Higher Suvmax values suggest a larger tumor burden and poor prognosis
In this study, the median follow-up duration was 49.0 months (range: 3.0-118.0 months). The median age of patients was 47 years (range: 11.0-78.0 years), and the majority of patients presented with AJCC stages III (43.4%) and IV (37.1%). Patient characteristics are summarized in Table 1. Throughout the follow-up period, 63 (8.7%) experienced local failure, 88 (12.1%) experienced distant metastasis, 161 (22.2%) encountered disease progression, and 75 (10.3%) succumbed to mortality. The 5-year rates for Local Recurrence-Free Survival (LRFS), Distant Metastasis-Free Survival (DMFS), Progression-Free Survival (PFS), and Overall Survival (OS) were 89.7%, 87.7%, 76.7%, and 87.8%, respectively.
The correlation between Suvmax and clinical variables was examined. Suvmax demonstrated associations with T stage (P < 0.01, Figure 1A), MTV (r=0.691, P < 0.001, Figure 1C), and LDH (r=0.178, P < 0.001, Figure 1D). Suvmax exhibited limited correlation with N-stage (Figure 1B).Subsequently, a survival analysis was conducted. The ROC curve determined the optimal cutoff value for Suvmax to predict overall survival, maximizing Youden's index. Patients were then categorized into high and low Suvmax groups. The results revealed that LRFS (HR=0.42; P=0.003, Figure 1E), PFS (HR=0.71; P=0.036, Figure 1G), and OS (HR=0.50; P=0.007, Figure 1H) in patients with high Suvmax were all significantly lower than those in patients with low Suvmax. Unexpectedly, there was no significant difference in DMFS between patients with high Suvmax and those with low Suvmax (P=0.318, Figure 1F).
The SUV-signature was formulated utilizing genes associated with glucose metabolism.
In a cohort of 29 patients undergoing transcriptome sequencing, individuals were categorized into two groups based on the median value of Suvmax. Differential gene analysis (Figure 2A), Gene Ontology (GO) analysis (Figure 2B), and Gene Set Enrichment Analysis (GSEA) (Figure 2C) revealed the enrichment of Suvmax differential genes in pathways related to glucose metabolism. Consequently, we explored the construction of the Suv-signature using genes associated with glucose metabolism.
Through correlation analysis between genes related to glucose metabolism and Suvmax, we identified 55 genes associated with glucose metabolism. Lasso-logistics modeling (Figure 2D and 2E) was conducted with 10-fold cross-validation, and the lambda value (0.00392) corresponding to the model with the smallest cross-validation mean error was selected. The coef coefficients were then extracted and combined to derive the Suv-signature (Figure 2F). The formulated formula is as follows: Suv-signature = 0.007 * PKM + 0.375 * SEH1L - 0.44 * LHPP + 0.007 * LDHA + 3.713 * PPFIA4 + 0.015 * PFKFKB3 - 0.099 * MPI - 11.022 * ZBTB20 - 0.672 * CLDN9 + 3.186 * PGAM4.
Verify whether Suv-signature can predict Suvmax and poor prgonosis
In order to prove that Suv-signature model can predict Suvmax, correlation analysis was conducted between Suv-signature and Suvmax using our dataset of 16 cases and 29 cases (Figure3A and FigureS1) and GSE135565 dataset (Figure3B).Suv-signature was found to be positively correlated with Suvmax, which proved that Suv-signature could represent Suvmax. Meanwhile, we conducted a prognosis validation using the GSE102349 dataset and stratified the cohorts based on the median Suv-signature values. Our findings revealed a significantly poorer prognosis in the high Suv-signature group compared to the low Suv-signature group. (P=0.006, Figure3C).
Subsequently, we utilized the single-cell RNA sequencing dataset (GSE150430) from NPC to investigate the intrinsic biological characteristics of Suvmax.Employing UMAP, we categorized the cells into six clusters, encompassing B cells, T cells, NK cells, myeloid cells, plasma cells, and epithelial cells (Figure 3D). SUV-signature scores were then computed across these distinct clusters, revealing the highest expression in epithelial tissues (Figure 3E). Further delineating epithelial cells into malignant and normal subtypes, we observed a significantly elevated Suv-signature expression in malignant epithelial cells compared to their non-malignant counterparts (Figures 3F-3I).
Suv-signature was associated with hypoxic and radiotherapy resistance
A correlation analysis was conducted on our center's dataset (Figure 4A) and the GSE102349 dataset (Figure 4D) to examine the relationship between Suv-signature and hypoxia score. The results revealed a significant correlation, with an R value of 0.76 in our center dataset and 0.62 in the GSE102349 dataset, indicating a close association between Suv-signature and hypoxia score.The outcomes from the Gene Set Enrichment Analysis (GSEA) further validate the enrichment of Suv-signature within hypoxia-associated hallmarks, as illustrated in Figure 4B and 4E. This observation supports the notion that Suv-signature could potentially serve as an indicator of hypoxia.
Additionally, our investigation revealed a notable enrichment score for Suv-signature within the radiotherapy (RT) resistance gene set, as depicted in Figure 4C and 4F. Given that radiotherapy stands as the primary treatment for NPC, and its insensitivity implies a poor prognosis, the connection between Suv-signature and RT resistance becomes pivotal. Notably, hypoxia serves as a known mechanism contributing to radiotherapy resistance. This implies that Suvmax may potentially predict an unfavorable prognosis by influencing radiotherapy sensitivity through oxygen deficiency.
SUV-signature was associated with immune function
In the single-cell RNA dataset, T cells were stratified into high and low Suv-signature groups based on the median values of Suv-signature. Our observations revealed a significant upregulation of immunosuppressive markers, including HAVCR2, PDCD1, LAG3, and TIGIT, in T cells exhibiting high Suv-signature scores (Figure 5A-D).
Expanding our analysis to the GSE102349 cohort, we further explored the association between Suv-signature and immune score, stromal score, predicted score, and tumor purity (Figure 5E-H). Remarkably, we found a negative correlation between Suv-signature and immune score, stromal score, and predicted score, while a positive correlation was observed with tumor purity. These findings suggest an inverse relationship between Suv-signature and immune function.