The LCN2/LOXL2/MMP9 ternary protein complex promotes migration and invasion through the FAK/AKT/GSK3β signaling pathway in esophageal squamous cell carcinoma

During tumor malignant development, the extracellular matrix (ECM) is usually abnormally regulated. Dysregulated expression of lysyl oxidase-like 2 (LOXL2), matrix metalloproteinase 9 (MMP9) and lipocalin 2 (LCN2) are associated with ECM remodeling. Nevertheless, the evidence of an LCN2/LOXL2/MMP9 protein complex and its potential molecular mechanisms remain unclear. extracellular LOXL2: MMP9: LCN2: deferoxamine mesylate; carcinoma; MMPs: Matrix neutrophil gelatinase-associated lipocalin; EMT: epithelial-mesenchymal

Protein extraction and concentration of conditioned medium Cells were seeded in a 10 cm dish (for co-IP) or 6 cm dish (for western blotting), cultured for 36 h following transfection, and then starved for 12 h in serum-free medium. Cells were then lysed in RIPA buffer (9806S, CST) to extract the total protein. Conditioned medium was concentrated using Amicon® Ultra centrifugal lters (UFC5010BK, MILLPORE) by centrifugation at 14,000×g for 10-30 min, to a nal volume of approximately 20 μl for each sample.

Western blotting
Approximately 30 μg of total protein was loaded, separated by 10% SDS-PAGE, and transferred to PVDF membranes (3010040001, Roche BR). The membranes were blocked with 5% skim milk powder (P0101, Maygene) in TBST (E175-01, GenStar) for 1 h and incubated with primary antibodies, then incubated with an HRP-conjugated secondary antibody. The speci c primary antibodies and secondary antibodies used in this study are provided in Supplementary Table 1. Finally, chemiluminescence was performed using SuperSignal TM West Pico PLUS Chemiluminescent Substrate (34579, Thermo) to visualize the protein bands of interest by a Bio-Rad ChemiDoc Imaging System (Bio-Rad, Hercules, CA, USA).
Mass spectrometry analysis LCN2-HA and LOXL2-Flag plasmids were co-transfected as an experimental group, and LOXL2-Flag plasmid was transfected as a control group in KYSE150 cells. The whole cell lysate was coimmunoprecipitated with anti-Flag antibody, separated by SDS-PAGE and stained with Coomassie Brilliant Blue. A differentially-expressed protein band around 25kD was excised and digested with trypsin for mass spectrometry (Q Exactive, Thermo) analysis. The peptides were identi ed by searching against the Uniprot database (https://www.uniprot.org/).

Wound healing assay
A scratch wound assay was carried out to measure the distance of cell migration. After 24 hours of transfection, cells were trypsinized, suspended and adjusted to a cell density of 2.5×10 5 cells/ml, inoculated into a 12-well plate, cultured for 12 hours, and then starved for 12 hours in serum-free medium. A straight-line scratch was made on the monolayer culture, approximately 80-90% con uence, using a sterile 200 µL pipette tip. The wells were photographed under 200X magni cation every 6 h or 12 h with microscope (Olympus, Tokyo, Japan).

Transwell migration and invasion assays
Migration and invasion assays to measure the mobility of cancer cells were performed using transwell chambers as previously described [11]. Cells were cultured 24 hours after transfection, and then starved for 12 hours in serum-free medium. The cells were digested and adjusted to 1.25×10 5 cells/ml (migration assay), 2.5×10 5 cells/ml (invasion assay), then 400 μl cell suspension was added to the upper chamber: uncoated transwells (353097, FALCON) were used for the migration assay, and Matrigel-coated transwells (356234, CORNING) were used for the invasion assay. Afterwards, 500 μl of 10% serum medium was added to the lower chamber. Cells were incubated for 24-36 h (migration assay) or 36-48 h (invasion assay). At the end of the culture, non-migrated cells in the upper chamber were carefully removed with a cotton swab. The migrated cells that passed through the membrane and adhered to the lower surface of the membrane were xed with 4% paraformaldehyde for 15 min, and then stained with crystal violet for 5 min. The number of migrated/invaded cells in 10 random elds was photographed using a 200× microscope (IX73, OLYMPUS, Japan).

MTS assay
Transfected cells were trypsinized, suspended and adjusted to 1× 10 5 cells/ml with serum-containing medium. Then, 100 μl of the cell suspension was added to each well of a 96-well plate (701001, NEST) in triplicate. At 0, 24, 48, 72 and 96 h after attachment, 20 μl of MTS reagent (G358B, Promega) was added to each well, and the OD of each well was measured using a Multiskan FC microplate reader (Thermo) after a 2 h incubation at 37°C.

Flow cytometry
For ow cytometry analysis, Cells were seeded into 6-well plates and co-transfected with desired plasmids. After 36 h of growth, the cells were trypsinized and centrifuged (700 rpm, 10 min), and the conditioned medium was removed. The cells were resuspended in 600 μl of 1× PBS and 1.4 ml of precooled 70% ethanol, then incubated at 4°C overnight. The next day, the cells were pelleted by centrifugation and resuspended in 500 μl PBS, 2 μl of propidium iodide (PI) (P4864-10ML, Sigma) and 12.5 μl of RNase (2 mg/ml), followed by incubation in the dark for 30 min at room temperature. Samples were analyzed on a BD Accuri C6 ow cytometer (BD, Bioscience, Franklin Lakes, NJ, USA) and data analysis was performed using FlowJo software (Tree Star, Ashland, OR, USA).

Colony-forming assay
At 36 hours after transfection, 1,000-2,000 cells were seeded into 6-well plates and cultured for 2 weeks. When macroscopic clones were formed, cells were xed in 4% paraformaldehyde for 20 min, stained with hematoxylin for 5 min and photographed with a ChemiDoc Imaging System. The number of clones was scored.
Zymography analysis MMP9 activity was measured by gelatin zymography as previously described [11]. The conditioned medium of transfected cells was concentrated and subjected to SDS-PAGE electrophoresis with 1% gelatin. Following electrophoresis, the separation gel was washed twice in washing buffer (2.5% Triton X-100) for 30 minutes each wash, then incubated in incubation buffer (40 mM Tris-HCl (pH 8.0), 10 mM CaCl 2 ) at 37°C for 24 hours. The gel was stained with 0.1% Coomassie blue dye for 1 h, destained and xed. The proteolytic activities of MMP9 and its complex were detected as a clear white zone against a Coomassie Blue-stained gel background.

Fluorescent matrix degradation assay
Glass coverslips were coated with 0.01% poly-L-lysine (P8920-100ML, Sigma), air-dried overnight, and xed with 0.5% glutaraldehyde. subcutaneously were observed for the rst 3 days, and then intraperitoneal injections of 30 μM deferoxamine mesylate (DFOM) (HY-B0988, MCE) were given daily, for a total of 14 days. At the end of the experiment, the nude mice were euthanized and xenograft tumors, including footpad primary tumors and popliteal-in ltrated tumors, and popliteal lymph nodes were dissected, weighed, and analyzed as described previously [11]. The tumor sizes and lymph node volumes were calculated by the formula of width × length × height/2.

Statistical analysis
Data analyses were performed with SPSS 13.0 (SPSS, Inc., USA). Two-group comparisons were performed using Student's t-test. Differences between groups were assessed by one-way analysis of variance when more than two groups were compared. Differences were considered statistically signi cant at *P < 0.05 and **P < 0.01. Pearson correlation analysis was used for the correlation test of the two groups of data. Experimental data are presented as the means ± SD for at least three independent experiments.

Protein-protein interaction network and cell model construction
To identify the proteins that interact with LOXL2, we performed co-IP using LOXL2 as bait followed by LC-MS analysis. Mass spectrometry results showed that multiple LCN2 protein peptides were identi ed from the differentially-expressed 25kD band in LOXL2-overexpressing cells. The PEP values of the experimental group peptides were far less than 0.1, and the cumulative coverage was as high as 42.3%, indicating the results were highly credible (Supplementary Table 2). The secondary ion mass spectrum of the VPLQQNFQDNQFQGK proteolysis product of LCN2 protein is shown in Supplementary Fig. S1A. The peaks represented the secondary ion peaks of the peptide measured by the mass spectrum. These results indicated that LOXL2 and LCN2 interacted to form a complex.
Next, to gain a complete view of the interactome between LCN2 and LOXL2, all literature reporting interacting proteins was obtained from HPRD (http://www.hprd.org/) and BioGRID (http://thebiogrid.org/), and were processed and merged into a "parent network" as previously described [18]. The parental network was entered into the Cytoscape program to construct a secondary protein interaction sub-network using LCN2 and LOXL2 as the seed proteins. LCN2 was linked with LOXL2 by at least 5 pair partner proteins, further supporting that LCN2 interacts directly or physically with LOXL2 ( Supplementary Fig. S1B) The endogenous expression level of LCN2 and LOXL2 was examined in eight esophageal cancer cell lines, with human 293T as a control. KYSE410 and KYSE150 cells showed relatively low endogenous LCN2 and LOXL2 levels, and were selected as the cell model for subsequent experiments ( Supplementary  Fig. S1D). KYSE410 and KYSE150 cells were co-transfected with LCN2-HA/LOXL2-Flag, LCN2-HA/MMP9-Flag, or LOXL2-HA/MMP9-Flag plasmid pairs, or the corresponding vector plasmids. High expression of the target proteins was observed in KYSE410 and KYSE150 after transfection, accordingly ( Supplementary Fig. S1E).
The interaction patterns of intracellular and extracellular LCN2/LOXL2/MMP9 The protein-protein interaction network (PPIN) results showed that MMP9 interacts with both LCN2 and LOXL2. To con rm this, the endogenous interactions between three proteins were detected by immuno uorescence (IF). It revealed that in the KYSE150 cell line, LCN2 and LOXL2, LCN2 and MMP9, and LOXL2 and MMP9 were all co-localized, with the Pearson correlation coe cients as 0.79, 0.92 and 0.91, respectively (Fig. 1A). Therefore, the ternary complex of LCN2/LOXL2/MMP9 might exist in esophageal cancer cells. To further con rm the presence of LCN2/LOXL2/MMP9 ternary complexes in ESCC cells, co-immunoprecipitation using whole cell lysates was performed after co-transfection of LCN2-HA/LOXL2-Flag, LCN2-HA/MMP9-Flag, or LOXL2-HA/MMP9-Flag.
Considering LCN2, LOXL2 and MMP9 are all secreted proteins, the culture supernatant after transfection was also collected and concentrated to detect their extracellular interactions. The results showed that both in KYSE410 and KYSE150 cells, protein-protein interaction between LCN2 and LOXL2, and between LCN2 and MMP9, occurred both intracellularly and extracellularly ( Fig. 1B-C). Nevertheless, the proteinprotein interactions between LOXL2 and MMP9 occurred only intracellularly (Fig. 1D). Next, we cooverexpressed LCN2-HA/LOXL2-HA/MMP9-Flag in KYSE410 cells and carried out co-immunoprecipitation using the whole cell lysates and culture supernatants. We found protein-protein interactions between MMP9 and LCN2 both intracellularly and extracellularly, whereas the protein-protein interaction between MMP9 and LOXL2 was identi ed only intracellularly ( Fig. S2D).
In addition to migration and invasion, we also investigated the functional role of the LCN2/LOXL2/MMP9 ternary complex in cell proliferation. Colony-forming assays showed that overexpression of LCN2/LOXL2, LCN2/MMP9, LOXL2/MMP9 enhanced cell colony formation in the KYSE150 cell line (Fig. 2D). Similar results were obtained with the KYSE410 cell line (Supplementary Fig. S2E). MTS assay results showed that proliferation of ESCC cells in the experimental group did not change signi cantly after their overexpression in either KYSE410 or KYSE150 cells (Supplementary Fig. S2F-G). In the KYSE150 cell line, ow cytometry showed no signi cant cell cycle change in the experimental group ( Supplementary Fig.  S2H).
To obtain a stably overexpressing cell line, the Neon TM Transfection System (Thermo) was used to electroporate the LCN2-HA/LOXL2-HA/MMP9-Flag (abbreviated as LLM) plasmids into the KYSE150 cell line, followed by two weeks of G418 selection. Western blotting showed that a cell line stably overexpressing all three proteins was successfully established (Fig. 2E). Consistently, wound healing experiments showed that LLM increased the migration of esophageal cancer cells (Fig. 2F). In order to study the effect of LLM on ESCC tumor progression in vivo, we separately injected the cells stably overexpressing the ternary LLM complex and control cells into the footpads of nude mice. Compared with the control group, LLM cells displayed enhanced tumor progression (Fig. 2G). The anatomy results showed that 4 out 6 nude mice overexpressing LLM developed tumor metastasis, where the tumor cells migrated from the footpad to the ankle, and the lymph nodes of 2 nude mice were swollen (Fig. 2H).

LCN2/ LOXL2 and MMP9/ LOXL2 interactions depend on the SRCR domains
To identify the speci c domains for LOXL2/LCN2, and LOXL2/MMP9 protein-protein interaction in the ternary complex, two truncated LOXL2 plasmids were constructed, DLOXL2-HA (545-774aa) (lacking the N-terminal scavenger receptor cysteine-rich (SRCR) domains, denoted D4-HA), and DLOXL2-HA (1-544aa) (lacking the C-terminal amine oxidase domain, denoted D9-HA), respectively. First, these two plasmids were able to be expressed in KYSE150 cells (Fig. 3A). Immuno uorescence showed that in the KYSE150 cell line, D4-HA did not co-localize with either LCN2 or MMP9, with Pearson correlation coe cients of -0.27 and 0.21, respectively (Fig. 3B). Confocal Z-stack analysis was applied to scan different sections of transfected cells to con rm their co-localization ( Supplementary Fig. S3). Further, co-immunoprecipitation results showed that, the protein-protein interactions between LOXL2 and either LCN2 or MMP9 disappeared when the SRCR domains of LOXL2 were deleted (Fig. 3C-D). These results suggested the protein-protein interactions for LCN2/LOXL2 and MMP9/LOXL2 both depended on the SRCR domains of LOXL2. Furthermore, the biological roles of the N-terminus and C-terminus of LOXL2 was investigated. Transwell and invasion assays showed that the two truncated plasmids (D4 and D9) could reduce the ability of ESCC cells to migrate and invade, as well as colonize, compared to full-length LOXL2 (Fig. 3E-G).

LCN2 inhibitor DFOM inhibits the migration and tumor growth of ESCC cells
Deferoxamine mesylate (DFOM) is an iron chelator that binds free ferric iron and is mainly used to treat diseases caused by iron overload (19). KYSE150 cells were treated with 0, 5, 10, 25, 50, 100 and 200 μM DFOM. After 48 h, an MTS assay was performed, and the IC50 was calculated to be 30.993 μM. Therefore, 30 μM DFOM was used for subsequent experiments. DFOM could decrease the expression level of LCN2 in KYSE150 cells (Fig. 4A). Wound healing experiments showed that LCN2 overexpressionmediated enhancement of ESCC cell migration could be reduced by DFOM treatment (Fig. 4B). Compared with the stable overexpression of LLM, these results show that the expression of LCN2 can be signi cantly reduced after DFOM treatment, indicating that DFOM may destroy the formation of the ternary complex LLM or reduce its expression (Fig. 4C). Compared with empty vector group, the migration of ESCC cells could be signi cantly promoted by LLM, which was decreased after DFOM treatment (Fig.  4D). Colony-forming assays found that LCN2 or LLM could promote the proliferation ability of single ESCC cells, which could be inhibited by DFOM treatment (Fig. 4E). In order to study the effect of LCN2 and the LLM ternary complex on tumor progression of ESCC in vivo, as well as the anti-tumor effect of DFOM in vivo, a xenograft model was applied to observe and evaluate. Compared with the control group, LCN2 promoted ESCC tumor growth in vivo. But after intraperitoneal administration of DFOM, it was found that both the tumor volume and weight were signi cantly reduced (Fig. 4F). Similarly, the LLM ternary complex promoted tumor progression in vivo compared to the vector group. Consistently, tumor growth was inhibited, and the tumor volume and weight were also reduced after DFOM treatment (Fig.  4G).

LCN2/LOXL2/MMP9 promote the degradation of ECM
The LCN2/LOXL2/MMP9 ternary complex can promote the migration and invasion of ESCC cells, but the speci c molecular mechanism is still unclear. In the experimental groups that overexpressed LCN2/MMP9 or LOXL2/MMP9, the ability to degrade gelatin for MMP9 was elevated in zymography experiments (Fig. 5A). It was also observed that gelatin degradation at 130kD occurred for extracts of LCN2/MMP9 co-overexpressing cells, which was corresponding to a LCN2/MMP9 complex. In cells cooverexpressing LCN2/MMP9 or LOXL2/MMP9, gelatin degradation was also seen at around 200kD, which coincides with the predicted molecular weight of the LCN2/LOXL2/MMP9 ternary complex. We presumed that when co-overexpressing LCN2/MMP9 or LOXL2/MMP9, a LCN2/LOXL2/MMP9 ternary complex synergistically degrades the matrix, promoting the migration and invasion of ESCC cells (Fig.  5A).
Next, visualization of ECM degradation in living cells was carried out using uorescent matrix degradation assays. In KYSE150 cells co-overexpressing LCN2/LOXL2, LCN2/MMP9 and LOXL2/MMP9, the ability to degrade uorescent Matrigel was enhanced, as a signi cant black ghost was formed after degradation, whereas Matrigel degradation was barely observed in control cells (Fig. 5B). Three sets of co-overexpressing cells were scanned from top to bottom using the Z-stack function of the confocal microscope. The results con rmed that cells co-overexpressing LCN2/LOXL2, LCN2/MMP9 or LOXL2/MMP9 degraded uorescent Matrigel around the cell spheroids ( Supplementary Fig. S4).
The growth environment of tumor cells in vivo is three-dimensional. For this reason, we further characterized the ability of tumor cells to degrade ECM in a 3D cell co-culture model. Cells in the control group displayed a round and smooth morphological structure, except for the enlargement of the cell body itself. In the LCN2/LOXL2-, LCN2/MMP9-or LOXL2/MMP9-overexpressing groups, in addition to enlargement of the cell body, the ability of cells to degrade Matrigel was enhanced as more Matrigel surrounding the cells was digested. With the prolongation of culture time, elongation and branching of lopodia around the cells became clearly visible (Fig. 5C). In the stably overexpressing LCN2/LOXL2/MMP9 cell line, a similar result was obtained that LLM can signi cantly degrade Matrigel and promote the extension of lopodia (Fig. 5D).
MMP9 activity in the ternary complex was examined. The zymography results indicated that the ternary complex LLM could signi cantly enhance the enzymatic activity of MMP9 to degrade gelatin and form a distinct degradation band, while DFOM treatment could signi cantly weaken the enzymatic activity of MMP9 (Fig. 5E). The results of 3D cell co-culture found that after DFOM treatment of ESCC cells, the ability of the LLM complex to degrade Matrigel was reduced along with the number of lopodia (Fig. 5F).

LCN2/LOXL2/MMP9 induce cytoskeletal micro lament remodeling
Tumor cell migration and invasion usually involve rearrangement of the cellular micro lament system. In order to better observe changes in the actin cytoskeleton of ESCC cells, F-actin in KYSE410 cells was uorescently labeled after the overexpression of target proteins. The results showed that stress bers in control cells were arranged in regular bundles, while the stress bers in the overexpression group were more disordered, suggesting that overexpression of LCN2, LOXL2 and MMP9 could promote cell migration by remodeling the cell micro lament cytoskeleton (Fig. 6A). Pro lin 1 (PFN1) is a member of the pro lin family of small actin-binding proteins that play an important role in actin dynamics by regulating actin polymerization in response to extracellular signals [20]. To observe the actin arrangement, F-actin and pro lin 1 were simultaneously uorescence labeled after LCN2/LOXL2, LCN2/MMP9 and LOXL2/MMP9 were transfected. The micro lament skeleton in the three overexpressing groups was more disordered, compared with the control group (Fig. 6B). The protein level of pro lin 1 was increased and accumulated in the cytoplasm and at the leading edge of the cell. These results indicated that LCN2/LOXL2/MMP9 promotes cell migration by enhancing pro lin 1 expression and promoting micro lament rearrangement (Fig. 6B).
In the KYSE410 cell line, LCN2-HA/LOXL2-HA/MMP9-Flag was overexpressed, and the HA and Flag tags were uorescently labeled. It could be observed that LCN2, LOXL2, and MMP9 were all successfully overexpressed. Compared with the empty vector group, the cytoskeleton of cells overexpressing LLM was disordered, and the number of lopodia was signi cantly increased (Fig. 6C). Next, we aimed to analyze whether DFOM treatment could rescue the cytoskeleton disorder. We found the uorescence intensity of pro lin1 increased in LLM-overexpressing cells, but was decreased after DFOM treatment and accompanied by restoration of the cytoskeleton organization to an orderly state (Fig. 6D). These results suggest that the LLM ternary complex may enhance the expression of pro lin 1. To con rm this, western blotting showed the expression of pro lin 1 was decreased after DFOM treatment (Fig. 6E).
For further con rmation, pathway activators or inhibitors were applied to con rm the involvement of these signaling pathways. Capivasertib (MCE, HY-15431), which increases phosphorylation of AKT at both Ser473 and Thr308 (21), was added to KYSE150 cells to mimic the activation by LCN2/LOXL2/MMP9 complex. Compared with the control group, AKT phosphorylation was enhanced by capivasertib, and the expression level of LCN2, LOXL2 and MMP9 were increased as well (Fig. 7D).
Wortmannin is a pan-inhibitor of PI3K. When cells overexpressing LCN2/LOXL2/MMP9 were grown mixed with Matrigel in a 3D model to observe the changes of cell morphology, no obvious changes were observed in lopodia number and length except the enlarge of cell body itself. However, lopodia in the overexpression group began to form at 3 h after transfection. After 6 h, most cells in overexpressing group had formed lopodia that extended around the cell surface, thus promoting cell spreading, morphological change and cell migration. On the other hand, LCN2/LOXL2/MMP9-overexpressing cells treated with wortmannin had barely extended lopodia after 6 h, and only part of cells began to form the short lamentous pseudopods (Fig. 7E).
The SPOCK1 gene encodes a matricellular glycoprotein that belongs to a family of novel Ca2+-binding proteoglycans that promote cell migration in many tumors [22]. Expression of SPOCK1 was clearly increased after LCN2/LOXL2/MMP9 overexpression, to reshape the ECM, then promote the migration and invasion of ESCC cells. Next, the PI3K inhibitor wortmannin was added to cells overexpressing LCN2/LOXL2/MMP9. Wortmannin treatment resulted in inhibition of SPOCK1 expression, suggesting that the ternary complex of LCN2/LOXL2/MMP9 activates the AKT signaling pathway to promote SPOCK1 expression (Fig. 7F). Interestingly, overexpression of SPOCK1 in KYSE150 cells enhanced phosphorylation of AKT, suggesting a positive regulation feed-back loop between SPOCK1 and p-AKT involved by LCN2/LOXL2/MMP9 (Fig. 7G).

Discussion
Dysregulation of ECM structure and components are key events in tumor progression. The ability of cancer cells to migrate and invade is one of the hallmarks of solid metastatic cancer [23]. It is critical to understand how cancer cells interact with their microenvironment to migrate and invade the surrounding tissue, move towards the vasculature (blood/lymphatic vessels), and extravasate to create distant metastases, to discovering e cient targets for anti-cancer therapy (24).
Numerous studies have shown that LCN2, a secreted protein, is closely related to the malignant progression of tumors. Elevated LCN2 expression has been observed in various human solid tumors, including breast, colorectal, ovarian, gastric, ovarian, bladder, kidney, lung cancers, and ESCC [9]. A higher expression level of LCN2 is usually associated with tumor size, tumor stage and invasion of carcinoma cells, serving as a poor prognostic factor. These characteristics strongly suggest LCN2 might be a potential biomarker and therapeutic target in malignancies [9]. Previous studies have reported that LCN2 and MMP9 can form a heterodimer through disul de bonding to protect MMP9 enzyme activity, contributing greatly to tumor invasion and metastasis [11,[25][26]. High levels of monomeric forms of LCN2 and MMP-9, and LCN2-MMP-9 heterodimers are secreted into the extracellular space, and their levels seem to correlate with the aggressive behavior of neoplastic cells in several types of cancer (9). On the other hand, LOXL2 can effectively enhance MMP9 enzyme activity by elevating MMP9 expression [27]. LOXL2 could catalyze the cross-linkage of extracellular collagen to change the stiffness of the ECM, facilitating the motility of cancer cells [27][28]. MMP9 plays a core role in the degradation of the extracellular matrix and basement membrane of cancers to contribute to tumor invasion and metastasis [29][30].
Considering the crosstalk of functional roles for LCN2, MMP9 and LOXL2 in tumor progression, we hypothesized that an LCN2/LOXL2/MMP9 ternary complex might exist in esophageal cancer and play a synergistic role in biological function. First, we used mass spectroscopy to show LCN2 could directly interact with LOXL2. An LCN2/LOXL2 PPI network suggested a physical interaction might occur between LCN2 and LOXL2, which might also involve other interacting proteins, such as MMP9. Our results show that an LCN2/LOXL2/MMP9 ternary complex is formed in esophageal cancer, with a distinctive intracellular and extracellular interaction pattern. The SRCR structure is a key domain through which LOXL2 interacts with both LCN2 and MMP9 to play an important biological role in esophageal cancer. In recent years, the N-terminal SRCR repeats for the LOX gene family has been found to play an important functional role [11]. Similarly, the SRCR domain in LOXL3, rather than the C-terminal oxidase catalytic domain, represents the major deacetylase/deacetyliminase activity in the modi cation of STAT3 [31].
Based on LCN2/LOXL2, LCN2/MMP9, LOXL2/MMP9 protein-protein interactions, we further elucidated the molecular and cellular mechanisms underlying the LCN2/LOXL2/MMP9 ternary complex on promoting migration and invasion of esophageal cancer cells, playing a synergistic role. Zymography showed that an elevated LCN2/LOXL2/MMP9 ternary complex was presented, especially in the case of enhanced expression of MMP9. MMP9 has three bronectin type II homologous repeat domains that bind to gelatin with high a nity, making gelatin the major substrate for MMP9 (32)(33). Degradation of uorescent bronectin (FN) experiments revealed that co-expression of LCN2/LOXL2, LCN2/MMP9, or LOXL2/MMP9 confers a greater ability to degrade FN. FN is a major non-collagen glycoprotein in the extracellular matrix and basement membrane, which has the function of adhering brin and collagen [34].
A previous study found a high stromal FN content facilitates tumor cell metastasis by promoting morphological change and improving the motility and migratory ability of ESCC cells [35]. Therefore, the LCN2/LOXL2/MMP9 ternary complex can promote the invasion of esophageal cancer cells by elevating the expression of MMPs to degrade gelatin and bronectin. Considering that growth of tumor cells in vivo occurs in a three-dimensional environment, 3D cell co-culture showed that the LCN2/LOXL2/MMP9 ternary complex degrades the extracellular matrix by promoting the formation and extension of lopodia, thereby enhancing the invasion of tumor cells. Matrigel® is widely used in cancer cell 3D cell co-culture models, which contains proteins commonly found in the basement membrane of epithelial structures, such as laminin, as well as type IV collagen and heparan sulfate proteoglycan [36][37]. MMP9 is able to degrade the laminin and type IV collagen in Matrigel. Therefore, we conclude that the LCN2/LOXL2/MMP9 ternary complex can promote the invasion of cancer cells by degrading laminin and type IV collagen.
In addition to degradation of the extracellular matrix, the mobility of tumor cells is also an important factor contributing to migration and invasion. Actin is a highly conserved protein that participates in various types of cell movement and is universally expressed in all eukaryotic cells. Polymerization of individual actin laments into organized parallel bundles occurs in lopodia, which are slender protrusions that can extend far beyond the cell edge and reach or sense distant targets [38]. Pro lin 1 (PFN1) is an abundant actin-binding protein that promotes nucleotide exchange of actin and converts ADP/G-actin to ATP/G-actin [39]. The PFN1-ATP/G-actin complex can interact with the fast-growing end of F-actin to increase ATP/G-actin for growing actin laments. Thus, pro lin 1 is a central mediator of actin micro lament and microtubule dynamics [20]. Cells overexpressing LCN2/LOXL2, LCN2/MMP9, LOXL2/MMP9 or LLM exhibit a disordered micro lament skeleton, increased lopodia, and increased expression of pro lin 1 distributed in the cell interior and the leading edge of the cell. These results suggest that LCN2/LOXL2/MMP9 could enhance cell migration by promoting micro lament rearrangement and enhancing pro lin 1 expression. The cytoskeleton, with its regulatory and structural proteins, had emerged as a novel and highly effective target to be exploited for speci c anti-metastatic drugs [40]. Though at a glance, it is contradictory that when considering their individual role in ECM remodeling, LCN2//MMP9 mainly contributes to matrix degradation, while LOXL2 strengthens ECM stiffness by catalyzing the cross-linking of collagen and elastin. Nevertheless, we provide evidence that LCN2/LOXL2, LCN2/MMP9, LOXL2/MMP9 or LLM affect tumor growth and progression in a synergistic manner. We presumed that ECM degradation opens a path, and ECM stiffness favors the assembly of lopodia and invadosome, both of which contribute to cancer cell migration and invasion.
Iron (Fe) plays a vital role in various biological processes. Cancer cells show a high rate of Fe metabolism and therefore require more Fe to proliferate. Evidence indicates that tumor cells are very sensitive to iron de ciency, much more sensitive than normal cells [41]. LCN2 is an important ferritin carrier, which can increase the iron level in cells to promote the progression of malignant behavior [42]. Fe chelating agents have been used to treat a variety of diseases, such as leukemia, neuroblastoma and breast cancer [43].
Among them, DFOM has been extensively studied and found to have anti-tumor effects by the treatment of iron overload [44][45].  [49]. In this study, we found that the LCN2/LOXL2/MMP9 ternary complex activates the FAK/AKT/GSK3β signaling pathway to promote expression of SPOCK1. SPOCK1 is closely related to cell proliferation, adhesion and metastasis, and it is highly expressed in a variety of cancers, including esophageal squamous cell carcinoma, liver, gallbladder, colon, and prostate cancer [50]. Liu et al. found that SPOCK1 upregulates the expression and activity of MMP9, causes remodeling of the ECM and promotes tumor cell migration and invasion [51]. Therefore, LCN2/LOXL2/MMP9 activates the FAK/AKT/GSK3β signaling pathway to enhance SPOCK1 expression and remodel ECM, thus promoting the migration and invasion of esophageal cancer cells. Moreover, degradation of the ECM could be inhibited after adding the PI3K inhibitor wortmannin, suggesting that inhibition of AKT phosphorylation at ser473 could inhibit the migration and invasion of esophageal cancer. So, AKT (ser473) phosphorylation could be applied as a potential therapeutic target for esophageal cancer. Taken together, we propose a schematic model that the LCN2/LOXL2/MMP9 complex promotes the migration and invasion of cancer cells, through intracellular and extracellular means (Fig. 8).

Conclusions
In this study, we found that an LCN2

Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.