Natural killer/T cell lymphoma (NKTCL) is a rare but highly aggressive malignancy classified within the non-Hodgkin lymphoma spectrum. It has a striking geographic distribution, being rare in North America and Europe but prevalent across Southeast Asia and parts of Latin America. Most malignant cells originate from NK cells, with a minority deriving from cytotoxic T cells, and nearly all cases are associated with Epstein-Barr virus (EBV) infection. Despite this defining viral driver, the mechanisms through which EBV reshapes the tumor microenvironment (TME) and drives immunosuppression have remained poorly characterized at the cellular level.
The treatment problem: Combination chemotherapy regimens incorporating L-asparaginase (such as SMILE and P-GemOx) remain the standard of care. However, approximately 50% of patients exhibit primary resistance or early relapse. Immunotherapy has shown promise in relapsed and refractory disease, but the lack of mechanistic understanding of the TME has limited the rational design of immunotherapeutic combinations. Identifying specific cell populations that drive immune evasion and tumor progression is a prerequisite for developing more targeted treatment strategies.
The single-cell and spatial gap: Prior studies using bulk genomic and transcriptomic profiling have revealed genetic heterogeneity and molecular subtypes in NKTCL. However, bulk methods average signals across millions of cells and cannot resolve the functional states of individual cell populations or reveal how those populations are spatially organized within the tumor. Single-cell RNA sequencing (scRNA-seq) addresses the first limitation, while spatial transcriptomics addresses the second. Together, these complementary technologies allow researchers to map the cellular landscape at the level of individual cells and tissue regions simultaneously.
This 2025 study published in iScience enrolled 13 patients with newly diagnosed NKTCL and subjected their tumor tissues to both scRNA-seq and spatial transcriptomic profiling, integrating the data with publicly available single-cell datasets from seven non-malignant nasopharyngeal control tissues for comparative analysis. The result is one of the most detailed cellular maps of NKTCL produced to date.
The study enrolled 10 patients with primary NKTCL for scRNA-seq, and for two of these patients, paired post-treatment specimens were also sequenced, enabling a preliminary look at treatment-related transcriptomic changes. An additional 7 non-malignant nasopharyngeal tissue samples were incorporated as controls from a publicly available dataset. After stringent quality control, a total of 96,369 cells were retained for analysis: 66,873 from NKTCL tissues and 29,496 from control tissues. This cell count is large enough to resolve rare cell populations but still reflects the reality that patient cohort sizes in rare malignancies are limited.
Malignant cell identification: A key methodological challenge in NKTCL is that malignant cells cannot be identified by a single canonical marker, because they express NK cell markers also found on normal NK cells. The authors addressed this by inferring large-scale chromosomal copy number variations (CNVs) from the transcriptome sequencing data. Cells with evident, large-scale CNV patterns consistent with malignancy were classified as malignant. Critically, the inferred CNV patterns showed high concordance with paired whole-exome sequencing data from the same patients, validating this approach. A total of 14,658 malignant cells were identified across seven patients (those with at least 200 malignant cells each).
Intratumoral heterogeneity analysis: To characterize the heterogeneity within the malignant cell population, the authors applied consensus non-negative matrix factorization (cNMF) to the 7 tumors with sufficient malignant cell counts. This unsupervised method decomposes gene expression data into biologically coherent programs without requiring predefined cell labels. Thirty-seven intra-tumoral programs were generated and subsequently classified into five "meta-programs" (MPs) using Pearson correlation coefficients. Pathway enrichment using the top 50 signature genes per MP was performed with Metascape to annotate the biological functions associated with each meta-program.
Spatial transcriptomics platform: Spatial profiling was performed on three primary NKTCL samples using Stereo-seq technology, which provides single-cell resolution spatial data through DNA nanoball sequencing. After quality control, 27,129 bin50 spots (each spanning a 50x50 DNB bin area, approximately 25 micrometers in diameter) were retained. The SPOTlight algorithm was used to deconvolute the cell-type composition of each spot based on the scRNA-seq reference, allowing the spatial mapping of all identified cell populations. Cell neighborhood analysis was then applied to identify spatially co-occurring cell type clusters across the tissue.
The cNMF analysis revealed five functionally distinct malignant meta-programs (MP1 through MP5) present across multiple patient samples, establishing that NKTCL tumors harbor substantial and reproducible intratumoral heterogeneity. MP1 is characterized by G2M checkpoint genes (CCNB1, CDC20, TPX2), reflecting a highly proliferative cycling state. MP2 is defined by E2F target genes (HIST1H4C, HIST1H1D, NUSAP1), associated with cell cycle progression and DNA replication. MP3 is defined by MYC target V1 genes (NPM1, FABP5, PTMA), pointing to MYC hyperactivation. MP4, expressing MACF1, NKTR, and GOLGA4, showed no significant enrichment in hallmark pathways and likely represents a damaged cell population. MP5 expresses NKG7, CCL5, and GZMK, with enrichment in the TNF-alpha via NF-kB pathway, consistent with a cytotoxic, more differentiated state.
Differentiation trajectory: Pseudotime analysis using Monocle2 reconstructed a continuous differentiation trajectory beginning at a high-proliferative state (MP3) and transitioning toward cytotoxic phenotypes (MP5). CytoTRACE scores confirmed this directionality: cells at the origin of the trajectory had lower CytoTRACE scores (indicating less differentiation), while cells at the terminus had higher scores (indicating more differentiation). This means MP3 represents the most immature, poorly differentiated malignant population, and MP5 represents the most mature, functionally cytotoxic end of the spectrum.
MP3 and poor prognosis: To assess clinical relevance, the top 50 signature genes for each meta-program were evaluated in an independent cohort of 97 bulk RNA-seq NKTCL datasets. Patients whose tumors expressed high levels of MP3 signature genes had significantly worse overall survival compared to those with lower MP3 expression (hazard ratio 3.71, p = 0.022 by log-rank test). No other meta-program reached statistical significance for overall survival in this analysis, underscoring MP3 as the clinically most consequential subpopulation.
Metabolic pathway enrichment analysis further showed that MP3 cells exhibit elevated fatty acid metabolism relative to other MPs. This metabolic state is consistent with what has been described in other aggressive cancers, where enhanced lipid metabolism supports rapid proliferation and contributes to an immunosuppressive microenvironment. The co-occurrence of MYC hyperactivation and fatty acid metabolic rewiring in MP3 set the stage for targeted functional experiments on FABP5.
Among the top genes in the MP3 signature, fatty acid-binding protein 5 (FABP5) emerged as a particularly interesting candidate for several reasons. First, its expression was significantly elevated in MP3 cells compared to all other meta-programs (Wilcoxon rank-sum test, p < 0.0001). Second, its expression decreased progressively along the pseudotime differentiation trajectory, meaning FABP5 is highest in the least-differentiated, most aggressive cells and lowest in the more mature cells. Third, a strong positive correlation was observed between FABP5 expression and MYC_TARGETS_V1 gene set scores specifically in the malignant NKTCL cell population (Pearson R = 0.67, p < 2.2e-16), suggesting a functional link between fatty acid binding and MYC pathway activity.
In vitro validation: To test whether FABP5 is functionally important rather than merely correlative, the researchers performed gain-of-function and loss-of-function experiments in the YT NKTCL cell line. Overexpression of FABP5 led to a marked increase in cell proliferation relative to vector controls, as measured by CCK-8 assay (p < 0.0001). Conversely, pharmacologic inhibition of FABP5 using the selective inhibitor SBFI-26 suppressed cell growth in a dose-dependent manner at concentrations of 50 and 100 micromolar (p < 0.0001). Genetic knockdown of FABP5 using two independent siRNAs (si-FABP5#1 and si-FABP5#2) produced consistent results, ruling out off-target effects of the pharmacologic inhibitor.
In vivo validation: The SBFI-26 inhibitor was then tested in a YT cell xenograft mouse model. Nude mice bearing subcutaneous YT tumors received daily intraperitoneal administration of SBFI-26 at 4 mg/kg or corn oil vehicle (n = 5 per group). At the study endpoint (day 28), SBFI-26-treated mice showed significantly reduced tumor volume and weight compared to controls (p < 0.01). Histological analysis of excised tumors confirmed that SBFI-26 treatment reduced Ki-67 positivity (a marker of active proliferation) and increased cleaved caspase-3 staining (a marker of apoptosis), and also reduced c-Myc protein levels in tumor tissue.
Mechanism - FABP5 stabilizes c-Myc: Western blot analysis confirmed that both pharmacologic FABP5 inhibition and genetic knockdown reduced c-Myc protein levels in YT cells. Critically, co-immunoprecipitation assays in HEK293T cells co-transfected with Flag-FABP5 and HA-c-Myc plasmids revealed a direct physical interaction between FABP5 and c-Myc proteins. This finding suggests that FABP5 modulates MYC signaling through a post-translational mechanism, likely by stabilizing the c-Myc protein rather than regulating its transcription. Because c-Myc itself lacks enzymatically active domains and has historically been considered undruggable, FABP5 represents a viable upstream surrogate target in MYC-driven NKTCL.
Compared with non-malignant nasopharyngeal control tissues, NKTCL tumor samples exhibited significantly higher proportions of myeloid cells and lower proportions of B cells and T cells (Wilcoxon rank-sum test). A total of 8,030 myeloid cells were clustered into ten subclusters based on canonical marker gene expression: mast cells (TPSB2, TPSAB1), two monocyte clusters (S100A9 and RETN), three macrophage clusters (APOE, C1QA, and CXCL10), two conventional dendritic cell clusters (XCR1 and CD1C), plasmacytoid dendritic cells (LILRA4, GZMB), and neutrophils (CXCL8, CXCL2). This myeloid expansion pattern is consistent with observations in murine models where EBV-encoded latent membrane protein 1 (LMP1) promotes myeloid propagation through interferon-gamma signaling.
Macro_APOE as the dominant pro-tumorigenic macrophage: Among the ten myeloid subclusters, the two macrophage subpopulations, Macro_APOE and Macro_CXCL10, were more abundant in tumor tissue than in controls and accounted for the largest share of the myeloid compartment. Differential gene expression analysis revealed that Macro_APOE expressed high levels of CD163, MSR1, and matrix metalloprotease-9 (MMP9), genes associated with anti-inflammatory responses and angiogenesis, and showed a bias toward the M2 macrophage phenotype. In contrast, Macro_CXCL10 expressed interferon-stimulated genes including guanylate-binding proteins (GBPs), IFITM3, and IDO1, placing it toward the M1 (pro-inflammatory) phenotype.
Developmental trajectory of macrophages: Partition-based graph abstraction (PAGA) analysis suggested that monocytes differentiate into Macro_CXCL10 and Macro_C1QA populations, which can then further transition into the pro-tumorigenic Macro_APOE subtype. This indicates a phenotypic shift from an anti-tumorigenic to a pro-tumorigenic macrophage state during NKTCL tumor progression. Consistent with this, Macro_APOE abundance was significantly reduced in the post-treatment paired specimens, suggesting that effective therapy may reverse macrophage polarization.
Macro_APOE and patient outcomes: Kaplan-Meier survival analysis showed that higher Macro_APOE gene signature scores in bulk RNA data from the independent NKTCL cohort were associated with significantly poorer overall survival (HR 2.63, p = 0.044 by log-rank test). These macrophages also expressed high levels of immune checkpoint molecules SIRPA and VSIG4. VSIG4 has been reported to inhibit activation of pro-inflammatory macrophages and T cells. The combination of M2 polarization, immune checkpoint expression, and adverse prognostic impact positions Macro_APOE as a potentially targetable immunosuppressive mediator in NKTCL.
Ligand-receptor interaction analysis between tumor cells and myeloid subsets identified several functionally relevant signaling axes. The most prominent interaction between malignant cells and Macro_APOE involved the SPP1-CD44 axis and the SIRPalpha-CD47 checkpoint. SPP1-CD44 interactions have been reported to drive chemoresistance in triple-negative breast cancer and may serve a similar chemoresistance-promoting function in NKTCL. The SIRPalpha-CD47 interaction is particularly significant: CD47 expressed on tumor cells transmits a "don't eat me" signal to macrophage SIRP-alpha receptors, dampening phagocytic activity and enabling tumor cells to evade macrophage-mediated destruction. This checkpoint is increasingly recognized as a target for anti-tumor immunotherapy across multiple cancer types.
T cell exhaustion correlation: Analysis of T and B cell subclusters revealed a significant increase in the proportion of CD8_LAG3 (exhausted CD8+ T cells) in NKTCL tumor tissue compared to controls. Furthermore, in bulk RNA data from the independent NKTCL cohort, the Macro_APOE gene signature score was significantly positively correlated with the CD8_LAG3 exhaustion score (Pearson correlation). This implies that Macro_APOE macrophages do not merely evade phagocytosis themselves but also actively suppress cytotoxic T cell function, creating a doubly immunosuppressive environment.
Additional immunosuppressive pathways: Ligand-receptor analysis also identified the Galectin-9/TIM-3 (LGALS9-HAVCR2) signaling pair between malignant and immune cells. TIM-3 is a well-established T cell exhaustion marker, and Galectin-9/TIM-3 signaling promotes T cell apoptosis and suppresses anti-tumor responses. The CCL5-CCR1 and CCL5-CCR5 chemokine axes were detected between malignant cells and immune cells, suggesting that tumor cells actively recruit immune populations through CCL5 secretion, including the Macro_APOE population itself. This creates a feed-forward loop in which tumor cells recruit immunosuppressive macrophages that in turn protect the tumor from immune destruction.
Together, these interaction networks describe a multi-layered immune evasion strategy in NKTCL: malignant cells actively recruit M2-polarized macrophages via CCL5, engage the SIRPalpha-CD47 axis to suppress phagocytosis, and drive T cell exhaustion through Macro_APOE-mediated signaling, all operating in concert within the tumor microenvironment.
Spatial transcriptomic analysis using Stereo-seq was performed on three primary NKTCL samples, adding a critical dimension to the single-cell findings by revealing where in the tissue these cell populations actually reside and interact. After deconvolution with SPOTlight, each of the 27,129 spots was assigned a dominant cell type from the scRNA-seq reference, and the resulting spatial maps showed that malignant MP1-5 cells, T cells, B cells, myeloid cells, fibroblasts, endothelial cells, and epithelial cells occupied distinct regional compartments within the tissue. The distribution of spot-level marker gene expression showed high consistency with the cell type assignments, validating the deconvolution approach.
MP3 spatial enrichment: MP3 tumor cells represented the highest proportion of malignant cells across all three spatial samples. Their spatial distribution co-localized with elevated MYC_TARGETS_V1 pathway scores and fatty acid metabolism scores, confirming that the transcriptomic profile of MP3 cells identified in scRNA-seq is maintained in their native tissue context. FABP5 expression showed a matching spatial distribution pattern with MP3 tumor cells, further solidifying FABP5 as an MP3-specific marker with spatial validity.
Cell neighborhood analysis: Six distinct cell neighborhoods (CN0 through CN5) were identified across the tissue using cell neighborhood analysis based on deconvolution-derived cell type proportions. CN1 and CN2 were enriched with tumor cells, myeloid cells, and stromal cells. CN5 was characterized by MP3 tumor cells and B cells. CN4 was dominated by epithelial cells. CN0 and CN3 showed high proportions of both MP3 tumor cells and myeloid cells, indicating direct spatial proximity between the most aggressive malignant population and the myeloid compartment.
Functional differences between MP3 neighborhoods: Comparing MP3 cells in CN0 versus CN3 revealed a nuance: MP3 cells in CN3 exhibited higher cytotoxicity features (higher expression of GZMB, GNLY, CCL3, CCL4), suggesting a tendency toward MP5-like differentiation in that microenvironment. Both Macro_APOE and CD8_LAG3 scores were significantly elevated in CN0 and CN3 compared to other neighborhoods, and dot plot analysis confirmed upregulated CD47, SIRPA, CCL5, and CCR1 expression in these same regions. The widespread co-expression of SIRPalpha-CD47 and CCL5-CCR1 pairs across all three samples supports the model that tumor cells recruit Macro_APOE through CCL5 and establish immunosuppressive niches through CD47-SIRPalpha signaling in spatially defined zones.
Sample size limitations: The most significant limitation of this study is the relatively small patient cohort. Only 10 patients contributed primary tumor specimens for scRNA-seq, with a subset of 7 providing sufficient malignant cells for meta-program analysis. Spatial transcriptomics was performed on only 3 samples. NKTCL is rare in Western populations, making large-scale enrollment challenging, but expanded cohorts from high-prevalence regions in Asia would strengthen the generalizability of the identified malignant subpopulations, macrophage phenotypes, and ligand-receptor interactions. The authors explicitly acknowledge that larger cohorts are needed to confirm the consistency of these findings.
Tissue sampling constraints: NKTCL predominantly presents as nasal and extranodal lesions with limited biopsy material available from nasopharyngeal sites. This restricts the ability to perform comprehensive functional validation of the specific cell-cell interactions identified in silico. Validating that CD47-SIRPalpha blockade or CCL5-CCR1 inhibition functionally reverses immune evasion in primary NKTCL tumor-immune co-culture systems or in animal models with intact immune components requires material that is difficult to obtain in sufficient quantities from nasopharyngeal biopsies.
EBV integration: Despite NKTCL being fundamentally an EBV-driven malignancy, the spatial and mechanistic contribution of EBV to the immunosuppressive microenvironment was not fully characterized in this study. The authors identify integrating EBV-specific molecular profiling into future studies as a priority, particularly to understand how EBV-encoded proteins such as LMP1 shape macrophage polarization, T cell exhaustion, and the tumor-immune ligand-receptor network. In vivo models with sufficient immune components, incorporating both FABP5 and CD47-SIRPalpha inhibitors tested in EBV-positive settings, are identified as needed next steps.
Therapeutic implications: The study identifies two complementary therapeutic opportunities. First, FABP5 inhibition (using agents such as SBFI-26 or next-generation FABP5 inhibitors) could target the MYC-driven, poorly differentiated MP3 subpopulation directly. Second, macrophage checkpoint blockade targeting the CD47-SIRPalpha or VSIG4 axes could relieve Macro_APOE-mediated immune evasion and restore T cell function. Whether combining these two strategies would produce synergistic anti-tumor activity in NKTCL models represents a logical and clinically relevant research direction. Translation into clinical trials will require prospective validation of both FABP5 overexpression and Macro_APOE infiltration as predictive biomarkers, along with safety assessment of FABP5 inhibitors in humans.