The β2-Subunit (AMOG) of Human Na+, K+-ATPase Is a Homophilic Adhesion Molecule

The β2 subunit of Na+, K+-ATPase was originally identified as the adhesion molecule on glia (AMOG) that mediates the adhesion of astrocytes to neurons in the central nervous system and that is implicated in the regulation of neurite outgrowth and neuronal migration. While β1 isoform have been shown to trans-interact in a species-specific mode with the β1 subunit on the epithelial neighboring cell, the β2 subunit has been shown to act as a recognition molecule on the glia. Nevertheless, none of the works have identified the binding partner of β2 or described its adhesion mechanism. Until now, the interactions pronounced for β2/AMOG are heterophilic cis-interactions. In the present report we designed experiments that would clarify whether β2 is a cell–cell homophilic adhesion molecule. For this purpose, we performed protein docking analysis, cell–cell aggregation, and protein–protein interaction assays. We observed that the glycosylated extracellular domain of β2/AMOG can make an energetically stable trans-interacting dimer. We show that CHO (Chinese Hamster Ovary) fibroblasts transfected with the human β2 subunit become more adhesive and make large aggregates. The treatment with Tunicamycin in vivo reduced cell aggregation, suggesting the participation of N-glycans in that process. Protein–protein interaction assay in vivo with MDCK (Madin-Darby canine kidney) or CHO cells expressing a recombinant β2 subunit show that the β2 subunits on the cell surface of the transfected cell lines interact with each other. Overall, our results suggest that the human β2 subunit can form trans-dimers between neighboring cells when expressed in non-astrocytic cells, such as fibroblasts (CHO) and epithelial cells (MDCK).


Introduction
The Na + , K + -ATPase (Na + -pump) is a transmembrane enzyme responsible for the active transport of Na + and K + ions through the plasma membrane of all animal cells [1]. The Na + -pump catalyzes the hydrolysis of ATP as an essential energy source to transport three Na + ions out of the cell and two K + ions into the cell against ion gradients [2]. This important pump is constituted of two indispensable subunits, α and β. The α subunit (110 kDa) contains the binding sites for ATP, Na + , K + , and sites for inhibitors and activators. All tissue-specific isoforms of the Na + , K + -ATPase α subunit (α 1 , α 2 , α 3 , and α 4 ) share Thus, N-glycans are probably required for stabilizing the amino acid-mediated interactions between the extracellular domains of the two β 1 subunits. In the case of β 2 , the cis-interaction-at least with Basigin/CD147-has been shown to involve the lectin domain of Basigin/CD147 and the oligomannosidic residues of β 2 [25]. Nevertheless, little is known about the role of N-glycans of β 2 in mediating cell-cell trans-interactions. Therefore, we decided to approach some of the open questions related to the β 2 adhesion mechanism using similar strategies that were used for demonstrating that β 1 is a homotypic transinteracting adhesion molecule in epithelial cells. In the present work, we expressed the human β 2 subunit of Na + , K + -ATPase in CHO fibroblasts and designed in vitro and in vivo experiments that strongly suggest that the β 2 subunit is also a homophilic trans-interacting cell adhesion molecule.

In Silico Prediction of Glycosylated β 2 -β 2 Trans-Interaction
As a first approach aimed to clarify the adhesive properties of the β 2 subunit of Na + , K + -ATPase, we decided to use bioinformatics tools to find out whether β 2 -β 2 interaction is relevant considering the physicochemical properties of the extracellular domain of the glycosylated protein (EDβ 2 ). To begin with, since there is no crystal structure available for ATP1B2, we generated a three-dimensional model. In search of a template for the β 2 subunit, we selected the β subunit of pig gastric proton pump due to its highest identity with the human β 2 isoform of Na + , K + -ATPase (39.82%). Thus, a 3D model of the EDβ 2 was generated as specified in Methods. Figure 1A illustrates the model obtained by the Swiss Model Program [26]. The corresponding Ramachandran plot depicted in Figure 1B shows that none of the residues are located in the disallowed region. These results suggest that our template-based model is valid. Thereafter, we proceeded to build the dimeric structure of the protein. Protein-protein docking was performed using HDOCK server, which is a specialized program for protein-protein docking analysis [27,28]. HDOCK uses the binding information from the PDB and a hybrid docking algorithm of template-based modeling and free docking [27]. As we were looking for a trans-interaction, we chose conformers in which the membranes are positioned in an opposing (parallel) way. Then, from the list of seven predictive models, model 1, with the higher ranking, and a free binding energy of −274.99 kcal/mol was selected to be illustrated in Figure 1C. The putative binding interface composed of 11 residues in each monomer is depicted in Figure 1D. Interestingly, for both Chains, at least four glycosylation sites of the EDβ 2 are located in the interface (Asn118, Asn153, Asn159, and Asn197) and their interactions are illustrated in Figure S1.

Generation of Experimental Tools
Encouraged by the in silico results, we designed in vitro and in vivo experiments that would further evaluate the trans-interaction between β2 isoforms. Thus, we expressed the human β2 subunit of Na + , K + -ATPase in CHO fibroblasts and in MDCK epithelial cells. We selected these cells, as they do not express that isoform of β subunit. A specific antibody against the human β2 subunit does not detect that protein in wild-type CHO fibroblasts or MDCK cells when assayed by immunofluorescence microscopy (IF) (Figure 2A) or by Western blot (WB) ( Figure 2B).

Generation of Experimental Tools
Encouraged by the in silico results, we designed in vitro and in vivo experiments that would further evaluate the trans-interaction between β 2 isoforms. Thus, we expressed the human β 2 subunit of Na + , K + -ATPase in CHO fibroblasts and in MDCK epithelial cells. We selected these cells, as they do not express that isoform of β subunit. A specific antibody against the human β 2 subunit does not detect that protein in wild-type CHO fibroblasts or MDCK cells when assayed by immunofluorescence microscopy (IF) (Figure 2A) or by Western blot (WB) ( Figure 2B). Transfection with the cDNA of human β2 subunit fused to the yellow fluorescent protein resulted in numerous stable clones of CHO β2 YFP and MDCK β2-YFP that were positive for β2 subunit by both WB analysis and fluorescence microscopy imaging. A representative stable clone of each is shown in Figure 2A. The YFP images were captured by confocal microscopy at 530 nm. As can be noticed, the β2 subunit is concentrated in the cell-cell contacts as observed for the β1 subunit in wild-type MDCK cells. Cell lysates of both CHO and MDCK β2-YFP cells showed an expected molecular weight of approximately 75 kDa when both anti-β2 (upper panel) and anti-GFP (lower panel) antibodies have been used ( Figure 2B). Moreover, that distribution at cell-cell contacts resembles one of many other cell-cell adhesion molecules. Therefore, we proceeded to study the adhesive properties of CHO β2 YFP cells. Transfection with the cDNA of human β 2 subunit fused to the yellow fluorescent protein resulted in numerous stable clones of CHO β 2 YFP and MDCK β 2 -YFP that were positive for β 2 subunit by both WB analysis and fluorescence microscopy imaging. A representative stable clone of each is shown in Figure 2A. The YFP images were captured by confocal microscopy at 530 nm. As can be noticed, the β 2 subunit is concentrated in the cellcell contacts as observed for the β 1 subunit in wild-type MDCK cells. Cell lysates of both CHO and MDCK β 2 -YFP cells showed an expected molecular weight of approximately 75 kDa when both anti-β 2 (upper panel) and anti-GFP (lower panel) antibodies have been used ( Figure 2B). Moreover, that distribution at cell-cell contacts resembles one of many other cell-cell adhesion molecules. Therefore, we proceeded to study the adhesive properties of CHO β 2 YFP cells.

Over-Expression of β 2 YFP Increases Cell Aggregation in a Glycosylation-Dependent Manner
The β 2 subunit has been shown to function as a cell adhesion and recognition molecule mainly in astrocytes. To test whether the expression of the β 2 subunit in CHO cells (CHO β 2 YFP) confers adhesiveness, we used two different adhesion assays.
One is a Dispase I adhesion assay. As described in Methods, a monolayer of CHO β 2 YFP adhered to the culture dish is treated with Dispase I to detach cell-substrate adhesion. A monolayer of wild-type CHO cells is detached as single cells suspended in the culture medium ( Figure 3). On the other hand, epithelial cells firmly attached by adhesion complexes (adherens and tight junctions) detach as a sheet [29]. In the case of CHO β 2 YFP and CHO β 1 , as shown in Figure 3B,C, the cells detach as big aggregates. These observations suggest that CHO cells overexpressing the β 2 YFP subunit form a pseudo-epithelial monolayer establishing both cell-substrate and cell-cell interactions. The attached cells are harvested with trypsin and re-suspended as single cells in the second adhesion assay. Then, the cells are allowed to aggregate by incubating at 37 • C overnight in constant agitation. Figure 4 illustrates the results of an aggregation assay performed with CHO β 2 YFP cells. As controls, we compared CHO wt cells and CHO cells transfected with the canine β 1 subunit (CHO dog β 1 ). The cells were cultured to 100% of confluence, as shown in the confocal microscopy images corresponding to each treatment ( Figure 4A). The relative amount of aggregates that have been formed was estimated by flow cytometry.
A representative dot plot is shown for each treatment ( Figure 4B). Figure 4C shows the average of aggregates created in three independent experiments. CHO fibroblasts form a small number of aggregates (less than 5%) while the CHO YFP-β 2 cells form a significantly higher proportion (30%) as well as the CHO dog β 1 cells (20%). As cell-cell adhesion mediated by β 1 -β 1 interaction depends on the N-glycosylation of the β 1 subunits [23,24], we analyzed whether N-glycosylation is involved in the cell-aggregation of CHO YFPβ 2 . As depicted in Figure 4B,C, treatment with Tunicamycin significantly decreases the aggregation of CHO YFP-β 2 cells. As expected, it also diminished the aggregation of CHO dog β 1 cells. These experiments show that the adhesive property of β 2 /AMOG of human Na + , K + -ATPase is conserved when expressed in fibroblasts from hamster ovary (CHO) and that it involves N-glycans. Therefore, we next addressed whether this aggregation could be mediated by β 2 -β 2 interaction. The β2 subunit has been shown to function as a cell adhesion and recognition molecule mainly in astrocytes. To test whether the expression of the β2 subunit in CHO cells (CHO β2 YFP) confers adhesiveness, we used two different adhesion assays.
One is a Dispase I adhesion assay. As described in Methods, a monolayer of CHO β2 YFP adhered to the culture dish is treated with Dispase I to detach cell-substrate adhesion. A monolayer of wild-type CHO cells is detached as single cells suspended in the culture medium ( Figure 3). On the other hand, epithelial cells firmly attached by adhesion complexes (adherens and tight junctions) detach as a sheet [29]. In the case of CHO β2 YFP and CHO β1, as shown in Figure 3B,C, the cells detach as big aggregates. These observations suggest that CHO cells overexpressing the β2 YFP subunit form a pseudo-epithelial monolayer establishing both cell-substrate and cell-cell interactions. The attached cells are harvested with trypsin and re-suspended as single cells in the second adhesion assay. Then, the cells are allowed to aggregate by incubating at 37 °C overnight in constant agitation. Figure 4 illustrates the results of an aggregation assay performed with CHO β2 YFP cells. As controls, we compared CHO wt cells and CHO cells transfected with the canine β1 subunit (CHO dog β1). The cells were cultured to 100% of confluence, as shown in the confocal microscopy images corresponding to each treatment ( Figure 4A). The relative amount of aggregates that have been formed was estimated by flow cytometry. A representative dot plot is shown for each treatment ( Figure 4B). Figure 4C shows the average of aggregates created in three independent experiments. CHO fibroblasts YFP-β2. As depicted in Figure 4B,C, treatment with Tunicamycin significantly decreases the aggregation of CHO YFP-β2 cells. As expected, it also diminished the aggregation of CHO dog β1 cells. These experiments show that the adhesive property of β2/AMOG of human Na + , K + -ATPase is conserved when expressed in fibroblasts from hamster ovary (CHO) and that it involves N-glycans. Therefore, we next addressed whether this aggregation could be mediated by β2-β2 interaction.  The percent of cell aggregates in R2 were averaged and illustrated in a comparative bar graph. The amount of cell aggregates of fully glycosylated CHO dog β 1 and CHO β 2 YFP cells is significantly higher than that of CHO wt cells. Error bars, ±S.D. (n = 5); **, significant difference from CHO dog β 1 , p < 0.005, ***, significant difference from CHO β 2 YFP, p < 0.005.

Cell Aggregation of CHO β 2 YFP Decreases When Cocultured with CHO wt Cells
If cell-cell aggregation of CHO β 2 YFP cells observed by flow cytometry is mediated by the β 2 -subunits of neighboring cells, we expected that diluting the cell suspension with non-transfected CHO cells would reduce the proportion of cell aggregates. On the other hand, if the β 2 subunit is binding to a unlike adhesion molecule on the plasma membrane of CHO fibroblasts, mixing these two cell types will not affect the relative amount of cell aggregates. Thus, we performed an aggregation assay combined with fluorescenceactivated cell sorting (FACS). The cytoplasm of CHO wt cells was pre-stained with CMTMR and, therefore, could be detected as red when analyzed by fluorescence microscopy or FACS. The aggregation assay in Figure 5 has been carried out as described in Figure 4. In addition to pure CHO wt and pure CHO β 2 YFP cells, we also used mixed suspensions of these two types of cells. The aggregates formed after three hours of agitation were sampled and imaged by confocal microscopy ( Figure 5A). While CHO wt cells (in red) produce very small if any aggregates, the β 2 YFP expressing cells form large ones. When mixing the two cell types in a 1:1 ratio, it can be observed that there are very few wt cells intercalated in those big (green) aggregates. Most of the wt cells (red) in the mixture were single cells. The same cell suspension was sorted simultaneously for size, granularity, and red and yellow fluorescence. The obtained data were analyzed as in Figure 4. The population of particles sorted as R2 ( Figure 4B) was then sorted for red, green, and mixed fluorescence ( Figure 5C). The graph in Figure 5D depicts the data obtained from three independent experiments. It can be seen that gradual dilution of β 2 YFP expressing cells with CHO wt cells decreased the relative amount of cell aggregates proportionally. The FACS data suggest that the formed aggregates were mainly of CHO β 2 YFP. The proportion of mixed aggregates (of both red and green color) is small and very similar in all dilutions (4% approximately). The fluorescence images and the FACS analysis shows that the aggregates are mainly formed by cells expressing β 2 YFP, therefore suggesting that the observed cell-cell adhesion and aggregation is mediated by the β 2 YFP on the surface of the transfected CHO cells.

Cell Aggregation of CHO β 2 YFP Is Due to β 2 -β 2 Interactions In Vivo
We have shown that CHO dog β 1 cells form aggregates in previous work. We demonstrated that the adhesiveness is due to a direct β 1 -β 1 interaction between adjacent cells [19].
To determine whether the β 2 subunits of Na + , K + -ATPase located on adjacent cell membranes can trans-interact and contribute to the observed cell aggregation of CHO β 2 YFP, we implemented an in vitro Pull-down assay. As a bait, we immobilized the β 2 subunit through the fused YFP on Sepharose beads coupled to an anti-GFP antibody. We then used as prey cell lysates that contain or do not contain β 2 subunits and as control lysates containing β 1 subunit. As shown in Figure 6A, none of the pull-down assays using β 2 YFP as bait resulted in an interacting β 2 subunit as prey. Then, we expressed a His 6 tagged human β 2 subunits in MDCK cells ( Figure 2A) and used it as bait, immobilized on Ni ++ beads. As prey, we used a cell extract expressing an identical β 2 subunit tagged with YFP. As shown in Figure 6B, there is no interaction between the two identical subunits in this case either. These results indicate that the apparent β 2 -β 2 interaction between live cells (Figures 2 and 3) cannot be reproduced in vitro by a pull-down assay, at least not under these specific experimental conditions. Therefore, we decided to try an in vivo assay of protein-protein interaction, which we named co-affinity precipitation (Co-AP) and described in detail in Materials and Methods. As shown in Figure 6C, a co-culture of MDCK cells expressing YFP tagged and His 6 tagged β 2 subunits form a continuous monolayer and localize the Na-pump expressing the β 2 subunit at both homotypic and heterotypic cell-cell contacts. A cell extract of co-cultured cells expressing the recombinant human β 2 subunits was incubated with Ni ++ -NTA beads. After washing and elution, both β 2 subunits were detected in western blot ( Figure 6D), suggesting that these two proteins interacted in vivo, throughout the co-culture.   For pull-down assay, β 2 YFP was immobilized on GFP-sepharose beads (bait). Then, the lysate of rat brain or ARPE-19 cells expressing β 2 subunit (prey) were incubated to form β 2 -β 2 complexes. Western blot analysis shows the expression of β 2 subunit in ARPE-19 cells and rat brain extract (TCE-Total Cell Extract). Both anti-GFP and anti-β 2 antibodies detect the β 2 subunit in CHO β 2 YFP cells. Nevertheless, none of the assayed β 2 subunits were pulled down by immobilized β 2 YFP. (B) A pull-down assay using β 2 His 6 as bait, immobilized on Ni-NTA HF resin. Total extracts of CHO β 2 YFP and MDCK β 2 YFP cells (prey) were incubated to form β 2 -β 2 complexes. WB analysis shows the expression of β 2 subunits in all transfected cell lines. None of the prey proteins (tagged with YFP) appears to form β 2 -β 2 interaction (Eluted) on the Ni-NTA HF resin. (C) IF images of a co-culture of MDCK cells transfected with β 2 YFP or β 2 -His 6, in a 70/30 ratio. On the left panel, an image of cells expressing β 2 YFP are shown in Yellow (528 nm), in the middle panel, β 2 -His 6 was stained with anti-β 2 antibody (Red), and on the right panel, a merged image showing continuous orange cell-cell contacts between the two cell-lines. (D) Co-affinity precipitation assay (Co-AP) performed with a co-culture as described in (C). As a negative control, a similar co-culture was made with CHO cells expressing β 1 -His 6 and MDCK β 2 YFP. Total cell extracts of the co-cultures were loaded on Ni-NTA HF resin to separate β 2-β 2 complexes that were formed in vivo. Eluted samples were analyzed by WB. Only the eluate of β 2 -His 6 and β 2 -YFP co-culture show a pattern of bands corresponding to that of the cell extracts of β 2 -YFP and β 2 -His 6 .

Discussion
The β 2 subunit of Na + , K + -ATPase has two different functions: it is an adhesion molecule on glia involved in the interaction between neural and glial cells; and a subunit of the Na + , K + -ATPase, essential for establishing a functional ion pump. In the present work, we have been focused on the adhesive function of the β 2 isoform. We studied the expression of the human β 2 isoform of the Na + , K + -ATPase transfected in CHO fibroblasts (CHO β 2 YFP) and MDCK epithelial cells. The expression of the recombinant protein was observed mainly at the plasma membrane. Our significant results include increased aggregation of CHO β 2 YFP fibroblasts that depend on N-glycosylation and a presence of trans-interaction with homotypic β 2 subunit in vivo and in silico.
Numerous investigations have shown the adhesion/recognition functions of β 2 /AMOG. Initially, β 2 /AMOG has been shown to promote neurite outgrowth [12]. Later, liposomes containing reconstituted β 2 /AMOG adhered to neurons in vitro and a specific antibody to β 2 /AMOG inhibited this adhesion [13]. The first question was whether these functions of β 2 /AMOG depend on the pump activity. Additional studies have demonstrated that inhibition of the pump by ouabain did not affect neuron-glial adhesion [14]. Moreover, fixed and nonviable cells expressing β 2 /AMOG can still promote neurite outgrowth [20]. The clear conclusion was that there must be a receptor molecule on neurons that recognizes the β 2 /AMOG on the astrocytes surface. Early studies aimed to identify the neuronal receptor of β 2 /AMOG were focused on the fact that β 2 /AMOG but not β 1 is responsible for that function. Since the main difference between β 1 and β 2 isoforms is their level of glycosylation, it was postulated that the adhesive function resides on the carbohydrates of β 2 /AMOG. The monoclonal AMOG antibody that blocks adhesion recognizes a nonglycosylated protein [12], suggesting that the carbohydrates are not responsible for the adhesion. However, co-culture experiments for neurite outgrowth on L-cells transfected with β 2 /AMOG show an increase in neurite outgrowth that was partially competed with the soluble, extracellular domain of β 2 /AMOG (recAMOG) but the non-glycosylated recAMOG had no effect [20]. Moreover, as β 2 /AMOG bears oligo-mannoses and carbohydrate epitopes (L3 and L4) common for neuronal cell adhesion molecules, it was proposed that β 2 /AMOG has a lectin activity. Nevertheless, no interaction has been detected between β 2 /AMOG and neuronal CAMs such as L1 or NCAM [13], and no lectin activity has yet been demonstrated for β 2 /AMOG.

The Role of N-Glycosylation
In the present work, the aggregation assay of CHO β 2 YFP cells reveals the adhesive property of the recombinant cells, suggesting a homophilic interaction between the β 2 subunits of Na + , K + -ATPase on the surface of neighboring cells. These results corroborate the observations previously made in transfected L-fibroblasts [20] and U87-MG glioma cells [30]. Experiments using Tunicamycin show that the N-glycosylation of β 2 plays an important role in β 2 -mediated cell-cell adhesion. In fact, the significance of the Nglycosylation sites of β 2 -subunit of Na + , K + -ATPase have been already demonstrated for the adhesion of Basigin/CD147 to the pump [25]. However, that interaction has been shown to occur in astrocytes, in cis, namely in the same plasma membrane. The group of Vagin, have published series of works focused on the roles of the N-glycosylation sites of the β subunits of Na + , K + -ATPase and H + ,K + -ATPase [10,11,23,24,31]. As they summarize in [11], "Individual N-glycans linked to the β subunits of the Na + , K + -ATPase and H + ,K + -ATPase are important for stable membrane integration of their respective α subunits, folding, stability, subunit assembly, and enzymatic activity of the pumps. They are also essential for the quality control of unassembled β subunits that results in either the exit of the subunits from the ER or their ER retention and subsequent degradation". Here, we show ( Figure 4A) that CHO β 2 YFP cells incubated with Tunicamycin rarely express the recombinant protein in their plasma membrane, in comparison to the non-treated cells. Therefore, it suggests that N-glycosylation of β 2 subunit of Na + , K + -ATPase is also important for the exit from ER and arrival of the Na + , K + -ATPase to the plasma membrane and thus for accomplishing cell-cell adhesion in CHO cells. Nonetheless, we cannot exclude the participation of a distinct glycosylated molecule because Tunicamycin treatment would also inhibit the Nglycosylation of other membrane proteins. Therefore, more specific experiments should be conducted. Our future studies will evaluate the in silico participation of the seven N-glycosylation sites in β 2 -β 2 interaction, and based on that information we will choose to mutate just those sites that are suggested to be involved. If cells expressing these mutants fail to make aggregates or are not pulled down in a co-culture assay with wild type β 2 subunit, we will be able to confirm and even to indicate which of the seven N-glycosylation sites are important for β 2 -β 2 trans-interaction.

β2/AMOG Is a Heterophilic Adhesion Molecule in Astrocytes
Early study from Melita Schachner laboratory in which they generated KO mice of β 2 /AMOG suggested that β 2 /AMOG in astrocytes is mainly necessary for the proper activity of the Na + , K + -ATPase [32]. Our studies [6] and others [7] related to the Na + , K + -ATPase in epithelia [33] and in cardiomyocytes [34] revealed that Na + , K + -ATPase is not just a pump. Na + , K + -ATPase plays an important role in cell-signaling mechanisms and in cell-cell adhesion through β-β interaction. Moreover, recent works report the participation of the β 2 /AMOG in cell signaling mechanisms in astrocytes [35] and in cerebellar granule neurons [36]. β 2 /AMOG-mediated Akt phosphorylation specifically activates the mTOR/p70S6 kinase pathway implicated in cell size regulation. Consistent with that, we noticed that CHO wt and CHO β 1 scattered cells are small and rounded particles, in comparison with the CHO β 2 YFP scattered cells that are bigger and with irregular form. In addition, some of the CHO β 2 YFP cells present membrane protrusions or blebs ( Figure 3). Therefore, we consider that the Na + , K + -ATPase composed of β 2 /AMOG subunit participates in both ion pumping and adhesion-based cell-signaling in astrocytes. All former studies have suggested that β 2 /AMOG interacts with a heterophilic binding partner on the neuronal cell surface. Our present work does not argue with that conclusion, and actually, we are trying to identify this heterophilic molecule in our lab.

β2/AMOG Is a Homophilic Cell-Adhesion Molecule
Based on their protein structures in the Protein Data Bank, all the β subunits of Na + , K + -ATPase and H + , K + -ATPase are classified as a subgroup of the IgG superfamily (CATH) given their β-sandwich domain. Since β 1 subunits are homophilic cell-adhesion molecules, we wondered about the capability of β 2 subunits to recognize and interact with homophilic β 2 subunits in neighboring cells. Therefore, we first evaluated this possibility using in silico approach, modeling the 3D structure of the human β 2 subunit and applying docking analysis. Our results, showing a stable interacting interface between two extracellular domains of β 2 subunits, encouraged us to explore this possibility further. Antonicek and Schachner [13] have already treated whether β 2 /AMOG is a homophilic adhesion molecule. They pre-incubated cerebellar neuron monolayers with an anti-AMOG antibody before the addition of AMOG-containing liposomes. They found that the antibody did not reduce AMOG-liposomes binding to neurons. Therefore, they concluded that β 2 /AMOG does not interact with itself. Our protein-protein interaction experiments in vitro neither show a homophilic interaction between β 2 subunits. The pull-down experiments performed in this study ( Figure 6A,B) suggest that the β 2 subunits of Na + , K + -ATPase would not be able to adhere to each other in vitro, out of a cell context. One possibility to explain it could be that another factor is involved in the interaction of the β 2 subunits between cells and that we lost it during the in vitro assay. The second possibility could be that protein folding is affected during in vitro pull-down, and as a result there is a change in the interaction site between β 2 subunits, turning it inaccessible. On the other hand, our experiments in vivo show that cultivating CHO cells expressing human β 2 /AMOG in their plasma membrane results in adhesive cells (Figure 3). Moreover, the aggregation and adhesion assays performed with CHO β 2 YFP cells (Figures 4 and 5) and the co-affinity precipitation assay ( Figure 6D) validate the homophilic property of cell-cell adhesion mediated by β 2 subunits. Overall, previous and present findings indicate that the trans-homophilic interaction of β 2 subunits cannot occur out of the plasma-membrane context of the cell. Since experiments evaluating the effect of anti-AMOG antibodies on astrocyte-astrocyte adhesion have shown that there is not such an interaction [13], a relevant question would be: why do astrocytes not adhere to each other by β 2 /AMOG subunits? There are several possible answers to that mystery. One can be related to recent findings reporting that in astrocytes, the Na + , K + -ATPase composed of α 2 /β 2 is part of a protein complex formed by PrP, GluR2, α 2 /β 2 -ATPase, basigin, and MCT1 that regulates lactate transport of astrocytes and may be functional in the metabolic cross-talk between astrocytes and neurons [22]. Interestingly, basigin and β 2 /AMOG-two components of that complex-are attached to each other by a lectinoligomannosidic interaction [25]. Therefore, this interaction could be interfering with a homophilic interaction between the β 2 subunits of adjacent astrocytes. On the other hand, neurons do not express the β 2 subunit, thus a homophilic cell-cell interaction between β 2 subunits would not be responsible for the described neuron-astrocyte adhesion. Therefore, we cannot exclude the existence of a receptor molecule at the neuron surface that recognizes and binds to β 2 /AMOG. It is well established that cell adhesion molecules interacting with extracellular matrix (ECM) components regulate and modulate the formation and maintenance of synapses in the nervous system (for a recent review, see [37]). Thus, an unexplored possibility could be an interaction of β 2 /AMOG with ECM.
Based on the new findings reported here, we infer that β 2 /AMOG is a heterophilic astrocyte-neuron adhesion molecule that developed a self-avoidance mechanism in a similar way to how neurons avoid forming synapses with themselves [38]. Interestingly, recent studies have related the β 2 /AMOG with glioblastoma (GB), the most devastating type of glioma [21,30,[39][40][41]. Those studies depict the β 2 /AMOG as a tumor suppressor cell adhesion molecule in astrocytes. In light of these findings, identifying an extracellular ligand for β 2 /AMOG, whether on neuron surface or at the surrounding ECM, would be of great therapeutic interest. Our great challenges in this field are understanding the self-avoiding mechanism of β 2 -β 2 binding in astrocytes and identifying the heterophilic partner for β 2 /AMOG in Neurons.

Cell Culture, Transfection, and Establishment of Stable Cell Line
MDCK dog kidney epithelial cells (ATCC CCL-34) were cultured in DMEM medium (Life Technologies, Grand Island, NE, USA). CHO-K1 Chinese hamster ovary cells (ATCC CCL-61) were cultured in a mixture of F12/DMEM media. All media were supplemented with 10% fetal Calf or Bovine serum (SFB), 100 U/mL penicillin, and 100 µg/mL streptomycin under a humidified atmosphere of 5% CO 2 and 95% air at 37 • C. The cDNAs of human β 2 and dog β 1 subunits of Na + , K + -ATPase cloned into pEYFP [10] were a generous gift from Dr. Olga Vagin (Department of Physiology, School of Medicine, UCLA and Veterans Administration, Los Angeles, CA, USA). In these constructs, the YFP is fused to the N-terminal (cytoplasmic tail) of the β subunits of Na + , K + -ATPase. To establish CHO cells stably expressing YFP-β 2 or YFP-β 1 , cells were transfected over night with pEYFP-β 2 and pEYF-β 1 using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Thereafter, the transfection medium was removed, and the cells were incubated for an additional 24 h with fresh medium containing 10% SFB to allow recovery. For generating stable clones, 48 h after transfection, cells were harvested by trypsinization on 100 mm dishes in DMEM medium containing 800 µg/mL G418 (Gibco, Grand Island, NY USA). After 2-3 weeks, resistant clones were selected and re-cloned with the aid of FACS. Stable colonies were maintained in the presence of 200 µg/mL G418.
For generating His 6 tagged human β 2 subunit, the cDNA of β 2 was amplified from pEYFP-β 2 plasmid using the following primers: FORWARD (5 AGCTGAATTCGTCATCC AGAAGAGAAGAAGAGCTG-3 ) and REVERS (5 TTTTGAGCTCTCAGGTTTTG TTGAT-GCGGAGTTTGAAG GC-3 ). The cDNA was re-cloned in pcDNA4 HisMaxB vector (at EcoRI and XhoI restriction sites) generating pCDNA β 2 -His 6 vector. This vector was further transfected as described above in MDCK cells and stable clones were selected using G-418 and verified by IF and WB using anti β 2 antibody (MDCK β 2 His 6 ).

Immunofluorescence Imaging
Immunofluorescence was performed as described [7]. Briefly, cells were washed icecold with phosphate-buffered saline PBS solution (CaCl 2 , 0.90 mM; MgCl 2 anhydrous, 0.49 mM; KCl, 2.66 mM; KH 2 PO 4 , 1.47 mM; NaCl, 137.93 mM; Na 2 HPO 4 , 8.09 mM), fixed and permeabilized with methanol for 10 min at −20 • C, washed three times with PBS, then blocked with 5% bovine serum albumin BSA (Bovine Serum Albumin Cohn fraction V powder, Equitech-Bio, INC., Kerrville, TX, USA) for 1 h at RT. The cells were incubated for 1 h at 37 • C with a specified primary antibody, washed, and then incubated (30 min, RT) with a secondary antibody; anti-mouse or anti-rabbit antibody according to the primary antibody used. The secondary antibodies used in this study were conjugated to Alexa 488, Alexa 594, or Cy5. When assaying for β 2 -YFP expression, no primary or secondary antibodies were used and fluorescence was observed at 530 nm. Coverslips were washed, fixed with methanol, and mounted for observation and imaging using confocal microscopy (SP2 Leica Microsystems). Captured images were processed with ImageJ (National Institutes of Health).

Cell Aggregation Assay
Cell aggregation assay was performed as described [6]. Briefly, cells were washed with PBS, incubated with trypsin 0.05% and verseno 0.05% at 37 • C for 5 min, and dispersed by gentle pipetting. Next, 1 × 10 6 cells were resuspended in P buffer (145 mM NaCl, 10 mM HEPES, pH 7.4, 1.0 mM Na-pyruvate, 10 mM glucose, 3.0 mM CaCl 2 ) and complemented with protease inhibitor mix (GE Healthcare Bio-Science Corp., Piscataway, NJ, USA). The cell suspension was placed in 1.5-mL microfuge and rotated on a gyratory shaker at 37 • C for 3 or 15 h as indicated in each figure. Aggregation was stopped by adding 2% (vol/vol) glutaraldehyde. The extent of aggregation was assessed by flow cytometry analysis (fluorescence-activated cell sorting: FACS Vantage; BD Biosciences, San Jose, CA, USA). Then, 10,000 events were sorted for size and granularity and plotted. The dead cells population was eliminated and the proportion of big and granulated particles out of the live particles was calculated. Cell treatment with Tunicamycin previous to aggregation assay was as follows: Confluent monolayers were incubated for 24 h without or with 1 µg/mL of Tunicamycin. Afterward, cells were washed three times with PBS and incubated for one hour in a culture medium. Aggregation assay proceeded as described above.

Cell Adhesion Assay (Dispase Assay)
The Dispase-based dissociation assay reported by [29] was adapted to CHO fibroblast as described by [42]. Cell cultures were seeded onto 24-well dishes. Twenty-four hours after reaching confluency, cultures were washed three times in phosphate-buffered saline (PBS) and incubated in the same for 30 min. Thereafter, the cells were incubated in 200 µL of Dispase I (0.6 U/mL; Sigma, St. Louis, MO, USA) for 40 min. The Dispase solution was carefully removed and 500 µL of culture medium (DMEM/F12 1:1 without complement) was added. Afterward, five rounds of pipetting through a 1 mL Gilson pipette were conducted. The dissociated cells and fragments were then sampled by taking three drops of 30 µL each and depositing on a slide. Fragments were imaged using an inverted microscope (Axiovert 200 M, Carl Zeiss, Oberkochen, Germany).

Mixed Monolayers (Co-Cultures)
Cell mixture was performed as described previously [5,43]. Briefly, the cell type different from β 2 YFP transfected CHO fibroblast (CHO wt or MDCK β 2 -His 6 ) was incubated with CellTracker Orange CMTMR (Molecular Probes, Eugene, OR, USA) for 1 h on 37 • C at a final concentration of 6.0 µM. Cells were then washed three times with PBS solution, re-incubated for 1 h in DMEM supplemented with 10% FCS or FBS. The day after, cells were harvested by trypsin and the suspension was mixed with CHO β 2 YFP cells suspension, in different proportions, as indicated in each case, to be co-cultured on glass coverslips and processed the day after for immunofluorescence assay. The same procedure was implemented when the cell mixture was analyzed for cell aggregation except for when the mixing ratio was 3:1, 1:1, and 1:3. For Co-AP assay, MDCK β 2 -His 6 cells were co-cultured with CHO β 2 YFP cells or with MDCK β 2 YFP cells in a 3:1 ratio without using a CellTracker.

Pull-Down Assay
Pull-down assays were conducted in two ways: (1) MDCK cells stably expressing the human β 2 -His 6 were cultured for 48 h. Thereafter, lysed with RIPA buffer (Santa Cruz Biotechnology, Santa Cruz, CA, USA) containing protease inhibitor mix (GE Healthcare Bio-Science Corp., Piscataway, NJ, USA), and protein content was determined with BCA Protein Assay (Thermo-Scientific Meridian Rd., Rockford, IL, USA). β 2 His 6 as bait was immobilized on Ni-NTA beads (His Trap FF column, GE Healthcare, Chicago, IL, USA) previously equilibrated with 10 mL RIPA containing protease inhibitors mix. Generally, 2 mg of total cell extract protein were loaded and allowed to interact overnight, at 4 • C, with gentle shaking. After 10 washes with 10 mL of 5 mM imidazole and 10 washes with 10 mL of 10 mM imidazole the lysates of CHO β 2 -YFP and MDCK β 2 -YFP cells were loaded as prey and interaction were allowed to occur overnight at 4 • C. After washing with 5 mM and 10 mM imidazole as above, the protein complexes were eluted using a solution of 500 mM imidazole. The eluted samples were mixed with loading buffer, heated at 90 • C for 5 min, then loaded on a 10% SDS-PAGE and analyzed by Western blot using anti-YFP antibody or anti-Na + , K + -ATPase β 2 antibodies. (2) CHO YFP-β 2 monolayers were cultured for 48 h and then lysed as described above. In this method, a YFP-tagged protein is immobilized by GFP sepharose beads and acts as the bait to capture a putative binding partner (prey). Here, lysates of ARPE-19 cells and of rat brain were loaded as prey, and interaction was allowed to occur overnight at 4 • C. The protein complexes are eluted boiling with loading buffer 2× for 5 min at 90 • C and then subjected to Western blot analysis as specified above.

Co-Affinity Precipitation (Co-AP)
MDCK and CHO cell lines, both stably expressing the human β 2 subunit, were cultured as described above. Next, 24 h after seeding, both cell lines were harvested by trypsinization and seeded in a 100 mm petri dish, 30% were MDCK β 2 -His 6 and 70% CHO β 2 YFP or MDCK β 2 YFP, and cultivated in an F12/DMEM mixture supplemented with 10% FBS and 100 U/mL of penicillin, 100 ug/mL of streptomycin. After 48 h, the cells were lysed with RIPA buffer (Santa Cruz Biotechnology, Santa Cruz, CA, USA) containing protease inhibitors (GE Healthcare Bio-Science Corporation, Piscataway, NJ, USA). The protein content was determined with BCA Protein Assay (Thermo-Scientific Meridian Rd., Rockford, IL, USA) 2 mg of total proteins from cell extract was placed in a 1.5 mL tube with 100 µL of HIS-Select ® HF Nickel Affinity Gel (Sigma H0537-25ML) and were incubated overnight at 4 • C with a gentle shaking. The next day, three washes with 250 µL of cold PBS, three washes with 250 µL of 5 mM imidazole, and an additional three washes with 10 mM imidazole were carried out. Finally, the protein was eluted with 200 µL of 500 mM imidazole (stirring for 10 min at 4 • C), centrifuged, and the eluate was mixed with 2× loading buffer and heated for 5 min at 90 • C. The whole sample was run on an 10% SDS-PAGE and analyzed by Western blot using an anti-β 2 antibody.

Molecular Docking Studies
We performed docking of EDβ 2 (Chain A)-EDβ 2 (Chain B) by using the HDOCK Server [27,28]; different dimer complexes were obtained and the most energetically favorable dimer was chosen. Analysis of the interactions in the interface of the dimeric structure of EDβ 2 was obtained employing the PDBsum web server [45]. UCSF Chimera Software was employed to get the images of the monomeric and dimeric structures of EDβ 2 [46]. Ramachandran plot was obtained by using SAVES 6.v.0 Software [47].

Statistical Analysis
This was performed using a chi-square test (GraphPad Prism 4 and Microsoft Excel). Statistical significance is specified in the figure legend. Error bars indicate standard deviation.

Conclusions
β 2 /AMOG has been known as a cell adhesion molecule for more than 30 years. So far, the known interactions of β 2 /AMOG expressed in astrocytes are heterophilic cisinteractions. A trans-heterophilic interaction with a yet unknown neuronal molecule is also strongly suggested. In the present work, we show that β 2 /AMOG is a homophilic adhesion molecule when expressed on the plasma membrane of non-astrocytic cells such as fibroblasts (CHO) or epithelial cells (MDCK). Recent findings indicating the participation of β 2 /AMOG in glioblastoma pathology makes it an interesting target for therapeutic approach. Our future studies will focus on characterizing the molecular mechanism for β 2 -β 2 homophilic interaction and the role of N-glycosylation in that mechanism.