Nm23-H1 Binds to Gelsolin and Inactivates Its Actin-Severing Capacity to Promote Tumor Cell Motility and Metastasis

Metastatic disease and the complications of its treatment are the major causes of death in patients with breast cancer (1). Metastasis suppressor genes have provided novel insights into the molecular control of metastasis. Upon overexpression, metastasis suppressors have no effect on primary tumor size, but significantly reduce metastasis formation (2).

Nm23 was the first metastasis suppressor gene to be identified (3). To date, 10 different Nm23 genes (Nm23-H1 to Nm23-H10) have been identified in humans; the -H1 and -H2 forms are the most studied (4). In vitro, a hallmark of Nm23 is its suppression of motility to a variety of chemoattractants. Transfection of Nm23 genes significantly inhibited metastasis in multiple model systems (5); where studied, primary tumor size was unaffected. In leukemias, lymphomas, and neuroblastoma, opposite trends were observed. Identifying the mechanism(s) of Nm23-H1 suppression of metastasis will be key to developing effective metastasis-preventive strategies for this pathway, but it remains incomplete. Potential contributors to Nm23-H1's suppression of metastasis include its histidine kinase activity, nucleotide diphosphate kinase activity, and 3′-5′-exonuclease activity (6), as well as protein–protein interactions (reviewed in ref. 7). The analysis of protein–protein interactions involving Nm23-H1 has been hampered by nonspecific associations, and has identified few interactions of obvious importance to the hallmark motility phenotype.

Here, we have taken an unbiased approach to identify potential binding partners for the human (-H1) and murine (-M1) forms of Nm23 in cultures, primary tumors, and metastases formed by murine 4T1 mammary carcinoma cells. We report that Nm23-H1 and -M1 bind to Gelsolin, the founding member of a family of actin-binding proteins involved in the remodeling of cellular actin filaments, regulating cell shape changes and movement (8). Structurally Gelsolin is composed of a Ca⁺⁺-dependent regulatory C-terminal fragment, an actin-severing N-terminal fragment, and a 70–amino acid connector containing the consensus for caspase-3–mediated cleavage (9). After Ca⁺⁺-dependent activation, Gelsolin binds and severs actin filaments and caps their fast-growing ends. Removal of Gelsolin from the actin filaments (uncapping) is mediated by phospholipids such as phosphatidyl inostil 4,5 bisphosphate (PIP2; ref. 10) and lysophosphatidic acid (LPA; ref. 11) permitting actin polymerization. Gelsolin null mice were viable (12); in dermal fibroblasts from null mice, reduced membrane lamellipodia and motility were observed, with enhanced Rac expression and increased F-actin in stress fibers (13, 14). Gelsolin also participates in lipid signaling (15) and the apoptotic process (16, 17) and various protein–protein interactions (18–21).

The correlation between Gelsolin expression and tumorigenesis is still controversial (22–25). To date, the role of Gelsolin expression in metastasis is incompletely described. In vitro studies report that Gelsolin overexpression promoted tumor cell motility and invasion through modulation of several pathways, including EGF receptor (EGFR), phosphoinositide 3-kinase (PI3K), and Ras–PI3K–Rac (14, 26, 27). In vivo, Gelsolin suppressed the epithelial–mesenchymal transition in mammary epithelial cells (28) and acted as a metastasis suppressor in B16 melanoma cells (29).

Herein, we show that Nm23-H1 binds to Gelsolin in multiple breast cancer cell lines, and abrogates Gelsolin's actin-depolymerization activity in vitro. In contrast with the limited published literature, Gelsolin overexpression in murine 4T1 and human MDA-MB-231T model systems had a significant effect on primary tumor formation and stimulated metastasis to the lungs and liver. Co-overexpression of Gelsolin and Nm23-H1 reduced Gelsolin's promotion of motility in vitro and metastasis in vivo.

Human breast cancer cell lines MDA-MB-435 and MCF7, and immortalized kidney cell line HEK293TN were obtained from American Type Culture Collection. A subline of human MDA-MB-231 cells, designated MDA-MB-231T cells, was used after authentication by the Core Genotyping Facility (National Cancer Institute, Bethesda, MD; ref. 30). Murine mammary adenocarcinoma cell line 4T1 expressing luciferase was obtained from Dr. Gary Sahagian (Tufts University, Medford, MA). The cells were cultured in Dulbecco's Modified Eagle Medium (DMEM; Invitrogen) supplemented with 10% FBS, and antibiotics (100 U/mL penicillin and 100 μg/mL streptomycin), under a humidified 37°C incubator at 5% CO₂. To reduce the presence of actin filaments in cell lysates, cells were incubated for 3 hours in the presence of 0.33 μg/mL latrunculin B (Millipore).

Nm23-H1 or Nm23-M1 cDNAs with an attached Flag tag were inserted into the BamH1 site in the Multiple Cloning Site of pcDNA3.1. The 4T1 cells were transfected with empty vector or plasmid containing Nm23-H1 or -M1 using Lipofectamine (Invitrogen) according to the manufacturer's recommended protocols. DMEM containing 600 μg/mL of G418 (Invitrogen) was used to select for clonal lines.

Gelsolin full-length or C- and N-terminus fragments (amino acids 1–360 and 361–742, respectively) cDNAs were amplified from pcDNA-Gelsolin, obtained from Dr. Shigeomi Shimizu (Medical Research Institute Tokyo, Tokyo, Japan). The cDNAs were first cloned into pENTR/D-TOPO vector and then inserted into an N-terminal GFP fusion vector, pcDNA 6.2/N-EmGFP using LR recombination reaction according to the manufacturer's recommendations (Gateway vector system; Invitrogen). Plasmids were transfected into cells using Effectene (Bio-Rad). Cells were subsequently cultured in complete DMEM with blasticidin (5 μg/mL) for selection.

MISSION The RNAi Consortium short hairpin RNA (shRNA) constructs were purchased from Sigma-Aldrich and included nontarget control shRNA (SHC216), shRNA to Gelsolin (TRCN0000029727), shRNA to Nm23-H1 (TRCN0000010061). Cell infection procedure was followed according to the manufacturer's protocol.

Lysates were prepared by washing cells twice in PBS, followed by incubation for 30 minutes on ice with a gentle lysis buffer [20 mmol/L Tris pH 8.0, 137 mmol/L NaCl, 10% glycerol, 0.2% Triton X-100, 1 mmol/L phenylmethylsulfonylfluoride (PMSF), 20 μmol/L leupeptin, 0.15 U/mL aprotinin, and 5 mmol/L sodium vanadate]. Frozen tumor samples were crushed before incubation in lysis buffer. Lysates were centrifuged at 13,000 × g for 15 minutes. Protein concentration of the supernatants was measured using a bicinchoninic acid (BCA) protein assay (Thermo Fisher Scientific).

Protein G Sepharose beads (GE Healthcare) were incubated with 10 μg of anti-Flag antibody (Sigma-Aldrich) for 1 hour at 4°C. After cross-linking the antibody to beads, they were washed with lysis buffer. The beads were incubated with 1 mg of protein suspended in lysis buffer on a rotating mixer overnight at 4°C. The beads were then washed three times with lysis buffer and proteins were eluted by boiling with NuPage loading buffer and sample reducing agent (Invitrogen). The supernatant was collected and separated using SDS-PAGE.

Gel bands were excised from the gel and digested with trypsin. The resultant peptides were extracted and analyzed using reversed-phase liquid chromatography (Agilent 1200; Agilent Technologies) coupled online with a LTQ XP mass spectrometer (Thermo Fisher Scientific). The seven most intense molecular ions in the mass spectrometry scan were sequentially selected for collision-induced dissociation using normalized collision energy of 35%. The tandem mass spectrometry (MS-MS) data were searched against UniProt Mus Musculus database from the European Bioinformatics Institute using BioWorks interfaced SEQUEST (Thermo Fisher Scientific).

Lysates were prepared from cells in culture by washing them twice in PBS, followed by incubation for 30 minutes on ice with NP-40 lysis buffer (20 mmol/L Tris–HCl pH 8.0, 100 mmol/L NaCl, 10% glycerol, 1% Nonidet P40, and 2 mmol/L EDTA). Lysates were centrifuged at 13,000 × g for 15 minutes. Protein concentration of the supernatants was measured using BCA assay. Dynabeads Protein-A/-G (Invitrogen) were incubated 10 minutes with primary antibody and 30 minutes with 1 mg of protein lysates, suspended in lysis buffer on a rotating mixer at room temperature. The beads were washed three times with PBS and proteins were eluted by boiling with NuPage loading buffer and sample reducing agent (Invitrogen). The supernatant was collected and separated using SDS-PAGE and processed as a Western blot analysis.

Primary antibodies used for either immunoprecipitation or Western blot analyses are the following: mouse anti-Flag (Sigma-Aldrich), rabbit anti-Flag (Cell Signaling Technology), rabbit anti-GFP (Cell Signaling Technology), mouse anti-GFP (Abcam), rabbit anti-Gelsolin (Abcam), mouse anti-Gelsolin (Sigma-Aldrich), mouse anti-Nm23-H1 (BD Biosciences), rabbit anti-Nm23-H1 (Santa Cruz Biotechnology), rabbit anti-PARP (Cell Signaling Technology), rabbit-anti-caspase-3 (Cell Signaling Technology), and rabbit anti-cleaved caspase-3 (Cell Signaling Technology). Mouse anti-α-tubulin (Sigma-Aldrich) was used as a loading control in Western blot analyses.

Transwell migration assays were conducted in Boyden chambers as described previously (31). The reported results represent the average of triplicate experiments.

The Gelsolin-severing activity was measured using the Pyrene–Actin Polymerization Kit (Cytoskeleton) following the manufacturer's protocol. Briefly, 1 mg/mL purified rabbit muscle pyrene–actin (Cytoskeleton) was diluted to 30 μg/mL and resuspended in polymerization buffer (50 mmol/L KCl, 2 mmol/L MgCl₂, 0.5 mmol/L ATP, and 2 mmol/L Tris, pH 8.0), incubated for 1 hour to form actin polymers. Cell lysates were prepared and 10 μL of equal amount of total protein was added to the final reaction volume (200 μL). The rate of florescence loss at 590 nm (530-nm excitation wavelength) was measured by fluorometry (VersaMax, Microplate Fluorescence Reader) using Softmax Pro 5.4 software. To measure the actin-polymerization activity, pyrene–actin was diluted to 2.3 μmol/L in the general actin buffer (5 mmol/L Tris–HCl pH 8.0 and 0.2 mmol/L CaCl₂) containing 0.2 mmol/L ATP and 0.5 mmol/L Dithiothreitol and stored on ice for 60 minutes to depolymerize actin oligomers. The tubes were centrifuged at 14,000 × g force for 30 minutes at 4°C. The supernatants containing pyrene-actin monomers were mixed with lysates, added into the 96-well plates, and held for 3 minutes. After the addition of 1/10 volume of 10× of polymerization buffer (500 mmol/L KCl and 10 mmol/L MgCl₂), the kinetics of pyrene fluorescence upon initiation of actin polymerization was monitored for 2 hours at 405 nm with an excitation wavelength of 355 nm using Microplate Fluorescence Reader.

Approximately, 1 × 10⁴ cells were plated in each chamber of 4-well chamber slides (BD Biosciences) and grown for 24 hours. The cells were fixed with buffered formalin, preblocked in blocking buffer (PBS containing 2% FBS, 2% goat serum, and 0.2% Triton X-100), and incubated with the following primary antibodies diluted in blocking buffer overnight: mouse anti-Flag (Sigma-Aldrich), rabbit anti-Flag (Cell Signaling Technology), rabbit anti-Nm23-H1 (Santa Cruz Biotechnology), rabbit anti-Gelsolin (Abcam), and mouse anti-Gelsolin (Sigma-Aldrich). Thereafter, cells were incubated with the following secondary antibodies diluted in blocking buffer: goat anti-mouse and goat anti-rabbit immunoglobulin G (IgG) antibodies conjugated to Alexa Fluor-488 and -546 (Molecular Probes), diluted at 1:500. Nuclei and F-Actin were visualized by the addition of 2 μg/mL of 4′,6-diamidino-2-phenylindole (DAPI; dilactate; Molecular Probes) and 5 units/mL of rhodamine–phalloidin (Invitrogen), respectively, to the secondary antibody mixture. After washing four times with PBS, sections were mounted in Fluorescent Mounting Medium (DakoCytomation), and photographed using either Zeiss AxioSkope 2 microscope or confocal microscope Zeiss 710UV and ZEN2009 software.

Experiments were carried out under an approved National Cancer Institute Animal Use Agreement. Six-week-old Balb/c and athymic nude females were obtained from Charles River Laboratories (National Cancer Institute-Frederick Animal Production Area). The spontaneous metastasis assay was conducted with 4T1 cells at a concentration of 5 × 10⁵ cells/50 μL, which were injected under anesthetic into the #4 mammary fat pad (mfp) of Balb/c mice. Primary tumors were measured by caliper every third day. On day 18 postinjection, primary tumors were resected under isoflurane anesthesia. On week 10 postinjection, mice were sacrificed and all relevant tumor and organ materials were fixed and embedded in optimal cutting temperature (OCT) medium. Ten mice per experimental arm were used. Mice were weighed weekly and monitored for signs of morbidity and labored breathing.

The experimental pulmonary metastasis study was conducted by injecting 5 × 10⁵ MDA-MB-231T cells into the lateral tail vein of athymic nude mice. Nine weeks postinjection, at necropsy, the lungs were collected and fixed in Bouins' solution. Surface metastatic lesions were counted on all lungs before embedding in paraffin, reported as a median for each group.

OCT-embedded lungs and livers of mice from the spontaneous metastasis 4T1 model were sectioned (7 μm) on a cryostat.

Lungs collected from the experimental metastasis MDA-MB-231T model were paraffin-embedded and sectioned (10 μm).

Hematoxylin and eosin (H&E) staining was conducted to count the metastatic lesions in two sections from each mouse, 200 μm apart. To measure proliferation rate in metastasis, lungs from both models were stained with rabbit anti-Ki67 antibody, following the manufacturer's protocol (Vector Laboratories).

Either ANOVA or Kruskal–Wallis test was conducted on the raw data. Statistical significance of difference between each sample and vector control arm was assessed using either Student t test or Mann–Whitney test. All statistical tests were two-sided and P values were considered statistically significant for P < 0.05. GraphPad Prism 4 software was used to analyze the data.

To identify the Nm23 coimmunoprecipitating proteins, we used murine 4T1 mammary carcinoma cells, previously transfected with a vector, nm23-H1 (human isoform 1) or nm23-M1 (murine), the latter two fused to a C-terminal Flag epitope (32). In agreement with other Nm23 transfections (33), overexpression of either human or murine Nm23 reduced in vitro motility in Boyden chambers by 57% (P = 0.0002) and 85% (P < 0.0001), respectively, compared with vector (Fig. 1A). Implantation of the vector- or Nm23-transfected 4T1 cells into the mfp of syngeneic mice resulted in comparable primary tumor sizes 10 days postinjection (32). When primary tumors were surgically removed and mice were kept until week 10 postinjection, Nm23 overexpression reduced the number of quantifiable metastases to the liver by 60% to 70% and lungs by 84% to 95% (32). Thus, in the 4T1 model system, Nm23-H1 or -M1 overexpression exhibited typical characteristics of a motility- and metastasis suppressor gene.

To identify the Nm23-binding proteins, Flag-tagged Nm23-H1 and -M1 proteins were immunoprecipitated from lysates of transfected 4T1 cells in vitro, from mfp primary tumors, and from grossly dissected lymph node-, spleen-, and liver metastases. Electrophoresed coimmunoprecipitating proteins (Fig. 1B) were eluted and identified using mass spectrometry. Table 1 lists selected potential coimmunoprecipitating proteins, noting the number of peptides sequenced from each cell line or organ preparation. This list included several proteins previously documented to bind Nm23, confirming the procedure used. Coimmunoprecipitation of Nm23 and the identified proteins was limited to tissue culture and primary tumors, and was largely lost in actual metastases. Notable candidates included several cytoskeletal proteins, which might underlie Nm23 function in tumor motility, invasion, and metastasis.

One of the candidate proteins resulting from this analysis was the actin-binding protein Gelsolin. Notably, 52 and 75 peptides of Gelsolin were sequenced from anti-Flag–Nm23-H1 and -M1 immunoprecipitated primary tumor lysates, respectively. Binding was reduced or lost in the lymph node, spleen, and liver metastases.

Coimmunoprecipitation assays were used to verify the interaction between Nm23 and Gelsolin in the murine mammary carcinoma cell line 4T1 overexpressing Flag-tagged Nm23-H1 and -M1 (Fig. 2A, left). Either anti-Flag or anti-Gelsolin antibodies were used to pull down the target protein from the cell lysates (Fig. 2A, right). Coimmunoprecipitation of Gelsolin and Nm23 was observed in both Nm23-H1 and M1 samples. Coimmunofluorescence of Nm23 and Gelsolin was conducted in 4T1 cells overexpressing Nm23-M1 along the edge of a scratch motility assay (Fig. 2B). Although Gelsolin was expressed in both cell edges and cytoplasm along the actin cytoskeleton structure, colocalization of the two proteins was observed exclusively in the cytoplasmatic compartment of the cells. Because Gelsolin is an actin-binding protein, the interaction of Nm23–Gelsolin may be an indirect result of Nm23 binding to the actin filaments. Treatment of the 4T1 cells with latrunculin B, an actin-depolymerization agent (34), destroyed the F-actin cytoskeleton but did not eliminate the Nm23–Gelsolin coimunoprecipitation (Fig. 2C and Supplementary Fig. S1).

To validate the Nm23-H1–Gelsolin interaction observed in the mammary cell line 4T1, a second cellular model, the human breast carcinoma cell line MDA-MB-231T, was used. Empty vector and Flag-tagged Nm23-H1 constructs were transfected in MDA-MB-231T cells (Fig. 2D, left). Anti-Nm23 and anti-Gelsolin antibodies were used to pull down the endogenous proteins together with exogenous Flag-tagged Nm23-H1 (Fig. 2D, right). The Nm23-H1–Gelsolin complex was detected only in Nm23-H1–overexpressing MDA-MB-231T cells, which was consistent with our inability to detect even low levels of Nm23-H1 immunoprecipitation in lysates in the vector-expressing controls. Immunoflorescence confirmed the colocalization of the two proteins in the perinuclear compartment (Fig. 2E), and the complex remained after destruction of the actin cytoskeleton by latrunculin B (Fig. 2F and Supplementary Fig. S1). The data confirm an interaction of transfected Nm23-H1 and Gelsolin in two breast carcinoma cell lines. When Gelsolin was fragmented into N- and C-terminal subunits, Nm23-H1 coimmunoprecipitated with the C-terminal fragment (Supplementary Fig. S2). The coimmunoprecipitation results were validated in other two cell lines, MDA-MB-435 and HEK293TN, overexpressing the empty vector and Nm23-H1 or both Nm23-H1 and Gelsolin, respectively (Supplementary Fig. S3).

To evaluate the interaction of Nm23-H1 and Gelsolin at endogenous levels, we used the MCF7 human breast carcinoma cell line. This cell line contains a higher Nm23 level compared with MDA-MB-231T cells (Fig. 2G, left). Both the immunoprecipitation and immunoflorescence assays showed that Nm23 and Gelsolin interacted in an F-actin–independent manner and colocalized in the cytoplasmic compartment (Fig. 2H and I and Supplementary Fig. S1).

Gelsolin is a potent F-actin–severing protein that, after filament cleavage, remains bound to the newly formed end (capping). In this way, Gelsolin facilitates depolymerization by blocking actin monomer addition at the growing end, whereas actin disassembly proceeds at the opposite end of the filament (35). Gelsolin also has two contrasting functions in apoptosis: (i) it has a proapoptotic role after cleavage by caspases, leading to uncontrolled F-actin severing; (ii) it can also prevent apoptosis by stabilizing mitochondria (9, 17). We hypothesized that Gelsolin's interaction with Nm23 would inhibit its actin-depolymerization function, given Nm23's profound effects on motility. Clonal cell lines overexpressing either a vector or Gelsolin, the latter with a GFP tag at its N-terminus, and either a vector or Nm23 (-H1 and -M1)-Flag were created in both the 4T1 and MDA-MB-231T cell lines. The GFP tag did not interfere with Gelsolin interaction with Nm23 (Supplementary Fig. S4). To evaluate whether Nm23 overexpression and interaction could affect actin cytoskeleton remodeling by Gelsolin, actin polymerization and depolymerization rates were measured using an in vitro, fluorescence-based assay for F-actin formation or severing activity. In 4T1 cells, transfection of either Gelsolin or Nm23-H1 had no effect on actin polymerization (Fig. 3A), in keeping with the lack of reported Gelsolin effect on actin polymerization (36). However, a decrease in fluorescent signal, corresponding to an increase in F-actin severing, was observed in the lysate from Gelsolin-overexpressing cells when the actin-depolymerization assay was conducted (Fig. 3B), consistent with the reported literature (35). In contrast, no florescence decay was observed in the reaction with lysates containing 4T1 cells overexpressing Nm23 alone or together with Gelsolin. Phalloidin staining was used to analyze the actin cytoskeleton structure (Fig. 3C). Severing activity associated with Gelsolin expression in 4T1 cells was visualized by the decreased number of actin filaments (stress fibers) and appearance of short filaments or actin monomer accumulation (red dots). Nm23 overexpression together with Gelsolin reversed this phenotype as the actin cytoskeleton appeared similar to that in the vector cells. Data were validated using lysates from MDA-MB-231T cells (Fig. 3D–F). Similar results were obtained when actin-severing activity was measured in protein lysates from a scratch-wounded confluent layer of either 4T1 or MDA-MB-231T cells (Supplementary Fig. S5). Although Nm23 showed only minor effect on reducing actin severing as compared with the control, its co-overexpression with Gelsolin resulted in a significant inhibition of Gelsolin actin-severing activity (39% in 4T1, P = 0.04; 47% in MDA-MB-231T, P = 0.05). These results indicate that Nm23 overexpression and interaction with Gelsolin prevented its actin-severing activity and modified stress fiber formation. Moreover, when endogenous Nm23-H1 expression was knocked down in MCF7 cells, actin severing increased 5-fold as compared with the control cells (Fig. 3G), thus confirming its involvement in this process.

Because actin cytoskeleton reorganization is important for cell shape and motility, Gelsolin has crucial roles in the control of these cellular functions. The effects of Gelsolin on tumor cell motility are complex (29, 37). We observed distinct patterns of in vitro motility upon Gelsolin overexpression. Gelsolin overexpression promoted cell migration in presence of growth factors such as EGF (27) and insulin (26), whereas no effect was observed when other chemoattractants such as LPA (38) and 1% FBS were used (Supplementary Fig. S6A and S6B). Motility to EGF was selected for further experiments, using 4T1 and MDA-MB-231T cells overexpressing Nm23, Gelsolin, or both proteins (Fig. 4). Overexpression of Gelsolin in 4T1 cells enhanced cell migration by 58% (P = 0.03), whereas Nm23 reduced it by 46% (P = 0.02), as compared with vector-transfected cells. Cells overexpressing both proteins showed 50% reduction in motility compared the Gelsolin-expressing cells (P = 0.03; Fig. 4A). MDA-MB-231T cells expressing Nm23, Gelsolin, or both proteins were also assayed. Cell migration was increased by 64% after Gelsolin overexpression, whereas it was reduced by 37% after Nm23-H1 overexpression as compared with the vector (P = 0.005 and 0.0009, respectively). The co-overexpression of Nm23-H1–Gelsolin reversed the Gelsolin phenotype, reducing motility by 53% compared with Gelsolin-expressing cells (P = 0.02; Fig. 4B). Thus, Nm23 inhibits the motility-promoting effects of Gelsolin, consistent with its abrogation of Gelsolin's actin-severing ability. In addition, knockdown of endogenous Nm23-H1 from MCF7 cells resulted in 4.8-fold increase in motility (P = 0.04; Supplementary Fig. S6C). Knockdown of Gelsolin in the vector-expressing MDA-MB-231T cells showed a 37% decreased motility (P = 0.03); it had no effect on motility in the Nm23-H1–expressing cells (Fig. 4C).

Other in vitro assays were conducted. No significant effect of Nm23 or Gelsolin overexpression on proliferation of 4T1 or MDA-MB-231T cells was observed in MTT assays (Fig. 4D and E). No effect of the Nm23 or Gelsolin overexpression on apoptosis was observed (Supplementary Fig. S7).

To evaluate the in vivo effects of Nm23-H1 and Gelsolin overexpression, 4T1 cells overexpressing either a vector or Gelsolin, and either a vector or Nm23-H1 were implanted into the mfp of syngeneic Balb/c mice. Differences in tumor growth between the cells were observed (Fig. 5A). In accord with its tumorigenic property in breast cancer (22, 23), Gelsolin-overexpressing 4T1 cells showed increased tumor growth, which was reversed when Nm23-H1 was co-overexpressed. Primary tumors were removed surgically on day 18 postinjection and the mice were sacrificed on day 69 postinjection to quantify metastasis (Fig. 5B–D). In the liver, two histologic forms of metastatic colonization were observed. In some mice, discreet metastatic lesions were quantifiable, whereas other mice contained florid metastatic disease throughout the liver (Fig. 5B); these two patterns of metastasis were quantified separately as previously reported (32). In the liver, Gelsolin overexpression increased distinct metastatic lesions by 24% (P = 0.187); diffuse liver metastases were observed in 60% (6 of 10 mice) of mice as compared with none in the vector transfectants. Nm23-H1 overexpression reduced distinct metastasis formation by 48% (P = 0.0002) and no mice had diffuse metastases. Co-overexpression of Gelsolin and Nm23-H1 reduced the percentage of mice with diffuse liver metastases to 20% (2 of 10 mice).

Data in the lungs showed similar trends (Fig. 5D). Gelsolin overexpression increased metastasis formation by 107% (median number of metastases 39 to 81; P = 0.001) as compared with the vector transfectants. Nm23-H1 overexpression reduced metastasis formation by 36% (median number of metastases 39 to 25; P = 0.0015). Co-overexpression of Gelsolin and Nm23-H1 reduced lung metastasis formation to levels significantly different from the Gelsolin transfectant (median number of metastases 81 to 34; P = 0.01). The data indicate a prometastatic function of Gelsolin in the 4T1 model system that is inhibited by Nm23-H1 overexpression.

To evaluate potential influence of proliferation on metastasis formation, Ki67 staining was conducted on pulmonary metastatic lesions. No difference in proliferation was observed between the experimental groups (Fig. 5E).

To exclude any influence on metastasis formation related to the differences observed in tumor growth, an experimental metastatic lung assay was conducted. To confirm an inhibitory effect of Nm23 on Gelsolin's prometastatic activity, MDA-MB-231T cells overexpressing either Nm23, Gelsolin, or both proteins were injected via the tail vein to produce lung metastases. Representative pictures of the lungs collected from mice after 9 weeks from the cell injection are shown in Fig. 6A, left. The median number of surface lung metastases/mouse was 20 in the vector arm and increased to 48 in the Gelsolin arm, confirming Gelsolin's metastasis promoting activity in two model systems. Surface experimental metastases declined 55% to 9 in Nm23-H1–overexpressing arm. Co-overexpression of Nm23-H1 and Gelsolin produced an average of 12 surface lung metastases, not statistically different from the vector transfectant (P = 0.83; Fig. 6A, right).

After paraffin-embedding and sectioning the lungs, H&E staining was conducted to visualize and count histologic metastases. This quantification of metastatic burden confirmed the surface metastasis findings: Gelsolin expression induced a 53.2% increase in metastasis formation (P = 0.041), whereas Nm23-H1 reduced the number of lung metastases by 74.1% (P = 0.035) as compared with the vector (Fig. 6B). Co-overexpression of Nm23-H1 and Gelsolin significantly reduced metastases by 75% as compared with Gelsolin arm (P = 0.035).

No difference in proliferation was observed in metastatic lesions between the experimental groups (Fig. 6C).

Although Nm23-H1 has been thoroughly validated as a metastasis suppressor in solid tumors, its mechanism of action remains elusive (5). Protein–protein interactions involving Nm23 have been widely reported (7), but few have been independently confirmed and, likewise, few modulate Nm23's hallmark phenotype of tumor cell motility. We conducted an unbiased search for Nm23-H1- and -M1–binding proteins using MS-MS sequencing of coimmunoprecipitating proteins from 4T1 lysates, primary tumors, and metastases. Of the Nm23-binding proteins identified from primary tumors, several are intimately connected to cellular motility, including Ezrin, Arp2/3, and Gelsolin. The interaction of Nm23-H1 and each of these proteins was undetectable in lymph nodes, liver, and spleen metastases from the same model system, which may reflect the downregulation of Nm23-H1 expression in metastatic lesions (39), leading to the loss of an association and unhindered function of the binding protein. Two-way coimmunoprecipitation experiments using transfected Nm23-H1 in 4T1 cells and human MDA-MB-231T breast carcinoma cells, MDA-MB-435 carcinoma cells and HEK293TN immortalized kidney cells confirmed the association of Nm23 and Gelsolin. Coimmunoprecipitation of the two proteins at endogenous levels was observed in human MCF7 breast carcinoma cells, which express relatively high levels of Nm23-H1. It remains possible that additional proteins are part of the Nm23-H1–Gelsolin complex; the prometastatic protein Prune has been reported to be a binding partner of each protein, for instance (40, 41).

Gelsolin is traditionally understood as an actin-binding protein. It severs and caps ends of actin filaments in a Ca⁺⁺–regulated manner, providing for dynamic actin assembly and disassembly in normal cell elasticity and in the leading and lagging aspects of cell motility. Less well understood are Gelsolin's protein–protein interactions with p53 (21), hormone receptors (18, 20), and other proteins (19), a lipid regulatory activity (15), and the apoptosis-inducing activity of its N-terminal fragment (16). Using lysates of 4T1 and MDA-MB-231T breast carcinoma cells transfected with either vector, Nm23-H1, Gelsolin, or Nm23-H1/Gelsolin, enzymatic assays for actin polymerization and depolymerization were conducted. Consistent with the literature, Gelsolin overexpression had no effect on actin polymerization, but augmented its depolymerization. It is not known whether Gelsolin directly severed the actin filaments in the assay, or capped filaments severed by other enzymes. In agreement with an interaction between Nm23-H1 and Gelsolin, co-overexpression of both proteins abrogated Gelsolin's depolymerization ability. Furthermore, after knocking down endogenous Nm23-H1 in MCF7 cells, an increase in actin severing was observed, confirming that Nm23-H1 may affect Gelsolin function. The data indicate a functional consequence of the Nm23-H1–Gelsolin interaction. Our coimmunoflourescence data point to an interaction of Gelsolin and Nm23-H1 in the perinuclear cytoplasm, permitting the hypothesis that Nm23-H1 and Gelsolin interact to regulate stress fiber formation, which occurs in this area. In experiments carried out using MDA-MB-231T cells, Nm23-H1 coimmunoprecipitated a transfected C-terminal portion of Gelsolin. This region has a single actin-binding site and regulates Gelsolin's activity (8), whereas its actual actin-severing activity is localized to Gelsolin's N-terminus.

Consistent with Gelsolin's role in actin dynamics, its overexpression has been reported widely to increase tumor cell motility and invasion or, conversely, its knockdown to inhibit motility (14, 26–28). In both the 4T1 and MDA-MB-231T model systems, we confirmed a stimulation of tumor cell motility by Gelsolin overexpression. The stimulatory effect of Gelsolin was abrogated by co-overexpression of Nm23, consistent with their binding and inactivation of actin depolymerization. Furthermore, while knockdown of endogenous Nm23-H1 in MCF7 cells resulted in increased motility, the knockdown of Gelsolin in Nm23-H1-overexpressing MDA-MB-231T cells showed no effect on cell migration, suggesting that the functional interaction is unidirectional in effect, that is, Gelsolin may not affect Nm23 function.

We did not observe any evidence for a proapoptotic role of the transfected Gelsolin, but cannot rule out that other of its functions could also be contributory.

In the face of Gelsolin's role in actin dynamics and tumor cell motility, in vivo data in the literature on the phenotypic consequences of its altered expression are both sparse and puzzling. Three reports showed a suppressive effect of Gelsolin overexpression, either wild-type or a Pro321→His mutation, on primary tumor formation (42–44). A lipid metabolism role of Gelsolin was implicated in its tumor-suppressive phenotype. We observed an increased in primary tumor size upon Gelsolin overexpression (22, 23). For metastasis, only one report has been published and it shows a surprising metastasis suppressive role for Gelsolin overexpression in B16 melanoma cells (29). This report stands in sharp contrast to two model systems reported herein, in which Gelsolin overexpression augmented 4T1 spontaneous metastasis to the lungs and liver, and experimental MDA-MB-231T metastasis to the lungs. The reason(s) for the discrepancy are unclear. Gelsolin overexpression in the B16 model was modest, and it is possible that its relative expression level can alter actin dynamics to produce divergent phenotypes. Other Gelsolin family members exert similar functions (27). In agreement with the coimmunoprecipitation of Gelsolin and Nm23-H1, and Nm23-H1's abrogation of Gelsolin's actin-depolymerization and motility induction properties, co-overexpression of Nm23-H1 reduced the metastasis-stimulatory phenotype of Gelsolin. Taken together, the data permit the hypothesis on a mechanism of action for Nm23-H1 metastasis suppression: an Nm23-H1 interaction with Gelsolin limits actin-depolymerization and dynamic function, thereby attenuating metastasis in breast cancer model systems.

No potential conflicts of interest were disclosed.

Conception and design: N. Marino, J.-C. Marshall, P.S. Steeg

Development of methodology: N. Marino, J.-C. Marshall, T. Veenstra

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Marino, J.-C. Marshall, Y. Qian, T. Veenstra

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Marino, J.-C. Marshall, M. Zhou, T. Veenstra, P.S. Steeg

Writing, review, and/or revision of the manuscript: N. Marino, J.-C. Marshall, T. Veenstra, P.S. Steeg

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Marino, J.W. Collins, M. Zhou

Study supervision: P.S. Steeg

The authors thank Dr. Gary Sahagian (Tufts University) for luciferase-labeled 4T1 cells and Dr. Shigeomi Shimizu (Medical Research Institute Tokyo) for pcDNA–Gelsolin plasmid.

This project has been funded with federal funds from the National Cancer Institute, NIH, under contract number HHSN261200800001E.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.