Nature | www.nature.com | 3 understanding of immunoregulatory mechanisms in TDLNs and how these might affect patient outcomes is lacking. We therefore performed comprehensive spatial proteogenomics studies on TDLNs from patients with melanoma, aiming to identify targetable regulators of systemic anti-tumour immunity. We found that TDLN contexture correlates with patient outcomes after surgery. Specifically, we identified lymph node (LN) myeloid cells—predominantly macrophages—that express the secreted phospholipase A2 family member PLA2G2D to be associ- ated with a poor prognosis. PLA2G2D acts as a LN-centred immune checkpoint that curtails anti-tumour T cell proliferation, and genetic or antibody-mediated inhibition of PLA2G2D suppressed tumour growth and synergized with anti-PD-1 ICI in preclinical models. Thus, PLA2G2D acts as a bona fide immune checkpoint in TDLNs and represents a new target for cancer immunotherapy. TDLN composition predicts poor prognosis We collected TDLNs from patients with resected stage II or III melanoma (non-metastatic or metastatic LNs, respectively) who did not receive adjuvant or neoadjuvant therapies and whose clinical outcomes were known. Patients were labelled as ‘recurrence’ (R; stage II: recurrence- free survival (RFS) ≤ 48 months, stage III: RFS ≤ 24 months) or ‘disease- free’ (DF; stage II: RFS > 96 months, stage III: RFS > 60 months). TDLNs were localized using perioperative radiotracer injection at the site of a previous tumour resection in a sentinel node procedure. Patients with melanoma ( n = 5 or 6 per group) were matched by tumour burden and known prognostic markers such as Breslow depth (Supplementary Table 1 and Extended Data Fig. 1a). Skin-draining (axillary) control LNs (cLNs) from patients who died from acute vascular disease in the absence of cancer, infection or autoimmune disease were also included. To map the cellular composition of TDLNs across different anatomi- cal sites, we designed a 40-plex imaging mass cytometry (IMC) antibody panel capable of distinguishing different immune and stromal cell sub- sets. Regions of interest (ROIs) from cortex, paracortex, peritumoural and intratumoural areas were selected (at least 5 ROIs per location and per patient; 372 ROIs were analysed in total) and protein expression intensities for all markers were acquired at the single-cell level (Fig. 1a and Extended Data Fig. 1b). Cell segmentation identified 870,277 indi- vidual cells, which were annotated on the basis of marker expression and included B cells, T cells, dendritic cells (DCs), macrophages and stromal cells (Fig. 1b,c and Extended Data Figs. 1c–h and 2a,b). Immune contexture in melanoma TDLNs differed from that in cLNs, even in the absence of metastasis (stage II) (Extended Data Fig. 2c,d)—in line with previous studies showing TDLN remodelling in patients developing metastases 4,18–20 . Unbiased comparisons of cell abundance revealed compositional changes mainly in the stage III TDLN paracortex, involv - ing a substantial increase in HLA-DR lo and HLA-DR hi DCs in patients in the R group, as opposed to increased levels of CD163 + and CD169 + macrophages in patients in the DF group (Fig. 1d). Increased DC– macrophage ratios were strongly linked to disease recurrence, but only in metastatic TDLNs (stage III; Extended Data Fig. 2e,f). Differences in cell abundance were independent of the cell segmentation method used (Extended Data Fig. 2g,h). To investigate networks of preferential cell–cell interactions, we used nearest-neighbour analysis to define ten cellular neighbourhoods (NBs) in the TDLN paracortex (Fig. 1e and Extended Data Fig. 3a). Sev- eral NBs were expanded or reduced in R versus DF patients (Extended Data Fig. 3b), and most of these consisted mainly of a single cell type (for example, T cells in NB1 and NB3; macrophages in NB6) or two cell types (for example, neutrophils interacting with endothelial cells in NB8). NB2, however, was particularly expanded in R compared with DF patients (Fig. 1f), and was characterized by an increased presence of CD4 + , CD8 + and regulatory T (T reg ) cells as well as HLA-DR hi DCs in R patients (Fig. 1g and Extended Data Fig. 3a,c,d). Notably, the frequen- cies of CD163 + and CD169 + macrophages—although generally reduced in the paracortex of R patients—also increased in NB2. Phenotypically, T cells within NB2 in R patients showed increased levels of activation and exhaustion-associated markers such as CTLA4 and TOX, whereas macrophage and DC subsets expressed increased levels of the immu- nosuppressive molecules PD-L1 and IDO (Extended Data Fig. 3e). Nearest-neighbour and cellular proximity analyses in the paracortex showed that the formation of DC–CD8 + T cell dyads increased in R com- pared with DF patients, and that this was often accompanied by close macrophage proximity (Fig. 1h–j and Extended Data Fig. 3f). Together, these data identify a distinct myeloid–T cell interface in the TDLN para- cortex of patients with melanoma who have a poor prognosis. Myeloid cells in TDLNs produce PLA2G2D To better characterize the observed myeloid–T cell phenotypes in the TDLN paracortex, we performed a targeted spatial transcriptomics anal- ysis of CD11c + DCs, CD68 + macrophages and CD8 + T cells co-localizing in paracortex regions from R and DF patients with stage III melanoma ( n = 5 per group) (Fig. 2a). This approach identified hundreds of dif- ferentially expressed genes (DEGs; P < 0.05) between R and DF patients for all three cell types (Fig. 2b and Extended Data Figs. 4 and 5 and Sup- plementary Table 2). Gene set enrichment analysis (GSEA) revealed that DCs in R patients upregulated genes that are strongly linked to mature DCs enriched in immunoregulatory molecules (mRegDCs) marked by the expression of CCR7 , LAMP3 , IL4I1 , FSCN1 and the IL-4R signalling pathway 21 (Fig. 2b,c and Extended Data Fig. 4a–d). Macrophages in R patients showed an inflammatory phenotype characterized by a strong induction of interferon-γ (IFNγ)-responsive genes (for example, CXCL9 , IRF1 and STAT1 ) and immunosuppressive molecules (for example IDO1 , LILRB4 and CD274 (which encodes PD-L1)) alongside reduced levels of genes related to phagocytosis 22 (Fig. 2b,c and Extended Data Fig. 4d–g). Fig. 1 | Location-specific changes in TDLN immune contexture are linked with distant disease metastasis. a ,TDLNs were collected from patients with stage II (non-metastatic) and stage III (metastatic) melanoma who remained disease free (DF; RFS > 60–96 months; mo) or who developed distant disease recurrence (R; RFS < 24–48 months) after surgery. Axillary cLNs from individuals without cancer were also collected. Tissues were stained with an IMC 40-plex antibody panel and imaged across ROIs in cortex, paracortex, peritumoural and intratumoural areas. b , Annotated uniform manifold approximation and projection (UMAP) of 870,277 single cells from 372 ROIs. BEC, blood endothelial cell; LEC, lymphatic endothelial cell. c , Scaled IMC marker expression for the different cell subsets identified. Bar graphs depict total cell numbers. d , Volcano plots showing differences in cell abundance (log 2 -transformed fold change of cells per mm 2 tissue per ROI) between patients with stage III melanoma in the DF group and in the R group. e , Representative IMC images including cell segmentation mask with annotations and cellular NB classifications. Scale bars, 100 μm. f , Spatial area occupied by cells associated with NB2 in patients with stage III melanoma in group DF or group R (R = 24 ROIs, DF = 30 ROIs, from 5 or 6 patients, respectively; 4–5 ROIs per patient). g , Relative abundance (scaled per cell type across NBs) of the indicated immune-cell subsets in NB2 in patients with stage III melanoma in group DF or group R (paracortex). h , Nearest-neighbour interaction analyses of the indicated cells in stage III DF or R paracortex. Colours indicate the proportions of ROIs with statistically significant interaction (+1) or avoidance (−1), corrected for cellular abundance. i , Quantification of DC–CD8 + T cell co-localization (dyads; left) and macrophages in close proximity (<15 μ m or <30 μ m) to DC–CD8 + T cell dyads (right) in DF or R paracortex (R = 24 ROIs, DF = 30 ROIs, from 5 or 6 patients, respectively). j , Visualization of DC–CD8 + T cell dyads and adjacent macrophages, showing annotated cells (top left), dyad distance (top right) and marker quantification of CD8a (T cells; red), CD11c (DCs; green) and CD68 (macrophages; cyan) (bottom). Scale bars, 100 μm. P values by unpaired two-sided t -test ( d , f , i ). *** P < 0.001, **** P < 0.0001 (exact P values in Supplementary Data).