ALCAM shedding at the invasive front of the tumor is a marker of myometrial infiltration and promotes invasion in endometrioid endometrial cancer

Endometrial cancer (EC) is the sixth deadliest cancer in women. The depth of myometrial invasion is one of the most important prognostic factors, being directly associated with tumor recurrence and mortality. In this study, ALCAM, a previously described marker of EC recurrence, was studied by immunohistochemistry at the superficial and the invasive tumor areas from 116 EC patients with different degree of myometrial invasion and related to a set of relevant epithelial and mesenchymal markers. ALCAM expression presented a heterogeneous functionality depending on its localization, it correlated with epithelial markers (E-cadherin/β-catenin) at the superficial area, and with mesenchymal markers at the invasive front (COX-2, SNAIL, ETV5, and MMP-9). At the invasive front, ALCAM-negativity was an independent marker of myometrial invasion. This negativity, together with an increase of soluble ALCAM in uterine aspirates from patients with an invasive EC, and its positive correlation with MMP-9 levels, suggested that ALCAM shedding by MMP-9 occurs at the invasive front. In vivo and in vitro models of invasive EC were generated by ETV5-overexpression. In those, we demonstrated that ALCAM shedding was related to a more invasive pattern and that full-ALCAM recovery reverted most of the ETV5-cells mesenchymal abilities, partially through a p-ERK dependent-manner.


INTRODUCTION
Endometrial cancer (EC) is the most common gynecologic cancer and the sixth most deathly cancer in women from western countries [1]. Myometrial invasion is one of the most important prognostic factors of EC and represents an increase in the rate of recurrence and a decrease in the 5-year survival [2]. The first step of myometrial invasion is characterized by the dissociation of tumor cells from the epithelial layer of the endometrial glands, and their subsequent infiltration through the basement membrane into the adjacent layer, the myometrium [3]. For that to happen, several authors suggest that epithelial tumor cells need to undergo an epithelial-to-mesenchymal transition (EMT). Moreover, it is known that in invading tumors this process takes place at the invasive front [4][5][6]. EMT is a dynamic process controlled by external signals coming from the tumor cells microenvironment, that involves cellular loss of polarity, loosen cell-cell contacts, reorganization of cytoskeleton and acquisition of plasticity, motility and invasive capabilities. Multiple factors have been described as being responsible for producing EMT in EC, such as ETV5, estrogen and progesterone receptors, and TGF-β, among others [3].
Activated-leukocyte cell adhesion molecule (ALCAM) is a cell-cell adhesion molecule of 105-kDa from the immunoglobulin superfamily (IgSF) that participates either in homotypic (ALCAM-ALCAM) or heterotypic (ALCAM-CD6) interactions between adjacent cells [7]. ALCAM presents 5 extracellular immunoglobulin domains (D1-D5), comprising 2 NH2-terminal, membrane-distal variable-(V)-type and 3 membrane-proximal constant-(C2)type Ig folds, a transmembrane-region and a short carboxyterminal cytoplasmic tail. The extracellular domains provide two structural and functional modules, responsible for ligand affinity (D1-D2) that mediates receptor transtrans-interactions between opposing cells, and avidity (D3-D5). The strength of the avidity is controlled by recruitment of ALCAM molecules to the site of cell-cell contact and by receptor cis-oligomerization via the C-type immunoglobulin domains. Coordination of both modules is necessary for stable ALCAM-ALCAM interaction, leading to the formation of a tight bilayered ALCAM network [8]. Both, homotypic and heterotypic interactions are mediated through its amino-terminal V domain (D1). ALCAM-ALCAM interactions are described as a D1-D1 interaction, while in ALCAM-CD6 interactions the ALCAM D1 domain binds to the CD6 membrane-proximal scavenger receptor cysteine rich (SRCR) domain [9,10].
Activation of ALCAM-mediated adhesion is dynamically regulated through actin cytoskeletondependent clustering. Zimmerman et al. described the interaction between ALCAM and the actin cytoskeleton [11]. This interaction could be allowed by the presence in its short cytoplasmic tail of a positive-charge-rich domain (PCRD) as well as a PDZ-binding motif KTEA placed at the C-terminus [12]. The inhibition of actin polymerization by low levels of cytochalasin D stimulates homotypic ALCAM-ALCAM interactions, increases protein lateral mobility, and the formation of ALCAM clusters in the cell surface, demonstrating that cytoskeleton regulates ALCAM-mediated adhesion [13]. Moreover, ALCAM has been observed to form a supra-molecular complex with Syntenin-1 and Ezrin [14]. PKCα also plays a role in the cytoskeleton-dependent avidity modulation of ALCAM and regulates the supra-molecular complex formed by ALCAM-Ezrin [15].
ALCAM expression in the cell surface can be regulated by endocytosis or protein cleavage. In fact, ALCAM co-localizes with clathrin in a process that is regulated by ubiquitination and dependent on ERK phosphorylation [16,17]. ALCAM levels can be also regulated by protein shedding via ADAM-17 metalloproteinase [18], generating a soluble 96-kDa form and a truncated membrane-bound. ALCAM is widely distributed within tissues and its function has been reported in extensive biological processes [19]. In the last years, it has been associated to the tumorigenesis of many cancers [20][21][22][23][24][25][26][27]. However, ALCAM expression and localization within the tumor cell have generated great controversy, and its relation to prognosis remain still unclear [28]. In addition, ALCAM shedding has been reported in the serum of breast, ovarian, thyroid and pancreatic cancer patients and in all the cases, related to poor clinical outcome [24,[29][30][31]. We have previously demonstrated that ALCAM is an important player in endometrioid endometrial cancer (EEC) dissemination [32]. In fact, we described that ALCAM-positivity is a marker of recurrence in early stage EEC and promotes migration and invasion in vitro and in vivo.
In this study, we go further in our understanding of ALCAM in EC tumorigenesis by evaluating its fullform expression in two different areas of the tumor and its relation with key molecules associated to the epithelial and mesenchymal context. Immunohistochemical staining www.oncotarget.com for ALCAM in tissues was conducted with an antibody directed against the extracellular domain of the protein. Since, ALCAM can be proteolytically processed by  metalloproteinases, thereby generating a soluble ALCAM  component and a truncated membrane-bound ALCAM  containing the transmembrane and cytoplasmic domain,  the antibody recognized the non-cleaved protein. We found a switch in the functional mode of ALCAM in EEC. Whereas at the superficial area of the tumor ALCAM correlated with the E-Cadherin/β-Catenin adhesion complex, at the invasive front ALCAM could be shed by MMP-9 in poor prognosis scenarios. Moreover, we evaluated the shed ALCAM, or soluble form (sALCAM), in uterine aspirates as a potential diagnostic tool to predict myometrial invasion.

ALCAM expression presented a different correlation profile depending on its localization within the tumor
We studied the immunohistochemical expression of ALCAM extracellular domain (non-cleaved protein) and a representative set of epithelial (E-cadherin/β-catenin major adhesion complex) and mesenchymal molecules involved in EEC dissemination (ETV5, COX-2, SNAIL, SLUG, MMP-2 and -9), in both the superficial and the invasive areas of the tumor. As our previous results indicated a different behavior of ALCAM regarding tumor differentiation, analyses were performed in the entire population, and in the well-differentiated (GI) and poorlydifferentiated (GII-III) subcohorts [32].
Univariate regression analyses were performed as a first approach to evaluate the association between ALCAM and epithelial and mesenchymal markers. Variables that significantly correlated with ALCAM (Supplementary Table 1) were included in multivariate linear regression models. ALCAM presented a significant positive correlation with the major adhesion complex at the superficial area of the tumor. Whilst β-catenin was retained as an outcome variable in all patients and in the well-differentiated cohort (R=0.437 and 0.394, respectively), E-cadherin was retained in the moderatelypoorly differentiated tumors (R=0.445) (Supplementary Table 2, Figure 1A). By contrast, at the invasive front of the tumor, SNAIL, COX-2 and MMP-9 were retained as outcome variables of the model, in all patients. COX-2 and SNAIL were positively correlated with ALCAM, whilst MMP-9 was negatively correlated. The best regression adjustment was reached in the moderately-poorly differentiated subcohort of patients, in which the model retained ETV5 and MMP-9 (R=0.559). Both molecules showed a negative correlation with ALCAM. Altogether, the results suggested a different functional profile of ALCAM according to its tumor localization, i.e. whilst it correlated with epithelial markers at the superficial tumor, correlations with mesenchymal markers were observed at the invasive front.

ALCAM-negativity at the invasive front of the tumor is a marker of myometrial invasion
We further investigated the role of ALCAM at the different tumor areas. At the superficial area, ALCAM expression was unrelated to the clinical parameters of myometrial invasion, tumor grade, and tumor progression. However, it acted as a marker of tumor progression and invasion when analyzed at the invasive front. In this tumor area, ALCAM significantly decreased along tumor stages and in tumors with myometrial invasion >50% (Table 1, Figure 1B-1C). In fact, the potential of ALCAM-negativity as an independent prognostic factor of myometrial invasion was confirmed by multivariate logistic regression analysis with an OR 3.273, together with the tumor grade (OR 3.484) ( Table 2).
ALCAM maintained a close and inverse relation with MMP-9 at the invasive front of the tumor (p=0.004; OR 0.26) ( Table 3). This relation was only preserved in poor prognosis tumors when the cohort was divided according to clinical parameters, i.e. either in tumors with moderately-poorly differentiated EEC histology (p=0.02; OR 0.08), in patients presenting >50% of myometrial infiltration (p=0.016; OR 0.19), and in ETV5-positive tumors (p=0.009; OR 0.22). Altogether, our results demonstrated that ALCAM-negativity is a marker for myometrial invasion at the invasive front. Its close and inverse relation with MMP-9 in poor prognosis scenarios, led us speculate that ALCAM-negativity might be due to an increased shedding of the protein by the metalloproteinase at the invasive front of poor prognostic tumors.

ALCAM shedding takes place in tumors with myometrial invasion and correlates with MMP-9 expression
In order to evaluate the possible association between ALCAM shedding and myometrial invasion, we analyzed the expression of soluble ALCAM (sALCAM) by ELISA, in 40 uterine aspirates from moderatelypoorly differentiated EEC patients with different depth of myometrial infiltration. The uterine aspirate is a bodyfluid that is collected from inside of the uterine cavity, and is mainly formed by the secretion of the tumor cells and other cells from the endometrium, enabling the detection of cleaved ALCAM produced by the tumor. We confirmed that sALCAM was significantly increased in patients presenting myometrial invasion >50% ( Figure  2A). Moreover, ROC analysis showed that sALCAM in uterine aspirates is a significant predictor of myometrial infiltration (AUC 0.8; p=0.001) ( Figure 2B). In fact, when www.oncotarget.com ALCAM is set at a cut-off of 9.375 ng/mg, the model presents a sensitivity of 87% and a specificity of 70.6%.
Then, the relation between sALCAM and MMP-9 was assessed in uterine aspirates by performing a gelatin zymography of the same patients ( Figure 2C). Linear regression analyses (Supplementary Table 3) evidenced a significant positive correlation between sALCAM with all MMP-9 detected forms, with the exception of the more labile active form. No correlation was found with the MMP-2 forms (data not shown). In order to avoid the problems derived from the simultaneous use of several measurements that reflect expressions of molecular forms of the same enzyme (Supplementary Table 3), PCA was addressed to explore the close relationship among the four MMP-9 forms and reduce their number before performing regression analysis. Two main components C1, which explains 75.18% of the variability of the original dataset, and C2 which only explains 13.86%, were extracted ( Figure 2D). Regression analyses showed a strong and positive correlation between sALCAM and C1, and specifically in patients presenting myometrial invasion >50% (R=0.575; p=0.004) ( Figure 2E, 2F). These results suggested that at the invasive front of the tumor, ALCAM extracellular shedding by MMP-9 could be an important process in myometrial invasion.

ALCAM and MMP-9 are important actors at the invasive front of an in vivo model of EEC dissemination
The Hec1A-ETV5 overexpression model has been extensively used to mimic the process of tumor invasion in EEC [33][34][35], as ETV5 overexpression is known to induce EMT in EEC and has been reported at the invasive front of EEC tumors, promoting migration and invasion in vitro and in vivo. Furthermore, the parental Hec1A cell line is representative of moderately-poorly differentiated EEC tumor with epithelial-like morphology [33,36]. To study ALCAM in a controlled environment, which represents a realistic approach towards the process of dissemination of EC, we generated orthotopic murine models with Hec1A (n=5) or its ETV5-overexpressing cells (n=4), directly injected into the uterus of the mice. In this model, tumor cells are localized in the same microenvironment of the original tumor, resembling the processes of tumor growth and myometrial infiltration under endometrial stimuli.
As expected, the tumors generated by the ETV5overexpressing cells presented larger myometrial invasion and tumor burden (Supplementary Figure 1A-1B).
While in the ETV5-overexpressing tumors, we observed disseminated cells in finger-like strands or single-cells, projecting into the stroma at the invasive front; in the Hec1A tumors, cells invaded forming clusters (see arrows in Figure 3A), as previously observed by Muinelo-Romay et al. in vitro [37]. The immunohistochemical staining of uncleaved ALCAM, MMP-9, and ETV5 in the mice's primary tumors unveiled that ALCAM expression is reduced at the invasive front of ETV5-overexpressing compared to Hec1A tumors (p<0.05; Figure 3B), and moreover, the pattern of expression is modified from a very membranous staining in Hec1A tumors to a diffuse cytoplasmic staining in ETV5-overexpressing cells ( Figure 3A). Interestingly, the intensity of ALCAM and MMP-9 showed no variation between the superficial or the invasive area in Hec1A tumors, but their intensity was altered in the ETV5-overexpressing tumors. In those, MMP-9 increased concomitant with a decrease in ALCAM from the superficial area to the invasive front of the tumor (Figure 3, Supplementary Figure 1C).
As a result of the in vivo model, we evidenced that the cell-cell contacts of the Hec1A invading cells seem to be preserved, as shown by the highly collective migration and a homogeneous ALCAM expression. However, the ETV5-overexpressing invading cells were more prone to present switching between thin cords and single-cells, both related to decreased or more transient contacts and higher rate of cleaved ALCAM expression. We finally confirmed that in an invasive scenario, ALCAM and MMP-9 are important actors at the invasive front of the tumor.

Recovery of ALCAM expression reduced the aggressiveness of invasive EEC cells
To understand the role of ALCAM at the invasive process, we recovered its expression in the invasive Hec1A-ETV5 model by using two ALCAM-overexpression vectors: one containing a full-length ALCAM (full-ALCAM) and another containing the extracellular and transmembrane regions (ALCAMcytoless), both cloned in pmCherry-N1 vectors ( Figure 4A). When transiently transfected in Hec1A-ETV5 cells, ALCAM-overexpression was localized predominantly in the plasma membrane ( Figure 4B-4C).
To evaluate the effects of ALCAM-recovery in ETV5overexpressing cells, we used 3D in vitro approaches that closely mimic the in vivo settings. We used a spheroid model to quantitatively study the spreading of cell aggregates on fibronectin-coated stripes. In addition to reproduce characteristics of the in vivo environment, this model allows us to analyze the competition between cell-cell and cell-substratum adhesion on tissue spreading [38]. In both conditions (ALCAM full-length and ALCAMcytoless) the speed was significantly decreased compared to the control cells ( Figure 4D-4E). However, the larger difference was found in the full-length ALCAMtransfected cells. In fact, we observed that the mean time of disaggregation of the Cherry-control spheroids was around 5 h, while ALCAM full-length and ALCAMcytoless expressing aggregates needed approximately 21 h and 18 h, respectively ( Figure 4D). Moreover, the pattern of spreading in control cells presented a more spindleshaped phenotype, with larger presence of individual cells and protrusions, highly dynamic transient cell-cell contacts, migration in a sheet fashion and presentation of chemotactic abilities. In contrast, full-ALCAM and ALCAMcytoless cells presented an increased intercellular adhesion phenotype and a more collective pattern of spreading ( Figure 4D, Supplementary Videos 1-3). Consequently, ALCAM recovery clearly impaired the migratory abilities of ETV5-overexpressing cells.
To verify these effects we evaluated how ALCAMrescue affects the formation and strength of cell-cell adhesion. We used two micropipettes to put into contact two isolated cells maintained in suspension, to avoid cellmatrix interactions, and initiate adhesion. Single cells were chosen under a fluorescence microscope to verify the Cherry-transfection. The mean separation force (SF) is used as a read-out of the strength of adhesion for a precise contact time. In all the cases, the SF required to disrupt the cell doublet increased with the contact time. We observed that at 4 min both ALCAM-transfected cells presented significantly larger adhesion than the control line (Figure . Cells expressing full-length ALCAM displayed stronger intercellular adhesion, as shown by the two-fold increase in their SF compared to control cells, and were more adhesive than cells expressing ALCAMcytoless. In contrast, at 10 min only the full-length ALCAMtransfected cells presented significantly stronger adhesion than the control. Thus, we showed that ALCAM shedding was important to reduce the adhesive properties of the cells. In this particular context, the cytoplasmic tail plays an important role in cell-cell adhesion formation or maturation. The molecular pathway by which ETV5 promotes its invasive functions in EC has been recently postulated by Alonso-Alconada et al. [39]. The authors described how ETV5 works through the BDNF-TrkB-ERK1/2 axis by inducing ERK1/2 phosphorylation, leading to the promotion of migration and invasion. In this work, we observed that ALCAM-overexpression was concomitant with a decrease in p-ERK, without reverting to an epithelial phenotype ( Figure 4G), but impairing the migratory and increasing the adhesive properties promoted by ETV5. Although overexpression of the ALCAMcytoless is sufficient to reduce the levels of p-ERK and to decrease spheroid spreading, ALCAM fulllength is necessary to increase cell-cell adhesion with time.

DISCUSSION
The principal components of adherens junctions in epithelial cells are E-cadherin and β-catenin. E-cadherin is a calcium-dependent transmembrane receptor that interacts homotypically to the E-cadherin of adjacent cells and heterotypically to cytoplasmic-binding proteins, including β-catenin, to transduce external signals within the cell and provide a physical link to the cytoskeleton [40]. Loss of epithelial polarity and decreased E-cadherin lead to EMT, they have been described to promote invasion and are related to poor prognosis and highgrade EC tumors [41]. In addition, the upregulation of the ETV5 Ets transcription factor has been linked to EMT and associated with myometrial infiltration in EC [42]. Our previous in vitro and in vivo studies evidenced that ETV5 promotes EMT by modulation of ZEB1 expression and consequent E-cadherin repression, leading to a complete reorganization of cell-cell and cell-substrate contacts and in the acquisition of migratory and invasive capabilities [33,34]. Moreover, ETV5 has been shown to co-localize with MMP-2 and -9 at the invasive front of the tumor [42]. The high expression of both metalloproteinase and COX-2, another direct target of ETV5, has been correlated to poor prognosis and relapse in EC [43,44].
A critical role has been evidenced in the process of EMT and E-cadherin expression downregulation, by SNAIL-related zinc-finger transcription factors (SNAIL and SLUG). The overexpression of SNAIL and SLUG has been seen closely associated with the reduced expression of E-cadherin in EC, and particularly at the invasive front of the tumor [45,46]. Specifically, SNAIL nuclear immunoexpression has been described to take place concomitantly with a decrease in E-cadherin expression at the invasive front of stage IB EEC [45].
In this study, we used an extracellular-ALCAM antibody to assess the non-cleaved protein expression by immunohistochemistry in two different tumor areas from 116 EEC primary tumors, i.e. at the superficial and the invasive front. Then, its expression was related to a set of epithelial (E-cadherin, β-catenin) and mesenchymal (ETV5, COX-2, MMP-2, MMP-9, SNAIL, and SLUG) EEC relevant markers. While at the superficial tumor, uncleaved ALCAM maintained a correlation with the major adhesion complex E-Cadherin/β-catenin; at the invasive tumor, where a more mesenchymal microenvironment takes place, it correlated with COX-2, SNAIL, MMP-9 and ETV5 mesenchymal markers. Interestingly, in the moderately-poorly differentiated tumors E-cadherin and ALCAM were positively correlated at the superficial area, whilst ETV5 and MMP-9 presented a significant negative correlation with extracellular ALCAM at the invasive front. Moreover, we evidenced that the correlation between ALCAM and MMP-9 was only significant in poor prognostic scenarios, i.e. in moderately-poorly differentiated tumors, in patients presenting myometrial invasion >50%, and ETV5positive tumors. Furthermore, at the invasive front, we demonstrated that ALCAM-negativity was an independent prognostic marker of myometrial invasion together with the tumor grade.
ALCAM-negativity was directly associated to an increased shedding of the protein, as demonstrated by the increased expression of sALCAM in uterine aspirates from patients with moderately-poorly differentiated EEC presenting myometrial invasion >50%. In our study, ROC analysis corroborated sALCAM as a significant predictor of myometrial invasion (AUC=0.80). These findings are particularly interesting as uterine aspirates are the preferable biopsy used in EC diagnosis and the assessment of sALCAM might open the avenue to develop a potential tool to determine pre-operatively the myometrial infiltration (sensitivity 87.0%, specificity 70.6%) at the time of diagnosis, even in earlier EC stages [47]. In addition, sALCAM and MMP-9 were significantly strongly correlated in the same patients presenting myometrial invasion >50%. Actually, sALCAM has been detected in FFPE [22] and in several biofluids [31,48,49], and related to tumor progression and poor patient outcome. Disruption of ALCAM-ALCAM interactions between cells promotes tumor cell motility and metastasis, then shedding may predict tumor progression at a molecular level. In fact, ectopic expression of mutant ΔN-ALCAM in melanoma cells induced loss of cellular aggregation, increased motility in skin reconstructs and lung metastasis formation at the expense of primary tumor [50].
In vivo and in vitro ETV5-overexpressing models of EEC were selected. Only in the ETV5-overexpressing invading tumors, uncleaved ALCAM was decreased at the invasive front, concomitant with an increase in MMP-9 expression. MMP-9 expression has been associated with increased metastatic potential in many cancer types [51]. Its activity modulates cell-cell and cell-matrix attachment and has been described to regulate the cell invasion and migration processes. Here, we could hypothesize that MMP-9 is responsible of ALCAM shedding, in invasive scenarios.
Moreover, we proved that ALCAM-recovery in ETV5-overexpressing cells decreased migration, increased cell-cell adhesion, and presented a more collective pattern of spreading. In addition, we evidenced that a larger adhesion force was necessary to separate the full-length ALCAM-transfected cell doublet, indicating its capacity to promote intercellular adhesion strengthening and manifesting the crucial role of its cytoplasmic tail in the maturations of cell-cell adhesion. In fact, the cytoplasmic tail of ALCAM has been described to interact with the actin cytoskeleton, and could be essential in translation of external mechanical forces into the cell to activate biochemical signals [50,52].
These differences in cell-cell adhesion, in the cell speed and the pattern of migration could be partially www.oncotarget.com explained by the decrease of the active levels of p-ERK. The link between the role of BDNF in the modulation of EMT induced by ETV5 and the acquisition of cell migratory/invasive capabilities through p-ERK phosphorylation has been established [39]. The known role of the BDNF-TrkB-ERK1/2 axis on neuronal plasticity and organogenesis might be mediating pro-metastatic events in endometrial cancer and as consequence the decrease in the p-ERK levels would lead to a less aggressive phenotype [39]. Moreover, some authors evidenced that sustained MAPK activation could lead to enhanced induction of proteolytic enzymes in the surrounding environment in tumors and specifically, the inhibition of ERK activation by MEK inhibitor PD098059 has been described to partially block MMP-9 production and attenuate the in vivo invasiveness of head and neck squamous cancer cells [53].
Although further research is needed to define the exact mechanisms, in vitro and in vivo models evidenced that ALCAM participates in cell-cell adhesion and collective cell migration, and suggested that to allow invasion in the frontier between tissues-restricted carcinoma and disseminated tumor cells, a dynamic and adaptive switch between ALCAM expression at the cell surface and cleaved ALCAM by MMP-9 might take place. In fact, cancer cell invasion is currently viewed as a heterogeneous and adaptive process [54] in which, the plasticity in cell adhesion, the dynamics of the cytoskeleton and the mechanotransduction of external stimuli are key processes. Taking into account that myometrial invasion is highly associated with poor prognosis and to a limited therapeutic response, our research opens an avenue for the use of ALCAM-recovery as a therapeutic approach to control tumor invasion.

Human endometrial cancer samples
Formalin-fixed paraffin-embedded tissue samples from 116 EEC patients were recruited from the Hospital del Mar (Barcelona, Spain). For each patient, we selected 4 macroscopically differentiated areas: 2 superficial, which were obtained from the core of tumor; and 2 invasive, which were obtained from the tumor in contact with the invasive front.
Forty uterine aspirates, from EEC patients were collected prior to surgery by aspiration in the Vall Hebron University Hospital (Barcelona, Spain), with a Cornier Pipelle (Eurogine Ref. 03040200). PBS1x was added (1:1) and centrifuged 20 min at 2,500 rcf to separate the soluble fraction.
All samples were obtained after the approval of the ethical committee of each institution. The clinicopathologic characteristics of all patients are detailed in Supplementary Table 4.

Immunohistochemical staining
Tissue microarray and immunohistochemistry were performed as previously described [34]. ALCAM, E-cadherin, β-catenin, ETV5, COX-2, MMP-2, MMP-9, SLUG, and SNAIL immunohistochemical expressions were analyzed in both areas of the tumor (superficial and invasive). Immunohistochemical staining for ALCAM was conducted with MOG/07 antibody (ab49496; Abcam, Cambridge, MA, USA) directed against the extracellular domain of the protein. Specifically, the antibody is raised against a 200-amino-acid-long peptide from the extracellular region of ALCAM, detecting the full-ALCAM or non-cleaved protein. Antigen retrieval and primary antibody incubation were performed as previously described [32]. To detect MMP-2, MMP-9, and ETV5, citrate buffer pH 9 was used for antigen retrieval.

Soluble ALCAM detection in uterine aspirates
The ALCAM Duoset ELISA kit (R&D Systems, Minneapolis, USA) was used to detect ALCAM levels in the supernatant of the uterine aspirates according to manufacturers' instructions. All samples were diluted 1:100 in BSA 1%.

Zymography
Proteins with metalloproteinase activity were identified by gelatin zymography and equal quantities of protein, from the uterine aspirates, were size fractionated as previously described [35].

Orthotopic murine model and bioluminescent imaging
All the procedures were performed according to the guidelines of the Spanish Council for Animal Care and the institutional guidelines for animal welfare (CEEA 23/16). A total of 9 six-week-old female athymic nude mice (Charles River Laboratories, Inc, Wilmington, MA) were inoculated by transmyometrial injection of Hec1A luciferase control cells (n=5) and stably transfected Hec1A-ETV5 luciferase cells (n=4) as described in [56]. Mice were sacrificed 3 weeks after the injection. All extracted tissues were formalin-fixed, stained with haematoxylin and eosin, and evaluated by a pathologist. After cells injection and before sacrifice, a weekly followup was carried out using the IVIS system (Xenogen Corporation) coupled to Living Imaging software 4.2 (Xenogen Corporation) in order to detect tumor growth by bioluminescent imaging. Luciferin (Firefly Luciferin, Caliper Lifescience Corp, Hopkinton, MA, USA) was used as the substrate for the luciferase expressing tumor cells and injected intraperitoneally at a concentration of 150 mg/kg in PBS.

Measurement of the separation force between cells
Forces were measured on the stage of an inverted epifluorescence microscope (Leica) equipped with objectives of 63x (PL FLUOTAR; NA/0.7; C PLAN NA/0.75) and with a cooled CCD C5985 (Hamamatsu) or Coolpix 5000 camera (Nikon) as previously described [58]. To preserve intact the cell surface proteins, cells were dissociated with Cell dissociation enzyme-free buffer (Gibco, NY, USA) and then transferred in working medium (CO2-independent medium, Invitrogen, CA, USA) and used immediately. Two isolated cells were put into contact and allow forming adhesion for specific period times. Aspiration was monitored continuously by a pressure sensor (model DP103-38; Validyne) during the separation process and the values recorded for each of the last two cycles in the series (Pn-1 and Pn) were used to calculate the separation force (SF) for each doublet using the equation: SF = π (d/2)2 (Pn-1+Pn)/2 where d is the inside diameter of the left pipette. Results for 15-30 measurements were used to obtain the mean SF for a specific contact time.

Aggregate spreading
Aggregates were obtained from 5 mL of cell suspension in CO2-equilibrated culture medium at a concentration of 4 × 10 5 cells per milliliters in 25 mL Erlenmeyer flasks and placed in a gyratory orbital shaker at 75 rpm at 37°C for 22 h. The flasks were pretreated with 2% dimethylchlorosilane in chloroform to prevent adhesion of cells to the glass surface.
For the preparation of coated glass substrates: twentyfive millimeter circular glass coverslips were sonicated in ethanol for 5 min, dried at ambient temperature, and exposed to deep UV for 5 min. Fibronectin (Sigma-Aldrich, Missouri, USA) coating was performed using a 0. As previously described [59], aggregates were deposited randomly on striped fibronectin-coated coverslips in a magnetic imaging chamber (Chamlide CMB, CM-B25-1) filled with CO2-equilibrated culture medium. Spreading was observed using a NIKON confocal microscope with an x10 air objective and a 37°C heating cube system. Bright field images were recorded for 36 h with a CCD camera (Luca-R, Andor) using NIS-Elements software. The increased spreading was measured at different time points with Image J software (National Institutes of Health, Bethesda), which allowed calculating the speed of migration for each cell line.

Immunofluorescence
After 48 h of transfection, cells seeded onto glass coverslips were fixed with 4% paraformaldehyde, blocked in PBS-BSA 4% and incubated with DAPI for 15 min at RT in dark. Coverslips were mounted using the Aqua/Poly Mount medium (Polysciences Europe GmbH, Eppelheim, Germany). Fluorescence images were captured with Spectral Confocal Microscope FV1000 (Olympus, Hamburg, Germany).

Statistical analysis
For continuous variables: univariate and multivariate linear regression analyses were performed to evaluate the association between ALCAM and epithelial or mesenchymal markers at the superficial and the invasive front. For each cohort, a stepwise method was used to select the explanatory variables based on analysis of variance, checking the normality (Kolmogorov-Smirnov) and the independence of residuals (Durbin-Watson).
Receiver operating characteristic (ROC) curve analysis was conducted on the individual soluble ALCAM (sALCAM) expression detected in uterine aspirates at diagnosis, to evaluate the potential of this parameter as a marker to discriminate patients presenting myometrial infiltration. To explore its relationship with the different forms of MMP-9 present in the uterine aspirates, we performed a factor analysis with extraction of principal component (PCA) and then, a regression analyses with sALCAM levels. The normal distribution of the data was verified (Kolmogorov-Smirnov) and levels were ln-transformed for linear analyses. Non-parametric test (Mann Whitney) were used for comparisons of data sets whose distributions deviate from normal. Measurements were made in triplicates in three independent experiments and presented with the Mean±SD (except for the separation force experiments, which were presented by Mean±SEM).
For categorical variables: the relationships between ALCAM expression at the superficial and the invasive front and clinicopathological parameters were tested in univariate analysis by Chisquare and Fisher's exact test. Stratified analysis was performed to study the molecular features ALCAM/MMP-9/ETV5 and the crosstalk between ALCAM/MMP-9 and clinicopathological parameters. The odds ratios (OR), with their confidence interval (IC 95%) were obtained through bootstrapping methods. Logistic regression multivariate analysis was done for the clinicopathological parameters and the biomarkers status to determine independent predictors for myometrial infiltration.
All statistical analyses were performed using the IBM SPSS Statistics 21. All two-sided p-values < 0.05 were considered statistically significant.

Author contributions
LD participated in the study design, carried out most of the experimental work and data analysis, and wrote the paper. EMG, CPM, MTQ, MAA, TVS, LAA, and WT performed experiments and their corresponding analysis, and were also involved in revising the manuscript. FBW, SD and WT designed part of the study and revised the manuscript. FA, BDF, GM, MA and AG were involved in the study design and data collection for the multicenter study biomarker research. JR, AGM, and EC conceived and designed the study, and revised the manuscript. EC also participated in data interpretation and literature search.