Cell lines were obtained from the American Type Culture Collection (Manassas, VA, United States). All culture media and reagents were from Gibco (Invitrogen, Carlsbad, CA, United States). BEAS-2B human lung cells and SKBR3 human mammary cells were cultured in DMEM containing 10% fetal bovine serum and 1% penicillin-streptomycin. THP-1 monocytic human cells were grown in RPMI medium supplemented with 20% FCS, and 1% penicillin-streptomycin.

The expression vector that encodes wild-type ZO-1 (ZO-1) has been previously described (Reichert et al., 2000). The plasmid was transfected using the X-tremeGENE nine DNA reagent (Roche Diagnostics, Mannheim, Germany) on 1 × 10⁵ cells plated in 6-well plates according to manufacturer instructions. Cells were harvested 48 h later for quantitative RT-PCR and western blotting analyses. Serum-free conditioned media were collected for in vivo sponge assay and cytokine array analyses.

The Proteome Profiler Human XL Cytokine Array kit (ARY022B, R&D) was used according to the manufacturer’s instructions. Briefly, conditioned media from BEAS-2B cells transfected with the ZO-1 expression vector or the empty vector was diluted with a mixture of biotinylated antibodies. Then, the mix was incubated overnight on a nitrocellulose membrane on which capture antibodies for 102 soluble factors had been coated in duplicate. After washing steps and incubation with horseradish peroxidase-coupled streptavidin, chemiluminescent detection was performed using ECL Prime (Pierce). The intensity of each spot was measured using MultiGauge (V3.0; Fujifilm; Tokyo; Japan). Each spot corresponding to a chemokine was quantified and normalized with the intensity of positive and negative control spots on the membrane. Data are expressed as fold induction for each chemokine in the ZO-1 transfected condition versus the pLNCX control condition.

Transmigration assays were performed as previously described (Brysse et al., 2012). THP-1 cells (10⁵) were suspended in 200 µl of serum-free DMEM containing 0.5% bovine serum albumin (BSA) and placed in the upper compartment of a transwell (6.5-mm diameter with 8 µm pore polycarbonate membrane insert; Costar). The lower compartment was filled with 600 µl of 48 h conditioned serum-free DMEM medium of BEAS-2B cells transfected or not with ZO-1 cDNA expression vector. After 24 h of incubation at 37°C, migrated THP-1 cells in the lower chamber of the transwell were counted with an ADAM automated cell counter. Results are expressed as fold induction to the empty expression vector control condition. Each experiment was carried out at least 3 times.

Sponge assays were performed as previously described (Van de Velde et al., 2018) with the approval of the ethical committee of the University of Liège (Approval No. 1599). Briefly, gelfoam sponges (Pfizer, New York, United States) were soaked in conditioned medium. Sponges were then subcutaneously inserted into the ears of BALB/c mice (Charles Rivers, Chatillon-sur-Chalarone, France) for 3 weeks. The experiment was repeated 3 times with five mice per group. At the end of the experiments, ears were collected, fixed in formol, and paraffin-embedded. Five micron-thick sections were realized across the sponge before immunostaining with a rabbit monoclonal anti-CD3 antibody (1:500; CD3; SP7; Ab16669; Abcam). Slides were scanned using the VS120 OLYMPUS (Olympus France SAS, Rungis, France) at a ×20 magnification. Image processing and quantification of cell density were performed using the images analysis toolbox of MATLAB 2019/B, according to the following steps: original images were registered in the full-color red, green, blue where sponge and cells appear in blue and red respectively. Classical and morphological filters were applied to eliminate the noise and enhance the contrast between the cells and the sponge. Finally, images were binarized using an automatic thresholding technique (Kapur et al., 1985). From binarized images, cells density was defined as the number of pixels belonging to all cells (total area occupied by cells) divided by the number of pixels belonging to the associated mask (filled area of the considered region of the sponge). The density values of labeled cells relative to the sponge area in the ZO1 groups were normalized to the mean density value of the control group.

Human tissue samples were obtained from 42 patients with NSCLC [21 adenocarcinomas (ADC) and 21 squamous cell carcinomas (SCC)] (Supplementary Table S1). The tumors were staged according to the eighth TNM UICC/AJCC edition. Clinical data such as age and gender were collected retrospectively. Access to patient data for this retrospective non-interventional study was approved by the French national commission CNIL (Comité National de l’Information et des Libertés) (NO.2049775 v0). Paraffin-embedded tumor samples were obtained from the Tumor Bank of Reims University Hospital Biological Ressource Collection (No. AC-2019-340) declared at the Ministry of Health according to the French Law, for use of tissue samples for research.

Immunohistochemistry for ZO-1, CD3, CD4, CD8, and Foxp3 was performed on serial sections of the 42 paraffin-embedded NSCLC tumor samples. ZO-1 and Foxp3 detection were performed as previously described (Lesage et al., 2017). CD3, CD4 and CD8 detection was performed with a Ventana Benchmark XT (Roche diagnostics GmbH). Each stage of the experiment was performed automatically according to the manufacturer’s instructions and driven by Ventana Medical Systems software, BenchMark XT module IHC/ISH (Ventana Medical Systems, Roche). Antibodies used for immunohistochemistry are listed in the supplementary material (Supplementary Table S2). A blind evaluation of the labeling and scoring was performed by two independent pathologists. Membrane-associated ZO-1 (tumor areas displaying a honeycomb ZO-1 staining), and cyto-nuclear ZO-1 staining, were scored as follows: 0 = no detection, 1 = detection in <10% of tumor cells, 2 = detection in 10–25% of tumor cells, 3 = detection in 26–50% of tumor cells, 4 = detection in >50% of tumor cells. From this scoring, cancers were categorized using a cutoff at 10%, as previously reported (Lesage et al., 2017), into a group combining cancers with scores 0 and 1 (<10%) and a group combining cancers with scores 2 to 4 (≥10%). CD3, CD4, CD8, and Foxp3 staining were scored as follows: 0 = no detection, 1 = detection in <10% of cells, 2 = detection in 10–25% of cells, 3 = detection in 26–50% of cells, 4 = detection in >50% of cells. Cancers were grouped in two categories: “low” expression group = cancers with scores from 0 to 2, “high” expression group = cancers with scores from 3 to 4.

Statistical analyses were performed with Prism (GraphPad Software, La Jolla, CA, United States). In vitro results expressed as fold induction were analyzed using the two-tailed one-sample Student’s t-test. In vivo results were analyzed using the two-tailed non-parametric Mann Whitney test. For human immunohistochemistry, the association between ZO-1 and CD3, CD4, CD8, and Foxp3 expression in NSCLC was studied by using Fisher’s exact test. A value of *p < 0.05 was considered statistically significant.

Aiming to decipher a potential functional role of cyto-nuclear ZO-1 on inflammatory cell recruitment, we transiently transfected BEAS-2B invasive lung cells, which express low levels of ZO-1 maintained as a cyto-nuclear pool (Lesage et al., 2017), with ZO-1 cDNA expression vector or the empty pLNCX control vector. As previously reported in this cell model, we observed an increase of the cyto-nuclear pool of ZO-1 in BEAS-2B cells transfected with the ZO-1 expression vector compared to controls cells (Lesage et al., 2017). To validate our hypothesis that cyto-nuclear ZO-1 may modulate the secretome of tumor cells and thereby control the chemotactic activity of tumor cells, we examined the expression of 104 human soluble factors by cytokine array in conditioned medium of cells transfected or not with ZO-1. The modulation of the secretome by ZO-1 was characterized by at least a 50% increase in the production of cytokines, while no cytokine was found decreased below 50% (Figure 1A). The six most over-produced pro-inflammatory molecules were growth-regulated oncogene-α (GROα), granulocyte-macrophage colony-stimulating factor (GM-CSF), intercellular adhesion molecule-1 (ICAM-1), IL-6, IL-8, and matrix metalloproteinase-9 (MMP-9) (Figures 1A,B; Supplementary Figure S1). As a proof of concept, we used a transmigration assay with the THP-1 monocytic cell line to assess the chemotactic activity of tumor cells modified as above for cyto-nuclear ZO-1 content. Our results revealed that THP-1 cells migrated 30% more in response to conditioned media from ZO-1 cDNA-transfected cells than to conditioned media from control cells (Figure 1C). These results together demonstrate that tumor cells expressing high levels of cyto-nuclear ZO-1 secrete soluble factors that can stimulate inflammatory cell migration in vitro.

Aiming to explore this ZO-1/soluble factors/inflammatory infiltrate axis further in an in vivo context, we used a mouse ear sponge assay which is particularly adequate to analyze the impact of tumor-produced soluble factors on immune cell recruitment. For this assay, bio-compatible sponges were soaked in conditioned media of ZO-1-transfected BEAS-2B cells or control cells, and inserted subcutaneously in mouse ears for 3 weeks. An initial quantification of DAPI labeling permitted to quantify overall levels of cell infiltration in the sponges. Our results showed a larger cellular infiltration of sponges soaked in conditioned media from ZO-1-transfected BEAS-2B cells (1.35 ± 0.07-fold) (Figures 2A,B). Examining closer cellular infiltrates, we observed a higher recruitment of CD3⁺ T cells (1.46 ± 0.17-fold; Figures 2C,D) in the “ZO-1 sponges”. Similar results were obtained with SKBR3 cells, another cell model which maintains ZO-1 as a cyto-nuclear pool (1.90 ± 0.15 fold; Supplementary Figure S2). Taken together, these results suggest that the secretome of tumor cells with high cyto-nuclear levels of ZO-1 promotes recruitment of immune T cells.

Given these in vitro and in vivo results, we examined the potential relationship between ZO-1 localization in tumor cells and the inflammatory and immune infiltrate by immunostaining in human NSCLC samples. First, cancers were categorized according to ZO-1 staining pattern as displaying a low (<10% of tumor area) or a high (in ≥10% of tumor area) distribution of membrane-associated ZO-1, and a high (in ≥10% of tumor area) or low (<10% of tumor area) distribution of cyto-nuclear staining (an illustration of a ZO-1 membrane staining versus a cyto-nuclear staining is provided in Figure 3A). As shown in the scoring table (Figure 3B), cancers with high membrane-associated ZO-1 staining showed low cyto-nuclear ZO-1 labeling, and cancers with a high cyto-nuclear distribution of ZO-1 predominantly associated with a low ZO-1 membrane-associated score. Noticeably, cancers displaying a low ZO-1 membrane-associated score equally distributed between a group displaying a high cyto-nuclear ZO-1 distribution and a group displaying a low distribution of cyto-nuclear ZO-1, suggesting an overall weaker expression of ZO-1 in the latter. We next compared the immune infiltrate in the low and high cyto-nuclear ZO-1 distribution groups. We thus analyzed CD3 to identify T lymphocytes, and CD4 and CD8 to further differentiate T helper and cytotoxic T lymphocytes respectively (Figure 3C). We observed that a high cyto-nuclear staining of ZO-1 in lung tumor cells was associated with an increased density of T lymphocytes identified as a cytotoxic CD8⁺ T sub-population into tumor microenvironment (Figure 3D). Although these results are rather suggestive of an anti-tumor immune infiltrate in human NSCLC samples, we also evidenced a higher density of Foxp3⁺ immunosuppressive regulatory T cells in tumors with high cyto-nuclear expression of ZO-1 (Figure 3D). Interestingly, this higher density of Foxp3 regulatory T cells also correlated with a higher density of CD8⁺ cytotoxic T cells into the human lung tumor microenvironment (Figure 3E). Altogether, these results showed that the presence of cyto-nuclear ZO-1 in tumor cells correlated with an increased density of CD8⁺ and Foxp3⁺ immune cells in NCSLC. These in vivo data suggest that, even though more numerous, effector T cells might be inhibited by a higher recruitment of immunosuppressive T cells.