bladder epithelia from *WT* mice (*n* = 3), *OMD*−/<sup>−</sup> (*n* = 5), and *PRELP*−/<sup>−</sup> (*n* = 3). Similarly to our previous gene expression analysis data (Figure 2), 148 genes were commonly affected both in *OMD*−/<sup>−</sup> and in *PRELP*−/<sup>−</sup> (Figure 8a), indicating their partial functional redundancy.

**Figure 8.** Expression profiling of *OMD*−/−, *PRELP*−/<sup>−</sup> mouse bladder epithelia. Expression profiling was performed using *OMD*−/−, *PRELP*−/<sup>−</sup> mouse bladder epithelia. Data were analyzed, and the following figures were made through the use of IPA (QIAGEN Inc., https://www.qiagenbioinformatics.com/products/ ingenuitypathway-analysis). (**a**) Significantly affected gene numbers, including both up and downregulated. (**b**) Expression of genes related to cell–cell adhesion and EMT.(**c,d**) Significantly affected cell adhesion-related pathways in *OMD*−/<sup>−</sup> (**c**) and *PRELP*−/<sup>−</sup> (**d**). The same pathways are connected by arrows. (**e**) Similarity of expression profiling data. Using Analysis Match software (Ingenuity Pathway Analysis, IPA, Qiagen), we examined the similarity of expression profiling data of OMD−/<sup>−</sup> retina with the already deposited publicly available expression profiling dataset and those of PRELP−/<sup>−</sup> retina. OMD−/<sup>−</sup> retina data showed high similarity with PRELP−/<sup>−</sup> retina. The public database search revealed that bladder cancer-related datasets showed high similarity in all categories. CP; canonical pathways, UR; upstream regulators, CN; causal networks, DE; downstream effectors. (**f**) Schematic drawing of "Regulation of the EMT pathway" in *OMD*−/<sup>−</sup> vs. *WT*. Drawing was slightly modified from the original of "Regulation of the EMT Pathway". This image was created by Ingenuity Pathway Analysis according to their rule.

These genes include components of cell–cell adhesion and EMT (Figure 8b). Ontological analysis using the IPA showed that EMT-related events such as "Regulation of the Epithelial–Mesenchymal Transition Pathway" (*OMD*−/−, *z* = 6.26, *PRELP*−/−, *z* = 1.31) (Figure 8c–e) were significantly affected in both *OMD*−/<sup>−</sup> and *PRELP*−/<sup>−</sup> bladder epithelia. Additionally, cell–cell adhesion-related pathways, which is a consequence of EMT, such as "Tight Junction Signaling", and "Germ Cell–Sertoli Cell Junction Signaling", were significantly affected both in *OMD*−/<sup>−</sup> and *PRELP*−/−, confirming their involvement in the maintenance of the epithelial junctional barrier.

The ontological analysis also revealed that many cancer-related pathways are more strongly affected (Figure S7a,b), even to a higher extend compared to the gene expression profiling performed in cell lines (Figure 2). Many oncogenes and tumor-suppressor genes are strongly affected (Figure S7c). Figure S6d shows the schematic diagram of "Molecular Mechanisms of Cancer" pathway, affected in *OMD*−/<sup>−</sup> (*z* = 15.2), indicating that a majority of cancer-related regulators such as NF-kB, p53, myc, Ras, c-Jun/c-Fos, TGF-β R1/2, and RB are significantly affected. Since the host mouse strain C57BL/6J is not known to hold tumorigenic mutations in the above proteins, these data confirm that in parallel with their ability to regulate EMT and cell–cell integrity, OMD and PRELP have the ability to influence cancer-related activities. Furthermore, in order to know how deeply the OMD suppression in bladder cancer contributes to the properties of bladder cancer, we searched already deposited publicly available expression profiling datasets that showed similarities with that of OMD−/<sup>−</sup> retina. This analysis revealed that many cancer-related public datasets showed strong similarity with our OMD−/<sup>−</sup> dataset. In particular, as shown in Figure 8e, both bladder transitional cell carcinoma and bladder carcinoma showed the strong similarity [35,36]. This result demonstrates the significant contribution of OMD suppression in human bladder cancer initiation and/or progression. In addition, we examined the similarity between the OMD−/<sup>−</sup> and PRELP−/<sup>−</sup> expression profiling datasets using the Analysis Match software. Figure 8e shows the high similarity between OMD−/<sup>−</sup> and PRELP−/−, supporting the results in Figure 8a–d.

### *2.9. Breakdown of the Umbrella Cell Layer in OMD*−/<sup>−</sup> *and PRELP*−/<sup>−</sup> *Mice*

We made 10 µm paraffin section series from whole bladder specimens of *WT*, *OMD*−/−, *PRELP*−/−, and double knockout mice and examined the fine structure of the urothelium. In the *WT* mice, bladder umbrella cells form a clear single epithelial layer at the apical side of the urothelium and function as a barrier to the toxic bladder fluid (Figure 9a–c). In contrast, all of the bladder tissue samples from *OMD*−/−, *PRELP*−/−, and the double knockout mice showed points of breakdown/dysplasia of the urothelium (Figure 9d–l). We here termed these histological structures as "epithelial bursts". Furthermore, histological observation and bladder marker staining showed that the spread cells of the epithelial bursts originated from umbrella cells expressing uroplakin-III (Figure 9m,n), while their number was significantly increased in *OMD*−/<sup>−</sup> or *PRELP*−/<sup>−</sup> mice (Figure 9o). Of note, no obvious abnormalities were seen in the basal and intermediate cell layers (Figure 9p). To investigate whether the epithelial bursts are associated with aberrant cell proliferation, we performed immunohistochemical analysis using the Ki67 proliferation marker. There are few Ki67-positive cells in the *WT* bladder urothelium, and their number is only slightly increased in the *OMD*−/<sup>−</sup> and the double knockout samples, suggesting that the epithelial bursts do not result from increased proliferation (Figure 9q).

In humans, carcinoma in situ (CIS) appears histologically as a flat dysplasia of umbrella cells and is recognized as an early sign of malignant bladder cancer. However, an epithelial burst-type dysplasia, as seen in the *OMD*−/<sup>−</sup> and *PRELP*−/<sup>−</sup> mouse bladders, has not been recognized. The luminal mouse bladder is consistently covered by convex mucosal folds, while the human bladder surface is relatively flat or slightly concave. During our histological analysis, we observed a simple flat dysplasia of umbrella cells in the concave areas of mouse bladder as in human CIS (Figure 9r–t), suggesting that the structural difference of dysplasia might result from the different urothelium structure: convex vs. concave. To address this, we developed a mathematical simulation to visualize the direction of epithelial layer breakdown through the calculation of the forces created on convex and concave structures (Figure 9u). The model demonstrated that in a convex structure, the basal side of the epithelial layer was sealed, and the epithelial cells tended to escape to the apical side, similar to an epithelial burst. On the other hand, in a concave structure, the apical side was sealed, and the dysplasia cells tended to move under the epithelial layer. Supporting our analysis, Messal et al. has recently reported that a mechanical tension model for tissue curvature can instruct the direction of cancer morphogenesis [37]. These model-based analyses suggest that OMD and/or PRELP deletion can result in a defect in maintenance of the umbrella cell layer, as observed in human bladder CIS.

**Figure 9.** *OMD*−/−, *PRELP*−/−, and their double knockout mice spontaneously initiate bladder papillary cancer. (**a**–**l**) 3-month-old bladder of *WT* (**a**–**c**), *OMD*−/<sup>−</sup> (**d**–**f**), *PRELP*−/<sup>−</sup> (**g**–**i**), and their double knockout (**j**–**l**). A low magnification and two high magnification images are shown in order. Epithelial bursts are indicated as arrows. Some enlarged areas are indicated as boxes in low-magnification images. Scale bar represents 500 µm (**a**,**d**,**g**,**j**) and 100 µm (**b**,**c**,**e**,**f**,**h**,**i**,**k**,**l**). (**m**–**o**) Uroplakin III antibody staining

of an epithelial burst of OMD−/−. Uroplakin staining (m), overlaid view of uroplakin III and DAPI (**n**). Scale bar represents 50 µm (**m**,**n**). Quantification of epithelial burst number per bladder; *WT* (*n* = 7), *OMD*−/<sup>−</sup> (*n* = 7), *PRELP*−/<sup>−</sup> (*n* = 6), *OMD*−/−, *PRELP*−/<sup>−</sup> (*n* = 3) (**o**). In quantification, we examined six to seven 10 µm slices from each bladder. Each slice was separated around 200 µm in the bladder, and these slices covered the whole bladder except their edges. (**p**) Laminin antibody staining of *WT* and *OMD*−/<sup>−</sup> bladders. (**q**) Ki-67 staining positive cells in bladder. (**r**–**t**) Carcinoma in situ (CIS)-like structures in *OMD*−/<sup>−</sup> (**r**) and *PRELP*−/<sup>−</sup> (**s**,**t**). (**u**) Computational models for the mouse and human epithelial dysplasia. Conditions of models (**u-i**). Calculated forces between cells (**u-ii**). Direction of dysplasia (**u-iii**). Epithelial burst-like dysplasia and carcinoma in situ-like dysplasia (**u-iv**). \*\*\* indicates *p* < 0.001.

### *2.10. Some PRELP*−/<sup>−</sup> *Mice Spontaneously Initiate Bladder Papillary Cancer*

On analysis of bladders from *OMD*−/−, *PRELP*−/−, and the double knockout mice, we found that *OMD*−/−, *PRELP*−/−, and double KO bladders showed a slightly increased number of mucosal folds with multiple branches (Figure 10a,b). Interestingly, in one-third of the *PRELP*−/<sup>−</sup> and double knockout mice but not in OMD−/<sup>−</sup> mice, the bladder developed abnormal urothelia with hyperplasia, resulting in a pattern of papillary growth on a normal muscularis (Figure 10c,e in WT, d, f–o in PRELP). This phenotype seen in some *PRELP*−/<sup>−</sup> bladders is similar to some types of human bladder papillary cancer (https://www.proteinatlas.org/learn/dictionary/pathology/urothelial+cancer).

We observed various stages of papillary cancer progression such as mucosal folds with multiple branches (Figure 10g), partially fused mucosal folds (Figure 10h), and completely fused mucosal folds (Figure 10i,j). The process of clot formation was also observed, including small aggregates of proteinaceous material secreted from umbrella cells (Figure 10i), larger aggregates in which clumps of cells were embedded (Figure 10k,l), and large acellular clots covered with a single layer of cells (Figure 10m). We observed early signs of cancer invasions into the underlying muscularis (Figure 10n,o).

**Figure 10.** *PRELP* knockout mice spontaneously initiate early stages of bladder cancer. (**a**) H&E-stained section showing a branched mucosal fold. (**b**) Number of mucosal folds with multiple branches. In each bladder, we have examined two sections in the medial region of bladder. (**c**) H&E-stained section of a *WT* mouse bladder at 3 months of age. (**d**) Bladder papillary cancer in *PRELP*−/<sup>−</sup> at 3 months. Scale bar represents 500 µm (**c,d**). The bladder lumen is almost completely filled by mucosal folds with multiple branches and fused mucosal folds. Clots formation is observed. Enlarged regions in the following panels are indicated by the dotted boxes. (**e**) *WT* bladder muscularis (Mus) and epithelial tissue (Epi). (**f**) *PRELP*−/<sup>−</sup> bladder muscularis. (**g**) Mucosal fold with multiple branches. (**h**) Partially fused mucosal folds with multiple branches. (**i**) Fused mucosal folds. The arrow points secretion of materials to lumen. (**j**) Fused mucosal folds. Deposited material is enriched in fused folds (arrow). (**k**) Separation of epithelial cells with sticky material. (**l**) Aggregation of separated cells with clot materials. (**m**) The clot is covered by a layer of cells. (**n**) T1 stage bladder cancer in *PRELP*−/<sup>−</sup> at 3 months. (**o**) Epithelial papillary cancer integration into muscularis (arrows). \*\* and \*\*\* indicate *p* < 0.005, and *p* < 0.001, respectively.

#### **3. Discussion**

#### *3.1. ECM Proteins and Cancer Initiation*

OMD and PRELP are secreted ECM proteins, belonging to the Class II SLRP subfamily [38–40]. SLRP family members were originally identified as abundant proteins within the ECM of cartilage, connecting tissues and differentiating osteoblasts [41–43]. ECM proteins of the tumor microenvironment play important roles in many aspects of cancer initiation and progression [44]. One member of the SLRP family, decorin expression, decreases on the malignant transformation of tumor cells. Thirty percent (30%) of decorin knockout mice developed spontaneous intestinal tumors [13]. On the other hand, in an inflammation murine model, decorin is upregulated in endothelial cells and facilitates the downregulation of tight junctions [45]. This suggests that inflammation may affect OMD and PRELP function.

Here, we have demonstrated that OMD and PRELP function to maintain epithelial cell–cell integrity in urothelial cells through the inhibition of partial EMT. At epithelial cancer initiation, EMT is required, while MET is observed at cancer metastasis. Recent comprehensive expression profiling analyses in bladder and other epithelial cancers have revealed a novel concept of partial EMT [46–49]. The typical partial EMT state is the loss of tight junctions without affecting adherens junctions [48]. This is particularly important for understanding cancer initiation. In bladder cancer, a loss of E-cadherin expression is used as a marker of advanced bladder cancer, suggesting that the partial EMT state might be associated with early-stage bladder cancer. The tight junctions between umbrella cells in *OMD*−/<sup>−</sup> and *PRELP*−/<sup>−</sup> mice disappeared, while adherens junctions were maintained, indicating a typical partial EMT state. The loss of tight junctions resulted in disruption of the apical–basal polarity of umbrella cells, which is demonstrated by uniform E-cadherin staining around umbrella cells. Moreover, the partial EMT state we observed is susceptible for breakdown of the umbrella-cell layer, which might be related to cancer initiation. Collectively, our findings might be the first demonstration of partial EMT state and associated bladder cancer initiation in mice.

#### *3.2. OMD and PRELP and NMIBC Initiation*

*OMD*−/<sup>−</sup> or *PRELP*−/<sup>−</sup> mice showed many breakdown sites in the umbrella-cell layer, and one-third of *PRELP*−/<sup>−</sup> developed large-scale papillary cancer without muscle invasion. A large region of chromosome 9q, including the *OMD* gene, is deleted in half of NMIBC cases [50]. The deletion is associated with the initiation of NMIBC [51]. *PTCH* and *TSC1* were proposed to be the critical tumor-suppressor genes in 9q deletions [52,53], but this hypothesis is controversial [54]. Rather, with the present study, we propose *OMD* as a novel 9q-residing tumor-suppressor gene involved in cancer bladder initiation.

NMIBC is clinically classified as Ta, T1, or CIS. CIS is proposed to originate from umbrella cells because the cells in CIS are positive to umbrella cell markers such as CK20 [55]. Recent comprehensive expression profiling analysis classified NMIBC into three classes. Among these, Class 2 has the expression of CIS type markers, and Class 2 is defined based on the expression of EMT marker genes [20]. *OMD*−/<sup>−</sup> or *PRELP*−/<sup>−</sup> showed two types of breakdown of the umbrella layer: epithelial bursts and CIS-like structures. Our mathematical model indicates that the difference between the two breakdowns reflects the structural differences of the epithelia. We propose that umbrella-layer breakdown mediated by the loss of OMD and PRELP may initiate CIS. Some bladder cancers are thought to originate from the umbrella cells [51], because selective overexpression of a mutant H-Ras in umbrella cells resulted in low-grade papillary tumors [56–58].

Additionally, *PRELP*−/<sup>−</sup> mice tended to form protein clots, including fibrin, in the bladder. The fibrin/fibrinogen degradation products in human urine samples have been used as a bladder cancer marker [59]. The leakage of fibrin is regulated by the blood–urine barrier in bladder epithelial cells. This suggests that damage to the blood urine barrier is associated with bladder cancer initiation and that PRELP may have the ability to regulate the blood–urine barrier. Interestingly, we have found

that in *OMD*−/<sup>−</sup> and *PRELP*−/<sup>−</sup> mice, umbrella cells are connected to each other by adherens junctions. It is known that the loss of E-cadherin is a marker of conversion from benign to malignant bladder cancer. Thus, double knockout of *OMD*/*PRELP* and E-cadherin may reveal the process of malignant cancer initiation.

#### *3.3. EMT*/*MET Regulated by OMD and PRELP*

During malignant transformation, cancer cells have acquired mesenchymal-like characteristics such as anoikis resistance and invade adjacent tissues. Our results showed that OMD or PRELP overexpression in bladder cancer cells resulted in an increase of epithelial-like properties such as tight junction induction and adherens junction activation as well as a change of EMT markers. A cardinal feature of cancer is the ability for anchorage-independent growth, which changes the properties of cell–cell and cell–matrix adhesion conferred at EMT.

Umbrella cells secrete signaling proteins such as EGF and TGF-β [60]. The concept that OMD or PRELP mediated the inhibitory activity of TGF-β and EGF pathways could be important for the regulation of EMT/MET, because the TGF-β/Smad2 pathway is the biggest common target of all SLRP family members [2] and is a well-known regulator of EMT/MET [46]. In addition, EGF is known as a major regulator of EMT/MET [46] and is one of the most established targets of cancer treatment [61]. Previously, we reported that the simultaneous regulation of Xnr2, FGF, and BMP pathways by Tsukushi, another SLRP member, had an increased synergistic effect compared to the single regulation of each pathway alone [6].

OMD and PRELP are selectively expressed in the ciliary body of the retina and in ependymal cells in the brain (paper in preparation) that are characterized by strong tight junctions forming the blood–CSF barrier. The expression of many components of tight junctions is associated with tumorigenesis [62]. However, so far, there is no report showing that the knockout of any tight junction component by itself can spontaneously lead to tumor formation, although, hyperplasia of the gastric epithelium has been observed in an occludin knockout model [63]. This suggests that the loss of tight junctions alone is not sufficient to initiate bladder cancer. TGF-β and EGF pathways are involved in the regulation of many cancer-associated signaling pathways, suggesting that in addition to the loss of tight junctions in an *OMD*−/<sup>−</sup> or *PRELP*−/<sup>−</sup> bladder, further regulation of TGF-β and EGF downstream signaling components might be required for cancer initiation. Of note, one limitation of our study is that although the TGF-β-flag protein bound with OMDmyc and PRELPmyc proteins directly, the binding affinities of secreted TGF-β to the OMD and PRELP is unknown; therefore, further studies are required.

### *3.4. The Similarity and Di*ff*erence between OMD and PRELP*

Both OMD and PRELP were downregulated, especially in bladder cancer. Our results indicated that although OMD and PRELP share considerable amount of signal pathways, there are some differences in the observed phenotypes: branching, proliferation, bladder cancer progression, and protein expression. Functional difference between OMD and PRELP may be associated with certain cancer phenotypes. This indicates that they would play a redundant and non-redundant function in bladder cancer.

#### *3.5. Diagnostic and Therapeutic Potential of OMD and PRELP in Bladder Cancer*

*OMD* and *PRELP* are expressed in normal human epithelia. However, in many epithelial cancers, they are strongly downregulated. Particularly, their expression in the bladder is drastically reduced even in very early stages of cancer. This potentially means that it is possible to classify a patient's clinical state based solely on their OMD and PRELP expression status from early-stage cancers. So far, several diagnostic markers of bladder cancer have been used in clinics such as BTA-Stat (sensitivity 50–70%, specificity 67–78%) and fibrin degradation products (FDP) (sensitivity 52–68.4%, specificity 79.6–91%) [64]. With our findings, we show that the assessment of *OMD* and *PRELP* expression status

can be used as a novel, more sensitive, criterion in assessing the initiation and progression of bladder cancer. We also observed a similar evaluation in renal cell carcinoma and retinoblastoma (paper in preparation), proposing their diagnostic potential in various epithelial cancers, possibly through using new technology such as quench bodies to detect loss-of-function regions [65]. This study demonstrates that the functions of OMD and PRELP are partially redundant in the regulation of both cell–cell integrity and cancer initiation/progression, and they are potentially important, especially for bladder cell therapeutics.

#### **4. Materials and Methods**

Materials and Methods are described in the Supplementary Materials and Methods. The accession number for the raw and processed data of microarray and RNA-seq data from *OMD* and *PRELP* knockdown experiments reported in this paper is GEO: GSE63955 and GSE144295. Other data supporting our findings can be found either in this article or in the supplementary materials. Please contact the corresponding author for all "unpublished data" and "paper in preparation" requests.

The research protocol was reviewed and approved by the Ethical Committee of Addenbrooke's Hospital, Cambridgeshire Local Research Ethics Committee (No. 03/018).

#### **5. Conclusions**

In this study, we demonstrated that two SLRP proteins, OMD and PRELP, are novel activators of the cell–cell integrity by inhibiting EMT through the simultaneous inhibition of TGF-β and EGF signaling. The downregulation of OMD and PRELP expression was observed in all of the cancers we analyzed, including bladder cancer. We showed that in association with a change of EMT states, OMD or PRELP suppression in mice resulted in an initiation of bladder cancer, while the activation of OMD or PRELP inhibited bladder cancer progression in vitro and in vivo. We propose that OMD and PRELP-mediated regulation of EMT is important for the initiation of human bladder cancer.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2072-6694/12/11/3362/s1, Figure S1: Microarray analysis of *OMD* and *PRELP* expression in various cancers and normal tissues; Figure S2: Expression analysis of OMD and PRELP in various human cancers using Gene Logic Inc. Figure S3: *OMD* and *PRELP* expression analysis in various cancer cells and ontological analysis of expression profiling data; Figure S4: Effect of OMD or PRELP on cell properties under standard cell culture conditions; Figure S5: Quantification of OMD and PRELP effects; Figure S6: PRELP is expressed in subpopulation of bladder umbrella epithelial cells; Figure S7: The ontological analysis in *OMD*−/<sup>−</sup> and *PRELP*−/<sup>−</sup> bladder epithelia; Figure S8: Original Western Blotting images used in Figure 5.; Table S1: Statistical analysis of *OMD* and *PRELP* expression levels in clinical bladder tissues; Table S2: Relationship between *OMD* and *PRELP* expression levels and carcinogenesis; Table S3: Primer sequences for quantitative RT-PCR.

**Author Contributions:** V.P., R.H., K.A., T.T., J.K.W., A.L., J.H., M.D., N.S., H.D., S.N.-Z., R.L., M.N., R.T., A.V. and S.-i.O. participated in project conception, performed the experiments, and analyzed the data. V.P., R.H., K.A., T.T., J.K.W., and S.-i.O. wrote the original draft. V.P., R.H., K.A., T.T., J.K.W., A.L., J.H., M.D., N.S., H.D., S.N.-Z., R.L., M.N., R.T., A.V., S.K., M.S.S., G.M., A.M., K.T., J.D.K., and S.-i.O. discussed the data, reviewed and edited the manuscript. R.H., G.M., K.T., J.D.K., and S.-i.O. supervised the experiments. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work is partially supported by Santen SensyT PhD studentships to A.L., Cancer Research UK (C1528/A2690), The Great Britain Sasakawa Foundation (B85), St Peter's Trust for Kidney Bladder & Prostate Research (NA), Fight for Sight (F94), and Childhood Eye Cancer Trust (24CEC12) to S.O.

**Acknowledgments:** We thank Bruce Ponder and David E. Neal for their initial contributions to this project and all members of the Ohnuma lab, particularly Stephen Bolsover, Ryohei Sekido, and Kevin Broad for critical reading of the manuscript. We also thank Tatsuhiko Tsunoda for assistance in cancer genome data analysis, to Karl Matter, Maria Balda, Vassiliki Saloura for helpful discussion and reagents, and to Stewart McArthur for the network analysis. We thank Alex Freeman for uropathological analysis.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


65. Abe, R.; Ohashi, H.; Iijima, I.; Ihara, M.; Takagi, H.; Hohsaka, T.; Ueda, H. "Quenchbodies": Quench-Based Antibody Probes That Show Antigen-Dependent Fluorescence. *J. Am. Chem. Soc.* **2011**, *133*, 17386–17394. [CrossRef]

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Cancers* Editorial Office E-mail: cancers@mdpi.com www.mdpi.com/journal/cancers

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18

www.mdpi.com ISBN 978-3-0365-3681-1