Nasopharyngeal carcinoma cells promote regulatory T cell development and suppressive activity via CD70-CD27 interaction

Article https://doi.org/10.1038/s41467-023-37614-6 Nasopharyngeal carcinoma cells promote regulatory T cell development and suppressive activity via CD70-CD27 interaction Lanqi Gong 1,2,10,J i eL u o1,10, Yu Zhang3,4,5,10,Y u m aY a n g1, Shanshan Li2, Xiaona Fang1, Baifeng Zhang 1,J i a oH u a n g1, Larry Ka-Yue Chow 1, Dittman Chung1,J i n l i nH u a n g1, Cuicui Huang1,2,Q i nL i u1,2,L uB a i1,2,Y u e nC h a kT i u1, Pingan Wu6, Yan Wang7, George Sai-Wah Tsao8, Dora Lai-wan Kwong1,2, Anne Wing-Mui Lee 1,2,9,W e iD a i 1,2 &X i n - Y u a nG u a n1,2,4,5,9 Despite the intense CD8+ T-cell infiltration in the tumor microenvironment of nasopharyngeal carcinoma, anti-PD-1 immunotherapy shows an unsatisfactory response rate in clinical trials, hindered by immunosuppressive signals. To understand how microenvironmentalcharacteristics alter immune home- ostasis and limit immunotherapy efficacy in nasopharyngeal carcinoma, here we establish a multi-center single-cell cohort based on public data, containing 357,206 cells from 50 patient samples. We reveal that nasopharyngeal carci- noma cells enhance development and suppressive activity of regulatory T cells via CD70-CD27 interaction. CD70 blocking reverts Treg-mediated suppression and thus reinvigorate CD8+ T-cell immunity. Anti-CD70+ anti-PD-1 therapy is evaluated in xenograft-derived organoids and humanized mice, exhibiting an improved tumor-killing efficacy. Mechanistically, CD70 knockout inhibits a collective lipid signaling network in CD4+ naïve and regulatory T cells invol- ving mitochondrial integrity, cholesterol homeostasis, and fatty acid meta- bolism. Furthermore, ATAC-Seq delineates that CD70 is transcriptionally upregulated by NFKB2 via an Epstein-Barr virus-dependent epigenetic mod- ification. Our findings identify CD70+ nasopharyngeal carcinoma cells as a metabolic switch that enforces the lipid-driven development, functional spe- cialization and homeostasis of Tregs, leading to immune evasion. This study also demonstrates that CD70 blockade can act synergistically with anti-PD-1 treatment to reinvigorate T-cell immunity against nasopharyngeal carcinoma. Nasopharyngeal carcinoma (NPC) is an epidemiologically pre- valent malignancy in Asia and North Africa, intensively driven by chronic Epstein-Barr virus (EBV) infection. In the past decade, the age-standardized NPC mortality has been continuously reduced from 0.93 to 0.86 1, as a result of regional economic growth, improved healthcare awareness, large-scale population screening of EBV, and optimization of chemoradiotherapy, but the survival and life quality of patients with advanced and treatment-resistant Received: 11 May 2022 Accepted: 24 March 2023 Check for updates A full list of af filiations appears at the end of the paper. e-mail: [email protected] Nature Communications| (2023) 14:1912 1 1234567890():,; 1234567890():,; NPC have not progressively improved 2,3. In our most recent ran- domized clinical trials at the Queen Mary Hospital and The Uni- versity of Hong Kong (HKU)-Shenzhen Hospital, more than 30% of the most advanced locoregional NPC patients relapsed and pro- gressed despite receiving the best chemoradiotherapy available 4. Therefore, recurrence, metastasis, and chemoradiotherapy resis- t a n c er e m a i ns i g n ificant challenges in the clinical management of NPC. The NPC microenvironment is a highly heterogenous ecosystem infiltrated with various effector subpopulations, including cytotoxic T cells, helper T cells, memory B cells and plasma cells, collectively caused by chronic EBV infection and locoregionally enriched lym- phoid structures 5. Thus, for patients with advanced and chemoradiotherapy-resistant NPC, immunotherapy has been theo- retically considered as a promising strategy that mobilizes in situ and peripheral effector cells to combat tumor progression. Nevertheless, recent clinical trials have demonstrated that anti-PD-1 monotherapy using pembrolizumab, nivolumab, and Camrelizumab, only gener- ated a <3% complete response rate and a < 35% partial response rate in NPC patients6–8. Hence, there is an unmet clinical need that should be addressed by characterizing vital tumor microenvironment (TME)-related factors contributing to immunotherapy resistance, and evaluating how to enhance the immunotherapy response by targeting these factors. Since 2020, the cellular constituents and functional dynamics of the NPC microenvironment have been initially identi fied and char- acterized by single-cell RNA sequencing (scRNA-seq) at multiple centers9–12. In the NPC microenvironment, many tumor-in filtrating effector T cells are highly speci fic to tumor and EBV antigens 13,14. However, single-cell analysis has shown that these anti-tumor T cells are in disturbed immune homeostasis, exhibiting defective prolifera- tion and cytotoxicity5. Such disrupted homeostasis might be modu- lated by various inhibitory signals, such as TGF- β,I L - 1 0 ,I L - 3 5 ,a n d cyclic adenosine monophosphate (cAMP), from abundant effector Tregs (eTregs) in the TME 5,15. In noncancerous in flammatory naso- pharyngeal (INP) tissues, only a minor fraction of Tregs is present to maintain immune homeostasis, whereas the TME contains a much higher proportion of Tregs, as tumor cells might recruit peripheral naïve Tregs (nTregs) and activate nTregs and resting Tregs (rTregs), which in turn facilitate the NPC cells to escape from immune surveillance10,15,16. The development and suppressive function of Tregs in the NPC microenvironment is a dynamic and context-speci fic pro- cess that has yet to be insuf ficiently investigated. The major obstacle for anti-PD-1 immunotherapy is that therapeutically expanded T cells can rapidly become dysfunctional due to in situ immunosuppressive signals unaffected by such treatment, leading to de novo and required resistance17,18. This phenomenon might explain the unsa- tisfied response rate of anti-PD-1 immunotherapy even in immune- inflamed NPC tumors. Therefore, targeting immunosuppressive sub- types, primarily FOXP3+ Tregs, might promote anti-tumor immunity and enhance the efficacy of PD-1 blockade16,19,20. In this work, to better understand how Tregs develop and exert suppressive functions in the NPC microenvironment from an inte- grated multi-omic perspective, we show that CD70+ NPC cells can enhance Treg development, homeostasis, and suppressive activity via lipid metabolism reprogramming. Active lipid metabolism was indis- pensable for the downstream immunosuppressive effect of CD70- CD27 signaling in Tregs. Therapeutic inhibition of CD70 overcomes Treg immunosuppression, enhances anti-tumor immunity, and synergizes with PD-1 blockade in pre-clinical animal models. Further- more, the EBV-NFKB2-CD70 axis is identified as of high importance in regulating immune evasion during NPC progression. Taken together, these results provide a molecular and pre-clinical basis for optimizing first-line immunotherapy regimens and reducing mortality in NPC patients. Results Increased intratumoral Treg abundance and activation are regulated by NPC-mediated interaction To comprehensively investigate the developmental and functional dynamics of intratumoral Tregs affected by NPC cells at single-cell resolution, we established a large-scale cohort containing 189,750 T cells clustered into 41 subtypes from 36 NPC tissues, 10 paired NPC peripheral blood samples, and 4 INP tissues, originated from 3 NPC studies9–11 (Fig. 1A, S1A, and S1B). Within the T-cell subpopulations, we identified three FOXP3+ Treg subtypes, namely nTregs, rTregs, and eTregs, with distinct transcriptome pro files15 (Supplementary Fig. 1c and Supplementary Data 1). The single-cell trajectory analysis inferred two developmental lineages of CD4+ T cells, where naïve T cells either differentiated into helper T cells or eTregs, transcriptionally driven by BATF and FOXP3 (Figs. 1b, 1c and Supplementary Fig. 1d). To understand how the lineage commitment of CD4 + T-cell sub- types was shifted in the TME, we first quanti fied the normalized abundance of each subtype across patients. Inflammatory tissues and NPC peripheral blood samples harbored enriched fractions of CD4+ naïve T cells and SELL + nTregs, whereas in the TME, these fractions were limited (Fig.1d). Among the differentiated CD4+ T-cell subtypes, rTregs and eTregs exhibited a TME preference, while follicular helper T (TFH) cells and central memory T (TCM) cells were more enriched in INP tissues (Fig. 1d and Supplementary Fig. 1e). Then, to con firm that the homeostasis between Tregs and naïve CD4+ T cells was impaired in the TME, we established two multivariate linear regression models using eTreg and naïve T cell-speci fic signatures to compute the Treg immunosuppressive activity and T-cell naïveness based on NPC bulk RNA-seq cohorts (Supplementary Figs. 1f–i). Based on the functional modules, we illustrated that a high Treg and low naïve T cell infiltration was a malignant hallmark in NPC patients and associated with inferior prognosis (Figs. 1e, f). CIBERSORTx deconvolution based on scRNA- seq signatures further validated that the higher intratumoral in filtra- tion and suppressive activity of Tregs collectively contributed to worse prognosis, but were not associated to other clinical parameters in NPC patients (Supplementary Figs. 1j–l and Supplementary Table 1). As the most developed subtype in the TME, eTregs had the highest expres- sion of immunosuppressive signatures, such as CTLA4, LAYN and TNFRSF4, as well as the most robust cytokine communication, meta- bolism and T-cell suppression activities, collectively induced by STAT, TGF-β and IFN signaling21 (Figs. 1g, h). In contrast, nTregs and rTregs, modulated by canonical WNT signaling22, exhibited minimal regulatory effects on the activation, proliferation and apoptosis of T cells (Fig.1h). Considering an eTreg-polarized CD4+ T-cell landscape in NPC, we hypothesized that increased intratumoral Treg abundance and sup- pressive activity might be collectively caused by tumor-mediated CD4+ naïve T cell-to-Treg development and Treg activation. Therefore, to delineate the tumor-dependent mechanism by which Tregs developed and activated in the TME, we established in vitro co-culture systems where CD4+naïve T cells were either directly or indirectly co-cultured with the NPC cell line C666 or normal nasopharyngeal epithelial (NPE) cell lines, NP460 and NP69 (Fig. 1I). Compared to induced Treg development from CD4+ naïve T cells in the absence of tumor cells and the transwell-based co-culture system (Supplementary Figs. 1m,n), the direct co-culture system demon- strated that C666 cells enhanced polarization from CD4+ naïve T cells to FOXP3+ Tregs and upregulated the vital eTreg marker, CTLA4 23 (Figs. 1j–l). The Treg-polarizing and activating effect of C666 cells was further validated by qRT-PCR of a selected gene panel related to lineage determination and functionality of Tregs (Fig. 1m), and by ELISAs of secreted immunosuppressive factors in the co-culture sys- tem, including IL-10, TGF-β, and adenosine (Fig. 1n). Altogether, tran- scriptome analysis and functional assays preliminarily suggested that the development and activation of Tregs in NPC were regulated by cell- cell contact instead of cytokine communication. Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 2 Tumor-restricted CD70 is correlated to Treg abundance and activation via interacting with CD27 To further identify the exact mode of action occurring among CD4+ naïve T cells, Tregs, and NPC cells, we used two well-established pro- grams, CellPhoneDB24 and CellChat25, that evaluated the strength and direction of cell-cell interactions based on scRNA data. CD70-CD27 signaling was one of the most predominant interactions between NPC cells, naïve CD4+ T cells and Treg subtypes and occurred in a CD70/ CD27 expression-dependent manner (Figs. 2A, B). Although other interactions that have been reported associated with Treg develop- ment and functions, including TIGIT26, CD22627, 4-1BB28 and LGALS929, were identified, they did not appear to be upregulated in both CD4+ 0 2 4 6 8 −5 0 5 10 −5 0 5 10 UMAP_1 UMAP_2 0 5 10 15 20 PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 0.27% Q3 46.96% PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 0.02% Q3 28.29% PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 0.84% Q3 67.53% PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 0.04% Q3 29.08% PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 2.15% Q3 31.79% PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 0.15% Q3 36.15% UMAP_1 UMAP_2 - 505 1 0 -5 0 5 10 189,750 T cells with 41 subtypes C1 C1-CTLA4 C2-FOXP3 C3-SELL CTLA4 g JAK-STAT Signaling KEGG TGF-β signaling Cell cycle Notch signaling TCR signaling Hallmark GO:BP Cytokine interaction Chemokine signaling C1 C2 C3 Antigen processing with C666 Transwell co-culture in 0.3 cm2 inserts with C666 with NP460 with NP460 with NP69 with NP69 Q2 2.91% Q2 15.41% Q2 2.24% Q4 50.66% FOXP3 IL2RA CTLA4 LAYN SOX4 MKI67TNFRSF4 TNFRSF9 ICOS IL10 C666NP460NP69 Q4 60.79% IL6 STAT3 signaling IL2 STAT5 signaling Wnt signaling Fatty acid metabolism Glycolysis IFNα response IFNγ response Cholesterol homeostasis −2 −1.5 −1 −0.5 0 0.5 1 C1 C2 C3 −2 −1.5 −1 −0.5 0 0.5 1 C4-AQP3 C5-LEF1 C6-TCF7 C7-SELL C8-IL7R C9-FOS C10-AIF1 C11-ANXA1 C12-GPR183 C13-IFI44L C14-CXCL13 C15-HSPH1 C16-IFIT3 C17-KLRB1 C19-TOX C20-HSPA1A C21-IFIT1 C22-RSAD2 C23-TNFRSF9 C24-HAVCR2 C25-ZNF683 C26-ITM2C C27-EOMES C28-CXCR4 C29-CCL4 C30-GZMK C31-FGFBP2 C32-XCL1 C33-NKG7 C34-CX3CR1 C35-TYROBP C36-FCGR3A C37-FCER1G C38-GNLY C39-KLRG1 C40-STMN1 C41-MKI67 C2 C3 FOXP3+ fraction (%) C1:eTreg C2:rTreg C3:nTreg Q4 68.64% Q2 2.49% Q4 29.15% Q2 1.32% Q4 70.36% Q2 1.53% Q4 51.24% T cell activation T cell homeostasis Cytokine production T cell differentiation Negative regulation of T cell proliferation T cell migration TCR signaling Positive regulation of T cell apoptosis −1 −0.5 0 0.5 1 C1 C2 C3 z-score z-score z-score C4 C5 C7 C8 C10 C11 C12 C17 C14 C18 C19 C20C21 C22 C23C24 C25 C26 C27 C28 C29 C30 C31 C33C34 C36 C35 C37 C38 C39 C40 C41 C32 C15 C6 C16 C13 CD4+ T Cells CD8+ T Cells C9 C18-CCR7 Proliferating 02 0 4 0 6 0 0.0 0.2 0.4 0.6 0.8 1.0Progression-free survival Suppressive score-High (n=44) Suppressive score-Low (n=44) Logrank p=0.038 Time (months) 4 6 8 10 12 14 16 Normal Tumor 4 5 6 7 8 Normal Tumor Treg suppressive score T cell naive score a d e h f k m Influence CD4+ T cells via cytokine communication Influence CD4+ T cells via contact-based interaction TNFRSF4 More suppressive More helper Pseudotime Developmental trajectory of CD4+ T cells 0 2 4 6 0 5 10 15 FOXP3 expression Pseudotime r = 0.59, P < 2.2e-16 eTreg rTreg nTreg c C5_naive C6_naive C7_naive Naive T CD4+ naive T cells → naive/resting Tregs → effector Tregs b LAYN SELL transwell co-culture in 0.3 cm2 inserts TGFB1 Normalized fold change Direct co-culture in 1.9 cm2 wells PE-FOXP3APC-CLTA4 PE-FOXP3APC-CLTA4 0 20 40 60 80 0 5 10 15 20 FOXP3+/CTLA4+ fraction (%) 0 20 40 60 80 0 5 10 15 20 IL-10 l Direct co-culture in 1.9 cm2 wells C666NP460NP69 FOXP3+ fraction (%) FOXP3+/CTLA4+ fraction (%) C666NP460NP69 C666NP460NP69 60 40 20 0 Concentration (pg/mL) 800 600 400 200 0 TGF-β Adenosine C666 NP460 NP69n 0 2 4 6 8Concentration (nmol/mL) C666 NP460 NP69 Normal PBMC Tumor -2 -1 0 1 2 Normalized T cell fracction P = 0.013 C1: Effector Tregs -1 0 1 2 Normal PBMC Tumor Normalized T cell fracction C2: Resting Tregs -2 -1 0 1 2 Normalized T cell fracction Normal PBMC Tumor C3: Naive Tregs C5, C6, C7: CD4+ naive T cells -1 0 1 2Normalized T cell fracction Normal PBMC Tumor j P = 2.140E-7 P = 5.631E-6 P = 0.007 P = 0.252 P = 3.468E-4 P = 0.029 P = 8.131E-5 P = 5.636E-6 P = 3.689E-5 P = 2.024E-5 P = 2.249E-5 P = 0.020 P = 0.003 P = 3.314E-6 P = 3.361E-5 P = 1.860E-5 P = 1.286E-5 P = 5.769E-5 P = 1.161E-4 P = 3.432E-5 P = 1.547E-4 P = 8.831E-4 P = 6.848E-4 P = 0.019 P = 0.004 P = 7.671E-5 P = 3.912E-5 P = 4.47E-4 P = 0.008 P = 0.005 P = 0.019 P = 0.004 P = 0.009 P = 0.040 P = 0.031 P = 0.013 P = 0.022 P = 0.138 P = 0.191 P = 7.423E-4 P = 0.002 P = 4.739E-4 P = 0.001 P = 4.946E-5 P = 9.448E-5 i Whole blood of healthy donors PBMCs CD4 + naive T cells Magnetic isolation Ficoll-paque Direct co-culture in a 1.9 cm2 24-well plate indirect co-culture in a 0.3 cm2 transwell insert Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 3 naïve T cells and Treg subtypes, nor signi ficantly strengthed in the eTregs in the NPC microenvironment (Supplementary Fig. 2a–d). Thus, CD70-CD27 signaling was identi fied as the only signi ficant contact interaction occurring among CD4+ naïve T cells, Tregs, and NPC cells, thus with the potential to regulate de novo Treg development and suppressive activity in the TME. CD70-CD27 signaling is a co-stimulatory pathway that has been shown to promote Treg proliferation 30,31, but the exact role and molecular mechanism of CD70-CD27 signaling in the development and activation of tumor-infiltrating Tregs remain inexplicit. Thus, we initi- ally analyzed the spatial proximity among NPC cells, Tregs and CD70 expression based on Visium data, where the integrated NPC scRNA-seq data was used as a reference to infer tumor and Treg fractions via anchor-based deconvolution32 and cell2location 33.S p a t i a l l y ,T r e g s were not only highly co-localized with NPC cells, but also were sig- nificantly correlated to CD70 expression (Figs.2c, 2d, Supplementary Figs. 2e,f), indicating an active role of CD70+ NPC cells in modulating proximal Tregs. We further examined CD70 expression on NPC cells isolated from fresh endoscopic biopsies via flow cytometry, showing that approximately 60% of EPCAM+ tumor cells were CD70+, whereas only 10% CD70+ cells in EPCAM- in filtrating immune/stromal cells (Fig. 2e). Moreover, IHC and IF staining exhibited significantly higher CD70 expression and FOXP3+/CTLA4+ eTreg in filtration in primary NPC tissues than in non-malignant nasopharyngeal lymphoid tissues (Figs. 2f, g and Supplementary Fig. 2g). Although minor fractions of effector T cells, B cells and dendritic cells transiently expressed CD70 upon activation34, single-cell analysis confirmed that CD70 expression in the TME was highly restricted to NPC cells (Fig. 2h). We stratified patients in the integrated scRNA-seq cohort into CD70-high and CD70-low groups based on median CD70 expression, and showed that CD70-high patients harbored enriched fractions of rTregs and eTregs, but decreased fractions of nTregs and CD4+ naïve T cells (Fig. 2i). Such phenomenon was corroborated by qRT-PCR on primary NPC biopsies, showing higher expression of Treg lineage and activation markers,FOXP3, IL2RA and CTLA4 in CD70-high patients (Supplementary Fig. 2h), who was stratified by the above flow cytometry result (Fig. 2e). Meanwhile, we found that the negative correlation between T cell cytotoxicity/proliferation and the Treg fraction was evident only in CD70-high patients (Fig. 2j), implying a feedback loop among CD70, Tregs and T-cell immunity. In bulk RNA-seq cohorts, CD70 expression was consistently upregulated in NPC patients and associated with inferior survival (Figs. 2k, l and Supplementary Fig. 2i). We further revealed that the CD70+ fraction in the C666 cells was 3-fold higher than that of NP460 and NP69 cells (Fig. 2m), thus inducing higher soluble CD27 (sCD27) cleaved from the cell surface of co-cultured CD4+ naïve T cells upon CD70-CD27 binding35 (Fig. 2N). In the time-resolved transcriptome analysis of induced CD4 + naïve T cell-to-Treg differentiation, the computed Treg suppressive score increased proportionally toFOXP3, CTLA4,a n d CD27 expression, whereas the naïve score exhibited an opposite trend during this process (Fig. 2o). In the bulk RNA-seq cohort, we determined that the CD70 expression was positively cor- related with the suppressive score and negatively correlated with the naïve score (Fig. 2p). These results implied a Treg-polarizing and activating effect of CD70+ NPC cells, but it might be dually dependent on the CD70+ fraction in NPC cells and the CD27+ fraction in CD4+ naïve T cells and Tregs. Since NPC is a unique EBV+ head and neck squamous cell car- cinoma (HNSCC), we later explored whether the above-predicted feedback loop between CD70+ tumor cells and Tregs existed in other human papillomavirus+ (HPV+) HNCs. By analyzing two independent HNSCC scRNA-seq cohorts36,37 (Supplementary Figs. 2m, n), we found that rTregs and eTregs were not signi ficantly infiltrated in the HPV+ HNSCC microenvironment, compared to the normal tonsil micro- environment (Supplementary Fig. 2o). In addition, CD70 expression in HNSCC cells was minimal and was not correlated to worse prog- nosis in HNC patients either (Supplementary Figs. 2p, q). These results supported that CD70-mediated Treg immunosuppression was an NPC-speci fic characteristic, and might not contribute to tumor progression or immunotherapy failure in HPV+ HNSCC patients. CD70 knockout in NPC cells alleviates immunosuppression by inhibiting Treg development and functionality To elucidate the direct effect of CD70 on Treg development and activation, we treated CD4+ naïve T cells with active recombinant CD70 protein. The agonist treatment upregulated FOXP3 and CTLA4 expression in CD4+ naïve T cells and increased secretion of immuno- suppressive factors through CD70-CD27 interaction (Figs. 3a–da n d Supplementary Fig. 3a). Subsequently, we performed CRISPR- mediated knockout of CD70 (CD70-KO) in C666 cells, which did not affect PD-L1 expression or cell proliferation in C666 cells (Supple- mentary Figs. 2k,l). However, CD70-KO signi ficantly reduced Treg development and activation from CD4+ naïve T cells by blocking CD27 interaction (Figs. 3e–g), and inhibited the Treg secretome pro file of sCD27, IL-10, TGF- β, and adenosine (Figs. 3g, h and Supplementary Fig. 3a). The impaired Treg suppressive activity was further confirmed by flow cytometry and qRT-PCR analysis on additional Treg activation markers, including ICOS,TNFRSF4, 4-1BB (encoded byTNFRSF9), GITR (encoded by TNFRSF18), and TIGIT (Fig. 3i, Supplementary Figs. 3b,c). TIGIT was the only marker in the panel unaffected by CD70-KO, but this result was consistent with previous studies showing expression of TIGIT on Treg was independent of CD70-CD27 co-stimulation38,39.T h e Treg suppression assay further corroborated that CD70-KO function- ally inhibited Treg suppressive activity on the proliferation of paired CD8+ T cells, echoing the above-mentioned results (Supplemen- tary Fig. 3d). Considering the impaired suppressive activity of Tregs might reinvigorate anti-tumor immunity mediated by CD8+ T cells, as sug- gested by the GSEA analysis on bulk RNA-seq data (Supplementary Fig. 3e), we established a tumor-peripheral blood mononuclear cell (PBMC) co-culture system which consisted of principally CD4+ and Fig. 1 | Polarized CD4+ T naïve-to-Treg development and activation is a tumor- specificc h a r a c t e r i s t i ci nN P C .aThe UMAP plot of 189,750 T cells with 41 subtypes identified. b The developmental trajectories of CD4+ T cells.c The change ofFOXP3 expression through the pseudotime developmental process from CD4+ naïve T cells to three Treg subtypes, modeled by two-sided polynomial regression ana- lysis with the 95% confidence band. d The normalized fractions of CD4+ naïve T cells and three Treg subtypes in NPC tissues ( n = 36) versus in NPC peripheral blood (n = 10) or INP tissues (n = 4) (two-sided unpaired t-test).e The Treg sup- pressive score and T cell naïve score computed from GSE68799 (tumor n =4 2 , normal n = 3, two-sided unpaired t-test).f The progression-free survival for NPC patients from GSE102349, stratified by the Treg suppressive score (highn =4 4 ,l o w n = 44, two-sided log-rank test).g Expression of Treg-specific and naïve signatures in CD4+ naïve T cells and Treg subtypes.h Major biological signaling and activities in Treg subtypes computed by GSVA.i Illustration of the in vitro co-culture systems of CD4+ naïve T cells, NPC cells, and NPE cells. j–l The fraction of total FOXP3+ Tregs, and FOXP3+/CTLA4 + activated Treg in the co-culture systems (k,t r a n s w e l l co-culture; l, direct co-culture) with C666 versus with NP460 or NP69 cells (n =3 , two-sided unpaired t test). m The normalized mRNA fold change of Treg lineage- and activation-specific markers in CD4+ naïve T cells co-cultured with C666 versus with NP460 or NP69 cells (n = 3, two-sided unpaired t test). n The change of immunosuppressive factors in the co-culture systems with C666 versus with NP460 or NP69 cells (n = 3, two-sided unpaired t test). The n number represents n biolo- gically independent samples/experiments in each group. The data are presented as the mean ± SD (bar plots), median ± IQR (whiskers = 1.5 × IQR, box & whiskers plots), and KDE (violin plots). Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 4 CD8+ T cells (Supplementary Figs. 3f, g). We found that CD70-KO in C666 cells greatly enhanced cytotoxicity-dependent tumor cell death (Figs. 3j–l), preferentially in a co-culture system similar to the physio- logical T-cell landscape in NPC patients, with a higher expression of CD27 on CD4+ T cells (Supplementary Figs. 3h –j). Higher tumor cell death might be directly due to loss of CD70 co-stimulation in co- cultured CD4+ naïve T cells and Tregs induced by CD70-NC C666 cells rather than CD4+ and CD8+ T cells in the co-culture system (Supple- mentary Figs. 3k–m). In the CD70-KO system, CD8+/granzyme A+ and CD8+/perforin+ T cell fractions, secretion of TNF-α and IFN-γ,a sw e l la s CD8+ T cell proliferation were elevated, and expression of PD-1 and TIM-3 was diminished (Figs. 3m–q and Supplementary Fig. 3n), 0.0 0.5 1.0 1.5 2.0CD70 expression level Myeloid T cellB cell Plasma NK cell FibroblastTumor CD70-CD27 interaction network Sender Receiver Influencer 0 1 Effector TregResting TregNaive Treg Tumor cluster 1Tumor cluster 2Tumor cluster 3Tumor cluster 4Normal epithelial Importance incoming signal outgoing signal 0.00 0.02 0.04 0.06 0 2 4 6 8 Tumor Normal Tumor Normal CD70 expression 0 20 40 60 80 100CD70+ fraction (%) C666 NP460 NP69 C666 NP460 NP69 sCD27 concentration (IU/mL) 0.0 0.2 0.4 0.6 0.8 1.0 Progression-free survival 0 Time (months) CD70-High (n=57) CD70-Low (n=31) Logrank P = 0.03 0.25 0.5 0.75 1.0 Tumor abundance 0.0 0 12 3 CD70 expression 01234 CD70 expression CD2 7-CD70 4-1BB/4 -1BBL TIGIT -CD112 CD4+ naive T-NPC tumor CD4+ naive T-normal epithelial Resting Treg-NPC tumor Resting Treg-normal epithelial Naive Treg-NPC tumor Naive Treg-normal epithelial Effector Treg-NPC tumor Effector Treg-normal epithelial -log10(P value) 0 1 2 3 -7.5 -5.0 -2.5 0.0 Log2 mean TIGI T-CD155 CD 4 + Naive T CD226 -CD112 LGALS9 -CD44 a b c h k m n l p 68 1 0 1 2 22 26 30 34 P value = 5.9E-10 r = 0.54 CD70 expression Treg suppressive score 68 1 0 1 2 4 5 6 7 8 CD70 expression T cell naive score P value = 8.4E-6 r = -0.41 0 2 4 6 8 CD70 expression 20 40 60 Suppressive NaïveFOXP3 CTLA4 Time of differentiation (hours) Normalized expression/score r=0.79 P=0.007 r=-0.87 P=0.001 o 0 2 4 6 8 10 12 14 CD27 r=0.87 P=0.001 r=0.57 P=0.087 Epithelial Mast GSE68799 Patient 1Patient 2 GSE13597 8 10 12 14 16 Tumor Normal CD70 expression GSE53819 FOXP3/CTLA4 expression in lymphoid FOXP3/CTLA4 expression in tumor 50 μm50 μm C1: eTregs -1 0 1 2Normalized cell fraction C2: rTregs C1: nTregs C5: naive T C6: naive T C7: naive T -1 0 1 2 -2 Normalized Treg fraction -2 -1 0 12 3 r = -0.80 r = 0.16 CD70-LowCD70-High CD70-High CD70-Low Normalized Treg fraction -2 -1 0 12 3- 2 - 1 0 12 3 -1 0 1 2 -2 PRF1 expression level T cell Grranzyme level -1 0 1 2 -2 T cell proliferation level 3 P = 2.0×10-5 P = 0.49 r = -0.52 r = -0.07 P = 0.019 P = 0.76 r = -0.48 r = 0.07 P = 0.034 P = 0.75 CD70-High CD70-Low CD70-High CD70-Low i j Normalized Treg fraction 0 5 10 15 C666 NP460 NP69 0.25 0.5 12 4 8 16 32 64 128 0.1 0.2 0.3 0.4 0.5 Treg abundance 0.0 0.0 0.2 0.4 0.6 0.8 Treg abundance 0 2 4 6 0.0 0.2 0.4 0.6 0 1 2 3 4 0.00 0.25 0.50 0.75 Treg abundance CD70 expression P value < 2.2E-16 r = 0.33 Treg abundance CD70 expression P value < 2.2E-16 r = 0.45 0 20 40 60 EpCAM- EpCAM+ CD70+ fraction (%) 10 2 10 3 10 4 10 5 10 6 0 20 40 60 80 100Count (%) PE-CD70 EpCAM+ EpCAM- CD70+ EpCAM-: 13.8% EpCAM+: 61.1% 0.25 0.5 0.75 1.0 Tumor abundance 0.0 1.000.750.500.250.00 Tumor abundance Treg abundance 0.75 0.50 0.25 0.00 P value < 2.2E-16 r = 0.82 de 1.000.750.500.250.00 Tumor abundance Treg abundance 0.60 0.40 0.20 0.00 P value < 2.2E-16 r = 0.82 0 20 40 60CD27+ cell fraction (%) g 100 μm Inflammatory nasopharynxNPC tissue Normal epithelium (--) Tumor core (++) Lymphoid tissue (-) 100 μm NPC tissue Normal epithelium (--) Tumor core (++) 100 μm f P = 1.510E-5 P = 6.960E-4 P = 0.060 P = 0.004 P = 7.302E-4 P = 0.019 P = 0.140 P = 0.004 P = 7.536E-8 P = 7.227E-8 P = 1.772E-6 P = 4.527E-7 P = 0.001 P = 4.902E-5 P = 0.003 P = 0.003 Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 5 suggesting that impaired Treg suppressive activity could re-activate anti-tumor CD8+ T cells. To more comprehensively demonstrate the immunosuppressive effect of CD70-CD27 interaction in NPC-in filtrating T cells, we designed an autologous co-culture system between CD19+CD70-NC/ CD19+ CD70-KO C666 cells with anti-CD19 CAR-T cells (Figures S3O,P). The autologous co-culture system enabled antigen-speci fic T-cell activation in both CD4+ and CD8+ T cells and antigen-specifick i l l i n gt o the CD19+ CD70-NC and CD19+ CD70-KO C666 cells. Compared to the results from the allogenic co-culture system, CD70-KO in C666 cells in the autologous co-culture system consistently inhibited Treg polar- ization, activation, and suppressive activity, leading to enhanced antigen-specific T-cell killing and cytotoxicity (Supplementary Figs. 3q–s). Furthermore, we established immuno-resistant C666 cells by sequential PBMC challenge to the parental C666 cells to mimic p r o g r e s s i v ei m m u n ee s c a p ed u r i n gN P Cp r o g r e s s i o n .W ef o u n da consistent upregulation of CD70 expression in the immuno-resistant cells (Supplementary Fig. 3t), validating the active role of CD70+ NPC cells in facilitating immune evasion. Additionally, the lack of murine NPC cell lines, spontaneous mouse models, and the inability of EBV infection in mice constantly hindered the translational research on NPC immunology and immu- notherapy. Thus, we established a PBMC-engrafted humanized NSG mouse model to study the in vivo effect of CD70-KO C666 cells on the development and suppressive activity of infiltrating Tregs. Consistent with the in vitro results, CD70-KO did not alter the in vivo tumor- igenicity of NPC in immunode ficient NSG mice (Fig. 3r), but led to significant tumor shrink in humanized NSG mice (Fig. 3s), further confirming an inhibitory role of CD70 on anti-tumor immunity. While H&E staining exhibited no signi ficance difference between total immune infiltration in CD70-NC and CD70-KO tumors (Supplementary Fig. 4a), CD70-KO tumors harbored a decreased fraction of FOXP3 + Tregs with impaired activation, thus lowering IL-10 and TGF- β con- centrations in the TME (Figs. 3t–v, Supplementary Figs. 4b,c). Thus, CD70-KO tumors revigorated the effector pro file of CD8+ T cells by promoting cytotoxic cytokine release and preventing T-cell exhaus- tion (Figs. 3t, u, Supplementary Figs. 4b–e). These results implied that therapeutic inhibition of CD70 in NPC patients might be an effective strategy to overcome Treg immunosuppression which in turn activates T-cell immunity. Cusatuzumab enhances antitumor immunity and acts synergis- tically with anti-PD-1 treatment in patient-derived models Considering the high CD27 expression on T cells, NK cells, and B cells that might lead to off-target effects when using CD27 blockade40,C D 7 0 has emerged as a more effective and safe target since it is not expressed on normal tissues nor hematopoietic lineages during homeostasis41. Cusatuzumab is a human αCD70 monoclonal antibody that can effectively block CD70-CD27 interaction and its downstream signaling. So far, cusatuzumab is being evaluated in patients with acute myeloid leukemia (AML) in multiple clinical trials (NCT04023526 and NCT04150887), and has shown a potent killing ef ficacy to CD70 + leukemia stem cells with manageable adverse events42. However, the efficacy of cusatuzumab has not been comprehensively described in solid tumors in pre-clinical settings. By treating C666 cells with 5 μg/ mL cusatuzumab, we found Treg development, activation, secretion of immunosuppressive factors, and the ability to suppress CD8+ T cell proliferation in the CD4+ naïve T cell co-culture system were inhibited to a comparable extent as CD70-KO (Figs. 4a–d and Supplementary Fig. 4f). Thus, cusatuzumab treatment induced higher tumor cell death by promoting proliferation and cytotoxicity of CD8+ T cells in the PBMC co-culture system (Figs. 4e–i, Supplementary Figs. 4g,h). We also evaluated the inhibitory role of a CD70-blocking-only antibody on immunosuppression in the PBMC co-culture system. CD70 blocking in C666 cells showed comparable tumor-killing and cytotoxicity- enhancing effects with cusatuzumab treatment (Supplementary Figs. 4i–l), further confirming CD70 inhibition is effective in inducing stronger anti-tumor immunity. Next, the immune-activating effect of cusatuzumab was further validated in the co-culture system between patient-derived primary NPC cells and autologous PBMCs. Compared to IgG treatment, cusatuzumab treatment inhibited the suppressive activity of Tregs, which in turn activated the effector function of CD8+ T cells, eventually resulting in an enhanced tumor-killing ef ficacy (Supplementary Figs. 4m–p). In this study, cusatuzumab was shown to enhance T-cell immunity by alleviating Treg development and suppressive activity in NPC. Alternatively, the anti-PD-1 treatment directly unleashes proliferation and activation of CD8+/PD-1+ T cells by preventing T-cell exhaustion. Therefore, a synergistic effect that could enhance antitumor immunity might be achieved by combining anti-CD70 and anti-PD-1 treatments together. To investigate this potential synergistic effect, wefirst used the TIDE method to predict ICB responsiveness and T-cell signatures in CD70-high andCD70-low NPC patients43. CD70-high NPC patients were predicted to be more resistant to ICB blockade, showing lower T-cell cytotoxicity and in flammation but higher T-cell exclusion (Supple- mentary Fig. 5a). Despite NPC, CD70 is also preferentially expressed in other solid tumors, such as melanoma 44. Some studies in melanoma have generated TME transcriptome profiles from responders and non- responders receiving ICB 45,46, and computational analysis of these profiles illustrated that CD70 expression was associated with T-cell dysfunction and inferior ICB outcomes (Supplementary Figs. 5b and 5c). To verify the pro-tumor effect of CD70 in melanoma, we per- formed CD70-KO in a murine melanoma cell line: B16-F10 (Supple- mentary Fig. 5d), and orthotopically injected CD70-NC and CD70-KO B16-F10 cells in C57BL/6J mice. CD70-KO led to significant melanoma shrink, but we also observed a tumor-speci fic response to CD70-KO, possibly due to the mouse-speci fic TME (Supplementary Figs. 5e and 5f). Fig. 2 | CD70 + NPC cells contribute to abundant and suppressive Treg infil- tration by interacting with CD4+/CD27+ T cells. a CellPhoneDB revealed enri- ched ligand-receptor pairs amongCD4+ naïve T cells, Tregs, NPC, and NPE cells (two-sided permutation test).b CellChat revealed NPC cells as the sender ofCD70- CD27 signaling toCD4+ naïve T cells and Tregs.c Spatial co-localization of NPC cells, Tregs, and CD70 (n = 1089 and 1331). d The Pearson correlation (two-sided) between NPC cells/CD70 expression and Tregs (n =1 0 8 9a n d1 3 3 1 ) .e The CD70+ fractions in EPCAM + NPC cells and EPCAM- cells from NPC biopsies (n = 5, two- sided unpaired t test).IHC (f) and IF staining (g) showed CD70, FOXP3, and CTLA4 expression in NPC, normal epithelium and lymphoid tissues.h Expression ofCD70 across cell lineages. i Normalized fractions of Treg and CD4+ naïve subtypes in CD70-high (n = 20) andCD70-low patients (tumorn = 16, normaln = 4) from scRNA- seq data (two-sided unpaired t-test).j The Pearson correlation (two-sided) between the normalized Treg fraction and average expression of granzymes,PRF1, and T-cell proliferation markers inCD70-high (n = 20) and CD70-low patients (tumor n =1 6 , normal n = 4) from scRNA-seq data. k Normalized CD70 expression in three NPC RNA-seq cohorts. Left: GSE68799 (tumorn =4 2 ,n o r m a ln = 3); Middle: GSE13597 (tumor n =2 5 ,n o r m a ln = 3); Right: GSE53819 (tumor n = 18, normal n =1 8 )( t w o - sided unpaired t test). l The progression-free survival for NPC patients from GSE102349 (high n =5 7 ,l o wn = 31, two-sided log-rank test).m The fraction of CD70+ cells in C666 cells versus in NP460 or NP69 cells (n = 3, two-sided unpaired t test). n Changes of surface and sCD27 in co-culture systems with C666 versus with NP460 or NP69 cells (n = 3, two-sided unpaired t test). o The change of Treg sup- pressive and T naïve scores and signatures during induced Treg differentiation (n = 3, two-sided Pearson correlation analysis).p The Pearson correlation (two- sided) between the Treg/T naïve scores and CD70 expression in GSE102349 (n = 112). The n number represents n biologically independent spots/samples/ experiments in each group. The data are presented as the mean ± SD (bar plots), median ± IQR (whiskers = 1.5 × IQR, box & whiskers plots), and KDE (violin plots). Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 6 Experimentally, we evaluated the efficacy of CD70+ PD-1 blockade in NPC organoids, which mimicked the physiological condition of primary NPC tumors. We established NPC organoids from an EBV+ NPC patient-derived xenograft (Xeno76) established in 2018 47,a n d evaluated CD70 and PD-L1 expression levels in Xeno76 accordingly (Fig. 4j). In the organoid-PBMC co-culture system, both anti-CD70 monotherapy and anti-PD-1 monotherapy signi ficantly increased tumor cell death, but combination therapy exhibited greater tumor- killing efficacy in terms of the organoid number, size, and viability (Figs. 4k–n). We further compared the in vivo ef ficacy of the combi- nation therapy to monotherapy in Xeno76-transplanted PBMC- engrafted humanized NSG mice. Consistently, the anti-CD70 + anti-PD- 1 combination therapy showed the highest ef ficacy in inhibiting patient-derived xenograft (PDX) growth (Figs. 4o, p). Meanwhile, the fractions of tumor-infiltrating Tregs and CD8+ cytotoxic T cells were significantly decreased and increased upon receiving the combination Count (%) 0 20 40 60 80 100 0 1 2 3 4 5 Q1 2.03% Q3 28.45% PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 1.08% Q3 47.59% PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 NSG mice without humanized immunity NSG mice with humanized immunity CD70-NC CD70-KO CD70-NC CD70-KO 91 2 1 5 1 8 2 1 0 200 400 600 CD70-NC CD70-KO Time (days) Tumor volume (mm3) 1 3 5 7 9 11 13 15 0 200 400 600 800 Time (days) Tumor volume (mm3) CD70-NC CD70-KO FOXP3+ GZMA+ PRF1+ 0 10 20 30 40Cell fraction (%) CD70-NC CD70-KO r s PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 2.76% Q3 36.10% PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 3.44% Q3 64.20% PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 3.62% Q3 55.06% PE-FOXP3APC-CTLA4 PE-Annexin VAPC-7AAD PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 1.61% Q3 21.91% Q2 15.09% Q4 46.05% CD70-NC CD70-KO Q2 4.57% Q4 36.76% CD70-NC+PBMC Q2 21.73% Q4 10.63% CD70-KO+PBMC Q2 72.07% Q4 4.42% 0.0 0.2 0.4 0.6 0.8 24 hrs 48 hrs Relative survival 0 5 10 15 20 NC KO sCD27 concentration (IU/mL) 0 20 40 60 80 100Apoptosis index (%) NC KO CD70-NC CD70-KOEarly apoptosis Late apoptosis 0 10 20 30 40 Granzyme A Perforin Cell fraction (%) CD70-NC CD70-KO e j 0 1 2 3 4 NC KO CD8+ proliferation index n 4,000 CD70-NC CD70-KO Count (%) 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 0 20 40 60 80 100 110 FITC-CFSE ParentalProliferating p CD70-NC CD70-KO Count (%) 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 0 20 40 60 80 100 110 PE-Granzyme A PE-Perforin CD70-NC CD70-KO GZMA+ NC: 26.19% KO: 34.18% Perforin+ NC: 24.24% KO: 16.77% CD70-NC CD70-KO m 3,000 2,000 1,000 0 TNF-α IFN-γ Concentration (pg/mL) NC KO CD70-NC CD70-KO Concentration (pg/mL) 0 10 20 30 40 50CD27+ cell fraction (%) FOXP3+ cell fraction (%) FOXP3+/CTLA4+ cell fraction (%) NC KO NC KO 0 15 30 45 60 0 5 10 15 2075 PE-FOXP3APC-CTLA4 Q2 5.80% Q4 45.52% 800 ng/mL Isotype Q2 10.13% Q4 59.39% a 800 ng/mL rCD70 0 20 40 60 80 IL-10 0 100 200 300 400 500 TGF-β Adenosine Concentration (pg/mL) 0 10 20 30 40 IL-10 Isotype 0 100 200 300 400 500 TGF-β rCD70 Adenosine 0 20 40 60 80CD27+ cell fraction (%) IsotyperCD70 0.0 0.5 1.0 1.5 Normalized mRNA fold change TNFRSF4 TNFRSF9 TNFRSF18 TIGIT CD70-NC CD70-KO 0 15 30 45 60 75FOXP3+ cell fraction (%) P = 0.003 0 2 4 6 8 10FOXP3+/CTLA4+ cell fraction (%) IsotyperCD70 IsotyperCD70 0 2 4 6 8 IsotyperCD70 sCD27 concentration (IU/mL) 0 1 2 3 4 5Concentration (nmol/mL)Concentration (nmol/mL) 0 10 20 30 40Concentration (pg/mL) 0 250 500 750 1,000 0 200 400 600 TNF-α IFN-γIL-10 TGF-β 0.0 0.5 1.0 1.5 2.0Normalized mRNA fold change 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 FOXP3 IL2RA IKZF2 CTLA4 LAYN TNFRSF4 TNFRSF18 ICOS CD70-NC CD70-KO b cd f g h i kl o q tu v P = 0.001 P = 8.439E-6 P = 0.004 P = 0.003 P = 6.599E-4 P = 9.558E-4 P = 1.283E-4 P = 8.694E-5 P = 4.276E-4 P = 0.009 P = 7.520E-4 P = 0.002 P = 0.012 P = 8.842E-4 P = 0.001 P = 6.363E-6 P = 0.4284 P = 0.003 P = 9.925E-5 P = 5.848E-7 P = 0.018 P = 0.045 P = 0.008 P = 0.034 P = 0.001 P = 0.873 P = 7.261E-5 P = 1.727E-4 P = 3.518E-4 P = 0.001 P = 0.004 P = 0.002 P = 1.075E-6 P = 6.303E-5 P = 4.348E-4 P = 0.018 P = 0.002 P = 0.019 P = 0.018 P = 0.001 P = 0.049 P = 0.003 P = 5.068E-5 P = 4.849E-4 P = 0.002 P = 0.007 Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 7 therapy, respectively (Fig.4q and Supplementary Fig. 5g), suggesting that combining anti-CD70 and anti-PD-1 treatments achieved a syner- gistic effect that stimulated T-cell immunity by overcoming Treg- mediated immunosuppression. CD70-CD27 signaling contributes to lipid metabolic repro- gramming in Tregs Tregs are a resilient T-cell subtype with strong tissue adaptiveness and molecular flexibility that facilitates their survival, homeostasis, and metabolism in different ecosystems, including periphery, lymph nodes, thymus and tumors48–50. As we demonstrated that CD70-KO and inhibition in NPC cells enhanced anti-tumor immunity by limiting Treg activities, we further investigated how CD70-CD27 signaling in flu- enced downstream homeostasis of Tregs, particularly in the TME of NPC. Hence, we applied additional 3′ scRNA-seq to PBMCs co-cultured with CD70-NC and CD70-KO C666 cells (Fig. 5a). Consistent with our in vitro and in vivo results, CD70-KO reduced the abundances of nTregs, rTregs and eTregs, and expanded pro-inflammatory subtypes, including CD8+cytotoxic T cells, CD4+T FH cells and CD4+ naïve T cells (Fig. 5b). Moreover, Tregs in the PBMCs co-cultured with CD70-KO C666 cells (KO-Tregs) exhibited lower expression of lineage-specific( FOXP3, IL2RA and SOX4), activation ( CTLA4 and LAYN)a n dc o - s t i m u l a t o r y (TNFRSF4and TNFRSF9) markers, but higher expression of naïve (SELL) signature markers (Fig. 5c). Particularly, some co-cultured eTreg transitioned into a fragile state which has been previously character- ized by retention of FOXP3 expression but aberrant IFN-γ production51 (Supplementary Figs. 6a,b). However, upregulation of PD-1 in CD70-co- cultured Tregs protected these cells from IFN-induced fragility while maintaining functional homeostasis (Fig. 5c and Supplementary Fig. 6c). Thus, the synergy observed upon inclusion of PD-1 blockade in the anti-CD70 treatment might also partially reduce immunosup- pression by inducing Treg fragility52. As for CD8+ cytotoxic T cells in the CD70-KO group, expression of proliferation (MKI67), cytotoxicity (GZMA and LTB) and activation ( TNFRSF1B and CD28) markers was elevated, whereas naïve (TCF7 and CCR7) and exhaustion (LAYN, LAG3, and HAVCR2) markers were decreased, further corroborating that disrupted suppressive activity of Tregs promoted proliferation, acti- vation, and alleviated exhaustion in CD8+ T cells (Fig. 5d). CD70- mediated Treg immunosuppression also hampered the inflammatory response of CD8+ cytotoxic T cells, thus lowering their anti-tumor effect (Supplementary Fig. 6d). To further delineate the downstream mechanisms involved with CD70-CD27 signaling in Tregs, we performed GSEA between CD70- induced Tregs and KO-Tregs. CD70+ C666 cells potently reprogramed metabolic pro files of co-cultured Tregs via enhancing oxidative phosphorylation (OXPHOS) driven by a complex lipid signaling net- work (Fig. 5e and Supplementary Fig. 6e). Tregs frequently require robust metabolism to sustain their development and suppressive activity in the TME. Particularly, Tregs possess high energy-driven plasticity, and thus can utilize different metabolic resources, including lactic acids, glucose, and fatty acids (FAs), to sustain functional homeostasis in the metabolite-deficient milieu20,53–55. Pathway analysis revealed that fatty acid metabolism, cholesterol homeostasis, and mitochondrial functions were top enriched signatures in CD70- induced Tregs, which have been recently reported to facilitate Treg lineage determination, functional fitness, and immunosuppression (Fig. 5e). For example, increased OXPHOS driven by fatty acid oxida- tion was a hallmark of Treg development and suppressive activity in the TME 53.M e a n w h i l e ,u p r e g u l a t e dd e t o x ification and redox home- ostasis protected CD70-induced Tregs from a high level of reactive o x y g e ns p e c i e sp r o d u c e db yO X P H O Sa n dA T Pp r o d u c t i o nw h i l e maintaining mitochondrial integrity (Fig. 5f). In contrast, KO-Tregs exhibited impaired functional homeostasis due to enhanced lipid efflux and clearance as well as loss of mitochondrial integrity (Fig.5f). Mitochondrial integrity, cholesterol homeostasis, and fatty acid oxidation collectively contribute to the functional specializa- tion of Tregs To gain mechanistic insights into how the CD70-induced complex lipid network strengthened the functional specialization and homeostasis of Tregs in the NPC microenvironment, we evaluated the intracellular changes of total lipid and key lipid constituents, such as cholesterol and fatty acids, in co-cultured CD4+ T cells with CD70-NC and CD70- KO C666 cells. First, genetic ablation and therapeutic inhibition of CD70 in NPC cells significantly reduced the total intracellular lipid and cholesterol in co-cultured Tregs (Figs. 6a, b). Second, there was an increase of cell-free cholesterol from FOXP3- non-Tregs, FOXP3+/ CTLA4- rTregs to FOXP3+/CTLA4+ eTregs, which conveyed that intracellular cholesterol accumulation was associated with Treg development and suppressive activity (Fig.6c). In the integrated NPC scRNA-seq cohort, we validated eTregs had more robust cholesterol metabolism and biosynthesis, but lower cholesterol export (Supple- mentary Fig. 7a). Contemplating that lipid signaling was a highly complex network involving subtle changes in various metabolites, we performed mass spectrometry-based metabolomics to quantify intracellular metabolic alterations of polar molecules and fatty acids in co-cultured CD4+ T cells (Fig. 6c). Mass spectrometry analysis revealed a consistent decrease of intracellular cholesterol and adenosine in the CD70-KO group with lower CD39 expression (Fig.6d, Supplementary Figs. 7b,c). Meanwhile, increased glycerol could protect Tregs from lipid- accumulated toxicity, which damaged FOXP3 stability56 (Fig. 6d). The OXPHOS-derived metabolites, including succinate, 2-hydroxyglutarate (2-HG) and malate, which have previously been reported associated with impaired Treg development and suppressive activity through epigenetic repression of PD-1 57, were actively metabolized (Fig. 6d). Whereas citrate, aspartate and glutamine remained unchanged in CD4+ T cells between CD70-NC and CD70-KO groups. For the fatty acid panel, the total fatty acid content in the CD70-KO group was markedly decreased (Fig. 6d). The concentrations of almost all intra- cellular fatty acid derivatives from C4 to C26 were reduced Fig. 3 | Genetic ablation of CD70 in NPC cells reverts Treg suppressive activity and enhances CD8+ T cell function in the TME. a, b Immunophenotyping of CD4+ naïve T cells treated with IgG antibody and recombinant CD70 antibody (rCD70, n = 3, two-sided unpaired t test).c Changes of surface and sCD27 in the co-cultured systems treated with IgG antibody and rCD70 (n = 3, two-sided unpaired t test). d The change of immunosuppressive factors in the IgG-treated and rCD70-treated co-culture systems (n = 3, two-sided unpaired t test). e, f Immunophenotyping of CD4+ naïve T cells co-cultured with CD70-NC and CD70-KO C666 cells (n =3 ,t w o - sided unpaired t test). g Changes of surface and sCD27 in the co-cultured systems with CD70-NC and CD70-KO C666 cells (n = 3, two-sided unpaired t test). h The change of immunosuppressive factors in the CD70-NC and CD70-KO co-culture systems (n = 3, two-sided unpaired t test). i The normalized mRNA fold change of Treg activation-specific markers in CD4+ naïve T cells co-cultured with CD70-NC and CD70-KO C666 cells (n = 3, two-sided unpaired t test).T-cell cytotoxicity measured in the CD70-NC and CD70-KO (j and k, flow cytometry,n =3 ; l, XTT assay, n = 5) PBMC co-culture systems (two-sided unpaired t test). The change of CD8+ T-cell cytotoxicity markers (m and n, flow cytometry), cytokines (o,E L I S A )a n d proliferation (p and q, CFSE) in the CD70-NC and CD70-KO PBMC co-culture sys- tems (n = 3, two-sided unpaired t test). (r, s) The CD70-NC and CD70-KO tumor growth (r, n = 5 for NSG mice;s, n = 6 for humanized mice) in immunodeficient NSG and PBMC-engrafted humanized mice (two-sided unpaired t test). The change of immunosuppressive and cytotoxic T cell subtypes (t, flow cytometry), cytokines (u, ELISA), and markers (v, qRT-PCR) in CD70-NC and CD70-KO tumors (n =3 ,t w o - sided unpaired t-test). The n number represents n biologically independent sam- ples/experiments in each group. The data are presented as the mean ± SD (bar plots). Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 8 (Supplementary Table 2), indicating that block of CD70-CD27 signaling in Tregs might undermine de novo fatty acid uptake, synthesis and oxidation. Based on the observed metabolic adaptation of Tregs in CD70/ CD27-mediated lipid signaling, we proposed that mitochondrial integrity, cholesterol homeostasis, and fatty acid metabolism collec- tively contributed to metabolic advantages and functional specializa- tion of Tregs via CD70-CD27 signaling in the NPC microenvironment. We speculated that CD70 served as a master metabolic switch that reprogramed lipid metabolism in CD4+ naïve T cells and Tregs to meet the metabolic demands committed these cells to Treg development, homeostasis and activation in a hostile milieu. To validate our hypothesis, we first showed that KO-Tregs had fewer and unhealthier mitochondria examined by electron microscopy (Fig.6e), resulting in decreased ATP generation, OXPHOS and mitochondrial potential shown by the Seahorse assay, and JC-1 staining (Figs. 6f–h). 0 20 40 60 Count (%) 10 0 10 1 10 2 10 3 10 4 10 5 10 6 0 20 40 60 80 100 Count (%) 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 0 20 40 60 80 100 PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 Q1 4.48% Q3 63.21% NPC PDX (Xeno 76) CD70+: 87.39% No stain NPC PDX APC-CD70 Cusatuzumab (+) Camrelizumab (+) 200 μm 200 μm200 μm 200 μm j mControl 0 10 20 30 40No. of organoids per well k anti-CD70 + + + -- -- +anti-PD-1 Control Xenotransplantation in humanized NSG mice Anti-CD70 Anti-PD-1 Combination Time (days) Tumor volume (mm3) Control anti-CD70 anti-PD-1 Combination Treatment starts 0 10 20 30 40 0 10 20 30 40 50 0 5 10 15 20 25 anti-CD70 + + + -- -- +anti-PD-1 Treg faction (%)GZMA faction (%)PRF1 faction (%) op q 369 1 2 1 5 1 8 2 1 0 500 1000 1500 0 4 8 12 PE-Annexin VAPC-7AAD Q2 43.75% Q4 8.77% Cusatuzumab sCD27 concentration (IU/mL) Early apoptosis Late apoptosis +- Control+PBMC Cusatuzumab+PBMC Cusatuzumab +-0 20 40 60 0.0 0.2 0.4 0.6 0.8Relative survival 24 hrs 48 hrs Cusatuzumab (-) Cusatuzumab (+) PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 0.80% Q3 63.95% PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 2.59% Q3 41.83% Q2 9.00% Q4 46.58% Q2 1.10% Q4 34.15% PE-FOXP3APC-CLTA4 Control Cusatuzumabab eg Granzyme A Perforin Cell fraction (%) h 0 10 20 30 40 Cusatuzumab (-) Cusatuzumab (+) Apoptotic fraction (%) 0 20 40CD27+ cell fraction (%) TNF-α IFN-γ Concentration (pg/mL) Concentration (pg/mL) IL-10 TGF-β Adenosine PD-L1+: 11.38% FITC-PD-L1 3,000 2,000 1,000 0 Cusatuzumab (-) Cusatuzumab (+) 0 200 400 600 Cusatuzumab (-) Cusatuzumab (+) PE-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 Q1 3.96% Q3 42.78% Q2 22.42% Q4 9.89% Q2 41.91% Q4 11.35% No stain NPC PDX NPC PDX (Xeno 76) 0 15 30 45 60 75FOXP3+ cell fraction (%) 0 FOXP3+/CTLA4+ cell fraction (%) 4 8 12 Cusatuzumab +- Cusatuzumab +- Cusatuzumab +- P = 0.002 60 0 2 4 6 8Concentration (nmol/mL) cd f i PE-H APC-H -10 3 01 0 3 10 4 10 5 10 6 10 7 10 8 10 -0.1 10 2 10 3 10 4 10 5 10 6 10 7.9 Q1 0.64% Q3 62.35% PE-H APC-H -10 3 01 0 3 10 4 10 5 10 6 10 7 10 8 10 -0.1 10 2 10 3 10 4 10 5 10 6 10 7.9 Q1 1.11% Q3 47.92% PE-H APC-H -10 3 01 0 3 10 4 10 5 10 6 10 7 10 8 10 -0.1 10 2 10 3 10 4 10 5 10 6 10 7.9 Q1 1.74% Q3 32.93% PE-H APC-H -10 3 01 0 3 10 4 10 5 10 6 10 7 10 8 10 -0.1 10 2 10 3 10 4 10 5 10 6 10 7.9 Q1 1.27% Q3 44.56% PE-Annexin VAPC-7AAD Q2 2.01% Q4 35.00% Q2 8.79% Q4 42.19% Camrelizumab (+)Control Cusatuzumab (+) Cusatuzumab (+) Camrelizumab (+) Q2 12.37% Q4 41.80% Q2 17.53% Q4 47.80% l 25 35 45 55 65 75Apoptosis index (%) anti-CD70 + + + -- -- +anti-PD-1 Early apoptosis Late apoptosisn Camrelizumab (+) Cusatuzumab (+) P = 6.354E-5 P = 3.919E-7 P = 0.003 P = 0.001 P = 0.021 P = 1.366E-5 P = 2.097E-4 P = 8.538E-9 P = 4.092E-7 P = 0.002 P = 0.005 P = 0.037 P = 0.045 P = 0.016 P = 0.012 P = 0.003 P = 0.047 P = 0.043 P = 0.840 P = 0.015 P = 0.031 P = 0.918 P = 0.002 P = 0.039 P = 0.005 P = 0.002 P = 0.013 P = 0.007 P = 0.035 P = 0.042 P = 2.897E-4 P = 0.004 P = 1.175E-4 P = 2.170E-4 P = 0.022 P = 3.040E-5 P = 7.749E-4 P = 0.003 P = 0.008 P = 0.172 P = 0.069 P = 0.025 P = 0.007 P = 0.029 P = 5.330E-4 P = 0.032 P = 0.010 P = 0.001 Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 9 Furthermore, we evaluated the changes of 46 representative genes involved in lipid synthesis, transport, and oxidation in Tregs, in which more than 70% of the selected signatures were significantly influenced by CD70-CD27 signaling (Fig. 6i). In light of elevated intracellular cholesterol in CD70-induced Tregs, we first inspected the change in LDLR,w h i c hw a si n v o l v e di n regulating cholesterol uptake58. Interestingly, expression of LDLR did not increase in CD70-induced Tregs, whereas cholesterol ef flux- associated genes, such as SCARB1 and SDC1,w e r ef o u n dd o w n - regulated, suggesting the cholesterol accumulation was caused by inhibited cholesterol export. Moreover, the key regulators and enzymes involved with cholesterol metabolism and Treg homeostasis via the mevalonate pathway, including SREBF1, SREBF2, HMGCS1, SOAT1, SOAT2, FNTB, PGGT1B, GGPS1, RAC2 and PD-1, which fostered Treg development and maintenance59,60, were also found augmented by CD70-CD27 signaling (Fig. 6i). Intracellular fatty acid accumulation in Tregs caused by fatty acid uptake and de novo synthesis was actively involved in IL-2-dependent proliferation and TCR-dependent activation 60,61.C D 7 0 - K Oh a d impaired fatty acid uptake and oxidation in co-cultured CD4+ T cells, as determined by BODIPY C 12 tracing and levels of three enzymes (ACADVL, ACADM, and HADHA) in the FAO pathway (Supplementary Figs. 7d and 7e). Increased ACACA, instead of FASN, in CD70-co- cultured CD4+ naïve T cells, might also contribute to intracellular fatty acid accumulation via de novo FA synthesis, which upregulated TCR- dependent Treg activation and maturation markers TNFRSF18 and CD44 (Fig. 6i). As the above metabolic assays demonstrated, the accumulated fatty acid pool in CD70-induced Tregs provided suf fi- cient resources for FAO via mitochondrial OXPHOS, which sustainably generated ATP suf ficient to meet the high energy demands of Treg functionalization and survival in the TME57. Mitochondrial complex III has been previously demonstrated to be essential for Treg suppressive activity and enhanced FAO-driven OXPHOS62. UQCRFS1, UQCRQ,a n d CYC1, associated with the integrity and function of mitochondrial complex III, were upregulated in Tregs activated by CD70-CD27 signaling, orchestrating with the electron microscope images (Figs. 6e, i). We also found upregulation of key regulators and enzymes in the TCA cycle and electron transport chain (ETC), including NR4A1, NR4A2, MDH1, MDH2, ECHS1 and ECH1 (Fig. 6i)63,64. Other transcription factors, includingSTAT5A, and cAMP- inducible factors, such as CREB1 and CEBPB, which were related to metabolic fitness of Tregs65,66, might ensure functional specialization of Tregs in the NPC microenvironment (Fig.6i). However, several other vital lipid metabolism-associated genes, including SQLE, SCAP, HMGCR, ATF4, CCDC5B, LAGLS3, CTNB1,a n d EGR1, were not differen- tially expressed (Fig. 6i), indicating an alternative and CD70- independent lipid signaling in Tregs within a tumor-specificn i c h e . We further constructed a lipid signaling module comprising 15 representative signature genes selected from the qPCR panel based on multivariate linear regression (Supplementary Fig. 7f), to quantita- tively evaluate the lipid signaling activity in Tregs and other CD4+ T cells in the single-cell and bulk transcriptome datasets. First, genetic ablation of CD70 in NPC cells signi ficantly impaired lipid signaling in the co-cultured Tregs (Fig. 6j). Meanwhile, in the integrated NPC single-cell cohort, eTregs possessed the highest lipid signaling activity, exceeding that of rTregs, nTregs and other CD4+ subtypes, including TFH cells, TCM cells and naïve cells (Fig. 6k). In the time-resolved Treg differentiation model, the lipid signaling activity was positively cor- related with the Treg suppressive score and CD25 expression during naïve T-cell-to-Treg differentiation and negatively correlated with the naïve score and IL7R expression (Fig. 6l). In the bulk RNA-seq cohort, the lipid signaling activity was highly correlated to the suppressive activity of Tregs and inferior prognosis in NPC patients (Figs. 6m, n). Furthermore, the correlations between the 15 representative signature genes and CD70, CD27, FOXP3, CTLA4,a n d CD25 were independently validated in transcriptome data for 337 whole blood samples from GTEx (Fig. 6o), suggesting that the lipid module was consistently related to Treg activities across individuals. In summary, thesefindings demonstrated intimate cooperation between lipid metabolism repro- gramming and CD70-CD27 signaling in the lineage determination, homeostasis, and the suppressive activity of NPC-infiltrating Tregs. To validate the importance of lipid metabolism in CD70-induced Treg development and activation in the NPC microenvironment, we assessed functional and metabolic changes in CD70-NC and CD70-KO co-culture with a lipid-depleted microenvironment. In the lipid- depleted co-culture system, CD70+ NPC cells could no longer induce a higher fraction of total and activated Tregs differentiated from CD4+ naïve T cells, nor induce stronger suppression on CD8+ T cell pro- liferation (Figs. 6p, q and Supplementary Fig. 7g). The immunosup- pressive secretome pro files between CD4+ T cells co-cultured with CD70-NC and CD70-KO C666 cells in a lipid-depleted system were not distinctive either (Fig. 6r). Therefore, lipid depletion also resulted in the incapability of CD70-CD27 interaction in regulating Treg immu- nosuppression in the NPC-PBMC co-culture system, causing elevated anti-tumor immunity in both CD70-NC and CD70-KO groups (Fig.6s). Moreover, the total intracellular lipid content in NC- and KO-Tregs remained comparable, due to the lack of free lipids that could be taken up and synthesized (Fig. 6t). Metabolic assays further exhibited that ATP generation, OXPHOS, mitochondrial integrity, and fatty acid oxi- dation in Tregs were significantly impaired upon lipid depletion, even with active CD70-CD27 signaling (Figs. 6u–x). These results success- fully demonstrated that lipid metabolism was indispensable for CD70- CD27 interaction to enhance development, homeostasis, and the suppressive activity of NPC-infiltrating Tregs. CD70 is transcriptionally upregulated in NPC cells by NFKB2 via the EBV-dependent epigenetic modification Since CD70 overexpression in NPC tissues was frequently observed in clinical practice, we wanted to explore the upstream mechanism contributing to CD70-mediated immune evasion. The single-cell ana- lysis found the transcriptome level of CD70 significantly over- expressed in EBV+ NPC cells (Fig.7a). We corroborated that the CD70+ fraction was higher in EBV+ NPC43 cells, compared to the EBV- coun- terparts (Fig. 7a). EBV infection in NPC causes genomic instability and thus epigenetically stimulates the transcription of oncogenes to facil- itate tumor progression and immune escape 67,68. The chromosomal Fig. 4 | Therapeutic inhibition of CD70 promotes anti-tumor immunity and anti-PD-1 efficacy. a, b Immunophenotyping of CD4+ naïve T cells treated with IgG antibody and cusatuzumab (n = 3, two-sided unpaired t test). c The change of surface and sCD27 in the co-cultured systems treated with IgG antibody and cusatuzumab (n = 3, two-sided unpaired t test). d The change of immunosuppres- sive factors in the co-cultured systems treated with IgG antibody and cusatuzumab (n = 3, two-sided unpaired t test). T-cell cytotoxicity measured in the IgG-treated and cusatuzumab-treated (e and f, flow cytometry,n =3 ; g, XTT assay, n = 5) C666/ PBMC co-culture systems (two-sided unpaired t test). The change of CD8 + T-cell cytotoxicity markers (h, flow cytometry) and cytokines (i, ELISA) in IgG-treated and cusatuzumab-treated C666/PBMC co-culture systems (n = 3, two-sided unpaired t test). j Expression of CD70 and PD-L1 in NPC PDX Xeno76. T-cell cytotoxicity measured in the IgG-treated and cusatuzumab-treated (k and m,m i c r o s c o p y ;l and n, flow cytometry) PDX/PBMC co-culture systems (n = 3, two-sided unpaired t test). o, p Tumor growth in PDX-bearing humanized mice treated with different regimens (n = 4 for each treatment group), measured from day 1 post intravenous PBMC injection (two-sided unpaired t test).q The change of Tregs and cytotoxic T cells in tumors compared between every two treatment groups (n = 3 per treatment group, two-sided unpaired t test). The n number represents n biologically independent samples/experiments in each group. The data are presented as the mean ± SD (bar plots). Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 10 locus of the CD70 gene on chromosome 19 has been reported to be frequently amplified in NPC 69,70. Thus, to delineate whether CD70 overexpression in NPC was epigenetically induced by EBV-mediated chromatin accessibility alterations, we performed ATAC-seq on EBV+ NPC cells, EBV- NPC cells, and NPE cells. The chromatin accessibility at the CD70 promoter region was enhanced in EBV+ NPC cells compared to the EBV- and normal counterparts, enabling a more effective tran- scription factor binding to promote CD70 mRNA transcription (Fig. 7b). Based on the ATAC analysis, we identi fied the top 15 tran- scription factors with the highest transcriptional activity in EBV+ NPC cells, that could potentially bind to the CD70 promoter region with higher accessibility (Fig. 7c). Among the 15 transcription factors, we 5,000 10,000 15,000 20,000Rank −10 −5 0 5 −15 −10 −5 0 5 UMAP_1 UMAP_2 C1 C3 C2 C5 C4 C6 C8 C7 C9 C10 C11 C12 C13 C1-Naive Tregs C2-Cytotoxic T cells_1 C3-Resting Tregs C4-Proliferating T cells C5-Cytotoxic T cells_2 C6-Naive T cells C7-Effector Tregs C8-Follicular helper T cells C9-Plasma B cells C10-Effector memory T cells C11-Fibroblasts C12-Unactivated T cells C13-Myeloid cells NPC co-cultured 19,682 PBMCs C1-Naive Tregs C2-Cytotoxic T cells_1 C3-Resting Tregs C4-Proliferating T cells C5-Cytotoxic T cells_2 C6-Naive T cells C7-Effector Tregs C8-Follicular helper T cells C10-Effector memory T cells C12-Unactivated T cells CD70-NC CD70-KO 1.00 0.00 0.50 0.75 0.25 Cell fraction 0 1 2 3 4Expression Level MKI67, Wilcoxon P < 2.2e−16 CD70-KOCD70-NC 0 1 2 3Expression Level LTB, Wilcoxon P < 2.2e−16 CD70-KOCD70-NC 0 1 2 3Expression Level 0 1 2 3 4 5 Expression Level GZMA, Wilcoxon P = 6.9e−10 TNFRSF1B, Wilcoxon P = 1e-12 CD70-KOCD70-NC CD70-KOCD70-NC 0 1 2Expression Level CD28, Wilcoxon P = 8.6e−08 CD70-KOCD70-NC 0 1 2 Expression Level 0 1 2 3Expression Level 0 1 2Expression Level 0 1 2 3 Expression Level 0 1 2 3Expression Level TCF7, Wilcoxon P = 5.5e−06 CD70-KOCD70-NC CD70-KO CD70-NC CD70-KO CD70-NC CD70-KO CD70-NC CD70-KO CD70-NC CCR7, Wilcoxon P = 4.4e−08 LAYN, Wilcoxon P = 4.2e−09 LAG3, Wilcoxon P = 8.3e−05 HAVCR2, Wilcoxon P = 3.1e−05 5,000 10,000 15,000 20,000 0.0 0.1 0.2 0.3 0.4 0.5 Fatty acid metabolism Enrichment score Rank zero cross at 8,114 Cholesterol homeostasisOxidative phosphorylation NES = 1.90 FDR = 0.001 0 1 2 Expression Level 0 1 2 3 4Expression Level 0 1 2Expression Level 0 1 2 Expression Level 0 1 2Expression Level CD70-KOCD70-NC CD70-KO CD70-NC CD70-KO CD70-NC CD70-KO CD70-NC CD70-KO CD70-NC FOXP3, Wilcoxon P < 2.2e−16 ILR2A, Wilcoxon P < 2.2e−16 SOX4, Wilcoxon P < 2.2e−16 CTLA4, Wilcoxon P = 2.1e−05 LAYN, Wilcoxon P = 0.0056 0 1 2Expression Level 0 1 2 3 4Expression Level 0 1 2 3 4Expression Level 0 1 2 Expression Level 0 1 2Expression Level CD70-KOCD70-NC SELL, Wilcoxon P < 2.2e−16 CD70-KOCD70-NC ISG15, Wilcoxon P = 9.1e−09 CD70-KOCD70-NC TNFRSF4, Wilcoxon P < 2.2e−16 CD70-KOCD70-NC TNFRSF9, Wilcoxon P = 0.0007 CD70-KOCD70-NC PDCD1, wilcoxon P = 0.039 Normalized gnee expression in Tregs (C1, C3 and C7)Normalized gnee expression in cytotoxic T (C2, C5 and C10) 5,000 10,000 15,000 20,000 0.0 0.1 0.2 0.3 0.4 0.5 0.6Enrichment score zero cross at 8,114 Rank 5,000 10,000 15,000 20,000 0.0 0.1 0.2 0.3 0.4 0.5 Rank Enrichment score zero cross at 8,114 NES = 1.95 FDR < 0.001 NES = 1.61 FDR = 0.014 NC-enriched KO-enriched NC-enriched KO-enriched NC-enriched KO-enriched 5̹000 10,000 15,000 20,000-0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Rank Mitochondrial respiratory electron transport NC-enriched KO-enriched zero cross at 8,114 NES = 1.84 FDR = 0.017 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Mitochondrial respiratory chain complex III Enrichment score Enrichment score NC-enriched KO-enriched NES = 1.55 FDR = 0.029 zero cross at 8,114 0.0 0.1 0.2 0.3 0.4 0.5Enrichment score 5,000 10,000 15,000Rank NC-enriched KO-enriched 20,000 Mitochondrial fatty acid β oxidation NES = 1.43 FDR = 0.170 Detoxification MT complex II assembly MT complex I assembly MT NADH to ubiquinone IL10 signaling IL2/STAT5 signaling Negative lipid metabolism WNT/β-catenin signaling Lipid export Cholesterol efflux Mitochondrial depolarisation a b c d e f -2 -1 0 1 2 Other top enriched signatures in Tregs Inflammatory response Citrate TCA cycle Redox homeostasis T cell homeostasis ATP metabolic process IFN signaling Glycolysis LDL clearance Apoptitic execution phase Caspase-mediated cleavage Normalized enrichment score (NES) P value < 0.05 and FDR < 0.2 Negative lipid transport zero cross at 8,114 C1: 20.20% C2: 11.27% C3: 15.12% C4: 14.15% C5: 8.72% C6: 8.70% C7: 11.49% C8: 5.07% C10: 5.07% C1: 15.04% C2: 19.90% C3: 11.98% C4: 11.58% C5: 11.72% C6: 11.25% C7: 4.49% C8: 9.02% C10: 4.49% 4 Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 11 characterized that NFKB2 was the one most correlated to CD70 expression validated in two NPC RNA-seq cohorts and Visium spatial data, and had the highest binding score to the CD70 promoter region predicted by JASPER (Figs. 7d, e and Supplementary Data 2). In EBV- infected lymphocytes, NFKB2 was also found highly correlated to CD70, indicating a consistent EBV-modulated epigenetics in both immune and epithelial cells (Fig. 7f). NFKB2 is a master regulator of the NF-κB pathway, which is widely considered as the homogeneous driver in NPC initiation and pro- gression. NFKB2 was frequently overexpressed in EBV+ NPC cells, instead of EBV- tumor cells, NPE cells, stromal cells, and immune cells in the TME (Fig.7g ) ,a n dh a sb e e np r e v i o u s l ys h o w nt op r o m o t et u m o r growth and survival 71. Nevertheless, its regulatory role on immune evasion of NPC and the molecular linkage to CD70 upregulation has not been comprehensively elucidated. Thus, to delineate the reg- ulatory role of NFKB2 on CD70 transcription, we identified a unique binding motif of NFKB2 on the CD70 promoter region with enhanced accessibility caused by EBV infection (Fig. 7b). Thus, we designed a point-mutatedCD70 binding motif with luciferase and demonstrated that NFKB2 could speci fically and effectively bind to this site to enhance CD70 transcription (Fig. 7h). To further validate NFKB2 as a transcriptional enhancer for CD70, we performed NFKB2 knockdown and inhibition in C666 cells, showing consistent decreases in both CD70 mRNA and CD70 protein (Figs. 7it o 7k). Loss of NFKB2 in C666 cells thus sensitized them to immune attack in the PBMC co-culture system (Figs.7l, m). These data unveiled the EBV-NFKB2-CD70 axis as a vital mechanism for NPC cells to escape immune surveillance by upregulating Treg immunosuppression that lowered T-cell immunity. Discussion Given the limited clinical impacts of ICB therapy in NPC patients, developing targets that best represent the TME characteristics remains essential for improving the clinical ef ficacy of precision immunotherapy. To address this need, we established a large-scale and multicentral NPC clinical cohort containing public single-cell information from 50 primary tissues and peripheral blood samples. Our analysis demonstrated that abundant and activated Tregs were the dominant source of immunosuppression in the NPC micro- environment. We proposed that most tumor-in filtrating Tregs were differentiated from in situ and peripheral CD4+ naïve T cells and activated via CD70-CD27 signaling. We identi fied and characterized that CD70+ NPC cells were the primary contributor to such Treg- mediated immunosuppression. Genetic ablation and therapeutic inhibition of CD70 in NPC cells undermined Treg abundance and suppressive activity, thus enhancing anti-tumor immunity mediated by CD8+ T cells and the ef ficacy of anti-PD-1 treatment (Fig. 7n), as comprehensively validated in cell lines, primary culture, organoids, PDX, and humanized mouse models. Therapeutically, CD70-targeted immunotherapy has been pri- marily evaluated in patients with hematologic malignancies, such as acute myeloid leukemia (AML) and B cell lymphoma72,73.H o w e v e r ,t h e efficacy of anti-CD70 treatment in solid tumors has not been com- prehensively described. Here, we exhibited that anti-CD70 mono- therapy and anti-CD70+ anti-PD-1 combination therapy overcame Treg-mediated immunosuppression and elevated the effector function of CD8+ T cells in NPC, providing sufficient experimental evidence of how anti-CD70 treatment exerted its tumor-killing effect in a Treg- enriched TME. We also unveiled that CD70-targeted cusatuzumab treatment destabilized Tregs in a manner dependent on the CD70 expression level on tumor cells, the CD27 expression level on CD4+ T cells, and the CD4+/CD8+ ratio in the TME. Clinical practice revealed that most NPC patients had CD70-high tumor histology, CD27-high lymphocyte infiltration, and a high CD4+/CD8+ ratio74,75, which could theoretically benefit from anti-CD70 monotherapy and anti-CD70/ anti-PD-1 combination therapy, and yet this hypothesis requires translational validation in the future. Mechanistically, at present, little is known about the down- stream molecular mechanism of CD70-CD27 co-stimulatory sig- naling in tumor-in filtrating Tregs. We identi fied CD70+ NPC cells as a metabolic switch that could turn on lipid metabolism reprogramming on CD4+/CD27+ T cells, particularly CD4+ naïve T cells and Tregs, via enhanced transcriptional and enzymatic activities in cholesterol homeost asis via the mevalonate pathway, de novo fatty acid uptake, synthesis, oxidation, and mitochon- drial integrity. Enhanced intracellular cholesterol accumulation and synthesis profoundly induced Treg suppressive activity and maintained Treg homeostasis in the TME via PD-1 induction to prevent IFNG-induced fragility, yet potentially being targeted by the synergistic PD-1 blockade. In CD70-induced Tregs, mito- chondrial fatty acid oxidation-driven OXPHOS and ATP genera- tion were signi ficantly enhanced to maintain Treg metabolic fitness in the TME by activating 2-HG and succinate metabolism and promoting epigenetic activation of immunosuppressive genes. Instead, we also identi fied some key regulators and enzymes associated with lipid signaling, such as SCAP, SQLE, LDLR,a n d FASN, were unaffected by CD70-CD27 interaction, implying that the lipid signaling network in Tregs is highly com- plicated than is currently deciphered. With recent advances in elucidating the lipid signaling involved in the functional specia- lization of Tregs in glycolysis-low and lipid-enriched TMEs 20,60,76, we are approaching ever closer to resolving the mystery of metabolic reprogramming and the development of more effective therapies to disrupt lipid signaling in Tregs, which might alleviate tumor progression and synergistically work with other ICBs. However, the metabolic pro file of tumor-in filtrating Tregs varies among cancer types due to the high tissue adaptivity and mole- cular flexibility of these cells 77. Nevertheless, how CD70-CD27 signaling coordinates lipid signaling in other malignancies and identification of the alternative lipid metabolic mechanisms in NPC-infiltrating Tregs are essential subjects for future investigations. Here, we particularly described the molecular mechanism asso- ciated with Treg development, homeostasis, and suppressive activity in one of the EBV-associated and CD70-high malignancies. But how EBV infection facilitates immune evasion of other malignancies, and whether other EBV+ cancers, including B cell lymphoma, T cell lym- phoma, gastric carcinoma, and leiomyosarcoma, exhibited a similar Treg landscape and immunosuppression mechanism as NPC remain insufficiently recognized. To facilitate the translation of CD70 block- ade in other CD70-high solid tumors, we found CD70 also significantly Fig. 5 | Single-cell analysis reveals that CD70-CD27 signaling enhances lipid- driven OXPHOS in Tregs. a The UMAP plot of 19,682 PBMCs with 13 subtypes identified. b The relative abundance of each T cell subtype in the PBMC co-culture system with CD70-NC and CD70-KO C666 cells. c Expression of Treg-specifica n d naïve signatures in Tregs (C1: nTregs + C3: rTregs + C7: eTregs) co-cultured with CD70-NC and CD70-KO C666 cells (n = 3731 and 3264, two-sided Wilcoxon signed- rank test).d Expression of cytotoxic, activation, naïve, and exhausted signatures in CD8+ cytotoxic T cells (C2, C5, and C10) co-cultured with CD70-NC and CD70-KO C666 cells (n = 1997 and 3741, two-sided Wilcoxon signed-rank test).e GSEA revealed enriched OXPHOS and lipid metabolism-associated signaling in NC-Tregs, compared to KO-Tregs (GSEA-computed false discovery rate).f Other top enriched hallmark, KEGG, REACTOME, and GP:BP activities (NES > 1.4/NES <−1.4, P value <0.05 and FDR < 0.2) in NC-Tregs and KO-Tregs (two-sided permutation test, the false discovery rate was computed by GSEA and used to avoid false positive path- ways). The n number represents n biologically independent cells in each group. The data are presented as the median ± IQR (whiskers = 1.5 × IQR, box & whiskers plots) and KDE (violin plots). Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 12 influenced ICB outcomes and T cell dysfunction in melanoma patients, and CD70-KO inhibited orthotopic melanoma growth without inter- fering with tumor cell proliferation. Likewise, future studies should focus on delineating how CD70-CD27 signaling enhances the sup- pressive activity and reprograms lipid metabolism in Tregs in other CD70-high solid tumors, such as melanoma. CD70-CD27 signaling plays a bifacial role in regulating T cell responses in the TME. Previous literature has reported that CD70 can reduce immune surveillance by inducing T cell exhaustion, Treg acti- v a t i o n ,N Kc e l ld e p l e t i o n ,a n dm a c r o p h a g er e c r u i t m e n ti nm u l t i p l e malignancies, whereas some other studies have suggested that the CD70-CD27 axis also contributes to the proliferation and survival of 10 4 10 5 10 6 10 5 10 6 10 6.5 10 4 10 5 10 6 10 5 10 6 10 6.5 0 2 4 6 8 0.25 0.5 1 2 4 8 16 32 64 128 0 2 4 6 8 10 12 14 Intracellular cholesterol (μM) 12k 14k 16k 18k 20k 22k Median filipin III intensity Non-Treg rTreg eTreg a d i Scaled module score NC KO Effector Treg 0.5 0.0 -0.5 Follicular helper T Resting Treg Central memory T IFN-induced naive T Naive Treg SELL naive T AQP3 naive T Scaled module score STAT5A Normalized fold change CREB1CEBPBFASN JUN NR4A1NR4A2ECHS1FABP5 SDC1MDH 1 MDH2 ATF 4 RAC 2 ECH1 EGR1 CD36TBX 21 MITF ACA CA 1.0 TCF7 naive T IL7R naive T Lipid signaling in CD4+ T subtypes Lipid signaling in Tregs Wilcoxon P < 2.2e-16 Wilcoxon P < 2.2e-16 Wilcoxon P < 2.2e-16 STAT5A UQCRFS1 MDH1 SOAT1 PDCD1 UQCRQ ECHS1 NR4A1 TNFRSF18 CREB1 CYC1 HMGCS1 SREBF2 ECH1 0.20.30.40.50.6 GTEx whole blood cohort 12 13 14 15 16 17 18 24 26 28 30 32 34 Lipid signaling score Suppressive score 02 0 4 0 60 0.0 1.0 Time (months) P value = 1.6E-10 r = 0.56 Progression-free survival Lipid signaling score-High (n=29) Lipid signaling score-Low (n=59) Logrank P = 0.01 FOXP3CD70 CTLA4 CD25 CD27 0.2 0.4 0.6 0.8 UQCRFS1UQCRQ CYC1 SQLESCA P GGPS 1 Normalized fold change LDLRHMGCR CD44SOAT1SOAT2 SREBF1 Time of differentiation (hours) Lipid Suppressive Naive CD25 IL7R Normalized expression/score r=0.95 P=2.535E-5 Pearson’s correlation -0.40 -0.20 0.00 0.20 0.40 j l k m n o FNT B PDCD 1 CPT1A 0 2 4 6 8 SREBF1SREBF2 SCARB1 HMGCS1 PGGT 1B TNFRSF 18 CCDC5BLGAL S3 CTNM B1 PPA RA PPA RG GSTP1 0.0 0.2 0.4 0.6Oil O red (absorbance at 492 nm) 0 2 4 6P = 5.318E-4 NC KO Cusatuzumab +- NC KO Cusatuzumab +- cb Cholesterol Glycerol Adenosine Succinate 2-HG Malate Fumarate Citrate Aspartate Glutamine Changes in polar molecules (CD70-KO/CD70-NC) Total fatty acids C12:0 Lauric acid C14:0 Myristic acid C16:0 Palmitic acid C18:0 Stearic acid C18:1 Oleic acid C18:2 Linoleic acid C20:0 Arachidic acid C20:4 n6 Arachidnic acid C22:0 Behenic acid 50% 100% 150% 200% Changes in fatty acids (CD70-KO/CD70-NC) CD70-NC CD70-KO PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 1.84% Q3 73.59% PE-H APC-H 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 10 1 10 2 10 3 10 4 10 5 10 6 10 7.2 Q1 2.25% Q3 70.18% CD70-NC+lipid-depleted medium CD70-KO+lipid-depleted medium PE-FOXP3APC-CTLA4 NC KO FOXP3+ fraction (%) 0 1 2 3 4 5 NC KO FOXP3+/CTLA4+ fraction (%) Q2 3.00% Q4 24.57% Q2 2.93% Q4 21.65% 6600X 11500X NC 500nm 500nm 500nm 500nm KO 6600X 11500X 500nm 500nm 500nm 500nm 0.00 0.05 0.10 0.15 0.20 0.25Oil O red absorbance NC KO 0 15 30 45 60 75 90 ATP pmol/min/1×104 cells GlycoATP mitoATP 17.61% 82.39% 3.21% 96.79% NC KO 0 10 20 30 40 50 NC KO ATP pmol/min/1×104 cells GlycoATP mi toATP 97.14% 97.96% 2.86% 2.04% TGF-β Concentration (pg/mL) 0 2 4 6 8 10 IL-10 CD70-NC CD70-KO 0 100 200 300 400 500 Adenosine NC KO Relative survival (%) FITC-JC-1 monomersPE-JC-1 aggregates NC KO 68.50% 40.07% 0 20 40 60 80 NC KO 0 5 10 15 CD70-NC CD70-KO JC-1 aggregate fraction (%) NCKO Median intensity (×104) NC KO 0 20 40 60 10 20 30 40 50 ACADVL ACADM 10 12 14 16 18 20 22 1.0 1.5 2.0 2.5 HADHA ef g h p qr s tu v w x JC-1 aggregate fraction (%) 0.5 1.0 1.5 2.0 0.0Concentration (nmol/mL) 0 10 20 30 P = 6.689E-5 P = 0.006 P = 0.025 P = 0.002 P = 2.229E-5 P = 4.364E-4 P = 0.004 P = 0.005 P = 0.010P = 0.016 P = 0.028 P = 0.003 P = 0.603 P = 0.636 P = 0.017 P = 0.005 P = 0.626 P = 0.509 P = 0.006 P = 3.718E-4 P = 0.024 P = 0.010 P = 1.524E-4 P = 0.020 P = 0.002 P = 0.007 P = 0.011 P = 0.522 P = 0.587P = 0.102 P = 0.003 P = 2.494E-5 P = 0.017 P = 0.739 P = 0.010 P = 0.001 P = 0.037 P = 0.999 P = 0.019 P = 0.006 P = 0.003 P = 0.004 P = 1.247E-4 P = 0.001 P = 8.426E-4 P = 5.766E-6 P = 0.668 P = 1.347E-4 P = 0.715 P = 0.007 P = 0.004 P = 0.010 P = 0.031 r=-0.83 P=0.003 r=-0.78 P=0.007 r=0.92 P=2.053E-4 P = 0.071 P = 0.650 P = 0.067 P = 0.539 P = 0.133 P = 0.088 P = 0.539 P = 0.808 P = 0.006 P = 0.109 P = 0.819 P = 0.107 Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 13 CD8+ T cells. In the present study, we have found the effector function of CD8 + T cells is elevated upon CD70-KO and inhibition, due to impaired Treg-mediated immunosuppression, whereas the direct influence of CD70-CD27 signaling on CD8 + T cells is negligible 41.W e have speculated that NPC has a CD4/CD27-enriched TME where the effect of CD70-CD27 interaction overwhelmingly takes place in CD4+ naïve T cells and Tregs. In addition, considering the high CD27 expression in B cell lineages and high B-cell in filtration in the NPC microenvironment, the study is limited to only investigating the interaction between NPC cells and T cells with active or blocked CD70- CD27 signaling, as PBMC-engrafted humanized mice only generate a CD3+ T cell-enriched immune system. In the future, a more compre- hensive study of CD70-CD27 signaling in NPC should utilize huma- nized mice generated by transplantation of human CD34 + hematopoietic stem cells (HSCs), which can induce a more systematic humanized immunity consisting of T cells, B cells, NK cells, and myeloid cells. In conclusion, our multiomics-driven functional investigations provide pre-clinical insights into how Tregs are developed, activated, and sustained in the NPC microenvironment, and show that CD70 inhibition is a therapeutically feasible approach to overcome the immunosuppressive TME, synergistically enhancing the ef ficacy of anti-PD-1 treatment. The anti-CD70+ anti-PD-1 combination therapy might produce particular bene fits in patients with advanced or treatment-resistant NPC, which optimizes the conventional manage- ment of these patients. We also consider that the multiomics NPC T-cell cohort, along with the established functional modules, such as Treg suppression, T cell naiveness, and Treg lipid signaling modules, can be applied in future studies of immunosuppression and metabolic reprogramming in tumor-infiltrating T cells in other malignancies or in routine clinical applications for the quantitative evaluation of immu- notherapy response. Finally, this lipid-driven immunosuppressive mechanism induced by CD70-CD27 signaling opens CD70-targeted precision therapy as additional avenue for NPC patients, and is potentially feasible and effective in patients with melanoma, depend- ing on their specific TME landscapes. Methods This research complies with all relevant ethical regulations, approved by the ethical committees at the University of Hong Kong, the Uni- versity of Hong Kong-Shenzhen Hospital, the Hong Kong Red Cross, a n dt h eC o m m i t t e eo ft h eU s eo fL i v eA n i m a l si nT e a c h i n ga n d Research at the University of Hong Kong. Patient enrollment The study was approved by the ethics committee at the University of Hong Kong, the University of Hong Kong-Shenzhen Hospital, and the Hong Kong Red Cross. We complied with all related ethical regulations. Written informed consent was obtained from healthy blood donors and all patients with primary NPC and non-malignant nasopharyngeal inflammation for their tissues to be used in the spatial transcriptome sequencing, IHC/IF staining, flow cytometry analysis, and primary culture in this study. Sample aggregation and batch effect correction in single-cell sequencing The raw fastq data of 50 samples from Gong et al., Liu et al., and Chen et al. NPC single-cell cohorts were processed and aggregated by Cell Ranger (v 6.1.2, 10x Genomics). Batch effects among single cells ori- ginated from different studies were corrected by Harmony (v 1.0, Broad Institute). After batch effect corretion in the aggregated data, no significant batch effect was observed across the original studies (Supplementary Fig. 1b). Quality control, normalization, and clustering of the aggregated single-cell data The gene expression matrix metadata was converted to a Seurat object using Seurat R package (v 4.0). Single cells with unique feature counts >4000 or <200, or >15% mitochondrial counts, werefiltered out from the downstream analysis. Doublets were removed by DoubletFinder (v 2.0). From the remaining 357,206 cells, the gene expression dataset was normalized, scaled, and subsequently dimensionally reduced based on 5000 variant genes and principal components ( n =3 0 ) , determined by the Elbow plot. Major cell lineages, including tumor cells, T cells, B cells, plasma cells, NK cells, myeloid cells, mast cells, fibroblasts, and normal epithelial cells, were projected on the two- dimensional UMAP representation and annotated using well- recognized signatures. To further identifyfiner subtypes within the T cell lineage, we performed the second-round UMAP reduction on 201,746 T cells with principal components (n = 30) and a resolution of 2. The doublets were further identi fied and removed by both graph- based clustering and DoubletFinder. The second-round UMAP reduc- tion revealed 41 distinct subtypes in 189,750 T cells. Pseudotime developmental trajectory in CD4+ Tc e l ls u b t y p e s Monocle 3 was applied to determine the developmental processes within the CD4+ T subpopulations78. The Seurat object of T cells was converted into a cell data set (cds) object and the original UMAP clusters were retained based on implemented functions available in Seurat wrappers. Subsequently, Monocle 3 learned the trajectory graph and ordered cells in pseudotime with a specified root node (C5- SELL-CD4+ naïve T cells) in the converted cds object. The pseudotime trajectory and development lineages were then projected on the ori- ginal UMAP plot. Cells with higher pseudotime indicated that they were in a later developmental stage. Two-sided Pearson correlation analysis and polynomial regression analysis were performed to eval- uate the change of Treg and T FH-specific transcription factors FOXP3 and BATF expression during Treg and TFH developmental processes. Fig. 6 | Lipid signaling promotes Treg development, homeostasis, and sup- pressive activity.Intracellular lipid (a) and cholesterol (b)i nc o - c u l t u r e dC D 4 + naïve T cells (n = 3, two-sided unpaired t test). c Median Filipin III intensity in each cell subtypes (n = 3, two-sided unpaired t test). d The change of intracellular metabolites in co-cultured CD4+ naïve T cells (n =1 ) .e Electron microscope images of co-cultured CD4+ naïve T cells. f OXPHOS and glycolysis ATP from co-cultured CD4+ naïve T cells (n = 4, two-sided unpaired t test).g, h JC-1 aggregates/monomers in co-cultured CD4+ naïve T cells (n = 3, two-sided unpaired t test).i mRNA changes of lipid signaling-associated genes in co-cultured CD4+ naïve T cells (n =3 ,t w o - sided unpaired t test).j Lipid signaling scores in Tregs (n = 3731 and 3264, two-sided Wilcoxon signed-rank test).k Lipid signaling scores in CD4+ T cell subtypes (n =5 0 , two-sided Wilcoxon signed-rank test).l Changes of T-cell module scores and genes during Treg differentiation (n = 3, two-sided Pearson correlation analysis). m Pearson correlation (two-sided) between lipid and suppressive scores (n = 112). n Progression-free survival for NPC patients (highn =2 9 ,l o wn = 59, two-sided log- rank test). o Pearson correlations (two-sided) between lipid/Treg-related genes in whole blood samples (n =3 3 7 ) .p, q Immunophenotyping of co-cultured CD4+ naïve T cells without lipid supplementation (n = 3, two-sided unpaired t test).r The change of immunosuppressive factors in lipid-depleted co-culture (n =3 ,t w o - s i d e d unpaired t test). s T-cell cytotoxicity in lipid-depleted co-culture (n = 5, two-sided unpaired t test). t Intracellular lipid in co-cultured CD4+ naïve T cells without lipid (n = 3, two-sided unpaired t test). u OXPHOS and glycolysis ATP from co-cultured CD4+naïve T cells without lipid (n = 4, two-sided unpaired t test).v JC-1 aggregates/ monomers in co-cultured CD4+ naïve T cells without lipid (n = 3, two-sided unpaired t test).w Median intensity of FA metabolism enzymes in co-cultured CD4+ naïve T cells without lipid (n = 3, two-sided unpaired t test).x Electron microscope images of co-cultured CD4+ naïve T cells without lipid. The n number representsn biologically independent samples/cells/experiments. The data are presented as the mean ± SD (bar plots), median ± IQR (whiskers = 1.5×IQR, box & whiskers plots), and KDE (violin plots). Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 14 Establishment of functional modules based on single-cell data Based on MAST analysis and functional results, we identi fied and selected genes that were most representative to Treg suppressive activity, T-cell naiveness, and lipid signaling, listed in Supplementary Fig. 1f, h, and f. We subsequently used these signatures to establish generalized binomial linear regression models in the eTreg cluster for constructing the Treg suppressive and lipid signaling modules, and in naïve T cell clusters for constructing the T cell naïve module, via the glm function in R (v 4.0). The glm function assigned each single cell with a categorical variable (1 and 0), stating whether it belongs to the designated cluster or not. The binomial linear regression model cal- culated a distinct coef ficient and P value for each representative chr19: 6,590,190 - 6,591,688 (+) 1498 bp 6,590,938 6,591,187 6,591,4366,590,6896,590,440 CD70 0.0 0.48 NPE footprints 0.0 0.48 EBV+ NPC footprints 0.0 0.48 EBV- NPC footprints 0.5 1.0 1.5 0.4 0.8 1.2 1.6 CD70 expression NFKB2 expression EBV- NPCEBV+ NPC 0 1 2 Expression Level CD70, Wilcoxon P < 2.2e−16 NFKB2, Wilcoxon P < 2.2E−16 (EBV+ NPC vs EBV- NPC) Expression Level P value < 2.2E-16 r = 0.51 TF binding score = 10.88 NPC vs. NPE EBV+ vs. EBV- EBV- vs. NPE a b c e NFKB2 0.0 0.5 1.0 1.5 2.0 CD70-luc NFKB2 + -- CD70-mut-luc -+ - -- ++-+ Firefly/Renilla luminescence h °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °°°° °° °° °° °° °° °° °° °° °° °° °° °° °°°° °° °° °° °° °° °°°° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °°°° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° °° 7.5 8.0 8.5 9.0 5.5 6.0 6.5 7.0 7.5 P value = 8.9E−16 EBV-transformed lymphocytes r = 0.68 CD70 expression NFKB2 expression f ZNF148 RELA REL NFKB2 NFKB1 MSGN1 CEBPD SP2 SP1 KLF5 TFAP2C ZKSCAN5 MYCN NFYB ZNF423 1.5 1.0 0.0 0.5 Log2 FC footprint scores ZNF148 RELA REL 0.6 Pearson’s correlation to CD70 0.4 0.2 0.0 -0.2 -0.4 -0.6 NFKB1MSGN1CEBPD SP1 SP2 KLF5TFAP2CZKSCAN5 MYCN NFYBZNF423 GSE102349 GSE68799 d Patient 1 Patient 2 Impaired immunity OXPHOS↑ EBV infection CTLA4Lipid signaling ↑ ↑ PD-1↑ Regulatory T cell EBV+ NPC cell sCD27 NFKB2Chromatin Lipid signaling ↑ CD4+ naive T cell CD70 CD27 Proliferation Cytotoxicity FOXP3 ↑ ↑ CD8+ T cell Granzyme A Perforinremodeling CD70 FOXP3 Suppressive activity ↑ anti- CD70 anti- PD-1 Lineage determination ↑ IL-10 n NFKB2 TNF-α, IFN-γ TGF-β Adenosine EBV+ EBV - 0 20 40 60 80 CD70+ fraction (%) P = 1.148E-6 shNFKB2 0.0 0.2 0.4 0.6 0.8 1.0 Relative survival 24 hrs 48 hrs shNC CD70+ shNC: 76.39% shNFKB2: 43.23% shNC shNFKB2 NFKB2 knockdown in C666 Count (%) 10 3 10 4 10 5 10 6 0 20 40 60 80 100 CD70+ Vehicle: 76.04% SN-52: 51.09% Vehicle SN-52 PE-CD70 SN-52 treatment of C666 SN-52 Relative survival Vehicle Count (%) 10 3 10 4 10 5 10 6 0 40 80 110 0 20 40 60 80CD70+ fraction (%) shNC shNFKB2 Vehicle SN-52 0 1 2 3 4 EBV+ NPC MyeloidEBV- NPC T cell B cell Fibroblast Epithelial NK cell 0.00 0.15 0.30 0.45 0.60 0.75 0.90 24 hrs 48 hrs APC-CD70 Development 0.0 0.5 1.0 1.5CD70 mRNA fold change shNC shNFKB2 Vehicle SN-52 g i j k l m P = 1.944E-6 P = 2.895E-6 P = 8.590E-4 P = 3.874E-4 P = 0.023 P = 0.001 P = 1.858E-6 P = 0.001 P = 4.455E-7 Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 15 signature based on its expression weight across all clusters. Only genes with p value lower than 0.05 were used in constructing the functional modules. The functional scores were then computed based on nor- malized gene matrix in bulk RNA-seq cohorts and single-cell cohorts, representing the quantified extent of Treg suppressive, T cell naïve, and lipid signaling activities. Processing of RNA-seq data The NPC RNA-seq datasets used in this study included: GSE102349 (tumor n = 112, one sample was removed by failing the quality control, and 88 samples with complete follow-up data were included in the subsequent survival analysis), GSE68799 (tumorn =4 2 ,n o r m a ln =4 ) , GSE13597 (tumorn =2 5 ,n o r m a ln = 3), GSE53819 (tumorn =1 8 ,n o r m a l n = 18) and GSE118719 (tumor n = 7, normal n = 4). The read counts in each RNA-seq cohort were generated by HTSeq (version 0.9.1) and normalized by DESeq2 (version 1.22.2). The quality of RNA-seq data was evaluated by Picard metrics (version 2.17.4) and RSeQC (version 2.6.4). Only the samples with at least 60% of reads mapped to coding regions were included in the downstream analysis. Single-cell gene set variation analysis (GSVA) The gene count data matrix with cluster annotation information was constructed in eTregs, rTregs, and nTregs. The gene list files were parsed from http://www.gsea-msigdb.org/gsea/downloads.jsp.G S V A on the single-cell data from the three Treg subtypes was performed with default settings and Poisson kernel argument, as implemented in the GSVA package (v 1.40.1). The pathway enrichment scores werefirst assigned to individual cells, and subsequently averaged within each Treg cluster and z-normalized across all Treg clusters. Isolation of peripheral blood mononuclear cells from healthy donors The collection of peripheral blood from healthy donors was approved and performed by the Hong Kong Red Cross. All donors formally consent for their blood to be used in experiments and analysis mentioned in this study. The collected blood was transferred from the Hong Kong Red Cross to the laboratory at room tempera- ture. The whole blood was mixed with citrate phosphate dextrose (CPD) and 0.5% EDTA in phosphate-buffered saline (PBS) with a 1:1 ratio. Subsequently, 38 mL blood mixture was transferred to each 50 mL conical tube with 12 mL Ficoll-Paque PLUS (Cytiva). The blood mixture was centrifuged at 400× g with the lowest acceleration and deceleration speed for 25 mins at room temperature. The PBMCs were carefully transferred into new 50 mL conical tubes and cen- trifuged at 300×g for 10 mins. The supernatant was aspirated, and PBMCs were washed with 50 mL PBS containing 1% FBS (Gibco) and 0.5% EDTA, and were centrifuged at 300×g for 10 mins. 2 mL red blood cell lysis buffer (Sigma) was added per tube, and the PBMC mixture was incubated for 5 mins at room temperature. 10 mL RPMI- 1640 supplemented with 10% FBS and 1% PS (Gibco) was added to neutralize the lysis buffer in each tube. The PBMCs were centrifuged at 300× g for 5 mins and were ready for isolation and activation. The remaining PBMCs were re-suspended in the freezing medium con- taining 90% FBS and 10% DMSO (Sigma) and frozen at −80 °C using the vertical gradient freeze method. The frozen PBMCs were trans- ferred into liquid nitrogen for long-term storage. Freshly isolated PBMCs and PBMCs that had been frozen for less than one month were used in the study. Magnetic isolation and TCR activation of CD4+ naïve T cells from PBMCs CD4+ naïve T cells were magnetically isolated from fresh PBMCs using the naive CD4 + T cell isolation kit II, human (Miltenyi), as per manu- facturer’s instructions. The purity and CD27 expression of isolated CD4+ naïve T cells was determined via flow cytometry by staining the isolated T cells with FITC anti-human CD45, APC/Cy7 anti-human CD45RA, and APC anti-human CD27 antibodies (BioLegend). The CD4+ naïve T cells were activated in a 96-well plate with a density of 3×10 5 cells/well in 150μLT e x M A C S™ GMP medium (Miltenyi) supplemented with 10% heat-inactivated FBS (HI-FBS), 1% PS, and 50 μM 2-mercaptoethanol (Gibco), in the presence of 10μg/mL plate-coated anti-human CD3 (TONBO) or 1:100 T Cell TransAct™, human (Miltenyi) for 2-3 days, as per manufacturer’si n s t r u c t i o n s . Cell lines and culture Human NPC cell lines C666 (EBV+) and NPC43 (EBV+ and EBV−), as well as human NPE cell lines NP460 and NP69 were provided by Professor George Sai-Wah Tsao at the University of Hong Kong. The 293FT cell line were purchased from Invitrogen (R70007). All NPC and NPE cell lines were cultured in RPMI-1640 (Gibco) supplemented with 10% FBS (Gibco) and 1% PS (Gibco), and the 293FT cell line were cultured in DMEM (Gibco) supplemented with 10% FBS (Gibco), 1% PS (Gibco), 1× MEM amino acids solution (Gibco) and 1 mM sodium pyruvate (Gibco). All cell lines were incubated in a humidified incubator at 37 °C with 5% CO2 supply. All the cell lines used in the study have been tested to be authentic by using STR pro filing and free of mycoplasma contamination. Establishment of in vitro transwell-based and direct co-culture systems NPC and NPE cells were seeded into a 24-well plate with a density of 5×104 cells/well in 1 mL complete RPMI-1640 24 hours prior to adding CD4+ naïve T cells. For the transwell-based co-culture system, 1 × 10 5 naive CD4+ T cells were seeded into a 0.3 cm2 transwell insert with 1μm pore size, avoiding T cell migration but allowing cytokine exchange. For the direct co-culture system, 2 × 10 5 naive CD4 + T cells were directly seeded into the 1.9 cm2 24-well plate together with C666 cells or NP69/NP460 cells. The negative control group has only 2 × 10 5 CD4 + naïve T cells per well without any NPC and NPE cells seeded in the 24-well plate. Cells were cultured in 1.5 mL TexMACS ™ GMP Fig. 7 | EBV infection increases chromatin accessibility at the CD70 promoter and promotesCD70 transcription via NFKB2. aLeft, the single-cell expression of CD70 in EBV+ and EBV- NPC cells (tumor samplen = 36, two-sided Wilcoxon signed- rank test); Right, the quantified CD70+ cell fractions in EBV+ and EBV- NPC43 cells (n = 3, two-sided unpaired t test).b The chromatin accessibility measured by ATAC- seq at the CD70 promoter in EBV+, EBV- NPC cells, and NPE cells.c Top 15 enriched TFs with the highest transcriptional activity in EBV+ NPC cells. d The Pearson cor- relation (two-sided) between each TF and CD70 in GSE102349 (n = 112) and GSE68799 (n = 42) cohorts. e The spatial co-localization ofNFKB2 and CD70 expression in down-sampled spatial spots (n = 100, two-sided Pearson correlation analysis with the 95% confidence band). f The Pearson correlation (two-sided) between NFKB2 and CD70 expression in EBV-transformed lymphocytes (n =1 0 7 , with the 95% confidence band).g Expression ofNFKB2 across major cell lineages in the NPC scRNA-seq cohort (n = 50, two-sided Wilcoxon signed-rank test).h Relative firefly luminescence normalized by renilla luminescence in each experimental group (n = 3, two-sided unpaired t test).i, j The change of CD70 + fraction in shNC, shNFKB2, vehicle-treated, and SN-52-treated C666 cells (n = 3, two-sided unpaired t test). k The 18s-normalized mRNA fold change ofCD70 in shNC, shNFKB2, vehicle- treated, and SN-52-treated C666 cells (n = 3, two-sided unpaired t test). l T-cell cytotoxicity measured in the shNC and shNFKB2 PBMC co-culture systems (n =5 , two-sided unpaired t test). m T-cell cytotoxicity measured in PBMCs co-cultured with vehicle and SN-52-treated C666 cells (n = 5, two-sided unpaired t test). n The schematic illustration of the molecular mechanism of the EBV-NFKB2-CD70-CD27 signaling axis and feedback loop between NPC cells, CD4+ naïve T cells/Tregs, and CD8+ T cells in the NPC microenvironment. Then number representsn biologically independent spots/samples/experiments in each group. The data are presented as the mean ± SD (bar plots), median ± IQR (whiskers = 1.5 × IQR, box & whiskers plots), and KDE (violin plots). Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 16 medium (Miltenyi) supplemented with 10% HI-FBS, 1% PS, 50 μM 2-mercaptoethanol (Gibco), 25 IU/mL IL-2 (Miltenyi) and 2 ng/mL TGF- β1 (PerproTech) for 3 days. Immunophenotyping of CD4+ and CD8+ Tc e l l sb yflow cytometry After 3-day co-culture, CD4+ T cells were collected from the super- natant, washed with cell staining buffer (BioLegend), and centrifuged at 300×g for 5 mins. Subsequently, the T cells were resuspended in cell staining buffer and stained with desired surface marker antibodies for 20 mins on ice. T cells were washed with 2 mL cell staining buffer and centrifuged at 300× g for 5 mins. Subsequently, the T cells were fixed and permeabilized using the true-nuclear transcription factor buffer set (BioLegend), as per manufacturer ’s instructions. Nuclear factor FOXP3 and intracellular CTLA4 were stained with the PE/Paci ficB l u e anti-human FOXP3 antibody (BioLegend) and APC anti-human CTLA4 antibody (Biolegend), respectively, as per manufacturer’s instructions. T h es t a i n e dc e l l sw e r es e q u e n t i a l l yw a s h e dw i t h1 ×P e r mb u f f e ra n dc e l l staining buffer (BioLegend) and then suspended in cell staining buffer for flow cytometry analysis. Surface GITR, ICOS, and PD-1 were stained with PE/Cy7 anti-human GITR (BioLegend), FITC anti-human ICOS (BioLegend), and APC anti-human PD-1 (BioLegend), as per manu- facturer’si n s t r u c t i o n s . For evaluating cytotoxicity of CD8+ T cells in the PBMC co-culture system, co-cultured PBMCs were resuspended in cell staining buffer and stained with APC anti-human CD8 (BioLegend) for 20 mins on ice, and then fixed and permeabilized using the Cyto-Fastfix-perm buffer set (BioLegend), as per manufacturer ’s instructions. Intracellular Granzyme A and Perforin were stained with PE anti-human granzyme A antibody (BioLegend) and PE anti-human perforin antibody (BioLe- gend) separately, as per manufacturer ’si n s t r u c t i o n s .N o v o C y t e Quanteon flow cytometer (Agilent) was used to detect conjugated fluorochrome intensity, and NovoExpress software (v 1.5.6, Agilent) was used to analyze the flow cytometry data. Detailed information on antibody catalog numbers, clones, and concentrations is listed in Supplementary Table 5. Enzyme-linked immunosorbent assay (ELISA) Cell culture supernatants from co-culture assays were collected after 48- or 78-hour co-culture. Subcutaneous tumors were dissected from humanized mice and snap-frozen in liquid nitrogen. The frozen tumors were minced in liquid nitrogen and dissociated in RIPA buffer (Sigma) + protease inhibitor (Sigma). The total protein in the tumor lysates was quantified using the BCA method by the CLARIOstar Plusfluorescence plate reader. Standard curves for our interested cytokines, including IL-10 (Abcam), TGF-β (Invitrogen), TNF-α (Invitrogen), IFN-γ (Abcam), and CD27 (Invitrogen), were determined by the standards supple- mented by the manufactures. The c oncentration of our interested cytokines in cell culture supernatants and tumor lysates from different experimental groups were quantified based on a measurement wave- length of 450 nm, a reference wavelength of 620 nm, and their corre- sponding standard curves. RNA extraction and quantitative reverse transcription PCR (qRT-PCR) Total RNA was extracted using the TRIZOL Reagent (Takara) and cDNA was synthesized by reverse transcription using the PrimeScript RT Reagent Kit with gDNA eraser (Takara). 0.5 ng cDNA was mixed with qRT-PCR primers (BGI genomics) and SYRB Green (Takara) in a 10 µL volume and analyzed by LightCycler 480 II Real-time PCR Detector (Roche). The normalized mRNA fold change 2 -ΔΔCt was cal- culated by normalizing Ct values of corresponding genes to the internal reference gene 18s, and then normalizing toΔCt values in the negative control group. The primer sequences were listed in Sup- plementary Table 3. Cell-cell communication analysis CellphoneDB (v 2.0) and CellChat (v 1.0) were applied to quantify and visualize the statistically and biologically enriched cell-cell commu- nication via ligand-receptor bindings among CD4+ naïve T cells, Treg subtypes, NPC, and NPE cells. The count file and meta file for Cell- PhoneDB were constructed by extracting the raw gene count and cell annotation from the Seurat object. Single-cell expression matrix and cluster annotation of the selected subtypes were processed by Cell- phoneDB in python (v 3.6). CellChat utilized the Seurat object in the R environment to identify overexpressed interacting pairs in each cluster and compute communication probability using the law of mass action. Cell-cell interacting pairs identified by CellChat in fewer than 10 cells were removed from the downstream analysis. To com- pute total cell-cell communication strength and differentially expressed interaction signals in Treg subtypes from primary NPC, INP, and NPC peripheral samples, CellChat analysis was performed on scRNA-seq data from these three sample origins separately, and merged via the mergeCellChat function. The default parameters were used throughout our interaction analysis described in https:// github.com/Teichlab/cellphonedb, https://htmlpreview.github.io/? https://github.com/sqjin/CellChat/blob/master/tutorial/CellChat- vignette.html. Spatial transcriptome sequencing and analysis Visium spatial transcriptome sequencing was performed on frozen tis- sue sections obtained from 2 patientswith primary and treatment-naïve NPC. Hematoxylin and eosin (H&E) staining was performed and imaged on each frozen section to determine the optimal sequencing location and used as a reference for the downstream spatial analysis. Then, tissue optimization experiments were conducted using the Visium spatial tis- sue optimization slide & reagents kit (10× Genomics) to determine the RNA quality (RIN > 7.0) and optimal permeabilization time. 6-minute tissue permeabilization yielded the highestfluorescence signal in both frozen sections. The frozen tissues were placed onto Visium spatial gene expression slide (10× Genomics), fixed, and permeabilized to release RNA, which bonded to adjacent capture probes. Next, RNA reverse transcription, cDNA synthesis, denaturation and amplification, quality control as well as spatial gene library construction were performed using Visium spatial gene expression reagents kit (10× Genomics). The spatial transcriptome library was sequenced on a NovaSeq 6000 (Illu- mina) and mapped to the human genome (hg38) using Spatial Ranger (v 1.3.1, 10× Genomics). The processed gene-voxel matrices contained spatial information of 1,089 (patient 1) and 1,331 (patient 2) spots and were further analyzed in Seurat (v 4.0). The spatial sequencing data were normalized using the SCTransform function followed by PCA reduction. Since each spatial spot contained multiple cells with varied lineages, we deconvoluted the spatial data using the SCTransformed NPC scRNA-seq data as a reference and estimated the underlying cellular composition in each voxel via the FindTransferAnchors function (normalization method =“SCT”) and the TransferData function (anchorset = anchors identified by the FindTransferAnchors function), implemented in Seurat (v 4.0). The fractions of NPC cells and Tregs in each spatial spot were projected on the slide image. The tumor core was collectively deter- mined by scRNA-seq-based spatial deconvolution analysis and patho- logical examination of the H&E staining images by two professional pathologists (JH and YW). Pearson’s correlation analysis was performed to evaluate the spatial co-local ization between NPC cells/CD70 and Tregs. To spatially map cell types by integrating Visium spatial data with scRNA-seq reference, we used cell2location (version 0.1). Brie fly, cell2location decompose multi-cell spatial transcriptomics data into the abundance estimation of reference cell types in a spatially coor- dinates by applying a principled Bayesian model. First, we build a high- quality reference of cell annotations using our integrated NPC scRNA- seq data, down-sampled to reference scRNA-seq data containing 1000 Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 17 cells per cluster and feed it to cell2location as the input. Second, cell2location perform gene selection and define expression signatures of each cell type. Next, we use the default option of negative binomial regression model in cell2location to estimate the reference cell type signatures and to obtain cell-type locations. Each Visium section was analyzed separately with default parameters, except train_args = ’ max_spochs’:3 0 0 ;‘N_cells_per_location’:3 0 ; ‘detection_alpha’:2 0 0 .A 5% percentile of the posterior distribution of mRNA counts was used to represent the number of mRNA molecules contributed by each cell type to each spot and to be plotted for visualization. Immunohistochemistry (IHC) and immunofluorescence (IF) staining The formalin-fixed paraffin-embedded (FFPE) NPC tissues used for IHC staining were collected from the Department of Pathology, The Uni- versity of Hong Kong-Shenzhen Hospital. The tissue slide was placed in a 70 °C oven overnight, and sequentially immersed in xylene, 1:1 xylene: ethanol, 100% ethanol, 95% ethanol, 75% ethanol, 50% ethanol, and tap water for 2 mins each step to completely dewax. The slides were immersed in 1× antigen retrieval buffer (Dako) for 45 mins at 95 °C, and then cooled down naturally for at least 2 hours. The slides were washed with PBS three times and incubated with 3% H 2O2 for 1 hour at room temperature. The slides were additionally washed with PBS and incubated with blocking buffer for 2 hours at room tem- perature. The slides were incubated with anti-CD70 antibody (Cell Signaling) in PBS containing 1% bovine serum albumin (BSA) at 4 °C overnight. After washing with PBS, the slides were treated with labeled polymer-HRP and DAB substrate using the RealTM EnVision Detection SystemsPeroxidase/DAB, Rabbit/Mouse (Dako), as per manufacturer’s instructions. Subsequently, the slides were washed, stained with hematoxylin (Thermofisher), and dehydrated by sequentially immer- sing the slides into 50% ethanol, 75% ethanol, 95% ethanol, 100% ethanol, 1:1 xylene: ethanol, and xylene. The slides were sealed with DPX mounting medium (Sigma), and the images were taken under a bright-field microscope. For IF staining, the slides were blocked with 3% BSA in PBS for 1 hour at room temperature, and incubated with primary antibodies, including mouse anti-human FOXP3 (Abcam) and rabbit anti-human CTLA4 (Abcam) antibodies at 4 °C overnight. The slides were washed with PBS three times and incubated with secondary antibodies, including Alexa 488 goat anti-mouse (Invitrogen) and Alexa 555 donkey anti-rabbit (Invitrogen) antibodies, for 1 hour at room tem- perature. After being washed with PBS, the slides were incubated with 1:1000 DAPI for 10 mins at room temperature. The slides were sealed with Fluoromount TM (Sigma), and the images were taken by the Nikon’s Digital Eclipse C1 Microscope System. The primary anti- bodies used for IHC and IF staining are listed in Supplementary Table 5. CRISPR-Cas9 CD70 knockout in NPC cells sgRNA sequences specifically targeting human and mouseCD70 were designed from https://cctop.cos.uni-heidelberg.de/.T h ef o r w a r da n d reverse sgRNAs were synthesized by Genewiz (China) and diluted to 100 μm in distilled water. The sgRNA was phosphorylated and annelid using the Anza TM T4 PNK kit (Invitrogen), as per manufacturer ’s instructions. Then, CRISPR-V2 plasmid underwent enzymatic digestion using the Fast Digest Esp3l (Thermo fisher) and was ligated with the phosphorylated sgRNA duplex using the T4 ligase kit (Takara), as per manufacturer’s instructions. The CD70-targeting CRISPR plasmid with ampicillin resistance was transformed in stabl3 (Invitrogen) grown on the ampicillin+ LB agar plate (Invitrogen). The transformed plasmid was extracted and purified using the plasmid extraction kit (Qiagen), as per manufacturer’s instructions. CRISPR-NC plasmid containing a non- specific sgRNA was transformed, extracted, and puri fied using the same method. 1×1 0 6 293FT cells were transfected with the CRISPR plasmids with two packaging plasmids, including PsPAX2 and pMD2.g, at a ratio of 3 μg: 2 μg: 1 μg, mixed with 300 μLO p t i - M E M™, GlutaMAX™ supple- ment (Thermo fisher), 6 μL Lipofectamine 2000 (Invitrogen) and 700 μL Dulbecco ’sM o d ified Eagle medium (Thermo fisher) supple- mented with 10% FBS (Gibco), 1× MEM amino acids solution (Gibco) and 1× sodium pyruvate (Gibco). After 72-hour transfection, lentivirus- containing supernatants were collected and centrifuged at 1100×g at 4 °C for 15 mins. C666 cells/B16-F10 cells were transduced with lenti- virus containing the CD70-targeting CRISPR plasmid or the CRISPR-NC plasmid in complete RMPI-1640 for 48 hours. 10 mM puromycin (Sigma) was used to select for stably transduced C666 cells/B16-F10 cells. Single clones for CD70- C666 cells/CD70- B16-F10 cells were sorted by BD FACSMelody (BD Biosciences) using the single-cell purity sorting mode. CD70-targeting sgRNA sequences used in C666 and B16- F10 cells are listed in Supplementary Table 4. The Treg suppression assay CD4+ naïve T cells were co-cultured with CD70-NC/CD70-KO C666 cells or IgG-treated/cusatuzumab-treated C666 cells for 3 days in 1.5 mL complete T cell culture medium supplemented with 25 IU/mL IL- 2 (Miltenyi) and 2 ng/mL TGF-β1 (PerproTech). After 3-day co-culture, paired CD8+ T cells were freshly isolated from the same donor and labelled with CellTrace TM pacific blue (Invitrogen), and were mixed with the co-cultured CD4+ T cells at a 1:1 ratio in T cell culture medium supplemented with 1:100 T cell Transact (Miltenyi) and 25 IU/mL IL-2 (Miltenyi). After 48-hour co-culture between CD4+ T cells and paired CD8+ T cells, the proliferation of CD8+ T cells was determined by pacific blue intensity in the CD8+ T cell population via NovoCyte Quanteon Flow Cytometer, and proliferation index was quantified by the built-in CFSE analysis module in NovoExpress software. Cell proliferation and survival analysis by the colorimetric XTT assay Without PBMC co-culture, NPC cells were seeded in a 96-well plate at a density of 1000 cells/well/100 μL RPMI-1640 complete medium. The cell proliferation was evaluated by using the XTT assay (Roche), as per manufacturer’s instructions, for a consecutive 5 days. With PBMC co- culture, PBMCs in the supernatant were removed, and the tumor cells were washed with PBS twice to remove the remaining PBMCs and dying cells. Fresh RPMI-1640 complete medium was added to each well with adherent tumor cells. The read absorbance (492 nm as the absorbance wavelength and 620 nm as the reference wavelength) was evaluated by using the XTT assay (Roche), as per manufacturer ’s instructions, at 24 hours and 48 hours post-PBMC co-culture. The relative absorbance was calculated by subtracting the read absorbance from the absorbance of an empty well. Relative cell survival was nor- malized by the relative absorbance of viable cells without PBMC co- culture. Establishment of the NPC-PBMC co-culture system and apop- tosis analysis CD70-NC and CD70-KO C666 cells were seeded into a 24-well plate with a density of 1 × 10 5 cells/well 24 hours prior to adding activated PBMCs. 5 × 105 PBMCs were seeded into the 24-well plate in TexMACS™ GMP medium (Miltenyi) supplemented with 10% HI-FBS, 1% PS, 50μM 2-mercaptoethanol and 25 IU/mL IL-2 (Miltenyi), for 48 hours. After 48 hours, the dead tumor cells were collected and viable adherent tumor cells were dissociated by TrypLE™ (Thermofisher) at 37 °C for 3 mins. The apoptosis of CD45- cells was determined by the PE annexin V apoptosis detection kit I (BD Biosciences) + FITC-CD45 anti-human antibody (BioLegend), as per manufacturer’s instructions. The cellular intensities of FITC-CD45, PE-annexin V and APC-7-AAD was compen- sated and measured by NovoCyte Quanteon flow cytometer and ana- lyzed by the NovoExpress software. Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 18 CFSE staining and analysis CellTrace™ CFSE cell proliferation kit (Invitrogen) was used to stain activated PBMCs with CFSE prior to co-culture with NPC cells, as per manufacturer’s instructions. After 48-hour co-culture, the proliferation of CD8+ T cells was determined by staining the APC anti-human CD8 antibody (BioLegend), and CFSE intensity in the CD8 + subpopulation was measured by NovoCyte Quanteon Flow Cytometer, and pro- liferation index was quantified by the built-in CFSE analysis module in NovoExpress software. Anti-CD19 CAR lentivirus packaging and concentration 293FT cells were seeded at the density of 1×106 cells/well in a six-well plate in the DMEM transfection medium. Cells were transfected with the packaging plasmids psPAX2 (Addgene,) and pMD2.G (Addgene) together with the CAR overexpression plasmid (pLV[Exp]-EGFP-EF1A- CD19-CAR) in a ratio of 1:2:4 using the Lipofectamine 3000 transfec- tion reagent (Invitrogen) according to the manufacturer ’s protocol. The lipid-DNA complex was incubated for 6 hours at 37 °C, 5% CO 2. Next, the cell medium was replaced with a fresh lentivirus packaging medium supplemented with 1× viral boost reagent (Alstem), following the manufacturer’s instructions. The lentivirus was collected at 48 hours and 72 hours post- transfection. Cell medium containing the lentivirus was centrifuged at 1,100×g, 4 °C for 15 min to remove the cell debris. To concentrate the lentivirus particles, we mixed the lentivirus with the precipitation solution (TaKaRa, Lenti-X Concentrator), as per the manufacturer ’s protocols. The mixture was incubated at 4 °C for 30 minutes and centrifuged at 1500×g for 45 minutes at 4 °C. The lentivirus pellet was resuspended in cold PBS to make the 100× concentrated virus. Anti-CD19 CAR-T cells production Fresh CD3+ T cells isolated from human PBMCs were activated with T cell TransACT (Miltenyi) before anti-CD19 CAR lentivirus transfection. Activated human T cells with a cell density of 2 × 106 /mL were cultured with the concentrated lentivirus (50μL/mL) supplemented with 8μg/ mL polybrene and centrifuged at 500× g at 32 °C for 90 min. The cell culture medium was refreshed 24 hours later and the transfected T cells were subsequently cultured in a complete T cell culture medium supplemented with 25 IU/mL IL-2 for additional 48 hours. Anti-CD19- CAR-expressing T cells sorted based on EGFP expression using BD FACS Aria SORP. Inhibition of CD70 by cusatuzumab and CD70-blocking-only antibody treatment The humanized monoclonal anti-CD70 antibody, cusatuzumab and anti-CD70 blocking-only antibody, was purchased from MyBioSource (MBS156578) and AdipoGen (ANC-222-050), respectively. Varied cusatuzumab concentrations, including 1μg/mL, 2.5 μg/mL, 5 μg/mL, and 10 μg/mL, were tested in the in vitro co-culture systems. 5 μg/mL cusatuzumab and the anti-CD70 blocking-only antibody diluted in the complete T cell culture medium was determined as the optimal con- centration to inhibit CD70-CD27 interaction between C666 cells with CD4+ naïve T cells and Tregs. 5 μg/mL IgG1, λ antibody treatment was used as vehicle control in vitro. In addition, 5 mg/kg cusatuzumab was determined as the optimal concentration in the in vivo settings. Mice All animal experiments were approved by the Committee of the Use of Live Animals in Teaching and Research at the University of Hong Kong under protocol 4924-19. NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG) mice and C57BL/6J mice were purchased from the Centre for Comparative Medicine Research, the University of Hong Kong. Female NSG mice at 6 weeks of age were used for PBMC-engrafted immune humanization. Male or female C57BL/6J mice at 4-6 weeks of age were used for mel- anoma inoculation. The maximal end-point tumor size permitted by the ethics committee is 2 cm in diameter and no tumor burden exceeded this limit in our study. Mice were housed on 12-hour light- dark cycles with the ambient temperature at 20 –25 °C and 40 –60% humidity at the Centre for Comparative Medicine Research, the Uni- versity of Hong Kong. Establishment of the PBMC humanized mouse model and xenotransplantation Freshly isolated PBMCs from healthy donors were activated with T Cell TransAct™, human (Miltenyi) for 3 days, and expanded in TexMACS™ GMP medium (Miltenyi) supplemented with 10% HI-FBS and 1% PS, 50 IU/mL IL-2 (Miltenyi) for additional 3 days. 1×10 7 activated PBMCs in 200 μL PBS were injected into 6-week-old female NSG mice via tail-vein injection. 3 × 10 6 CD70-NC and CD70-KO C666 cells were injected subcutaneously into humanized mice on day 9 post-PBMC injection. The humanized immunity started to re-establish one-week post-PBMC injection and was determined on day 14 by staining mouse PBMCs with FITC anti-human CD45 antibody. For the xenotransplanted humanized mouse model, Xeno76 grown on nude mice was dissected into small chunks with an identical volume (2 mm × 2 mm × 2 mm). 4-week-old female NSG mice were anesthetized by intraperitoneal injection of 150μL ketamine and xylazine mixture. The left dorsalflank of each mouse was cut with sterile scissors, and the PDX was placed inside the mouse ’s body. The incision was closed by Nylon. After 2 weeks when the PDX started to grow, 1 × 10 7 activated PBMCs in 200 μL PBS were injected into the xenotransplanted NSG mice. Different regimens, including 5 mg/kg IgG1,λ antibody, 5 mg/kg cusatuzumab, 3 mg/kg camrelizumab, and 5 mg/kg cusatuzumab + 3 mg/kg camrelizumab, were given to the PDX-bearing mice intraper- itoneally once every three days, starting on day 7 post-PBMC injection. Primary NPC cell culture After the endoscopic biopsy, the fresh NPC tissue samples (2–3m m3) were rinsed with RPMI-1640 medium supplemented with 10% FBS and 2% PS three times on ice, and then placed on a 6-well culture plate in 2 mL RPMI-1640 medium supplemented with 10% FBS, 1% PS and 4μM Y-27632 (MCE). The adherent primary NPC cells were cultured for 3 days. Meanwhile, autologous PBMCs were isolated from the paired peripheral blood using the Ficoll-Paque method mentioned above and activated in T cell medium supplemented with 25 IU/mL IL-2 for 3 days. On day 3, the activated PBMCs were added to co-culture with primary NPC cells at a ratio of 4:1 in the T cell medium supplemented with 25 IU/ mL IL-2 and 5 μg/mL cusatuzumab (MyBioSource) or 5 μg/mL IgG1, λ (Sigma) for 48 hours. After then, dead cells were gently washed away by PBS three times, and the viable adherent cells were measured by the colorimetric XTT assay. The immune cells were collected and stained with FITC-CD4, Pacific Blue-FOXP3, and APC-CTLA4 to evaluate Treg fraction changes after cusatuzumab treatment. Establish NPC organoids and the organoid-PBMC co-culture system Xeno76 grown on nude mice was dissociated into single cells by using Liberase™ (Sigma) containing 0.5μg/mL DNAse I (Sigma) at 37 °C for 30 min. Dissociated cells were mixed with ADF+/+/+ (AdDMEM/ F12 supplemented with HEPES, GlutaMax, and Penicillin-Streptomycin) and BME with a ratio of 1:3 and seeded per 24 wells. After solidification, the organoid medium was added. Organoid medium contained B27 supplement, Primocin, 1.0 mmol/L N-acetyl-l-cysteine, 10 mmol/L Nicotinamide, 50 ng/mL human EGF, 500 nmol/L A83-01, 10 ng/mL human FGF10, 5 ng/mL human FGF2, 1 μmol/L Prostaglandin E2, 3 μmol/L CHIR 99021, 1 μmol/L Forskolin, 10 μM Rho kinase inhibitor Y-27632 (MCE), 4% R-spondin, and 4% Noggin. During culturing, the medium was refreshed at most every three days. For passaging, organoids were collected from the plate by disrupting the BME dro- plets with a P1000, collecting and washing in 10 mL ADF+/+/+. Pellet Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 19 was resuspended in 1 mL of 10X TrypLE Express (Thermo fisher) and incubated at 37 °C. Digestion was closely monitored, and the suspen- sion was pipetted up and down every 5 minutes to aid disruption of the organoids. TrypLE digestion was stopped when organoids were dis- rupted into single cells by adding 10 mL ADF+/+/+. Cells were subse- quently resuspended in ice-cold 75% BME in ADF+/+/+ and plated with a ratio of 1:3 to allow ef ficient outgrowth of new organoids. The medium was changed every 2 –3 days, and organoids were usually passaged with a split ratio of 1:3 every 7–10 days. The organoids were gently collected, centrifuged at 300× g for 10 mins at 4 °C, and subsequently mixed with 1 × 105 activated PBMCs i n1 . 5m LA d D M E M / F 1 2s u p p l e m e n t e dw i t h1 0 %H I - F B S ,1 %P S ,5 0μM2 - mercaptoethanol, 25 IU/mL IL-2. After 48-hour co-culture, the orga- noids were dissociated into single-cell suspension using TrypLE Express (Thermofisher). Apoptosis of CD45- organoid-derived NPC cells was determined by the PE annexin V apoptosis detection kit I (BD Biosciences) + FITC-CD45 anti-human antibody (BioLegend), as per manufacturer’s instructions. The cellular intensities of FITC-CD45, PE- annexin V, and APC-7-AAD were compensated and measured by the NovoCyte Quanteonflow cytometer and analyzed by the NovoExpress software. Single-cell sequencing of PBMCs The PBMCs co-cultured with C666-NC and C666-KO NPC cells after 48 hours were isolated from the culture medium by centrifuging at 300× g for 5 mins, and re-suspended in PBS before single-cell encap- sulation. The single-cell encapsulation and library preparation was done at the Centre for PanorOmic Sciences, the University of Hong Kong. The single-cell suspension was converted to uniquely barcoded RNA, using the Chromium single cell 3′ library, gel Bead & multiplex kit, and Chromium single cell chip kit (10× Genomics), as per manu- facturer’s instructions. The libraries were sequenced on a NovaSeq 6000, and mapped to the human genome (hg38) using Cell Ranger (v 6.1.2, 10x Genomics). Gene positions were annotated as per Ensembl build 85 and filtered for biotype. The sample aggregation and analysis were performed using the aforementioned methods. GSEA analysis on Tregs and CD8+ T cells We constructed gct and clsfiles for GSEA analysis based on the single- cell expression matrix and metadata of NC-Tregs and KO-Tregs. GSEA for Windows (v 4.2.3) was downloaded fromhttps://www.gsea-msigdb. org/gsea/downloads.jsp.T h ep a t h w a ye n r i c h m e n ts c o r e si nh a l l m a r k , KEGG, REATOME, and GO:BP gene sets were computed based on Sig- nal2Noise analysis. Oil O red staining and quantification For staining of total intracellular lipid in CD4+ T cells co-cultured with CD70-NC and CD70-KO C666 cells or IgG-treated and cusatuzumab- treated C666 cells, CD4+ naïve T cells after 3-day co-culture were washed by PBS and fixed with 10% formalin for 30 mins at room tem- perature. Thefixed T cells were then washed with 60% isopropanol for 5 mins and stained with 0.5% Oil O red staining solution for 20 mins at room temperature. Then, T cells were washed with ddH 2Oa n d6 0 % isopropanol. Lastly, the Oil O red stain in T cells were extracted by 100% isopropanol for 5 mins. The absorbance of Oil O red was quan- tified on a multi-plate reader at a 492 nm measurement wavelength and a 620 nm reference wavelength. Cellular cholesterol content measurement Cellular-free cholesterol content was measured using the cholesterol cell-based detection assay kit (Cayman). Co-cultured CD4+ T cells were fixed and permeabilized using the true-nuclear transcription factor buffer set (BioLegend), as per manufacturer ’s instructions. Then the cells were stained with filipin III (Cayman), PE anti-human FOXP3 (BioLegend), and APC anti-human CTLA4 (BioLegend). The median filipin III intensity in FOXP3-/CTLA4-, FOXP3 + /CTLA4- and FOXP3+/CTLA4+ cells was measured by the BD Aria SORP flow cyt- ometer. For the oxidation-based quanti fication of cholesterol, total cholesterol was extracted using the cholesterol extraction kit (Sigma), and then oxidized using the Amplex Red Cholesterol Assay Kit (Invitrogen), per manufacturer ’s instructions. The oxidized cho- lesterol was analyzed by the CLARIOstar Plus fluorescence plate reader. Mass spectrometry-based metabolomics in co-cultured CD4+ naïve T cells and supernatant Sample processing and gas chromatography-mass spectrometry (GC- MS) analysis were performed at the Centre for PanorOmic Sciences- Proteomics and Metabolomics Core, the University of Hong Kong. The number of co-cultured CD4+ naïve T cells in each sample was deter- mined by the Countess 3 Cell Counter (Thermo fisher). 4-5 × 10 5 co- cultured CD4+ naïve T cells were collected and washed with ice-cold sterile 0.9% saline 3 times and snap-frozen in liquid nitrogen. 1 mL of co-cultured supernatant was collected and stored in −80°C prior to sample submission. The input cell pellet was then extracted in 2 ml of 0.02 mg/L internal standard in 80% ice-cold methanol in dry ice. The lysate was then centrifuged at 14,000×g for 20 min at 4°C for protein precipitation. 1 ml of supernatant was dried under a gentle stream of nitrogen at room temperature. The dried aliquot was then re- suspended with 500 µl 50% acetonitrile. Each sample was vortexed for 10 s and centrifuged for 2 min at 14,000× g. The supernatant was dried under a gentle stream of nitrogen at room temperature and submitted to derivatization. The dried residue was redissolved and derivatized for 2 hours at 37 °C in 40 μl of methoxylamine hydro- chloride (30 mg/ml in pyridine) followed by trimethylsilylation for 1h o u r a t 3 7° C i n 7 0μlM S T F Aw i t h1 %T M C S .U pt o1µl sample was injected for GC-MS/MS analysis. Data analysis was performed using the Agilent MassHunter Workstation Quantitative Analysis Software (v B.04.xx). Linear cali- bration curves for each analyte were generated by plotting the peak area ratio of external/internal standard against standard concentration at different concentration levels. Analytes were con firmed by com- paring the ratio of characteristic fragment ions in the sample and standard. Thefinal intracellular concentration of targeted metabolites was normalized by the total number of input cells in each sample. The mass spectrometry results are provided in Supplementary Data 3. Electron microscopy scanning CD4+ naïve T cells co-cultured with CD70-NC and CD70-KO C666 cells in the lipid-depleted or normal co-culture culture system after 3 days were washed, gently centrifuged, and re-suspended in PBS, then cells were fixed in 2.5% glutaraldehyde (Sigma-Aldrich), post fixed in 2% osmium tetroxide (Sigma-Aldrich), stained with uranyl acetate (Sigma- Aldrich), and dehydrated with a graded ethanol series. Sections were cut after embedding in Epon Araldite (Sigma-Aldrich), collected on uncoated nickel grids after sectioning and the mitochondria were observed and photographed using Philips CM-100 Transmission Electron Microscope. The sample preparation and electron micro- scopy scanning were performed by the Electron Microscope Unit at the University of Hong Kong. Seahorse real-time ATP rate assay CD4+ naïve T cells were co-cultured with CD70-NC and CD70-KO C666 cells in the complete T cell culture system for 3 days. 16-hour prior to the assay, XFe96 sensor cartridge (Agilent) was hydrated with auto- claved water in a non-CO 2 37 °C incubator overnight. 1-hour prior to the assay, the XFe96 cell culture plate (Agilent) was coated with 30µL 0.1 mg/mL of poly-D-lysine (Gibco) for T cell adhesion. After 3-day co- culture, 1 × 105 CD4+ T cells were seeded into the cell culture plate with 200 μL Seahorse XF RPMI medium (Agilent) supplemented with Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 20 10 mM of XF glucose (Agilent), 1 mM of XF pyruvate (Agilent) and 2 mM of XF glutamine (Agilent) and incubated in a non-CO 2 37 °C incubator for 60 mins. Subsequently, the T cells were washed with the complete Seahorse XF RPMI medium to a final volume of 180 μL. The XFe96 sensor cartridge was added with 200 μL Seahorse XFe96 Cali- brant/well. Following the manufacturer’si n s t r u c t i o n so nt h eS e a h o r s e XF Real-Time ATP Rate Assay (Agilent), 20μL diluted Oligomycin and 22 μL diluted Rotenone + Antimycin A were added into the injection port A and B, respectively. The XFe96 sensor cartridge covered with the injection plate was placed into the Agilent Seahorse XFe96 Ana- lyzer for calibration. When the calibration was finished, replaced the utility plate containing the Seahorse XFe96 Calibrant with the cell culture plate. The ATP production rate in Tregs was dynamically measured by the Agilent Seahorse XFe96 Analyzer. ATP production via OXPHOS and glycolysis was evaluated based on the built-in analysis software. Evaluation of mitochondrialpotential via JC-1 staining CD4+ naïve T cells co-cultured with CD70-NC and CD70-KO C666 cells in lipid-depleted or normal culture medium after 3 days were stained with JC-1 dye (Sigma) for 20 mins in the dark at 37 °C, as per manu- facturer’s instructions. T cells were washed twice with cold washing buffer and centrifuged at 800×g f o r5m i n .A f t e rt h efinal wash, T cells were re-suspend in 0.5 ml cold washing buffer for acquiring and ana- lyzing the PE (JC-1 aggregates) and FITC (JC-1 monomers)fluorescence by ACEA Novocyte Quanteon. Establishment of the lipid-depleted co-culture system For experiments conducted in the lipid-depleted co-culture system, the complete T cell medium was supplemented with 10% heat-inactivated lipid-depleted FBS (Neuromics) instead of regular FBS (Gibco). CD4+ naïve T cells, Treg and PBMC co-culture assays with CD70-NC and CD70- KO C666 cells were performed as described above. Assay for Transposase-Accessible Chromatin using sequencing on NPC and NPE cells 50,000 cells were used for each experiment. Transposase Tn5 enzy- matic digestion was undergone at 37 °C for 30 mins using Illumina Tagment DNA Enzyme and Buffer Small Kit (Cat no. 20034197) fol- lowing the protocol previously described. Sequencing was performed at 150 bp PE Illumina X-Ten platform. The ENCODE ATAC-seq data analysis standards and processing pipeline was followed, and TOBIAS was used for the motif analysis. The signals were adjusted and merged based on the conditions to estimate the log-transformed fold changes of the footprint scores for comparison among the EBV+ NPC cells, EBV- NPC cells, and normal NPE cells. The differential motif analysis was performed based on the JASPAR database. The peaks were identi fied using the ENCODE analysis pipeline. The motif analysis using TOBIAS based on the JASPAR database was performed to evaluate the binding of the potential transcription factors. Site-directed mutagenesis and the dual-luciferase reporter assay 2000 bp (-1500 bp- 500 bp) of the promoter sequence of CD70 were cloned into the pGL3-basic vector (Promega, E1751). For site- directed mutagenesis of the potential binding motif, the identi- fied NFKB2 binding site (GGGGATTCCCC) on the CD70 promoter was mutated with a QuikChange II site-directed mutagenesis kit (Stratagene). The mutation of the construct was con firmed by Sanger sequencing. 293FT cells were seeded in a 24-well plate at the density of 1×10 5/well. The luciferase reporter constructs (5:5:1 mixture of CD70-promoter-luc or CD70-promoter-mut-luc, NFKB2-OE or NFKB2-NC and Renil la plasmid (Promega)) were transfected together into 293FT cells using X-tremeGENE HP reagent (Roche) according to the manufacturer ’s instructions. pGL3-basic vector was used as t he negative control. 48 hours later, the culture supernatant w as aspirated and replaced with 1× passive lysis buffer (Promega), the plate was then incubated at 4 °C for 20 mins for complete cell lysis. Subsequently, 40 μLc e l l lysate was mixed with 20 μL luciferase assay reagent (Promega) in the Lockwell maxisorp plate, and the firefly luminescence was immediately detected using the m icroplate reader. Dual-Glo® Stop & Glo® reagent (Promega) was added to the plate, mixed, and incubated for 3 mins before measuring the Renilla lumines- cence. The firefly luminescence was normalized to the Renilla signals. NFKB2 knockdown via shRNA To characterize the role of NFKB2 on transcriptional regulation of CD70 expression, we performed the small hairpin RNA (shRNA) knockdown in C666 cells. shRNA that specifically targeted NFKB2 was designed, synthesized and inserted into the lentiviral vector pLKO.1. Lentivirus were packaged using 293FT cells as described above. C666 cells were infected with shNC and sh NFKB2 lentivirus for 48 hours, supplemented with 10 μg/mL polybrene (Sigma). Following the lenti- virus infection, puromycin was added in the cell culture medium for the selection and maintenance of shNC and NFKB2-knockdown C666 cells. NFKB2-targeting shRNA sequences used are listed in Supple- mentary Table 6. Statistical analysis and reproducibility Statistical analysis was performed, and plots were generated using GraphPad Prism 8 and IBM SPS S Statistics 22. For data with as a m p l es i z e≤6, bar plots were used and all data points were shown. For data with a sample size >6, box & whiskers plots were used to visualize the data distri bution. Bar plot data were pre- sented as the mean values ± SD. Box & whiskers plots were pre- sented as the median values ± IQR (25 th and 75 th percentiles) and w h i s k e r se n c o m p a s s1 . 5t i m e st h eI Q R ,a sd e t e r m i n e db yt h e Tukey method. Violin plots were presented as kernel density estimations (KDE) annotated with median values ± IQR. For parametric data, P values were evaluated by the two-sided Stu- dent’s t-test. For non-parametric data from scRNA-seq data, P values were evaluated by the two-sided Wilcoxon signed-rank test or Kruskal-Wallis one-way analysis of variance. Two-sided Pearson and Spearman correlation analysis was used to compute the correlation and statistical signi ficance between two variables. In GSEA, both false discovery rates (FDR) and P values were used to determine statistical signi ficance. In survival analysis, P values were evaluated by the two-sided l og-rank test. The microscopic i m a g e ss h o w ni nF i g s .2f, g, 4k, 6e, x and Supplementary Figs. 2g and 4a are representative for three independent experiments which generated similar results. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Data availability The NPC single-cell sequencing data used in the study are publicly available in Gene Expression Omnibus (GEO) under accession numbers GSE150825, GSE150430,a n d GSE162025. The NPC bulk RNA sequen- cing data used in the study are publicly available in GEO under accession numbers GSE68799, GSE102349, GSE53819, GSE13597,a n d GSE118719. The RNA sequencing data for time-resolved induced Treg differentiation used in the study is publicly available in GEO under accession number GSE96538. The HNSCC single-cell sequencing data used in the study are publicly available in GEO under accession num- bers GSE139324and GSE164690.T h er a wa n dp r o c e s s e dV i s i u ms p a t i a l sequencing data of primary NPC tissues and single-cell sequencing data of the co-cultured PBMCs have been deposited in GEO under Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 21 accession numbers GSE200310 and GSE200315. Source data are pro- vided with this paper. The remaining data are available within the Article, Supplementary Information or Source Data file. Source data are provided with this paper. References 1. Yu, H. et al. The global burden of nasopharyngeal carcinoma from 2009 to 2019: an observational study based on the Global Burden of Disease Study 2019. Eur. Arch. Otorhinolaryngol.279, 1519– 1533 (2021). 2 . C h e n ,Y .P .e ta l .N a s o p h a r y n g e a lc a r c i n o m a .Lancet 394, 64– 80 (2019). 3. Liu, Z., Chen, Y., Su, Y., Hu, X. & Peng, X. Nasopharyngeal carci- noma: clinical achievements andconsiderations among treatment options. Front. Oncol. 11, 635737 (2021). 4. Lee, A. W., Ma, B. B., Ng, W. T. & Chan, A. T. Management of naso- pharyngeal carcinoma: current practice and future perspective.J. Clin. Oncol. 33,3 3 5 6– 3364 (2015). 5. Gong, L. et al. The stromal and immune landscape of nasophar- yngeal carcinoma and its implications for precision medicine tar- geting the tumor microenvironment.Front. Oncol. 11,7 4 4 8 8 9 (2021). 6. Fang, W. et al. Camrelizumab (SHR-1210) alone or in combination with gemcitabine plus cisplatinfor nasopharyngeal carcinoma: results from two single-arm, phase 1 trials.Lancet Oncol. 19, 1338– 1350 (2018). 7. Hsu, C. et al. Safety and antitumor activity of Pembrolizumab in patients with programmed Death-Ligand 1-Positive Nasopharyngeal Carcinoma: Results of the KEYNOTE-028 Study.J. Clin. Oncol. 35, 4050– 4056 (2017). 8 . M a ,B .B .Y .e ta l .A n t i t u m o ra c t i v i t yo fN i v o l u m a bi nr e c u r r e n ta n d metastatic nasopharyngeal carcinoma: An International, Multi- center Study of the Mayo Clinic Phase 2 Consortium (NCI-9742).J. Clin. Oncol. 36,1 4 1 2– 1418 (2018). 9. Liu, Y. et al. Tumour heterogeneit y and intercellular networks of nasopharyngeal carcinoma at single cell resolution.Nat. Commun. 12,7 4 1( 2 0 2 1 ) . 10. Gong, L. et al. Comprehensive sin gle-cell sequencing reveals the stromal dynamics and tumor-specific characteristics in the micro- environment of nasopharyngeal carcinoma.Nat. Commun. 12, 1540 (2021). 11. Chen, Y. P. et al. Single-cell transcriptomics reveals regulators underlying immune cell diversity and immune subtypes associated with prognosis in nasopharyngeal carcinoma.Cell Res. 30, 1024– 1042 (2020). 12. Jin, S. et al. Single-cell transcriptomic analysis defines the interplay between tumor cells, viral infection, and the microenvironment in nasopharyngeal carcinoma.Cell Res. 30,9 5 0– 965 (2020). 13. Li, J. et al. Functional inactivation of EBV-speci fic T-lymphocytes in nasopharyngeal carcinoma: implications for tumor immunother- apy. PLoS One 2, e1122 (2007). 14. Nilsson, J. S. et al. Immune phenotypes of nasopharyngeal cancer. Cancers 12, 3428 (2020). 15. Liston, A. & Gray, D. H. Homeostatic control of regulatory T cell diversity.Nat. Rev. Immunol. 14,1 5 4– 165 (2014). 16. Li, C., Jiang, P., Wei, S., Xu, X. & Wang, J. Regulatory T cells in tumor microenvironment: new mechanisms, potential therapeutic strate- gies and future prospects. Mol. Cancer 19, 116 (2020). 17. Pitt, J. M. et al. Resistance mechanisms to immune-checkpoint blockade in cancer: tumor-intrinsic and -extrinsic factors.Immunity 44,1 2 5 5– 1269 (2016). 18. Zhang, Y., Guan, X. Y. & Jiang, P. Cytokine and Chemokine signals of T-Cell exclusion in tumors.Front Immunol. 11, 594609 (2020). 19. Amoozgar, Z. et al. Targeting Treg cells with GITR activation alle- viates resistance to immunotherapy in murine glioblastomas.Nat. Commun. 12,2 5 8 2( 2 0 2 1 ) . 20. Kumagai, S. et al. Lactic acid p r o m o t e sP D - 1e x p r e s s i o ni nr e g - ulatory T cells in highly glycolytic tumor microenvironments.Can- cer Cell 40,2 0 1– 218.e9 (2022). 21. Wang, Z. et al. Role of IFN-gamma in induction of Foxp3 and con- version of CD4+ CD25- T cells to CD4+ Tregs. J. Clin. Invest. 116, 2434– 2441 (2006). 22. van Loosdregt, J. et al. Canonical Wnt signaling negatively mod- ulates regulatory T cell function.Immunity 39,2 9 8– 310 (2013). 23. Mijnheer, G. et al. Conserved human effector Treg cell tran- scriptomic and epigenetic signature in arthritic joint inflammation. Nat. Commun. 12, 2710 (2021). 24. Efremova, M., Vento-Tormo, M., Teichmann, S. A. & Vento-Tormo, R. CellPhoneDB: inferring cell-cell communication from combined expression of multi-subunit ligand-receptor complexes.Nat. Pro- toc. 15,1 4 8 4– 1506 (2020). 25. Jin, S. et al. Inference and analysis of cell-cell communication using CellChat. Nat. Commun. 12, 1088 (2021). 26. Kurtulus, S. et al. TIGIT pred ominantly regulates the immune response via regulatory T cells.J. Clin. Invest.125,4 0 5 3–4062 (2015). 27. Fourcade, J. et al. CD226 opposes TIGIT to disrupt Tregs in mela- noma. JCI Insight 3, e121157 (2018). 28. Smith, S. E., Hoelzinger, D. B., Dominguez, A. L., Van Snick, J. & Lustgarten, J. Signals through 4-1BB inhibit T regulatory cells by blocking IL-9 production enhancing antitumor responses.Cancer Immunol. Immunother.60,1 7 7 5– 1787 (2011). 29. Wu, C. et al. Galectin-9-CD44 interaction enhances stability and function of adaptive regulatory T cells.Immunity 41, 270– 282 (2014). 30. Claus, C. et al. CD27 signaling increases the frequency of regulatory T cells and promotes tumor growth. Cancer Res. 72, 3664– 3676 (2012). 31. Yang, Z. Z., Novak, A. J., Ziesmer, S. C., Witzig, T. E. & Ansell, S. M. CD70+ non-Hodgkin lymphoma B cells induce Foxp3 expression and regulatory function in intratumoral CD4+CD25 T cells.Blood 110,2 5 3 7– 2544 (2007). 32. Hao, Y. et al. Integrated analysis of multimodal single-cell data.Cell 184,3 5 7 3– 3587.e29 (2021). 33. Kleshchevnikov, V. et al. Cell2location maps fine-grained cell types in spatial transcriptomics.Nat. Biotechnol.40,6 6 1– 671 (2022). 34. Kawamura, T. et al. CD70 is selectively expressed on Th1 but not on Th2 cells and is required for Th1-type immune responses.J. Invest. Dermatol. 131,1 2 5 2– 1261 (2011). 35. Jacobs, J. et al. CD70: An emerging target in cancer immunother- apy. Pharm. Ther. 155,1 – 10 (2015). 36. Cillo, A. R. et al. Immune landscape of viral- and carcinogen-driven head and neck cancer. Immunity 52,1 8 3– 199.e9 (2020). 37. Kurten, C. H. L. et al. Investigating immune and non-immune cell interactions in head and neck tumors by single-cell RNA sequen- cing. Nat. Commun. 12, 7338 (2021). 38. Arroyo Hornero, R. et al. CD70 e xpression determines the ther- apeutic efficacy of expanded human regulatory T cells. Commun. Biol. 3,3 7 5( 2 0 2 0 ) . 39. Anta, N. et al. Chronic CD27-CD70 costimulation promotes type 1- specific polarization of effector Tregs.Front. Immunol. 14, 1023064 (2023). 40. Hendriks, J. et al. CD27 is required for generation and long-term maintenance of T cell immunity. Nat. Immunol. 1,4 3 3– 440 (2000). 41. Flieswasser, T. et al. The CD70-CD27 axis in oncology: the new kids on the block. J. Exp. Clin. Cancer Res. 41, 12 (2022). Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 22 42. Riether, C. et al. Targeting CD70 with cusatuzumab eliminates acute myeloid leukemia stem cellsi np a t i e n t st r e a t e dw i t hh y p o - methylating agents.Nat. Med. 26,1 4 5 9– 1467 (2020). 43. Jiang, P. et al. Signatures of T cell dysfunction and exclusion predict cancer immunotherapy response.Nat. Med 24,1 5 5 0– 1558 (2018). 44. Pich, C. et al. Melanoma-expressed CD70 is involved in invasion and metastasis.Br. J. Cancer 114,6 3– 70 (2016). 45. Liu, D. et al. Integrative molecul ar and clinical modeling of clinical outcomes to PD1 blockade in patients with metastatic melanoma. Nat. Med. 25,1 9 1 6– 1927 (2019). 46. Sade-Feldman, M. et al. De fining T cell states associated with response to checkpoint immunotherapy in melanoma.Cell 175, 998– 1013.e20 (2018). 47. Lin, W. et al. Establishment and characterization of new tumor xenografts and cancer cell lines from EBV-positive nasopharyngeal carcinoma.Nat. Commun. 9, 4663 (2018). 48. Whibley, N., Tucci, A. & Powrie, F. Regulatory T cell adaptation in the intestine and skin. Nat. Immunol. 20,3 8 6– 396 (2019). 49. Wing, J. B. & Sakaguchi, S. Multiple treg suppressive modules and their adaptability.Front Immunol. 3,1 7 8( 2 0 1 2 ) . 50. Miragaia, R. J. et al. Single-Cell Transcriptomics of regulatory T cells reveals trajectories of tissue adaptation.Immunity 50, 493– 504.e7 (2019). 51. Overacre-Delgoffe, A. E. et al. Interferon-Gamma drives Treg fra- gility to promote anti-tumor immunity.Cell 169, 1130– 1141.e11 (2017). 52. Yoshida, K. et al. Anti-PD-1 antibody decreases tumour-in filtrating regulatory T cells. BMC Cancer 20, 25 (2020). 53. Kempkes, R. W. M., Joosten, I., Koenen, H. & He, X. Metabolic pathways involved in regulatory T cell functionality.Front. Immunol. 10,2 8 3 9( 2 0 1 9 ) . 54. Shi, H. & Chi, H. Metabolic control of treg cell stability, plasticity, and tissue-specific heterogeneity.Front. Immunol. 10,2 7 1 6 (2019). 55. Watson, M. J. et al. Metabolic support of tumour-in filtrating reg- ulatory T cells by lactic acid. Nature 591,6 4 5– 651 (2021). 56. Howie, D. et al. A novel role for triglyceride metabolism in Foxp3 expression.Front. Immunol. 10,1 8 6 0( 2 0 1 9 ) . 57. Das, M., Alzaid, F. & Bayry, J. Regulatory T cells under the mercy of mitochondria.Cell Metab. 29,2 4 3– 245 (2019). 58. Go, G. W. & Mani, A. Low-density lipoprotein receptor (LDLR) family orchestrates cholesterol homeostasis.Yale J. Biol. Med. 85, 19– 28 (2012). 59. Su, W. et al. Protein prenylation drives discrete signaling programs for the differentiation and maintenance of effector Treg cells.Cell Metab. 32, 996– 1011.e7 (2020). 60. Lim, S. A. et al. Lipid signalling enforces functional specialization of Treg cells in tumours. Nature 591,3 0 6– 311 (2021). 61. Pacella, I. et al. Fatty acid metabolism complements glycolysis in the selective regulatory T cell expansion during tumor growth. P r o c .N a t lA c a d .S c i .U S A115, E6546– E6555 (2018). 62. Weinberg, S. E. et al. Mitochondrial complex III is essential for suppressive function of regulatory T cells.Nature 565, 495– 499 (2019). 63. Palsson-McDermott, E. M. & O ’Neill, L. A. J. Targeting immunome- tabolism as an anti-inflammatory strategy.Cell Res. 30, 300– 314 (2020). 64. Martinez-Reyes, I. & Chandel, N. S. Mitochondrial TCA cycle metabolites control physiology and disease.Nat. Commun.11, 102 (2020). 65. Yan, Y. et al. Metabolic pro files of regulatory T cells and their adaptations to the tumor microenvironment: implications for anti- tumor immunity. J. Hematol. Oncol. 15, 104 (2022). 66. Zhang, Z. et al. An OGT-STAT5 axis in regulatory T cells controls energy and iron metabolism.Front Immunol. 13, 874863 (2022). 6 7 . T s a o ,S .W . ,T s a n g ,C .M .&L o ,K .W .E p s t e i n - B a r rv i r u si n f e c t i o na n d nasopharyngeal carcinoma.P h i l o s .T r a n s .RS o c .L o n d .BB i o l .S c i. 372,2 0 1 6 0 2 7 0( 2 0 1 7 ) . 6 8 . L e o n g ,M .M .L .&L u n g ,M .L .T h eImpact of Epstein-Barr VIrus I n f e c t i o nO nE p i g e n e t i cR e g u l a tion Of Host Cell Gene Expression In Epithelial And Lymphocytic Malignancies.Front Oncol. 11, 629780 (2021). 69. Ding, R. B. et al. Molecular landscape and subtype-speci fic therapeutic response of nasopharyngeal carcinoma revealed by integrative pharmacogenomics. Nat. Commun. 12, 3046 (2021). 7 0 . H u i ,A .B .e ta l .D e t e c t i o no frecurrent chromosomal gains and losses in primary nasopharyngeal carcinoma by comparative genomic hybridisation.Int J. Cancer 82, 498– 503 (1999). 71. Wang, C. et al. Genome-wide CRIS PR-based gene knockout screens reveal cellular factors and pathways essential for nasopharyngeal carcinoma.J. Biol. Chem. 294,9 7 3 4– 9745 (2019). 72. Sauer, T. et al. CD70-speci fic CAR T cells have potent activity against acute myeloid leukemia without HSC toxicity.Blood 138, 318– 330 (2021). 73. Tu, S. et al. CD19 and CD70 dual-target chimeric antigen receptor T-cell therapy for the treatment of relapsed and refractory primary central nervous system diffuse large B-cell lymphoma.Front Oncol. 9, 1350 (2019). 74. Agathanggelou, A. et al. Expression of immune regulatory mole- cules in Epstein-Barr virus-associated nasopharyngeal carcinomas with prominent lymphoid stroma. Evidence for a functional inter- action between epithelial tumor cells and infiltrating lymphoid cells. Am. J. Pathol. 147, 1152– 1160 (1995). 7 5 . L a u ,K .M .e ta l .I n c r e a s ei nc i r c u l a t i n gF o x p 3 + C D 4 + C D 2 5 ( h i g h ) regulatory T cells in nasopharyngeal carcinoma patients.Br. J. Cancer 96,6 1 7– 622 (2007). 76. Kumagai, S. et al. An oncogenic alteration creates a micro- environment that promotes tumor progression by conferring a metabolic advantage to regulatory T cells. Immunity 53, 187– 203.e8 (2020). 77. Sasidharan Nair, V., Saleh, R., Toor, S. M., Cyprian, F. S. & Elkord, E. Metabolic reprogramming of T regulatory cells in the hypoxic tumor microenvironment.Cancer Immunol. Immunother.70, 2103– 2121 (2021). 78. Qiu, X. et al. Reversed graph embedding resolves complex single- cell trajectories.Nat. Methods 14,9 7 9– 982 (2017). Acknowledgements This work was supported by grants from the Hong Kong Research Grant Council (RGC), including Collaborative Research Funds (C7065-18GF, C 7 0 2 6 - 1 8 G F ,a n dC 4 0 3 9 - 1 9 G F ,X . G . ) ,R e s e a r c hI m p a c tF u n d s( R 4 0 1 7 - 1 8 , R1020-18F and R7022-20, X.G.), Theme-based Research Funds (TRS (T12- 703/22-R), X.G.), National Natural Science Foundation of China (81772554, X.G., 82072738, X.G., 82103479, L.G.), Natural Science Foundation of Guangdong Province (2023A1515010109, Z.Y.), Shenzhen Key Laboratory Program (ZDSYS20210623091811035, X.G.), The Shenz- hen Science and Technology program (KQTD20180411185028798, X.G.) and the Program for Guangdong Introducing Innovative and Entrepre- neurial Teams (2019BT02Y198, X.G.). Xin-Yuan Guan is the Sophie YM Chan Professor in Cancer Research. Author contributions L.G., and X.G. designed the study. L.G., and X.G. wrote the manuscript. D.L.K., P.W., and S.L. supervised the sample collection. L.G., J.L., Z.Y., Y.Y., and X.F. performed the experiments. L.G., W.D., B.F., L.C., and D.C. Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 23 analyzed the data. J.L.H., and Y.W. collected the FFPE slides and per- formed the pathological examination. G.T. provided the valuable PDX. A.W.L., G.T., J.H., C.H., Q.L, L.B., and Y.T. participated in the discussion. X.G. supervised the study. All the authors have read and approved the manuscript. Competing interests The authors declare no competing interests. Additional information Supplementary informationThe online version contains supplementary material available at https://doi.org/10.1038/s41467-023-37614-6. Correspondenceand requests for materials should be addressed to Xin-Yuan Guan. Peer review information: Nature Communicationsthanks Adrian Och- senbein, Sander de Kivit, Qin Ma and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Reprints and permissions informationis available at http://www.nature.com/reprints Publisher’s note Springer Nature remains neutral with regard to jur- isdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visithttp://creativecommons.org/ licenses/by/4.0/. © The Author(s) 2023 1Department of Clinical Oncology, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong, China.2Department of Clinical Oncology, The University of Hong Kong-Shenzhen Hospital, Shenzhen, China.3Department of Pediatric Oncology, Sun Yat-sen University Cancer Center, Guangzhou, China. 4State Key Laboratory of Oncology in South China, Guangdong Key Laboratory of Nasopharyngeal Carcinoma Diagnosis and Therapy, Sun Yat-sen University Cancer Center, Guangzhou, China.5Collaborative Innovation Center for Cancer Medicine, Sun Yat-sen University Cancer Center, Guangzhou, China. 6Department of Surgery, The University of Hong Kong-Shenzhen Hospital, Shenzhen, China.7Department of Pathology, The University of Hong Kong- Shenzhen Hospital, Shenzhen, China.8School of Biomedical Sciences, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong, China. 9Advanced Energy Science and Technology Guangdong Laboratory, Huizhou, China.10These authors contributed equally: Lanqi Gong, Jie Luo, Yu Zhang. e-mail: [email protected] Article https://doi.org/10.1038/s41467-023-37614-6 Nature Communications| (2023) 14:1912 24

The release as fetched from its publisher. oncoverity.com ↗