Identification of molecular subtypes and a novel prognostic model of diffuse large B-cell lymphoma based on a metabolism-associated gene signature

Abstract

Background: Diffuse large B cell lymphoma (DLBCL) is the most common lymphoma in adults. Metabolic reprogramming in tumors is closely related to the immune microenvironment. This study aimed to explore the interactions between metabolism-associated genes (MAGs) and DLBCL prognosis and their potential associations with the immune microenvironment.

Methods: Gene expression and clinical data on DLBCL patients were obtained from the GEO database. Metabolism-associated molecular subtypes were identified by consensus clustering. A prognostic risk model containing 14 MAGs was established using Lasso-Cox regression in the GEO training cohort. It was then validated in the GEO internal testing cohort and TCGA external validation cohort. GO, KEGG and GSVA were used to explore the differences in enriched pathways between high- and low-risk groups. ESTIMATE, CIBERSORT, and ssGSEA analyses were used to assess the immune microenvironment. Finally, WGCNA analysis was used to identify two hub genes among the 14 model MAGs, and they were preliminarily verified in our tissue microarray (TMA) using multiple fluorescence immunohistochemistry (mIHC).

Results: Consensus clustering divided DLBCL patients into two metabolic subtypes with significant differences in prognosis and the immune microenvironment. Poor prognosis was associated with an immunosuppressive microenvironment. A prognostic risk model was constructed based on 14 MAGs and it was used to classify the patients into two risk groups; the high-risk group had poorer prognosis and an immunosuppressive microenvironment characterized by low immune score, low immune status, high abundance of immunosuppressive cells, and high expression of immune checkpoints. Cox regression, ROC curve analysis, and a nomogram indicated that the risk model was an independent prognostic factor and had a better prognostic value than the International Prognostic Index (IPI) score. The risk model underwent multiple validations and the verification of the two hub genes in TMA indicated consistent results with the bioinformatics analyses.

Conclusions: The molecular subtypes and a risk model based on MAGs proposed in our study are both promising prognostic classifications in DLBCL, which may provide novel insights for developing accurate targeted cancer therapies.

Keywords: Diffuse large B-cell lymphoma; Immune microenvironment; Metabolism; Molecular subtype; Prognosis.

Fig. 1

Consensus clustering and the different immune profiles between two clusters. A Consensus matrix heatmap indicating that the optimal value for consensus clustering is K = 2. B Heatmap visualizing the different expression pattern of the 92 MAGs in the two clusters. C Survival curve of the patients in the two clusters. D CIBERSORT analysis in the two clusters. E The expression of immune checkpoints among two clusters. P values were showed as: ns not significant; *p < 0.05; **p < 0.01; ***p < 0.001

Fig. 2

Construction of the risk model in the GSE10846 training cohort and validation of the risk model in the GSE10846 testing cohort and TCGA cohort. A, D, G Distribution of the risk score, survival status, and gene expression of 14 MAGs in the GSE10846 training cohort (A), GSE10846 testing cohort (D) and TCGA cohort (G). B, E, H Kaplan–Meier curves of OS of patients in the high- and lowrisk groups in the GSE10846 training cohort (B), GSE10846 testing cohort (E) and TCGA cohort (H). C, F, I ROC curves for predicting the 1/2/3-year overall survival in the GSE10846 training cohort (C), GSE10846 testing cohort (F) and TCGA cohort (I)

Fig. 3

Clinical correlations of risk score and development of the nomogram in the GSE10846 dataset. A Survival curve of patients in the high- and low-risk groups in the GSE10846 immunotherapy cohort (232 patients who had received R-CHOP treatment). B–H Relationships between the risk score and clinicopathological features (including age, ECOG status, stage, LDH level, IPI score, gender and extranodal sites). 305 patients with complete clinicopathological features from the GSE10846 dataset were analyzed. The distance of both ends of boxes represents the interquartile range of values and the thick line represents the median value. I Univariate and multivariate analyses revealed the risk score was an independent prognostic factor for DLBCL patients. J Nomogram for predicting the 3- and 5-year OS of DLBCL patients. K Calibration curves of the nomogram for OS prediction at 3- and 5- year. L ROC curves indicating the comparisons of the risk score and the IPI score in predicting 1-year OS

Fig. 4

The different immune profiles between the low- and high- risk groups in the GSE10846 dataset. Two risk groups were divided based on the median risk score. A ESTIMATE algorithm. B ssGSEA analysis. C CIBERSORT analysis. D Correlation between risk score and immune cell content. E Expression variation of immune checkpoint. p values were showed as: ns not significant; *p < 0.05; **p < 0.01; ***p < 0.001

Fig. 5

Identification of two hub genes and prediction of their relationship with immune cell content and experimental validation of their expression in the DLBCL TMA cohort. A Venn diagram analysis showed that the overlap of WGCNA analysis and LASSO model led to two hub genes being identified: PLTP and PHKA1. B Prediction of correlations between hub genes and immune cell content. C Prediction of correlations between hub genes and immune checkpoints. D Differences in PLTP expression between reactive hyperplasia tissues and DLBCL tissues with mIHC (p value by Wilcoxon rank-sum test). E Differences in PHKA1 expression between reactive hyperplasia tissues and DLBCL tissues with mIHC (p value by Wilcoxon rank-sum test)

Fig. 6

Experimental verification of the relationship between PLTP and prognosis and immune microenvironment in the DLBCL TMA cohort. A Kaplan–Meier curve for the PLTP high- and low- expression groups in our TMA cohort. The optimal cutoff point was obtained from X-tile 3.6.1 software. B Characterization of cell immunophenotypes with mIHC. A staining panel was developed to visualize DAPI, CD68, CD163 and PLTP simultaneously on the same tissue slide. C M2 macrophages content in the PLTP high- and low-expression groups. CD68+CD163+ indicated the content of M2 macrophages (p value by Wilcoxon rank-sum test). D Characterization of cell immunophenotypes with mIHC. A staining panel was developed to visualize DAPI, CD11B and PLTP simultaneously on the same tissue slide. E TAMs content in the PLTP high- and low- expression groups. CD11B+ indicated the content of TAMs (p value by Wilcoxon rank-sum test). F Characterization of cell immunophenotypes with mIHC. A staining panel was developed to visualize DAPI, PD-1, PD-L1, LAG3 and PLTP simultaneously on the same tissue slide. CD274 indicated the content of PD-L1 and PDCD1 indicated the content of PD-1. G–I Correlations between PLTP expression and immune checkpoints (including PD-1, PD-L1, and LAG3)

Fig. 7

Experimental verification of the relationship between PHKA1 and prognosis and immune microenvironment in the DLBCL TMA cohort. A Kaplan–Meier curve for the PHKA1 high- and low-expression groups in our TMA cohort. The optimal cutoff point was obtained from X-tile 3.6.1 software. B Characterization of cell immunophenotypes with mIHC. A staining panel was developed to visualize DAPI, CD68, CD163 and PHKA1 simultaneously on the same tissue slide. C M2 macrophages content in the PHKA1 highand low- expression groups. CD68+CD163+ indicated the content of M2 macrophages (p value by Wilcoxon rank-sum test). D Characterization of cell immunophenotypes with mIHC. A staining panel was developed to visualize DAPI, CD11B and PHKA1 simultaneously on the same tissue slide. E TAMs content in the PHKA1 high- and low- expression groups. CD11B+ indicated the content of TAMs (p value by Wilcoxon rank-sum test). F Characterization of cell immunophenotypes with mIHC. A staining panel was developed to visualize DAPI, PD-1, PD-L1, LAG3 and PHKA1 simultaneously on the same tissue slide. CD274 indicated the content of PD-L1 and PDCD1 indicated the content of PD-1. G–I Correlations between PHKA1 expression and immune checkpoints (including PD-1, PD-L1, and LAG3)