Extracellular Hsp90α and clusterin synergistically promote breast cancer epithelial-to-mesenchymal transition and metastasis via LRP1 Free

Heat shock protein 90 (Hsp90) is a ubiquitously expressed molecular chaperone that plays essential roles in various biological processes (Whitesell and Lindquist, 2005). Currently, more than 300 proteins have been identified as Hsp90 interactors with a range of diverse functions (http://www.picard.ch/downloads). The activity of Hsp90 can be precisely regulated by co-chaperones and post-translational modifications (Lu et al., 2014; Trepel et al., 2010; Wang et al., 2012a). Hsp90 has two isoforms in the cytosol, Hsp90α (also known as HSP90AA1) and Hsp90β (also known as HSP90AB1) (Chen et al., 2005). Intriguingly, the Hsp90α isoform can be secreted extracellularly (Eustace et al., 2004; Yang et al., 2014) and promotes tumor cell migration, invasion and metastasis in various cancer types such as colon cancer (Chen et al., 2013), breast cancer (Stellas et al., 2010) and prostate cancer (Nolan et al., 2015). Antibodies or cell-impermeable inhibitors against eHsp90α dramatically inhibit cancer metastasis (Devarakonda et al., 2015; Tsutsumi et al., 2008). The positive correlation between extracellular (e)Hsp90α level and tumor malignancy has previously been reported in both tumor cell lines and human cancer patients by our group (Wang et al., 2009). Moreover, plasma Hsp90α has recently been shown to be an excellent biomarker for liver and lung cancer in the clinic (Fu et al., 2017; Luo et al., 2016).

The significant role of eHsp90α in tumor metastasis raises another critical question of how eHsp90α executes its pro-motility function. eHsp90α is generally considered to perform its function by interacting with other proteins. Several studies have demonstrated that eHsp90α interacts with and activates matrix metalloproteinase 2 (MMP2) (Song et al., 2010; Stellas et al., 2010), which can be assisted by extracellular members of the heat shock protein 70 (Hsp70) family (Sims et al., 2011). Other proteins involved in extracellular matrix degradation, including matrix metalloproteinase 9 (MMP9) (Stellas et al., 2010), tissue plasminogen activator (tPA, also known as PLAT) (McCready et al., 2010), and lysyl oxidase-like protein 2 (LOXL2) (McCready et al., 2014) have also been reported to be activated by eHsp90α.

Besides functions in modulating extracellular matrix proteins, accumulating evidence has shown that eHsp90α is a signaling factor, which allows alternative mechanisms for its migration-promoting function (Cheng et al., 2008; Li et al., 2007). An important consequence triggered by eHsp90α signaling is the activation of EMT in tumor cells, as previously demonstrated in prostate and colorectal cancer models (Chen et al., 2013; Hance et al., 2012). eHsp90α contributes to EMT activation through increasing the expression of EMT-related transcription factors such as Snail (also known as Snai1), Slug (also known as Snai2), Zeb1 and Twist1, and downregulating E-cadherin (Chen et al., 2013; Nolan et al., 2015).

Several cell­-surface receptors, such as low-density lipoprotein receptor-related protein 1 (LRP1) (Cheng et al., 2008), Toll-like receptor 4 (TLR4) (Thuringer et al., 2011) and human epidermal growth factor receptor 2 (HER2, also known as ERBB2) (Sidera et al., 2008) have been reported to mediate eHsp90α signaling. LRP1 is a ubiquitously expressed cell-surface receptor that either acts as a scavenger receptor to regulate lipoprotein metabolism and protease homeostasis, or a signaling receptor that modulates cell survival and motility (Herz and Strickland, 2001). Elevated expression of LRP1 is associated with numerous human diseases such as breast cancer, prostate cancer and Alzheimer's disease (Lin and Hu, 2014). It has been demonstrated that LRP1 mediates eHsp90α-induced cell migration and signaling activation by AKT, ERK and NF-κB proteins (Chen et al., 2010; Tsen et al., 2013). However, it is unknown whether there are other factors involved in LRP1-mediated eHsp90α signaling.

Here, we explored new extracellular interaction partner(s) of eHsp90α and obtained evidence for clusterin as a newly discovered binding partner of eHsp90α, which modulates eHsp90α functions through increasing its binding affinity to LRP1. eHsp90α and clusterin synergistically induced EMT and cell migration through activating the AKT, ERK and NF-κB signaling pathways, which ultimately results in enhanced tumor metastasis.

To identify extracellular proteins that interact with eHsp90α, a pull-down assay with human serum was performed. eHsp90α was obtained from conditioned medium (CM) of MCF-7 cells using a NI-NTA purification system, then coupled to cyanogen bromide-activated sepharose beads. Human serum was first pre-treated with an Albumin/IgG Depletion Kit, which specifically removes albumin and IgG (Fig. S1A, lane 2 versus lane 1), and incubated with eHsp90α-coupled sepharose beads. The precipitated fractions were resolved by SDS-PAGE and subsequently underwent mass spectrometry analysis (Fig. 1A, lane 4). Proteins with a mass spectrometry protein score over 50 were further analyzed by DAVID functional annotation (Table S1), showing that these eHsp90α-interacting proteins were involved in various biological processes including immunity, hemostasis and lipid metabolism (Fig. 1B). We reviewed the literature on nine proteins that were reported to promote tumor metastasis (Table 1), and chose clusterin (peptides shown in Fig. S1B) for further investigation because: 1) clusterin is an extracellular chaperone (Humphreys et al., 1999) that regulates extracellular protein homeostasis (Wyatt et al., 2013); 2) it can modulate signaling activation (Byun et al., 2014; Gil et al., 2013) through cell-surface receptors such as LRP2 (Kounnas et al., 1995); and 3) expression levels of clusterin are correlated with metastasis in multiple types of cancer, including breast cancer (Redondo et al., 2000), liver cancer (Lau et al., 2006), as well as colon cancer (Pucci et al., 2009). Therefore, we hypothesized that clusterin may modulate eHsp90α functions.

To confirm the above results, we also detected eHsp90α-interacting proteins in tumor cell CM, among which clusterin was also detected with a high percentage of peptide coverage (Table S2, Fig. S1C). To verify the physical interaction between eHsp90α and clusterin, we overexpressed Myc-tagged clusterin in MCF-7 cells and collected the CM to perform co-immunoprecipitation (co-IP) using anti-Hsp90α and anti-Myc antibodies. Immunoblotting results further confirmed that eHsp90α physically interacts with clusterin in the CM (Fig. 1C,D, lane 2).

One major function of eHsp90α is to promote cell migration, involving the activation of ERK, AKT, NF-κB, and Src signaling pathways (Wong and Jay, 2016). Consistently, our results demonstrated that eHsp90α increased tumor cell migration and phosphorylation of ERK, AKT and NF-κB proteins in a dose-dependent manner in breast cancer cell lines MCF-7 and MDA-MB-231 (Fig. S2A,B). To investigate whether clusterin is involved in eHsp90α-promoted tumor cell migration, a transwell migration assay was performed with eHsp90α and clusterin. As shown in Fig. 2A,B, treatment with purified eHsp90α (5 μg/ml) or clusterin (5 μg/ml) alone enhanced tumor cell migration, while co-administration of eHsp90α and clusterin elicited a synergistic effect. Similarly, the phosphorylation of ERK, AKT and NF-κB proteins was elevated after co-administration of eHsp90α and clusterin in both MCF-7 and MDA-MB-231 cells (Fig. 2C, lane 4 versus lanes 1, 2, 3). To further investigate the relationship between eHsp90α and clusterin, we knocked down clusterin with siRNA and examined the pro-migration effect of eHsp90α. While knockdown of clusterin did not affect Hsp90α levels in the CM (Fig. 2D, lane 4 versus lane 3), eHsp90α-induced tumor cell migration was significantly attenuated (Fig. 2E,F), implying decreased eHsp90α activity in the absence of clusterin. To exclude possible off-target effects of clusterin siRNA, we repeated the experiment using another siRNA sequence (si CLU #2) and observed consistent results (Fig. S2C–E). In agreement with these results, treatment of MCF-7 and MDA-MB-231 cells with anti-Hsp90α and anti-clusterin neutralizing antibodies reduced cell migration (Fig. 2G; Fig. S2F). Additionally, we did not observe any synergistic effect of eHsp90α and clusterin on cell proliferation (Fig. S2G).

One mechanism of eHsp90α-induced cell motility is through modulating EMT (Hance et al., 2012). Thus, we examined whether clusterin could cooperate with eHsp90α to synergistically promote EMT. Epithelial-like MCF-7 cells exhibited a spindle-shaped morphology after co-treatment with eHsp90α and clusterin, as shown by microscope images and cytoskeleton staining (Fig. 3A). Furthermore, we evaluated protein and mRNA levels of EMT markers, and found that co-treatment of eHsp90α and clusterin decreased E-cadherin expression but increased the expression of N-cadherin, Snail, Slug and Zeb1, all of which were minimally influenced by eHsp90α treatment alone (Fig. 3B; Fig. S3A; Fig. 3C, lane 4 versus lane 1). Meanwhile, we also examined the distribution of E-cadherin by immunofluorescence. The results showed that eHsp90α and clusterin decreased E-cadherin-based cell–cell contacts in MCF-7 cells (Fig. 3D), further demonstrating the occurrence of EMT induced by eHsp90α and clusterin in MCF-7 cells. Even in mesenchymal-like MDA-MB-231 cells, co-treatment of eHsp90α and clusterin induced an elongated morphology and F-actin enrichment at cell protrusions (Fig. 3E), along with increased mesenchymal marker expressions (Fig. 3F; Fig. S3B; Fig. 3G, lane 4 versus lane 1), as well as decreased and diffused E-cadherin expression (Fig. 3H). Moreover, knocking down of Snail, Slug or Zeb1 inhibited cell migration induced by eHsp90α and clusterin both in MCF-7 and MDA-MB-231 cells (Fig. 3I; Fig. S3E,F). These results demonstrate that clusterin interacts with eHsp90α to synergistically activate the EMT process, which is responsible for eHsp90α- and clusterin-induced tumor cell migration.

To explore which signaling pathway(s) that mediate eHsp90α and clusterin-induced EMT, we treated cells with inhibitors of AKT (LY294002), NF-κB (BAY11-7082) or ERK (U0126) proteins. Immunoblotting and qRT-PCR results showed that blockade of these signaling pathways inhibited the upregulation of Snail, Slug and Zeb1 caused by eHsp90α and clusterin (Fig. 3J, lane 4, 5, 6 versus lane 3; Fig. S3C,D). Meanwhile, these inhibitors also decreased cell migration induced by eHsp90α and clusterin (Fig. 3K; Fig. S3G). These results demonstrate that eHsp90α and clusterin promote EMT and tumor migration through activation of signaling pathways in both MCF-7 and MDA-MB-231 cells.

It has been widely reported that eHsp90α elicits autocrine signaling through its receptor LRP1, which is responsible for activation of multiple signaling pathways, induction of EMT and enhancement of cell motility (Woodley et al., 2009). To elucidate whether clusterin was involved in the interaction between eHsp90α and LRP1, we used immunofluorescence to detect co-localization between eHsp90α and LRP1, in the presence or absence of clusterin. As shown in Fig. 4A,B, eHsp90α and LRP1 co-localized with each other, and the co-localization is more obvious in clusterin-treated cells. To confirm the result, we performed in situ proximity ligation assay (PLA) in both MCF-7 and MDA-MB-231 cells. PLA signal, which represents the association between eHsp90α and LRP1, was significantly increased by co-treatment with eHsp90α and clusterin as compared with eHsp90α treatment alone, suggesting that clusterin facilitated the interaction between eHsp90α and LRP1 (Fig. 4C,D). To further verify the result, we pre-treated MCF-7 and MDA-MB-231 cells with receptor-associated protein (RAP, also known as LRPAP1), a ligand-binding antagonist of LRP1 (Bu, 2001). In this case, the binding of eHsp90α to LRP1 was abolished regardless of the presence or absence of clusterin (Fig. S4A,B).

As we had determined that clusterin interacts with eHsp90α and increases the binding affinity of eHsp90α to LRP1, we next explored whether clusterin was part of the eHsp90α–LRP1 complex. To detect any interaction between clusterin and LRP1, a PLA assay was performed using anti-Myc and anti-LRP1 antibodies. Increased PLA signal was detected in cells co-treated with eHsp90α and clusterin but not in cells treated with clusterin alone (Fig. S4C,D), suggesting that the interaction between clusterin and LRP1 occurs only in the presence of eHsp90α. To confirm a physical interaction of clusterin with the eHsp90α–LRP1 complex, a co-IP experiment was also performed. Cells were first incubated with eHsp90α and clusterin, and protein–protein interaction was stabilized by a crosslinker. Consistently, we detected LRP1 and eHsp90α in the co-IP system immunoprecipitated with anti-clusterin antibody (Fig. 4E, lane 2). Similar results were also obtained in cells immunoprecipitated with anti-Hsp90α and anti-LRP1 antibodies (Fig. 4E, lanes 3 and 4). Taken together, these results show that clusterin and eHsp90α form a complex with LRP1.

Next, we conducted a pull-down assay to provide further insights into the interaction model of the clusterin–eHsp90α–LRP1 complex. LRP1 protein was purified by immunoprecipitation and incubated with clusterin alone or in combination with eHsp90α. The pull-down assay showed that LRP1 could directly interact with eHsp90α (Fig. 4F, lane 2), but weakly interact with clusterin (Fig. 4F, lane 3). Intriguingly, when LRP1-coated beads were co-incubated with eHsp90α and clusterin, clusterin was found to interact with LRP1, accompanied with an increased interaction between eHsp90α and LRP1 (Fig. 4F, lane 4), suggesting that clusterin indirectly interacts with LRP1 through eHsp90α, which in return increases the binding affinity of eHsp90α to LRP1.

To further investigate whether the migration-promoting effects of eHsp90α and clusterin were mediated specifically by LRP1, we silenced LRP1 using siRNA and examined the migration of MCF-7 and MDA-MB-231 cells exposed to eHsp90α (5 μg/ml) and clusterin (5 μg/ml). We detected that the positive effect of eHsp90α and clusterin on tumor cell migration was abrogated in LRP1 knockdown cells (Fig. 5A; Fig. S5A). A similar result was also observed in cells treated with RAP (Fig. S5B,C). As expected, blocking LRP1 attenuated the phosphorylation of AKT, ERK, and NF-κB caused by eHsp90α and clusterin as well (Fig. 5B, lane 8 versus lane 4), demonstrating that clusterin facilitated eHsp90α signaling in an LRP1-dependent manner.

Next, we tested whether LRP1 overexpression could rescue the effects of eHsp90α and clusterin in LRP1 knockdown cells. Considering that the 13 kb cDNA for full-length LRP1 is too large to be accommodated and expressed (Lillis et al., 2005), we used ‘mini-LRP1’ to perform this experiment. The mini-LRP1, containing an extracellular ligand-binding subdomain (subdomain II at the α chain) fused to the intracellular β chain (Obermoeller-McCormick et al., 2001), has been demonstrated to mediate eHsp90α binding and rescue its migration-promoting function in LRP1 knockdown cells (Tsen et al., 2013). As shown in Fig. 5C and Fig. S5D, re-expressing mini-LRP1 could rescue the clusterin- and eHsp90α-induced migration-promoting effects in LRP1 knockdown cells. Similarly, mini-LRP1 overexpression also restored the activation of AKT, ERK and NF-κB proteins in response to eHsp90α and clusterin stimulation (Fig. 5D, lane 8 versus lane 4), which further confirmed the important role of LRP1 in mediating eHsp90α and clusterin function. In addition, to exclude any possible off-target effects of LRP1 siRNA, we repeated the experiments using another siRNA sequence (siLRP1 #2) and observed consistent results (Fig. S5E-H).

Taken together, we propose a novel regulatory mechanism of eHsp90α signaling, in which clusterin physically interacts with eHsp90α and enhances the binding affinity of eHsp90α to LRP1.

eHsp90α has been linked with tumor metastasis in many studies (Wang et al., 2009). Since we have demonstrated that clusterin facilitates eHsp90α function and synergistically promotes tumor migration in vitro, we speculated that they might also have a synergistic effect on promoting tumor metastasis in vivo. GFP-labeled MCF-7 cells were orthotopically inoculated into nude mice, and tumor-bearing mice were intravenously administered twice a week with purified eHsp90α (1 mg/kg body weight), clusterin (1 mg/kg body weight) or a combination of eHsp90α and clusterin. Consistent with the in vitro results, primary tumor volumes and weights showed no obvious difference among the four treatment groups (Fig. 6A,B). Sentinel lymph node metastasis was analyzed by measuring GFP-positive tumor cells. Immunofluorescence results showed that treatment with eHsp90α or clusterin alone produced a modest elevation in tumor metastasis, whereas combinational treatment elicited synergistic metastasis-promoting effects (Fig. 6C,D, right-hand panels). Meanwhile, the weight of sentinel lymph nodes was significantly increased by the co-administration of eHsp90α and clusterin (Fig. 6C,D, left-hand panels). Moreover, liver metastasis was also significantly increased by the co-treatment of eHsp90α and clusterin as shown by H&E staining in Fig. 6E,F.

Subsequently, we examined the influence of anti-Hsp90α and anti-clusterin neutralizing antibodies on tumor metastasis in a MDA-MB-231 xenograft model. Tumor volumes and tumor weights showed no obvious change after the administration of the two antibodies (Fig. 6G,H). Blocking either eHsp90α or clusterin led to 50% inhibition of sentinel lymph node metastasis, while co-treatment with anti-Hsp90α and anti-clusterin antibodies further decreased the metastasis, causing 60% inhibition compared with the control group (Fig. 6I,J, right-hand panels). A similar inhibitory effect was also observed on the weight of sentinel lymph nodes (Fig. 6I,J, left-hand panels). In addition, mice treated with anti-clusterin and anti-Hsp90α antibodies exhibited decreased liver metastasis compared with control mice (Fig. 6K,L).

We further validated the contributions of eHsp90α and clusterin to EMT in vivo. The immunohistochemistry results showed that eHsp90α and clusterin treatment promoted EMT process in MCF-7 tumor tissues, as reflected by increased expression levels of Zeb1, Slug, Snail, and decreased expression of E-cadherin (Fig. S6A,C). In contrast, anti-Hsp90α and anti-clusterin antibodies inhibited EMT activation in MDA-MB-231 tumor tissues (Fig. S6B,D). From these experiments, we conclude that eHsp90α and clusterin synergistically promote tumor cell metastasis in vivo, which provides evidence for extracellular clusterin and eHsp90α as potential therapeutic targets for breast cancer treatment.

The functions of eHsp90α are tightly correlated with cell migration and tumor metastasis. Thus, revealing the regulatory mechanism of eHsp90α activity will enlighten optimal design of eHsp90α-based anti-cancer therapies and optimize eHsp90α-based biomarkers for early detection of cancer. In this study, we used mass spectrometry to identify possible interacting proteins of eHsp90α in serum and tumor cell CM, and newly identified clusterin as an interacting protein of eHsp90α. Further investigations revealed that clusterin facilitated eHsp90α signaling through increasing the binding affinity of eHsp90α for its receptor LRP1, thus promoting tumor cell migration and metastasis synergistically.

Since the extracellular environment lacks ATP and co-chaperones, which are essential for modulating the intracellular functions of Hsp90 proteins, eHsp90α requires a series of different interacting proteins in the extracellular environment to exert its functions. Zou and colleagues identified Lys-270 and Lys-277 as the active sites for tumor-promoting functions of eHsp90α both in vitro and in vivo. Surprisingly, lysine substitutions at the corresponding residues cause Hsp90β to mimic eHsp90α function in vitro, but not in vivo (Zou et al., 2017), indicating that the full function of eHsp90α requires the assistance of other partners. In the present study, we have demonstrated an interaction between clusterin and Hsp90α both in tumor cell CM (Fig. 1C,D) and tumor tissues (Fig. S6E). The secreted clusterin is an 80 kDa disulfide-linked heterodimeric glycoprotein consisting of an α chain and a β chain with equal molecular weights. It has been reported that clusterin has at least three discrete ligand binding sites (Lakins et al., 2002). At present, although specific binding motifs of clusterin to its various ligands are not well understood, sites located in the C-terminus of the α chain and/or the N-terminus of the β chain may be responsible for mediating binding of un-stressed ligands (Lakins et al., 2002). Thus, these regions may also act as potential targets mediating the interaction between eHsp90α and clusterin under low-stress conditions.

Functionally, clusterin is recognized as a cytoprotective protein in the cytoplasm and extracellular spaces (Shannan et al., 2006), and is responsible for drug resistance and metastases formation in many tumor models (Garcia-Aranda et al., 2017). Lamoureux and colleagues have found that Hsp90 inhibition causes an elevation of clusterin expression, while silencing clusterin using OGX011 synergistically enhances the effect of Hsp90 inhibitor on prostate cancer metastasis (Lamoureux et al., 2011). Several studies have demonstrated that levels of clusterin in serum also elevate in cancer patients (Guo et al., 2014; Pucci et al., 2009). In our studies, using antibodies against Hsp90α and clusterin does not obviously inhibit tumor growth, but significantly inhibits cancer metastasis, which is consistent with previous reports (Wang et al., 2009) and reveals the potential of extracellular Hsp90α and clusterin as targets for cancer therapies.

Extracellular clusterin has been widely studied as a chaperone that stabilizes proteins under high-stress conditions (Humphreys et al., 1999) and regulates the endocytic degradation of unfolded proteins through its receptor LRP2 (Kounnas et al., 1995). Besides, extracellular clusterin also acts as a co-factor of extracellular signaling molecules to synergistically trigger intracellular signaling activation (Bajari et al., 2003; Byun et al., 2014; Jo et al., 2008). Given the fact that eHsp90α interacts with clusterin, we hypothesize that clusterin is a candidate eHsp90α signaling modulator. In this work, we find that clusterin assists in signaling activation of eHsp90α, which contributes to EMT and cell migration in MCF-7 and MDA-MB-231 breast cancer cells. These findings agree well with previous work showing that extracellular clusterin promotes EMT and migration in prostate cancer (Shiota et al., 2012) and liver cancer (Wang et al., 2012b). However, the underlying mechanisms of how extracellular clusterin facilitates EMT are still poorly characterized. Lenferink and colleagues have demonstrated that clusterin is required for TGFβ1-induced EMT and metastasis in breast cancer (Lenferink et al., 2010). However, they did not detect any direct cell-surface binding of clusterin, and an antibody that specifically blocks the binding of clusterin to LRP2 also failed to prevent clusterin-induced EMT, indicating that some other proteins in the extracellular space participate in this process (Lenferink et al., 2010). Our results agree with their hypothesis and provide a possible explanation for the mechanism of how clusterin promotes EMT. Specifically, clusterin induces EMT through interacting with eHsp90α and facilitating the activity of eHsp90α in an LRP1-dependent manner.

LRP1 is a receptor that binds to a variety of ligands with diverse roles during cancer progression (Van Gool et al., 2015). Many studies have demonstrated that eHsp90α stimulates intracellular signaling pathways through interacting with LRP1 (Chen et al., 2010; Tsen et al., 2013). Our results reveal that the synergistic effects of eHsp90α and clusterin also require LRP1, where clusterin acts as a modulator that enhances the binding affinity of eHsp90α to LRP1. However, the binding sites of eHsp90α and clusterin remain to be determined, as well as how clusterin binding influences the eHsp90α–LRP1 complex. It has been reported that eHsp90α interacts with LRP1 through the F-5 fragment, a 115-amino acid peptide retaining the whole pro-motility function of eHsp90α (Cheng et al., 2011). The first half of the F-5 fragment is located at the linker region of Hsp90α (a sequence between the N-terminal and middle domains of Hsp90α) and forms a dynamically disordered structure, while the second half of the F-5 fragment on the middle domain of Hsp90α shows a conserved structure (Sahu et al., 2012). Since Hsp90α is a highly flexible and dynamic protein (Penkler et al., 2018), we speculate that the binding of clusterin might cause a conformational change of eHsp90α that enhances the exposure of the F-5 fragment to LRP1, thus enhancing the binding of eHsp90α to LRP1.

In conclusion, our study demonstrates a novel regulatory mechanism of eHsp90α function by association with clusterin. Since clusterin is widely distributed in human tissues and extracellular fluids (de Silva et al., 1990), the clusterin pool produced by normal cells may also dynamically regulate the bioavailability of eHsp90α. Moreover, the discovery of a functional correlation between eHsp90α and clusterin also indicates the potential for future development of clusterin and eHsp90α as co-biomarkers for cancer diagnosis and prognosis.

MCF-7, MDA-MB-231 and HEK-293T cells were obtained from the American Type Culture Collection, and maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (FBS) (Thermo Fisher Scientific, Waltham, MA) in a cell culture incubator at 37°C with 5% CO₂.

Monoclonal antibody against Hsp90α for western blotting, cell migration assays and animal studies was developed by our laboratory as previously described (Song et al., 2010). Antibodies against clusterin (42143; 1:1000), E-cadherin (14472; 1:1000), N-cadherin (13116; 1:1000), Slug (9585; 1:1000), Zeb1 (3396; 1:1000), AKT (9272; 1:1000), p-AKT (4060; 1:1000), ERK (4695; 1:1000), p-ERK (4377; 1:1000), NF-κB (8242; 1:1000), p-NF-κB (3033; 1:1000), Src (2123; 1:1000), and p-Src (6943; 1:1000) were from Cell Signaling Technologies (Danvers, MA). Antibodies against p38 (sc7149; 1:500), p-p38 (sc7973; 1:500), JNK (sc7345; 1:500), p-JNK (sc6254; 1:500), and neutralizing antibody against clusterin (sc8354; 1:500) were from Santa Cruz Biotechnology (Dallas, TX). Antibodies against Snail (ab53519; 1:1000) and LRP1β (ab92544; 1:1000) were from Abcam (Cambridge, MA). Anti-LRP1α antibody (L2295; 1:1000) was from Sigma-Aldrich (St. Louis, MI). Anti-Myc antibody (11667149001; 1:2000) was from Roche (Basel, Switzerland). FITC-conjugated anti-F-actin antibody (bs-1571R-FITC; 1:100) was from Biosynthesis (Beijing, China). Anti-GAPDH antibody (TA-08; 1:2000), and FITC- (ZF-0312; 1:2000), TRITC- (ZF-0316; 1:2000) or HRP-linked (ZDR-5306, ZDR-5307; 1:2000) anti-mouse IgG and anti-rabbit IgG were from ZSGB-BIO Company (Beijing, China). PI3K/AKT inhibitor LY294002, NF-κB inhibitor BAY11-7082, and MEK/ERK inhibitor U0126 were from Selleck (Houston, TX). Recombinant Hsp90α was provided by PROTGEN (Yantai, China). Receptor-associated protein (RAP) was from Invitrogen (Carlsbad, CA).

Human serum samples were obtained from our previous clinical trial (NCT02324101) in Tianjin Cancer Hospital. This study was performed in accordance with the Declaration of Helsinki and approved by the Tianjin Cancer Hospital Medical Ethics Committee. All patients enrolled in this study gave written informed consent after a discussion of the possible risks and the use of the samples for scientific research.

Human Hsp90α with an N-terminal His tag or human clusterin with a C-terminal Myc–His tag were separately constructed into virus transfer plasmid plenty-CMV-MCSSV-BSD. 4 μg packaging plasmid pPAX2, 400 ng envelope plasmid pVSVG, and 4 μg transfer plasmid were transfected into HEK-293T cells using polyethylenimine. Cell medium were collected 48 h after transfection and precipitated by means of lentivirus concentration solution. MCF-7 cells were infected with virus for 24 h, and cells with stable Hsp90α or clusterin overexpression were selected using blasticidin. The cells expressing protein constructs were washed with PBS three times, and cultured in serum-free DMEM for 24 h. The conditioned medium (CM) was collected and concentrated by centrifugation (Millipore, Billerica, MA) at 3800 g, followed by Ni-NTA purification system (Qiagen, Hilden, Germany). His-tagged proteins were dialyzed against PBS. Protein purity was over 90% as assessed by Coomassie Blue staining of SDS-PAGE.

For siRNA transient transfection, cells at 40–50% confluence were transfected with siRNA using Lipofectamine 2000 (Invitrogen) according to the standard procedure. Cells were harvested 48 h after transfection. Scramble siRNA (siCtrl) was used as the negative control. All siRNA sequences are listed in Table S3, and purchased from GenePharma (Suzhou, China).

The mini-LRP1 plasmid was constructed by Ruibiotech (Beijing, China) as previously described (Obermoeller-McCormick et al., 2001), which includes a signal peptide (the first five amino acids), a hemagglutinin (HA) tag, the extracellular ligand-binding subdomain II, and the intracellular p85 subunit gene. The plasmid was transiently transfected into cells using Lipofectamine 2000 following the same procedure as siRNA transfection.

Serum samples from breast cancer patients were diluted and allowed to flow through an Albumin/IgG Depletion column (Calbiochem, Darmstadt, Germany). Tumor cell CM was prepared as mentioned above. The His-tagged eHsp90α protein was covalently coupled to Cyanogen Bromide-Activated Matrices (Sigma-Aldrich), and incubated with either albumin/IgG-depleted serum or concentrated CM overnight with constant rotation at 4°C. Samples were washed three times with Tris buffer (50 mM Tris, 150 mM NaCl, 0.5% NP40), resolved by SDS-PAGE and stained with Coomassie Blue. The entire lane was excised from the gel for mass spectrometry analysis as previously described (Daquinag et al., 2007). Each protein score was calculated using Thermo Proteome Discoverer software based on the following formula: protein score=(sum of all cross-correlation factors of 0.8 or above)+(peptide charge×peptide relevance factor), with the peptide relevance factor set to 0.4.

Cells were washed with PBS, and incubated in DMEM containing Hsp90α (5 μg/ml) and clusterin (5 μg/ml) for 15 min and washed with PBS. Protein–protein interactions were stabilized by incubation with 3 mg/ml dimethyl 3,3-dithiobispropionimidate (DTBP, Thermo Fisher Scientific, Waltham, MA) crosslinker for 2 h at 4°C. All cells were collected and centrifuged at 700 g for 5 min, and lysed with lysis buffer (20 mM Tris, 150 mM NaCl, 0.5% NP40, pH 7.4) supplemented with phosphatase and protease inhibitors (Roche, Basel, Switzerland) at 4°C for 20 min. The supernatants were collected by centrifuging at 16,200 g for 10 min and incubated with indicated antibodies and protein A-sepharose beads (Roche, Basel, Switzerland) overnight at 4°C with constant rotation. The protein A-sepharose beads were washed with lysis buffer and eluted in sample buffer (1% SDS, 1 mM dithiothreitol) by boiling at 100°C for 10 min, followed by western blot analysis. Similar procedures were used for co-IP of proteins in the CM.

Samples from whole-cell lysate or sepharose beads were subjected to SDS-PAGE. Proteins were transferred onto a polyvinylidenedifluoride membrane (Millipore) and incubated with primary antibodies overnight at 4°C, followed by incubation with HRP-conjugated secondary antibodies for 1 h at room temperature. The antibody-labeled proteins were detected using either Western Blotting Luminol Reagent (Santa Cruz Biotechnology, Dallas, TX) or SuperSignal West Femto Chemiluminescent Substrate Kit (Thermo Fisher Scientific). Specifically, proteins that have low basal expression levels in certain cells were detected using the SuperSignal West Femto Chemiluminescent Substrate Kit and cell lysate samples were loaded at 40 μg per well. Detailed descriptions for these proteins are: N-cadherin (dilution, 1:1000; exposure time, 5 min), Snail (dilution, 1:1000; exposure time, 2 min), Slug (dilution, 1:1000; exposure time, 5 min) and Zeb1 (dilution, 1:1000; exposure time, 2 min) in MCF-7 cells, as well as E-cadherin (dilution, 1:1000; exposure time, 10 min), N-cadherin (dilution, 1:1000; exposure time, 2 min) in MDA-MB-231 cells. Other proteins were detected using Western Blotting Luminol Reagent at appropriate exposure time (from 10 s to 5 min), and cell lysate samples were loaded at 20 μg per well.

Total RNA was extracted using TRIzol Reagent (Invitrogen), and converted to cDNAs using Revert Aid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific) according to the manufacturer's instructions. qRT-PCR was performed using the SYBR Green qRT-PCR Master Mix Kit (Stratagene, San Diego, CA). The results were normalized to glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and the control group. All primers were synthesized by Invitrogen, and primer sequences are listed in Table S4.

MCF-7 and MDA-MB-231 cells were first cells were first cultured in medium with Hsp90α (5 μg/ml, purified from CM), clusterin (5 μg/ml, purified from CM) or the combination of Hsp90α and clusterin for 24 h, then trypsinized and seeded on glass coverslips and cultured with indicated proteins for 24 h. Cells were fixed with 4% paraformaldehyde for 30 min, blocked with 2% goat serum for 1 h, incubated with primary antibodies overnight at 4°C, followed by incubation with TRITC- or FITC-conjugated secondary antibodies for 1 h at room temperature. Nuclei were stained with 4,6-diamidino-2-phenylindole (DAPI) for 2 min at room temperature. Images (12 fields from three independent wells for each condition) were captured and analyzed using NIS-Elements analysis software (Nikon, Tokyo, Japan). For the co-localization experiments, cells were incubated with FITC-conjugated Hsp90α, with or without clusterin, for 15 min, and fixed with 3% paraformaldehyde for 20 min, then subjected to the same procedures as described above.

MCF-7 and MDA-MB-231 cells were seeded on glass coverslips and cultured overnight. Cells were incubated in serum-free DMEM with Hsp90α (5 μg/ml, purified from CM), clusterin (5 μg/ml, purified from CM) or the combination of Hsp90α and clusterin for 15 min, fixed with 3% paraformaldehyde for 20 min, blocked with the blocking solution supplied by the Duolink in situ PLA kit (Sigma-Aldrich), and incubated with anti-Hsp90α and anti-LRP1 antibodies overnight at 4°C. Subsequent procedures were completed following the manufacturer's instructions. Nuclei were stained with DAPI for 2 min at room temperature. Images (12 fields from three independent wells for each condition) were captured and analyzed using NIS-Elements analysis software (Nikon).

Cells were seeded into 96-well plates (2000 cells per well) and cultured overnight in DMEM containing 10% serum. Medium was replaced with DMEM containing 1% serum and purified Hsp90α at different dosages (0 μg/ml, 5 μg/ml, 10 μg/ml, 15 μg/ml) for 24 h. Cell viability was evaluated using Cell Counting Kit-8 (Dojindo, Tokyo, Japan) according to the manufacturer's instructions.

The transwell migration assay was performed using Millicell inserts (8 μm pore size, Millipore). 5×10⁴ MDA-MB-231 cells or 1×10⁵ MCF-7 cells were inoculated into the upper chamber. Both the upper chamber and the lower chamber were supplemented with DMEM containing 1% FBS and indicated treatment. Cells were allowed to migrate for 24 h. All the inserts were fixed with 4% paraformaldehyde and stained with Crystal Violet. Cells at the lower surface of the filter membranes were counted in six random fields per insert. Three independent inserts were used for each condition. All experiments were performed at least three times.

All animal studies were approved by the Institutional Animal Care and Use Committee of Tsinghua University. 1×10⁶ GFP-MCF-7 cells or 2×10⁶ GFP-MDA-MB-231 cells mixed with Matrigel (BD Biosciences, Bedford, MA) were subcutaneously inoculated into the fat pad of six-week-old female nude mice. Mice were randomly divided into different groups (five mice per group), Hsp90α (1 mg/kg) and clusterin (1 mg/kg), neutralizing antibodies against Hsp90α and clusterin (1 mg/kg) were administrated intravenously twice a week. Tumor volumes were measured every four days, and calculated as volume=0.5×length×(width)². All mice were euthanized when the largest tumor volume reached 1500 mm³ or the longest tumor length reached 15 mm. The tumors and the sentinel lymph nodes were isolated and weighed. The lymph nodes were frozen (face up embedding), cut into 6-μm-thick sections with a cryostat (Leica CM1950), fixed with acetone and stained with DAPI. GFP-labeled tumor cells in sentinel lymph nodes were detected using confocal microscopy. Images of each lymph node were randomly captured for six fields and the metastasis levels were measured by the ratio of GFP area to DAPI area. The tumor tissues were fixed with 4% paraformaldehyde and embedded in paraffin. Tissue sections were subjected to immunohistochemistry assay (IHC) using antibodies against E-cadherin, Snail, Slug, or Zeb1. Each tumor section was randomly captured for three fields and the expression levels of each protein were quantified using Image-Pro Plus software (Media Cybernetics). For liver metastasis experiments, primary tumors were surgically resected at 30 days after tumor inoculation, and liver metastasis was detected at 60 days after tumor inoculation. Liver tissues were isolated and fixed with 4% paraformaldehyde. Tissue sections were stained with H&E, and quantified by measuring the ratio of tumor metastasis area to total liver area.

All data are presented as mean±s.d. or mean±s.e.m. Statistical analyses for multiple group comparisons were performed using one-way or two-way ANOVA, followed by Bonferroni's post-hoc comparisons tests or two-tailed, unpaired Student's t-tests when appropriate, and were performed using GraphPad Prism software. P<0.05 was considered to be significantly different.

We thank all members in the Luo laboratory for their helpful discussions and technical assistance.