Decoding melanoma-immune cell communication through spatially resolved ligand-receptor interaction analyses

Decoding melanoma-immune cell communication through spatially resolved ligand-receptor interaction analyses | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Decoding melanoma-immune cell communication through spatially resolved ligand-receptor interaction analyses Florian Große, Christoph Kämpf, Dennis Löffler, Max A Lingner Chango, and 12 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9075388/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Melanoma is a highly immunogenic tumor where interactions between tumor and immune cells critically influence disease progression and treatment response. Traditional single-cell transcriptomics lack spatial context, limiting the understanding of these interactions within the tumor microenvironment. Here we integrated 10x Genomics Visium and Visium HD spatial transcriptomics with multiple single-cell RNA-seq datasets to map ligand-receptor interactions at the tumor-immune border region. This approach identified immune cell interactions already known in melanoma and highlighted unknown but prominent LGALS1 – PTPRC (proteins: galectin-1–CD45) interactions localized at the tumor-immune border region. Immunofluorescence and biolayer interferometry confirmed galectin-1 expression on melanoma cells and its binding to CD45 on infiltrating T cells. High LGALS1 expression correlated with reduced overall survival in patients undergoing checkpoint inhibitor therapy. These findings suggest that galectin-1–CD45 interactions contribute to immune modulation in melanoma and represent potential targets for immunotherapeutic strategies. Dermatology skin cancer melanoma spatial sequencing ligand-receptor interactions immunotherapy Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Main Melanoma is a highly immunogenic tumor, and the interaction between immune cells and tumor cells is of central importance for tumor development and control. Significant treatment responses using immune checkpoint inhibitors have been achieved in recent years with antibodies targeting cytotoxic T-lymphocyte antigen 4 (CTLA-4) and programmed cell death 1 (PD-1) in immune cells, which reactivate the compromised immune system ( Schadendorf et al., 2018 ). At present, checkpoint inhibitor treatment has gained increasing relevance because of its mutation-independence and long-‍term effects. However, most patients (60%) do not respond to these treatments, and recurrence rates are high ( Lim et al., 2023 ). New clinical trials have shown promising results with antibodies directed against further immune checkpoint targets, such as lymphocyte activation gene‑3 (LAG‑3) and T-cell immunoglobulin and mucin-domain containing-3 (TIM‑3) on immune cells, which led to the approval of anti-LAG3 treatment in combination with anti-PD1 treatment. However, the treatment effects are only slightly better than those of anti-PD1 treatment alone ( Tawbi et al., 2022 ). Recent studies have shown that immune checkpoint inhibitors may also provide beneficial effects as neoadjuvant treatments ( Saad et al., 2023 ). The search for new immune targets and treatment modalities for future treatment is ongoing. In this context, a better understanding of the melanoma-‍immune cell interactions at the immunological interface is needed. In recent years, the spatiotemporal organization of cells within a complex tissue has been analyzed using single-cell and spatial transcriptomics technology ( Lim et al., 2020 ). Several earlier studies have used single-cell transcriptome analysis to analyze melanoma and immunological heterogeneity ( Tirosh et al., 2016; Rambow et al., 2018; Sade-Feldmann et al., 2019; Li et al., 2020 ). In particular, the analysis of melanoma treatment resistance and response in humans has been the focus of these studies ( Jerby-Arnon et al., 2018; Sade-Feldman et al., 2019; Li et al., 2020 ). One of the first studies utilizing single-cell RNA sequencing (scRNA-seq) analyses identified a T cell signature that allowed the classification of bulk-sequenced tumors as tumors harboring a (T cell)-exclusion program ( Jerby- ‍Arnon, et al., 2018 ). This exclusion program was then mapped onto ipilimumab and anti- ‍ PD1– ‍ treated samples analyzed by scRNA-seq to identify co-expressed genes in individual cells. In a further study, single-cell transcriptomes of metastatic melanoma patients were generated from biopsies taken at baseline and under anti-PD1 inhibitor treatment, either alone or in combination with anti-CTLA4 treatment ( Sade-Feldman et al., 2019 ). Treatment-resistant clusters were enriched for genes linked to T cell exhaustion ( LAG3 , PDCD1 , HAVCR2 , TIGIT , CD38) . A more recent study focused on dysfunctional T cells in melanoma lesions undergoing immune checkpoint therapy ( Li et al., 2020 ). This study included patients with prior treatment against CTLA‑4 or PD-1 or a combination of both and revealed that CD8 + T cells partly transitioned into a dysfunctional T cell pool. These dysfunctional T ‍ cells are characterized by the expression of PDCD1 , LAG3 and molecules shared with CD4 + Tregs (e.g., ‍ CSF1 and ZBED2) . However, the dysfunctional T cells had the highest levels of clonal expansion and were still active in ex vivo experiments. Overall, single-cell analyses of melanoma and other tumors have enabled a deeper understanding of tumor heterogeneity and the transcriptional states of immune cells in the tumor microenvironment. Several studies have shown that basic mechanisms of cell–cell communication can be inferred from single-cell analyses using knowledge of known ligand–receptor interactions in the spatial context ( Larsson et al., 2021 ). We recently analyzed transcriptional programs and ligand–receptors interaction in different cell types in high (hot), intermediate and low (cold) immune melanoma tumors ( Stubenvoll ‍et‍ al., 2025 ). Among prominent ligand-receptor interactions in hot tumors were CD58– ‍ CD2 and CD59–CD2, of which CD58–CD2 showed functional relevance. However, in classical scRNA-seq datasets, the physical relationships between cells remain unknown, which makes it difficult to place cell–cell interactions in an appropriate tissue context. To address this issue and extend our recent studies, we performed spatial transcriptomics (ST) analyses in primary melanomas and cutaneous melanoma metastases ( Figure 1 ), complemented by a recent data set of melanoma lymph node metastases ( Pozniak et al., 2024 ). In fact, a few recent studies have already used spatial transcriptomics for melanoma tissue, but did not specifically address the local melanoma –immune cell – interactions at the immunological interface ( Thrane et al., 2018; Hunter et al., 2021; Karras et al., 2022; Pozniak et al., 2024 ). We particularly focused on melanoma-‍T/NK-cell interactions as T/NK cells are clinically relevant in immunotherapies for melanoma. We identified several new candidates for melanoma–immune cell interactions, such as LGALS1–PTPRC, LGALS1–CD69, and LGALS3–CD6 using two different 10x Genomics® spatial transcriptomics technologies (Visium and Visium HD) and ligand–receptor interaction analysis ( Dimitrov et al., 2022 ). These interaction partners suggest new targets for future immunotherapeutic approaches in the therapeutic or adjuvant setting in melanoma. Results Spatial transcriptomics at mini-bulk scale captures the immunological interface of melanoma–immune-cells in primary melanomas and lymph node metastases To uncover melanoma–immune cell interactions, we first performed spatial transcriptomics (ST) of four primary melanoma lesions representing different subtypes (two superficial spreading, one nodular and one acrolentiginous melanoma) with varying immune cell infiltration levels (two immune hot and two immune intermediate melanoma lesions, Figure 1 ). Samples were analyzed utilizing the 10x Genomics Visium platform enabling whole transcriptome assessment at mini-bulk scale of the immunological interface at the tumor–immune border region ( Supplementary Table 1, and Extended data Figure 1 ). To guide spatial transcriptomics data analysis, an experienced histopathologist (MZ) determined cell annotations and histopathological areas of all tissue samples. We further verified immune cell infiltrates by immunofluorescence staining of serial sections with anti-CD3 antibody (CD3 ‍is a part of the T cell receptor complex). The samples were processed according to the manufacturer’s (10x Genomics) specifications for the Visium platform. We used 10x Genomics Space Ranger for initial data analysis and QC (for per-sample statistics see Supplementary Table 1C-D ). Figure 2A shows H&E staining of one of the two superficial spreading melanoma (SSM) with a tumor (Breslow) thickness of 2.8 mm (PM1). First, we identified 7 co-expression clusters by unsupervised Louvain clustering ( Blondel et al., 2008 ) on log-transformed size-normalized gene expression ( Figure ‍2B ), according to the standard Seurat workflow ( Hao et al., 2021 ). To obtain biologically relevant co-expression clusters, we chose the resolution parameter such that the clustering granularity was comparable to the histopathological annotation (see, Methods). Second, we annotated co-‍expression clusters according to enriched genes identified via Wilcoxon rank sum test of each cluster versus all remaining clusters (adjusted p 0.15). Among the top ten enriched marker genes ( Methods , p < 0.01), we found classical cluster-defining genes consistent with the histopathological annotation of sample PM1: IGHA1 , IGHG1 and IGKC for expression of complement factors and immunoglobulins referring to immune cells (cluster 1), SOX10, PMEL and MLANA for melanoma cells (cluster 2), S100A7A, KRT1 , KRT10 , and KRT15 for keratinocytes (clusters 3, 4 and 5) ( Extended ‍ data Figure ‍2A ). Additionally, we conducted gene set enrichment analysis (GSEA) ( Subramanian ‍et al., 2005 ) of marker genes based on the Gene Ontology (GO) ( Ashburner et al., 2000 ) biological processes (GO:BP, FDR < 0.01) database ( Methods and Extended data Figure 2 ). GSEA ‍confirmed cluster annotations by revealing significant enrichment (FDR < 0.01) of genes involved in leukocyte cell-‍cell adhesion and T cell activation in cluster 1, genes related to pigmentation in cluster 2, and genes associated with keratinocyte proliferation in clusters 3 and 4 and keratinocyte differentiation in clusters 4 and 5, respectively ( Extended data Figure 2B ). Overall, our unsupervised clustering strategy of spatial gene expression patterns reliably identified distinct histopathological tissue regions. However, ‍non-specific cell type annotation partly persists, since the 10x Visium platform lacks single-‍cell resolution. For instance, clusters 1, 3 and 4 exhibit low-level expression of classical markers for melanoma cells like SOX10 , PMEL and MLANA , albeit not at statistically significant levels( Extended data ‍Figure 2B ). For a refined view of melanoma and T/NK cell localization in 10x Visium spots, we combined two distinct approaches of integrating single-cell gene expression references with ST data (see, Methods). First, we applied spot-wise cell type deconvolution using our recently published single-cell gene expression dataset ( Stubenvoll et al., 2025, for details see, Methods). It contains distinct subtypes of melanoma cells derived from a study analyzing gene expression patterns in patient-derived melanoma cell cultures ( Tsoi et al., 2018 ). The deconvolution results confirm the histological and unsupervised co-‍expression clustering annotation of sample PM1. Spots belonging to co-expression cluster 2, which correspond to the histologically annotated tumor region, are primarily composed of melanoma cells. Spots enriched for T/NK cells are in the region histologically annotated as inflammatory, which corresponds to cluster 1 ( Figures 2A-C) . Notably, the percentage of T/NK cells is not uniform in the inflammatory region but contains multiple nests of tightly packed inflammatory cells along with a moderately high T/NK cell content along the remaining tumor interface. Second, we integrated the single-cell gene expression reference with two existing single-cell gene expression melanoma datasets ( Jerby-Arnon et al., 2018; Solé-Boldo et al., 2020 ) to derive panels of cell type-specific and sample-‍independent marker genes. Using these gene panels, we calculated module scores utilizing Seurat that represent the cell-type specific expression over background ( Figure 2D ). All resulting marker gene panels are listed in Supplementary Table 2 . For example, the marker gene panel for melanoma cells contains among others PMEL , TYR , MLANA , and MITF supporting biological relevance of retrieved marker gene panels. Both approaches, cell type deconvolution and module scores, localized tumor cells in similar regions. Deconvolution tended to overestimate melanoma cell prevalence in regions with low total expression, as indicated by low tumor cell module scores in low-‍expression regions ( Figures 2D-E ). To identify the tumor-immune border region, we therefore used a combined approach to identify the tumor core region and regions with highly active immune cells. Spots are labelled as melanoma positive, if deconvolution shows more than 10% of melanoma content and the melanoma module score exceeds the threshold defined by a gaussian mixture model. Spots are labelled as T/NK positive, if deconvolution shows more than 10% of T/NK cell content and the respective module score exceeds the threshold defined by a robust background estimation ( Figure 2F , Methods ). With this combined information, we identified spots with high tumor and/or immune cell activity and then marked the spots positive for both or with direct adjacency of both as the region with active immune infiltration or border region for short. Figure 2H shows the result of our approach, separating the active region (yellow, bright green and bright red) from tumor and immune cell positive spots not part of the inflammatory border (dark red and dark green). We verified immune cell infiltrates in this region by immunofluorescence staining of a serial paraffin section, showing areas of CD3-positivity adjacent to the tumor area ( Figure 2G ). We followed the same data analysis workflow for the three remaining melanoma samples (PM2-PM4) processed by 10x Visium ST (Extended data Figures 3-5). These samples consisted of the second SSM of 0.6 ‍mm tumor thickness (PM2), an acrolentiginous melanoma (ALM) of 4.3 mm tumor thickness (PM3) and an SSM of 1.0 mm tumor thickness (PM4). PM2 showed a histopathologically detectable inflammatory infiltrate in the upper dermis, indicated by a yellow encircled area ( Extended data Figure 3A ). This area was represented by co-expression cluster 9 (inflammatory cells) and was also verified by immunofluorescence staining for CD3 expression ( Extended data Figure 3G ). Epidermal keratinocytes were present in co-expression clusters 1, 5 and 7, showing expression of KRT1 , KRT10 , LOR and KRT15 ; which were collocated with melanoma cells as indicated by the classical melanoma cells markers MLANA and TYRP1 ( Extended Figure 3C ). These findings were supported by GSEA regarding the expression of genes involved in leukocyte-mediated immunity in cluster 9 and pigmentation in clusters 5 and 7 ( Extended Figure 3D ). Immune infiltration was also detected in a serial H&E section ( Extended data Figure 3G ). In PM3, an acrolentiginous melanoma of 4.3 mm Breslow thickness, the border region was indicated by cluster 5, showing an immune cluster immediately under the upper major melanoma mass (cluster ‍7) and immediately adjacent to the lower melanoma part (cluster 4) ( Extended Figure 4A and 4B ). The ‍melanoma clusters showed varying expression levels of melanoma cell markers such as PMEL , MLANA and TYR ( Extended data Figure 4C ), and GSEA showed enrichment of genes involved in melanin metabolic processes; whereas GSEA of the immunological cluster showed enrichment of genes involved in positive regulation of T cell activation ( Extended data Figure 4D ). Extended data Figure 4E shows H&E staining of a serial section and Extended data Figure 4F shows an anti-CD3 immunofluorescence staining of the inflammatory infiltrate. The area of T cell infiltration was also verified by the T cell module score ( Extended data Figure 4G ). Although acrolentiginous melanomas are poorly inflammatory in general, we identified a clear inflammatory infiltrate in this case. In PM4, a superficial spreading melanoma of 1.0 mm tumor thickness ( Extended data Figure 5 ) inflammatory cells were present in cluster 0, and gene expression was detected for known immune regulators like CCL19 and CCL21 , as well as a complement factor C3 in this cluster ( Extended data Figure 5B and 5C ). Cluster 1, 4 and 5 harbored melanoma cells and expressed PRAME , SOX10 and MLANA , as classical melanoma cell markers ( Extended Figure 5C ), supported by GSEA of genes involved in cellular pigmentation (clusters ‍1 ‍and 5) and positive regulation of T cell activation (cluster 0), respectively ( Extended data Figure 5D ). All ‍melanoma cell areas and areas of T and NK cell infiltration were further confirmed by deconvolution and module score analyses as described above ( Extended data Figure 5G ). We followed the same approach as for sample PM1 to identify tumor-immune border regions in primary melanoma samples PM2-4 (Figure 3A). To also cover tumor-immune cell communication in metastasis, we extended our analyses to a published independent dataset of six melanoma lymph-‍node metastases analyzed with 10x Visium ST ( Pozniak et al., 2024 ). Using the same workflow as for PM1-4, we analyzed the samples from Pozniak and co-workers (Pozniak et al., 2024) and found varying degrees of infiltration and tumor activity. These samples mostly contained large dense tumor masses covering most of the area with infiltration ranging from almost-absence of detectable immune cells to large inflammatory regions bordering the tumor masses ( Supplementary Figure 1 ). Three samples (POZNIAK1-3) showed distinctly weaker T/NK cell scores and deconvolved T/NK content and were excluded from further analysis due to the absence of a defined tumor-immune border region. The other three samples (POZNIAK4-6) showed stronger T/NK activity and a defined border region and were retained for ligand-‍receptor analysis ( Figure 3A ). Overall, we defined a spatially resolved map of the tumor-immune border region in four primary melanomas of different subtypes and three independent lymph node metastases enabling unbiased, in-depth and putatively clinically relevant analysis of ligand–receptor interactions at the tumor-‍immune border region. Spatially resolved immunological interfaces between melanoma and T/NK cells revealed physiologically relevant ligand-receptor interactions We resolved ligand-receptor interactions at the melanoma-T/NK cell borders for all seven samples (PM1-4 and POZNIAK4-6) by testing overrepresentation of known ligand-receptor interactions. We utilized the NICHES R package ( Raredon et al., 2023 ) to compute interaction scores ( product-‍of-‍expressions interaction model) while incorporating the spatial proximity of spots. Combining the ligand–receptor interaction databases from LIANA ( Dimitrov et al., 2022 ) and NicheNet ( Browaeys et al, 2020 ) yielded a total of 14,977 unique mechanisms (after accounting for changed gene symbols and deduplication), for which we calculated NICHES scores and overrepresentation. For ‍an unbiased detection of ligand-receptor interactions at the tumor-immune border region, we evaluated all known interactions regardless of previously identified cell types. Finally, the result matrices were combined across samples and annotated with additional information, such as the expression of ligand and receptor molecules in different cell populations in the single cell reference dataset (see Methods). Figure 3B depicts all interacting partners significantly enriched in the border regions (adjusted p-‍value < 0.05) in all four primary melanomas (PM1-4) and their according enrichment values in three melanoma lymph node metastases (POZNIAK4-6). An expanded list of significantly enriched interactions in at least 3 out of 4 primary melanomas is given in Extended data Figure 6. We ‍identified 153 unique mechanisms in at least 3 primary melanomas, also visualized in a circus plot ( Figure 3C ). A summary of all ligand–receptor interactions between T/NK and melanoma cell types that were significantly enriched according to the LIANA analysis of the border region (aggregated p-value < 0.01) and supported in at least 2 Visium HD samples for one source-target cell type pair is available in Supplementary Table 3A , and a summary of all interactions differentially expressed (adjusted p-value < 0.05) in the border region at least one melanoma sample (Visium platform) is available in Supplementary Table 3B . A substantial fraction of detected interactions and/or interaction partners have been described in earlier studies on immune-regulated tumor control, such as CD58 and CD59 ( Frangieh et al., 2021 ), CD58 –CD2 ( Stubenvoll et al., 2025 ), CCL19 and CXCL9 ( Jacquelot et al., 2018; Gowhari et al., 2022; Ding et al., 2016 ), as well as HMGB1–CXCR4 ( Li Pomi F et al., 2022 ) and MIF –CD74 ( de Azevedo et al., 2020; Figueiredo et al., 2018 ), underpinning the physiological relevance of identified interactions. Prominent expression was observed for LGALS1 (galectin-1) on melanoma cells and PTPRC (CD45) on immune cells ( Figure 3B ). PTPRC showed the strongest expression on immune cells, especially on T ‍cells, similar to CD3 and CD2. Galectin-1 (gene product of LGALS1 ) is a known immunosuppressive molecule and an interaction partner of CD45 (gene product of PTPRC ) (Novák et al., 2025 ). Both are expressed in proximity at the border regions between melanoma and T cells. NICHES interaction scores align well with the previously defined border regions in all seven samples ( Figure 3D and 3E, compare 3A and ‍2H; full panels as in Figure 3D for all primary samples for LAGALS1–PTPRC and three additional LR pairs are available as Supplementary Figure 2 . However, evidence for the role of the galectin-‍1–‍CD45 interaction in melanoma immunology is scarce. This possible interaction in vivo was found in our data set and by reanalysis of the mentioned independent study ( Pozniak et al., 2024 ) . Multicolor immunofluorescence validates LGALS1 (galectin-1)–PTPRC (CD45) expression and membrane-localization We performed multicolor immunofluorescence (IF) staining to test whether LGALS1 (galectin-1) expressing melanoma cells were in proximity to PTPRC (CD45) expressing T cells. IF staining assessed expression of SOX10, CD3, galectin-1 and CD45 in 20 additional melanoma samples. The percentage of CD3 ‍positive (CD3 + ) cells classified the immune status of samples as cold ( 20 %). Subsequently, we measured distances between CD45 positive (CD45 + ) and galectin-1 and SOX10 (galectin-1 + ) double positive tumor cells for at least three representative areas per sample. Figure 4A and Supplementary Figures 3A-C show representative examples with CD45 + immune cells in proximity to galectin-1 + melanoma cells. A digitalized picture is shown in Figure 4B and Supplementary Figures 3A-C . Cell segmentation coordinates of stained sections were analyzed by Qupath and exported to analyze the distance between galectin-1 + melanoma cells and CD45 + immune cells. The majority of CD45 + immune cells were within a range from 0 to 20 ‍µM ( Figure 4C ). Further quantification of all samples showed that the density of CD45 + immune cells surrounding galectin-1 + melanoma cells within 20 µm was significantly higher in immune hot melanoma samples compared to immune cold samples ( Figure 4D ). However, the median distance between galectin-1 + and CD45 + was not different between samples of different immune cell infiltration. Together, the majority of CD45 + immune cells are located in proximity to galectin-1 + tumor cells with high prevalence in immune hot samples, which may support immune interactions. To further characterize the galectin-1–CD45 interaction on a protein level, we performed in-vitro interaction assays. For this purpose, soluble extracellular domains of galectin-1 and CD45 were expressed and purified from E. coli and HEK 293 cells respectively. Complementary in-silico structure prediction supported a plausible extracellular interaction interface between galectin-1 and CD45. Using Boltz-2-based structure prediction ( Passaro et al., 2025 ), we observed that both proteins are placed in matching positions in the most confident models (confidence score ≥ 0.91). These predictions suggest an interaction in the fibronectin type-III domain I (residues 391-483) of CD45 and the potential C-Terminal dimerization site of galectin-1 opposite from the carbohydrate binding site ( Figure 4E and Supplementary Figure 4 and Supplementary Table 4 ). The binding affinity of CD45 to galectin-1, and galectin-3, respectively, which was included due to high sequence and presumably functional similarity, and negative control anti-SARS-CoV2 T1 antibody was determined via bio-layer interferometry (BLI) ( Figures 4E-F ). Association and dissociation were relatively rapid in case of CD45–‍galectin-1 and CD45–‍galectin-3 binding, arguing for low but specific binding. Taken together, galectin-1 shows rapid and specific binding to CD45, which may influence immune cell function during immune control of melanoma cells. High-definition spatial transcriptomics validates LGALS1 and PTPRC co-expression at the melanoma–‍immune cell border region In a next step, a set of seven additional melanoma samples, four primary melanoma (HD_PM1-‍HD_PM4) and three cutaneous melanoma metastases (HD_MM1-HD_MM3), were subjected to ST at subcellular resolution utilizing the Visium HD technology. It allows for higher spatial resolution and thereby direct analysis of putative intratumor cell-cell interactions and may thereby complement the above analyses using classical Visium technology ( Olivera et al., 2025 ). The four primary melanomas had also been analyzed in our recently published melanoma single-cell study ( Stubenvoll et al., 2025 ). In short, samples were processed according to the manufacturer’s (10x ‍Genomics) specifications. We used SpaceRanger for initial data analysis and QC (sample statistics in Supplementary Table 1 ). Afterwards, we continued with a modified bin2cell (Polanski et al., 2024 ) workflow to continue our analysis with quasi-single-cell resolution (see, Methods). We used a custom segmentation of the H&E image in order to robustly identify cell nuclei and aggregated 2x2µm capture bins into cells using bin2cell. We then used TACCO OT ( Mages et al., 2023 ) to annotate cell types based on the single-cell reference ( Stubenvoll et al., 2025 ) already used for deconvolution of samples PM1-‍4. After quality control and filtering, we identified the tumor-immune border region. The border region is defined by T/NK cells and tumor cells being located within 50 µm distance from each other. All cells within the tumor-immune border region were then analyzed with LIANA using the same combined database of ligand-receptor interactions as above, which identifies ligand–receptor interactions at single-cell level and results were aggregated across samples (see, Methods). Despite the different technological and statistical approach, this analysis gives a similar insight into cell–cell interactions at the border region as our previous analysis using the classical 10x Visium technology, albeit with a more direct attribution of ligand–receptor interactions to interacting cell types. Naturally, there are many more potentially relevant interactions in the table, some of which were recently described. For example, a pair of ligand-receptors with high expression was APP (Amyloid Beta Precursor Protein) in various combinations with CCR8, CCR4, IL18RAP, CXCR4, and CD74, was present in all three melanoma differentiation states. APP – CD74 showed particularly high expression in transitory melanoma (Mel_trans) cells. Recently, APP has been described as an immunosuppressive molecule in glioblastoma via the APP–CD74 axis ( Ma et al., 2024 ). Among the interactions significant in most samples involving Mel_trans_melan and Mel_trans cells with activated cytotoxic T cells were also PCDH7 – CXCR4 . PCDH7 is known to be involved in cell-‑cell recognition and appears to be involved in pancreatic cancer as a cold tumor induction-related gene ( Mochida et al., 2025). In a more recent study using a CRISPR knockout screen of melanoma cells in co-culture with cytotoxic T cells, CD58 , CD59 , IFNGR and CXCR4 were among the top perturbations in melanoma cells associated with immune checkpoint inhibitor resistance, emphasizing the principal role of CXCR4 in melanoma immune control ( Frangieh et al., 2021 ). In Figure 5D , we summarized LIANA data for the LGALS1 – PTPRC interaction across all combinations of cell types, indicating that this mechanism is not exclusive to melanoma—T cell interactions but can also be found between dendritic cells (DCs) and T cells. Indeed, DCs have been detected at immunological borders to interfere with immune activation and galectin-1 has been regarded as a marker for immunosuppressive DCs ( Ilarregui et al., 2009; Ludin et al., 2025; Novák et al., 2025 ). We further investigated whether our approach could directly identify potentially interacting cells by calculating NICHES interaction scores. Due to the high spatial resolution, we could set the edge distance cutoff to 50 µm to be consistent with our border region definition. Figure 5E shows a representative Visium HD sample (HD_PM3), completed with annotated cell types, region classification, and NICHES interaction scores. Since rendering all interaction edges across the whole sample would not work visually, we highlighted cells with high-valued incident edges in the overview plot (left) and only rendered non-zero edges for selected regions (right). Qualitatively, these findings support our LIANA results. The same analyses were performed for the other Visium HD samples ( Supplementary Figure 5 ). We observed numerous interactions involving melanoma, T, and dendritic cells along the tumor–immune border region. Overall, high-definition ST at subcellular scale strongly supported our results and also identified a number of other ligand-receptor interactions that have been shown to play a role in cancer biology, but may also be of importance for melanoma immunity. High LGALS1 and PTPRC expression associate with differential survival in immunologically hot tumors Finally, the possible impact of LGALS1 and PTPRC expression on the prognosis of melanoma patients was analyzed to further substantiate the clinical relevance of our findings ( Figure 6 ). We collected the skin cancer cohort data from The Cancer Genome Atlas ( Cancer Genome Atlas Network, 2025) via The ‍Cancer Immunome Atlas (TCIA, http://tcia.at) webserver. The TCIA enhances the TCGA data by providing estimates of cell type proportions using the quanTIseq deconvolution algorithm ( Finotello et al., 2019 ). Additionally, we downloaded another cohort of melanoma patients published by Gide and co-workers ( Gide et al., 2019 ) from the Sequence Read Archive (SRA) and used quanTIseq for the deconvolution of these samples. We examined the effect of LGALS1 and PTPRC expression in the context of immune infiltration on patient survival using Kaplan–Meyer survival analyses ( Figure 6 ). For both cohorts, we used the largest homogenous subgroup to reduce interfering effects: for the TCGA cohort, we included only metastatic samples with non-zero detectable infiltration (as per quanTiseq estimates, n = 291) and for the Gide cohort ( Gide et al., 2019 ), we focused on Pembrolizumab patients only (n = 71), which show a markedly different survival pattern compared to Nivolumab patients. In both cases, we separated samples into LGALS1 / PTPRC high and low groups based on a median split first. High PTPRC is generally associated with favorable outcomes ( LGALS1 low subgroup: log-rank test p = 0.00032 in TCGA cohort and p = 0.00041 in the Gide cohort, respectively) ( Figure 6A and B ), but this effect is attenuated in patients with high LGALS1 expression ( LGALS1 high subgroup: log-rank test p = 0.22 in TCGA cohort and p = 0.12 in Gide et al. cohort, respectively). It is known that PTPRC is predominantly expressed by lymphocytes, which prompted the question of whether the positive effect on patient survival is largely the positive effect of higher T lymphocyte infiltration on patient survival. For the larger TCGA cohort, we were able to divide patients by estimated CD8 + T cell content and then separate LGALS1 / PTPRC high and low subgroups for CD8 + high and low patients separately ( Figure 6C ). In the CD8 + high status and high PTPRC expression group, high LGALS1 expression was associated with bad prognosis (log-rank p = 0.0046). In contrast, the PTPRC low subgroup showed no significant difference in survival based on LGALS1 levels (log-rank p = 0.61), which means that the level of CD8 + cytotoxic T cells alone is not sufficient to explain the survival differences. In case of CD8 + low status, there was no significant difference between high and low LGALS1 expression, both in case of high and low PTPRC expression (log-rank p = 0.46 and p = 0.61, respectively). This argues for a direct impact of galectin-1 ( LGALS1) on CD45 ( PTPRC) via an immune inactivating pathway, which is, however, without consequences in the absence of relevant CD8 + T cell infiltration (cold tumors). Taken together, high LGALS1 expression with a negative impact on PTPRC may negatively influence melanoma patient survival. Discussion Melanoma cell–immune cell interactions are a central mechanism of tumor control and treatment response ( Schadendorf et al., 2018 ). Recent single-cell studies have revealed several different mechanisms active in this situation, defining different subpopulations of CD4 + and CD8 + T cells in these tumors, among which exhausted T cells appear to play a central role ( Jerby-Arnon et al., 2018; Sade-‍Feldman et al., 2019; Li et al., 2020 ). However, the spatial relationship of this interaction could not be defined in these studies. With the advent of spatial transcriptomic technology in cancer science, cell-cell interactions can be analyzed in detail ( Larsson et al., 2021; Seferbekova et al., 2023 ). The 10x ‍Genomics Visium and Visium HD platforms enable whole-transcriptome assessment at spatial resolution, facilitating unbiased description of cell-cell interactions. Here, we analyzed melanoma–‍immune cell interactions in primary melanomas, melanoma lymph node metastases, and cutaneous melanoma metastases by utilizing both platforms. We were able to identify and localize various cell populations, such as keratinocytes, melanoma cells, lymphocytes, macrophages, monocytes and endothelial cells within the melanoma and metastatic lesions and define regions of inflammatory infiltration at the proximity to melanoma cells (border regions), as well as ligand-‍receptor interactions between tumor and immune cells in these regions. Many of these interactions at the melanoma–immune border region identified in the present study were interactions between different HLA molecules and CD3 (as part of the T cell receptor complex). This further supported the physiological relevance of our findings, since HLA-mediated antigen presentation to T cell receptor molecules is a main mechanism at the immunological synapse. However, we identified also several other putative interactions at this melanoma–immune border region. Among these were MIF – CD74 , LGALS1 – CD69 and LGALS3 – CD6 , and CCL19 – S1PR2 and CCL19 –‍ S1PR3 interactions, which had already been described to play a role in melanoma immunology ( Laidlaw et al., 2019; de Azevedo et al., 2020; Femel et al., 2022; ). MIF is a lymphokine involved in cell-‍mediated immunity and an immunosuppressive factor secreted in the tissue microenvironment of melanomas ( de Azevedo et al., 2020 ). CD74 is a chaperon for the class II MHC-complex and is thereby involved in T cell antigen presentation. In two recent studies, the interaction between MIF and CD74 induced tolerogenic dendritic cells and M2 pro-tumorigenic macrophages in melanoma ( de Azevedo et al., 2020; Figueiredo et al., 2018 ). Thus, this interaction may finally support melanoma growth. Galectin-3 (product of LGALS3 gene) has been shown to act as a binding partner for the T cell receptor (TCR) and leads to downregulation of the TCR and inhibition of early T cell activation ( Gilson et al., 2019 ). Galectin-3 also inhibits interferon gamma diffusion into the melanoma and ex vivo generation of tumor-specific T cells ( Zubieta et al., 2006 ). Our candidate list of putative interactions also contained other less well-known interaction partners in melanoma. Among these were APOE and SORL1 . APOE ‍is ‍originally described as a protein involved in triglyceride lipid protein-rich metabolism. SORL1 ‍belongs to the low-density lipoprotein receptor (LDLR) family. The biological relevance of these findings in the tumor context remains to be studied in more detail, but may so far be supported by the fact that different APOE variants were linked to melanoma treatment response ( Ostendorf et al., 2020 ). We recently showed that CD58 , but not CD59 , acts as a positive immune modulator in melanoma ( Stubenvoll et al., 2025 ). However, the investigation of CD59 in this context requires further investigations, since CD59 is a versatile molecule and specific conformational changes may be necessary for active CD59–CD2 interaction. Interestingly, many of the interactions found to be enriched in the border region in the primary analysis using the Visium platform could be verified in the second cohort of samples using the Visium HD platform and attributed to specific interacting cell types. The latter provides higher spatial resolution and therefore allows analysis of cell-cell interactions also in compact tumor masses such as melanomas. In the present study, we put an emphasis on LGALS1 and PTPRC expression which was among top interaction partners in the first sample set of four primary melanomas and present in all four samples, and in an additional set of melanoma lymph node metastases of an independent cohort ( Pozniak et al., 2025 ). Moreover, both of these molecules show an elevated expression at the tumor-‍immune border region. By use of an additional larger number of melanoma samples stained by immunofluorescence, it was shown that CD45 + T cells are predominantly located at a distance between 10-50 µM to the next melanoma cell and also numerically enriched in close proximity to melanoma cells in immune hot tumor areas, as compared to cold areas. These findings further support the notion that this interaction may be of clinical relevance, as CD8 + T cell infiltration is a major prognostic factor for melanoma prognosis ( Jacquelot et al., 2017 ; Fu et al., 2019 ). More recently, the complex role of galectin-1 as an immunomodulator has been reviewed in more detail ( Novák et al., 2025 ). Within the tumor tissue environment, galectin-1 may impact the differentiation of tolerogenic dendritic cells and induces the apoptosis of effector T cells and enhances the proliferation of regulatory T cells. Principally, galectin-1 is heterogeneously expressed in different types of cancers, with melanomas showing prominent upregulation compared to normal tissue and to other tumor entities ( Novák et al., 2025 ). We found significant upregulation of LGALS1 (galectin-1) in three different data sets, in our own Visium and Visium HD data set and a Visium data set from an independent group ( Pozniak et al., 2024 ). Moreover, galectin-1 (and galectin-3) showed strong and immediate molecular binding activity to CD45 ( PTPRC gene product) in biolayer interferometry measurements, further arguing for an important role at the melanoma–immune border region regarding cell-cell interactions. Binding was rapid and transient but may still be strong enough to impede a CD45-mediated immune reaction. Here, blocking antibodies may improve the CD45-mediated tumor immune response. Experimental studies have shown that the galectin1–CD45 interaction exerts a number of different functions. E.g., it is known that dimeric galectin-1 binds to activated T cells and induces IL-10, which leads to the generation of regulatory T cells and subsequent immune suppression ( Cedeno-Laurent et al. 2012 ). Moreover, galectin-1 may suppress IL-2 and IFN-γ production in Th1 cells ( Rabinovic et al., 2002 ). In a more recent study, the use of the galectin-1 small molecule inhibitor LLS30 increased the anti-tumor activity of anti-PD1 treatment in immunotherapy-resistant prostate cancer ( Wang et al., 2024 ). Little is known about LGALS1 expression in other cancer contexts such as melanoma. Here, we show that LGALS1 expression was associated with a significant negative impact on overall survival of melanoma patients, when using a large TCGA cohort (n=291). In contrast, PTPRC expression was associated with improved overall survival. These findings were associated with high CD8 + T cell infiltration and essentially absent in immune cold tumors, arguing for a specific immune regulatory mechanism towards CD8 + T cells. Consistent with this, overall survival increased with higher PTPRC expression when focusing on patients under pembrolizumab (anti-PD1 antibody) immunotherapy in melanoma ( Gide et al., 2019 ). In a recent study, PTPRC expression was associated with improved survival and response to immunotherapy in melanoma patients ( Li et al., 2023 ). LGALS1 has also been reported as a negative prognostic factor in other cancers, e.g., ovarian, colon, and liver cancer ( Novák et al., 2025 ). Taken together based on data from our own and several other studies, the interaction between galectin-1 and CD45 (protein product of PTPRC ) might be a suitable target for future melanoma treatment approaches. We integrated a recent single-cell and spatial transcriptomic study in our analyses ( Pozniak et al., 2024 ) which put an emphasis on gene regulatory mechanisms (regulons) in treatment response and resistance. We exploited this data set, which also contained spatial transcriptomics data of melanoma lymph node metastases to further substantiate our ligand–receptor findings. In line with our data, an immune compartmentalization with a close proximity of LGALS1 + melanoma cells and T cells (including CD8 + T cells) could also be shown for the samples with strong infiltration and a clearly defined border region. In another recent study using melanoma spatial transcriptomic technology in a murine melanoma setting ( Karras et al., 2022 ), the proximity of pre-EMT neural crest stem-like cells (NCSCs) to blood vessels (perivascular niche) was analyzed, showing an anti-correlation of NCSC activity and distance to the next blood vessel. Regarding ligand-receptor interactions, authors found evidence for a Dll4–NOTCH3 interaction. Further analyses in this study were supportive for a model in which Dll4 + endothelial cells stimulate melanoma growth via activation of NC stem-like melanoma cells. Finally, one earlier spatial transcriptomics study used a mutated BRAF-‍based zebrafish melanoma model and the 10x Genomics Visium technology ( Hunter et al., 2024 ). When looking at melanoma microenvironment interfaces, HMGB2 came up as one of the major genes expressed by melanoma cells, which correlates well with HMGB1 expression in human melanomas found in our study. Both HMGB1 and HMGB2 encode for damage-associated proteins and may induce or repress extracellular inflammatory processes. However, a detailed analysis of melanoma cell interactions with immune cells was not presented in these studies, highlighting the relevance of our work. Taken together, unbiased whole-transcriptome spatial transcriptomics across three different data sets enabled us to identify ligand-receptor interactions enriched at the tumor-immune border region and to pinpoint the interacting cell types. This yields an important resource for understanding cell-cell communication in the melanoma tumor microenvironment and potential tumor evasion mechanisms. We provide substantial evidence that the interaction between galectin-1 and CD45 occurs at the melanoma–immune border region and may thereby impact the melanoma immune response. Future studies should be performed to prove their functional relevance in preclinical models and subsequent clinical trials. Materials and Methods Patient samples The use of patient samples for spatial and single-cell transcriptomics was approved by the local ethics committee of the Medical Faculty of the University of Leipzig Medical Center (AZ023-16-01022016; AZ349/18-ek). Samples from humans will be used for these experiments after informed consent from the patients and according to the rules and regulations of the Declaration of Helsinki regarding Ethical Principles for Medical Research 2013. A summary of patient samples and clinical information is provided in Extended data Figure 1 and Supplementary Table 1 . Fresh frozen tissue slices were derived from four primary melanomas. In addition, four primary melanomas and three melanoma metastases were analyzed by 10x Genomics Visium HD technology (10x Genomics). An additional set of 20 primary melanoma samples was used for immunofluorescence analysis ( Supplementary ‍Table 1 ). Previously published single-cell sequencing data from ten primary melanomas, of which four are paired with the four primary melanomas used for Visium HD, are used as a reference data set for cell type annotation ( Stubenvoll et al., 2025 ). Spatial Transcriptomics ‒ 10x Genomics Visium Platform Tissue Handling and Sectioning Samples were histologically annotated by an experienced histopathologist (MZ). Fresh frozen samples were cryosectioned at -10 °C (Thermo Cryostar, Fisher Scientific, Schwerte, Germany). Sections (10 µm) of four primary melanomas were placed on chilled Visium Tissue Optimization Slides and Visium Spatial Gene Expression Slides (10x Genomics, Pleasanton, CA, U.S.A.) and adhered by warming the back of the slide. Tissues were processed as recommended by the manufacturer. Tissue optimization For tissue optimization, tissue sections were fixed in chilled methanol and stained according to the 10x ‍Genomics Visium spatial tissue optimization user guide. Fluorescent images were taken using a Nikon Eclipse Ti with a Cy3 filter (ex/em brand) using a PlanFluro10X Ph1 objective and 900 ms exposure time. The optimal permeabilization time was determined to be 18 min. Spatial gene expression For spatial gene expression, tissue sections were fixed in chilled methanol and H&E stained according to the Visium Spatial Gene Expression User Guide. Brightfield histology images were obtained using a PlanFluro10X Ph1 objective (400 µs) on a Nikon Eclipse Ti (11746 × 11746 pixels). Raw images were stitched together using NIS-Elements AR 5.21.02 (Nikon) and exported as .tif files with high- resolution settings. The sequencing libraries were generated according to the user guide with an 18-minute permeabilization time, they were quantified (dsDNA HS Kit, Thermo Fisher), the molarity of each library was calculated, and equal molar amounts were pooled. Sequencing was performed with a 28-10-10-90 read setup on an Illumina HiSeq (11 pM loading concentration including 1% PhiX) using a Rapid SBS (200 cycles) Kit v2; or a NextSeq 2000 (650 pM loading concentration including 2% PhiX) using a P2 Reagent (200 cycles) kit. Raw data processing Sequencing data were analyzed using 10x Genomics SpaceRanger software version 1.3.1, with default parameter settings. We used spaceranger mkfastq to demultiplex and convert BCL files into FASTQ files. For read mapping and gene expression quantification, spaceranger count was performed. Reads were mapped to the human genome version GRCh38/hg38 provided by 10x Genomics. For gene expression quantification, the reference provided by 10x Genomics was used (2020‑A Human GRCh38 (GENCODE v32/Ensembl98)). We used the Loupe browser v5.1 to transfer histological sample annotations onto the spots. During this process, we manually removed a few disconnected spots from the main tissue region. Data analysis, quality control and preprocessing Further analyses of the quantified gene expression matrices were carried out in R v4.4.2 ( R Core Team, 2021 ). We used Seurat v5.3.0 ( Hao et al., 2024 ) and semla v1.3.2 ( Larsson et al., 2021 ) to import gene counts per spot matrix and tissue images. To assess sample quality, we visualized the distribution of genes and counts per spot and percentages of mitochondrial and ribosomal genes per spot across all spots of all samples and determined conservative cutoffs for removing low-quality spots. Subsequently, all spots were removed with less than 80 detected genes, less than 100 total counts, or more than 20% of counts mapped to mitochondrial genes. The number of spots removed per sample is shown in Supplementary Table 1 . Furthermore, the average expression of the housekeeping genes curated by Tirosh and co-workers was computed ( Tirosh et al., 2016 ) for each spot. Sufficient expression was observed in all spots, with variations depending on the tissue region. We applied multiple task-specific pre-processing schemes. For Louvain clustering and generation of Uniform Manifold Approximation and Projection (UMAP) embeddings ( Becht et al., 2018 ), we used the recently improved Seurat SCTransform workflow. In short, counts were preprocessed by applying the variance-stabilizing transformation (provided by the Seurat function SCTransform with settings vst.flavor=”v2” ) using the recently improved version ( Choudhary and Satija, 2022 ). Based on this, the first 30 principal components were computed using the highest variance explanation. A ‍two-‍dimensional Uniform Manifold Approximation and Projection (UMAP) embedding using 30 ‍principal components were calculated. To compute the mean expression values and NICHES interaction scores later on, we additionally performed the Seurat standard preprocessing workflow using natural log transformation. Log-transformed values were computed with and without library size normalization. Although it is best practice for single-cell data, the value of normalizing the observed counts per spot to a common value is still a debated topic. We observed, in agreement with two recent studies ( Saiselet et al., 2020, Bhuva et al., 2024 ), that the total gene expression levels per spot appear to have strong biological relevance. While scoring the mean expression or module scores of cell type-specific genes, we observed artifacts introduced by library-size normalization between regions with very different total expression (immune cells in tumor vs. stromal regions). For library-size normalization, we used the Seurat function NormalizeData with a scale factor of 10,000. For data without library-size normalization, we added a pseudocount of 1 to the observed counts and then computed the natural logarithm. After the initial computation of quality control metrics, all counts belonging to mitochondrial and ribosomal genes were filtered before any further analyses were performed. Deriving clusters of co-expressing spots Clusters were computed for co-expressing spots (representing regions with similar cell type composition and state) for each Visium sample using Louvain clustering of spots. We used principal component reduction based on variance-stabilized data described earlier. Using the selected 30 ‍principal components, a shared nearest-neighbor graph was generated, which was subsequently clustered using the Louvain Algorithm. For each sample, multiple clusters were computed with seven different resolutions evenly spaced between 0.1 and 1.3. Cluster stability was assessed using a cluster tree plot ( Zappia and Oshlack, 2018 ). Clustering results were compared with histological annotation to determine the clustering with the best concordance to histological annotation and highest plausibility. Correspondence to histological annotation was assessed by plotting the histological annotation vs. cluster number correspondence matrix, as well as a variety of correspondence metrics (notably Adjusted Rand Index, Normalized Mutual Information, Normalized Variation of Information and Overlap Coefficient) using the mclustcomp package. Histological annotations were not regarded as a ground truth to match completely, since it was expected that we would obtain a finer resolution of tissue region borders and possibly a finer resolution of overall structures in the tissue. Therefore, resolutions were selected that would not join annotated histological features into a single cluster based on visual assessment and otherwise matched the histologically annotated region borders well (by correspondence metrics, mostly Overlap Coefficient) while avoiding excessive subclustering. In doubt, we selected the resolution that gave the best overview of the tissue structures. The resulting choices of clustering resolutions are summarized in Supplementary Table 1C . Co-expression cluster annotation We annotated the spot clusters using a three-step approach: comparison with histological annotations, overexpression of marker genes, and pathway enrichment. To determine the overexpressed genes, we performed a differential gene expression analysis for each cluster in each sample by performing a Wilcoxon rank-sum test on the log-transformed size-normalized expression data. Using the FindMarkers function implemented in Seurat with logfc.threshold=0.15 and default parameters otherwise, we compared the expression of all spots in each cluster to the union of all other clusters. Adjusted p-values for multiple testing were calculated using Bonferroni’s method. A gene was considered to be significantly differentially expressed if the adjusted p-value was < 0.01. Gene enrichment analysis for significantly overexpressed genes was performed using the R package clusterProfiler ( Wu et al., 2021 ) and the Gene Ontology (GO) database of biological processes (GO:BP) ( Ashburner et al., 2000 ). The significance of enrichment was assessed by a hypergeometric test and adjusted p-values for multiple testing were calculated based on the Benjamini-Hochberg ( Benjamini and Hochberg, 1995 ) method. All GO terms with FDR<0.01 are considered significantly over-represented. In addition, overrepresented GO terms were further filtered to contain at least five of the significantly differentially expressed genes of a cluster or, in case of clusters with few differentially expressed marker genes, GO terms were required to contain at least 10% (with a minimum of two) of the differentially expressed genes in a cluster. Based on this information, cluster names were determined by an experienced histopathologist (MK). Spot deconvolution and annotation Since regular Visium spots are considerably larger than single cells and thus contain mixed expression signals from different cell types, we used reference-based deconvolution to estimate the cell type proportions for each spot. Reference-based deconvolution algorithms use the information from an annotated single cell reference dataset to estimate spot composition. We used our recently published single-cell dataset ( Stubenvoll et al., 2025 ) as reference to deconvolve Visium samples using RCTD ( Cable et al., 2022 ) and CARD ( Ma et al., 2022 ) algorithms. Since we know that deconvolution algorithms can be sensitive to the quality and granularity of the cell type annotation of the reference, we ran deconvolution on the original annotations and two coarser annotation levels, where cell subtypes were combined. The deconvolution results used in this work are averages of two CARD deconvolution runs on medium and coarse annotation granularity. Because of considerable uncertainty of the deconvoluted cell type proportions in preliminary tests and to get an estimate of cell type activity in absolute terms (not composition), we further computed per spot scores indicating the presence or absence of specific cell types based on marker gene panels. Two distinct sources of marker genes were used: the Panglao database ( Franzén et al., 2019 ) and a custom panel of genesets derived from our single-cell reference ( Stubenvoll et al., 2025 ) plus the published melanoma datasets GSE115978 generated by Jerby-Arnon and co-workers ( Jerby-Arnon et al., 2018 ) and GSE130973 generated by Solé-Boldo and co-workers ( Solé-Boldo et al., 2020 ). In short, we used the cell types provided by the authors in each case and, where necessary, added a coarser annotation with collapsed cell subtypes. We then updated gene symbols in order to match datasets, removed untyped cells and used downsampling to a maximum of 5000 cells per type for the Stubenvoll and co-workes dataset to combat cell type imbalance. We then employed deconvolution preprocessing from the BayesPrism ( Chu et al., 2022 ) pipeline to exclude genes with low expression and susceptibility to batch effects and finally used BayesPrism’s get.exp.stat function to conduct pairwise t-tests between all pairs of subtypes from different cell types and combine results to get highly cell type‒specific marker genes for each type across the three references. Finally, we aggregated results from all three references, collecting maximum p-values and minimum log-fold changes across the references each cell type appeared in, then retained the 30 most specific marker genes from the combined data. With the marker gene sets obtained in that way, we used the Seurat function AddModuleScore to compute the average expression minus the expression of a set of randomly selected control genes, stratified by average expression level via binning ( Tirosh et al., 2016 ). We calculated all module scores for size-normalized and non-normalized log-transformed counts. Delineation of the tumor–immune border in Visium samples To determine which spots were marked as positive for immune or tumor cells, we combined the information gained from deconvolution and cell type‒specific module scores. We marked spots as positive for either type only if both approaches support this. For deconvolution results, we applied a cutoff of 0.1 on the estimated cell type fractions to determine spots as positive for a given cell type (the thresholds on the y-axes in Figure 2F ). Because the tumor/immune cell contents and expression levels varied considerably between different samples, we aimed to find a robust model for the module score distributions that delineates spots with strong tumor/immune signatures from background spots without requiring manual determination of thresholds for each sample. For lymphocyte module scores, we could reasonably assume small areas with high expression and a large number of neutral background spots. We therefore modelled the background distribution using the median and Q n as robust estimators of location and scale (using the robustbase R package). We used the inferred location and scale of the background distribution (assumed gaussian) to compute z-scores from our module scores followed by p-values. P-values were then FDR adjusted for the number of spots in each sample. We marked spots with adjusted p-‑values < 0.05 as potentially positive for the cell type in question (this corresponds to the thresholds on the x-axes in Figures 2E and F middle and right ). Because we cannot make this assumption for tumor cells (whose content varied widely between samples), we fitted a custom Gaussian mixture model to the tumor-cell scores in each sample using the mixtools package (3 equal-variance plus one independent component, which empirically worked best). After fitting, the component with highest mean represents the main tumor component, and we determined the threshold for assigning tumor-positive cells as the lowest point where the tumor component density exceeds the aggregate density of all other components (this corresponds to the threshold on the x-axes in Figures 2E and F left ; the tumor component is marked in red and the sum of background components dull green). Spots were finally marked as positive for a given cell type if they surpassed both thresholds on the deconvolved percentage and the module score. We defined the tumor-immune-border region as the collection of spots positive for tumor and immune cells, spots positive for tumor cells that are directly adjacent to spots positive for immune cells, and spots positive for immune cells that are directly adjacent to spots positive for tumor cells (“interfacing spots”, see Figure 2H ). All other spots that were positive for either cell type but not in the border region served as comparison group for enrichment tests later (“non-interfacing”). Spots not enriched for any cell type of interest were marked as background and not used any further. Computation of spatial interaction scores for ligand–receptor pairs on Visium samples To analyze ligand–receptor interactions (LRI) between cells in close proximity, we used the NICHES ( Raredon et al., 2023 ) framework to compute ligand–receptor interaction scores. NICHES computes interaction scores as the product of the expression of the ligand and receptor genes, such that a computed interaction score is always zero if either the ligand or the receptor is not expressed at all and high if both are strongly expressed. This framework incorporates spatial information by allowing ligands and receptors to be expressed in different spots by calculating an interaction score for each directed edge between interacting spots (once for self-edges). For the 10x Genomics Visium® platform, each capture spot had a maximum of six neighboring spots (55 µm center-to-center distance), with fewer neighbors at the edge of the capturing area. Therefore, we restricted interacting spot pairs to those that are physically neighboring, or a spot interacting with itself (self-edges). We used the union of the curated databases of ligand–receptor mechanisms provided in the NicheNet ( Browaeys et al., 2020 ) and LIANA ( Dimitrov et al., 2022 ) packages to determine biologically plausible ligand-receptor interactions for which interaction scores were to be computed. This yielded a total of 14,977 distinct mechanisms after accounting for changed gene symbols and deduplication. Enrichment test for ligand–receptor interactions in Visium samples To test for the specific enrichment of LRI between immune cells and tumor cells in our Visium samples, permutation-based enrichment tests of the NICHES interaction scores were conducted. We calculated the enrichment between spots in the tumor-immune border region (“interfacing”) versus tumor and immune-positive spots not in that border region (“non-interfacing”). Enrichment was calculated using a model-free permutation-based test as implemented by the diff_mean_test function in the SCTransform R package. For this test, a null distribution is built by randomly permuting the labels (in/not in the border region) of all spots in the compared groups and recording the observed difference in arithmetic mean interaction scores between the resulting groups. The mean and standard deviation of the null distribution are subsequently used to turn the real observed difference in mean interaction scores between the compared regions into z-scores and then into p-values. Finally, p-values for all the scored putative ligand–receptor pairings are adjusted using the Benjamini-Hochberg method. We aggregated the ligand–receptor interaction enrichment results across all our Visium samples plus the 3 samples from ( Pozniak et al., 2024) with the most prominent tumor–immune border region. To better attribute the expression of all ligands and receptors to cell types in melanoma samples, we added annotations derived from our single-cell reference. In short, single-cell counts were summed across cells with the same type, then normalized to 10,000 counts per cell and log-transformed with an added pseudocount of 1, to get estimated expression per cell type. Expression of ligand- or receptor complexes was calculated as the geometric mean of the component expressions. Using the geometric mean here has the benefit that a complex ends up with zero expression if any of its components have zero expression, avoiding artifactual expression of complexes that cannot possibly exist in our samples. For visualization, we computed the empirical quantiles in the distribution of expressions and calculated z-scores as quantiles of a normal distribution matching the empirical expression quantiles. To determine the list of top candidates considered for further validation, we retained only candidates that were significantly enriched (adjusted p-value < 0.05) in the tumor–immune border region in 3 or more melanoma samples. Circos plots of LRIs were generated by aggregating all retained ligand–receptor pairs and connecting them in a circular diagram using the circlize R package ( Gu et al., 2014 ). Genes were separated into sections for ligands, receptors and genes that appear in ligand and receptor position in different mechanisms. The link colors for Figure 3C were generated by generating a 3-dimensional UMAP transformation of the expression z-score by cell type matrix, normalizing each component to the unit interval and assigning each axis to one color in RGB space. In this way, genes with similar expression pattern across cell types (as in Figure 3B right ) end up with similar colors in the circos plot and links are colored according to their ligand. Spatial Transcriptomics ‒ 10x Genomics Visium HD Platform Spatial gene expression For Visium HD, formalin-fixed paraffin embedded (FFPE) tissues from seven samples were analyzed. The RNA quality of the embedded samples was checked using the Agilent 5200 fragment analyzer. The DV200 value was at least 48% for primary melanoma and 61% for melanoma metastasis samples. FFPE tissue blocks were sectioned at a thickness of 5 µm using a rotary microtome and placed on glass slides. Hematoxylin and eosin (H&E) staining was performed according to the 10× Genomics protocol. After staining, the slides were imaged at 40× magnification via an AxioScan (Zeiss, Germany). The captured RNA was transferred to Visium HD slides using the CytAssist. cDNA synthesis and library construction were performed using the Visium HD Library Preparation Kit (10× Genomics). Briefly, mRNA was captured by barcoded probes, reverse transcribed into cDNA and amplified for sequencing library preparation Finally,sequencing was performed on a NextSeq 2000 instrument (Illumina) using P4 XLEAP SBS reagent kits (100 cycles) with a 43-10-10-50 read setup. The final molarity of the library pool on the sequencing flow cell was 650 pM including 1% PhiX library. Raw data processing We used Illumina BCL Convert in version 4.1.7 to demultiplex and convert BCL files into FASTQ files. Sequencing data analysis used 10x Genomics SpaceRanger software version 4.0.1, with default parameter settings. For read mapping and gene expression quantification, spaceranger count was performed. Reads were mapped and quantified using transcriptome reference 2024-A Human GRCh38 (GENCODE v44/Ensembl110 annotations) and the probe set Visium Human Transcriptome Probe Set v2.1.0 provided by 10x Genomics. Data analysis, quality control and preprocessing The Visium HD processed samples have been analyzed by first applying stain deconvolution to the histological microscopy images. Stain deconvolution was performed using scikit-image ( van der Walt et al., 2014 ) to separate the hematoxylin and eosin color channels. The images with the hematoxylin color channel are used as input to the bin2cell workflow for cell segmentation using stardist ( Weigert and Schmidt, 2022 ) and bin count aggregation. The cell by gene count matrix is stored as anndata ( Virshup et al., 2024 ) object. Cell type label transfer was performed using the TACCO OT (optimal transport) method. As input for the cell type label transfer, the dataset from Stubenvoll and co-workers ( Stubenvoll et al., 2025 ) was used. This dataset contains single-cell RNA sequencing data from 10 primary melanoma samples of which pm6 is paired with HD_PM1, pm7 is paired with HD_PM2, pm9 is paired with HD_PM3 and pm10 is paired with HD_PM4. We repeated label transfer using two different annotation levels of the reference (with and without cell subtypes) and removed spots with non-matching combinations of coarse and fine annotation. Extended Figure 7 shows the transcript distribution for some genes of interest. In these plots, cells that contain at least a single read of a gene are plotted as a spot. Zoomed-in plots show LGALS1 and PTPRC expressing cells at identical locations as zoomed in plots in Figure 5E and Supplementary Figures 5A-F . Ligand–receptor analysis of Visium HD samples We used the preprocessed and annotated Visium HD data to support our analysis with direct evidence of the interacting cell types through the superior resolution of the Visium HD technology. After removing low-quality cells (at least 5 and less than 150 bins per cell, at least 40 gene counts), we annotated the tumor–immune border region by calculating, for each cell, the distances to closes tumor and T/NK cell and then assigned classes based on a distance threshold of 50 µm. We used all spots in the border region to run the whole LIANA analysis pipeline using the same database of ligand–receptor interactions that we used for the analysis of the Visium data. Aggregated p-values from the pipeline output were filtered with a cutoff of 0.01 and retained results aggregated and visualized with custom functions based on the LIANA plotting functions. To pinpoint interactions directly, we additionally used the NICHES package with slight modifications to the code to ensure feasible runtime and memory usage due to the high number of cells in the HD samples. We again used 50 µm as the distance cutoff for interactions in NICHES. Average interaction values for each cell were calculated for LGALS1–PTPRC by averaging all adjacent edges for each cell. Visualizations for these data were done based on the ggplot2 and ggraph packages. Multicolor immunofluorescence Formalin-fixed, paraffin-embedded (FFPE) sections (7 µm) of a total set of 20 malignant melanoma samples were mounted on charged microscopy slides, were used for immunofluorescence (IF) staining. After deparaffinization, heat-mediated antigen retrieval was performed by placing the slides in hot citrate buffer (pH 6.0) for 45 minutes. The sections were then blocked with 10% fetal bovine serum (FBS) for 45 minutes before applying the primary antibodies at the appropriate concentrations. Following overnight incubation, the secondary antibodies (1:500) and DAPI (1:500) were applied, followed by mounting with Prolong™ Gold antifade mountant. The following antibodies and concentrations were used: CoraLite®594-conjugated mouse anti-SOX10 (1:200) (Proteintech Group, Rosemont, IL, USA, CL594-66786, 1D2C8), CoraLite®488-conjugated mouse anti-galectin-1 (1:400) (Proteintech Group, Rosemont, IL, USA, CL488-60223, 3G10D2), AlexaFluor®647-conjugated rabbit anti-CD45 (1:400) (Abcam, Cambridge, U.K., ab305209, EPR20033), and rabbit anti-CD3 (Novus Biologicals, Centennial, CO, U.S.A., NB600-1441, SP7). Secondary antibodies included goat anti-rabbit Alexa Fluor 647 (Invitrogen, Darmstadt, Germany, A21245) Fluorescence microscopy was performed at 20x and 40x magnification. Images were captured in multiple fluorescence channels, and JPEG formats were saved for subsequent analysis. Analysis was conducted using QuPath (version 0.5.1). For the quantification of CD3 + cells to determine immune status, a minimum of three 20x images per section were analyzed. One to three representative 20x images were used for galectin-1‑CD45 distance measurements. Positive cell detection was based on DAPI staining to identify nuclei, and measurements of marker expression were extracted from the corresponding color channel (RGB). Thresholds were set manually for each marker to ensure accurate detection and quantification and were applied across all images. For cell distance measurements, image segmentation results based on anti-CD45, anti-SOX10, and anti-galectin-1 staining were exported from QuPath as .tsv files and subsequently processed in Python (v3.8.8) using pandas v1.5.3, numpy v1.24.2, matplotlib v3.7.1, seaborn v0.13.2, and scipy v1.10.1 (spatial analysis via distance_matrix and cKDTree). CD3 quantification was performed to classify the melanoma samples according to their immune status, based on the ratio of CD3⁺ positive cells to the total number of cells. Tumors were classified based on CD3⁺ cell infiltration as follows: 'cold' tumors contained ≤10% CD3⁺ cells, 'intermediate' tumors had 10–20% CD3⁺ cells, and 'hot' tumors had >20% CD3⁺ cells. Group differences (cold, intermediate, hot) were assessed using Welch’s ANOVA following a Brown-Forsythe test for variance heterogeneity. Dunnett’s T3 post hoc test was used for multiple comparisons. Statistical significance was defined as p < 0.05 (*), p < 0.01 (), and p < 0.001 (*) Determination of protein-protein Interactions Expression and purification of recombinant LGALS1 and LGALS3 Gene sequences for LGALS1 and LGALS3 were obtained through custom synthesis from IDT as eBlocks and were subsequently cloned into pET21a (+) vector via Gibson Assembly. Proteins were expressed with an N-terminal his-tag from E. coli BL21 (DE3) cultures. Protein overproduction was induced at OD ‍~ 0.7 with a final concentration of 50 µM isopropyl-β-D-thiogalactopyranoside (IPTG). Cells were harvested through centrifugation at 4 °C after 4 h of incubation at 37 °C and 210 rpm and lysed by sonication. The clarified lysate was purified via immobilized metal affinity chromatography (IMAC) (Ni‑NTA Agarose column) (Quiagen, Hilden, Germany) followed by size-exclusion chromatography (SEC) over a Superdex 75 pg gel filtration column (Cytiva, Marlborough, MA, U.S.A.). LGALS1 purification was performed under reducing conditions. All Elution and washing reagents for reducing purification contained 8 mM Dithiothreitol (DTT) and the imidazole containing elution solution contained 10 mM 2‑Mercaptoethanol (BME). Determination of binding affinity with bio-layer interferometry (BLI) of PTPRC and LGALS1 and LGALS3 Affinity measurements were conducted in an Octet® R8 system with Octet® ProA Biosensors (Sartorius) at 27 °C. Samples were diluted in HBS-EP+ buffer (0.01 M 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid, 0.15 M NaCl, 3 mM ethylenediaminetetraacetic acid, 0.05% Tween 20, 1% bovine serum albumin, pH 7.4). T1 Anti-NY-ESO-1 Antibody was obtained after expression in human derived HEK Expi 293 cells and subsequent Protein A affinity purification and buffer exchange. Fc-‑tagged CD45 (SinoBiology, 10086-H02H) and the control antibody were diluted to 5 µg/mL and loaded onto the ProA Biosensors. The tips with bound protein were washed in HBS-EP+ buffer for 60 s and separately transferred to LGALS1 or LGALS3 solutions with concentrations ranging from 3000 nm to 0 nM for an association period of 120 s, followed by dissociation in HBS-EP+ buffer for 120 s. Tips were regenerated with 10 mM glycine pH 1.5. Data Processing and determination of dissociation constant K D Data processing was performed in Octet® BLI Analysis version 12. Preprocessing of the raw data involved subtraction of reference well, sensor axis alignment and Savitzky-Golay-Filtering. Data was then fitted to 1:1 binding curves with global fitting in kinetic analysis allowing different R max to ensure best fitting model. The steady state affinity was determined using the theoretical equilibrium binding response data (R eq ) at the end of the association. Ordered amino acid sequences: Galectin-1: HHHHHHMACGLVASNLNLKPGECLRVRGEVAPDAKSFVLNLGKDSNNLCLHFNPRFNAHGDANTIVCNSKDGGAWGTEQREAVFPFQPGSVAEVCITFDQANLTVKLPDGYEFKFPNRLNLEAINYMAADGDFKIKCVAFD Galectin-3: HHHHHHMFHILRLESTVDLSEPLKDNGIIVFQSDKLDLEPSPNLGPTGIDNTNVNLINAKGDVLLHIGIRRRENAFVFNSIPYGESRGPEERIPLEGTFGDRRDPSITIFDHPDRYQIMIDYKTVYYYKKRLEGRCEKVSYKINEGQTPPFSDVLGVTVLYFANVMPRAN In-silico structure prediction To identify a possible interaction interface between galectin-1 and CD45 first, we searched in the Human PPI database created by ( Zhang et al., 2025 ) for existing complex predictions. They used a striped version of CD45 (P08575_S1:226-575) without parts of the extracellular, membrane-spanning, and intracellular domains (model ID: P08575_S1__P09382_S0), sequence see Supplementary Table ‍4A , ID P08575_S1) with a contact probability of 0.06573 (Alphafold2 prediction). The model is shown in Supplementary Figure 4A , with galectin-1 positioned inside the membrane, indicating a failed complex prediction. To remove the bias for the galectin-1 position predicted at the lower end of CD45, the membrane spanning region and parts of the intracellular domain were introduced in a new structure prediction run, using Boltz2 ( Passaro et al., 2025 ) in template mode for CD45 with enabled potentials, 20 predictions, and 10 seeds. The model showed that galectin-1 was no longer predicted at the lower end of the extracellular domain, but rather at the intracellular part of CD45 ( Supplementary Figure 4B ) with acceptable overall model confidence but low interface reliability (see Supplementary Table 4B , Pair ID: P08575_strip + Gal1). Consequently, the intracellular part in the sequence was removed, and a new model was generated with the same settings and seeds, resulting in a high-confidence structural model with near-high-quality predicted relative positions between the extracellular domains of CD45 and galectin-1 (see Supplementary Figure 4C-D and Supplementary Table 4B , Pair ID: P08575_stripIntra + Gal1). Even though not all models display these metrics, it is still possible to identify some highly confident (confidence score ≥ 0.909, ipTM ≥ 0.759) models in this binding interface ( Supplementary Figure 4D-F ). Survival analysis The possible impact of selected ligand–receptor interaction partners on the prognosis of melanoma patients was analyzed. For this purpose, we collected the skin cancer cohort data from The Cancer Genome Atlas (TCGA ; Skin Cancer Melanoma (SKCM) cohort) from The Cancer Immunome Atlas (TCIA, http://tcia.at) webserver, which enhances the TCGA data by providing estimates of cell type proportions using the quanTIseq deconvolution algorithm ( Finotello et al., 2019 ). RNA-seq data from TCGA were obtained from the recount2 database ( Collado-Torres et al., 2017a ). Gene expression values and sample metadata were downloaded via the R/Bioconductor package recount ( Collado-‍Torres et al., 2017b ). The expression values were normalized to transcripts per million (TPM) using the getTPM function. Additionally, another cohort of melanoma patients published by Gide and co-workers was downloaded ( Gide et al., 2019 ) from the Sequence Read Archive (SRA) and used quanTIseq for the deconvolution of these samples. We examined the effect of LGALS1 and PTPRC expression on patient survival using the survival and survminer R packages to conduct Kaplan–Meyer survival analyses and log-rank tests. After brief analysis of the mayor subtypes in each cohort (primary melanoma/metastasis for TCGA, primary treatment in the Gide cohort), we retained only the larger, more homogenous group in each case (metastases for TCGA, Pembrolizumab-treated patients for the Gide cohort) and additionally excluded infiltration-free samples from the TCGA cohort (zero CD8 + T cell content as per quanTiseq estimates). In all cases, we separated LGALS1 , PTPRC and CD8 + T cell content into high and low categories based on the group median. For the TCGA cohort, we additionally split the CD8 + T cell subgroups first and then calculated LGALS1 and PTPRC high/low labels based on the subgroup medians again. For the Gide cohort, this split was attempted but led to the subgroups of interest not including any events due to the low total sample size. Declarations Conflicts of interest: M. Kunz received honoraria from the Speaker Bureau of Roche Pharma and travel support from Novartis Pharma GmbH and Bristol-Myers Squibb GmbH. J.C. Simon obtained speaker fees from Bristol-Myers Squibb, Roche Pharma AG, Novartis, and MSD Sharp & Dome, as well as financial support for congress attendance from Bristol-Myers Squibb, MSD Sharp & Dome, and Novartis. M. Ziemer received lecture fees from Bristol-Myers Squibb, MSD Sharp & Dohme GmbH, Pfizer Pharma GmbH, and Sanofi-Aventis Deutschland GmbH, received financial support for congress participation from Bristol-Myers Squibb, and serves as a member of expert panels on cutaneous adverse reactions for Pfizer INC. K. Reiche received honoraria from Novartis Pharma GmbH. Data and Code Availability The datasets generated and/or analyzed during the current study are available in the NCBI GEO (https://www.ncbi.nlm.nih.gov/geo/) repository under accession number GSE314509[1] (Visium HD datasets; access token: obktegagrtmlrax) and GSE314538[2] (Visium datasets; access token: ejwfqeimzholzgx).The code used to analyze the data and generate the figures included in this study is available at https://github.com/fraunhofer-izi/Grosse_et_al_2026. Author Contributions F. Große: Conceptualization, data curation, formal analysis, investigation, methodology, software, visualization, and writing of the original draft C. Kämpf: Data curation, formal analysis, investigation, methodology, software, visualization, and writing of the original draft D. Löffler: Investigation, methodology, validation and visualization M. Lingner Chango: Data curation, investigation, validation, visualization C. Blumert: Data curation, formal analysis, funding acquisition, methodology A. Scholz: Data curation, formal analysis A. Stubenvoll: Data curation, investigation, validation, visualization H. Löffler-Wirth: Data curation, formal analysis, software, supervision H. Binder: Conceptualization, formal analysis, software, supervision C. Schultz: Data curation, investigation, validation, visualization M. Beining: Data curation, investigation, modeling, visualization J. C Simon: Resources, supervision, validation, writing – review & editing C. T. Schoeder: Data curation, investigation, validation, visualization M. Ziemer: Resources, investigation, writing – review & editing K. Reiche: Conceptualization, data curation, funding acquisition, investigation, project, administration, supervision, writing – original draft, review and editing. M. Kunz: Conceptualization, data curation, funding acquisition, investigation, project administration, resources, supervision, writing – original draft, writing – review, and editing. Acknowledgements: This work was supported by the Sächsische Aufbaubank, grant number 100714507, to M. Kunz, and grant number 100714512, to K. Reiche and C. Blumert . Further, the authors acknowledge the financial support by the Federal Ministry of Research, Technology and Space of Germany and by Sächsische Staatsministerium für Wissenschaft, Kultur und Tourismus in the program Center of Excellence for AI-research „Center for Scalable Data Analytics and Artificial Intelligence Dresden/Leipzig“, project identification number: ScaDS.AI. The results of the survival analyses are based in parts upon data generated by the TCGA Research Network (https://www.cancer.gov/tcga). [1] https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE314509 [2] https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE314538 References Ashburner M , Ball CA, Blake JA, et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet. 2000;25(1):25-29. doi:10.1038/75556 Becht E , McInnes L, Healy J, Dutertre CA, Kwok IWH, Ng LG, Ginhoux F, Newell EW. Dimensionality reduction for visualizing single-cell data using UMAP. Nat Biotechnol. 2018. doi: 10.1038/nbt.4314. Benjamini Y, Hochberg Y . Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. J R Statist Soc 1995 (57), 289-300. http://www.jstor.org/stable/2346101 Bhuva DD , Tan CW, Salim A, Marceaux C, Pickering MA, Chen J, Kharbanda M, Jin X, Liu N, Feher K, Putri G, Tilley WD, Hickey TE, Asselin-Labat ML, Phipson B, Davis MJ. Library size confounds biology in spatial transcriptomics data. Genome Biol 2024; 25(1), 99 https://doi.org/10.1186/s13059-024-03241-7. Blondel VD , Guillaume, JL, Lambiotte, R,Lefebvre, E. Fast unfolding of communities in large networks. In: J. Stat. Mech. 2008 (10), P10008. DOI: 10.1088/1742-5468/2008/10/P10008. Browaeys R , Saelens W, Saeys Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat Methods. 2020;17(2):159-162. doi: 10.1038/s41592-019-0667-5. Cable DM , Murray E, Zou LS, Goeva A, Macosko EZ, Chen F, Irizarry RA. Robust decomposition of cell type mixtures in spatial transcriptomics. Nat Biotechnol. 2022 Apr;40(4):517-526. doi: 10.1038/s41587-021-00830-w. Cancer Genome Atlas Network . Genomic Classification of Cutaneous Melanoma. Cell. 2015;161(7):1681-96. doi: 10.1016/j.cell.2015.05.044 Cedeno-Laurent F , Opperman M, Barthel SR, Kuchroo VK, Dimitroff CJ. Galectin-1 triggers an immunoregulatory signature in Th cells functionally defined by IL-10 expression. J Immunol 2012; 188:3127-3137. doi: 10.4049/jimmunol.1103433. Choudhary S , Satija R. Comparison and evaluation of statistical error models for scRNA-seq. Genome Biol. 2022; 23(1):27.https://doi.org/10.1186/s13059-021-02584-9 Chu T , Wang Z, Pe'er D, Danko CG. Cell type and gene expression deconvolution with BayesPrism enables Bayesian integrative analysis across bulk and single-cell RNA sequencing in oncology. Nat Cancer. 2022;3(4):505-517. doi: 10.1038/s43018-022-00356-3. Collado-Torres L, Nellore A, Kammers K, Ellis SE, Taub MA, Hansen KD, Jaffe AE, Langmead B, Leek JT.Reproducible RNA-seq analysis using recount2. Nat Biotechnol. 2017a;35(4):319-321. doi:10.1038/nbt.3838 Collado-Torres L , Nellore A, Jaffe AE. recount workflow: Accessing over 70,000 human RNA-seq samples with Bioconductor. F1000Res. 2017b;6:1558. doi: 10.12688/f1000research.12223.1. de Azevedo RA , Shoshan E, Whang S, Markel G, Jaiswal AR, Liu A, Curran MA, Travassos LR, Bar-Eli M. MIF inhibition as a strategy for overcoming resistance to immune checkpoint blockade therapy in melanoma. Oncoimmunology. 2020;9(1):1846915. doi: 10.1080/2162402X.2020.1846915 Dimitrov D, Türei D, Garrido-Rodriguez M, Burmedi PL, Nagai JS, Boys C, Ramirez Flores RO, Kim H, Szalai B, Costa IG, Valdeolivas A, Dugourd A, Saez-Rodriguez J. Comparison of methods and resources for cell-cell communication inference from single-cell RNA-Seq data. Nat Commun. 2022;13(1):3224. doi:10.1038/s41467-022-30755-0 Ding Q , Lu P, Xia Y, Ding S, Fan Y, Li X, Han P, Liu J, Tian D, Liu M. CXCL9: evidence and contradictions for its role in tumor progression. Cancer Med. 2016;5(11):3246-3259. doi: 10.1002/cam4.934. Femel J , van Hooren L, Herre M, Cedervall J, Saupe F, Huijbers EJM, Verboogen DRJ, Reichel M, Thijssen VL, Griffioen AW, Hellman L, Dimberg A, Olsson AK. Vaccination against galectin-1 promotes cytotoxic T-cell infiltration in melanoma and reduces tumor burden. Cancer Immunol Immunother. 2022;71(8):2029-2040. doi: 10.1007/s00262-021-03139-4. Figueiredo CR, Azevedo RA, Mousdell S, Resende-Lara PT, Ireland L, Santos A, Girola N, Cunha RLOR, Schmid MC, Polonelli L, Travassos LR, Mielgo A. Blockade of MIF-CD74 on Macrophages and Dendritic Cells Restores the Antitumour Immune Response Against Metastatic Melanoma. Front Immunol. 2018;9:1132. doi: 10.3389/fimmu.2018.01132. Finotello F , Mayer C, Plattner C, Laschober G, Rieder D, Hackl H, Krogsdam A, Loncova Z, Posch W, Wilflingseder D, Sopper S, Ijsselsteijn M, Brouwer TP, Johnson D, Xu Y, Wang Y, Sanders ME, Estrada MV, Ericsson-Gonzalez P, Charoentong P, Balko J, de Miranda NFDCC, Trajanoski Z. Molecular and pharmacological modulators of the tumor immune contexture revealed by deconvolution of RNA-seq data. Genome Med. 2019;11(1):34. doi: 10.1186/s13073-019-0638-6. Frangieh CJ, Melms JC, Thakore PI, Geiger-Schuller KR, Ho P, Luoma AM, Cleary B, Jerby-Arnon L, Malu S, Cuoco MS, Zhao M, Ager CR, Rogava M, Hovey L, Rotem A, Bernatchez C, Wucherpfennig KW, Johnson BE, Rozenblatt-Rosen O, Schadendorf D, Regev A, Izar B. Multimodal pooled Perturb-CITE-seq screens in patient models define mechanisms of cancer immune evasion. Nat Genet. 2021;53(3):332-341. doi: 10.1038/s41588-021-00779-1. Franzén O , Gan LM, Björkegren JLM. PanglaoDB: a web server for exploration of mouse and human single-cell RNA sequencing data. Database (Oxford). 2019;2019:baz046. doi:10.1093/database/baz046 Fu Q, Chen N, Ge C, Li R, Li Z, Zeng B, Li C, Wang Y, Xue Y, Song X, Li H, Li G. Prognostic value of tumor-infiltrating lymphocytes in melanoma: a systematic review and meta-analysis. Oncoimmunology. 2019 Apr 3;8(7):1593806. doi: 10.1080/2162402X.2019.1593806 Gide TN , Quek C, Menzies AM, Tasker AT, Shang P, Holst J, Madore J, Lim SY, Velickovic R, Wongchenko M, Yan Y, Lo S, Carlino MS, Guminski A, Saw RPM, Pang A, McGuire HM, Palendira U, Thompson JF, Rizos H, Silva IPD, Batten M, Scolyer RA, Long GV, Wilmott JS. Distinct Immune Cell Populations Define Response to Anti-PD-1 Monotherapy and Anti-PD-1/Anti-CTLA-4 Combined Therapy. Cancer Cell. 2019;35(2):238-255.e6. doi: 10.1016/j.ccell.2019.01.003. Gilson RC , Gunasinghe SD, Johannes L, Gaus K. Galectin-3 modulation of T-cell activation: mechanisms of membrane remodelling. Prog Lipid Res. 2019;76:101010. doi: 10.1016/j.plipres.2019.101010. Gowhari Shabgah A , Al-Obaidi ZMJ, Sulaiman Rahman H, Kamal Abdelbasset W, Suksatan W, Bokov DO, Thangavelu L, Turki Jalil A, Jadidi-Niaragh F, Mohammadi H, Mashayekhi K, Gholizadeh Navashenaq J. Does CCL19 act as a double-edged sword in cancer development? Clin Exp Immunol. 2022;207(2):164-175. doi: 10.1093/cei/uxab039. PMID: 35020885; PMCID: PMC8982982. Gu Z , Gu L, Eils R, Schlesner M, Brors B. circlize Implements and enhances circular visualization in R. Bioinformatics. 2014;30(19):2811-2. doi: 10.1093/bioinformatics/btu393. Epub 2014 Jun 14. Hao Y , Hao S, Andersen-Nissen E, Mauck WM 3rd, Zheng S, Butler A, Lee MJ, Wilk AJ, Darby C, Zager M, Hoffman P, Stoeckius M, Papalexi E, Mimitou EP, Jain J, Srivastava A, Stuart T, Fleming LM, Yeung B, Rogers AJ, McElrath JM, Blish CA, Gottardo R, Smibert P, Satija R.Integrated analysis of multimodal single-cell data. Cell. 2021;184(13):3573-3587.e29. doi:10.1016/j.cell.2021.04.048 Hao Y , Stuart T, Kowalski MH, Choudhary S, Hoffman P, Hartman A, Srivastava A, Molla G, Madad S, Fernandez-Granda C, Satija R.Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat Biotechnol 2024;42; 293–304.https://doi.org/10.1038/s41587-023-01767-y Hunter MV, Moncada R, Weiss JM, Yanai I, White RM. Spatially resolved transcriptomics reveals the architecture of the tumor-microenvironment interface. Nat Commun. 2021;12(1):6278. doi: 10.1038/s41467-021-26614-z Ilarregui JM , Croci DO, Bianco GA, Toscano MA, Salatino M, Vermeulen ME, Geffner JR, Rabinovich GA. Tolerogenic signals delivered by dendritic cells to T cells through a galectin-1-driven immunoregulatory circuit involving interleukin 27 and interleukin 10. Nat Immunol. 2009;10(9):981-91. doi: 10.1038/ni.1772. Jacquelot N , Pitt JM, Enot DP, Roberti MP, Duong CPM, Rusakiewicz S, Eggermont AM, Zitvogel L. Immune biomarkers for prognosis and prediction of responses to immune checkpoint blockade in cutaneous melanoma. Oncoimmunology. 2017 Mar 7;6(8):e1299303. doi: 10.1080/2162402X.2017.1299303. Jacquelot N , Duong CPM, Belz GT, Zitvogel L. Targeting Chemokines and Chemokine Receptors in Melanoma and Other Cancers. Front Immunol. 2018;9:2480. doi: 10.3389/fimmu.2018.02480. Jerby-Arnon L , Shah P, Cuoco MS, Rodman C, Su MJ, Melms JC, Leeson R, Kanodia A, Mei S, Lin JR, Wang S, Rabasha B, Liu D, Zhang G, Margolais C, Ashenberg O, Ott PA, Buchbinder EI, Haq R, Hodi FS, Boland GM, Sullivan RJ, Frederick DT, Miao B, Moll T, Flaherty KT, Herlyn M, Jenkins RW, Thummalapalli R, Kowalczyk MS, Cañadas I, Schilling B, Cartwright ANR, Luoma AM, Malu S, Hwu P, Bernatchez C, Forget MA, Barbie DA, Shalek AK, Tirosh I, Sorger PK, Wucherpfennig K, Van Allen EM, Schadendorf D, Johnson BE, Rotem A, Rozenblatt-Rosen O, Garraway LA, Yoon CH, Izar B, Regev A.A Cancer Cell Program Promotes T Cell Exclusion and Resistance to Checkpoint Blockade. Cell. 2018;175(4):984-997.e24. doi:10.1016/j.cell.2018.09.006 Karras P , Bordeu I, Pozniak J, Nowosad A, Pazzi C, Van Raemdonck N, Landeloos E, Van Herck Y, Pedri D, Bervoets G, Makhzami S, Khoo JH, Pavie B, Lamote J, Marin-Bejar O, Dewaele M, Liang H, Zhang X, Hua Y, Wouters J, Browaeys R, Bergers G, Saeys Y, Bosisio F, van den Oord J, Lambrechts D, Rustgi AK, Bechter O, Blanpain C, Simons BD, Rambow F, Marine JC. A cellular hierarchy in melanoma uncouples growth and metastasis. Nature. 2022;610(7930):190-198. doi: 10.1038/s41586-022-05242-7. Laidlaw BJ , Gray EE, Zhang Y, Ramírez-Valle F, Cyster JG. Sphingosine-1-phosphate receptor 2 restrains egress of γδ T cells from the skin. J Exp Med. 2019;216(7):1487-1496. doi: 10.1084/jem.20190114 Larsson L , Frisén J, Lundeberg J. Spatially resolved transcriptomics adds a new dimension to genomics. Nat Methods. 2021;18(1):15-18. Li H , van der Leun AM, Yofe I, Lubling Y, Gelbard-Solodkin D, van Akkooi ACJ, van den Braber M, Rozeman EA, Haanen JBAG, Blank CU, Horlings HM, David E, Baran Y, Bercovich A, Lifshitz A, Schumacher TN, Tanay A, Amit I.Dysfunctional CD8 T Cells Form a Proliferative, Dynamically Regulated Compartment within Human Melanoma. Cell. 2020;181(3):747. doi: 10.1016/j.cell.2020.04.017. Li Pomi F , Borgia F, Custurone P, Vaccaro M, Pioggia G, Gangemi S. Role of HMGB1 in Cutaneous Melanoma: State of the Art. Int J Mol Sci. 2022;23(16):9327. doi: 10.3390/ijms23169327. Li X , Yue Z, Wang D, Zhou L. PTPRC functions as a prognosis biomarker in the tumor microenvironment of cutaneous melanoma. Sci Rep. 2023;13(1):20617. doi: 10.1038/s41598-023-46794-6. Lim B , Lin Y, Navin N. Advancing Cancer Research and Medicine with Single-Cell Genomics. Cancer Cell. 2020 Apr 13;37(4):456-470. doi: 10.1016/j.ccell.2020.03.008. Lim SY , Shklovskaya E, Lee JH, Pedersen B, Stewart A, Ming Z, Irvine M, Shivalingam B, Saw RPM, Menzies AM, Carlino MS, Scolyer RA, Long GV, Rizos H. The molecular and functional landscape of resistance to immune checkpoint blockade in melanoma. Nat Commun. 2023;14(1):1516. doi: 10.1038/s41467-023-36979-y. Ludin A , Stirtz GL, Tal A, Nirmal AJ, Pfaff KL, Manos M, Besson N, Eskndir N, Porter B, Jones SM, Faulkner HM, Gong Q, Liu S, Barrera I, Wu L, Pessoa Rodrigues C, Sahu A, Jerison E, Alessi JV, Ricciuti B, Richardson DS, Weiss JD, Moreau HM, Stanhope ME, Afeyan AB, Sefton J, McCall WD, Formato E, Yang S, Zhou Y, Hoytema van Konijnenburg DP, Cole HL, Cordova M, Deng L, Rajadhyaksha M, Quake SR, Awad MM, Chen F, Wucherpfennig KW, Sorger PK, Hodi FS, Rodig SJ, Murphy GF, Zon LI. CRATER tumor niches facilitate CD8+ T cell engagement and correspond with immunotherapy success. Cell. 2025;188(24):6720-6736.e26. doi: 10.1016/j.cell.2025.09.021. Ma C, Chen J, Ji J, Zheng Y, Liu Y, Wang J, Chen T, Chen H, Chen Z, Zhou Q, Hou C, Ke Y. Therapeutic modulation of APP-CD74 axis can activate phagocytosis of TAMs in GBM. Biochim Biophys Acta Mol Basis Dis. 2024;1870(8):167449. doi: 10.1016/j.bbadis.2024.167449. Ma Y , Zhou X. Spatially informed cell-type deconvolution for spatial transcriptomics. Nat Biotechnol. 2022 Sep;40(9):1349-1359. doi: 10.1038/s41587-022-01273-7. Mages S , Moriel N, Avraham-Davidi I, Murray E, Watter J, Chen F, Rozenblatt-Rosen O, Klughammer J, Regev A, Nitzan M.TACCO unifies annotation transfer and decomposition of cell identities for single-cell and spatial omics. Nat Biotechnol. 2023 Oct;41(10):1465-1473. doi: 10.1038/s41587-023-01657-3. Mochida Y , Ohuchida K, Zhang B, Yamada Y, Tsutsumi C, Kubo A, Oyama K, Shinkawa T, Iwamoto C, Torata N, Abe T, Ideno N, Ikenaga N, Nakata K, Oda Y, Nakamura M. Identification of cold tumor induction-related markers in pancreatic cancer and the clinical implication of PCDH7. J Cancer Res Clin Oncol. 2025;151(2):45. doi: 10.1007/s00432-025-06095-z. Novák J , Takács T, Tilajka Á, László L, Oravecz O, Farkas E, Than NG, Buday L, Balogh A, Vas V. The sweet and the bitter sides of galectin-1 in immunity: its role in immune cell functions, apoptosis, and immunotherapies for cancer with a focus on T cells. Semin Immunopathol. 2025;47(1):24. doi: 10.1007/s00281-025-01047-8 Ostendorf BN , Bilanovic J, Adaku N, Tafreshian KN, Tavora B, Vaughan RD, Tavazoie SF. Common germline variants of the human APOE gene modulate melanoma progression and survival. Nat Med. 2020; 26(7):1048-1053. doi: 10.1038/s41591-020-0879-3. Passaro S , Corso G, Wohlwend J, Reveiz M, Stephan Thaler S, Somnath VR, Getz N, Portnoi T, Roy J, Stark H, Kwabi-Addo D, Beaini D, Jaakkola T, Barzilay R. Boltz-2: Towards Accurate and Efficient Binding Affinity Prediction. bioRxiv [Preprint]. 2025 Jun 18:2025.06.14.659707. doi: 10.1101/2025.06.14.659707. Polański K , Bartolomé-Casado R, Sarropoulos I, Xu C, England N, Jahnsen FL, Teichmann SA, Yayon N. Bioinformatics. 2024 Sep 2;40(9):btae546. doi: 10.1093/bioinformatics/btae546 Pozniak J , Pedri D, Landeloos E, Van Herck Y, Antoranz A, Vanwynsberghe L, Nowosad A, Roda N, Makhzami S, Bervoets G, Maciel LF, Pulido-Vicuña CA, Pollaris L, Seurinck R, Zhao F, Flem-Karlsen K, Damsky W, Chen L, Karagianni D, Cinque S, Kint S, Vandereyken K, Rombaut B, Voet T, Vernaillen F, Annaert W, Lambrechts D, Boecxstaens V, Saeys Y, van den Oord J, Bosisio F, Karras P, Shain AH, Bosenberg M, Leucci E, Paschen A, Rambow F, Bechter O, Marine JC. A TCF4-dependent gene regulatory network confers resistance to immunotherapy in melanoma. Cell. 2024;187(1):166-183.e25. doi: 10.1016/j.cell.2023.11.037 R Core Team . R: A language and environment for statistical computing. R Foundation for Statistical Computing 2021, Vienna, Austria. URL: https://www.R-project.org/. Rabinovich GA , Ramhorst RE, Rubinstein N, Corigliano A, Daroqui MC, Kier-Joffé EB, Fainboim L. Induction of allogenic T-cell hyporesponsiveness by galectin-1-mediated apoptotic and non-apoptotic mechanisms. Cell Death Differ. 2002;9(6):661-70. doi: 10.1038/sj.cdd.4401009 Rambow F , Rogiers A, Marin-Bejar O, Aibar S, Femel J, Dewaele M, Karras P, Brown D, Chang YH, Debiec-Rychter M, Adriaens C, Radaelli E, Wolter P, Bechter O, Dummer R, Levesque M, Piris A, Frederick DT, Boland G, Flaherty KT, van den Oord J, Voet T, Aerts S, Lund AW, Marine JC.Toward Minimal Residual Disease-Directed Therapy in Melanoma. Cell. 2018; 174(4):843-855.e19, doi: 10.1016/j.cell.2018.06.025. Raredon MSB , Yang J, Kothapalli N, Lewis W, Kaminski N, Niklason LE, Kluger Y.Comprehensive visualization of cell-cell interactions in single-cell and spatial transcriptomics with NICHES. Bioinformatics. 2023;39(1):btac775. doi:10.1093/bioinformatics/btac775 Saad M , Castellano E, Tarhini AA. Clinical updates in neoadjuvant immunotherapy for melanoma before surgery. Expert Rev Clin Immunol. 2024 Aug;20(8):927-943 doi: 10.1080/1744666X.2023.2248392. Sade-Feldman M , Yizhak K, Bjorgaard SL, Ray JP, de Boer CG, Jenkins RW, Lieb DJ, Chen JH, Frederick DT, Barzily-Rokni M, Freeman SS, Reuben A, Hoover PJ, Villani AC, Ivanova E, Portell A, Lizotte PH, Aref AR, Eliane JP, Hammond MR, Vitzthum H, Blackmon SM, Li B, Gopalakrishnan V, Reddy SM, Cooper ZA, Paweletz CP, Barbie DA, Stemmer-Rachamimov A, Flaherty KT, Wargo JA, Boland GM, Sullivan RJ, Getz G, Hacohen N.Defining T Cell States Associated with Response to Checkpoint Immunotherapy in Melanoma. Cell. 2019 Jan 10;176(1-2):404. doi: 10.1016/j.cell.2018.12.034. Saiselet M , Rodrigues-Vitória J, Tourneur A, Craciun L, Spinette A, Larsimont D, Andry G, Lundeberg J, Maenhaut C, Detours V.Transcriptional output, cell-type densities, and normalization in spatial transcriptomics. J Mol Cell Biol. 2020;12(11):906-908. doi:10.1093/jmcb/mjaa028 Schadendorf D , van Akkooi ACJ, Berking C, Griewank KG, Gutzmer R, Hauschild A, Stang A, Roesch A, Ugurel S. Melanoma. Lancet 2018; 392(10151):971-984. doi: 10.1016/S0140-6736(18)31559-9. Seferbekova Z , Lomakin A, Yates LR, Gerstung M. Spatial biology of cancer evolution. Nat Rev Genet. 2023 May;24(5):295-313. doi: 10.1038/s41576-022-00553-x Stubenvoll A , Schmidt M, Moeller J, Chango MAL, Schultz C, Antoniadou O, Loeffler-Wirth H, Bernhart S, Große F, Thier B, Paschen A, Anderegg U, Simon JC, Ziemer M, Schoeder CT, Binder H, Kunz M. Single-cell transcriptomics and epigenomics point to CD58-CD2 interaction in controlling primary melanoma growth and immunity. Cancer Commun (Lond). 2025;45(4):465-470. doi: 10.1002/cac2.12651 Solé-Boldo L , Raddatz G, Schütz S, Mallm JP, Rippe K, Lonsdorf AS, Rodríguez-Paredes M, Lyko F. Single-cell transcriptomes of the human skin reveal age-related loss of fibroblast priming. Commun Biol. 2020;3(1):188. doi: 10.1038/s42003-020-0922-4. Subramanian A , Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP.Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005;102(43):15545-50. doi: 10.1073/pnas.0506580102. Tawbi HA , Schadendorf D, Lipson EJ, Ascierto PA, Matamala L, Castillo Gutiérrez E, Rutkowski P, Gogas HJ, Lao CD, De Menezes JJ, Dalle S, Arance A, Grob JJ, Srivastava S, Abaskharoun M, Hamilton M, Keidel S, Simonsen KL, Sobiesk AM, Li B, Hodi FS, Long GV; RELATIVITY-047 Investigators. Relatlimab and Nivolumab versus Nivolumab in Untreated Advanced Melanoma. N Engl J Med. 2022;386(1):24-34. doi: 10.1056/NEJMoa2109970. Thrane K, Eriksson H, Maaskola J, Hansson J, Lundeberg J. Spatially Resolved Transcriptomics Enables Dissection of Genetic Heterogeneity in Stage III Cutaneous Malignant Melanoma. Cancer Res. 2018;78(20):5970-5979. doi: 10.1158/0008-5472.CAN-18-0747. Tirosh I , Izar B, Prakadan SM, Wadsworth MH 2nd, Treacy D, Trombetta JJ, Rotem A, Rodman C, Lian C, Murphy G, Fallahi-Sichani M, Dutton-Regester K, Lin JR, Cohen O, Shah P, Lu D, Genshaft AS, Hughes TK, Ziegler CG, Kazer SW, Gaillard A, Kolb KE, Villani AC, Johannessen CM, Andreev AY, Van Allen EM, Bertagnolli M, Sorger PK, Sullivan RJ, Flaherty KT, Frederick DT, Jané-Valbuena J, Yoon CH, Rozenblatt-Rosen O, Shalek AK, Regev A, Garraway LA.Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science 2016; 352(6282):189-96. doi: 10.1126/science.aad0501. Tsoi J , Robert L, Paraiso K, Galvan C, Sheu KM, Lay J, Wong DJL, Atefi M, Shirazi R, Wang X, Braas D, Grasso CS, Palaskas N, Ribas A, Graeber TG. Multi-stage Differentiation Defines Melanoma Subtypes with Differential Vulnerability to Drug-Induced Iron-Dependent Oxidative Stress. Cancer Cell. 2018 May 14;33(5):890-904.e5. doi: 10.1016/j.ccell.2018.03.017. Van der Walt S , Schönberger JL, Nunez-Iglesias J, Boulogne F, Warner JD, Yager N, Gouillart E, Tony Yu, the scikit-image contributors. scikit-image: image processing in Python. PeerJ 2014 (6); e453, doi.org/10.7717/peerj.453 Virshup I , Rybakov S, Theis Fabian J, Angerer P, Wolf FA. anndata: Access and store annotated data matrices. JOSS 2024;9 (101), 4371. DOI: 10.21105/joss.04371. Wang HC , Xia R, Chang WH, Hsu SW, Wu CT, Chen CH, Shih TC. Improving cancer immunotherapy in prostate cancer by modulating T cell function through targeting the galectin-1. Front Immunol. 2024;15:1372956. doi: 10.3389/fimmu.2024.1372956. Weigert M, Schmidt U. Nuclei Instance Segmentation and Classification in Histopathology Images with Stardist. 2022 IEEE International Symposium on Biomedical Imaging Challenges (ISBIC). Kolkata, India, 28.03.2022 - 31.03.2022: IEEE, S. 1–4. DOI: 10.1109/ISBIC56247.2022.9854534 Wu T , Hu E, Xu S, Chen M, Guo P, Dai Z, Feng T, Zhou L, Tang W, Zhan L, Fu X, Liu S, Bo X, Yu G.clusterProfiler 4.0: A universal enrichment tool for interpreting omics data. Innovation (Camb). 2021 Jul 1;2(3):100141. doi: 10.1016/j.xinn.2021.100141 Zappia L , Oshlack A. Clustering trees: a visualization for evaluating clusterings at multiple resolutions. Gigascience. 2018;7(7):giy083. doi:10.1093/gigascience/giy083 Zhang J , Humphreys IR, Pei J, Kim J, Choi C, Yuan R, Durham J, Liu S, Choi HJ, Baek M, Baker D, Cong Q. Predicting protein-protein interactions in the human proteome. Science. 2025;390(6771):eadt1630. doi: 10.1126/science.adt1630. Zappia L , Oshlack A. Clustering trees: a visualization for evaluating clusterings at multiple resolutions. 1357 Gigascience. 2018;7(7):giy083. doi:10.1093/gigascience/giy083 Zubieta MR , Furman D, Barrio M, Bravo AI, Domenichini E, Mordoh J. Galectin-3 expression correlates with apoptosis of tumor-associated lymphocytes in human melanoma biopsies. Am J Pathol. 2006;168(5):1666-75. doi: 10.2353/ajpath.2006.050971. Additional Declarations The authors declare potential competing interests as follows: M. Kunz received honoraria from the Speaker Bureau of Roche Pharma and travel support from Novartis Pharma GmbH and Bristol-Myers Squibb GmbH. J.C. Simon obtained speaker fees from Bristol-Myers Squibb, Roche Pharma AG, Novartis, and MSD Sharp & Dome, as well as financial support for congress attendance from Bristol-Myers Squibb, MSD Sharp & Dome, and Novartis. M. Ziemer received lecture fees from Bristol-Myers Squibb, MSD Sharp & Dohme GmbH, Pfizer Pharma GmbH, and Sanofi-Aventis Deutschland GmbH, received financial support for congress participation from Bristol-Myers Squibb, and serves as a member of expert panels on cutaneous adverse reactions for Pfizer INC. K. Reiche received honoraria from Novartis Pharma GmbH. Supplementary Files SupplementaryFigures.pdf Supplementary Figures 1 to 5 SupplementaryTable1.xlsx Supplementary Table 1 SupplementaryTable2.xlsx Supplementary Table 2 SupplementaryTable3engl.xlsx Supplementary Table 3 SupplementaryTable4.xlsx Supplementary Table 4 ExtendedData.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9075388","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":603650882,"identity":"64288f20-badd-456d-8703-7c4b0cb5c245","order_by":0,"name":"Florian Große","email":"","orcid":"","institution":"Fraunhofer Institute of Cell Therapy and Immunology , Leipzig","correspondingAuthor":false,"prefix":"","firstName":"Florian","middleName":"","lastName":"Große","suffix":""},{"id":603650883,"identity":"d959c83d-e63f-4e9f-a5d8-8dcd738b41dc","order_by":1,"name":"Christoph Kämpf","email":"","orcid":"","institution":"Fraunhofer Institute of Cell Therapy and Immunology , Leipzig","correspondingAuthor":false,"prefix":"","firstName":"Christoph","middleName":"","lastName":"Kämpf","suffix":""},{"id":603650884,"identity":"ec7f7a9c-4098-4828-92cc-7f9462a8b8eb","order_by":2,"name":"Dennis Löffler","email":"","orcid":"","institution":"Fraunhofer Institute of Cell Therapy and Immunology , Leipzig","correspondingAuthor":false,"prefix":"","firstName":"Dennis","middleName":"","lastName":"Löffler","suffix":""},{"id":603650885,"identity":"dc49fa99-3e66-44fa-914e-982f6f4eab4a","order_by":3,"name":"Max A Lingner Chango","email":"","orcid":"","institution":"Department of Dermatology, Venereology and Allergology, University of Leipzig Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Max","middleName":"A Lingner","lastName":"Chango","suffix":""},{"id":603650892,"identity":"a209da38-f74c-4359-a69b-0f28a58fd1c5","order_by":4,"name":"Conny Blumert","email":"","orcid":"","institution":"Fraunhofer Institute of Cell Therapy and Immunology , Leipzig","correspondingAuthor":false,"prefix":"","firstName":"Conny","middleName":"","lastName":"Blumert","suffix":""},{"id":603650893,"identity":"6ce8a483-b511-4bd4-b530-00db12833c03","order_by":5,"name":"Alexander Scholz","email":"","orcid":"","institution":"Institute of Clinical Immunology, University of Leipzig","correspondingAuthor":false,"prefix":"","firstName":"Alexander","middleName":"","lastName":"Scholz","suffix":""},{"id":603650898,"identity":"1af1d5a0-12d0-4288-a9a5-9bedb3f12187","order_by":6,"name":"Antonia Stubenvoll","email":"","orcid":"","institution":"Department of Dermatology, Venereology and Allergology, University of Leipzig Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Antonia","middleName":"","lastName":"Stubenvoll","suffix":""},{"id":603650900,"identity":"8146fd4e-4fd4-473d-90e2-df6d9728f365","order_by":7,"name":"Henry Loeffler-Wirth","email":"","orcid":"","institution":"Interdisciplinary Center for Bioinformatics, University of Leipzig","correspondingAuthor":false,"prefix":"","firstName":"Henry","middleName":"","lastName":"Loeffler-Wirth","suffix":""},{"id":603650902,"identity":"08481372-922a-4e00-9de1-dcf5c140b96d","order_by":8,"name":"Hans Binder","email":"","orcid":"","institution":"Interdisciplinary Center for Bioinformatics, University of Leipzig","correspondingAuthor":false,"prefix":"","firstName":"Hans","middleName":"","lastName":"Binder","suffix":""},{"id":603650903,"identity":"e7ac1921-8c9f-4772-8add-aef81002fe54","order_by":9,"name":"Carolyn Schultz","email":"","orcid":"","institution":"Department of Dermatology, Venereology and Allergology, University of Leipzig Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Carolyn","middleName":"","lastName":"Schultz","suffix":""},{"id":603650904,"identity":"fdedaecb-88b7-42d0-a3f0-74d2399a1931","order_by":10,"name":"Max Beining","email":"","orcid":"","institution":"Institute for Drug Discovery, University of Leipzig Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Max","middleName":"","lastName":"Beining","suffix":""},{"id":603650905,"identity":"3f1a0252-0134-4e36-a795-f95ff16c01ba","order_by":11,"name":"Jan C Simon","email":"","orcid":"","institution":"Department of Dermatology, Venereology and Allergology, University of Leipzig Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Jan","middleName":"C","lastName":"Simon","suffix":""},{"id":603650906,"identity":"c3a8ee4e-4591-4421-99aa-9add3f01704a","order_by":12,"name":"Clara T Schoeder","email":"","orcid":"","institution":"Institute for Drug Discovery, University of Leipzig Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Clara","middleName":"T","lastName":"Schoeder","suffix":""},{"id":603650907,"identity":"a53b84aa-a843-4a22-b8b2-3cf52fee2df4","order_by":13,"name":"Mirjana Ziemer","email":"","orcid":"","institution":"Department of Dermatology, Venereology and Allergology, University of Leipzig Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Mirjana","middleName":"","lastName":"Ziemer","suffix":""},{"id":603650908,"identity":"d3708f28-d200-4553-afd5-3751aa443050","order_by":14,"name":"Kristin Reiche","email":"","orcid":"","institution":"Fraunhofer Institute of Cell Therapy and Immunology , Leipzig","correspondingAuthor":false,"prefix":"","firstName":"Kristin","middleName":"","lastName":"Reiche","suffix":""},{"id":603650909,"identity":"d5fff11b-c266-4f1d-8bbb-7b65718ef8bf","order_by":15,"name":"Manfred Kunz","email":"data:image/png;base64,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","orcid":"","institution":"Department of Dermatology, Venereology and Allergology, University of Leipzig Medical Center","correspondingAuthor":true,"prefix":"","firstName":"Manfred","middleName":"","lastName":"Kunz","suffix":""}],"badges":[],"createdAt":"2026-03-09 16:30:52","currentVersionCode":1,"declarations":{"humanSubjects":true,"vertebrateSubjects":false,"conflictsOfInterestStatement":true,"humanSubjectEthicalGuidelines":true,"humanSubjectConsent":true,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-9075388/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9075388/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104385671,"identity":"933f672d-c7ec-4e4c-8359-aed25736db52","added_by":"auto","created_at":"2026-03-11 08:41:37","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":325773,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eOverall study design to decode spatially resolved melanoma-immune cell communication. \u003c/strong\u003eWe analyzed primary melanomas and melanoma metastases using 10x Genomics Visium and Visium HD spatial transcriptomics platforms. First column: Four primary melanoma samples were processed by Visium spatial transcriptomics. For cell-type annotation, public single-cell transcriptomic data sets were used (Stubenvoll et al., 2025). Validation was done with a publicly available 10x Genomics Visium data set of lymph node metastases (\u003cstrong\u003ePozniak et al., 2024\u003c/strong\u003e). Data analysis focused on cellular and ligand-receptor (LR) interactions at the tumor-immune border region. Second column: Validation of LR interactions was done in an additional own sample set of four primary melanomas and three melanoma metastases using 10x Genomics Visium HD technology using a single-cell transcriptomics data set for cell type annotations. Third column: Twenty primary melanomas were subsequently analyzed by multicolor immunofluorescence for proximity analysis of cells expressing LR pairs. Fourth column: The impact of an identified ligand-receptor pair on melanoma patient survival.\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/4363ccd87fd27f7aa2796e57.png"},{"id":104385674,"identity":"1b69fefb-81d6-4f34-b492-4a37a0535f30","added_by":"auto","created_at":"2026-03-11 08:41:38","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":564020,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSpatial transcriptomics of a superficial spreading melanoma of 2.8 mm Breslow thickness\u003c/strong\u003e. A melanoma sample of a superficial spreading melanoma, Breslow thickness 2.8 mm (PM1), was analyzed by spatial transcriptomics using the 10x Genomics Visium technology. \u003cstrong\u003e(A)\u003c/strong\u003e H\u0026amp;E staining of the cryosection on the Visium slide with expert annotation of histopathological areas using differently colored bordered areas. \u003cstrong\u003e(B)\u003c/strong\u003eUnsupervised clustering of spot expression via Louvain clustering (Seurat workflow, resolution 0.5). Clusters were named based on overexpressed genes and GSEA (Extended data Figure 2). \u003cstrong\u003e(C)\u003c/strong\u003e Spatial distribution of deconvolved melanoma and T/NK cell content. \u003cstrong\u003e(D)\u003c/strong\u003e Spatial distribution of cell type specific module scores for melanoma, T, and NK cells calculated using log-transformed counts of custom gene panels. \u003cstrong\u003e(E)\u003c/strong\u003e The Histograms of the module score values in Figure 2C. Distribution estimators and calculated thresholds for determining cell type positive spots are overlayed. \u003cstrong\u003e(F)\u003c/strong\u003eCombination of deconvolution results (y-axis) and cell type specific module scores (x-axis) with distribution modelling. Spots where presence of a cell type is supported by both methods are colored (upper right quadrants) and used to identify the border region as shown in Figure 2H. \u003cstrong\u003e(G)\u003c/strong\u003eImmunofluorescence staining of a serial section of the mentioned sample with anti-CD3 antibody to identify T cell infiltrates. T cell infiltrates are identified immediately adjacent to the melanoma cell area. Scalebar equals 500 µm. \u003cstrong\u003e(H)\u003c/strong\u003e Active regions defined by close proximity of tumor and immune cell positive spots (yellow, bright green and bright red) and tumor and immune cell spots not part of the inflammatory border (dark red and dark green).\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/bda5995356c9c9bc46f27721.png"},{"id":104385676,"identity":"e0e78d39-ac25-4ad7-82cb-078b3657c3ce","added_by":"auto","created_at":"2026-03-11 08:41:38","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1020815,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eInterfacing melanoma and immune spots in three further own and 3 independent melanoma samples and LR interactions at the tumor-immune border region. (A)\u003c/strong\u003e Annotated tumor and T/NK positive spots and resulting border regions for samples PM2-PM4 and POZNIAK4-POZNIAK6 (\u003cstrong\u003ePozniak et al., 2024\u003c/strong\u003e). Active regions defined by close proximities of tumor and immune cells are indicated by colored spots (yellow, bright green and bright red for direct interaction at the \u003cstrong\u003etumor-immune \u003c/strong\u003eborder region; dark red and dark green more distant to the tumor-immune\u003cstrong\u003e \u003c/strong\u003eborder region). \u003cstrong\u003e(B)\u003c/strong\u003e Summary of LR interactions consistently enriched at the tumor-immune border region in all four of our primary melanomas analyzed by use of the NICHES framework (adjusted p-value \u0026lt; 0.05). Heatmap panels from left to right: observed log2FC of interaction scores in border region vs. non-border region spots in our samples; in samples from Pozniak and co-workers (\u003cstrong\u003ePozniak et al. 2024\u003c/strong\u003e); z-scores of normalized expressions per broad cell type, as described recently (Stubenvoll et al. 2025). \u003cstrong\u003e(C)\u003c/strong\u003e Circos plots of LR interaction that are significant in at least 3 of 4 of our samples. For better visualization, human leukocyte antigen (HLA) and T cell receptor (TCR) interactions were removed. \u003cstrong\u003e(D)\u003c/strong\u003e \u003cem\u003eLGALS1\u003c/em\u003e and \u003cem\u003ePTPTRC\u003c/em\u003e expression, respectively, in PM1 and NICHES interaction scores of the \u003cem\u003eLGALS1\u003c/em\u003e–\u003cem\u003ePTPRC\u003c/em\u003einteraction. \u003cstrong\u003e(E)\u003c/strong\u003e Niches interaction scores for \u003cem\u003eLGALS1\u003c/em\u003e and \u003cem\u003ePTPRC\u003c/em\u003eof samples PM2-PM4 and POZNIAK4-POZNIAK6.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/99a47a5abcbacd368de66731.png"},{"id":104385754,"identity":"3663fd54-a6a1-4fb6-aabd-557bdf627960","added_by":"auto","created_at":"2026-03-11 08:41:59","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":5233904,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eValidation of galectin-1 and CD45 protein-protein-interaction using melanoma tissue distribution, in-silico model and in-vitro BLI assay.\u003c/strong\u003e Overall, 20 samples of primary melanomas were stained by immunofluorescence (IF) for SOX10 (yellow), galectin-1 (green) and CD45 (red), and cell distances between SOX10 and galectin-1 double-positive (SOX10 +/galectin-1+) melanoma cells and CD45 positive (CD45+) immune cells were determined by QuPath analysis. \u003cstrong\u003e(A)\u003c/strong\u003e A representative staining of an immune-hot melanoma sample (IF_PM12) is shown at 20x magnification. \u003cstrong\u003e(B)\u003c/strong\u003e Cell segmentation coordinates from the IF-stained sample in (A) were exported in QuPath to analyze the distance between galectin-1+ melanoma cells and CD45+ immune cells. \u003cstrong\u003e(C)\u003c/strong\u003e Distribution of closest distance between galectin-1+ melanoma cells and CD45+ immune cells are shown, indicating median distance at 11µm (blue dotted line) and most immune cells within a range of 20 µM (red straight line). \u003cstrong\u003e(D)\u003c/strong\u003e The left plot shows the density of CD45+ immune cells surrounding SOX10+/galectin-1+ melanoma cells within 20 µm distance, revealing a significantly higher immune cell density in immune hot melanoma samples compared to immune cold samples (left panel), however, with similar median distance (right panel). Adjusted p-values (Dunnett’s test) are displayed in scatter plots. P-values ≤ 0.05 were regarded as statistically significant. \u003cstrong\u003e(E)\u003c/strong\u003e Model of interaction of monomeric extracellular domains of galectin-1 (light brown) and CD45 (blue), generated via Boltz2 (\u003cstrong\u003ePassaro et al., 2025\u003c/strong\u003e). \u003cstrong\u003e(F)\u003c/strong\u003e Bio layer interferometry (BLI) measurements for the interaction of CD45 with galectin-1 and galectin-3, respectively. Fc-tagged CD45 and negative control antibody was loaded to a protein A coated biosensor and protein binding was observed with different concentrations of galectin-1 and galectin-3. Baseline at c = 0 µM was determined and subtracted from the displayed results. The dissociation constant (KD) at the steady state was determined after 120 sec of association and dissociation.\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/f357a4da5267e12c01d9e4da.png"},{"id":104385696,"identity":"0acd836b-f5bf-4bc5-829c-44503a009d07","added_by":"auto","created_at":"2026-03-11 08:41:43","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":2653491,"visible":true,"origin":"","legend":"\u003cp\u003eLR interactions between melanoma cells and T cells in a different set of primary and metastatic melanoma samples. LR interactions between melanoma cells and T cells in close proximity were analyzed in four additional primary melanomas and three additional cutanous melanoma metastases using the 10x Genomics Visium HD technology. (A-C) Results of LIANA ligand-receptor analyses of cells in the border region. Included interactions were significant (aggregated p-value \u0026lt; 0.01) in at least 3 samples for any combination of sender and receiver cell type. Melanoma cell subtypes were derived from Stubenvoll and co-workers (Stubenvoll et al., 2025). (A) Dot plots of interactions between neural crest (NC)-like melanocytic cells (Mel_nc-like) and different T/NK cell subtypes. (B) Dot plots of interactions between Mel_trans melanoma cells and different T cell subtypes. (C) Dot plots of interaction between Mel_trans_melan melanoma cells and different T cell subtypes. (D) Cell type interaction network for LGALS1-PTPRC interactions based on the LIANA analyses. The line thickness represents the number of samples in which this interaction was significant (aggregated p-value \u0026lt; 0.01) in the LIANA analysis for each sender-receiver cell type combination. Arrows point from cells expressing LGALS1 to cells expressing PTPRC. (E) Representative pictures of Visium HD data (sample HD_PM3) with annotated cell types and border region indication (background color). NICHES was used to compute an interaction score for the LGALS1-PTPRC interaction for all cells within 50 µm distance. The average interaction value of all incident edges per cell is indicated as blue dots (left picture). Magnified views of indicated frames (right panels) show interactions between individual cells indicated by blue lines identified by NICHES.\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/64c3de82849ec6478fd69da6.png"},{"id":104385621,"identity":"40bfef2a-af26-4245-880e-50bbaa2aee36","added_by":"auto","created_at":"2026-03-11 08:41:21","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":212854,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eImpact of \u003c/strong\u003e\u003cem\u003e\u003cstrong\u003eLGALS1\u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003e/\u003c/strong\u003e\u003cem\u003e\u003cstrong\u003ePTPRC\u003c/strong\u003e\u003c/em\u003e\u003cstrong\u003e expression on melanoma patient survival. \u003c/strong\u003eMelanoma patient survival data were collected from skin cancer cohort data from The Cancer Genome Atlas (TCGA) via The Cancer Immunome Atlas and from a further cohort of melanoma patients treated with immune checkpoint inhibition. (\u003cstrong\u003eA\u003c/strong\u003e) Kaplan-Meyer survival curves for subgroups stratified by LGALS1 and PTPRC gene expression in metastatic melanoma patient samples of the TCGA skin cancer cutaneous melanoma (SKCM) cohort. (\u003cstrong\u003eB\u003c/strong\u003e) Kaplan-Meyer survival curves for subgroups stratified by LGALS1 and PTPRC gene expression in Pembrolizumab-treated patient samples of a melanoma patient cohort treated with immune checkpoint inhibition (\u003cstrong\u003eGide et al., 2019\u003c/strong\u003e). (\u003cstrong\u003eC\u003c/strong\u003e) Kaplan-Meyer survival curves of metastatic melanoma patients of the TCGA cohort, first separated by estimated CD8\u003csup\u003e+\u003c/sup\u003e T cell content (to account for the correlation between PTPRC expression and infiltration) and then split into LGALS1 and PTPRC high/low subgroups like in (A).\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/98f627596202c6205e6d3d1d.png"},{"id":104385883,"identity":"740ff92b-1e7c-49ad-9337-1a3833fae967","added_by":"auto","created_at":"2026-03-11 08:42:24","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":13779218,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/b08545dd-3001-4d1f-9776-e887505dc5a6.pdf"},{"id":104385730,"identity":"6fcc30be-549d-42fb-b173-f59ac5f6b4b4","added_by":"auto","created_at":"2026-03-11 08:41:48","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":21434143,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Figures 1 to 5\u003c/p\u003e","description":"","filename":"SupplementaryFigures.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/7ee954b9ad6777cde906b2c8.pdf"},{"id":104385667,"identity":"d6862d25-2cd2-422c-ad5b-064a96a76719","added_by":"auto","created_at":"2026-03-11 08:41:35","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":33105,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Table 1\u003c/p\u003e","description":"","filename":"SupplementaryTable1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/f3e34159980fb0bf86a0e008.xlsx"},{"id":104385697,"identity":"83e4c7d3-1db9-4d25-a344-423c42fa4ec5","added_by":"auto","created_at":"2026-03-11 08:41:44","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":19975,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Table 2\u003c/p\u003e","description":"","filename":"SupplementaryTable2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/c998e97dba375a0b26816ff0.xlsx"},{"id":104385681,"identity":"9e56e25f-f6fd-4c8d-9fb3-50e600136717","added_by":"auto","created_at":"2026-03-11 08:41:40","extension":"xlsx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":1267565,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Table 3\u003c/p\u003e","description":"","filename":"SupplementaryTable3engl.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/0af72c05131e529ffd9dfb18.xlsx"},{"id":104385673,"identity":"1eff86b2-2ca8-4ef0-94c5-f57218706256","added_by":"auto","created_at":"2026-03-11 08:41:38","extension":"xlsx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":22228,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary Table 4\u003c/p\u003e","description":"","filename":"SupplementaryTable4.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/49f2b12808f7a0d0974ea50a.xlsx"},{"id":104385662,"identity":"200e0a86-3d34-4ed2-a3b9-54ed2daf3f3a","added_by":"auto","created_at":"2026-03-11 08:41:34","extension":"docx","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":7626358,"visible":true,"origin":"","legend":"","description":"","filename":"ExtendedData.docx","url":"https://assets-eu.researchsquare.com/files/rs-9075388/v1/a6cc408a37f6d7f49870ef39.docx"}],"financialInterests":"The authors declare potential competing interests as follows: M. Kunz received honoraria from the Speaker Bureau of Roche Pharma and travel support from Novartis Pharma GmbH and Bristol-Myers Squibb GmbH. J.C. Simon obtained speaker fees from Bristol-Myers Squibb, Roche Pharma AG, Novartis, and MSD Sharp \u0026 Dome, as well as financial support for congress attendance from Bristol-Myers Squibb, MSD Sharp \u0026 Dome, and Novartis. M. Ziemer received lecture fees from Bristol-Myers Squibb, MSD Sharp \u0026 Dohme GmbH, Pfizer Pharma GmbH, and Sanofi-Aventis Deutschland GmbH, received financial support for congress participation from Bristol-Myers Squibb, and serves as a member of expert panels on cutaneous adverse reactions for Pfizer INC. K. Reiche received honoraria from Novartis Pharma GmbH. ","formattedTitle":"\u003cp\u003eDecoding melanoma-immune cell communication through spatially resolved ligand-receptor interaction analyses\u003c/p\u003e","fulltext":[{"header":"Main","content":"\u003cp\u003eMelanoma is a highly immunogenic tumor, and the interaction between immune cells and tumor cells is of central importance for tumor development and control. Significant treatment responses using immune checkpoint inhibitors have been achieved in recent years with antibodies targeting cytotoxic T-lymphocyte antigen 4 (CTLA-4) and programmed cell death 1 (PD-1) in immune cells, which reactivate the compromised immune system (\u003cstrong\u003eSchadendorf et al., 2018\u003c/strong\u003e). At present, checkpoint inhibitor treatment has gained increasing relevance because of its mutation-independence and long-‍term effects. However, most patients (60%) do not respond to these treatments, and recurrence rates are high (\u003cstrong\u003eLim et al., 2023\u003c/strong\u003e). New clinical trials have shown promising results with antibodies directed against further immune checkpoint targets, such as lymphocyte activation gene‑3 (LAG‑3) and T-cell immunoglobulin and mucin-domain containing-3 (TIM‑3) on immune cells, which led to the approval of anti-LAG3 treatment in combination with anti-PD1 treatment. However, the treatment effects are only slightly better than those of anti-PD1 treatment alone (\u003cstrong\u003eTawbi et al., 2022\u003c/strong\u003e). Recent studies have shown that immune checkpoint inhibitors may also provide beneficial effects as neoadjuvant treatments (\u003cstrong\u003eSaad et al., 2023\u003c/strong\u003e). The search for new immune targets and treatment modalities for future treatment is ongoing. In this context, a better understanding of the melanoma-‍immune cell interactions at the immunological interface is needed.\u003c/p\u003e\n\u003cp\u003eIn recent years, the spatiotemporal organization of cells within a complex tissue has been analyzed using single-cell and spatial transcriptomics technology (\u003cstrong\u003eLim et al., 2020\u003c/strong\u003e). Several earlier studies have used single-cell transcriptome analysis to analyze melanoma and immunological heterogeneity (\u003cstrong\u003eTirosh\u0026nbsp;et\u0026nbsp;al., 2016; Rambow et al., 2018; Sade-Feldmann et al., 2019; Li et al., 2020\u003c/strong\u003e). In particular, the analysis of melanoma treatment resistance and response in humans has been the focus of these studies (\u003cstrong\u003eJerby-Arnon et al., 2018; Sade-Feldman et al., 2019; Li et al., 2020\u003c/strong\u003e). One of the first studies utilizing single-cell RNA sequencing (scRNA-seq) analyses identified a T\u0026nbsp;cell signature that allowed the classification of bulk-sequenced tumors as tumors harboring a (T\u0026nbsp;cell)-exclusion program (\u003cstrong\u003eJerby-\u003c/strong\u003e\u003cstrong\u003e‍Arnon, et al., 2018\u003c/strong\u003e). This exclusion program was then mapped onto ipilimumab and anti-\u003cstrong\u003e‍\u003c/strong\u003ePD1–\u003cstrong\u003e‍\u003c/strong\u003etreated samples analyzed by scRNA-seq to identify co-expressed genes in individual cells.\u003c/p\u003e\n\u003cp\u003eIn a further study, single-cell transcriptomes of metastatic melanoma patients were generated from biopsies taken at baseline and under anti-PD1 inhibitor treatment, either alone or in combination with anti-CTLA4 treatment (\u003cstrong\u003eSade-Feldman et al., 2019\u003c/strong\u003e). Treatment-resistant clusters were enriched for genes linked to T\u0026nbsp;cell exhaustion (\u003cem\u003eLAG3\u003c/em\u003e, \u003cem\u003ePDCD1\u003c/em\u003e, \u003cem\u003eHAVCR2\u003c/em\u003e, \u003cem\u003eTIGIT\u003c/em\u003e, \u003cem\u003eCD38)\u003c/em\u003e. A more recent study focused on dysfunctional T\u0026nbsp;cells in melanoma lesions undergoing immune checkpoint therapy (\u003cstrong\u003eLi et al., 2020\u003c/strong\u003e). This study included patients with prior treatment against CTLA‑4 or PD-1 or a combination of both and revealed that CD8\u003csup\u003e+\u003c/sup\u003e T cells partly transitioned into a dysfunctional T cell pool. These dysfunctional T \u003cstrong\u003e‍\u003c/strong\u003ecells are characterized by the expression of \u003cem\u003ePDCD1\u003c/em\u003e, \u003cem\u003eLAG3\u003c/em\u003e and molecules shared with CD4\u003csup\u003e+\u003c/sup\u003e Tregs (e.g., \u003cstrong\u003e‍\u003c/strong\u003e\u003cem\u003eCSF1\u003c/em\u003e and \u003cem\u003eZBED2)\u003c/em\u003e. However, the dysfunctional T cells had the highest levels of clonal expansion and were still active in \u003cem\u003eex vivo\u003c/em\u003e experiments.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOverall, single-cell analyses of melanoma and other tumors have enabled a deeper understanding of tumor heterogeneity and the transcriptional states of immune cells in the tumor microenvironment. Several studies have shown that basic mechanisms of cell–cell communication can be inferred from single-cell analyses using knowledge of known ligand–receptor interactions in the spatial context (\u003cstrong\u003eLarsson et al., 2021\u003c/strong\u003e). We recently analyzed transcriptional programs and ligand–receptors interaction in different cell types in high (hot), intermediate and low (cold) immune melanoma tumors (\u003cstrong\u003eStubenvoll\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e‍et‍ al., 2025\u003c/strong\u003e). Among prominent ligand-receptor interactions in hot tumors were CD58–\u003cstrong\u003e‍\u003c/strong\u003eCD2 and CD59–CD2, of which CD58–CD2 showed functional relevance.\u003c/p\u003e\n\u003cp\u003eHowever, in classical scRNA-seq datasets, the physical relationships between cells remain unknown, which makes it difficult to place cell–cell interactions in an appropriate tissue context. To address this issue and extend our recent studies, we performed spatial transcriptomics (ST) analyses in primary melanomas and cutaneous melanoma metastases (\u003cstrong\u003eFigure 1\u003c/strong\u003e), complemented by a recent data set of melanoma lymph node metastases (\u003cstrong\u003ePozniak et al., 2024\u003c/strong\u003e). In fact, a few recent studies have already used spatial transcriptomics for melanoma tissue, but did not specifically address the local melanoma –immune\u0026nbsp;cell – interactions at the immunological interface (\u003cstrong\u003eThrane et al., 2018; Hunter\u0026nbsp;et\u0026nbsp;al., 2021; Karras et al., 2022; Pozniak et al., 2024\u003c/strong\u003e). We particularly focused on melanoma-‍T/NK-cell interactions as T/NK cells are clinically relevant in immunotherapies for melanoma.\u003c/p\u003e\n\u003cp\u003eWe identified several new candidates for melanoma–immune cell interactions, such as LGALS1–PTPRC, LGALS1–CD69, and LGALS3–CD6 using two different 10x\u0026nbsp;Genomics® spatial transcriptomics technologies (Visium and Visium HD) and ligand–receptor interaction analysis (\u003cstrong\u003eDimitrov et al., 2022\u003c/strong\u003e). These interaction partners suggest new targets for future immunotherapeutic approaches in the therapeutic or adjuvant setting in melanoma.\u003c/p\u003e"},{"header":"Results","content":"\u003ch3\u003eSpatial transcriptomics at mini-bulk scale captures the immunological interface of melanoma\u0026ndash;immune-cells in primary melanomas and lymph node metastases\u003c/h3\u003e\n\u003cp\u003eTo uncover melanoma\u0026ndash;immune cell interactions, we first performed spatial transcriptomics (ST) of four primary melanoma lesions representing different subtypes (two superficial spreading, one nodular and one acrolentiginous melanoma) with varying immune cell infiltration levels (two immune hot and two immune intermediate melanoma lesions, \u003cstrong\u003eFigure 1\u003c/strong\u003e). Samples were analyzed utilizing the 10x Genomics Visium platform enabling whole transcriptome assessment at mini-bulk scale of the immunological interface at the tumor\u0026ndash;immune border region (\u003cstrong\u003eSupplementary Table\u0026nbsp;1, and Extended data Figure 1\u003c/strong\u003e). To guide spatial transcriptomics data analysis, an experienced histopathologist (MZ) determined cell annotations and histopathological areas of all tissue samples. We further verified immune cell infiltrates by immunofluorescence staining of serial sections with anti-CD3 antibody (CD3 \u0026zwj;is a part of the T cell receptor complex). The samples were processed according to the manufacturer\u0026rsquo;s (10x Genomics) specifications for the Visium platform. We used 10x Genomics Space Ranger for initial data analysis and QC (for per-sample statistics see \u003cstrong\u003eSupplementary Table 1C-D\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 2A\u003c/strong\u003e shows H\u0026amp;E staining of one of the two superficial spreading melanoma (SSM) with a tumor (Breslow) thickness of 2.8 mm (PM1). First, we identified 7 co-expression clusters by unsupervised Louvain clustering (\u003cstrong\u003eBlondel et al., 2008\u003c/strong\u003e) on log-transformed size-normalized gene expression (\u003cstrong\u003eFigure \u0026zwj;2B\u003c/strong\u003e), according to the standard Seurat workflow (\u003cstrong\u003eHao et al., 2021\u003c/strong\u003e). To obtain biologically relevant co-expression clusters, we chose the resolution parameter such that the clustering granularity was comparable to the histopathological annotation (see, Methods). Second, we annotated co-\u0026zwj;expression clusters according to enriched genes identified via Wilcoxon rank sum test of each cluster versus all remaining clusters (adjusted\u0026nbsp;p\u0026nbsp;\u0026lt;\u0026nbsp;0.01, log2FC\u0026nbsp;\u0026gt;\u0026nbsp;0.15). Among the top ten enriched marker genes (\u003cstrong\u003eMethods\u003c/strong\u003e, p \u0026lt; 0.01), we found classical cluster-defining genes consistent with the histopathological annotation of sample PM1: \u003cem\u003eIGHA1\u003c/em\u003e, \u003cem\u003eIGHG1\u003c/em\u003e and \u003cem\u003eIGKC\u0026nbsp;\u003c/em\u003efor expression of complement factors and immunoglobulins referring to immune cells (cluster 1), \u003cem\u003eSOX10, PMEL and MLANA\u003c/em\u003e for melanoma cells (cluster 2), \u003cem\u003eS100A7A, KRT1\u003c/em\u003e, \u003cem\u003eKRT10\u003c/em\u003e, and \u003cem\u003eKRT15\u003c/em\u003e for keratinocytes (clusters 3, 4 and 5) (\u003cstrong\u003eExtended \u0026zwj;\u003c/strong\u003e\u003cstrong\u003edata\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eFigure \u0026zwj;2A\u003c/strong\u003e). Additionally, we conducted gene set enrichment analysis (GSEA) (\u003cstrong\u003eSubramanian \u0026zwj;et al., 2005\u003c/strong\u003e) of marker genes based on the Gene Ontology (GO) (\u003cstrong\u003eAshburner et al., 2000\u003c/strong\u003e) biological processes (GO:BP, FDR\u0026nbsp;\u0026lt;\u0026nbsp;0.01) database (\u003cstrong\u003eMethods and Extended data Figure 2\u003c/strong\u003e). GSEA \u0026zwj;confirmed cluster annotations by revealing significant enrichment (FDR\u0026nbsp;\u0026lt;\u0026nbsp;0.01) of genes involved in leukocyte cell-\u0026zwj;cell adhesion and T cell activation in cluster 1, genes related to pigmentation in cluster 2, and genes associated with keratinocyte proliferation in clusters 3 and 4 and keratinocyte differentiation in clusters 4 and 5, respectively (\u003cstrong\u003eExtended data Figure 2B\u003c/strong\u003e). Overall, our unsupervised clustering strategy of spatial gene expression patterns reliably identified distinct histopathological tissue regions. However, \u0026zwj;non-specific cell type annotation partly persists, since the 10x Visium platform lacks single-\u0026zwj;cell resolution. For instance, clusters 1, 3 and 4 exhibit low-level expression of classical markers for melanoma cells like \u003cem\u003eSOX10\u003c/em\u003e, \u003cem\u003ePMEL\u003c/em\u003e and \u003cem\u003eMLANA\u003c/em\u003e, albeit not at statistically significant levels(\u003cstrong\u003eExtended data \u0026zwj;Figure 2B\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFor a refined view of melanoma and T/NK cell localization in 10x Visium spots, we combined two distinct approaches of integrating single-cell gene expression references with ST data (see, Methods). First, we applied spot-wise cell type deconvolution using our recently published single-cell gene expression dataset (\u003cstrong\u003eStubenvoll et al., 2025,\u0026nbsp;\u003c/strong\u003efor details see, Methods). It contains distinct subtypes of melanoma cells derived from a study analyzing gene expression patterns in patient-derived melanoma cell cultures (\u003cstrong\u003eTsoi et al., 2018\u003c/strong\u003e). The deconvolution results confirm the histological and unsupervised co-\u0026zwj;expression clustering annotation of sample PM1. Spots belonging to co-expression cluster 2, which correspond to the histologically annotated tumor region, are primarily composed of melanoma cells. Spots enriched for T/NK cells are in the region histologically annotated as inflammatory, which corresponds to cluster 1 (\u003cstrong\u003eFigures 2A-C)\u003c/strong\u003e. Notably, the percentage of T/NK cells is not uniform in the inflammatory region but contains multiple nests of tightly packed inflammatory cells along with a moderately high T/NK cell content along the remaining tumor interface. Second, we integrated the single-cell gene expression reference with two existing single-cell gene expression melanoma datasets (\u003cstrong\u003eJerby-Arnon et al., 2018; Sol\u0026eacute;-Boldo et al., 2020\u003c/strong\u003e) to derive panels of cell type-specific and sample-\u0026zwj;independent marker genes. Using these gene panels, we calculated module scores utilizing Seurat that represent the cell-type specific expression over background (\u003cstrong\u003eFigure 2D\u003c/strong\u003e). All resulting marker gene panels are listed in \u003cstrong\u003eSupplementary Table 2\u003c/strong\u003e. For example, the marker gene panel for melanoma cells contains among others \u003cem\u003ePMEL\u003c/em\u003e, \u003cem\u003eTYR\u003c/em\u003e, \u003cem\u003eMLANA\u003c/em\u003e, and \u003cem\u003eMITF\u003c/em\u003e supporting biological relevance of retrieved marker gene panels. Both approaches, cell type deconvolution and module scores, localized tumor cells in similar regions. Deconvolution tended to overestimate melanoma cell prevalence in regions with low total expression, as indicated by low tumor cell module scores in low-\u0026zwj;expression regions (\u003cstrong\u003eFigures 2D-E\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo identify the tumor-immune border region, we therefore used a combined approach to identify the tumor core region and regions with highly active immune cells. Spots are labelled as melanoma positive, if deconvolution shows more than 10% of melanoma content and the melanoma module score exceeds the threshold defined by a gaussian mixture model. Spots are labelled as T/NK positive, if deconvolution shows more than 10% of T/NK cell\u0026nbsp;content and the respective module score exceeds the threshold defined by a robust background estimation (\u003cstrong\u003eFigure 2F\u003c/strong\u003e, \u003cstrong\u003eMethods\u003c/strong\u003e). With this combined information, we identified spots with high tumor and/or immune cell activity and then marked the spots positive for both or with direct adjacency of both as the region with active immune infiltration or border region for short. \u003cstrong\u003eFigure 2H\u003c/strong\u003e shows the result of our approach, separating the active region (yellow, bright green and bright red) from tumor and immune cell positive spots not part of the inflammatory border (dark red and dark green). We verified immune cell infiltrates in this region by immunofluorescence staining of a serial paraffin section, showing areas of CD3-positivity adjacent to the tumor area (\u003cstrong\u003eFigure 2G\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe followed the same data analysis workflow for the three remaining melanoma samples (PM2-PM4) processed by 10x Visium ST \u003cstrong\u003e(Extended data Figures 3-5).\u0026nbsp;\u003c/strong\u003eThese samples consisted of the second SSM of 0.6 \u0026zwj;mm tumor thickness (PM2), an acrolentiginous melanoma (ALM) of 4.3 mm tumor thickness (PM3) and an SSM of 1.0 mm tumor thickness (PM4).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePM2 showed a histopathologically detectable inflammatory infiltrate in the upper dermis, indicated by a yellow encircled area (\u003cstrong\u003eExtended data Figure 3A\u003c/strong\u003e). This area was represented by co-expression cluster 9 (inflammatory cells) and was also verified by immunofluorescence staining for CD3 expression (\u003cstrong\u003eExtended data Figure 3G\u003c/strong\u003e). Epidermal keratinocytes were present in co-expression clusters 1, 5 and 7, showing expression of \u003cem\u003eKRT1\u003c/em\u003e, \u003cem\u003eKRT10\u003c/em\u003e, \u003cem\u003eLOR\u003c/em\u003e and \u003cem\u003eKRT15\u003c/em\u003e; which were collocated with melanoma cells as indicated by the classical melanoma cells markers \u003cem\u003eMLANA\u003c/em\u003e and \u003cem\u003eTYRP1\u003c/em\u003e (\u003cstrong\u003eExtended Figure 3C\u003c/strong\u003e). These findings were supported by GSEA regarding the expression of genes involved in leukocyte-mediated immunity in cluster 9 and pigmentation in clusters 5 and 7 (\u003cstrong\u003eExtended Figure 3D\u003c/strong\u003e). Immune infiltration was also detected in a serial H\u0026amp;E section (\u003cstrong\u003eExtended data Figure 3G\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn PM3, an acrolentiginous melanoma of 4.3 mm Breslow thickness, the border region was indicated by cluster 5, showing an immune cluster immediately under the upper major melanoma mass (cluster \u0026zwj;7) and immediately adjacent to the lower melanoma part (cluster 4) (\u003cstrong\u003eExtended Figure 4A and 4B\u003c/strong\u003e). The \u0026zwj;melanoma clusters showed varying expression levels of melanoma cell markers such as \u003cem\u003ePMEL\u003c/em\u003e, \u003cem\u003eMLANA\u0026nbsp;\u003c/em\u003eand \u003cem\u003eTYR\u0026nbsp;\u003c/em\u003e(\u003cstrong\u003eExtended data Figure 4C\u003c/strong\u003e), and GSEA showed enrichment of genes involved in melanin metabolic processes; whereas GSEA of the immunological cluster showed enrichment of genes involved in positive regulation of T\u0026nbsp;cell activation (\u003cstrong\u003eExtended data Figure 4D\u003c/strong\u003e). \u003cstrong\u003eExtended data Figure 4E\u003c/strong\u003e shows H\u0026amp;E staining of a serial section and \u003cstrong\u003eExtended data Figure\u003c/strong\u003e \u003cstrong\u003e4F\u003c/strong\u003e shows an anti-CD3 immunofluorescence staining of the inflammatory infiltrate. The area of T cell infiltration was also verified by the T cell module score (\u003cstrong\u003eExtended data Figure 4G\u003c/strong\u003e). Although acrolentiginous melanomas are poorly inflammatory in general, we identified a clear inflammatory infiltrate in this case.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn PM4, a superficial spreading melanoma of 1.0 mm tumor thickness (\u003cstrong\u003eExtended data Figure 5\u003c/strong\u003e) inflammatory cells were present in cluster 0, and gene expression was detected for known immune regulators like \u003cem\u003eCCL19\u003c/em\u003e and \u003cem\u003eCCL21\u003c/em\u003e, as well as a complement factor \u003cem\u003eC3\u003c/em\u003e in this cluster (\u003cstrong\u003eExtended data Figure 5B and 5C\u003c/strong\u003e). Cluster 1, 4 and 5 harbored melanoma cells and expressed \u003cem\u003ePRAME\u003c/em\u003e, \u003cem\u003eSOX10\u0026nbsp;\u003c/em\u003eand\u003cem\u003e\u0026nbsp;MLANA\u003c/em\u003e, as classical melanoma cell markers (\u003cstrong\u003eExtended Figure 5C\u003c/strong\u003e), supported by GSEA of genes involved in cellular pigmentation (clusters \u0026zwj;1 \u0026zwj;and 5) and positive regulation of T cell activation (cluster 0), respectively (\u003cstrong\u003eExtended data Figure 5D\u003c/strong\u003e). All \u0026zwj;melanoma cell areas and areas of T and NK cell infiltration were further confirmed by deconvolution and module score analyses as described above (\u003cstrong\u003eExtended data Figure 5G\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eWe followed the same approach as for sample PM1 to identify tumor-immune border regions in primary melanoma samples PM2-4 \u003cstrong\u003e(Figure\u0026nbsp;3A).\u003c/strong\u003e To also cover tumor-immune cell communication in metastasis, we extended our analyses to a published independent dataset of six melanoma lymph-\u0026zwj;node metastases analyzed with 10x Visium ST (\u003cstrong\u003ePozniak\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2024\u003c/strong\u003e). Using the same workflow as for PM1-4, we analyzed the samples from Pozniak and co-workers (Pozniak et al., 2024) and found varying degrees of infiltration and tumor activity. These samples mostly contained large dense tumor masses covering most of the area with infiltration ranging from almost-absence of detectable immune cells to large inflammatory regions bordering the tumor masses (\u003cstrong\u003eSupplementary Figure 1\u003c/strong\u003e). Three samples (POZNIAK1-3) showed distinctly weaker T/NK cell scores and deconvolved T/NK content and were excluded from further analysis due to the absence of a defined tumor-immune border region. The other three samples (POZNIAK4-6) showed stronger T/NK activity and a defined border region and were retained for ligand-\u0026zwj;receptor analysis (\u003cstrong\u003eFigure 3A\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOverall, we defined a spatially resolved map of the tumor-immune border region in four primary melanomas of different subtypes and three independent lymph node metastases enabling unbiased, in-depth and putatively clinically relevant analysis of ligand\u0026ndash;receptor interactions at the tumor-\u0026zwj;immune border region.\u003c/p\u003e\n\u003ch3\u003eSpatially resolved immunological interfaces between melanoma and T/NK cells revealed physiologically relevant ligand-receptor interactions\u0026nbsp;\u003c/h3\u003e\n\u003cp\u003eWe resolved ligand-receptor interactions at the melanoma-T/NK cell borders for all seven samples (PM1-4 and POZNIAK4-6) by testing overrepresentation of known ligand-receptor interactions. We utilized the NICHES R package (\u003cstrong\u003eRaredon et al., 2023\u003c/strong\u003e) to compute interaction scores (\u003cem\u003eproduct-\u0026zwj;of-\u0026zwj;expressions\u003c/em\u003e interaction model) while incorporating the spatial proximity of spots. Combining the ligand\u0026ndash;receptor interaction databases from LIANA (\u003cstrong\u003eDimitrov\u0026nbsp;et\u0026nbsp;al., 2022\u003c/strong\u003e) and NicheNet (\u003cstrong\u003eBrowaeys et al, 2020\u003c/strong\u003e) yielded a total of 14,977 unique mechanisms (after accounting for changed gene symbols and deduplication), for which we calculated NICHES scores and overrepresentation. For \u0026zwj;an unbiased detection of ligand-receptor interactions at the tumor-immune border region, we evaluated all known interactions regardless of previously identified cell types. Finally, the result matrices were combined across samples and annotated with additional information, such as the expression of ligand and receptor molecules in different cell populations in the single cell reference dataset (see Methods).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 3B\u003c/strong\u003e depicts all interacting partners significantly enriched in the border regions (adjusted p-\u0026zwj;value \u0026lt; 0.05) in all four primary melanomas (PM1-4) and their according enrichment values in three melanoma lymph node metastases (POZNIAK4-6). An expanded list of significantly enriched interactions in at least 3 out of 4 primary melanomas is given in \u003cstrong\u003eExtended data Figure 6.\u0026nbsp;\u003c/strong\u003eWe \u0026zwj;identified 153 unique mechanisms in at least 3 primary melanomas, also visualized in a circus plot (\u003cstrong\u003eFigure\u0026nbsp;3C\u003c/strong\u003e). A summary of all ligand\u0026ndash;receptor interactions between T/NK and melanoma cell types that were significantly enriched according to the LIANA analysis of the border region (aggregated p-value \u0026lt; 0.01) and supported in at least 2 Visium HD samples for one source-target cell type pair is available in \u003cstrong\u003eSupplementary Table 3A\u003c/strong\u003e, and a summary of all interactions differentially expressed (adjusted p-value \u0026lt; 0.05) in the border region at least one melanoma sample (Visium platform) is available in \u003cstrong\u003eSupplementary Table 3B\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eA substantial fraction of detected interactions and/or interaction partners have been described in earlier studies on immune-regulated tumor control, such as CD58 and CD59 (\u003cstrong\u003eFrangieh et al., 2021\u003c/strong\u003e), CD58 \u0026ndash;CD2 (\u003cstrong\u003eStubenvoll et al., 2025\u003c/strong\u003e), CCL19 and CXCL9 (\u003cstrong\u003eJacquelot et al., 2018; Gowhari et al., 2022; Ding et al., 2016\u003c/strong\u003e), as well as HMGB1\u0026ndash;CXCR4 (\u003cstrong\u003eLi Pomi F et al., 2022\u003c/strong\u003e) and MIF \u0026ndash;CD74 (\u003cstrong\u003ede Azevedo et al., 2020; Figueiredo et al., 2018\u003c/strong\u003e), underpinning the physiological relevance of identified interactions.\u003c/p\u003e\n\u003cp\u003eProminent expression was observed for \u003cem\u003eLGALS1\u003c/em\u003e (galectin-1) on melanoma cells and \u003cem\u003ePTPRC\u003c/em\u003e (CD45) on immune cells (\u003cstrong\u003eFigure 3B\u003c/strong\u003e). \u003cem\u003ePTPRC\u003c/em\u003e showed the strongest expression on immune cells, especially on T \u0026zwj;cells, similar to CD3 and CD2. Galectin-1 (gene product of \u003cem\u003eLGALS1\u003c/em\u003e) is a known immunosuppressive molecule and an interaction partner of CD45 (gene product of \u003cem\u003ePTPRC\u003c/em\u003e) (Nov\u0026aacute;k \u003cstrong\u003eet al., 2025\u003c/strong\u003e). Both are expressed in proximity at the border regions between melanoma and T cells. NICHES interaction scores align well with the previously defined border regions in all seven samples (\u003cstrong\u003eFigure 3D and 3E, compare 3A and \u0026zwj;2H;\u0026nbsp;\u003c/strong\u003efull panels as in \u003cstrong\u003eFigure 3D\u003c/strong\u003e for all primary samples for LAGALS1\u0026ndash;PTPRC and three additional LR pairs are available as \u003cstrong\u003eSupplementary Figure 2\u003c/strong\u003e. However, evidence for the role of the galectin-\u0026zwj;1\u0026ndash;\u0026zwj;CD45 interaction in melanoma immunology is scarce. This possible interaction in vivo was found in our data set and by reanalysis of the mentioned independent study (\u003cstrong\u003ePozniak\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2024\u003c/strong\u003e)\u003cstrong\u003e.\u003c/strong\u003e\u003c/p\u003e\n\u003ch3\u003eMulticolor immunofluorescence validates\u0026nbsp;LGALS1\u0026nbsp;(galectin-1)\u0026ndash;PTPRC\u0026nbsp;(CD45) expression and membrane-localization\u003c/h3\u003e\n\u003cp\u003eWe performed multicolor immunofluorescence (IF) staining to test whether \u003cem\u003eLGALS1\u003c/em\u003e (galectin-1) expressing melanoma cells were in proximity to \u003cem\u003ePTPRC\u003c/em\u003e (CD45) expressing T cells. IF staining assessed expression of SOX10, CD3, galectin-1 and CD45 in 20 additional melanoma samples. The percentage of CD3 \u0026zwj;positive (CD3\u003csup\u003e+\u003c/sup\u003e) cells classified the immune status of samples as cold (\u0026lt;\u0026nbsp;10\u0026nbsp;%), intermediate (10-\u0026zwj;20\u0026nbsp;%) or hot (\u0026gt;\u0026nbsp;20\u0026nbsp;%). Subsequently, we measured distances between CD45 positive (CD45\u003csup\u003e+\u003c/sup\u003e) and galectin-1 and SOX10 (galectin-1\u003csup\u003e+\u003c/sup\u003e) double positive tumor cells for at least three representative areas per sample. \u003cstrong\u003eFigure\u0026nbsp;4A\u003c/strong\u003e and \u003cstrong\u003eSupplementary Figures 3A-C\u003c/strong\u003e show representative examples with CD45\u003csup\u003e+\u003c/sup\u003e immune cells in proximity to galectin-1\u003csup\u003e+\u0026nbsp;\u003c/sup\u003emelanoma cells. A digitalized picture is shown in \u003cstrong\u003eFigure 4B and Supplementary Figures 3A-C\u003c/strong\u003e. Cell segmentation coordinates of stained sections were analyzed by Qupath and exported to analyze the distance between galectin-1\u003csup\u003e+\u003c/sup\u003emelanoma cells and CD45\u003csup\u003e+\u003c/sup\u003e immune cells. The majority of CD45\u003csup\u003e+\u003c/sup\u003e immune cells were within a range from 0 to 20 \u0026zwj;\u0026micro;M (\u003cstrong\u003eFigure 4C\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFurther quantification of all samples showed that the density of CD45\u003csup\u003e+\u003c/sup\u003e immune cells surrounding galectin-1\u003csup\u003e+\u003c/sup\u003e melanoma cells within 20 \u0026micro;m was significantly higher in immune hot melanoma samples compared to immune cold samples (\u003cstrong\u003eFigure 4D\u003c/strong\u003e). However, the median distance between galectin-1\u003csup\u003e+\u0026nbsp;\u003c/sup\u003eand CD45\u003csup\u003e+\u0026nbsp;\u003c/sup\u003ewas not different between samples of different immune cell infiltration. Together, the majority of CD45\u003csup\u003e+\u003c/sup\u003e immune cells are located in proximity to galectin-1\u003csup\u003e+\u0026nbsp;\u003c/sup\u003etumor cells with high prevalence in immune hot samples, which may support immune interactions.\u003c/p\u003e\n\u003cp\u003eTo further characterize the galectin-1\u0026ndash;CD45 interaction on a protein level, we performed \u003cem\u003ein-vitro\u003c/em\u003e interaction assays. For this purpose, soluble extracellular domains of galectin-1 and CD45 were expressed and purified from \u003cem\u003eE. coli\u0026nbsp;\u003c/em\u003eand HEK 293 cells respectively. Complementary \u003cem\u003ein-silico\u0026nbsp;\u003c/em\u003estructure prediction supported a plausible extracellular interaction interface between galectin-1 and CD45. Using Boltz-2-based structure prediction (\u003cstrong\u003ePassaro et al., 2025\u003c/strong\u003e), we observed that both proteins are placed in matching positions in the most confident models (confidence score\u0026nbsp;\u0026ge;\u0026nbsp;0.91). These predictions suggest an interaction in the fibronectin type-III domain I (residues 391-483) of CD45 and the potential C-Terminal dimerization site of galectin-1 opposite from the carbohydrate binding site (\u003cstrong\u003eFigure 4E\u003c/strong\u003e and \u003cstrong\u003eSupplementary Figure 4 and Supplementary Table 4\u003c/strong\u003e). The binding affinity of CD45 to galectin-1, and galectin-3, respectively, which was included due to high sequence and presumably functional similarity, and negative control anti-SARS-CoV2 T1 antibody was determined via bio-layer interferometry (BLI) (\u003cstrong\u003eFigures\u0026nbsp;4E-F\u003c/strong\u003e). Association and dissociation were relatively rapid in case of CD45\u0026ndash;\u0026zwj;galectin-1 and CD45\u0026ndash;\u0026zwj;galectin-3 binding, arguing for low but specific binding. Taken together, galectin-1 shows rapid and specific binding to CD45, which may influence immune cell function during immune control of melanoma cells.\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003eHigh-definition spatial transcriptomics validates LGALS1 and PTPRC co-expression at the melanoma\u0026ndash;\u0026zwj;immune cell border region\u0026nbsp;\u003c/h3\u003e\n\u003cp\u003eIn a next step, a set of seven additional melanoma samples, four primary melanoma (HD_PM1-\u0026zwj;HD_PM4) and three cutaneous melanoma metastases (HD_MM1-HD_MM3), were subjected to ST at subcellular resolution utilizing the Visium HD technology. It allows for higher spatial resolution and thereby direct analysis of putative intratumor cell-cell interactions and may thereby complement the above analyses using classical Visium technology (\u003cstrong\u003eOlivera et al., 2025\u003c/strong\u003e). The four primary melanomas had also been analyzed in our recently published melanoma single-cell study (\u003cstrong\u003eStubenvoll et al., 2025\u003c/strong\u003e). In short, samples were processed according to the manufacturer\u0026rsquo;s (10x \u0026zwj;Genomics) specifications. We used SpaceRanger for initial data analysis and QC (sample statistics in \u003cstrong\u003eSupplementary Table 1\u003c/strong\u003e). Afterwards, we continued with a modified bin2cell \u003cstrong\u003e(Polanski et al., 2024\u003c/strong\u003e) workflow to continue our analysis with quasi-single-cell resolution (see, Methods). We used a custom segmentation of the H\u0026amp;E image in order to robustly identify cell nuclei and aggregated 2x2\u0026micro;m capture bins into cells using bin2cell. We then used TACCO OT (\u003cstrong\u003eMages\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2023\u003c/strong\u003e) to annotate cell types based on the single-cell reference (\u003cstrong\u003eStubenvoll et al., 2025\u003c/strong\u003e) already used for deconvolution of samples PM1-\u0026zwj;4. After quality control and filtering, we identified the tumor-immune border region. The border region is defined by T/NK cells and tumor cells being located within 50 \u0026micro;m distance from each other. All cells within the tumor-immune border region were then analyzed with LIANA using the same combined database of ligand-receptor interactions as above, which identifies ligand\u0026ndash;receptor interactions at single-cell level and results were aggregated across samples (see, Methods). Despite the different technological and statistical approach, this analysis gives a similar insight into cell\u0026ndash;cell interactions at the border region as our previous analysis using the classical 10x Visium technology, albeit with a more direct attribution of ligand\u0026ndash;receptor interactions to interacting cell types.\u003c/p\u003e\n\u003cp\u003eNaturally, there are many more potentially relevant interactions in the table, some of which were recently described. For example, a pair of ligand-receptors with high expression was \u003cem\u003eAPP\u003c/em\u003e (Amyloid Beta Precursor Protein) in various combinations with CCR8, CCR4, IL18RAP, CXCR4, and CD74, was present in all three melanoma differentiation states. \u003cem\u003eAPP\u003c/em\u003e\u0026ndash;\u003cem\u003eCD74\u003c/em\u003e showed particularly high expression in transitory melanoma (Mel_trans) cells. Recently, \u003cem\u003eAPP\u003c/em\u003e has been described as an immunosuppressive molecule in glioblastoma via the APP\u0026ndash;CD74 axis (\u003cstrong\u003eMa et al., 2024\u003c/strong\u003e). Among the interactions significant in most samples involving Mel_trans_melan and Mel_trans cells with activated cytotoxic T cells were also \u003cem\u003ePCDH7\u003c/em\u003e\u0026ndash;\u003cem\u003eCXCR4\u003c/em\u003e. PCDH7 is known to be involved in cell-‑cell recognition and appears to be involved in pancreatic cancer as a cold tumor induction-related gene (\u003cstrong\u003eMochida et al., 2025).\u003c/strong\u003e In a more recent study using a CRISPR knockout screen of melanoma cells in co-culture with cytotoxic T cells, \u003cem\u003eCD58\u003c/em\u003e, \u003cem\u003eCD59\u003c/em\u003e, \u003cem\u003eIFNGR\u003c/em\u003e and \u003cem\u003eCXCR4\u003c/em\u003e were among the top perturbations in melanoma cells associated with immune checkpoint inhibitor resistance, emphasizing the principal role of CXCR4 in melanoma immune control (\u003cstrong\u003eFrangieh et al., 2021\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eIn \u003cstrong\u003eFigure 5D\u003c/strong\u003e, we summarized LIANA data for the \u003cem\u003eLGALS1\u003c/em\u003e\u0026ndash;\u003cem\u003ePTPRC\u003c/em\u003e interaction across all combinations of cell types, indicating that this mechanism is not exclusive to melanoma\u0026mdash;T cell interactions but can also be found between dendritic cells (DCs) and T cells. \u0026nbsp;Indeed, DCs have been detected at immunological borders to interfere with immune activation and galectin-1 has been regarded as a marker for immunosuppressive DCs (\u003cstrong\u003eIlarregui et al., 2009;\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eLudin et al., 2025;\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;Nov\u0026aacute;k\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;et al., 2025\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eWe further investigated whether our approach could directly identify potentially interacting cells by calculating NICHES interaction scores. Due to the high spatial resolution, we could set the edge distance cutoff to 50 \u0026micro;m to be consistent with our border region definition. \u003cstrong\u003eFigure 5E\u003c/strong\u003e shows a representative Visium HD sample (HD_PM3), completed with annotated cell types, region classification, and NICHES interaction scores. Since rendering all interaction edges across the whole sample would not work visually, we highlighted cells with high-valued incident edges in the overview plot (left) and only rendered non-zero edges for selected regions (right). Qualitatively, these findings support our LIANA results. The same analyses were performed for the other Visium HD samples (\u003cstrong\u003eSupplementary Figure 5\u003c/strong\u003e). We observed numerous interactions involving melanoma, T, and dendritic cells along the tumor\u0026ndash;immune border region.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOverall, high-definition ST at subcellular scale strongly supported our results and also identified a number of other ligand-receptor interactions that have been shown to play a role in cancer biology, but may also be of importance for melanoma immunity.\u003c/p\u003e\n\u003ch3\u003eHigh LGALS1 and PTPRC expression associate with differential survival in immunologically hot tumors\u003c/h3\u003e\n\u003cp\u003eFinally, the possible impact of \u003cem\u003eLGALS1\u0026nbsp;\u003c/em\u003eand \u003cem\u003ePTPRC\u0026nbsp;\u003c/em\u003eexpression on the prognosis of melanoma patients was analyzed to further substantiate the clinical relevance of our findings (\u003cstrong\u003eFigure 6\u003c/strong\u003e). We collected the skin cancer cohort data from \u003cem\u003eThe Cancer Genome Atlas\u003c/em\u003e (\u003cstrong\u003eCancer Genome Atlas Network, 2025)\u003c/strong\u003e via \u003cem\u003eThe\u003c/em\u003e\u003cem\u003e\u0026nbsp;\u0026zwj;Cancer Immunome Atlas\u003c/em\u003e (TCIA, http://tcia.at) webserver. The TCIA enhances the \u003cem\u003eTCGA\u003c/em\u003e data by providing estimates of cell type proportions using the quanTIseq deconvolution algorithm (\u003cstrong\u003eFinotello\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2019\u003c/strong\u003e). Additionally, we downloaded another cohort of melanoma patients published by Gide and co-workers (\u003cstrong\u003eGide et al., 2019\u003c/strong\u003e) from the Sequence Read Archive (SRA) and used quanTIseq for the deconvolution of these samples.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe examined the effect of \u003cem\u003eLGALS1\u003c/em\u003e and \u003cem\u003ePTPRC\u003c/em\u003e expression in the context of immune infiltration on patient survival using Kaplan\u0026ndash;Meyer survival analyses (\u003cstrong\u003eFigure 6\u003c/strong\u003e). For both cohorts, we used the largest homogenous subgroup to reduce interfering effects: for the TCGA cohort, we included only metastatic samples with non-zero detectable infiltration (as per quanTiseq estimates, n\u0026nbsp;=\u0026nbsp;291) and for the Gide cohort (\u003cstrong\u003eGide et al., 2019\u003c/strong\u003e), we focused on Pembrolizumab patients only (n = 71), which show a markedly different survival pattern compared to Nivolumab patients. In both cases, we separated samples into \u003cem\u003eLGALS1\u003c/em\u003e/\u003cem\u003ePTPRC\u003c/em\u003e high and low groups based on a median split first. High \u003cem\u003ePTPRC\u003c/em\u003e is generally associated with favorable outcomes (\u003cem\u003eLGALS1\u003c/em\u003e low subgroup: log-rank test p = 0.00032 in TCGA cohort and p = 0.00041 in the Gide cohort, respectively) (\u003cstrong\u003eFigure 6A and B\u003c/strong\u003e), but this effect is attenuated in patients with high \u003cem\u003eLGALS1\u003c/em\u003e expression (\u003cem\u003eLGALS1\u003c/em\u003e high subgroup: log-rank test p = 0.22 in TCGA cohort and p = 0.12 in Gide et al. cohort, respectively).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIt is known that \u003cem\u003ePTPRC\u003c/em\u003e is predominantly expressed by lymphocytes, which prompted the question of whether the positive effect on patient survival is largely the positive effect of higher T lymphocyte infiltration on patient survival. For the larger TCGA cohort, we were able to divide patients by estimated CD8\u003csup\u003e+\u003c/sup\u003e T cell content and then separate \u003cem\u003eLGALS1\u003c/em\u003e/\u003cem\u003ePTPRC\u003c/em\u003e high and low subgroups for CD8\u003csup\u003e+\u003c/sup\u003e high and low patients separately (\u003cstrong\u003eFigure 6C\u003c/strong\u003e). In the CD8\u003csup\u003e+\u003c/sup\u003e high status and high \u003cem\u003ePTPRC\u003c/em\u003e expression group, high \u003cem\u003eLGALS1\u003c/em\u003e expression was associated with bad prognosis (log-rank p = 0.0046). In contrast, the \u003cem\u003ePTPRC\u003c/em\u003e low subgroup showed no significant difference in survival based on \u003cem\u003eLGALS1\u003c/em\u003e levels (log-rank p = 0.61), which means that the level of CD8\u003csup\u003e+\u0026nbsp;\u003c/sup\u003ecytotoxic T cells alone is not sufficient to explain the survival differences. In case of CD8\u003csup\u003e+\u003c/sup\u003e low status, there was no significant difference between high and low \u003cem\u003eLGALS1\u003c/em\u003e expression, both in case of high and low \u003cem\u003ePTPRC\u003c/em\u003e expression (log-rank p = 0.46 and p = 0.61, respectively). This argues for a direct impact of galectin-1 (\u003cem\u003eLGALS1)\u003c/em\u003e on CD45 (\u003cem\u003ePTPRC)\u003c/em\u003e via an immune inactivating pathway, which is, however, without consequences in the absence of relevant CD8\u003csup\u003e+\u003c/sup\u003e T cell infiltration (cold tumors). Taken together, high \u003cem\u003eLGALS1\u003c/em\u003e expression with a negative impact on \u003cem\u003ePTPRC\u003c/em\u003e may negatively influence melanoma patient survival.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eMelanoma cell–immune cell interactions are a central mechanism of tumor control and treatment response (\u003cstrong\u003eSchadendorf et al., 2018\u003c/strong\u003e). Recent single-cell studies have revealed several different mechanisms active in this situation, defining different subpopulations of CD4\u003csup\u003e+\u003c/sup\u003e and CD8\u003csup\u003e+\u003c/sup\u003e T cells in these tumors, among which exhausted T cells appear to play a central role (\u003cstrong\u003eJerby-Arnon et al., 2018;\u003c/strong\u003e \u003cstrong\u003eSade-‍Feldman et al., 2019; Li et al., 2020\u003c/strong\u003e). However, the spatial relationship of this interaction could not be defined in these studies. With the advent of spatial transcriptomic technology in cancer science, cell-cell interactions can be analyzed in detail (\u003cstrong\u003eLarsson et al., 2021;\u003c/strong\u003e \u003cstrong\u003eSeferbekova et al., 2023\u003c/strong\u003e). The 10x ‍Genomics Visium and Visium HD platforms enable whole-transcriptome assessment at spatial resolution, facilitating unbiased description of cell-cell interactions. Here, we analyzed melanoma–‍immune cell interactions in primary melanomas, melanoma lymph node metastases, and cutaneous melanoma metastases by utilizing both platforms. We were able to identify and localize various cell populations, such as keratinocytes, melanoma cells, lymphocytes, macrophages, monocytes and endothelial cells within the melanoma and metastatic lesions and define regions of inflammatory infiltration at the proximity to melanoma cells (border regions), as well as ligand-‍receptor interactions between tumor and immune cells in these regions.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMany of these interactions at the melanoma–immune border region identified in the present study were interactions between different HLA molecules and CD3 (as part of the T cell receptor complex). This further supported the physiological relevance of our findings, since HLA-mediated antigen presentation to T cell receptor molecules is a main mechanism at the immunological synapse. However, we identified also several other putative interactions at this melanoma–immune border region. Among these were \u003cem\u003eMIF\u003c/em\u003e–\u003cem\u003eCD74\u003c/em\u003e, \u003cem\u003eLGALS1\u003c/em\u003e–\u003cem\u003eCD69\u003c/em\u003e and \u003cem\u003eLGALS3\u003c/em\u003e–\u003cem\u003eCD6\u003c/em\u003e, and \u003cem\u003eCCL19\u003c/em\u003e–\u003cem\u003eS1PR2\u003c/em\u003e and \u003cem\u003eCCL19\u003c/em\u003e–‍\u003cem\u003eS1PR3\u003c/em\u003e interactions, which had already been described to play a role in melanoma immunology (\u003cstrong\u003eLaidlaw et al., 2019; de Azevedo et al., 2020; Femel et al., 2022;\u003c/strong\u003e). \u003cem\u003eMIF\u003c/em\u003e is a lymphokine involved in cell-‍mediated immunity and an immunosuppressive factor secreted in the tissue microenvironment of melanomas (\u003cstrong\u003ede Azevedo et al., 2020\u003c/strong\u003e). \u003cem\u003eCD74\u003c/em\u003e is a chaperon for the class II MHC-complex and is thereby involved in T cell antigen presentation. In two recent studies, the interaction between \u003cem\u003eMIF\u003c/em\u003e and \u003cem\u003eCD74\u003c/em\u003e induced tolerogenic dendritic cells and M2 pro-tumorigenic macrophages in melanoma (\u003cstrong\u003ede Azevedo et al., 2020; Figueiredo et al., 2018\u003c/strong\u003e). Thus, this interaction may finally support melanoma growth. Galectin-3 (product of\u003cem\u003e\u0026nbsp;LGALS3\u003c/em\u003e gene) has been shown to act as a binding partner for the T cell receptor (TCR) and leads to downregulation of the TCR and inhibition of early T cell activation (\u003cstrong\u003eGilson et al., 2019\u003c/strong\u003e). Galectin-3 also inhibits interferon gamma diffusion into the melanoma and \u003cem\u003eex vivo\u003c/em\u003e generation of tumor-specific T cells (\u003cstrong\u003eZubieta et al., 2006\u003c/strong\u003e). Our candidate list of putative interactions also contained other less well-known interaction partners in melanoma. Among these were \u003cem\u003eAPOE\u003c/em\u003e and \u003cem\u003eSORL1\u003c/em\u003e. \u003cem\u003eAPOE\u003c/em\u003e ‍is ‍originally described as a protein involved in triglyceride lipid protein-rich metabolism. \u003cem\u003eSORL1\u003c/em\u003e ‍belongs to the low-density lipoprotein receptor (LDLR) family. The biological relevance of these findings in the tumor context remains to be studied in more detail, but may so far be supported by the fact that different APOE variants were linked to melanoma treatment response (\u003cstrong\u003eOstendorf et al., 2020\u003c/strong\u003e). We recently showed that \u003cem\u003eCD58\u003c/em\u003e, but not \u003cem\u003eCD59\u003c/em\u003e, acts as a positive immune modulator in melanoma (\u003cstrong\u003eStubenvoll et al., 2025\u003c/strong\u003e). However, the investigation of CD59 in this context requires further investigations, since CD59 is a versatile molecule and specific conformational changes may be necessary for active CD59–CD2 interaction.\u003c/p\u003e\n\u003cp\u003eInterestingly, many of the interactions found to be enriched in the border region in the primary analysis using the Visium platform could be verified in the second cohort of samples using the Visium HD platform and attributed to specific interacting cell types. The latter provides higher spatial resolution and therefore allows analysis of cell-cell interactions also in compact tumor masses such as melanomas.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn the present study, we put an emphasis on \u003cem\u003eLGALS1\u003c/em\u003e and \u003cem\u003ePTPRC\u003c/em\u003e expression which was among top interaction partners in the first sample set of four primary melanomas and present in all four samples, and in an additional set of melanoma lymph node metastases of an independent cohort (\u003cstrong\u003ePozniak\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2025\u003c/strong\u003e). Moreover, both of these molecules show an elevated expression at the tumor-‍immune border region. By use of an additional larger number of melanoma samples stained by immunofluorescence, it was shown that CD45\u003csup\u003e+\u003c/sup\u003e T cells are predominantly located at a distance between 10-50 µM to the next melanoma cell and also numerically enriched in close proximity to melanoma cells in immune hot tumor areas, as compared to cold areas. These findings further support the notion that this interaction may be of clinical relevance, as CD8\u003csup\u003e+\u003c/sup\u003e T cell infiltration is a major prognostic factor for melanoma prognosis (\u003cstrong\u003eJacquelot et al., 2017\u003c/strong\u003e; \u003cstrong\u003eFu et al., 2019\u003c/strong\u003e). More recently, the complex role of galectin-1 as an immunomodulator has been reviewed in more detail (\u003cstrong\u003eNovák et al., 2025\u003c/strong\u003e). Within the tumor tissue environment, galectin-1 may impact the differentiation of tolerogenic dendritic cells and induces the apoptosis of effector T cells and enhances the proliferation of regulatory T cells. Principally, galectin-1 is heterogeneously expressed in different types of cancers, with melanomas showing prominent upregulation compared to normal tissue and to other tumor entities (\u003cstrong\u003eNovák et al., 2025\u003c/strong\u003e). We found significant upregulation of \u003cem\u003eLGALS1\u003c/em\u003e (galectin-1) in three different data sets, in our own Visium and Visium HD data set and a Visium data set from an independent group (\u003cstrong\u003ePozniak et al., 2024\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMoreover, galectin-1 (and galectin-3) showed strong and immediate molecular binding activity to CD45 (\u003cem\u003ePTPRC\u003c/em\u003e gene product) in biolayer interferometry measurements, further arguing for an important role at the melanoma–immune border region regarding cell-cell interactions. Binding was rapid and transient but may still be strong enough to impede a CD45-mediated immune reaction. Here, blocking antibodies may improve the CD45-mediated tumor immune response. Experimental studies have shown that the galectin1–CD45 interaction exerts a number of different functions. E.g., it is known that dimeric galectin-1 binds to activated T cells and induces IL-10, which leads to the generation of regulatory T cells and subsequent immune suppression (\u003cstrong\u003eCedeno-Laurent et al. 2012\u003c/strong\u003e). Moreover, galectin-1 may suppress IL-2 and IFN-γ production in Th1 cells (\u003cstrong\u003eRabinovic et al., 2002\u003c/strong\u003e). In a more recent study, the use of the galectin-1 small molecule inhibitor LLS30 increased the anti-tumor activity of anti-PD1 treatment in immunotherapy-resistant prostate cancer (\u003cstrong\u003eWang et al., 2024\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eLittle is known about \u003cem\u003eLGALS1\u003c/em\u003e expression in other cancer contexts such as melanoma. Here, we show that \u003cem\u003eLGALS1\u003c/em\u003e expression was associated with a significant negative impact on overall survival of melanoma patients, when using a large TCGA cohort (n=291). In contrast, \u003cem\u003ePTPRC\u003c/em\u003e expression was associated with improved overall survival. These findings were associated with high CD8\u003csup\u003e+\u003c/sup\u003e T cell infiltration and essentially absent in immune cold tumors, arguing for a specific immune regulatory mechanism towards CD8\u003csup\u003e+\u003c/sup\u003e T cells. Consistent with this, overall survival increased with higher \u003cem\u003ePTPRC\u003c/em\u003e expression when focusing on patients under pembrolizumab (anti-PD1 antibody) immunotherapy in melanoma (\u003cstrong\u003eGide\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2019\u003c/strong\u003e). In a recent study, \u003cem\u003ePTPRC\u003c/em\u003e expression was associated with improved survival and response to immunotherapy in melanoma patients (\u003cstrong\u003eLi et al., 2023\u003c/strong\u003e). \u003cem\u003eLGALS1\u003c/em\u003e has also been reported as a negative prognostic factor in other cancers, e.g., ovarian, colon, and liver cancer (\u003cstrong\u003eNovák et al., 2025\u003c/strong\u003e). Taken together based on data from our own and several other studies, the interaction between galectin-1 and CD45 (protein product of \u003cem\u003ePTPRC\u003c/em\u003e) might be a suitable target for future melanoma treatment approaches.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe integrated a recent single-cell and spatial transcriptomic study in our analyses (\u003cstrong\u003ePozniak et al., 2024\u003c/strong\u003e) which put an emphasis on gene regulatory mechanisms (regulons) in treatment response and resistance. We exploited this data set, which also contained spatial transcriptomics data of melanoma lymph node metastases to further substantiate our ligand–receptor findings. In line with our data, an immune compartmentalization with a close proximity of \u003cem\u003eLGALS1\u003c/em\u003e\u003csup\u003e+\u003c/sup\u003e melanoma cells and T cells (including CD8\u003csup\u003e+\u003c/sup\u003e T cells) could also be shown for the samples with strong infiltration and a clearly defined border region. In another recent study using melanoma spatial transcriptomic technology in a murine melanoma setting (\u003cstrong\u003eKarras et al., 2022\u003c/strong\u003e), the proximity of pre-EMT neural crest stem-like cells (NCSCs) to blood vessels (perivascular niche) was analyzed, showing an anti-correlation of NCSC activity and distance to the next blood vessel. Regarding ligand-receptor interactions, authors found evidence for a Dll4–NOTCH3 interaction. Further analyses in this study were supportive for a model in which Dll4\u003csup\u003e+\u003c/sup\u003e endothelial cells stimulate melanoma growth via activation of NC stem-like melanoma cells. Finally, one earlier spatial transcriptomics study used a mutated BRAF-‍based zebrafish melanoma model and the 10x Genomics Visium technology (\u003cstrong\u003eHunter et al., 2024\u003c/strong\u003e). When looking at melanoma microenvironment interfaces, \u003cem\u003eHMGB2\u003c/em\u003e came up as one of the major genes expressed by melanoma cells, which correlates well with \u003cem\u003eHMGB1\u003c/em\u003e expression in human melanomas found in our study. Both \u003cem\u003eHMGB1\u003c/em\u003e and \u003cem\u003eHMGB2\u003c/em\u003e encode for damage-associated proteins and may induce or repress extracellular inflammatory processes. However, a detailed analysis of melanoma cell interactions with immune cells was not presented in these studies, highlighting the relevance of our work.\u003c/p\u003e\n\u003cp\u003eTaken together, unbiased whole-transcriptome spatial transcriptomics across three different data sets enabled us to identify ligand-receptor interactions enriched at the tumor-immune border region and to pinpoint the interacting cell types. This yields an important resource for understanding cell-cell communication in the melanoma tumor microenvironment and potential tumor evasion mechanisms. We provide substantial evidence that the interaction between galectin-1 and CD45 occurs at the melanoma–immune border region and may thereby impact the melanoma immune response. Future studies should be performed to prove their functional relevance in preclinical models and subsequent clinical trials.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003ch3\u003ePatient samples\u003c/h3\u003e\n\u003cp\u003eThe use of patient samples for spatial and single-cell transcriptomics was approved by the local ethics committee of the Medical Faculty of the University of Leipzig Medical Center (AZ023-16-01022016; AZ349/18-ek). Samples from humans will be used for these experiments after informed consent from the patients and according to the rules and regulations of the Declaration of Helsinki regarding Ethical Principles for Medical Research 2013. A summary of patient samples and clinical information is provided in\u003cstrong\u003e\u0026nbsp;Extended data Figure 1\u0026nbsp;\u003c/strong\u003eand\u003cstrong\u003e\u0026nbsp;Supplementary Table 1\u003c/strong\u003e. Fresh frozen tissue slices were derived from four primary melanomas. In addition, four primary melanomas and three melanoma metastases were analyzed by 10x Genomics Visium HD technology (10x Genomics). An additional set of 20 primary melanoma samples was used for immunofluorescence analysis (\u003cstrong\u003eSupplementary\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003e‍Table\u0026nbsp;1\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003ePreviously published single-cell sequencing data from ten primary melanomas, of which four are paired with the four primary melanomas used for Visium HD, are used as a reference data set for cell type annotation (\u003cstrong\u003eStubenvoll et al., 2025\u003c/strong\u003e).\u003c/p\u003e\n\u003ch3\u003eSpatial Transcriptomics\u0026nbsp;‒\u0026nbsp;10x Genomics Visium Platform\u003c/h3\u003e\n\u003ch4\u003eTissue Handling and Sectioning\u003c/h4\u003e\n\u003cp\u003eSamples were histologically annotated by an experienced histopathologist (MZ). Fresh frozen samples were cryosectioned at -10\u0026nbsp;°C (Thermo Cryostar, Fisher Scientific, Schwerte, Germany). Sections (10\u0026nbsp;µm) of four primary melanomas were placed on chilled Visium Tissue Optimization Slides and Visium Spatial Gene Expression Slides (10x Genomics, Pleasanton, CA, U.S.A.) and adhered by warming the back of the slide. Tissues were processed as recommended by the manufacturer.\u003c/p\u003e\n\u003ch4\u003eTissue optimization\u003c/h4\u003e\n\u003cp\u003eFor tissue optimization, tissue sections were fixed in chilled methanol and stained according to the 10x ‍Genomics Visium spatial tissue optimization user guide. Fluorescent images were taken using a Nikon Eclipse Ti with a Cy3 filter (ex/em brand) using a PlanFluro10X Ph1 objective and 900 ms exposure time. The optimal permeabilization time was determined to be 18 min.\u0026nbsp;\u003c/p\u003e\n\u003ch4\u003eSpatial gene expression\u003c/h4\u003e\n\u003cp\u003eFor spatial gene expression, tissue sections were fixed in chilled methanol and H\u0026amp;E stained according to the Visium Spatial Gene Expression User Guide. Brightfield histology images were obtained using a PlanFluro10X Ph1 objective (400 µs) on a Nikon Eclipse Ti (11746 × 11746 pixels). Raw images were stitched together using NIS-Elements AR 5.21.02 (Nikon) and exported as .tif files with high- resolution settings.\u003c/p\u003e\n\u003cp\u003eThe sequencing libraries were generated according to the user guide with an 18-minute permeabilization time, they were quantified (dsDNA HS Kit, Thermo Fisher), the molarity of each library was calculated, and equal molar amounts were pooled. Sequencing was performed with a 28-10-10-90 read setup on an Illumina HiSeq (11 pM loading concentration including 1% PhiX) using a Rapid SBS (200 cycles) Kit v2; or a NextSeq\u0026nbsp;2000 (650 pM loading concentration including 2% PhiX) using a P2 Reagent (200 cycles) kit.\u003c/p\u003e\n\u003ch4\u003eRaw data processing\u003c/h4\u003e\n\u003cp\u003eSequencing data were analyzed using 10x Genomics SpaceRanger software version 1.3.1, with default parameter settings. We used \u003cem\u003espaceranger mkfastq\u003c/em\u003e to demultiplex and convert BCL files into FASTQ files. For read mapping and gene expression quantification, \u003cem\u003espaceranger count\u003c/em\u003e was performed. Reads were mapped to the human genome version GRCh38/hg38 provided by 10x Genomics. For gene expression quantification, the reference provided by 10x Genomics was used (2020‑A Human GRCh38 (GENCODE v32/Ensembl98)). We used the Loupe browser v5.1 to transfer histological sample annotations onto the spots. During this process, we manually removed a few disconnected spots from the main tissue region.\u0026nbsp;\u003c/p\u003e\n\u003ch4\u003eData analysis, quality control and preprocessing\u0026nbsp;\u003c/h4\u003e\n\u003cp\u003eFurther analyses of the quantified gene expression matrices were carried out in R v4.4.2 (\u003cstrong\u003eR Core Team, 2021\u003c/strong\u003e). We used Seurat v5.3.0 (\u003cstrong\u003eHao et al., 2024\u003c/strong\u003e) and semla v1.3.2 (\u003cstrong\u003eLarsson et al., 2021\u003c/strong\u003e) to import gene counts per spot matrix and tissue images.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo assess sample quality, we visualized the distribution of genes and counts per spot and percentages of mitochondrial and ribosomal genes per spot across all spots of all samples and determined conservative cutoffs for removing low-quality spots. Subsequently, all spots were removed with less than 80 detected genes, less than 100 total counts, or more than 20% of counts mapped to mitochondrial genes. The number of spots removed per sample is shown in \u003cstrong\u003eSupplementary Table 1\u003c/strong\u003e. Furthermore, the average expression of the housekeeping genes curated by Tirosh and co-workers was computed (\u003cstrong\u003eTirosh et al., 2016\u003c/strong\u003e) for each spot. Sufficient expression was observed in all spots, with variations depending on the tissue region.\u003c/p\u003e\n\u003cp\u003eWe applied multiple task-specific pre-processing schemes. For Louvain clustering and generation of Uniform Manifold Approximation and Projection (UMAP) embeddings (\u003cstrong\u003eBecht\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2018\u003c/strong\u003e), we used the recently improved Seurat SCTransform workflow. In short, counts were preprocessed by applying the variance-stabilizing transformation (provided by the Seurat function \u003cem\u003eSCTransform\u003c/em\u003e with settings \u003cem\u003evst.flavor=”v2”\u003c/em\u003e) using the recently improved version (\u003cstrong\u003eChoudhary\u0026nbsp;and\u0026nbsp;Satija,\u0026nbsp;2022\u003c/strong\u003e). Based on this, the first 30 principal components were computed using the highest variance explanation. A ‍two-‍dimensional Uniform Manifold Approximation and Projection (UMAP) embedding using 30 ‍principal components were calculated. To compute the mean expression values and \u003cem\u003eNICHES\u003c/em\u003e interaction scores later on, we additionally performed the \u003cem\u003eSeurat\u003c/em\u003e standard preprocessing workflow using natural log transformation.\u003c/p\u003e\n\u003cp\u003eLog-transformed values were computed with and without library size normalization. Although it is best practice for single-cell data, the value of normalizing the observed counts per spot to a common value is still a debated topic. We observed, in agreement with two recent studies (\u003cstrong\u003eSaiselet\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2020, Bhuva\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2024\u003c/strong\u003e), that the total gene expression levels per spot appear to have strong biological relevance. While scoring the mean expression or module scores of cell type-specific genes, we observed artifacts introduced by library-size normalization between regions with very different total expression (immune cells in tumor vs. stromal regions). For library-size normalization, we used the \u003cem\u003eSeurat\u003c/em\u003e function \u003cem\u003eNormalizeData\u0026nbsp;\u003c/em\u003ewith a scale factor of 10,000. For data without library-size normalization, we added a pseudocount of 1 to the observed counts and then computed the natural logarithm. After the initial computation of quality control metrics, all counts belonging to mitochondrial and ribosomal genes were filtered before any further analyses were performed.\u003c/p\u003e\n\u003ch4\u003eDeriving clusters of co-expressing spots\u003c/h4\u003e\n\u003cp\u003eClusters were computed for co-expressing spots (representing regions with similar cell type composition and state) for each Visium sample using Louvain clustering of spots. We used principal component reduction based on variance-stabilized data described earlier. Using the selected 30 ‍principal components, a shared nearest-neighbor graph was generated, which was subsequently clustered using the Louvain Algorithm. For each sample, multiple clusters were computed with seven different resolutions evenly spaced between 0.1 and 1.3.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCluster stability was assessed using a cluster tree plot (\u003cstrong\u003eZappia and Oshlack, 2018\u003c/strong\u003e). Clustering results were compared with histological annotation to determine the clustering with the best concordance to histological annotation and highest plausibility. Correspondence to histological annotation was assessed by plotting the histological annotation vs. cluster number correspondence matrix, as well as a variety of correspondence metrics (notably Adjusted Rand Index, Normalized Mutual Information, Normalized Variation of Information and Overlap Coefficient) using the \u003cem\u003emclustcomp\u003c/em\u003e package. Histological annotations were not regarded as a ground truth to match completely, since it was expected that we would obtain a finer resolution of tissue region borders and possibly a finer resolution of overall structures in the tissue. Therefore, resolutions were selected that would not join annotated histological features into a single cluster based on visual assessment and otherwise matched the histologically annotated region borders well (by correspondence metrics, mostly Overlap Coefficient) while avoiding excessive subclustering. In doubt, we selected the resolution that gave the best overview of the tissue structures. The resulting choices of clustering resolutions are summarized in \u003cstrong\u003eSupplementary Table 1C\u003c/strong\u003e.\u003c/p\u003e\n\u003ch4\u003eCo-expression cluster annotation\u0026nbsp;\u003c/h4\u003e\n\u003cp\u003eWe annotated the spot clusters using a three-step approach: comparison with histological annotations, overexpression of marker genes, and pathway enrichment. To determine the overexpressed genes, we performed a differential gene expression analysis for each cluster in each sample by performing a Wilcoxon rank-sum test on the log-transformed size-normalized expression data. Using the \u003cem\u003eFindMarkers\u003c/em\u003e function implemented in Seurat with \u003cem\u003elogfc.threshold=0.15\u003c/em\u003e and default parameters otherwise, we compared the expression of all spots in each cluster to the union of all other clusters. Adjusted p-values for multiple testing were calculated using Bonferroni’s method. A gene was considered to be significantly differentially expressed if the adjusted p-value was \u0026lt; 0.01.\u003c/p\u003e\n\u003cp\u003eGene enrichment analysis for significantly overexpressed genes was performed using the R package clusterProfiler (\u003cstrong\u003eWu et al., 2021\u003c/strong\u003e) and the Gene Ontology (GO) database of biological processes (GO:BP) (\u003cstrong\u003eAshburner et al., 2000\u003c/strong\u003e). The significance of enrichment was assessed by a hypergeometric test and adjusted p-values for multiple testing were calculated based on the Benjamini-Hochberg (\u003cstrong\u003eBenjamini and Hochberg,\u0026nbsp;1995\u003c/strong\u003e) method. All GO terms with FDR\u0026lt;0.01 are considered significantly over-represented. In addition, overrepresented GO terms were further filtered to contain at least five of the significantly differentially expressed genes of a cluster or, in case of clusters with few differentially expressed marker genes, GO terms were required to contain at least 10% (with a minimum of two) of the differentially expressed genes in a cluster.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBased on this information, cluster names were determined by an experienced histopathologist (MK).\u003c/p\u003e\n\u003ch4\u003eSpot deconvolution and annotation\u0026nbsp;\u003c/h4\u003e\n\u003cp\u003eSince regular Visium spots are considerably larger than single cells and thus contain mixed expression signals from different cell types, we used reference-based deconvolution to estimate the cell type proportions for each spot. Reference-based deconvolution algorithms use the information from an annotated single cell reference dataset to estimate spot composition. We used our recently published single-cell dataset (\u003cstrong\u003eStubenvoll et al., 2025\u003c/strong\u003e) as reference to deconvolve Visium samples using RCTD (\u003cstrong\u003eCable et al., 2022\u003c/strong\u003e) and CARD (\u003cstrong\u003eMa et al., 2022\u003c/strong\u003e) algorithms. Since we know that deconvolution algorithms can be sensitive to the quality and granularity of the cell type annotation of the reference, we ran deconvolution on the original annotations and two coarser annotation levels, where cell subtypes were combined. The deconvolution results used in this work are averages of two CARD deconvolution runs on medium and coarse annotation granularity.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBecause of considerable uncertainty of the deconvoluted cell type proportions in preliminary tests and to get an estimate of cell type activity in absolute terms (not composition), we further computed per spot scores indicating the presence or absence of specific cell types based on marker gene panels. Two distinct sources of marker genes were used: the Panglao database (\u003cstrong\u003eFranzén et al., 2019\u003c/strong\u003e) and a custom panel of genesets derived from our single-cell reference (\u003cstrong\u003eStubenvoll et al., 2025\u003c/strong\u003e) plus the published melanoma datasets GSE115978 generated by Jerby-Arnon and co-workers (\u003cstrong\u003eJerby-Arnon et al., 2018\u003c/strong\u003e) and GSE130973 generated by Solé-Boldo and co-workers (\u003cstrong\u003eSolé-Boldo et al., 2020\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn short, we used the cell types provided by the authors in each case and, where necessary, added a coarser annotation with collapsed cell subtypes. We then updated gene symbols in order to match datasets, removed untyped cells and used downsampling to a maximum of 5000 cells per type for the Stubenvoll and co-workes dataset to combat cell type imbalance. We then employed deconvolution preprocessing from the BayesPrism (\u003cstrong\u003eChu et al., 2022\u003c/strong\u003e) pipeline to exclude genes with low expression and susceptibility to batch effects and finally used BayesPrism’s get.exp.stat function to conduct pairwise t-tests between all pairs of subtypes from different cell types and combine results to get highly cell type‒specific marker genes for each type across the three references. Finally, we aggregated results from all three references, collecting maximum p-values and minimum log-fold changes across the references each cell type appeared in, then retained the 30 most specific marker genes from the combined data.\u003c/p\u003e\n\u003cp\u003eWith the marker gene sets obtained in that way, we used the Seurat function AddModuleScore to compute the average expression minus the expression of a set of randomly selected control genes, stratified by average expression level via binning (\u003cstrong\u003eTirosh et al., 2016\u003c/strong\u003e). We calculated all module scores for size-normalized and non-normalized log-transformed counts.\u003c/p\u003e\n\u003ch4\u003eDelineation of the tumor–immune border in Visium samples\u003c/h4\u003e\n\u003cp\u003eTo determine which spots were marked as positive for immune or tumor cells, we combined the information gained from deconvolution and cell type‒specific module scores. We marked spots as positive for either type only if both approaches support this.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFor deconvolution results, we applied a cutoff of 0.1 on the estimated cell type fractions to determine spots as positive for a given cell type (the thresholds on the y-axes in \u003cstrong\u003eFigure 2F\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eBecause the tumor/immune cell contents and expression levels varied considerably between different samples, we aimed to find a robust model for the module score distributions that delineates spots with strong tumor/immune signatures from background spots without requiring manual determination of thresholds for each sample.\u003c/p\u003e\n\u003cp\u003eFor lymphocyte module scores, we could reasonably assume small areas with high expression and a large number of neutral background spots. We therefore modelled the background distribution using the median and Q\u003csub\u003en\u003c/sub\u003e as robust estimators of location and scale (using the robustbase R package). We used the inferred location and scale of the background distribution (assumed gaussian) to compute z-scores from our module scores followed by p-values. P-values were then FDR adjusted for the number of spots in each sample. We marked spots with adjusted p-‑values \u0026lt; 0.05 as potentially positive for the cell type in question (this corresponds to the thresholds on the x-axes in \u003cstrong\u003eFigures 2E and F middle and right\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBecause we cannot make this assumption for tumor cells (whose content varied widely between samples), we fitted a custom Gaussian mixture model to the tumor-cell scores in each sample using the \u003cem\u003emixtools\u0026nbsp;\u003c/em\u003epackage (3 equal-variance plus one independent component, which empirically worked best). After fitting, the component with highest mean represents the main tumor component, and we determined the threshold for assigning tumor-positive cells as the lowest point where the tumor component density exceeds the aggregate density of all other components (this corresponds to the threshold on the x-axes in \u003cstrong\u003eFigures 2E and F left\u003c/strong\u003e; the tumor component is marked in red and the sum of background components dull green).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSpots were finally marked as positive for a given cell type if they surpassed both thresholds on the deconvolved percentage and the module score. We defined the tumor-immune-border region as the collection of spots positive for tumor and immune cells, spots positive for tumor cells that are directly adjacent to spots positive for immune cells, and spots positive for immune cells that are directly adjacent to spots positive for tumor cells (“interfacing spots”, see \u003cstrong\u003eFigure 2H\u003c/strong\u003e). All other spots that were positive for either cell type but not in the border region served as comparison group for enrichment tests later (“non-interfacing”). Spots not enriched for any cell type of interest were marked as background and not used any further.\u003c/p\u003e\n\u003ch4\u003eComputation of spatial interaction scores for ligand–receptor pairs on Visium samples\u003c/h4\u003e\n\u003cp\u003eTo analyze ligand–receptor interactions (LRI) between cells in close proximity, we used the NICHES (\u003cstrong\u003eRaredon\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2023\u003c/strong\u003e) framework to compute ligand–receptor interaction scores.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNICHES computes interaction scores as the product of the expression of the ligand and receptor genes, such that a computed interaction score is always zero if either the ligand or the receptor is not expressed at all and high if both are strongly expressed. This framework incorporates spatial information by allowing ligands and receptors to be expressed in different spots by calculating an interaction score for each directed edge between interacting spots (once for self-edges). \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFor the 10x Genomics Visium® platform, each capture spot had a maximum of six neighboring spots (55\u0026nbsp;µm center-to-center distance), with fewer neighbors at the edge of the capturing area. Therefore, we restricted interacting spot pairs to those that are physically neighboring, or a spot interacting with itself (self-edges).\u003c/p\u003e\n\u003cp\u003eWe used the union of the curated databases of ligand–receptor mechanisms provided in the NicheNet (\u003cstrong\u003eBrowaeys\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2020\u003c/strong\u003e) and LIANA (\u003cstrong\u003eDimitrov\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2022\u003c/strong\u003e) packages to determine biologically plausible ligand-receptor interactions for which interaction scores were to be computed. This yielded a total of 14,977 distinct mechanisms after accounting for changed gene symbols and deduplication.\u003c/p\u003e\n\u003ch4\u003eEnrichment test for ligand–receptor interactions in Visium samples\u003c/h4\u003e\n\u003cp\u003eTo test for the specific enrichment of LRI between immune cells and tumor cells in our Visium samples, permutation-based enrichment tests of the NICHES interaction scores were conducted. We calculated the enrichment between spots in the tumor-immune border region (“interfacing”) versus tumor and immune-positive spots not in that border region (“non-interfacing”).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eEnrichment was calculated using a model-free permutation-based test as implemented by the \u003cem\u003ediff_mean_test\u003c/em\u003e function in the \u003cem\u003eSCTransform\u003c/em\u003e R package. For this test, a null distribution is built by randomly permuting the labels (in/not in the border region) of all spots in the compared groups and recording the observed difference in arithmetic mean interaction scores between the resulting groups. The mean and standard deviation of the null distribution are subsequently used to turn the real observed difference in mean interaction scores between the compared regions into z-scores and then into p-values. Finally, p-values for all the scored putative ligand–receptor pairings are adjusted using the Benjamini-Hochberg method.\u003c/p\u003e\n\u003cp\u003eWe aggregated the ligand–receptor interaction enrichment results across all our Visium samples plus the 3 samples from (\u003cstrong\u003ePozniak et al., 2024)\u003c/strong\u003e with the most prominent tumor–immune border region. To better attribute the expression of all ligands and receptors to cell types in melanoma samples, we added annotations derived from our single-cell reference. In short, single-cell counts were summed across cells with the same type, then normalized to 10,000 counts per cell and log-transformed with an added pseudocount of 1, to get estimated expression per cell type. Expression of ligand- or receptor complexes was calculated as the geometric mean of the component expressions. Using the geometric mean here has the benefit that a complex ends up with zero expression if any of its components have zero expression, avoiding artifactual expression of complexes that cannot possibly exist in our samples. For visualization, we computed the empirical quantiles in the distribution of expressions and calculated z-scores as quantiles of a normal distribution matching the empirical expression quantiles.\u003c/p\u003e\n\u003cp\u003eTo determine the list of top candidates considered for further validation, we retained only candidates that were significantly enriched (adjusted p-value \u0026lt; 0.05) in the tumor–immune border region in 3 or more melanoma samples.\u003c/p\u003e\n\u003cp\u003eCircos plots of LRIs were generated by aggregating all retained ligand–receptor pairs and connecting them in a circular diagram using the \u003cem\u003ecirclize\u003c/em\u003e R package (\u003cstrong\u003eGu\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2014\u003c/strong\u003e). Genes were separated into sections for ligands, receptors and genes that appear in ligand and receptor position in different mechanisms. The link colors for \u003cstrong\u003eFigure 3C\u003c/strong\u003e were generated by generating a 3-dimensional UMAP transformation of the expression z-score by cell type matrix, normalizing each component to the unit interval and assigning each axis to one color in RGB space. In this way, genes with similar expression pattern across cell types (as in \u003cstrong\u003eFigure 3B right\u003c/strong\u003e) end up with similar colors in the circos plot and links are colored according to their ligand.\u003c/p\u003e\n\u003ch3\u003eSpatial Transcriptomics\u0026nbsp;‒\u0026nbsp;10x Genomics Visium HD Platform\u003c/h3\u003e\n\u003ch4\u003eSpatial gene expression\u003c/h4\u003e\n\u003cp\u003eFor Visium HD, formalin-fixed paraffin embedded (FFPE) tissues from seven samples were analyzed. The RNA quality of the embedded samples\u0026nbsp;was checked using the Agilent 5200 fragment analyzer. The DV200 value was at least 48% for primary melanoma and 61% for melanoma metastasis samples. FFPE tissue blocks were sectioned at a thickness of 5\u0026nbsp;µm using a rotary microtome and placed on glass slides. Hematoxylin and eosin (H\u0026amp;E) staining was performed according to the 10× Genomics protocol. After staining, the slides were imaged at 40× magnification via an AxioScan (Zeiss, Germany). The captured RNA was transferred to Visium HD slides using the CytAssist. cDNA synthesis and library construction were performed using the Visium HD Library Preparation Kit (10× Genomics). Briefly, mRNA was captured by barcoded probes, reverse transcribed into cDNA and amplified for sequencing library preparation Finally,sequencing was performed on a NextSeq 2000 instrument (Illumina) using P4 XLEAP SBS reagent kits (100 cycles) with a 43-10-10-50 read setup. The final molarity of the library pool on the sequencing flow cell was 650 pM including 1% PhiX library.\u003c/p\u003e\n\u003ch4\u003eRaw data processing\u003c/h4\u003e\n\u003cp\u003eWe used Illumina BCL Convert in version 4.1.7 to demultiplex and convert BCL files into FASTQ files. Sequencing data analysis used 10x Genomics SpaceRanger software version 4.0.1, with default parameter settings. For read mapping and gene expression quantification, \u003cem\u003espaceranger count\u003c/em\u003e was performed. Reads were mapped and quantified using transcriptome reference 2024-A Human GRCh38 (GENCODE v44/Ensembl110 annotations) and the probe set Visium Human Transcriptome Probe Set v2.1.0 provided by 10x Genomics.\u0026nbsp;\u003c/p\u003e\n\u003ch4\u003eData analysis, quality control and preprocessing\u0026nbsp;\u003c/h4\u003e\n\u003cp\u003eThe Visium HD processed samples have been analyzed by first applying stain deconvolution to the histological microscopy images. Stain deconvolution was performed using scikit-image (\u003cstrong\u003evan\u0026nbsp;der\u0026nbsp;Walt\u0026nbsp;et\u0026nbsp;al., 2014\u003c/strong\u003e) to separate the hematoxylin and eosin color channels. The images with the hematoxylin color channel are used as input to the bin2cell workflow for cell segmentation using stardist (\u003cstrong\u003eWeigert and Schmidt, 2022\u003c/strong\u003e) and bin count aggregation. The cell\u0026nbsp;by\u0026nbsp;gene count matrix is stored as anndata (\u003cstrong\u003eVirshup\u0026nbsp;et\u0026nbsp;al., 2024\u003c/strong\u003e) object. Cell type label transfer was performed using the TACCO OT (optimal transport) method. As input for the cell type label transfer, the dataset from Stubenvoll and co-workers (\u003cstrong\u003eStubenvoll\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2025\u003c/strong\u003e) was used. This dataset contains single-cell RNA sequencing data from 10 primary melanoma samples of which pm6 is paired with HD_PM1, pm7 is paired with HD_PM2, pm9 is paired with HD_PM3 and pm10 is paired with HD_PM4. We repeated label transfer using two different annotation levels of the reference (with and without cell subtypes) and removed spots with non-matching combinations of coarse and fine annotation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eExtended Figure 7\u003c/strong\u003e shows the transcript distribution for some genes of interest. In these plots, cells that contain at least a single read of a gene are plotted as a spot. Zoomed-in plots show LGALS1 and PTPRC expressing cells at identical locations as zoomed in plots in \u003cstrong\u003eFigure 5E\u003c/strong\u003e and \u003cstrong\u003eSupplementary Figures 5A-F\u003c/strong\u003e.\u003c/p\u003e\n\u003ch4\u003eLigand–receptor analysis of Visium HD samples\u003c/h4\u003e\n\u003cp\u003eWe used the preprocessed and annotated Visium HD data to support our analysis with direct evidence of the interacting cell types through the superior resolution of the Visium HD technology. After removing low-quality cells (at least 5 and less than 150 bins per cell, at least 40 gene counts), we annotated the tumor–immune border region by calculating, for each cell, the distances to closes tumor and T/NK cell and then assigned classes based on a distance threshold of 50 µm.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe used all spots in the border region to run the whole LIANA analysis pipeline using the same database of ligand–receptor interactions that we used for the analysis of the Visium data. Aggregated p-values from the pipeline output were filtered with a cutoff of 0.01 and retained results aggregated and visualized with custom functions based on the LIANA plotting functions.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo pinpoint interactions directly, we additionally used the NICHES package with slight modifications to the code to ensure feasible runtime and memory usage due to the high number of cells in the HD samples. We again used 50 µm as the distance cutoff for interactions in NICHES. Average interaction values for each cell were calculated for LGALS1–PTPRC by averaging all adjacent edges for each cell. Visualizations for these data were done based on the ggplot2 and ggraph packages.\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003eMulticolor immunofluorescence\u003c/h3\u003e\n\u003cp\u003eFormalin-fixed, paraffin-embedded (FFPE) sections (7\u0026nbsp;µm) of a total set of 20 malignant melanoma samples were mounted on charged microscopy slides, were used for immunofluorescence (IF) staining. After deparaffinization, heat-mediated antigen retrieval was performed by placing the slides in hot citrate buffer (pH\u0026nbsp;6.0) for 45\u0026nbsp;minutes. The sections were then blocked with 10% fetal bovine serum (FBS) for 45\u0026nbsp;minutes before applying the primary antibodies at the appropriate concentrations. Following overnight incubation, the secondary antibodies (1:500) and DAPI (1:500) were applied, followed by mounting with Prolong™ Gold antifade mountant. The following antibodies and concentrations were used: CoraLite®594-conjugated mouse anti-SOX10 (1:200) (Proteintech Group, Rosemont, IL, USA, CL594-66786, 1D2C8), CoraLite®488-conjugated mouse anti-galectin-1 (1:400) (Proteintech Group, Rosemont, IL, USA, CL488-60223, 3G10D2), AlexaFluor®647-conjugated rabbit anti-CD45 (1:400) (Abcam, Cambridge, U.K., ab305209, EPR20033), and rabbit anti-CD3 (Novus Biologicals, Centennial, CO, U.S.A., NB600-1441, SP7). Secondary antibodies included goat anti-rabbit Alexa Fluor 647 (Invitrogen, Darmstadt, Germany, A21245)\u003c/p\u003e\n\u003cp\u003eFluorescence microscopy was performed at 20x and 40x magnification. Images were captured in multiple fluorescence channels, and JPEG formats were saved for subsequent analysis. Analysis was conducted using QuPath (version 0.5.1). For the quantification of CD3\u003csup\u003e+\u003c/sup\u003e cells to determine immune status, a minimum of three 20x images per section were analyzed. One to three representative 20x images were used for galectin-1‑CD45 distance measurements. Positive cell detection was based on DAPI staining to identify nuclei, and measurements of marker expression were extracted from the corresponding color channel (RGB). Thresholds were set manually for each marker to ensure accurate detection and quantification and were applied across all images.\u003c/p\u003e\n\u003cp\u003eFor cell distance measurements, image segmentation results based on anti-CD45, anti-SOX10, and anti-galectin-1 staining were exported from QuPath as .tsv files and subsequently processed in Python (v3.8.8) using pandas v1.5.3, numpy v1.24.2, matplotlib v3.7.1, seaborn v0.13.2, and scipy v1.10.1 (spatial analysis via distance_matrix and cKDTree). CD3 quantification was performed to classify the melanoma samples according to their immune status, based on the ratio of CD3⁺ positive cells to the total number of cells. Tumors were classified based on CD3⁺ cell infiltration as follows: 'cold' tumors contained ≤10% CD3⁺ cells, 'intermediate' tumors had 10–20% CD3⁺ cells, and 'hot' tumors had \u0026gt;20% CD3⁺ cells. Group differences (cold, intermediate, hot) were assessed using Welch’s ANOVA following a Brown-Forsythe test for variance heterogeneity. Dunnett’s T3 post hoc test was used for multiple comparisons. Statistical significance was defined as p \u0026lt; 0.05 (*), p \u0026lt; 0.01 (), and p \u0026lt; 0.001 (*)\u003c/p\u003e\n\u003ch3\u003eDetermination of protein-protein Interactions\u0026nbsp;\u003c/h3\u003e\n\u003ch4\u003eExpression and purification of recombinant LGALS1 and LGALS3\u003c/h4\u003e\n\u003cp\u003eGene sequences for LGALS1 and LGALS3 were obtained through custom synthesis from IDT as eBlocks and were subsequently cloned into pET21a (+) vector via Gibson Assembly. Proteins were expressed with an N-terminal his-tag from \u003cem\u003eE. coli\u003c/em\u003e BL21 (DE3) cultures. Protein overproduction was induced at OD ‍~ 0.7 with a final concentration of 50 µM isopropyl-β-D-thiogalactopyranoside (IPTG). Cells were harvested through centrifugation at 4 °C after 4 h of incubation at 37 °C and 210 rpm and lysed by sonication. The clarified lysate was purified via immobilized metal affinity chromatography (IMAC) (Ni‑NTA Agarose column) (Quiagen, Hilden, Germany) followed by size-exclusion chromatography (SEC) over a Superdex 75 pg gel filtration column (Cytiva, Marlborough, MA, U.S.A.). LGALS1 purification was performed under reducing conditions. All Elution and washing reagents for reducing purification contained 8 mM Dithiothreitol (DTT) and the imidazole containing elution solution contained 10 mM 2‑Mercaptoethanol (BME).\u0026nbsp;\u003c/p\u003e\n\u003ch4\u003eDetermination of binding affinity with bio-layer interferometry (BLI) of PTPRC and LGALS1 and LGALS3\u003c/h4\u003e\n\u003cp\u003eAffinity measurements were conducted in an Octet® R8 system with Octet® ProA Biosensors (Sartorius) at 27 °C. Samples were diluted in HBS-EP+ buffer (0.01 M 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid, 0.15 M NaCl, 3 mM ethylenediaminetetraacetic acid, 0.05% Tween 20, 1% bovine serum albumin, pH 7.4). T1 Anti-NY-ESO-1 Antibody was obtained after expression in human derived HEK Expi 293 cells and subsequent Protein A affinity purification and buffer exchange. Fc-‑tagged CD45 (SinoBiology, 10086-H02H) and the control antibody were diluted to 5 µg/mL and loaded onto the ProA Biosensors. The tips with bound protein were washed in HBS-EP+ buffer for 60 s and separately transferred to LGALS1 or LGALS3 solutions with concentrations ranging from 3000 nm to 0 nM for an association period of 120 s, followed by dissociation in HBS-EP+ buffer for 120 s. Tips were regenerated with 10 mM glycine pH 1.5.\u0026nbsp;\u003c/p\u003e\n\u003ch4\u003eData Processing and determination of dissociation constant K\u003csub\u003eD\u003c/sub\u003e\u003c/h4\u003e\n\u003cp\u003eData processing was performed in Octet® BLI Analysis version 12. Preprocessing of the raw data involved subtraction of reference well, sensor axis alignment and Savitzky-Golay-Filtering. Data was then fitted to 1:1 binding curves with global fitting in kinetic analysis allowing different R\u003csub\u003emax\u003c/sub\u003e to ensure best fitting model. The steady state affinity was determined using the theoretical equilibrium binding response data (R\u003csub\u003eeq\u003c/sub\u003e) at the end of the association.\u003c/p\u003e\n\u003ch4\u003eOrdered amino acid sequences:\u0026nbsp;\u003c/h4\u003e\n\u003cp\u003eGalectin-1:\u003c/p\u003e\n\u003cp\u003eHHHHHHMACGLVASNLNLKPGECLRVRGEVAPDAKSFVLNLGKDSNNLCLHFNPRFNAHGDANTIVCNSKDGGAWGTEQREAVFPFQPGSVAEVCITFDQANLTVKLPDGYEFKFPNRLNLEAINYMAADGDFKIKCVAFD\u003c/p\u003e\n\u003cp\u003eGalectin-3:\u003c/p\u003e\n\u003cp\u003eHHHHHHMFHILRLESTVDLSEPLKDNGIIVFQSDKLDLEPSPNLGPTGIDNTNVNLINAKGDVLLHIGIRRRENAFVFNSIPYGESRGPEERIPLEGTFGDRRDPSITIFDHPDRYQIMIDYKTVYYYKKRLEGRCEKVSYKINEGQTPPFSDVLGVTVLYFANVMPRAN\u003c/p\u003e\n\u003ch3\u003eIn-silico structure prediction\u003c/h3\u003e\n\u003cp\u003eTo identify a possible interaction interface between galectin-1 and CD45 first, we searched in the Human PPI database created by (\u003cstrong\u003eZhang et al., 2025\u003c/strong\u003e) for existing complex predictions. They used a striped version of CD45 (P08575_S1:226-575) without parts of the extracellular, membrane-spanning, and intracellular domains (model ID: P08575_S1__P09382_S0), sequence see \u003cstrong\u003eSupplementary Table ‍4A\u003c/strong\u003e, ID P08575_S1) with a contact probability of 0.06573 (Alphafold2 prediction). The model is shown in \u003cstrong\u003eSupplementary Figure 4A\u003c/strong\u003e, with galectin-1 positioned inside the membrane, indicating a failed complex prediction. To remove the bias for the galectin-1 position predicted at the lower end of CD45, the membrane spanning region and parts of the intracellular domain were introduced in a new structure prediction run, using Boltz2\u0026nbsp;(\u003cstrong\u003ePassaro et al., 2025\u003c/strong\u003e) in template mode for CD45 with enabled potentials, 20 predictions, and 10 seeds. The model showed that galectin-1 was no longer predicted at the lower end of the extracellular domain, but rather at the intracellular part of CD45 (\u003cstrong\u003eSupplementary Figure 4B\u003c/strong\u003e) with acceptable overall model confidence but low interface reliability (see \u003cstrong\u003eSupplementary Table 4B\u003c/strong\u003e, Pair ID: P08575_strip + Gal1). Consequently, the intracellular part in the sequence was removed, and a new model was generated with the same settings and seeds, resulting in a high-confidence structural model with near-high-quality predicted relative positions between the extracellular domains of CD45 and galectin-1 (see \u003cstrong\u003eSupplementary Figure 4C-D\u003c/strong\u003e and \u003cstrong\u003eSupplementary Table 4B\u003c/strong\u003e, Pair ID: P08575_stripIntra + Gal1). Even though not all models display these metrics, it is still possible to identify some highly confident (confidence score ≥ 0.909, ipTM ≥ 0.759) models in this binding interface (\u003cstrong\u003eSupplementary Figure 4D-F\u003c/strong\u003e).\u003c/p\u003e\n\u003ch3\u003eSurvival analysis\u003c/h3\u003e\n\u003cp\u003eThe possible impact of selected ligand–receptor interaction partners on the prognosis of melanoma patients was analyzed. For this purpose, we collected the skin cancer cohort data from \u003cem\u003eThe Cancer Genome Atlas\u003c/em\u003e (TCGA\u003cem\u003e; Skin Cancer Melanoma\u0026nbsp;\u003c/em\u003e(SKCM) cohort) from \u003cem\u003eThe Cancer Immunome Atlas\u003c/em\u003e (TCIA, http://tcia.at) webserver, which enhances the \u003cem\u003eTCGA\u003c/em\u003e data by providing estimates of cell type proportions using the quanTIseq deconvolution algorithm (\u003cstrong\u003eFinotello et al., 2019\u003c/strong\u003e). RNA-seq data from TCGA were obtained from the recount2 database (\u003cstrong\u003eCollado-Torres\u0026nbsp;et\u0026nbsp;al.,\u0026nbsp;2017a\u003c/strong\u003e). Gene expression values and sample metadata were downloaded via the R/Bioconductor package recount (\u003cstrong\u003eCollado-‍Torres et al., 2017b\u003c/strong\u003e). The expression values were normalized to transcripts per million (TPM) using the \u003cem\u003egetTPM\u003c/em\u003e function.\u003c/p\u003e\n\u003cp\u003eAdditionally, another cohort of melanoma patients published by Gide and co-workers was downloaded (\u003cstrong\u003eGide et al., 2019\u003c/strong\u003e) from the Sequence Read Archive (SRA) and used quanTIseq for the deconvolution of these samples.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe examined the effect of \u003cem\u003eLGALS1\u003c/em\u003e and \u003cem\u003ePTPRC\u003c/em\u003e expression on patient survival using the survival and survminer R packages to conduct Kaplan–Meyer survival analyses and log-rank tests. After brief analysis of the mayor subtypes in each cohort (primary melanoma/metastasis for TCGA, primary treatment in the Gide cohort), we retained only the larger, more homogenous group in each case (metastases for TCGA, Pembrolizumab-treated patients for the Gide cohort) and additionally excluded infiltration-free samples from the TCGA cohort (zero CD8\u003csup\u003e+\u003c/sup\u003e T cell content as per quanTiseq estimates). In all cases, we separated \u003cem\u003eLGALS1\u003c/em\u003e, \u003cem\u003ePTPRC\u003c/em\u003e and CD8\u003csup\u003e+\u003c/sup\u003e T cell content into high and low categories based on the group median. For the TCGA cohort, we additionally split the CD8\u003csup\u003e+\u003c/sup\u003e T cell subgroups first and then calculated LGALS1 and PTPRC high/low labels based on the subgroup medians again. For the Gide cohort, this split was attempted but led to the subgroups of interest not including any events due to the low total sample size.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eConflicts of interest:\u003c/strong\u003e M. Kunz received honoraria from the Speaker Bureau of Roche Pharma and travel support from Novartis Pharma GmbH and Bristol-Myers Squibb GmbH. J.C. Simon obtained speaker fees from Bristol-Myers Squibb, Roche Pharma AG, Novartis, and MSD Sharp \u0026amp; Dome, as well as financial support for congress attendance from Bristol-Myers Squibb, MSD Sharp \u0026amp; Dome, and Novartis. M. Ziemer received lecture fees from Bristol-Myers Squibb, MSD Sharp \u0026amp; Dohme GmbH, Pfizer Pharma GmbH, and Sanofi-Aventis Deutschland GmbH, received financial support for congress participation from Bristol-Myers Squibb, and serves as a member of expert panels on cutaneous adverse reactions for Pfizer INC. K. Reiche received honoraria from Novartis Pharma GmbH.\u003c/p\u003e\u003ch2\u003eData and Code Availability\u003c/h2\u003e\n\u003cp\u003eThe datasets generated and/or analyzed during the current study are available in the NCBI GEO (https://www.ncbi.nlm.nih.gov/geo/) repository under accession number \u0026nbsp;GSE314509[1] (Visium HD datasets; access token: obktegagrtmlrax) and GSE314538[2] (Visium datasets; access token: ejwfqeimzholzgx).The code used to analyze the data and generate the figures included in this study is available at \u0026nbsp;https://github.com/fraunhofer-izi/Grosse_et_al_2026.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eAuthor Contributions\u003c/h2\u003e\n\u003cp\u003eF. Gro\u0026szlig;e: Conceptualization, data curation, formal analysis, investigation, methodology, software, visualization, and writing of the original draft\u003c/p\u003e\n\u003cp\u003eC. K\u0026auml;mpf: Data curation, formal analysis, investigation, methodology, software, visualization, and writing of the original draft\u003c/p\u003e\n\u003cp\u003eD. L\u0026ouml;ffler: Investigation, methodology, validation and visualization\u003c/p\u003e\n\u003cp\u003eM. Lingner Chango: Data curation, investigation, validation, visualization\u003c/p\u003e\n\u003cp\u003eC. Blumert: Data curation, formal analysis, funding acquisition, methodology\u003c/p\u003e\n\u003cp\u003eA. Scholz: Data curation, formal analysis\u003c/p\u003e\n\u003cp\u003eA. Stubenvoll: Data curation, investigation, validation, visualization\u003c/p\u003e\n\u003cp\u003eH. L\u0026ouml;ffler-Wirth: Data curation, formal analysis, software, supervision\u003c/p\u003e\n\u003cp\u003eH. Binder: Conceptualization, formal analysis, software, supervision\u003c/p\u003e\n\u003cp\u003eC. Schultz: Data curation, investigation, validation, visualization\u003c/p\u003e\n\u003cp\u003eM. Beining: Data curation, investigation, modeling, visualization\u003c/p\u003e\n\u003cp\u003eJ. C Simon: Resources, supervision, validation, writing \u0026ndash; review \u0026amp; editing\u003c/p\u003e\n\u003cp\u003eC. T. Schoeder: Data curation, investigation, validation, visualization\u003c/p\u003e\n\u003cp\u003eM. Ziemer: Resources, investigation, writing \u0026ndash; review \u0026amp; editing\u003c/p\u003e\n\u003cp\u003eK. Reiche: Conceptualization, data curation, funding acquisition, investigation, project, administration, supervision, writing \u0026ndash; original draft, review and editing.\u003c/p\u003e\n\u003cp\u003eM. Kunz: Conceptualization, data curation, funding acquisition, investigation, project administration, \u0026nbsp;resources, supervision, writing \u0026ndash; original draft, writing \u0026ndash; review, and editing. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements:\u003c/strong\u003e This work was supported by the S\u0026auml;chsische Aufbaubank, grant number 100714507, to M. Kunz, and grant number 100714512, to K. Reiche and C. Blumert\u003cstrong\u003e.\u0026nbsp;\u003c/strong\u003eFurther, the authors acknowledge the financial support by the Federal Ministry of Research, Technology and Space of Germany and by S\u0026auml;chsische Staatsministerium f\u0026uuml;r Wissenschaft, Kultur und Tourismus in the program Center of Excellence for AI-research \u0026bdquo;Center for Scalable Data Analytics and Artificial Intelligence Dresden/Leipzig\u0026ldquo;, project identification number: ScaDS.AI. The results of the survival analyses are based in parts upon data generated by the TCGA Research Network (https://www.cancer.gov/tcga).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e[1] https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE314509\u003c/p\u003e\n\u003cp\u003e[2] https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE314538\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003e\u003cstrong\u003eAshburner M\u003c/strong\u003e, Ball CA, Blake JA, et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet. 2000;25(1):25-29. doi:10.1038/75556\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eBecht E\u003c/strong\u003e, McInnes L, Healy J, Dutertre CA, Kwok IWH, Ng LG, Ginhoux F, Newell EW. Dimensionality reduction for visualizing single-cell data using UMAP. Nat Biotechnol. 2018. doi: 10.1038/nbt.4314.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eBenjamini Y, Hochberg Y\u003c/strong\u003e. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. \u003cem\u003eJ R Statist Soc\u0026nbsp;\u003c/em\u003e1995 (57), 289-300.\u0026nbsp;http://www.jstor.org/stable/2346101\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eBhuva DD\u003c/strong\u003e, Tan CW, Salim A, Marceaux C, Pickering MA, Chen J, Kharbanda M, Jin X, Liu N, Feher K, Putri G, Tilley WD, Hickey TE, Asselin-Labat ML, Phipson B, Davis MJ. \u0026nbsp;Library size confounds biology in spatial transcriptomics data. Genome Biol 2024; 25(1), 99\u0026nbsp;https://doi.org/10.1186/s13059-024-03241-7.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eBlondel VD\u003c/strong\u003e, Guillaume, JL, Lambiotte, R,Lefebvre, E. Fast unfolding of communities in large networks. In: \u003cem\u003eJ. Stat. Mech.\u0026nbsp;\u003c/em\u003e2008 (10), P10008. DOI: 10.1088/1742-5468/2008/10/P10008.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eBrowaeys R\u003c/strong\u003e, Saelens W, Saeys Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat Methods. 2020;17(2):159-162. doi: 10.1038/s41592-019-0667-5.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eCable DM\u003c/strong\u003e, Murray E, Zou LS, Goeva A, Macosko EZ, Chen F, Irizarry RA. Robust decomposition of cell type mixtures in spatial transcriptomics. Nat Biotechnol. 2022 Apr;40(4):517-526. doi: 10.1038/s41587-021-00830-w.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eCancer Genome Atlas Network\u003c/strong\u003e. Genomic Classification of Cutaneous Melanoma. Cell. 2015;161(7):1681-96. doi: 10.1016/j.cell.2015.05.044\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eCedeno-Laurent F\u003c/strong\u003e, Opperman M, Barthel SR, Kuchroo VK, Dimitroff CJ. Galectin-1 triggers an immunoregulatory signature in Th cells functionally defined by IL-10 expression. J Immunol 2012; 188:3127-3137. doi: 10.4049/jimmunol.1103433.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eChoudhary S\u003c/strong\u003e, Satija R. Comparison and evaluation of statistical error models for scRNA-seq. Genome Biol. 2022; 23(1):27.https://doi.org/10.1186/s13059-021-02584-9\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eChu T\u003c/strong\u003e, Wang Z, Pe\u0026apos;er D, Danko CG. Cell type and gene expression deconvolution with BayesPrism enables Bayesian integrative analysis across bulk and single-cell RNA sequencing in oncology.\u0026nbsp;Nat\u0026nbsp;Cancer. 2022;3(4):505-517. doi: 10.1038/s43018-022-00356-3.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eCollado-Torres L,\u0026nbsp;\u003c/strong\u003eNellore A, Kammers K, Ellis SE, Taub MA, Hansen KD, Jaffe AE, Langmead B, Leek JT.Reproducible RNA-seq analysis using recount2. Nat Biotechnol. 2017a;35(4):319-321. doi:10.1038/nbt.3838\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eCollado-Torres L\u003c/strong\u003e, Nellore A, Jaffe AE. recount workflow: Accessing over 70,000 human RNA-seq samples with Bioconductor. F1000Res. 2017b;6:1558. doi: 10.12688/f1000research.12223.1.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003ede Azevedo RA\u003c/strong\u003e, Shoshan E, Whang S, Markel G, Jaiswal AR, Liu A, Curran MA, Travassos LR, Bar-Eli M. MIF inhibition as a strategy for overcoming resistance to immune checkpoint blockade therapy in melanoma. Oncoimmunology. 2020;9(1):1846915. doi: 10.1080/2162402X.2020.1846915\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eDimitrov\u003c/strong\u003e D, T\u0026uuml;rei D, Garrido-Rodriguez M, Burmedi PL, Nagai JS, Boys C, Ramirez Flores RO, Kim H, Szalai B, Costa IG, Valdeolivas A, Dugourd A, Saez-Rodriguez J. Comparison of methods and resources for cell-cell communication inference from single-cell RNA-Seq data. Nat Commun. 2022;13(1):3224. doi:10.1038/s41467-022-30755-0\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eDing Q\u003c/strong\u003e, Lu P, Xia Y, Ding S, Fan Y, Li X, Han P, Liu J, Tian D, Liu M. CXCL9: evidence and contradictions for its role in tumor progression. Cancer Med. 2016;5(11):3246-3259. doi: 10.1002/cam4.934.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eFemel J\u003c/strong\u003e, van Hooren L, Herre M, Cedervall J, Saupe F, Huijbers EJM, Verboogen DRJ, Reichel M, Thijssen VL, Griffioen AW, Hellman L, Dimberg A, Olsson AK. Vaccination against galectin-1 promotes cytotoxic T-cell infiltration in melanoma and reduces tumor burden. Cancer Immunol Immunother. 2022;71(8):2029-2040. doi: 10.1007/s00262-021-03139-4.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eFigueiredo CR,\u003c/strong\u003e Azevedo RA, Mousdell S, Resende-Lara PT, Ireland L, Santos A, Girola N, Cunha RLOR, Schmid MC, Polonelli L, Travassos LR, Mielgo A. Blockade of MIF-CD74 on Macrophages and Dendritic Cells Restores the Antitumour Immune Response Against Metastatic Melanoma. Front Immunol. 2018;9:1132. doi: 10.3389/fimmu.2018.01132.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eFinotello F\u003c/strong\u003e, Mayer C, Plattner C, Laschober G, Rieder D, Hackl H, Krogsdam A, Loncova Z, Posch W, Wilflingseder D, Sopper S, Ijsselsteijn M, Brouwer TP, Johnson D, Xu Y, Wang Y, Sanders ME, Estrada MV, Ericsson-Gonzalez P, Charoentong P, Balko J, de Miranda NFDCC, Trajanoski Z. Molecular and pharmacological modulators of the tumor immune contexture revealed by deconvolution of RNA-seq data. Genome Med. 2019;11(1):34. doi: 10.1186/s13073-019-0638-6.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eFrangieh\u003c/strong\u003e CJ, Melms JC, Thakore PI, Geiger-Schuller KR, Ho P, Luoma AM, Cleary B, Jerby-Arnon L, Malu S, Cuoco MS, Zhao M, Ager CR, Rogava M, Hovey L, Rotem A, Bernatchez C, Wucherpfennig KW, Johnson BE, Rozenblatt-Rosen O, Schadendorf D, Regev A, Izar B. Multimodal pooled Perturb-CITE-seq screens in patient models define mechanisms of cancer immune evasion. Nat Genet. 2021;53(3):332-341. doi: 10.1038/s41588-021-00779-1.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eFranz\u0026eacute;n O\u003c/strong\u003e, Gan LM, Bj\u0026ouml;rkegren JLM. PanglaoDB: a web server for exploration of mouse and human single-cell RNA sequencing data. Database (Oxford). 2019;2019:baz046. doi:10.1093/database/baz046\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eFu Q,\u0026nbsp;\u003c/strong\u003eChen N, Ge C, Li R, Li Z, Zeng B, Li C, Wang Y, Xue Y, Song X, Li H, Li G. Prognostic value of tumor-infiltrating lymphocytes in melanoma: a systematic review and meta-analysis. Oncoimmunology. 2019 Apr 3;8(7):1593806. doi: 10.1080/2162402X.2019.1593806\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eGide TN\u003c/strong\u003e, Quek C, Menzies AM, Tasker AT, Shang P, Holst J, Madore J, Lim SY, Velickovic R, Wongchenko M, Yan Y, Lo S, Carlino MS, Guminski A, Saw RPM, Pang A, McGuire HM, Palendira U, Thompson JF, Rizos H, Silva IPD, Batten M, Scolyer RA, Long GV, Wilmott JS. Distinct Immune Cell Populations Define Response to Anti-PD-1 Monotherapy and Anti-PD-1/Anti-CTLA-4 Combined Therapy. Cancer Cell. 2019;35(2):238-255.e6. doi: 10.1016/j.ccell.2019.01.003.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eGilson RC\u003c/strong\u003e, Gunasinghe SD, Johannes L, Gaus K. Galectin-3 modulation of T-cell activation: mechanisms of membrane remodelling. Prog Lipid Res. 2019;76:101010. doi: 10.1016/j.plipres.2019.101010.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eGowhari Shabgah A\u003c/strong\u003e, Al-Obaidi ZMJ, Sulaiman Rahman H, Kamal Abdelbasset W, Suksatan W, Bokov DO, Thangavelu L, Turki Jalil A, Jadidi-Niaragh F, Mohammadi H, Mashayekhi K, Gholizadeh Navashenaq J. Does CCL19 act as a double-edged sword in cancer development? Clin Exp Immunol. 2022;207(2):164-175. doi: 10.1093/cei/uxab039. PMID: 35020885; PMCID: PMC8982982.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eGu Z\u003c/strong\u003e, Gu L, Eils R, Schlesner M, Brors B. circlize Implements and enhances circular visualization in R. Bioinformatics. 2014;30(19):2811-2. doi: 10.1093/bioinformatics/btu393. Epub 2014 Jun 14.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eHao Y\u003c/strong\u003e, Hao S, Andersen-Nissen E, Mauck WM 3rd, Zheng S, Butler A, Lee MJ, Wilk AJ, Darby C, Zager M, Hoffman P, Stoeckius M, Papalexi E, Mimitou EP, Jain J, Srivastava A, Stuart T, Fleming LM, Yeung B, Rogers AJ, McElrath JM, Blish CA, Gottardo R, Smibert P, Satija R.Integrated analysis of multimodal single-cell data. Cell. 2021;184(13):3573-3587.e29. doi:10.1016/j.cell.2021.04.048\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eHao Y\u003c/strong\u003e, Stuart T, Kowalski MH, Choudhary S, Hoffman P, Hartman A, Srivastava A, Molla G, Madad S, Fernandez-Granda C, Satija R.Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat Biotechnol 2024;42; 293\u0026ndash;304.https://doi.org/10.1038/s41587-023-01767-y\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eHunter MV,\u0026nbsp;\u003c/strong\u003eMoncada R, Weiss JM, Yanai I, White RM. Spatially resolved transcriptomics reveals the architecture of the tumor-microenvironment interface. Nat Commun. 2021;12(1):6278. doi: 10.1038/s41467-021-26614-z\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eIlarregui JM\u003c/strong\u003e, Croci DO, Bianco GA, Toscano MA, Salatino M, Vermeulen ME, Geffner JR, Rabinovich GA. Tolerogenic signals delivered by dendritic cells to T cells through a galectin-1-driven immunoregulatory circuit involving interleukin 27 and interleukin 10. Nat Immunol. 2009;10(9):981-91. doi: 10.1038/ni.1772.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eJacquelot N\u003c/strong\u003e, Pitt JM, Enot DP, Roberti MP, Duong CPM, Rusakiewicz S, Eggermont AM, Zitvogel L.\u003cbr\u003e\u0026nbsp;Immune biomarkers for prognosis and prediction of responses to immune checkpoint blockade in cutaneous melanoma. Oncoimmunology. 2017 Mar 7;6(8):e1299303. doi: 10.1080/2162402X.2017.1299303.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eJacquelot N\u003c/strong\u003e, Duong CPM, Belz GT, Zitvogel L. Targeting Chemokines and Chemokine Receptors in Melanoma and Other Cancers. Front Immunol. 2018;9:2480. doi: 10.3389/fimmu.2018.02480.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eJerby-Arnon L\u003c/strong\u003e, Shah P, Cuoco MS, Rodman C, Su MJ, Melms JC, Leeson R, Kanodia A, Mei S, Lin JR, Wang S, Rabasha B, Liu D, Zhang G, Margolais C, Ashenberg O, Ott PA, Buchbinder EI, Haq R, Hodi FS, Boland GM, Sullivan RJ, Frederick DT, Miao B, Moll T, Flaherty KT, Herlyn M, Jenkins RW, Thummalapalli R, Kowalczyk MS, Ca\u0026ntilde;adas I, Schilling B, Cartwright ANR, Luoma AM, Malu S, Hwu P, Bernatchez C, Forget MA, Barbie DA, Shalek AK, Tirosh I, Sorger PK, Wucherpfennig K, Van Allen EM, Schadendorf D, Johnson BE, Rotem A, Rozenblatt-Rosen O, Garraway LA, Yoon CH, Izar B, Regev A.A Cancer Cell Program Promotes T Cell Exclusion and Resistance to Checkpoint Blockade. Cell. 2018;175(4):984-997.e24. doi:10.1016/j.cell.2018.09.006\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eKarras P\u003c/strong\u003e, Bordeu I, Pozniak J, Nowosad A, Pazzi C, Van Raemdonck N, Landeloos E, Van Herck Y, Pedri D, Bervoets G, Makhzami S, Khoo JH, Pavie B, Lamote J, Marin-Bejar O, Dewaele M, Liang H, Zhang X, Hua Y, Wouters J, Browaeys R, Bergers G, Saeys Y, Bosisio F, van den Oord J, Lambrechts D, Rustgi AK, Bechter O, Blanpain C, Simons BD, Rambow F, Marine JC. A cellular hierarchy in melanoma uncouples growth and metastasis. Nature. 2022;610(7930):190-198. doi: 10.1038/s41586-022-05242-7.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eLaidlaw BJ\u003c/strong\u003e, Gray EE, Zhang Y, Ram\u0026iacute;rez-Valle F, Cyster JG. Sphingosine-1-phosphate receptor 2 restrains egress of \u0026gamma;\u0026delta; T cells from the skin. J Exp Med. 2019;216(7):1487-1496. doi: 10.1084/jem.20190114\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eLarsson L\u003c/strong\u003e, Fris\u0026eacute;n J, Lundeberg J. Spatially resolved transcriptomics adds a new dimension to genomics. Nat Methods. 2021;18(1):15-18.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eLi H\u003c/strong\u003e, van der Leun AM, Yofe I, Lubling Y, Gelbard-Solodkin D, van Akkooi ACJ, van den Braber M, Rozeman EA, Haanen JBAG, Blank CU, Horlings HM, David E, Baran Y, Bercovich A, Lifshitz A, Schumacher TN, Tanay A, Amit I.Dysfunctional CD8 T Cells Form a Proliferative, Dynamically Regulated Compartment within Human Melanoma. Cell. 2020;181(3):747. doi: 10.1016/j.cell.2020.04.017.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eLi Pomi F\u003c/strong\u003e, Borgia F, Custurone P, Vaccaro M, Pioggia G, Gangemi S. Role of HMGB1 in Cutaneous Melanoma: State of the Art. Int J Mol Sci. 2022;23(16):9327. doi: 10.3390/ijms23169327.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eLi X\u003c/strong\u003e, Yue Z, Wang D, Zhou L. PTPRC functions as a prognosis biomarker in the tumor microenvironment of cutaneous melanoma. Sci Rep. 2023;13(1):20617. doi: 10.1038/s41598-023-46794-6.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eLim B\u003c/strong\u003e, Lin Y, Navin N. Advancing Cancer Research and Medicine with Single-Cell Genomics. Cancer Cell. 2020 Apr 13;37(4):456-470. doi: 10.1016/j.ccell.2020.03.008.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eLim SY\u003c/strong\u003e, Shklovskaya E, Lee JH, Pedersen B, Stewart A, Ming Z, Irvine M, Shivalingam B, Saw RPM, Menzies AM, Carlino MS, Scolyer RA, Long GV, Rizos H. The molecular and functional landscape of resistance to immune checkpoint blockade in melanoma. Nat Commun. 2023;14(1):1516. doi: 10.1038/s41467-023-36979-y.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eLudin A\u003c/strong\u003e, Stirtz GL, Tal A, Nirmal AJ, Pfaff KL, Manos M, Besson N, Eskndir N, Porter B, Jones SM, Faulkner HM, Gong Q, Liu S, Barrera I, Wu L, Pessoa Rodrigues C, Sahu A, Jerison E, Alessi JV, Ricciuti B, Richardson DS, Weiss JD, Moreau HM, Stanhope ME, Afeyan AB, Sefton J, McCall WD, Formato E, Yang S, Zhou Y, Hoytema van Konijnenburg DP, Cole HL, Cordova M, Deng L, Rajadhyaksha M, Quake SR, Awad MM, Chen F, Wucherpfennig KW, Sorger PK, Hodi FS, Rodig SJ, Murphy GF, Zon LI. CRATER tumor niches facilitate CD8+ T cell engagement and correspond with immunotherapy success. Cell. 2025;188(24):6720-6736.e26. doi: 10.1016/j.cell.2025.09.021.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eMa C,\u0026nbsp;\u003c/strong\u003eChen J, Ji J, Zheng Y, Liu Y, Wang J, Chen T, Chen H, Chen Z, Zhou Q, Hou C, Ke Y. Therapeutic modulation of APP-CD74 axis can activate phagocytosis of TAMs in GBM. Biochim Biophys Acta Mol Basis Dis. 2024;1870(8):167449. doi: 10.1016/j.bbadis.2024.167449.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eMa Y\u003c/strong\u003e, Zhou X. Spatially informed cell-type deconvolution for spatial transcriptomics. Nat Biotechnol. 2022 Sep;40(9):1349-1359. doi: 10.1038/s41587-022-01273-7.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eMages S\u003c/strong\u003e, Moriel N, Avraham-Davidi I, Murray E, Watter J, Chen F, Rozenblatt-Rosen O, Klughammer J, Regev A, Nitzan M.TACCO unifies annotation transfer and decomposition of cell identities for single-cell and spatial omics. Nat Biotechnol. 2023 Oct;41(10):1465-1473. doi: 10.1038/s41587-023-01657-3.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eMochida Y\u003c/strong\u003e, Ohuchida K, Zhang B, Yamada Y, Tsutsumi C, Kubo A, Oyama K, Shinkawa T, Iwamoto C, Torata N, Abe T, Ideno N, Ikenaga N, Nakata K, Oda Y, Nakamura M. Identification of cold tumor induction-related markers in pancreatic cancer and the clinical implication of PCDH7. J Cancer Res Clin Oncol. 2025;151(2):45. doi: 10.1007/s00432-025-06095-z.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eNov\u0026aacute;k J\u003c/strong\u003e, Tak\u0026aacute;cs T, Tilajka \u0026Aacute;, L\u0026aacute;szl\u0026oacute; L, Oravecz O, Farkas E, Than NG, Buday L, Balogh A, Vas V. The sweet and the bitter sides of galectin-1 in immunity: its role in immune cell functions, apoptosis, and immunotherapies for cancer with a focus on T cells. Semin Immunopathol. 2025;47(1):24. doi: 10.1007/s00281-025-01047-8\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eOstendorf BN\u003c/strong\u003e, Bilanovic J, Adaku N, Tafreshian KN, Tavora B, Vaughan RD, Tavazoie SF. Common germline variants of the human APOE gene modulate melanoma progression and survival. Nat Med. 2020; 26(7):1048-1053. doi: 10.1038/s41591-020-0879-3.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003ePassaro S\u003c/strong\u003e, Corso G, \u0026nbsp;Wohlwend \u0026nbsp;J, Reveiz M, \u0026nbsp; Stephan Thaler S, Somnath VR, Getz N, \u0026nbsp; Portnoi T, Roy J, \u0026nbsp;Stark H, Kwabi-Addo D, \u0026nbsp;Beaini D, Jaakkola T, Barzilay R. Boltz-2: Towards Accurate and Efficient Binding Affinity Prediction. bioRxiv [Preprint]. 2025 Jun 18:2025.06.14.659707. doi: 10.1101/2025.06.14.659707.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003ePolański K\u003c/strong\u003e, Bartolom\u0026eacute;-Casado R, Sarropoulos I, Xu C, England N, Jahnsen FL, Teichmann SA, Yayon N. Bioinformatics. 2024 Sep 2;40(9):btae546. doi: 10.1093/bioinformatics/btae546 \u003cstrong\u003ePozniak J\u003c/strong\u003e, Pedri D, Landeloos E, Van Herck Y, Antoranz A, Vanwynsberghe L, Nowosad A, Roda N, Makhzami S, Bervoets G, Maciel LF, Pulido-Vicu\u0026ntilde;a CA, Pollaris L, Seurinck R, Zhao F, Flem-Karlsen K, Damsky W, Chen L, Karagianni D, Cinque S, Kint S, Vandereyken K, Rombaut B, Voet T, Vernaillen F, Annaert W, Lambrechts D, Boecxstaens V, Saeys Y, van den Oord J, Bosisio F, Karras P, Shain AH, Bosenberg M, Leucci E, Paschen A, Rambow F, Bechter O, Marine JC. A TCF4-dependent gene regulatory network confers resistance to immunotherapy in melanoma. Cell. 2024;187(1):166-183.e25. doi: 10.1016/j.cell.2023.11.037\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eR Core Team\u003c/strong\u003e. R: A language and environment for statistical computing. R Foundation for Statistical Computing 2021, Vienna, Austria. URL: https://www.R-project.org/.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eRabinovich GA\u003c/strong\u003e, Ramhorst RE, Rubinstein N, Corigliano A, Daroqui MC, Kier-Joff\u0026eacute; EB, Fainboim L. Induction of allogenic T-cell hyporesponsiveness by galectin-1-mediated apoptotic and non-apoptotic mechanisms. Cell Death Differ. 2002;9(6):661-70. doi: 10.1038/sj.cdd.4401009\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eRambow F\u003c/strong\u003e, Rogiers A, Marin-Bejar O, Aibar S, Femel J, Dewaele M, Karras P, Brown D, Chang YH, Debiec-Rychter M, Adriaens C, Radaelli E, Wolter P, Bechter O, Dummer R, Levesque M, Piris A, Frederick DT, Boland G, Flaherty KT, van den Oord J, Voet T, Aerts S, Lund AW, Marine JC.Toward Minimal Residual Disease-Directed Therapy in Melanoma. Cell. 2018; 174(4):843-855.e19, doi: 10.1016/j.cell.2018.06.025.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eRaredon MSB\u003c/strong\u003e, Yang J, Kothapalli N, Lewis W, Kaminski N, Niklason LE, Kluger Y.Comprehensive visualization of cell-cell interactions in single-cell and spatial transcriptomics with NICHES. Bioinformatics. 2023;39(1):btac775. doi:10.1093/bioinformatics/btac775\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eSaad M\u003c/strong\u003e, Castellano E, Tarhini AA. Clinical updates in neoadjuvant immunotherapy for melanoma before surgery. Expert Rev Clin Immunol. 2024 Aug;20(8):927-943 doi: 10.1080/1744666X.2023.2248392.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eSade-Feldman M\u003c/strong\u003e, Yizhak K, Bjorgaard SL, Ray JP, de Boer CG, Jenkins RW, Lieb DJ, Chen JH, Frederick DT, Barzily-Rokni M, Freeman SS, Reuben A, Hoover PJ, Villani AC, Ivanova E, Portell A, Lizotte PH, Aref AR, Eliane JP, Hammond MR, Vitzthum H, Blackmon SM, Li B, Gopalakrishnan V, Reddy SM, Cooper ZA, Paweletz CP, Barbie DA, Stemmer-Rachamimov A, Flaherty KT, Wargo JA, Boland GM, Sullivan RJ, Getz G, Hacohen N.Defining T Cell States Associated with Response to Checkpoint Immunotherapy in Melanoma. Cell. 2019 Jan 10;176(1-2):404. doi: 10.1016/j.cell.2018.12.034.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eSaiselet M\u003c/strong\u003e, Rodrigues-Vit\u0026oacute;ria J, Tourneur A, Craciun L, Spinette A, Larsimont D, Andry G, Lundeberg J, Maenhaut C, Detours V.Transcriptional output, cell-type densities, and normalization in spatial transcriptomics. J Mol Cell Biol. 2020;12(11):906-908. doi:10.1093/jmcb/mjaa028\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eSchadendorf D\u003c/strong\u003e, van Akkooi ACJ, Berking C, Griewank KG, Gutzmer R, Hauschild A, Stang A, Roesch A, Ugurel S. Melanoma. Lancet 2018; 392(10151):971-984. doi: 10.1016/S0140-6736(18)31559-9.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eSeferbekova Z\u003c/strong\u003e, Lomakin A, Yates LR, Gerstung M. Spatial biology of cancer evolution. Nat Rev Genet. 2023 May;24(5):295-313. doi: 10.1038/s41576-022-00553-x\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eStubenvoll A\u003c/strong\u003e, Schmidt M, Moeller J, Chango MAL, Schultz C, Antoniadou O, Loeffler-Wirth H, Bernhart S, Gro\u0026szlig;e F, Thier B, Paschen A, Anderegg U, Simon JC, Ziemer M, Schoeder CT, Binder H, Kunz M. Single-cell transcriptomics and epigenomics point to CD58-CD2 interaction in controlling primary melanoma growth and immunity. Cancer Commun (Lond). 2025;45(4):465-470. doi: 10.1002/cac2.12651\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eSol\u0026eacute;-Boldo L\u003c/strong\u003e, Raddatz G, Sch\u0026uuml;tz S, Mallm JP, Rippe K, Lonsdorf AS, Rodr\u0026iacute;guez-Paredes M, Lyko F. Single-cell transcriptomes of the human skin reveal age-related loss of fibroblast priming. Commun Biol. 2020;3(1):188. doi: 10.1038/s42003-020-0922-4.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eSubramanian A\u003c/strong\u003e, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP.Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005;102(43):15545-50. doi: 10.1073/pnas.0506580102.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eTawbi HA\u003c/strong\u003e, Schadendorf D, Lipson EJ, Ascierto PA, Matamala L, Castillo Guti\u0026eacute;rrez E, Rutkowski P, Gogas HJ, Lao CD, De Menezes JJ, Dalle S, Arance A, Grob JJ, Srivastava S, Abaskharoun M, Hamilton M, Keidel S, Simonsen KL, Sobiesk AM, Li B, Hodi FS, Long GV; RELATIVITY-047 Investigators. Relatlimab and Nivolumab versus Nivolumab in Untreated Advanced Melanoma. N Engl J Med. 2022;386(1):24-34. doi: 10.1056/NEJMoa2109970.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eThrane K,\u0026nbsp;\u003c/strong\u003eEriksson H, Maaskola J, Hansson J, Lundeberg J. Spatially Resolved Transcriptomics Enables Dissection of Genetic Heterogeneity in Stage III Cutaneous Malignant Melanoma. Cancer Res. 2018;78(20):5970-5979. doi: 10.1158/0008-5472.CAN-18-0747.\u003cstrong\u003eTirosh I\u003c/strong\u003e, Izar B, Prakadan SM, Wadsworth MH 2nd, Treacy D, Trombetta JJ, Rotem A, Rodman C, Lian C, Murphy G, Fallahi-Sichani M, Dutton-Regester K, Lin JR, Cohen O, Shah P, Lu D, Genshaft AS, Hughes TK, Ziegler CG, Kazer SW, Gaillard A, Kolb KE, Villani AC, Johannessen CM, Andreev AY, Van Allen EM, Bertagnolli M, Sorger PK, Sullivan RJ, Flaherty KT, Frederick DT, Jan\u0026eacute;-Valbuena J, Yoon CH, Rozenblatt-Rosen O, Shalek AK, Regev A, Garraway LA.Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science 2016; 352(6282):189-96. doi: 10.1126/science.aad0501.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eTsoi J\u003c/strong\u003e, Robert L, Paraiso K, Galvan C, Sheu KM, Lay J, Wong DJL, Atefi M, Shirazi R, Wang X, Braas D, Grasso CS, Palaskas N, Ribas A, Graeber TG. Multi-stage Differentiation Defines Melanoma Subtypes with Differential Vulnerability to Drug-Induced Iron-Dependent Oxidative Stress. Cancer Cell. 2018 May 14;33(5):890-904.e5. doi: 10.1016/j.ccell.2018.03.017.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eVan der Walt S\u003c/strong\u003e, Sch\u0026ouml;nberger JL, Nunez-Iglesias J, Boulogne F, Warner JD, Yager N, Gouillart E, Tony Yu, the scikit-image contributors. scikit-image: image processing in Python.\u0026nbsp; PeerJ 2014 (6);\u0026nbsp;e453, doi.org/10.7717/peerj.453\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eVirshup I\u003c/strong\u003e, Rybakov S, Theis Fabian J, Angerer P, Wolf FA. anndata: Access and store annotated data matrices. JOSS 2024;9 (101), 4371. DOI: 10.21105/joss.04371.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eWang HC\u003c/strong\u003e, Xia R, Chang WH, Hsu SW, Wu CT, Chen CH, Shih TC. Improving cancer immunotherapy in prostate cancer by modulating T cell function through targeting the galectin-1. Front Immunol. 2024;15:1372956. doi: 10.3389/fimmu.2024.1372956.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eWeigert M,\u0026nbsp;\u003c/strong\u003eSchmidt U. Nuclei Instance Segmentation and Classification in Histopathology Images with Stardist. 2022 IEEE International Symposium on Biomedical Imaging Challenges (ISBIC). Kolkata, India, 28.03.2022 - 31.03.2022: IEEE, S. 1\u0026ndash;4. DOI: 10.1109/ISBIC56247.2022.9854534\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eWu T\u003c/strong\u003e, Hu E, Xu S, Chen M, Guo P, Dai Z, Feng T, Zhou L, Tang W, Zhan L, Fu X, Liu S, Bo X, Yu G.clusterProfiler 4.0: A universal enrichment tool for interpreting omics data. Innovation (Camb). 2021 Jul 1;2(3):100141. doi: 10.1016/j.xinn.2021.100141\u003cstrong\u003eZappia L\u003c/strong\u003e, Oshlack A. Clustering trees: a visualization for evaluating clusterings at multiple resolutions. Gigascience. 2018;7(7):giy083. doi:10.1093/gigascience/giy083\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eZhang J\u003c/strong\u003e, Humphreys IR, Pei J, Kim J, Choi C, Yuan R, Durham J, Liu S, Choi HJ, Baek M, Baker D, Cong Q. Predicting protein-protein interactions in the human proteome. Science. 2025;390(6771):eadt1630. doi: 10.1126/science.adt1630.\u003cstrong\u003eZappia L\u003c/strong\u003e, Oshlack A. Clustering trees: a visualization for evaluating clusterings at multiple resolutions. 1357 Gigascience. 2018;7(7):giy083. doi:10.1093/gigascience/giy083\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eZubieta MR\u003c/strong\u003e, Furman D, Barrio M, Bravo AI, Domenichini E, Mordoh J. Galectin-3 expression correlates with apoptosis of tumor-associated lymphocytes in human melanoma biopsies. Am J Pathol. 2006;168(5):1666-75. doi: 10.2353/ajpath.2006.050971.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Sächsische Aufbaubank","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"skin cancer, melanoma, spatial sequencing, ligand-receptor interactions, immunotherapy","lastPublishedDoi":"10.21203/rs.3.rs-9075388/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9075388/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMelanoma is a highly immunogenic tumor where interactions between tumor and immune cells critically influence disease progression and treatment response. Traditional single-cell transcriptomics lack spatial context, limiting the understanding of these interactions within the tumor microenvironment. Here we integrated 10x Genomics Visium and Visium HD spatial transcriptomics with multiple single-cell RNA-seq datasets to map ligand-receptor interactions at the tumor-immune border region. This approach identified immune cell interactions already known in melanoma and highlighted unknown but prominent \u003cem\u003eLGALS1\u003c/em\u003e–\u003cem\u003ePTPRC \u003c/em\u003e(proteins: galectin-1–CD45) interactions localized at the tumor-immune border region. Immunofluorescence and biolayer interferometry confirmed galectin-1 expression on melanoma cells and its binding to CD45 on infiltrating T cells. High \u003cem\u003eLGALS1 \u003c/em\u003eexpression correlated with reduced overall survival in patients undergoing checkpoint inhibitor therapy. These findings suggest that galectin-1–CD45\u003cem\u003e \u003c/em\u003einteractions contribute to immune modulation in melanoma and represent potential targets for immunotherapeutic strategies.\u003c/p\u003e","manuscriptTitle":"Decoding melanoma-immune cell communication through spatially resolved ligand-receptor interaction analyses","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-11 08:39:08","doi":"10.21203/rs.3.rs-9075388/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"5687b432-c94e-4405-95e0-8d6650d72b24","owner":[],"postedDate":"March 11th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":64230972,"name":"Dermatology"}],"tags":[],"updatedAt":"2026-03-11T08:39:11+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-11 08:39:08","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9075388","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9075388","identity":"rs-9075388","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}