Cancer-Intrinsic Protein Neddylation Modulates the Intratumoral Immune Landscape to Constrain Immune Checkpoint Blockade Therapy

The authors show a role for cancer-intrinsic neddylation in controlling the intratumoral immune landscape of TNBC and response to ICB. The demonstration of a mechanistic link between neddylation and lipid metabolism suggests targets to enhance response to immunotherapy.

Abstract

The cancer-intrinsic network controlling the response to immune checkpoint blockade therapy remains elusive. Using a published single-cell RNA sequencing (scRNA-seq) dataset from patients with breast cancer, we found that pretherapy mRNA expression of NEDD8, which encodes the regulatory protein essential for neddylation, in cancer cells was negatively associated with T-cell expansion upon pembrolizumab treatment. Genetic ablation of protein neddylation in murine tumors induced a curative response to PD-1 or PD-L1 blockade in mice. Combined analysis of neddylation-deficient murine tumors using scRNA-seq and high-resolution spatial transcriptomics revealed distinct spatial immune co-localization and immune cell signatures. In a high-content phenotypic screen of known compounds, protein neddylation was required for the induction of phenotypic changes by lipid inhibitors and ferroptosis inducers in human breast cancer cells. In accordance, these compounds enhanced the efficacy of PD-1 blockade and protected mice from tumor rechallenge. Altogether, we delineate the unknown functions of protein neddylation on antitumor immunity and provide therapeutic options that can enhance immunotherapy.

Introduction

Immune checkpoint blockade (ICB) therapy provides clinical benefits in patients with solid tumors. However, the tumor microenvironment contains a range of suppressive mechanisms to dampen sufficient immune activation (1). On one hand, cancer-intrinsic pathways could render insufficient cytotoxicity mediated by T cells. On the other hand, cancer-derived chronic inflammatory signals could induce myeloid cells with immunosuppressive properties. Therefore, overcoming cancer cell–intrinsic and cancer-driven immune resistance mechanisms is crucial to enhance cancer susceptibility to immunotherapy.

High-throughput genetic screens have been instrumental in identifying genes that confer resistance to immunotherapy in cancer cells. For instance, genome-wide CRISPR/Cas9 and epigenetic screens in tumor-bearing mice have uncovered key regulators of immune evasion and resistance to ICB therapy (2–9), paving the way for the development of targeted immunotherapies for patients (10). Using genetic screens in a human coculture system, we previously revealed the function of cancer-intrinsic protein neddylation in resistance against nivolumab (11). Protein neddylation is a multienzymatic process that involves substrates being conjugated with the neural precursor cell expressed developmentally downregulated protein 8 (NEDD8); it is a crucial posttranslational modification mechanism that modulates protein activity and various biological processes, including tumorigenesis (12, 13). However, the precise molecular network controlled by protein neddylation during cross-talk between cancer and immune cells remains elusive.

In this study, we utilized a published single-cell RNA sequencing (scRNA-seq) dataset generated from patients with breast cancer, as well as genetically modified murine tumors, to demonstrate the impact of cancer-intrinsic protein neddylation on the intratumoral immune landscape upon PD-1 blockade therapy. Our analysis revealed that high protein neddylation in cancer cells limited the expression of antigen-presenting molecules and rewired lipid metabolism in both cancer cells and macrophages. Moreover, single-cell spatial transcriptomics analysis using the CosMX technology illustrated the dynamic spatial–temporal changes in macrophage infiltration and co-localization. In human triple-negative breast cancer (TNBC) cells, protein neddylation was found to be required for the phenotypic changes induced by statin drugs and ferroptosis inducers. Indeed, combining these compounds with PD-1 blockade enhanced their antitumor efficacy and immune memory in tumor-bearing mice.

Materials and Methods

Detailed information on the inhibitors, antibodies, proteins, reagents, and sequences used in the study is summarized in Supplementary Tables S1–S4.

Sex as a biological variable

Our study only employed female mice as a model system due to the prevalence of breast cancer in female patients. It is unknown whether the findings are relevant to male mice.

Cell lines and cultures

The human TMBC cell line MDA-MB-231 (RRID: CVCL_0062) was purchased from Sigma-Aldrich in 2019. The HCC1937 cell line (RRID: CVCL_0290) was a gift from Prof. Óscar Fernández-Capetillo in 2020 (Karolinska Institutet, Sweden). The EO771 murine breast cancer cell line (RRID: CVCL_GR23, ATCC) was generously provided by Dr. Maria Ulvmar in 2019 (Uppsala University, Sweden). The murine acute myeloid leukemia (AML) cell line C1498 (RRID: CVCL_3494) was purchased from ATCC in 2023. Stock cell lines were stored in liquid nitrogen, and cells in culture were replaced after 10 passages. CRISPR knockout (KO) and control isogenic cell lines for the NEDD8 gene in human MDA-MB-231 and murine EO771 and C1498 cell lines were generated according to a previously published protocol (11). Information on the source of the reagents can be found in Supplementary Table S3. In brief, ribonucleoprotein (RNP) complexes containing the CRISPR RNAs (crRNA) targeting human or mouse genes of interest (Supplementary Table S4), trans-activating crRNAs, and an endonuclease Cas9 protein were transfected into cancer cells. RNP complexes lacking crRNAs were used as a negative reaction. Cell pellets (5 × 10⁵ cells) were resuspended in 5 μL of resuspension buffer R or buffer T and mixed with an equal volume of the RNP complex. The Neon transfection system was set up following the manufacturer’s instructions, after which the cell mixture was transferred into Neon pipette tips, and electroporation was conducted using specific programs. Cells were repeatedly transfected with either gene-targeting RNP complexes or control complexes to achieve complete gene deletion. All cell lines were cultured in Iscove’s Modified Dulbecco’s Medium (IMDM; Thermo Fisher Scientific) supplemented with 10% heat-inactivated fetal bovine serum (FBS; Cytiva) and 1% penicillin–streptomycin (Thermo Fisher Scientific) at 37°C in a humidified incubator with 5% CO₂, unless otherwise specified. Cell line authenticity was verified by DNA fingerprinting (Eurofins), and routine mycoplasma testing was performed using the MycoAlert kit (Lonza).

Western blotting

To determine the expression of NEDD8 protein, cancer cell pellets were lysed on ice for 15 minutes using RIPA buffer containing 1 mmol/L EGTA, 20 mmol/L Tris, 150 mmol/L NaCl, 1 mmol/L EDTA, 1% NP-40, 1 mmol/L NaF, 1 mmol/L NaVO3, and 1 mmol/L sodium phosphate without additional reducing agents and supplemented with a protease inhibitor cocktail (Thermo Fisher Scientific). The lysates were then centrifuged at 17,000 × g for 12 minutes at 4°C to remove cellular debris. Protein concentration was determined by the bicinchoninic acid assay (Thermo Fisher Scientific). Next, proteins were treated with 1× SDS loading dye, boiled at 70°C for 10 minutes, and then loaded into the 4% to 12% Bis-Tris precast gels (Invitrogen) for SDS-PAGE. Proteins were transferred to a nitrocellulose membrane (Invitrogen) using the iBlot Transfer System (Invitrogen), followed by blocking with 5% nonfat skim milk powder (OXOID) for 1 hour. The membranes were then incubated with primary antibodies against the target proteins or loading control proteins (Supplementary Table S2) at 4°C overnight. Next, membranes were incubated in corresponding horseradish peroxidase–conjugated secondary antibodies at room temperature for 1 hour, followed by visualization using SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo Fisher Scientific) in an Amersham Imaging System (GE HealthCare). TBS (1×) with 0.05% Tween 20 and dH₂O was used as a wash buffer.

Testing of antitumor efficacy in immunocompetent mice

The mice utilized in this study were kept at the animal facility located in the Department of Immunology, Genetics, and Pathology at Uppsala University. Experiments were conducted under ethical permits 5.8.18-06394/2020 and 5.2.18-05492.2025, which were reviewed and approved by the Swedish Board of Agriculture, Jönköping, Sweden.

To investigate the impact of Nedd8 deletion on tumor growth, age-matched female C57BL/6NTac mice (6–10 weeks old, purchased from Taconic) were inoculated with syngeneic murine cancer cell lines, EO771 or C1498, respectively. Briefly, mouse Nedd8 KO or control cells were injected subcutaneously into the right flank at 6 × 10⁵ cells (EO771) or 4 × 10⁵ cells (C1498) per mouse. Tumor volumes and body weights were checked regularly throughout the study until the maximum humane endpoint of 1,500 mm³. Tumor volumes were calculated using the following formula: (length × width²)/2. When tumors had reached ∼50 mm³ or 7 days after inoculation, mice were treated intraperitoneally with either an anti–PD-1 (clone RMP1-14, BioXcell, RRID: AB_10949053) or a rat isotype IgG2a control (clone 2A3, BioXcell, RRID: AB_1107769), 50 μg per mouse every 3 days. For the anti–PD-L1 study, mice were treated intraperitoneally with anti–PD-L1 (200 μg, clone 10F.9G2, BioXcell, RRID: AB_10949073) or an IgG2b isotype control (clone LTF-2, BioXcell, RRID: AB_1107780), 4 days after tumor inoculation and then every 3 days. In cell depletion experiments in the EO771 Nedd8 KO tumor model, mice were intraperitoneally infused with a rat IgG isotype control antibody (RRID: AB_1107769), a CSF1R-depleting antibody (150 μg, clone AFS98, BioXcell, RRID: AB_2687699), or a CD8a-depleting antibody (150 μg, clone 2.43, BioXcell, RRID: AB_1125541) once per week, starting from the day before anti–PD-1 treatment. Anti–PD-1 (100 μg, RRID: AB_10949053) or rat IgG2a isotype (RRID: AB_1107769) control was administered intraperitoneally on days 7, 10, and 13. For the combination depletion experiment, mice were injected intraperitoneally with 70 μg anti–PD-1 or rat IgG2a isotype antibody on days 7, 10, and 13. For rechallenge studies, EO771 (1.5 × 10⁶ cells) or C1498 (3 × 10⁵ cells) were injected subcutaneously into the left flank of cured mice in 100 μL serum-free IMDM (Thermo Fisher Scientific), and tumor growth was followed.

For the inhibitor treatment studies, simvastatin (8.2 mg/kg) or RSL3 (10 mg/kg; both MedChemExpress) was administered intraperitoneally at indicated human equivalent doses in 100 μL of vehicle control (PBS with 10% DMSO and 20% β-cyclodextrin, cat. #463361000, Thermo Fisher Scientific), together with 50 μg anti–PD-1 or rat IgG2a isotype antibody on days 7, 10, and 13 after tumor implantation. Compounds were administered one additional time on day 16. EO771 control cells (2 × 10⁶ cells per mouse) were injected intraperitoneally in cured mice in rechallenge studies, followed by flow cytometry analysis of the splenocytes. Interchange of units (mg/kg to mg/m²) of human dose and calculation of animal equivalent doses (AED) were carried out using the following formulas: mg/m² = K_m × mg/kg, and AED (mg/kg) = Human dose (mg/kg) × K_m ratio, respectively (14).

Flow cytometry analysis

To determine the effect of recombinant human IFNα (rhIFNA) and recombinant human IFNβ (rhIFNB) on human MDA-MB-231 control (ctrl)/NEDD8 KO cell lines, cells (2 × 10⁵) were cultured in six-well flat-bottom plates overnight to allow cell adherence and subsequently treated with or without rhIFNA or rhIFNB (50 ng/mL, PeproTech). In some experiments, pharmacologic inhibitors against JAK1/2 (Ruxolitinib, 1 μmol/L, MedChemExpress), STING (H-151, 2 μmol/L, MedChemExpress), or cGAS (G140, 2 μmol/L, TargetMol) were added to cells for 2 hours before stimulating with rhIFNB. After 24 hours of incubation, cells were harvested and centrifuged at 350 × g for 4 minutes for flow cytometry analysis. In addition, culture media were collected, and soluble factors were analyzed using a human antivirus response LEGENDplex kit (BioLegend). Information about the antibodies used for flow cytometry can be found in Supplementary Table S2.

To determine the expression of surface proteins, the cells were transferred to a 96-well V-bottom plate in triplicates (200 μL/well). Cell pellets were resuspended with 20 μL PBS containing aqua fixable live/dead marker (1:100, Thermo Fisher Scientific) and human TruStain FcX Fc receptor blocking antibody (1:100, BioLegend, RRID: AB_2818986) for 25 minutes at 4°C. After washing with PBS, cells were stained with detection antibodies (1:100) for surface proteins for 25 minutes at 4°C, followed by washing with PBS. Then, antibody-stained cells were resuspended with 150 μL PBS. For the analysis of cells from in vivo studies, single-cell suspensions of splenocytes were isolated from the cured mice on day 7 after rechallenge treatment by passing spleens through 40 μm cell strainers. Next, red blood cells were washed away by incubation with the red blood lysis buffer (BioLegend, RRID: AB_2863073) for 2 minutes on ice and subsequent washing with PBS. Frozen single-cell suspensions isolated from murine tumors using a tumor dissociation kit (Miltenyi Biotec) and a gentleMacs instrument (Supplementary Table S3) were thawed and washed twice with PBS. Single cells were then loaded into a 96-well V-bottom plate and incubated with 20 μL PBS containing Aqua Fixable LIVE/DEAD Marker (1:200, Invitrogen) and Fc recptor blocker (1:100, Thermo Fisher Scientific, RRID: AB_467134) for 30 minutes at 4°C, followed by washing with PBS. Subsequently, cells were stained in 20 μL of a mastermix containing a panel of fluorochrome-conjugated antibodies targeting surface proteins for 30 minutes at 4°C. To detect intracellular proteins, cells were fixed and permeabilized using the True-Nuclear buffer set (BioLegend), followed by incubation with fluorochrome-conjugated antibodies (1:50) for 45 minutes at 4°C. Antibody-stained samples were quantified using a CytoFLEX LX flow cytometer (Beckman Coulter), and the data were analyzed using FlowJo software v10.

scRNA-seq

To analyze the mode of action for Nedd8 deletion upon PD-1 blockade, 6 × 10⁵ EO771 Nedd8 KO/ctrl cells were injected subcutaneously into syngeneic C57BL/6NTac mice. Mice were then treated intraperitoneally with 50 μg anti–PD-1 (clone RMP1-14, BioXcell, RRID: AB_10949053) or a rat isotype IgG2a control (clone 2A3, BioXcell, RRID: AB_1107769) on days 10, 13, and 16.

For the scRNA-seq analysis, size-matched tumors were harvested, and single cells were isolated using a tumor dissociation kit (Miltenyi Biotec) and a gentleMacs instrument. Cell pellets were resuspended in 1 mL Aqua Fixable LIVE/DEAD Marker (1:100, Invitrogen) and incubated for 30 minutes at 4°C, followed by washing with PBS. Next, 1 million viable cells were sorted using a CytoFLEX SRT (Beckman Coulter), and the rest of the cells were stored in liquid nitrogen for subsequent use. Then, the live cells were fixed using the Chromium Next GEM Single Cell Fixed RNA Sample Preparation Kit, 16 rxns (PN-1000414, 10x Genomics), prior to library preparation, according to the manufacturer’s instructions. Gene expression of fixed samples was evaluated following Chromium Fixed RNA Profiling (Gene Expression FLEX). Fixed samples were hybridized with barcoded probes, and the sequencing library was generated following the user guide (CG000527). Fixed cell sample concentration was evaluated with the automated cell counter LUNA-FX7 before the hybridization step. Samples were hybridized with the 16-barcoded probe kit. Incubation of the samples at 42°C was performed for 22 hours and 30 minutes. After hybridization, samples were pooled using an equal number of cells for 15 out of 16 samples. Fifteen samples contributed to the multiplexed pool with 5 × 10⁴ cells, whereas one sample contributed with 5 × 10³ cells due to the reduced cell number. The multiplexed pool, containing 7.55 × 10⁵ cells, was washed and filtered following the user guide. The cell concentration of the multiplexed pool was determined before loading the Chromium X chip, using the automated cell counter, with targeted cell recovery aimed at 1.28 × 10⁵ cells. The library was sequenced as 90 bp paired-end reads on an Illumina NovaSeq X Plus system, based on user guide recommendations. A base quality score >30 was used as a threshold, and Cell Ranger was used as a trimming tool. FASTQ files were processed with Cell Ranger v8.0.1 (10x Genomics) using the Mouse Reference Genome Mm10 2020 A.

scRNA-seq data preprocessing and analysis

Information about cell type annotation, time point of biopsy, and T-cell expansion status in patients with breast cancer was included in the original datasets (15). To define T-cell expansion, the T-cell receptor (TCR) sequence needs to be identified in at least two T cells from the same patient, and the frequency needs to increase after the pembrolizumab treatment. If the T-cell clone is only identified after the treatment, it is also defined as an expanded clone. Similar to the original study, patients with more than 30 expanded T-cell clones were considered to be experiencing significant T-cell expansion. Cells were filtered, normalized, and clustered using the pipeline described below. The general cell type labels in the data were used in differential gene discoveries and cell subclustering. For the pseudobulk analysis, cells from the same patient collected at the same time point were aggregated into a single sample and analyzed using DESeq2 (16) to identify differentially expressed genes (DEG). To stratify patients based on NEDD8 mRNA expression, the average normalized expression of NEDD8 in pretherapy tumor cells was calculated for each patient. The mean of these patient-level averages was used as the cutoff to classify patients into NEDD8-low (low expression in pretherapy tumor cells) and NEDD8-high (high expression in pretherapy tumor cells) groups. This classification, derived from the pretherapy samples, was subsequently applied to the patient-paired on-therapy samples.

The scRNA-seq expression matrix, barcode, and metadata files derived from 29 human breast tumors [processed count matrix with metadata from ref. (15)] and 12 murine EO771 tumors (Cell Ranger output and processed matrix from SRR37123745) were integrated and analyzed using the Seurat (v5.1.0) pipeline (17). Sample BC011 was discarded due to its significantly low number of sequenced cells compared with other samples. Low-quality cells were filtered based on the number of unique molecular identifiers (UMI), the number of genes detected, mitochondrial gene ratio, hemoglobin gene ratio, heat shock protein gene ratio, and cell complexity. Specifically, for the number of UMIs, the number of genes detected, and the mitochondrial gene ratio, cells below the 5th percentile or above the 95th percentile were removed. For the hemoglobin gene and heat shock protein ratio, cells above the 95th percentile were removed. For cell complexity, cells below the 5th percentile were removed. The cell complexity score was calculated for each cell using the following formula

$$\text{Cell}\mathrm{\hspace{0.17em}}\text{complexity}\mathrm{\hspace{0.17em}}\mathrm{=}\frac{{\log }_{10}\left(\text{number}\mathrm{\hspace{0.17em}}\text{of}\mathrm{\hspace{0.17em}}\text{gene}\mathrm{s}\mathrm{\hspace{0.17em}}\text{detected}\right)}{{\log }_{10}\left(\text{number}\mathrm{\hspace{0.17em}}\text{of}\mathrm{\hspace{0.17em}}\text{UMIs}\right)}$$

SCTransform (v0.4.1; ref. 18) was applied to normalize the expression counts, and the metrics mentioned above were regressed out. Next, 50 principal components (PC) were computed, and the top 20 PCs were used to generate cell clusters.

Cell types and reclustered cell subtypes were annotated based on the expression of canonical cell markers. InferCNV (v 1.20.0, https://github.com/broadinstitute/inferCNV) was utilized to calculate cellular copy-number variation (CNV) and assess epithelial cell malignancy, with randomly downsampled immune cells used as the reference. Gene signatures for immune cells were obtained from the Molecular Signature Database (MSigDB) depository and published datasets (19, 20). Human gene symbols were converted to mouse aliases first using geneconverter (v1.1.0; ref. 21), then input into the Seurat AddModuleScore() function to compute pathway enrichment scores for each cell. For tumor cells, the STRING database (RRID: SCR_005223) was used to perform gene set enrichment analysis. Normalized P values and FDRs were calculated using tests embedded in the R packages (RRID: SCR_001905). DEGs were accessed with the Seurat FindMarkers() function, using the nonparametric Wilcoxon rank-sum test or the Wilcoxon rank-sum test with Bonferroni P value adjustment.

Single-cell spatial transcriptomics using CosMX Spatial Molecular Imager

For spatial transcriptomics analysis, control or Nedd8 KO EO771 tumor tissues were harvested from mice treated with either IgG or PD-1 blockade and placed in 4% paraformaldehyde overnight, followed by dehydration with 70% ethanol. To prepare formalin-fixed, paraffin-embedded (FFPE) tumor blocks, tumors were embedded and blocked with paraffin using the standard procedure for tissue paraffin embedding. Next, FFPE blocks were sectioned at 5 μm thickness and mounted on VWR Premium Superfrost Plus Micro slides (48311-703) using a standard rotary microtome (UC7, Leica). Slides with the largest area that best represented the tumors were selected. After sectioning, slides were baked at 37°C overnight to improve tissue adherence. Tissue sections were then prepared for CosMx Spatial Molecular Imager (SMI) profiling following the protocol outlined by Denisenko and colleagues (22). Briefly, tissues underwent fixation, dehydration, and heat-induced epitope retrieval, followed by digestion with Proteinase K provided as part of the slide preparation kit (Bruker). Fiducials were applied for spatial alignment, and blocking was performed before in situ hybridization using the CosMX Mouse Universal Cell Characterization Panel (Bruker). After overnight hybridization at 37°C, excess probes were washed off, and the tissues were stained with DAPI and a 4-fluorophore-conjugated antibody cocktail, namely segmentation antibodies (Bruker). Tissue sections were then prepared for imaging by affixing a custom flow cell for analysis on the CosMX SMI instrument. RNA target readout was performed by loading the flow cell and selecting a total of 330 fields of view (FOV; 169 Slide 1 and 161 Slide 2) based on hematoxylin and eosin staining. RNA detection was achieved through sequential hybridization and imaging of 16 reporter pools, with UV cleavage and washing cycles to improve sensitivity.

Analysis of single-cell spatial transcriptomics

Spatial data preprocessing was performed on the NanoString AtoMx platform. Specifically, cells that failed any of the following tests were removed: cell area, number of counts, FOV quality, negative probe proportion, and cell complexity. To annotate cell types in the spatial dataset, we applied the RCTD (23) cell decomposition algorithm embedded in the R package spacexr (v2.2.1, https://github.com/dmcable/spacexr). The scRNA-seq data from EO771 mouse tumors described above were downsampled to up to 300 cells for each cell type to be used as the reference data.

To visualize the relative depth of cells inside the tumor, we computed the convex hull of tumors with the chull() function (grDevices, v4.4.1) and then calculated the infiltration score based on the following formula:

$$\mathrm{S}\left(\mathrm{i}\right)\mathrm{\hspace{0.17em}}\mathrm{=}\frac{\text{dis}}{\sqrt{\text{Area}\mathrm{\hspace{0.17em}}\mathrm{/}\mathrm{\hspace{0.17em}}\mathrm{\hspace{0.17em}}\mathrm{\pi }}}$$

For a specific tumor, dis represents the length between the centroid of a cell inside the tumor and its closest hull line, which can be used to estimate the cell’s distance to the tumor edge. The area parameter is the area of the convex hull, and the square root of area divided by $\mathrm{\pi }$ was used as the approximate radius of the tumor.

We next analyzed the co-localization tendency of different cell types in individual tumors. For cell–cell pairs A − B, the score was calculated as follows:

$$\mathrm{S}\left(\mathrm{A}\mathrm{-}\mathrm{B}\right)\mathrm{=}\frac{\text{average}\mathrm{\hspace{0.17em}}\text{number}\mathrm{\hspace{0.17em}}\text{of}\mathrm{\hspace{0.17em}}\mathrm{B}\mathrm{\hspace{0.17em}}\text{in}\mathrm{\hspace{0.17em}}\text{radius}\mathrm{\hspace{0.17em}}\text{of}\mathrm{\hspace{0.17em}}\mathrm{A}}{\text{average}\mathrm{\hspace{0.17em}}\text{number}\mathrm{\hspace{0.17em}}\text{of}\mathrm{\hspace{0.17em}}\mathrm{B}\mathrm{\hspace{0.17em}}\text{in}\mathrm{\hspace{0.17em}}\text{radius}\mathrm{\hspace{0.17em}}\text{of}\mathrm{\hspace{0.17em}}\mathrm{a}\mathrm{\hspace{0.17em}}\text{random}\mathrm{\hspace{0.17em}}\text{cell}}\mathrm{\hspace{0.17em}}\mathrm{-}\mathrm{\hspace{0.17em}}1$$

We randomly sampled up to 500 cells from each type of cell as central cell A and 2,000 cells from the total cells as the random background central cells.

To determine macrophage-enriched areas, we ran dbscan (v1.2.2; ref. 24) on macrophage spatial coordinates to compute macrophage clusters; all cells close to these clusters were included as well to construct the macrophage niche. Data were normalized with SCTransform() (v0.4.1) and then cellchat (v2.1.2; ref. 25); cell–cell interaction analysis was performed at the niche level.

Cell painting

A morphologic profiling assay using multiplexed fluorescent dyes was used to reveal the effects of known compounds on protein neddylation (see Supplementary Table S1 for information about the compounds). Briefly, compounds were added in a 384-well flat-bottom plate at indicated doses or 0.1% DMSO (control). Next, 9 × 10² MDA-MB-231 ctrl or NEDD8 KO cells were added to the plate in 40 μL of growth medium followed by 96 hours of incubation. Cells were stained with six fluorescent dyes, namely Hoechst for the nucleus, SYTO14 for nucleoli and cytoplasmic RNA, phalloidin and WGA for F-actin cytoskeleton and Golgi apparatus, MitoTracker for mitochondria, and concanavalin A for the endoplasmic reticulum. Automated imaging of the cells was subsequently performed using a SQUID microscope (Cephla). Next, semiautomated image analysis software (Cell Profiler) identified individual cells and extracted 2,110 morphologic features, of which 1,324 were selected (features with median absolute deviation >0.001) to produce a rich profile suitable for detecting subtle phenotypes. Profiles of cell populations treated with different compounds were compared with identify the phenotypic impact of drugs.

Transmission electron microscopy analysis

To analyze subcellular organelles, transmission electron microscopy (TEM) was performed using the MDA-MB-231 or HCC1937 NEDD8 KO/ctrl cell line pairs. Briefly, cells were seeded in six-well flat-bottom plates and incubated overnight to allow adherence. Next, inhibitors or 0.1% DMSO were added to the plates at indicated doses. The plates were then incubated for 3 days (simvastatin) or 4 days (ML210) at 37°C. Cells (1 × 10⁶ per condition) were fixed with 2.5% glutaraldehyde (Ted Pella) + 1% formaldehyde in 0.1 mol/L phosphate buffer (pH 7.4) overnight at 4°C, followed by washing with 0.1 mol/L phosphate buffer. Slices of fixed cells were postfixed with 1% osmium tetroxide (Agar Scientific) for 30 minutes, dehydrated in increasing concentrations of ethanol, followed by a second dehydration with propylene oxide (Agar Scientific) for an additional 5 minutes. The samples were embedded in Agar 100 resin (Agar Scientific) overnight and then polymerized at 60°C for 2 days. Ultrathin sections were cut (50–60 nm) with an ultra-microtome (UC7, Leica) and placed on formvar-coated grids (Ted Pella), followed by staining with 5% uranyl acetate and 3% lead citrate (Science Services). The samples were examined under a TEM microscope (Tecnai Spirit G² BioTwin TEM, FEI/Thermo Fisher Scientific) with an Orius SC200 CCD camera (Gatan) at 80 kV.

Statistical analysis

Flow cytometry data were processed using FlowJo v10 software (RRID: SCR_008520), whereas all results were summarized and analyzed with GraphPad Prism 10 (RRID: SCR_002798). Statistical analyses included the two-group test, unpaired (independent groups) two-tailed t test, with significance set at 0.05. Comparisons across multiple experimental groups (treated or untreated) as independent factors were conducted using a two-way ANOVA (Tukey post hoc), as described in the figure legends. Differential gene expression analyses among groups in the scRNA-seq datasets were performed using the nonparametric Wilcoxon rank-sum test or the nonparametric Wilcoxon rank-sum test with Bonferroni P value adjustment, as stated in the figure legends. Survival differences in tumor-bearing mice between treated and untreated groups were calculated using the Kaplan–Meier curves with Mantel–Cox.

Results

Pretherapy cancer-intrinsic NEDD8 mRNA expression negatively correlates with T-cell expansion upon pembrolizumab treatment

We recently revealed that genetic deletion of NEDD8 in breast cancer cells enhanced the response to PD-1 blockade in experimental models (11). To examine whether cancer-intrinsic NEDD8 mRNA expression affects T-cell expansion in patients, we employed a public dataset in which breast tumor tissues were analyzed using combined single-cell transcriptomics and single-cell TCR sequencing (scTCR-seq) prior to and after one dose of pembrolizumab treatment (15). Given that pathologic tumor response could not be evaluated at this early stage (biopsy obtained on day 10 of pembrolizumab therapy), we followed the analysis method reported in the original study and used T-cell repertoire expansion determined by scTCR-seq as an immunologic surrogate of therapeutic response. Tumors were considered to harbor T-cell expansion when more than 30 T-cell clones increased in abundance in the on-treatment biopsy compared with the pretreatment biopsy (Supplementary Fig. S1A). Pretherapy breast tumors containing cancer cells expressing high levels of antigen-presenting machinery genes (HLA class I and II genes, TAP1, B2M) and IFN response genes (STAT1, IRF1) were associated with T-cell expansion after pembrolizumab (Fig. 1A). Moreover, genes associated with breast cancer cell survival and immune resistance (CCND1, ATF3, ERBB2/3, CTNND1, MUC1) were expressed at low levels in pretherapy samples that demonstrated T-cell expansion after pembrolizumab (Supplementary Fig. S1B). In addition, we confirmed that high expression of NEDD8 mRNA in pretherapy breast cancer cells was negatively associated with T-cell expansion upon pembrolizumab treatment (Fig. 1A; Supplementary Fig. S1B). Indeed, pretherapy patient tumors with T-cell expansion upon treatment contained cancer cells with significantly lower NEDD8 mRNA expression across subtypes of breast cancer (Fig. 1B). Moreover, tumors containing NEDD8-low cancer cells (red symbols) showed higher levels of T-cell expansion. Within this group, the level of T-cell expansion inversely correlated with the level of cancer intrinsic NEDD8 expression (Supplementary Fig. S1C).

Figure 1.

Pretherapy cancer intrinsic expression of NEDD8 mRNA negatively correlates with T-cell expansion upon pembrolizumab treatment. To examine cancer intrinsic genes that regulate immune activation, we analyzed published datasets from patients with breast cancer (BrCa) prior to and after one dose of pembrolizumab. A, Expression of mRNAs at the pseudo-bulk level in cancer cells is retrieved from pretherapy patient tumors and compared between patients with (n = 9) or without (n = 20) T-cell expansion after pembrolizumab using DESeq2. Statistical differences are compared using the Wilcoxon rank-sum test with Bonferroni P value adjustment. B, Expression of cancer intrinsic NEDD8 mRNA in breast cancer cells from tumors with (n = 9) or without (n = 20) T-cell expansion is shown. Each dot represents the average expression of NEDD8 mRNA in cancer cells from a patient’s tumor, and statistical differences are determined using the nonparametric Wilcoxon rank-sum test. C, Patient tumors are grouped according to the pretherapy NEDD8 mRNA levels (high, n = 20; low, n = 9) in cancer cells, and the expression of selected genes in CD8⁺ T cells in these tumors prior to and after one dose of pembrolizumab is shown. D, The expression (Exp) of selected genes in macrophages from tumors with pretherapy high or low NEDD8-expressing cancer cells prior to and after one dose of pembrolizumab is shown. The size of the circles represents the percentage of positive cells, and the color represents the average expression (Ave. Exp.). E, Control or NEDD8 KO MDA-MB-231 cells are treated with PBS or 50 ng/mL rhIFNA or rhIFNB. F, Pharmacologic inhibitors against JAK1/2 (Ruxolitinib, 1 μmol/L), STING (H-151, 2 μmol/L), or cGAS (G140, 2 μmol/L) are incubated with cells for 2 hours before the addition of rhIFNB. Soluble levels of CXCL10 or IL8/CXCL8 in culture supernatants are quantified using the LEGENDplex cytokine array. Each dot represents a biological replicate, and results are shown as mean ± SD, n = 3. Statistical differences are determined using the unpaired t test. Ctrl, control; E, expansion; ER, estrogen receptor; FC, fold change; HER2, human epidermal growth factor receptor 2; NE, no expansion; UMAP, Uniform Manifold Approximation and Projection for Dimension Reduction.

Next, we determined the expression of NEDD8 mRNA in breast cancer cells and stratified the patients into NEDD8-low or NEDD8-high expression groups according to the average expression levels in pretherapy tumors. We observed that patients with pretherapy NEDD8-low tumors demonstrated a significant upregulation of NEDD8 mRNA levels after pembrolizumab treatment (Supplementary Fig. S1D, P = 0.039), resulting in comparable on-therapy NEDD8 mRNA expression levels between the groups. NEDD8-low cancer cells in pretherapy tumors exhibited enhanced expression of inflammatory response and TNF/IFN response signatures (Supplementary Fig. S1E). Moreover, transcriptomic analysis of CD8⁺ T cells in pretherapy tumors identified elevated expression of genes related to cytotoxicity (GZMA, IFNG, TNF), IFN response (ISG15, IRF6), and immune checkpoints (PDCD1, TIGIT, LAG3) that were upregulated in the NEDD8-low tumors (Fig. 1C; Supplementary Fig. S2A). These differences in gene expression in NEDD8-low pretherapy tumors became more evident after pembrolizumab treatment (Fig. 1C). A similar transcriptomic profile was observed in CD8⁺ T cells if patients were grouped based on T-cell expansion (Supplementary Fig. S2B), suggesting that cancer-intrinsic NEDD8 levels could be used to identify potential therapy responders.

Because tumor-infiltrating myeloid cells play a pivotal role in the in situ expansion of T cells, we focused on the macrophage population and investigated whether their transcriptomic signatures differed in NEDD8-high or NEDD8-low patient tumors. Intratumoral macrophages from NEDD8-low pretherapy tumors showed reduced fatty acid metabolism, which remained low after pembrolizumab treatment (Supplementary Fig. S2C). Because high lipid metabolism contributes to the immunosuppressive function in myeloid cells (26, 27), we hypothesized that NEDD8-low cancer cells promoted the proinflammatory phenotype in myeloid cells upon PD-1 blockade. Indeed, macrophages in NEDD8-low tumors exhibited enhanced antigen-presentation capacity and IFN/IL1B-responsive genes (Supplementary Fig. S2C). In particular, genes associated with the IFN response (STAT1, IRF1/5, NFKBIA, CXCL9/10/11) and antigen presentation to CD8⁺ T cells (HLA-A/B/C, B2M, TAP1/2) were upregulated in macrophages after pembrolizumab if the pretherapy tumors contained NEDD8-low cancer cells (Fig. 1D). In contrast, HLA class II genes (HLA-DRA/DPA1/DPB1 and CD74) were the highest in macrophages after pembrolizumab treatment if the tumors showed high pretherapy NEDD8 expression (Fig. 1D).

To experimentally validate whether NEDD8-deficient human cancer cells could acquire a more inflammatory phenotype in vitro, we genetically removed the NEDD8 gene in MDA-MB-231 cells (Supplementary Fig. S2D) and found that the KO cells expressed higher HLA class I and class II molecules on the surface both at baseline and in response to stimulation with recombinant IFNA or IFNB (Supplementary Fig. S2E). Moreover, NEDD8-deficient human TNBC cells released significantly higher levels of CXCL10 and IL8/CXCL8 upon IFNB treatment (Fig. 1E), which are chemokines that are important for the recruitment of immune cells. Pharmacologic inhibition of either JAK1/2 (ruxolitinib) or STING (H-151) significantly impaired the induction of these proinflammatory chemokines by KO cells (Fig. 1F), indicating elevated cancer-intrinsic type I IFN signaling upon NEDD8 deletion.

Genetic deletion of protein neddylation elicits a curative response to PD-1/L1 blockade

Next, we sought to dissect the molecular mechanisms and response to ICB therapy by genetic ablation of the Nedd8 gene in murine EO771 breast cancer cells (Fig. 2A). Nedd8-deficient tumors showed similar growth compared with the control tumors (Fig. 2B). However, PD-1 blockade was ineffective in controlling the growth of control tumors but elicited potent antitumor immunity against Nedd8-deficient tumors, leading to complete responses (CR) in 75% of the mice (Fig. 2C) and significantly prolonged survival (Fig. 2D). Moreover, Nedd8 deletion in cancer cells resulted in a delay in the aggressive growth of the C1498 AML model in response to PD-1 blockade (Supplementary Fig. S3A–S3C). These data suggest that the loss of the Nedd8 gene in murine cancer cells did not directly impair the in vivo proliferative capacity of cancer cells but sensitized the cancer cells to immune-mediated cytotoxicity.

Figure 2.

Genetic deletion of the Nedd8 gene sensitizes murine tumors to ICB therapy in tumor-bearing mice. A, The Nedd8 gene in EO771 cells is deleted using CRISPR/Cas9, and the expression of the NEDD8 protein is assessed by Western blotting in two different cell passages. A representative image of two biological replicates is shown. Control (Ctrl) or Nedd8 KO EO771 (6 × 10⁵ cells) are injected s.c. in 100 μL FBS-free medium in the right flank of 6–10-week-old female C57BL/6NTac mice. Mice are treated with (B) rat IgG2a isotype control (2A3) or (C) 50 μg of αPD-1 antibody (RMP1-14) on days 7, 10, and 13 i.p., and the tumor growth is shown, n = 8 mice per group. D, Survival probabilities of mice treated as described above are calculated using Kaplan–Meier curves. Representative of two independent experiments is shown. Ctrl or Nedd8 KO EO771 cells are injected as above, and mice are treated i.p. either with (E) rat IgG2b isotype control (LTF-2) or (F) 200 μg of αPD-L1 antibody (10F.9G2) on days 7, 10, and 13. Tumor growth is monitored until the study endpoint, n = 8 mice per group. In the rechallenge study, tumor-free mice from the αPD-L1 study are injected s.c. with EO771 control cells (1.8 × 10⁶ cells per mouse) in the left flank, and tumor volumes are recorded regularly. G, Survival probabilities of mice treated as described above are calculated using Kaplan–Meier curves. Representative of two independent experiments is shown. Survival differences were calculated using Kaplan–Meier curves and a log-rank test (Mantel–Cox). ns, not significant.

Although the use of PD-1 blocking antibodies yielded improved survival rates in clinical trials (28), PD-L1 blockade induced less immune-related adverse events across cancer types (29). Given that targeting protein neddylation may introduce additional safety concerns (30), we sought to investigate whether the superior antitumor immunity in Nedd8 KO tumors could be relevant to PD-L1 blockade. Although the PD-L1 blocking antibody resulted in a 25% CR in mice bearing control tumors, this percentage was doubled in mice bearing the Nedd8-deficient tumors (Fig. 2E and F), leading to improved survival in these mice (Fig. 2G).

To study whether immune memory was formed in cured mice, we rechallenged the tumor-free mice after the PD-L1 blockade therapy with three times more EO771 cells than used in the initial implantation. All rechallenged mice remained tumor-free, regardless of the treatment group (Fig. 2E–G), indicating potent immune memory against EO771-derived tumor antigens.

PD-1 blockade drives the influx of inflammatory myeloid cells in Nedd8-deficient tumors

To comprehensively elucidate the intratumoral landscape controlled by the cancer intrinsic protein neddylation, we performed a combined analysis on size-matched EO771 tumors using scRNA-seq and single-cell transcriptomics on the CosMx platform (Fig. 3A; Supplementary Fig. S3D). To ensure sufficient materials could be collected, treatments were initiated on day 10, and tumors were collected 3 days after the last dose of PD-1 blockade.

Figure 3.

scRNA-seq reveals a distinct immune landscape governed by cancer-intrinsic protein neddylation. A, Schematic illustration of tumors collected for scRNA-seq and CosMx spatial transcriptomics analysis using treated control or Nedd8 KO EO771 tumors. B, Classification of 15 × 10⁴ single cells from 15 tumors and Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) of the transcriptionally defined clusters derived from the scRNA-seq analysis is shown. C, Differential gene expression analysis highlighting upregulated and downregulated pathways in cancer cells from Nedd8 KO or control EO771 tumors upon αPD-1 treatment, with four samples per group. The size of the circle represents the gene counts in the pathway. Red and blue colors represent upregulated or downregulated pathways in KO tumors treated with PD-1 blockade, respectively. D and E, A heatmap showing signature scores in (D) macrophages and (E) CD8⁺ T cells from control or Nedd8 KO tumors collected from mice treated with IgG isotype control or PD-1 blocking therapy. The color represents the level of the signature scores. F, Tumors are harvested from mice in (A), and frequencies of the CD80⁺IL12/23p40⁺ double-positive cells in CD11c⁺MHCII+ DCs or CD11b⁺ non-DC myeloid cells are shown in the tumors from different treatment groups (5–6 tumors per group). G, The percentage of CD8⁺ T cells expressing the intracellular cytotoxic protein granzyme B (GZMB) across different treatment groups is shown, with 4–6 tumors per group. Data are shown as mean ± SD, and an unpaired t test is used. H, EO771 Nedd8 KO cells (6 × 10⁵ cells) are injected s.c. into the right flank of immunocompetent mice in 100 μL FBS-free medium. Mice are then treated i.p. with an αCD8a depleting antibody (150 μg, clone 2.43, BioXcell) or an αCSF-1R depleting antibody (150 μg, clone AFS98, BioXcell), once per week, starting from 1 day before the injection of a rat IgG2a isotype control (2A3) or an αPD-1 antibody (70 μg, RMP1-14) on days 7, 10, and 13. Tumor growth is monitored and shown as mean ± SEM, with at least seven mice per group. Statistical tests were performed using two-way ANOVA. Ctrl, control; GOBP, Gene Ontology Biological Process; Inflam, inflammatory; Reg, regular; TAM; tumor-associated macrophage.

On average, 10,000 cells were sequenced from each sample in scRNA-seq, and at least 2,500 genes were detected per cell. Cell clustering identified eight cell populations (Fig. 3B; Supplementary Fig. S3E), and epithelial cells and macrophages were more abundant in EO771 tumors compared with other cell types (Supplementary Fig. S3F). However, the percentages of these cells were not influenced by Nedd8 deficiency nor the treatment. Because cancer cells showed overlapping transcriptomic markers with epithelial cells, we calculated CNV in the scRNA-seq dataset (Supplementary Fig. S3G). This approach facilitated the separation between malignant and nonmalignant cells in EO771 tumors (Supplementary Fig. S4A and S4B). In mice treated with the isotype control antibody, deletion of Nedd8 in cancer cells demonstrated enhanced regulation of immune responses and vasculature development but showed reduced cell cycle and DNA replication (Supplementary Fig. S4C). However, Nedd8-deficient cancer cells entered an inflammatory state with an enhanced ability to enable cell migration and cytokine response when treated with PD-1 blockade (Fig. 3C).

We then asked whether the immune landscape was remodeled by Nedd8 deletion in cancer cells. In macrophages from Nedd8-deficient tumors, PD-1 blockade increased the expression of antigen-presenting genes, as well as fatty acid metabolism, compared with macrophages from control tumors treated with PD-1 blockade, but did not modulate the inflammatory status (Fig. 3D). In contrast, deletion of Nedd8 in cancer cells reduced fatty acid metabolism in macrophages and enhanced their activation in tumors treated with the isotype control antibody (Fig. 3D). Furthermore, tumor-infiltrating macrophages in Nedd8-KO tumors treated with PD-1 blockade exhibited a high potential for MHC-I antigen presentation, as well as activation and inflammatory status (Fig. 3D), but showed low expression of genes associated with lipid metabolism (Apoe, Lipa, Lpl) and MHC-II antigen presentation (Cd74, H2 genes; Supplementary Fig. S4D and S4E).

Although percentages of CD8⁺ T-cell subsets were comparable among treatment groups (Supplementary Fig. S5A and S5B), gene signatures representing cytotoxicity and glycolysis were enhanced in CD8⁺ T cells in KO tumors treated with PD-1 blockade compared with treated control tumors (Fig. 3E). Although cytotoxicity genes (Gzmb, Pfr1, Tnf) and immune checkpoint genes (Ctla4, Havcr2, Lag3) were enhanced in CD8⁺ T cells, Tox, which is the master regulator for T-cell exhaustion (31, 32), was expressed at low levels when KO tumors were treated with PD-1 blockade (Supplementary Fig. S5C). In KO tumors treated with PD-1 blockade, natural killer cells showed enhanced expression of cytotoxic genes (Gzmc, Ifng, and Prf1), as well as markers associated with exhaustion (Ctla4, Lag3, and Cd96) compared with treated control tumors (Supplementary Fig. S5D). In addition, more than 60% of CD4⁺ T cells in EO771 tumors showed a signature for regulatory properties (Supplementary Fig. S6A and S6B) and demonstrated upregulated costimulatory functions in treated KO tumors, as compared with the control tumors upon PD-1 blockade (Supplementary Fig. S6C).

In accordance, additional experiments using flow cytometry confirmed the abundance of myeloid cells (as assessed by CD11b expression) in EO771 tumors, with no significant differences observed between groups (Supplementary Fig. S6D). However, Nedd8-deficient tumors showed a significant decrease in infiltrating CD11c⁺MHCII⁺ dendritic cells (DC) as compared with control tumors, regardless of ICB therapy (Supplementary Fig. S6E), whereas the non-DC myeloid cells were comparable in abundance to control tumors (Supplementary Fig. S6E). We did observe that myeloid cells in Nedd8-deficient tumors demonstrated higher immune-stimulatory potential, as indicated by the significantly increased proportions of CD80⁺IL12/23p40⁺ cells (Fig. 3F) and a marked decrease of macrophages expressing CD206, which is a marker for anti-inflammatory myeloid cells (Supplementary Fig. S6F). In line with the transcriptomics data (Supplementary Fig. S5C), tumor-infiltrating T cells did not change significantly in percentages (Supplementary Fig. S6G), but CD8⁺ T cells in Nedd8-deficient tumors upregulated the intracellular expression of granzyme B (Fig. 3G). To demonstrate the functional contribution of CD8⁺ T cells and macrophages, mice bearing Nedd8 KO tumors were infused with anti-CD8 or anti-CSF1R prior to PD-1 blockade. Removal of CD8⁺ T cells, but not macrophages, abrogated the antitumor efficacy of PD-1 blockade in mice bearing Nedd8 KO tumors (Supplementary Fig. S6H). Depletion of both CD8⁺ T cells and macrophages in ICB-treated KO tumors showed comparable tumor growth in mice as compared with tumors treated with the isotype control (Fig. 3H).

The intratumoral spatial architecture is orchestrated by neddylation

Because the spatial distribution of immune cells is known to fine-tune antitumor immunity, we performed single-cell spatial transcriptomics analysis in more than 500,000 cells using the CosMx platform, which detects 1,000 genes per cell (33). Cells were annotated using gene signatures validated in the scRNA-seq datasets derived from EO771 tumors of the same in vivo study. We determined that spatial transcriptomics recovered more fibroblasts and endothelial cells as compared with scRNA-seq (Supplementary Fig. S7A), and fibroblasts defined the outer boundary of the tumor tissues (Fig. 4A). Cancer cells were predominantly located in the tumor core (Fig. 4A), whereas macrophages were enriched at the invasive margins (Fig. 4B). In line with scRNA-seq analysis, CD4⁺ and CD8⁺ T cells were less abundant than macrophages in all groups of EO771 tumors (Supplementary Fig. S7A), and they were located across the tumor tissue (Supplementary Fig. S7B).

Figure 4.

Cancer intrinsic neddylation reshapes the spatial architecture of EO771 tumors. Size-matched EO771 control or Nedd8 KO tumors are harvested 3 days after the last dose of αPD-1 antibody or the isotype control. FFPE sections from the tumors are subjected to single-cell spatial transcriptomics analysis on the CosMx platform. A, fibroblasts (red) and cancer cells (blue) or (B) macrophages (cyan) and CD8⁺ T cells (orange) are visualized in EO771 tumors treated with PD-1 blockade. The white dashed line depicts the tumor core, and the red dashed line shows the outer tumor boundary. C, Infiltration scores of each immune cell type in EO771 tumors from different groups are shown as histograms. A higher infiltration score represents a closer position to the center of the tumor core. D, Co-localization among immune cells is calculated at the single-cell resolution. A higher coefficient [Coeff (red)] represents a closer proximity between two cells. E, Macrophages are shown on EO771 tumors from different groups, and cellular niches are depicted in colors. Gray dots represent macrophages that do not form a niche. Images in (A), (B), and (E) are generated from the same tumor treated with PD-1 blockade, and the spatial location of different immune cell types is visualized. A detailed analysis using CellChat is performed in macrophage niches, and strengths of receptor–ligand interactions (F) from CD8⁺ T cells to macrophages or (G) from macrophages to CD8⁺ T cells are shown. The size of the dots represents the percentage of expression in macrophage niches, whereas the color represents the average communication probability (Avg comm prob) between the receptor and the ligand.

Based on the smallest convex hull of the tumor core, we calculated the depth of immune infiltration in different groups. Although Nedd8 KO in cancer cells or PD-1 blockade of control tumors showed minimal effects on immune infiltration, PD-1 blockade of KO tumors substantially mobilized macrophages, CD8⁺ T cells, CD4⁺ T cells, and DCs toward the center of the tumors (Fig. 4C). To assess the spatial distribution of cell types, we quantified the co-localization coefficient among different cell types. Although T cells were found in proximity to cancer cells in untreated control tumors, regulatory T cells (Treg), macrophages, and fibroblasts were also abundant in the cellular neighborhood [Fig. 4D (top left)], which indicated an immunosuppressive milieu. Moreover, CD8⁺ and CD4⁺ T cells were distant from other immune cells in wild-type (WT), IgG-treated tumors, which could cause anergy due to the lack of costimulation and cytokines. PD-1 blockade or Nedd8 deletion in EO771 tumors induced a clear spatial–temporal change and increased the co-localization among immune cell types [Fig. 4D (top right and bottom left)]. However, fibroblasts were adjacent to immune cells, which could indicate that immune cells were primarily located at the periphery of tumor tissues (Fig. 4D). Compared with other groups, PD-1 blockade triggered distinct spatial reorganization among immune cell types in Nedd8 KO tumors. In particular, CD8⁺ T cells exclusively co-localized with macrophages and cancer cells but were distant from Tregs and fibroblasts [Fig. 4D (bottom right)]. Given that macrophages exhibited lower lipid metabolism and a more inflammatory phenotype in this group (Fig. 3D), the cross-talk between CD8⁺ T cells and macrophages could strengthen cytotoxicity against cancer cells in this group.

Because macrophages played a central role in KO tumors upon PD-1 blockade, we extracted the spatial distribution of macrophages in all groups of tumors and depicted cellular niches based on the distance among cells. Although macrophages were present in IgG-treated control and Nedd8 KO tumors, very few niches were identified in these tumors (Fig. 4E), which was in line with the relatively low co-localization among macrophages (Fig. 4D). In contrast, PD-1 blockade in control or KO tumors substantially enhanced the quantity and area of macrophage niches (Fig. 4E).

Next, we sought to elucidate the ligand–receptor interactions among cell types within the macrophage niches using CellChat. As shown in Fig. 4F, CD8⁺ T cells demonstrated a higher probability of interacting with CCR5 on macrophages through the release of CCL3/4/5 in Nedd8 KO tumors upon PD-1 blockade than in WT EO771 tumors upon PD-1 blockade. On the other hand, macrophages in the niches showed reduced communication probability with CD74 and CXCR6 on CD8⁺ T cells (Fig. 4G). Moreover, Tregs in macrophage niches in Nedd8 KO tumors upon PD-1 blockade exhibited less strength and probability of interacting with CD8⁺ T cells (Supplementary Fig. S7C) or macrophages (Supplementary Fig. S7D) than the equivalent cells in WT EO771 tumors upon PD-1 blockade.

Neddylation governs human TNBC cell sensitivity to lipid metabolism and glutathione peroxidase 4 inhibitors

Because existing neddylation inhibitors show inhibitory effects on immune cells (11), we sought to identify alternative functional pathways that were dependent on protein neddylation in human TNBC cells. This was achieved by employing the high-content Cell Painting assay, which measured >1,300 morphologic features at cellular and subcellular levels (34), in the control/NEDD8 KO isogenic MDA-MB-231 cell line pair. A set of 60 chemical compounds with well-defined targets (Supplementary Table S1) were included in the screen (Fig. 5A).

Figure 5.

Lipid metabolism and ferroptosis are alternative targets for protein neddylation in human TNBC cells. A, Sixty known compounds are added to a 384-well plate at varying concentrations, using MDA-MB-231 control (ctrl) or NEDD8 KO cells (9 × 10² cells per well). After 96 hours of incubation, automated cell painting and image acquisition are performed, followed by image analysis using CellProfiler software to identify individual cells and extract morphologic features. Nuclei counts are compared between control and NEDD8 KO cells treated with increasing concentrations of (B) ML210 or (C) simvastatin and lovastatin after 96 hours of incubation; each dot represents a technical replicate in the plate, mean ± SD. D, Heatmap visualization of 1,324 morphologic features in control and NEDD8 KO cells treated with ML210 (10 μmol/L), simvastatin (Simva; 0.93 μmol/L), lovastatin (Lova; 3.12 μmol/L), or pravastatin (Prava; 3.12 μmol/L) is generated using an unpaired t test (P < 0.01). Data shown are representative of three technical replicates. MDA-MB-231 control or NEDD8 KO cells were treated with (E) ML210 (3 μmol/L, 4 days) or (F) simvastatin (1 μmol/L, 3 days), and 0.1% DMSO was used as a control in both conditions. Morphologic features of 1 × 10³ cells per condition were analyzed using TEM. Green arrows, healthy mitochondria; red arrows, ferroptotic mitochondria; yellow arrows, lipid droplets. Representative images from three technical replicates are shown. Top in E: scale bar, 1 μm; Bottom in F: scale bar, 2 μm.

Among the 60 compounds, the selective glutathione peroxidase 4 (GPX4) inhibitor, ML210 (35), and two statin drugs (simvastatin and lovastatin) showed decreased potency on NEDD8-deficient cells (Fig. 5B and C). A detailed analysis of all 1,324 morphologic features extracted from the Cell Painting experiments demonstrated that ML210 (10 μmol/L), simvastatin (0.9375 μmol/L), and lovastatin (3.125 μmol/L) induced differential changes in 624, 506, and 531 cellular features, respectively, in control cells compared with the KO cells (Fig. 5D). Meanwhile, treatment with the hydrophilic statin drug pravastatin (3.125 μmol/L) induced only nine differential cellular features in control cells (Fig. 5D), possibly due to its lower potency on cancer cells (36). Among the subcellular organelles, the endoplasmic reticulum, mitochondria, and Golgi apparatus were the most affected (Supplementary Fig. S8A).

To visualize the cellular response to ML210 and simvastatin at subcellular resolution, we performed TEM using the isogenic cell line pair. When NEDD8-proficient cells were treated with ML210, we observed a loss of mitochondrial integrity (Fig. 5E), leading to the formation of ferroptotic bodies (Supplementary Fig. S8B). In contrast, genetic deletion of NEDD8 in MDA-MB-231 cells prevented the destruction of mitochondria in the presence of ML210 (Fig. 5E; Supplementary Fig. S8B). Treatment with simvastatin resulted in the accumulation of lipid droplets in the cytoplasm and mitochondria, which was not observed when NEDD8 was absent in MDA-MB-231 cells (Fig. 5F; Supplementary Fig. S8C). The lack of ferroptosis induction upon NEDD8 deletion was confirmed in an additional human TNBC cell line, HCC1937 (Supplementary Fig. S9). Therefore, we concluded that cancer intrinsic lipid homeostasis and ferroptosis could be explored as alternative targets for protein neddylation.

Disruption of lipid metabolism and peroxidation enhances the efficacy of PD-1 blockade

We tested whether simvastatin or RSL3, which is a covalent GPX4 inhibitor that has been shown to delay tumor growth in mice (37), could affect the response to PD-1 blockade in EO771 tumor-bearing mice (Fig. 6A). Although none of the treatments were effective in controlling tumor growth as a monotherapy, simvastatin or RSL3 increased the CR rates when combined with PD-1 blockade (42.9% or 44%, respectively; Fig. 6B and C). Mice treated with either of the combination therapies demonstrated significantly prolonged survival compared with the mice treated with the inhibitors alone (Fig. 6D and E). Moreover, combining simvastatin with PD-1 blockade significantly delayed the aggressive growth of C1498 AML tumors in immunocompetent mice (Supplementary Fig. S10A).

Figure 6.

Simvastatin or RSL3 enhances the response to PD-1 blockade in tumor-bearing mice. A, Treatment schedule of mice bearing control breast EO771 cancer cells. Tumor growth in mice treated with (B) simvastatin (8.2 mg/kg) or (C) GPX4 inhibitor RSL3 (10 mg/kg) together with 50 μg rat IgG2a isotype control (2A3) or PD-1 blockade (RMP1-14) is shown, with at least seven mice per group. D and E, Survival probabilities of the treated mice are calculated using Kaplan–Meier curves. A representative of two independent experiments is shown. For the rechallenge study, cured mice are injected i.p. with EO771 control cells (2 × 10⁶ cells per mouse). After 7 days, splenocytes are harvested for FACS analysis. F, Frequencies and gating strategy of CD86⁺ CD80⁺ and CD86⁺ MHCII⁺ cells (shown as mean ± SD, with at least three tumors per group) within the Ly6G⁻ myeloid cell population are compared between mice treated with αPD-1 alone and those receiving the combination therapy. G, Mice cured by αPD-1 alone (n = 3) or in combination with simvastatin (8.2 mg/kg, n = 4) or RSL3 (10 mg/kg, n = 3) are rechallenged with C1498 cells (3 × 10⁵ per mouse, s.c.). Tumor volumes are measured until the end of the study, and statistical significances are tested using two-way ANOVA (shown as mean ± SD). H, Survival differences of mice rechallenged with C1498 cells were calculated using Kaplan–Meier curves and a log-rank test (Mantel–Cox). APC, allophycocyanin; Ctrl, control; Macro, macrophage; Simva, simvastatin; Veh, vehicle.

Next, we investigated the immunologic memory in the spleens of cured mice by systemic rechallenge of EO771 cells in the abdominal cavity (Fig. 6A). Neither the percentages of lymphocyte subsets nor the expression of PD-1 or CD25 were significantly different in cured mice after rechallenge (Supplementary Fig. S10B and S10C). As compared with mice cured by PD-1 blockade, we observed elevated percentages of splenic myeloid cells coexpressing CD86 and CD80 or MHC-II (Fig. 6F) but not MHC-II⁺ or CD80⁺MHC-II⁺ myeloid cells (Supplementary Fig. S10D). Further analysis showed that this was due to changes in F4/80⁺CD11c^dim macrophages but not in CD11c^high DCs (Fig. 6F; Supplementary Fig. S10E).

Because rechallenge with EO771 cells failed to form tumors in cured mice (Fig. 2E–G), we tested whether different combination treatments (Supplementary Fig. S10F) could lead to protection against C1498 cells in cured mice. As shown in Fig. 6G, mice cured by simvastatin plus PD-1 blockade showed significantly better immune protection against C1498 rechallenge compared with mice cured by the anti–PD-1 monotherapy. RSL3 together with anti–PD-1 primed a potent immune protection and prevented C1498 tumor growth (Fig. 6G), leading to significantly prolonged survival in this group (Fig. 6H).

Discussion

A collection of previous studies has elucidated the intratumoral heterogeneity of TNBC using scRNA-seq to support the discovery of predictive biomarkers or therapeutic targets for ICB therapy in TNBC (15, 38, 39). Although these datasets establish an encyclopedia of tumor-infiltrating immune cells, illustrations of the spatial–temporal features of TNBC tumors were needed to reveal novel biological insights (40). For example, multiplexed immunofluorescent imaging (41) and imaging mass spectrometry (42) were employed to show the importance of interactions between cancer and immune cells in predicting treatment outcomes (42). Recently, Shiao and colleagues (43) combined scRNA-seq and CODEX spatial profiling to delineate the therapy response trajectories of radio-immunotherapy in patients with early-stage TNBC. With the development of single-cell spatial transcriptomics, it is now possible to depict the architecture of breast cancer at high resolution (44).

By analyzing patient datasets and performing genome-wide genetic screens (11), we recently validated the NEDD8 gene as a cancer-intrinsic barrier for efficient immune activation upon PD-1 blockade. However, the functional hierarchy and precise molecular network controlled by neddylation remain elusive. Here, we combine scRNA-seq and single-cell transcriptomics on the CosMx platform (33) to decode the immunologic landscape in control and Nedd8 KO murine tumors upon treatment with PD-1 blockade. EO771 tumors demonstrated distinct spatial features and cellular heterogeneity, with cancer cells located at the core of the tumors, whereas fibroblasts were enriched at the invasive margin. Myeloid cells were abundant in EO771 tumors and were mobilized by PD-1 blockade to the tumor tissues.

The breast cancer microenvironment presents distinct myeloid cell profiles, which contribute functionally to the development of primary and acquired resistance to ICB therapy (45). In the EO771 model, Yofe and colleagues (46) showed the spatial–temporal role of regulatory myeloid cells in the establishment of metastatic niches. When comparing scRNA-seq datasets from patient and murine tumors, we identified several overlapping changes that were associated with cancer-intrinsic neddylation. For example, cancer cells demonstrated a more inflammatory phenotype in NEDD8-low patient tumors and Nedd8-deficient murine tumors, which coincided with reduced lipid metabolism potential in macrophages and enhanced T-cell expansion and expression of cytotoxic genes. In the future, it would be relevant to dissect whether NEDD8-low patient tumors support the expansion of neoantigen-specific T-cell clones.

The metabolic status of cancer cells is known to regulate antitumor immunity (47). In particular, macrophages can take up exogenous fatty acids from cancer cells and upregulate intracellular lipid metabolism (48). In our study, we found that Nedd8-deficient cancer cells demonstrated the ability to modulate vasculature and angiogenesis, but PD-1 blockade was required to potentiate KO cancer cells to attract myeloid cells through the release of chemokines. Furthermore, we revealed that the intensified interplay between CD8⁺ T cells and inflammatory macrophages could be key to the maximized tumor destruction upon PD-1 blockade in Nedd8-deficient tumors. Therefore, the association between cancer-intrinsic neddylation, myeloid cell status, and TCR clonal expansion/exhaustion should be further investigated using additional patient datasets or functional assays.

Upon NEDD8 deletion, we found that human cancer cells upregulated HLA class I and II molecules and released proinflammatory chemokines CXCL10 and IL8/CXCL8 at higher levels upon treatment with IFNB. This notion was further supported by neddylation-deficient cancer cells in mouse tumors, in which transcripts associated with immune cell migration and inflammatory response were elevated upon PD-1 blockade. These observations suggest that genetic deletion of neddylation could potentiate the immune signaling network in cancer cells, supporting the infiltration of proinflammatory immune cells. Further experiments are warranted to reveal the precise functional role of chemokines, as well as antigens presented by MHC class II molecules in antitumor immunity, which remain to be elucidated.

The cyclic GMP-AMP synthase–stimulator of interferon genes (cGAS/STING) pathway is an intracellular DNA sensing machinery and is crucial in eliciting potent antitumor immunity through the activation of the type I IFN response (49). The neddylation inhibitor pevonedistat has been shown to dampen cGAS/STING activation in response to viral challenge (50), but this inhibitor could stabilize nuclear cGAS, which induces an IFN response in cancer cells (51). In our study, inhibition of STING, but not cGAS, abolished the enhanced release of proinflammatory chemokines from neddylation-deficient cancer cells. Although this suggests a functional link between neddylation and cGAS/STING, additional studies using genetic deletion of cGAS, as well as evaluating the phosphorylation of signaling proteins, for example, pSTING/pTBK1/pIRF3, could help clarify the mechanistic interplay between the pathways.

Protein neddylation is a process of conjugating NEDD8 to its target enzymes and substrates, and it is a crucial posttranslational regulatory mechanism that sustains tumor growth (12). Pharmacologic inhibition of the NEDD8-activating enzyme by pevonedistat (52) has demonstrated antitumor activity in patients with cancer (53–57). However, the combination of pevonedistat and azacitidine failed to provide clinical benefits in patients with myeloid malignancies in the phase III PANTHER trial (NCT03268954; ref. 58). In preclinical models, pevonedistat upregulated the expression of PD-L1 on glioblastoma cells (59) and increased the response rates to ICB therapy (59, 60). However, we showed that neddylation inhibitors hampered the activation of lymphocytes (11), and toxicity has been noted in patients at high doses (30), potentially due to neddylation-independent mechanisms (11).

Due to challenges in the clinical development of neddylation inhibitors, we sought to identify known compounds that target pathways regulated by protein neddylation. Our phenotypic screen demonstrated that statin drugs and the GPX4 inhibitor preferentially inhibited NEDD8-competent cells and increased the response rates with PD-1 blockade in mice. High-resolution TEM revealed that NEDD8 protein was required to interrupt mitochondrial integrity by simvastatin or the GPX4 inhibitor in human TNBC cells. GPX4 is a lipid-repair enzyme that prevents lipid peroxidation and therefore protects cancer cells from ferroptosis (37). A recent study showed that GPX4 inhibitors enhanced CD8⁺ T cell–mediated cancer cell killing through the downregulation of cystine transporters when cancer cells encountered IFNG released by activated T cells (61). Although the NEDD8 KO cancer cells gained resistance to GPX4 inhibitors in vitro, ferroptosis may be enhanced in tumor-bearing mice due to the enhanced cancer immunogenicity and the release of IFNG. Moreover, statin drugs have been shown to induce ferroptosis in cancer cells through inhibition of the mevalonate pathway, leading to downregulation of GPX4 protein (62). However, the molecular mechanisms underlying the regulation of lipid metabolism and ferroptosis by protein neddylation remain to be functionally validated.

Although our study mainly employs the EO771 and C1498 syngeneic mouse models, previous reports have demonstrated the synergistic effects between statin drugs and immunotherapy in other mouse models (63), as well as the potential beneficial effects in ICB-treated patients (64, 65). Because these compounds could manipulate several pathways in cancer cells, further studies are warranted to reveal the contribution of neddylation-independent properties of these drugs in combination with PD-1 blockade.

Supplementary Material

Figure S1

Supplementary Figure S1. Pre-therapy NEDD8 mRNA levels in cancer cells are associated with T-cell expansion.

Figure S2

Supplementary Figure S2. Neddylation in cancer cells regulates immune cell phenotype.

Figure S3

Supplementary Figure S3: Characterization of tumor-infiltrating cells using scRNAseq.

Figure S4

Supplementary Figure S4. Nedd8 deletion modulates gene expression in EO771 cancer cells and infiltrating macrophages.

Figure S5

Supplementary Figure S5. Characterization of tumor-infiltrating T and NK cells in EO771 tumors using scRNA-seq.

Figure S6

Supplementary Figure S6. Phenotypes of tumor-infiltrating immune cells are modulated by Nedd8-deficiency in EO771 cancer cells.

Figure S7

Supplementary Figure S7. Characterization of immune cell infiltration in EO771 tumors using CosMX single-cell spatial transcriptomics.

Figure S8

Supplementary Figure S8. Simvastatin-treated MDA-MB-231 human TNBC cells show similar phenotype in TEM as the NEDD8 deletion.

Figure S9

Supplementary Figure S9. Simvastatin-treated HCC1937 human TNBC cells show similar phenotype in TEM as the NEDD8 deletion.

Figure S10

Supplementary Figure S10. Assessment of combinatorial effects of simvastatin or RSL3 with PD-1 blockade in C1498 or EO771 tumor-bearing mice.

Supplemental Table S1

Supplemental Table S1: Compounds for the Cell Painting screen on control and NEDD8 KO MDA-MB-231 cells.

Supplemental Table S2

Supplemental Table S2: Antibodies and proteins.

Supplemental Table S3

Supplemental Table S3: Other reagents.

Supplemental Table S4

Supplemental Table S4: CRISPR RNA sequences.

Data Availability

The raw scRNA-seq and CosMx single-cell spatial transcriptomics results of mouse control or Nedd8 KO tumors treated with αPD-1 or the isotype control have been made available at the Sequence Read Archive and Gene Expression Omnibus at NCBI (RRID: SCR_005012). The accession numbers are PRJNA1419176 (scRNA-seq) and GSE318797 (CosMx), respectively.

Code Availability

Codes associated with this work can be found on GitHub through the following link: https://github.com/MaoLab-UU/2026-CIR-NEDD8-PAPER.

Authors’ Disclosures

Y. Mao reports grants from SciLifeLab, the Swedish Foundation for Strategic Research, the Swedish Cancer Society, and the Swedish Research Council during the conduct of the study, as well as grants from the Novo Nordisk Foundation, personal fees from Quotient Therapeutics, and other support from AstraZeneca outside the submitted work. No disclosures were reported by the other authors.

Authors’ Contributions

I. Papakyriacou: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. Y. Lu: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. M. Jarvius: Formal analysis, validation, investigation, methodology, writing–review and editing. M. Lapins: Data curation, software, formal analysis, validation, writing–review and editing. J. Carreras-Puigvert: Resources, data curation, software, formal analysis, supervision, investigation, methodology, writing–review and editing. Y. Mao: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.