Ethics approval

This study was approved by the Erasmus MC Ethics Committee (MEC-2017-375, METC 2020-054), allowing for retrospectively collected data and residual material from tissue collected for routine diagnostic purposes to be used, waiving the need for explicit written informed consent. Human tissues and patient data were used according to the Code for Proper Secondary Use of Human Tissue and the Code of Conduct for the Use of Data in Health Research as stated by the Federation of Dutch Medical Scientific Societies. All animal studies were approved by the Institutional Animal Care and Use Committee at Mispro (South San Francisco) or the animal welfare committee (IvD) of the Erasmus MC (Rotterdam), with approval by the CCD (Centrale Commissie Dierproeven) under the permit number AVD101002017867.

Patient inclusion

From patients with melanoma who underwent a sentinel LN procedure at the Erasmus MC Cancer Institute between 2005 and 2017, individuals with or without nodal metastasis (that is, positive TDLNs) were identified. For comparative purposes, we selected patients with melanoma who either presented with early (stage II, RFS ≤ 48 months; stage III, RFS ≤ 24 months) distant disease recurrence or did not develop disease recurrence (stage II, RFS > 96 months; stage III, RFS > 60 months) after the sentinel LN procedure. To avoid including false-negative TDLNs, we only included stage II patients with a negative TDLN who developed metastatic disease within the regional LNs (similar to the TDLN basin) after 9 months or longer. Patients with melanoma did not receive neoadjuvant or adjuvant treatment (which was not routinely administered during the inclusion period), excluding effects of immunotherapy on TDLN contexture or disease outcome. We also selected patients with resectable stage IB–III NSCLC who underwent surgical resection at the Erasmus MC Cancer Institute between 2018 and 2021, with either distant disease recurrence (>6 months and ≤60 months after surgery) or who did not develop disease recurrence (>60 months after surgery). Adjuvant chemotherapy for NSCLC was allowed but is known to improve overall survival only minimally after surgery⁴⁸. Patients with melanoma or with NSCLC were excluded if they had synchronous malignancies of other origin, chronic infections, active autoimmune disease or other disease aetiologies requiring immunosuppressive treatment (including prednisone or equivalent >5 mg), or if they received neoadjuvant treatment before resection. Most patients with NSCLC in both prognostic groups received adjuvant chemotherapy according to local treatment protocols.

Tissue sample processing and ROI selection

Formalin-fixed paraffin-embedded (FFPE) LN samples were retrieved from the Pathology department at the Erasmus MC (standard protocols for diagnostic work-up) and tissue sections were prepared at a thickness of 4 µm and 5 µm for IMC and IHC, respectively. ROIs (500 × 500 µm) were selected at four distinct LN anatomical locations: cortex, paracortex, peritumoural and intratumoural. ROI selection was guided by (i) IHC staining for SOX10 (tumour), CD3 (T cells) and CD20 (B cells), to prevent an exclusive focus on the highly prevalent B cell follicles; and (ii) haematoxylin and eosin (H&E), CD3, CD20 and SOX10-stained 5-µm sections from each sample. ROI selection was performed under the supervision of an experienced certified pathologist. Tissue microarrays (TMAs) were created using the fully automated TMA Grand Master (3DHISTECH) system. From the paracortex of TDLNs from patients with stage III melanoma, 2-mm-diameter cores were selected from the above-mentioned ROIs and embedded in paraffin.

Design of the IMC antibody panel

IMC antibodies were either acquired pre-labelled with metal isotopes (Standard BioTools) or labelled in-house using MaxPar conjugation kits (Standard BioTools) according to the manufacturer’s protocol. Antibodies were titrated on TDLN material containing metastases and were evaluated by an experienced pathologist to determine optimal antibody concentrations. Details of antibodies and dilutions are provided in Supplementary Table 5.

IMC tissue staining

For IMC, the antibody staining protocol was adapted from Standard BioTools. FFPE LN material was sliced at a thickness of 4 µm. Slides were baked for 2 h at 60 °C before being dewaxed in fresh xylene (Sigma-Aldrich, 534056) for 20 min. Slides were then dehydrated in descending grades of ethanol (ethanol 100%, Boom, 84901622.5000) (100%, 95%, 80%, 70%) for 5 min each and washed in sterile water for 5 min by gentle agitation. Slides were inserted in preheated antigen retrieval solution at 96 °C (Dako, S2367) and incubated for 30 min with loose lids. After incubation, slides were cooled down to 70 °C before being washed in sterile water for 10 min in 50-ml Falcon tubes by gentle agitation. Slides were then inserted in Epredia Shandon plastic cover plates (Epredia) and moved to the Sequenza Immunostaining Center System (Epredia). Slides were washed twice with phosphate-buffered saline (PBS) by filling the Sequenza cover plates and allowing the fluid to flow through the system. Next, the tissue was blocked using 3% bovine serum albumin (BSA) in PBS for 45 min at room temperature. Meanwhile, the antibody cocktail was prepared in 0.5% BSA in PBS. To each slide, 150 µl antibody cocktail was added, after which the Sequenza Immunostaining Center System was covered with a lid and incubated at 4 °C overnight. The next day, slides were washed with 0.2% Triton X-100 (Thermo Fisher Scientific, WD319895) in PBS until the fluid ran through the system. The slides were washed twice with PBS in the same manner before being stained with 150 µl Intercalator-Ir in PBS for 30 min at room temperature. Finally, the slides were washed with sterile water before being dried at room temperature. Slides were stored in a dry place until acquisition. Details of antibodies and dilutions are provide in Supplementary Table 5.

Image acquisition

For IMC, images were acquired on the Hyperion Imaging System using CyTOF software v.7.1 (Standard BioTools). Laser strength was determined using representative tissue samples. The machine was tuned before every acquisition and reference energy was calibrated once a month. All intra- and peritumoural ROIs from patient P002 were excluded from further analysis owing to abnormal staining patterns.

Cell segmentation

To generate single-cell data from IMC images, segmentation was performed in accordance with the Bodenmiller IMC segmentation pipeline⁴⁹, including the use of CellProfiler v.4.2.5⁵⁰ and Ilastik v.1.4.0⁵¹. In brief, MCD and .txt files were converted into individual tiff files using the IMC preprocessing script. Hot pixels were filtered by comparing the intensity of each pixel against the maximum intensity of the 3 × 3 neighbouring pixels; if differences were larger than a threshold of 50 (default settings), the pixel intensity was clipped to the maximum intensity in the 3 × 3 neighbourhood. Next, images were cropped in CellProfiler for use in supervised pixel classification using Ilastik, which was trained to recognize the nucleus, cytoplasm and background using selected markers (DNA, CD45, CD20, CD3, SOX10, and ICSK1–3 from the Cell segmentation kit III). We trained Ilastik on 50 × 50-µm cropped images of multiple ROIs from each patient. In brief, the average intensity across all channels is calculated and multiplied by 100 to help identify background regions. Next, the average expression was clipped to a range between 0 and 1. Ilastik stacks were upscaled by factor 2 and cropped to 50 × 50 pixels. After feature reduction, cell masks were generated, and images were segmented using CellProfiler to extract single-cell staining intensity data. This segmentation strategy was validated using Steinbock (v.0.16.3)—a toolkit for the processing of multiplex tissue images—which implements DeepCell’s Mesmer⁵² (v.0.4.1) for automated cell segmentation based on TissueNet, a dataset for training segmentation models with more than one million manually labelled cells. In brief, Steinbock was run using the Steinbock Docker container in Linux as provided by the authors. Raw .mcd files were preprocessed and filtered for hot pixels with a threshold of 50. Next, cells were segmented using Mesmer based on CD11c, CD14, CD45, CD68, CD20, CD8a, CD169 and CSK-III as cytoplasm markers and SOX10, DNAI and DNAII as nuclear markers. Single-cell data were extracted from segmented images by calculating the mean object intensities per channel, as well as spatial object properties (area, centroid, major axis length, minor axis length and eccentricity).

Processing and normalization of IMC data

Data were processed using an R-based workflow for multiplex imaging data, as described previously⁴⁹. In brief, data were transformed using an inverse hyperbolic sine function (asinh function from the rrscale v.1.0 package). The dataset then underwent median cell size calculation, with very small cells (diameter less than 10 pixels) excluded from further analysis. Shot noise was filtered mostly by performing all downstream analyses using median signal intensities per cell, with cells consisting, on average, of 80–90 pixels per µm²—minimizing the effect of rare random positive pixel artefacts. As a quantitative metric of image quality, we generated signal-to-noise ratio (SNR) values for all markers in our panel (Extended Data Fig. 1b). SNR values were calculated using a two-component Gaussian mixture model for each marker for each cell, calculated as SNR = I_s/I_n, in which I_s represents the signal intensity (mean intensity of pixels with a true signal) and I_n represents the noise intensity (mean intensity of pixels containing noise). Using the Otsu thresholding method, we identified pixels of foreground (signal) and background (noise), in which signal intensity is divided by noise intensity for each cell as described previously⁴⁹. Most hallmark markers for cell-type identification (for example, CD4, CD8, CD20, CD68 and CD11c) showed relatively strong signal intensities and SNR values (>2). Transcription factors and specific cell-surface receptors showed lower staining intensities, yet all had positive SNR values (> 1.5). ROI coverage was calculated and any ROIs with a cell coverage lower than 50% were inspected to identify potential selection errors. To ensure data integrity, batch correction was then applied to the data per ablation run, as well as Ki-67 expression, using the fastMNN method⁵³ (v.1.29).

Cell-type annotation

FastMNN-corrected data were clustered using the Rphenograph package (v.0.1.1)⁵⁴, with k = 40. The resulting clusters were then annotated using marker proteins to distinguish general lineages (B cells, CD4⁺ T cells, CD8⁺ T cells, T_reg cells, tumour, endothelial cells, myeloid cells, fibroblasts and stroma). These lineages were subclustered to remove cells that were wrongly annotated and to further stratify the lineages into more defined cell phenotypes. We refrained from using markers with relatively low signal intensity and SNR values (Extended Data Fig. 1b) for phenotyping (for example, BCL6 and PD-L1). To increase annotation accuracy, we limited definitions of key cell types to naive and non-naive T cells (based on CD3, CD4, CD8A, FOXP3, CD45RA, Ki-67, CTLA4, PD-1 and HLA-DR); DCs (CD11c^hiCD68^lo) separated on the basis of high or low HLA-DR levels; and macrophages (CD11c^loCD68^hi) separated on the basis of high levels of CD163 or CD169. The phenotyping of cells was validated by careful inspection of labels projected on cell masks in combination with marker expression on images, often resulting in several rounds of visual inspection and subsequent parameter optimization (for example, marker-based gating of individual cell phenotypes using cytomapper).

Cell–cell interaction and neighbourhood analysis

Interaction graphs were calculated using ‘buildSpatialGraph’ from the imcRtools package v.0.1.8 (ref. ⁴⁹), with the Delaunay triangulation method, resulting in a definition of the nearest neighbours for each cell in the IMC dataset (maximum distance parameter = 20). Per ROI image, we used ‘testInteractions’ to compute the average cell-type–cell-type interaction counts and compared this count against an empirical null distribution generated by permuting all cell labels, while maintaining the tissue structure. This way, every pairwise interaction was tested for statistically significant interaction (+1) or avoidance (−1) in every ROI. Sums of interact and avoid scores were visualized in a heat map as percentages of ROIs with a particular label. ‘testInteractions’ was also used to compute the average interaction count for each cell phenotype pair across the entire dataset, including label permutation to enable statistical testing. Next, we inferred cellular NBs using ‘aggregateNeighbors’ based on the Delauney interaction graph, in which cells were counted by their phenotypic label and separated by disease stage as well as by location. Data were then k-means-clustered (k = 10), from which the resulting NBs were visualized using a heat map for each patient group (stage II or III; R or DF) and location (for example, paracortex, intratumoural) separately. Cell distances were calculated using ‘minDistToCells’ from the imcRtools package (v.0.1.8)⁴⁹. Next, dyads were defined as surrounding cells within 15 μm of target cells as measured from the centroid of cells. We then calculated cell distances to cells labelled as DC–CD8 dyads and quantified macrophage abundance near these dyads (within 15 μm or 30 μm), separated by patient group.

Image visualization

Composite images of selected channels were generated using Fiji ImageJ (v.1.54t)⁵⁵. Images were processed to remove outlier pixels, filtered using a median or Gaussian filter (0.5-pixel radius) and scaled to enhance contrast as described previously⁵⁶, or visualized using the plotPixels function in the cytomapper⁵⁷ package.

GeoMx digital spatial profiling

Digital spatial profiling (DSP) using the GeoMx Whole Transcriptome Atlas (WTA) (NanoString Technologies) was performed following the manufacturer’s instructions. In brief, 5-µm-thick slices were cut from TMAs and stained for SYTO 13 (NanoString Technologies, 1:20), CD11c–Alexa Fluor 647 (Abcam, EP1347Y, 1:40), CD8a–Alexa Fluor 594 (BioLegend, C8/144B, 1:40) and CD68–Alexa Fluor 532 (Novus Biologicals, C68/684, 1:40), and incubated overnight with GeoMx WTA probes. Within the TDLN paracortex, regions in which CD11c⁺, CD68⁺ and CD8⁺ cells co-localized were selected. From these regions, WTA probes from CD11c⁺, CD8⁺ and CD68⁺ cells were collected separately and sequenced using an Illumina NextSeq2000. Paired-end clusters were generated, producing reads that were 26 bases in length. We chose to remove segments with less than 15% of all genes detected. The resulting counts were normalized using Q3 normalization and DEGs (P < 0.05) were calculated using a linear mixed model (LMM) as implemented in the GeomxTools v.3.10.0 R package (https://doi.org/10.18129/B9.bioc.GeomxTools). Pathway enrichment analyses on the DEGs were done using Metascape⁵⁸.

GSEA

We ran GSEA as implemented by the clusterProfiler⁵⁹ (v.4.3.3) R package against custom gene signatures for DCs^21,60,61,62, macrophages²² and CD8⁺ T cells^33,34. The GSEA function was implemented with default settings, using a Benjamini–Hochberg correction for multiple testing of the resulting P values.

Analyses of publicly available transcriptome data

TCGA transcriptome data for skin cutaneous melanoma (SKCM) were obtained from the UCSC Xena portal (https://tcga.xenahubs.net). To calculate the levels of PLA2GD2 in various types of melanoma sample, we used log₂(x + 1)-transformed RNA-Seq by Expectation Maximization (RSEM)-normalized counts and curated phenotype data as directly available from the UCSC Xena portal. To investigate which cytokines in the TME correlate with PLA2G2D expression, Pearson’s R coefficients and P values were determined for a range of cytokines and PLA2G2D across various tumour types included in TCGA. In the case of type I (α and β subtypes were included) and type III (λ) interferons, and in the case of TGF-β isoforms, the gene with the highest R coefficient per cytokine was plotted. Analyses were performed using the GEPIA2 TCGA data browser⁶³. Log₂-transformed fold changes in the RNA expression of genes in NSCLC biopsies before anti-PD-1 therapy (pre) and after (post) onset of acquired therapy resistance were obtained from a previous study¹⁶.

IHC and multiplex immunofluorescence

PLA2G2D was stained by automated IHC using the Ventana Benchmark ULTRA v.12.1 (Ventana Medical Systems). FFPE sections of 4-µm thickness were stained for PLA2G2D (Supplementary Table 6) using OptiView (760-700, Ventana). In brief, after deparaffinization and heat-induced antigen retrieval with CC1 (950-500, Ventana) for 32 min, the tissue samples were incubated with PLA2G2D antibody for 32 min at 37 °C. Incubation was followed by Optiview detection and haematoxylin II counterstaining for 8 min followed by a blue colouring reagent for 8 min according to the manufacturer’s instructions (Ventana, Roche). Multiplex immunofluorescence analysis was performed by automated immunofluorescence using the Ventana Benchmark Discovery ULTRA (Ventana Medical Systems). Slides of 4-µm-thick FFPE sections were stained for CD11c, CD8, CD68, LAMP3 and CXCL13 (Supplementary Table 6). In brief, after deparaffinization and heat-induced antigen retrieval with CC1 (950-224, Ventana) for 32 min, incubation with anti-CD8 was performed for 32 min at 37 °C, followed by Omnimap anti-mouse HRP (760-4310, Ventana) and detection with R6G (950-240, Ventana). An antibody denaturation step was performed with CC2 (950-123, Ventana) at 100 °C for 20 min. Second, incubation with anti-CD68 was performed for 12 min at 37 °C, followed by Omnimap anti-mouse (760-3210 Ventana) and detection with DCC (760-244, Ventana). An antibody denaturation step was performed with CC2 (950-123, Ventana) at 100 °C for 20 min. Third, incubation with anti-LAMP3 was performed for 60 min at 37 °C, followed by Omnimap anti-rabbit (760-3211 Ventana) and detection with Red610 (760-245, Ventana) for 8 min. An antibody denaturation step was performed with CC2 at 100 °C for 20 min. Fourth, incubation with anti-CD11c was performed for 12 min at 37 °C, followed by Omnimap anti-rabbit HRP (760-4311, Ventana) and detection with Cy5 (760-238, Ventana) for 8 min. An antibody denaturation step was performed with CC2 at 100 °C for 20 min. Finally, anti-CXCL13 was incubated for 20 min at 37 °C, followed by Omnimap anti-rabbit HRP (760-4311, Ventana) and detection with FAM (760-243, Ventana) for 8 min. Finally, slides were washed in PBS and mounted with Vectashield containing 4′,6-diamidino-2-phenylindole (Vector Laboratories). Slides were imaged with an Axioscan 7 (ZEISS).

Automated quantification of PLA2G2D⁺ cells

PLA2G2D-stained whole slides were scanned using the NanoZoomer (20×, Hamamatsu, HT2.0) and analysed using the open-source digital pathology software QuPath (v.0.5.1)⁶⁴. All slides were set as ‘Brightfield (H-DAB)’, and for each slide, stain vectors were set for haematoxylin and DAB. ROIs of 200 µm by 200 µm in TDLNs and primary tumour tissues were selected for positive cell detection. Interfollicular regions and the centre of the primary tumour or the LN metastasis were selected for ROIs, and germinal centres, tissue borders and anthracotic areas were excluded from ROI selection. The number of ROIs per tissue depended on the tissue size. Cell segmentation and positive cell detection was performed on the basis of nuclear detection and mean DAB stain intensity using the built-in QuPath positive cell detection command, with the following settings: detection image, haematoxylin OD; requested pixel size, 0.5 µm; background radius, 8.0 µm, performing background by reconstruction; median radius, 0.0 µm; sigma, 1.5 µm; minimum area, 10.0 µm, maximum area, 200.0 µm; threshold, 0.1; maximum background, 2.0; cell expansion, 4.0 µm, including nuclei and smooth boundaries. Three staining intensity thresholds were set using the following settings: threshold compartment: cytoplasm: DAB OD mean, threshold positive 1+, 0.08; threshold positive 2+, 0.16; threshold positive 3+, 0.24. For each slide, ROI measurements were exported and used for analysis. Per TDLN or tumour, as well as per patient, the average of all ROIs per tissue was used to calculate the percentage of positive cells, positive cells per mm² and the number of different positive cell intensities (1+, 2+ or 3+).

Collection of TDLN tissue from patients with melanoma or NSCLC for cell culture

We collected fresh TDLN tissue from patients with NSCLC or melanoma who underwent surgery at the Erasmus MC. Viable cells were obtained using the scrape cytology method, as described previously^65,66. In brief, excised LNs were bisected during the standard pathology workflow, and a scalpel blade was used to shave along the cut surface and rinsed in 3 ml RPMI culture medium to obtain cells. This way, the LN samples could still be used for further processing without interfering with diagnostic workflows. The cell suspensions were filtered through a 35-μm filter cap (Falcon) and centrifuged at 400g for 7 min. The medium was discarded and the cells were cryopreserved at −196 °C in RPMI containing 40% fetal bovine serum (FBS) and 10% dimethyl sulfoxide. All TDLN tissue samples were stored after resection at 4 °C and processed within 18 h.

Flow-cytometry analysis

TDLN samples were thawed in RPMI + 10% FBS, centrifuged for 7 min at 400g, resuspended in RPMI with 10% FBS and incubated for 60 min at 37 °C. Then, cells were washed and resuspended in FACS buffer (0.5% BSA, 10% NaN₃ sodium azide and 2 mM EDTA in PBS) and stained for cell-surface markers (Supplementary Table 7) in Brilliant Stain Buffer (BD) for 30 min at 4 °C. Cells were washed with PBS and labelled with Fixable Viability Stain 575V (BD) for 15 min at 4 °C. For intracellular staining, the FoxP3 fixation/Permeabilization Kit (eBioscience) guidelines were followed. All cells were then washed and resuspended in FACS buffer before analysis on a BD FACSymphony A5 flow cytometer (FACS Diva software v.9; BD).

TDLN T cell in vitro stimulation assay

TDLN samples were thawed and stained for cell-surface molecules as described above. Cells were washed and resuspended in MACS buffer (0.5% BSA and 2 mM EDTA in PBS). Before cell sorting, Helix NP Blue was added according to the manufacturer’s guidelines, and sorting was performed using a BD FACSAria III. We sorted three CD8⁺ T cell populations: CD8⁺ T naive (alive lineage⁻CD5⁺CD8⁺CCR7⁺CD45⁺), CD8⁺ T non-naive PD-1^int (alive lineage⁻CD5⁺CD8⁺non(CCR7⁺CD45RA⁺)PD-1^int) and CD8⁺ T non-naive PD-1^hi (alive lineage⁻CD5⁺CD8⁺non(CCR7⁺CD45RA⁺)PD-1^hi). The lineage cocktail contained antibodies against CD19, CD20, CD11c, CD68, CD56, CD14, CD16 and CD123. CD5 was used instead of CD3 to allow stimulation with anti-CD3 after sorting. In the TDLN samples, the lineage-negative CD5⁺ cell population overlapped with the lineage-negative CD3⁺ cell population by more than 98%, as verified by flow cytometry. Sorted cells were collected in 100% FBS. Then, cells were resuspended in culture medium consisting of RPMI with 10% FBS, 1% penicillin–streptomycin and primocin (100 µg ml⁻¹). Per patient, concentrations were determined on the basis of the condition with the lowest cell yield, so that the other conditions contained cells x times the lowest cell count. The antibodies and culture materials used are listed in Supplementary Table 7. To mimic chronic T cell activation, stimulation was performed using plates coated with anti-CD3 at 5 µg ml⁻¹ 24 h before culture. Cells were seeded in 100 µl culture medium with the addition of 2 µg ml⁻¹ soluble anti-CD28 at day 0. Cells were centrifuged, resuspended in culture medium and passaged onto a fresh anti-CD3-coated plate on days 2, 3 and 4, and analysed on day 5. This assay was adapted from a previous study⁶⁷. On day 5, cells were washed and stained for cell-surface and intracellular molecules as described above. Cells were analysed using a BD FACSymphony A5. FlowJo v.10.0 was used for data analysis.

Production of recombinant PLA2G2D protein

Recombinant human or mouse PLA2G2D proteins were generated by fusing the cDNA of human (NM_012400.4) or mouse (NM_011109.3) PLA2G2D with the Fc region of human IgG1 on the C terminus in the pcDNA3.4 expression vector (Thermo Fisher Scientific). An enzymatic-dead mutant of PLA2G2D–Fc was generated by introducing an H47Q mutation into the human PLA2G2D cDNA. An effector-less Fc-tag variant was generated by introducing L234A, L235A and P329G mutations into the human IgG1 fusion tag. PLA2G2D–Fc proteins were expressed by transiently transfecting Expi293 cells using the Expi293 Expression System Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. After 5 days of culture, supernatants were collected and recombinant proteins were purified by protein A affinity chromatography using a HiTrap PrismA column (Cytiva) on a Bio-Rad NGC FPLC system.

In vivo tumour models

C57BL/6 and BALB/c mice were purchased from the Jackson Laboratory or Envigo, housed in specific pathogen-free conditions and fed ad libitum. Pla2g2d^−/− mice were generated by CRISPR–Cas9-mediated gene editing in which deletion of exon 2 of the Pla2g2d gene results in loss of function through a frameshift from exon 1 to exon 3. Proximal (5′-CGAAAACCTAAGAGCCTGCG-3′) and distal (5′-TAGGTGGCTGGAACAACCAT-3′) Pla2g2d guide RNAs and Cas9 protein were injected into C57BL/6NTac zygotes using established methods. PLA2G2D-humanized mice on a C57BL/6 background were generated by Ozgene using gene targeting by homologous recombination. In brief, the mouse Pla2g2d genomic locus from residues L21 within exon 2 to the stop codon in exon 4 was replaced with human cDNA encoding residues I22 through to the stop codon. Gene targeting was performed on C57BL/6 ES cells. LN6-987AL and B16F0 cells¹⁹ were provided by N. Reticker-Flynn and cultured in Dulbecco’s modified Eagle’s medium (DMEM). MC38 and B16F10 (ATCC) tumour-cell lines were cultured in DMEM, and E.G7-OVA and CT26 (ATCC) tumour-cell lines were cultured in RPMI medium. All media were supplemented with Glutamax, gentamycin (50 µg ml⁻¹) and 10% FBS. Cell lines were not subjected to further authentication but were routinely tested for mycoplasma contamination. To prepare mice for tumour inoculation, they were shaved on the right flank. MC38, B16F10, CT26 or E.G7-OVA (10⁶ cells) and LN6-987AL or B16F0 (2 × 10⁵ cells) were collected and washed twice with PBS before subcutaneous inoculation (in 100 µl PBS) into mice. Mice were monitored daily, and tumours were measured by calliper every 3–4 days. When tumours reached 50–150 mm³ and were palpable, mice were randomized into groups and antibodies or proteins were administered i.p. every 3–4 days for a total of four doses. Mice with palpable tumours were injected with anti-PLA2G2D, anti-IFNγ (BioXCell) or isotype control antibodies (all 10 mg kg⁻¹) in 200 µl PBS. Anti-mouse PD-1 antibody (clone RMP1-14, Leinco) was dosed at 5 mg kg⁻¹ using the same schedule. Recombinant PLA2G2D–Fc protein was administered i.p. at 50 µg in 100 µl PBS per dose every 4 days for a total of four doses. For CD8⁺ T cell depletion experiments, anti-CD8 depleting antibodies (200 µg) were administered i.p. one day before the initiation of anti-PLA2G2D treatment. Antibody administration (mouse IgG2a isotype control; anti-mouse CD8a, clone YTS169, BioXCell) was repeated every 3–4 days and continued throughout the duration of the experiment. In some experiments, blood, LNs, TDLNs and tumours were collected at the indicated time points after tumour inoculation. Single-cell suspensions of LNs and TDLNs were prepared by mechanically dispersing the LNs through a 100-μm nylon-mesh cell strainer. For tumours, single cells were prepared using the Tumour Dissociation Kit (Miltenyi). After staining with antibodies as described above, cell suspensions were analysed by flow cytometry. Mice were killed when tumours greater than 2,000 mm³ developed or when tumour ulceration was observed. Mice (only females) were 8–12 weeks old when used in experiments. Sample sizes were determined on the basis of previous experience with the tumour models (for example, variability in growth kinetics in vivo), and on the basis of the practical availability of specific mice. Blinding was not applied in this study. Source data for all in vivo experiments are provided.

Single-cell RNA-seq and data analysis

C57BL/6 mice were inoculated with B16F0-tdTomato or LN6-987AL-tdTomato cells as described above and treated with isotype or anti-PLA2G2D antibodies at days 8 and 11. At day 12, tumours, TDLNs and non-draining LNs were collected and single-cell suspensions were prepared and stained for extracellular antibodies as described above. Using a FACSAria III (BD), cells were pooled from four sorted LN (or TDLN) populations that were mixed in equal ratios: (1) alive cells; (2) CD45⁻ cells; (3) CD45⁺CD3⁺CD44⁺ cells; and (4) combined CD45⁺CD3⁻CD11b⁺ + CD45⁺CD3⁻CD11c⁺ cells. From processed tumours, cells were pooled and mixed from three sorted fractions: (1) alive cells; (2) CD45⁺CD3⁺CD44⁺ cells; and (3) CD45⁺CD3⁻CD11b⁺ + CD45⁺CD3⁻CD11c⁺ cells. Next, pooled cell fractions were encapsulated for cDNA synthesis and barcoded with multiplexed runs on a Chromium X using the GEM-X 3′ OCM Chip (on-chip multiplexing). Final sequencing libraries were generated following the instructions of the GEM-X Universal 3′ Gene Expression v4 4-plex user guide (CG000768). Sequencing was performed on a Novaseq 6000 platform (Illumina) using an S2 (100 cycle) flow cell with a 28-10-10-90 cycle setting. The target sequencing depth of each library was >25,000 reads per cell. Sequencing reads were aligned to the mouse mm39 (GRCm39-2024-A) reference genome and quantified using CellRanger (10X Genomics, v.9.01). CellRanger output was loaded into R using Seurat (v.5.3.0) and filtered (200–6,000 features per cell, less than 15% mitochondrial DNA). Next, counts were log-normalized, variable features were called (nfeatures = 3000) and the data were scaled. Next, a principal component analysis was run and a sample batch correction was performed using Harmony (v.1.2.3). Clusters were identified on the basis of shared-nearest-neighbour clustering using the first 30 principal components (PCs), with a resolution of 0.3 and k = 30. The first 30 PCs were then used to generate UMAP projections with a minimum distance of 0.3. Clusters were annotated using known gene markers: T cells (Cd3e, Cd3g, Cd3d, Cd4, Cd8a and Cd8b), B cells (Ms4a1 and Cd19), DCs (Cst3 and Flt3), macrophages (Cd68 and Csf1r), neutrophils (S100a9), NK cells (Klrb1c) and tumour (Stmn1). In addition, FindMarkers with a Wilcoxon rank-sum test and an absolute log₂-transformed fold change cut-off of 1 was used to identify cluster marker genes and further stratify cell annotation. Previously published gene signatures of DCs and macrophage populations^68,69 were used to validate the identity of myeloid subsets. Published breast cancer (GSE167036) and lung adenocarcinoma (GSE131907) datasets containing both tumour and LN (or TDLN)-derived cells were processed, clustered and annotated as described in the original papers, with some simplification of complex annotations to improve comparisons between datasets. PLA2G2D expression was projected onto a UMAP of cells from the LNs (or TDLNs) using Seurat.

Bone-marrow chimera experiments

Pla2g2d^+/+ (CD45.1⁺) and Pla2g2d^−/− (CD45.2⁺) mice (8–12 weeks old) were lethally irradiated with two doses of 500 cGy administered 4–6 h apart (total 1,000 cGy) to induce myeloablation. After irradiation, mice received antibiotics in their drinking water for 2–3 weeks. Bone marrow was collected from the tibiae and femora of congenic Pla2g2d^+/+ C57BL/6 (CD45.1⁺) and Pla2g2d^−/− (CD45.2⁺) donor mice (6–8 weeks old). Cells were flushed with RPMI supplemented with 10% FBS, filtered through a 70-μm strainer and subjected to ACK lysis to remove erythrocytes. Cells were washed, counted and resuspended in PBS at 1 × 10⁶ cells per 100 μl. Recipient mice received bone-marrow cells by retro-orbital injection. Donor chimerism was assessed 6 weeks later by flow cytometry. After chimerism was confirmed, mice were subcutaneously implanted with MC38 tumour cells for downstream analysis of tumour growth.

Generation of anti-PLA2G2D antibodies

Antibodies against PLA2G2D were generated by immunizing BALB/c mice with recombinant human PLA2G2D–Fc-tagged proteins. Mice were immunized at multiple sites using complete Freund’s adjuvant for multiple boosts. Test bleeds were performed by saphenous vein lancing seven days after the last boost. When the antibody titre was high enough, mice were given a final intravenous boost in the lateral tail vein, after which immunized mice were killed and spleens isolated. Hybridomas were generated by electrofusion of splenocytes with Sp2/0 myeloma cells. Fused cells were plated into 96-well plates in HAT selective medium and grown for 10–14 days to generate hybridoma clones, which were assayed for binding to human PLA2G2D proteins by ELISA. Positive binders were expanded and antibodies in the supernatants were purified by protein G (HiTrap Protein G HP, GE Healthcare). The specificity of anti-PLA2G2D antibodies was determined by verifying their lack of binding to other related sPLA2 family members by ELISA. In brief, recombinant PLA2G1B, PLA2G2A, PLA2G2E, PLA2G2F, PLA2GV and PLA2G2X Fc-tagged proteins were generated as described above for PLA2G2D–Fc. Binding to sPLA2 family members was tested by ELISA. Purified proteins were coated at 0.5 µg ml⁻¹ on 96-well plates and titrations of anti-PLA2G2D antibodies or isotype control antibodies were applied, followed by HRP-conjugated secondary detection and visualization with TMB-Ultra reagent (Thermo Fisher Scientific). The specificity of the antibodies was further tested by evaluating their ability to inhibit the enzymatic activity of PLA2G2D–Fc or the various sPLA2 family members in a phospholipase enzyme assay as previously described³⁷.

Isolation and culture of human peripheral blood cells

Human white blood cell concentrate was obtained from Stanford Blood Center or the Sanquin Blood Bank, and peripheral blood mononuclear cells (PBMCs) were prepared by Ficoll density gradient. Pan T cells, CD8⁺ T cells and monocytes were isolated from human PBMCs using Pan T, CD8⁺ T Cell and Pan Monocyte isolation kits (Miltenyi) according to the manufacturer’s instructions, respectively.

Stimulation of T cells in PBMC-based assays

PBMCs were labelled with CellTrace CFSE (Invitrogen) and T cells were stimulated with 1 µg ml⁻¹ anti-CD3 and 0.2 µg ml⁻¹ anti-CD28 antibodies in the presence of recombinant PLA2G2D–Fc, PLA2G2D(H47Q)–Fc or Fc control in RPMI. After 48 h of incubation, cell-culture supernatants and cells were collected for cytokine production and flow-cytometry analysis, respectively. IL-2 and IFN-γ levels were measured using Meso Scale Discovery Human or Mouse V-Plex/U-Plex kits. T cell proliferation was measured by multicolour flow cytometry on a BD LSRFortessa X-20 flow cytometer using CD3, CD4, CD8 and live–dead markers for live T cell gating, followed by CFSE dilution analysis for proliferation analysis according to the manufacturer’s instructions. The binding of PLA2G2D to T cells was determined by incubating various concentrations of PLA2G2D–Fc or Fc control proteins to resting or activated T cells for 30 min, followed by washing and secondary detection with fluorophore-labelled anti-human IgG antibodies and subsequent analysis on a flow cytometer.

Stimulation of isolated T cells for surface–PLA2G2D binding kinetics

CD8⁺ cells were isolated and cultured as described above and stimulated with T cell TransAct (Miltenyi). At the indicated time points, cells were collected, washed and stained with PLA2G2D–His for 30 min at 4 °C in PBS. Next, the cells were stained with anti-His-APC together with other extracellular antibodies and live–dead markers as described above. All cells were then washed and resuspended in FACS buffer before analysis on a BD FACSymphony A5 flow cytometer.

Stimulation of isolated T cells for intracellular cytokine staining and proliferation assays

CD8⁺ cells were isolated and cultured as described above and stimulated with T cell TransAct (Miltenyi) in the presence or absence of 10 µg ml⁻¹ PLA2G2D protein, which was added again after 48 h. For proliferation assays, isolated CD8⁺ cells were labelled with the CellTrace Violet Cell Proliferation Kit (0.05 µM, Thermo Fisher Scientific) before starting the culture. After 72 h of incubation, cells were collected, washed and stained with antibodies and live–dead markers as described above. For intracellular cytokine staining, cells were then fixed with 2% paraformaldehyde for 20 min at room temperature and permeabilized with 0.5% saponin for 10 min at 4 °C, after which the cells were stained for intracellular cytokines in 0.5% saponin for 60 min at 4 °C. All cells were then washed and resuspended in FACS buffer before analysis on a BD FACSymphony A5 flow cytometer.

Bulk RNA-seq of stimulated T cells

CD8⁺ cells were isolated and cultured as described above and stimulated with T cell TransAct (Miltenyi) in the presence or absence of 10 µg ml⁻¹ PLA2G2D protein. After 24 h of incubation, cells were collected and washed, and RNA was isolated using the RNeasy Micro kit (QIAGEN) according to the manufacturer’s instructions. Samples for bulk RNA-seq were sequenced and analysed as described before⁷⁰. In brief, raw counts and reads per kilobase million (RPKM) were determined on the exons of RefSeq-annotated genes using HOMER’s (v.5.1)⁷¹ analyzeRepeats.pl command. The R package DESeq2 v.1.48.2 was used for the identification of DEGs, sample scaling and statistical analysis. Genes with an RPKM < 1 in 50% of replicates in one condition were excluded. DEGs were determined by an absolute log₂-transformed fold change value > 0.5 and an adjusted P value < 0.05. Pathway enrichment analysis on DEGs was performed using Metascape⁵⁸ (v.3.5).

Mixed lymphocyte reactions

For mouse allogeneic mixed lymphocyte reactions, human monocyte-derived DCs were differentiated from monocytes using the DC differentiation toolbox (Miltenyi) according to the manufacturer’s instructions. Monocyte-derived DCs were matured in the presence of TNF where indicated. DCs were then treated with mitomycin-C for 1 h at 37 °C, followed by three washes with RPMI, then added to a 96-well round-bottomed plate. T cells were added to DCs at a DC-to-T cell ratio of 1:10 in RPMI, and were incubated for 5 days before analysis of cytokine levels by MSD or collection of conditioned medium. Where indicated, 0.5 µg ml⁻¹ recombinant PLA2G2D–Fc, PLA2G2D(H47Q)–Fc, Fc control or anti-PD-1 (pembrolizumab; 5 mg ml⁻¹ or 10 mg ml⁻¹) were added during the allogeneic coculture.

Quantitative PCR

Total RNA was isolated from in vitro differentiated macrophages, MDMs and DCs using the RNeasy Mini Kit (QIAGEN) according to the manufacturer’s instructions. Total RNA was synthesized into cDNA using the High-Capacity cDNA Reverse Transcription Kits (Applied Biosystems). Quantitative PCR with reverse transcription (qRT–PCR) was performed using Power SYBR Green PCR Master Mix (Applied Biosystems) per the manufacturer’s instructions. The amplification condition was set at 40 cycles at 95 °C for 15 s and 60 °C for 60 s using an ABI Prism 7900HT Sequence Detection System (Applied Biosystems). β-actin (ACTB) and hypoxanthine phosphoribosyl transferase (HPRT1) were used for normalization. Cq values for the two reference genes were determined by ABI SDS software (v.1.5.1) and used to calculate the geometric mean of ACTB and HPRT1 as the normalization factor using GeNorm. The normalization factor was applied to PLA2G2D data, and values were converted into conventional fold changes by setting the value of a control sample to 1 as indicated, scaling all other values proportionally. Primer sequences used:

PLA2G2D forward: 5′-CAACCCAAAGATGCCACG-3′ and reverse: 5′-CCAGCTTCCCTTGTCAGAG-3′.

ACTB forward: 5′-ATTGCCGACAGGATGCAGAA-3′ and reverse: 5′-GCTGATCCACATCTGCTGGAA-3′. HPRT1 forward: 5′-GACCAGTCAACAGGGGACAT-3′ and reverse: 5′-AACACTTCGTGGGGTCCTTTTC-3′.

Statistical analysis

For IMC data analyses, differential cell abundances were calculated using a two-sided t-test (Rstatix v.0.7.2; https://rpkgs.datanovia.com/rstatix/). Wilcoxon signed-rank tests and unpaired t-tests were implemented for non-normally and normally distributed data, respectively, using the stat_compare_means() function included in ggpubr v.0.6.0. Correlations were calculated as Pearson’s correlation coefficient (r) as implemented by ggpubr v0.6.0 (https://rpkgs.datanovia.com/ggpubr/) using the stat_cor function. For DSP whole-transcriptome analysis, DEGs were calculated using a LMM as implemented in the GeomxTools v.3.10.0 R package. Statistical analyses of the flow cytometry, tumour growth curves, survival analysis and IHC and immunofluorescence results were performed in GraphPad Prism v.10. Error bars reported for bar graphs in the figures denote s.e.m. unless indicated otherwise. For box plots, boxes indicate the median and interquartile range, with whiskers representing the data distribution excluding outliers. Paired or unpaired t-tests were used for comparisons of paired data with Gaussian distributions, and paired or unpaired Wilcoxon tests were performed for non-normally distributed paired data. P values (and adjusted P values) smaller than 0.05 were considered statistically significant; exact P values are provided in the Supplementary Data. Data points used in the figures were obtained from independent biological samples. When representative data are shown, similar results were obtained from one or more independent replicate experiments.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.