Table3_A prognostic 15-gene model based on differentially expressed genes among metabolic subtypes in diffuse large B-cell lymphoma.XLSX

The outcomes of patients with diffuse large B-cell lymphoma (DLBCL) vary widely, and about 40% of them could not be cured by the standard first-line treatment, R-CHOP, which could be due to the high heterogeneity of DLBCL. Here, we aim to construct a prognostic model based on the genetic signature of metabolic heterogeneity of DLBCL to explore therapeutic strategies for DLBCL patients. Clinical and transcriptomic data of one training and four validation cohorts of DLBCL were obtained from the GEO database. Metabolic subtypes were identified by PAM clustering of 1,916 metabolic genes in the 7 major metabolic pathways in the training cohort. DEGs among the metabolic clusters were then analyzed. In total, 108 prognosis-related DEGs were identified. Through univariable Cox and LASSO regression analyses, 15 DEGs were used to construct a risk score model. The overall survival (OS) and progression-free survival (PFS) of patients with high risk were significantly worse than those with low risk (OS: HR 2.86, 95%CI 2.04–4.01, p < 0.001; PFS: HR 2.42, 95% CI 1.77–3.31, p < 0.001). This model was also associated with OS in the four independent validation datasets (GSE10846: HR 1.65, p = 0.002; GSE53786: HR 2.05, p = 0.02; GSE87371: HR 1.85, p = 0.027; GSE23051: HR 6.16, p = 0.007) and PFS in the two validation datasets (GSE87371: HR 1.67, p = 0.033; GSE23051: HR 2.74, p = 0.049). Multivariable Cox analysis showed that in all datasets, the risk model could predict OS independent of clinical prognosis factors (p < 0.05). Compared with the high-risk group, patients in the low-risk group predictively respond to R-CHOP (p = 0.0042), PI3K inhibitor (p < 0.05), and proteasome inhibitor (p < 0.05). Therefore, in this study, we developed a signature model of 15 DEGs among 3 metabolic subtypes, which could predict survival and drug sensitivity in DLBCL patients.

弥漫性大B细胞淋巴瘤（diffuse large B-cell lymphoma, DLBCL）患者的预后结局差异显著，约40%的患者无法通过标准一线治疗方案R-CHOP获得治愈，这一现象可能与DLBCL的高度异质性密切相关。本研究旨在基于DLBCL代谢异质性的基因特征构建预后模型，以探索DLBCL患者的治疗策略。本研究从基因表达汇编数据库（Gene Expression Omnibus, GEO）获取了1个训练队列与4个验证队列的DLBCL患者临床数据及转录组数据。在训练队列中，研究者针对7大代谢通路中的1916个代谢基因进行围绕中心点划分聚类（Partitioning Around Medoids, PAM），以此鉴定DLBCL的代谢亚型；随后对不同代谢簇间的差异表达基因（differentially expressed genes, DEGs）进行分析，最终筛选得到108个与预后相关的差异表达基因。通过单变量Cox回归与最小绝对收缩和选择算子回归（Least Absolute Shrinkage and Selection Operator, LASSO）分析，最终纳入15个差异表达基因构建风险评分模型。结果显示，高风险组患者的总生存期（overall survival, OS）与无进展生存期（progression-free survival, PFS）均显著劣于低风险组（OS：风险比HR=2.86，95%置信区间CI=2.04–4.01，p < 0.001；PFS：HR=2.42，95%CI=1.77–3.31，p < 0.001）。该风险模型在4个独立验证数据集（GSE10846：HR=1.65，p=0.002；GSE53786：HR=2.05，p=0.02；GSE87371：HR=1.85，p=0.027；GSE23051：HR=6.16，p=0.007）中均与总生存期显著相关；在2个验证数据集（GSE87371：HR=1.67，p=0.033；GSE23051：HR=2.74，p=0.049）中与无进展生存期显著相关。多变量Cox回归分析显示，在所有数据集中，该风险模型均可独立于临床预后因素预测总生存期（p < 0.05）。与高风险组相比，低风险组患者对R-CHOP方案（p=0.0042）、PI3K抑制剂（p < 0.05）及蛋白酶体抑制剂（p < 0.05）具有更优的预测响应效果。综上，本研究构建了基于3种代谢亚型中15个差异表达基因的特征模型，该模型可有效预测DLBCL患者的生存期与药物敏感性。