As shown in the flowchart ( Figure 1 ), a total of 658 newly diagnosed DLBCL patients treated in Ruijin Hospital from 2006 to 2019 were included in this study, namely 64 patients with CD5+ and 594 patients negative for CD5 (CD5−). Patients with primary CNS lymphoma, with primary mediastinal lymphoma, or who received supportive care were excluded. All patients were treated with the R-CHOP–based therapy (rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone) and were included in the survival analysis. There were 529 non-DE DLBCL patients, namely 48 CD5+ non-DE patients and 481 CD5− non-DE patients. Among the 48 CD5+ non-DE patients, eight were recruited in a prospective, single-arm phase II clinical trial (ChiCTR-OIN-17012130) to evaluate the therapeutic effect of metformin as maintenance therapy, in which patients received metformin orally at a dose of 1.0 g, twice a day for 2-year maintenance after complete remission (23). The remaining 40 patients did not receive metformin maintenance therapy. Among 658 patients, 355 had available DNA sequencing, and 208 had available RNA sequencing data.

Clinical data including age, Eastern Cooperative Oncology Group performance status, Ann Arbor stage, serum lactate dehydrogenase, IPI, and sites of extranodal involvement were collected. Immunohistochemistry was performed for CD10, BCL6, MUM1, BCL2, MYC, and CD5 using an indirect immunoperoxidase method on the paraffin sections of tumor tissues according to standard protocols. The histological diagnosis and percentage of positive tumor cells were confirmed by two experienced pathologists. Germinal center B-cell or non–germinal center B-cell origin was classified by Hans algorithm. MYC/BCL2 DE lymphoma was defined as MYC expressed on ≥40% of tumor cells and BCL2 expressed on ≥50% of tumor cells (24).

Genomic DNA from frozen tumor tissues was extracted using a QIAamp DNA Mini Kit (Qiagen, Hilden, Germany); DNA from formalin-fixed paraffin-embedded tumor tissues was extracted using a GeneRead DNA Formalin-Fixed Paraffin-Embedded Tissue Kit (Qiagen). For whole genome sequencing (WGS), sequencing was performed with 150-bp paired-end strategy on an Illumina HiSeq platform in WuXi NextCODE, Shanghai. For whole exome sequencing (WES), exome regions were captured by a SeqCap EZ Human Exome Kit (version 3.0), and sequencing was performed with 150-bp paired-end strategy on a HiSeq 4000 platform in Righton, Shanghai. Burrows–Wheeler Aligner version 0.7.13-r1126 was applied to align read pairs to Human Reference Genome version hg19 (downloaded from UCSC Genome Browser, URLs). Samtools version 1.3 was used to remove PCR duplications and generate chromosomal coordinate-sorted bam files. Genome Analysis Toolkit version 3.4 Haplotype Caller and Genome Analysis Toolkit Unified Genotyper were applied to call single-nucleotide variations (SNVs) and indels. Annotation of SNVs and indels was applied using the UCSC Genome Browser (http://genome.ucsc.edu). The filter of SNVs and indels was based on the following pipelines: (1) germline mutations detected in paired blood samples were excluded, (2) mutations with low frequency (<0.05) were excluded, and (3) mutations included in the single-nucleotide polymorphism database and not reported in COSMIC (the Catalogue of Somatic Mutations in Cancer) version v77 were excluded.

For targeted sequencing, PCR primers were designed by Primer 5.0 software. Multiplexed libraries of tagged amplicons from tumor tissue samples were generated by Shanghai Righton Bio-Pharmaceutical Multiplex–PCR Amplification System.

All data of DNA sequencing including targeted sequencing (n = 178), WES (n = 119), and WGS (n = 58) were collected from our previous report (14).

RNA was extracted from frozen tumor tissue samples with a Trizol and RNeasy Mini Kit (Qiagen) and quantified with NanoDrop. Details of RNA sequencing procedures and RNA sequencing data of 208 patients were all cited from our previous report (14). The raw reads were aligned to human reference genome hg19 with a Burrows‐Wheeler Aligner (v0.7.13‐r1126). Transcript counts table files were generated with HTSeq. Normalization of raw reads and analysis of differentially expressed genes were carried out using R package “limma” (v3.48.1) (25). Gene set enrichment analysis (GSEA) was performed with R package “clusterProfiler” (v4.0.0) based on MSigDB-curated gene sets (c2.cp.keggADD.v7.0.symbols.gmt and c5.bp.v7.0.symbols.gmt) (26). Metabolic signature scores were calculated using single-sample GSEA through R package “GSVA” (1.40.1); gene sets of lipid metabolism were acquired from the KEGG database. A CIBERSORT deconvolution algorithm (v 1.03) was applied in the calculation of immune cell populations based on its reference list. A total of 22 immune cell phenotypes were included in the analysis according to the signatures of bulk RNA-sequencing profile.

The B-lymphoma cell line OCI-LY-10 was provided by Huang CX. DB and THP-1 cells were purchased from the American Type Culture Collection. PBMCs from a healthy donor were obtained via Ficoll–Paque density gradient centrifugation. Cells were cultivated in IMDM or RPMI 1640, with the addition of 10% heat-inactivated fetal bovine serum and 1% penicillin/streptomycin (15140122, Gibco, Carlsbad, CA, USA) under a humidified atmosphere containing 95% air–5% CO₂ at 37 °C. Metformin was obtained from Selleck (Houston, TX, USA). Sulfosuccinimidyl oleate (SSO) was obtained from MedChemExpress (Shanghai, China).

CD5 overexpression cells or negative control cells were constructed using lentiviral transfection. A purified plasmid MSCV-IRES-EGFP-CD5 (GV358) retroviral vector or control vector was transfected into package HEK-293T cells. Lentiviral particles were collected from the supernatant of HEK-293T cells and condensed to a viral concentration of ~2 × 10⁸ transducing units/ml. Then, LY-10 cells or DB cells were cultivated, with the addition of lentiviral particles for 72 h in IMDM or RPMI 1640 containing polybrene (8 μg/ml). Flow cytometry was applied to select stably transduced clones expressing EGFP fluorescence protein.

PBMCs were cocultured with CD5 overexpressed OCI-LY-10 or vector OCI-LY-10 in IMDM and CD5 overexpressed DB in RPMI 1640 for 72 h. The following anti-human antibodies were used: APC anti-CD14 (555397, BD Biosciences), PE-CY7 anti-CD68 (565595, BD Biosciences), PE anti-CD163 (556018, BD Biosciences), and Percp-Cy5.5 anti-CD36 (336224, Biolegend). Flow cytometry was applied on a FACS Calibur cytometer (BD Biosciences) and analyzed by FlowJo software (v10.4).

To induce M2 macrophages, 320 nM phorbol myristate acetate was added to THP-1 cells for 6 h, followed by phorbol myristate acetate plus 20 ng/ml of IL-4 and IL-13 for 18 h (27).

Oil red O staining was applied using a modified Lipid Staining Kit (C0158S, Beyotime) following its staining protocol. Cells were fixed with 10% formalin for 10 min and washed with phosphate-buffered saline twice. Staining wash buffer was added for 20 s and then removed. Then, cells were stained with oil red O solution for 20–30 min, and then the stain solution was removed. Oil red O staining was observed under a microscope.

Total RNA was extracted from M2 cells using a Trizol reagent and reverse-transcribed into cDNA using a HiScript III RT SuperMix for qPCR with a gDNA wiper (R323-01, Vazyme). Quantitative real-time PCR was performed with a ChamQ Universal SYBR qPCR Master Mix (Q711-02, Vazyme) using ABI ViiA7 (Applied Biosystems, Bedford). Forward primer of CD36: 5′-GGCTGTGACCGGAACTGTG-3′; reverse primer: 5′-AGGTCTCCAACTGGCATTAGAA-3′; forward primer of FASN: 5′-CCGAGACACTCGTGGGCTA-3′; reverse primer: 5′-CTTCAGCAGGACATTGATGCC′; forward primer of FABP5: 5′-TGAAGGAGCTAGGAGTGGGAA-3′; reverse primer: 5′-TGCACCATCTGTAAAGTTGCAG′. Relative expression was calculated and described using ^2−ΔΔCT.

The clinical characteristics of patients were assessed using Pearson’s χ² test or Fisher’s exact test. Differences of immune cell populations and normalized gene expression in two groups were carried out using unpaired t-test or Mann–Whitney U test. Progression-free survival (PFS) was determined as the time from the date of diagnosis to the date when the disease progression of relapse was recognized or the date of last follow-up. Overall survival (OS) was determined as the time from the date of diagnosis to the date of death or the date of last follow-up. Survival functions were analyzed using the Kaplan–Meier method and compared by the log-rank test. Statistical significance was defined as p < 0.05. All statistical analyses were performed using R software (version 4.1.0; http://www.R-project.org) and Statistical Package for the Social Sciences (SPSS, 25.0) software (SPSS, Inc., Chicago, IL, USA).

The median follow-up of 658 patients was 41.3 months, and the 5-year PFS and the 5-year OS were 70.6% and 80.0%, respectively. However, CD5+ DLBCL patients had poor 5-year PFS (54.2% vs. 72.4%, p = 0.003; Figure 2A ), but without statistical difference in the 5-year OS (71.0% vs. 80.9%, p = 0.077; Figure 2B ) when compared with CD5− DLBCL patients. Among the DE patients, no significant difference was observed between CD5+ and CD5− patients in terms of PFS (5-year PFS 40.0% vs. 59.3%, p = 0.110; Figure 2C ) and OS (5-year OS 65.6% vs. 65.9%, p = 0.879; Figure 2D ). Nevertheless, CD5+ non-DE patients had poor outcome, as compared with CD5− non-DE patients: 5-year PFS 58.8% vs. 76.3%, p = 0.016 ( Figure 2E ) and 5-year OS 72.9% vs. 84.7%, p = 0.037 ( Figure 2F ).

Based on these findings that CD5+ had prognostic indications in non-DE DLBCL, further clinical, genetic, and microenvironment profiles were analyzed in non-DE patients. Among non-DE, CD5+ patients had advanced Ann Arbor stage (III–IV) (p = 0.021), high-risk IPI (p = 0.010), more frequent bone marrow involvement (p = 0.018), and high CNS relapse (p = 0.040; Table 1 ).

Oncogenic mutations related to DLBCL progression were detected in WES, WGS, and targeted sequencing. Mutation patterns in non-DE patients were displayed in Figure 3A . The mutation frequencies of MYD88 (28.1% vs. 13.0%, p = 0.020), FOXO1 (18.8% vs. 6.5%, p = 0.013), and TMSB4X (15.6% vs. 4.6%, p = 0.010) were significantly higher in CD5+ patients than in CD5− patients.

Gene ontology analysis revealed the activation of multiple pathways, including the MYD88-dependent Toll-like receptor pathway, monocyte- and macrophage-related pathways (monocyte chemotaxis, macrophage migration, macrophage activation, positive regulation of response to cytokine stimulus, regulation of myeloid cell differentiation, tumor necrosis factor superfamily cytokine production, and CCR chemokine receptor binding), and pathways related to inhibit immune responses (negative regulation of immune system process, TGF-β receptor signaling pathway), in CD5+ non-DE patients ( Figures 3B,C ). Following analysis of an immune cell proportion in TME was carried out through CIBERSORT. As displayed in Figure 3D , an elevated proportion of monocytes (p = 0.019), M2 macrophage (p < 0.001), and a decreased proportion of NK cells were observed (p < 0.001) in CD5+ non-DE when compared with CD5− non-DE patients. Meanwhile, the normalized expression level of HAVCR2 (also called TIM3) was significantly increased in CD5+ than CD5− non-DE patients (p = 0.018) ( Figure 3E ). Moreover, the normalized expression levels of HAVCR2 (p = 0.033) and TGFBI (p < 0.001) were positively correlated with the M2 proportion in TME, and the normalized expression of HAVCR2 was positively correlated with TGFBI (p < 0.001) ( Figure 3F ).

Regarding the variation in metabolic pathways between CD5+ and CD5− non-DE DLBCL, as revealed by GSEA, upregulation of lipid metabolic pathways, including the PPAR signaling pathway, the ATP-binding cassette transporter pathway, ether lipid metabolism, the fatty acid metabolism pathway, and arachidonic acid metabolism, was observed in CD5+ patients. However, the metabolic pathway enriched in CD5− patients included pyrimidine metabolism, porphyrin and chlorophyll metabolism, pentose and glucuronate interconversions, and glyoxylate and dicarboxylate metabolism ( Figure 4A ). Further comparing the correlation between metabolic pathways and immune cell proportion, all these upregulated metabolism pathways in CD5+ non-DE patients were positively correlated with the M2 proportion ( Figure 4B ). Among differentially expressed genes, the normalized expression level of CD36 was significantly higher in CD5+ than CD5− non-DE patients (p = 0.035; Figure 4C ) and positively correlated with the M2 proportion in TME (p < 0.001; Figure 4D ). Additionally, the expression level of CD36 was also positively correlated with the expressions of multiple chemokines and chemokine receptors including CCL16, CCL18, CCL2, CCL8, CCR2, CXCR1, and CXCR2 ( Figure 4E ).

After PBMC cocultured with CD5-overexpressed (LY-10/CD5) or vector-transfected (LY-10/vector) cells at the ratio of 5:1 for 72 h, flow cytometry was applied to calculate the proportion of M2 and CD36 expression on M2. As shown in Figure 5A , the coculture with LY-10/CD5 or LY-10/vector increased the M2 proportion (13.0% ± 3.3%, p = 0.004 or 6.9% ± 1.7%, p = 0.007) in PBMC as compared with PBMC alone (1.4% ± 0.9%). Of note, the M2 proportion was significantly high in the LY-10/CD5 coculture system than in the LY-10/vector coculture system (p = 0.047; Figure 5A ). Metformin treatment (200 μM) significantly decreased the proportion of M2 in both coculture systems, from 13.0% ± 3.3% to 4.8% ± 1.4% (p = 0.017) in the LY-10/CD5 coculture system and from 6.9% ± 1.7% to 3.6% ± 0.9% (p = 0.039) in the LY-10/vector coculture system ( Figure 5B ). Further exploring the efficacy of metformin on the expression of CD36, flow cytometry analysis revealed that, after treatment with metformin, the CD36 expression level on M2 was significantly downregulated in the LY-10/CD5 coculture system (p = 0.025) than the LY-10/vector coculture system (p = 0.052; Figure 5C ). PBMCs cocultured with DB/CD5 cells were also treated with 200 μM metformin for 72 h, M2 cells decreased from 10.8% ± 2.1% to 6.3% ± 1.5% (p = 0.043), and CD36 expression on M2 cells also decreased significantly (p = 0.018) ( Supplementary Figure S1 ). M2 cells induced from THP-1 cells were treated with 200 μM metformin for 72 h. The mRNA expression levels of CD36, FABP5, and FASN were significantly decreased after metformin treatment (p = 0.019, p = 0.021, and p = 0.037, respectively) ( Figure 5D ). Also, oil red O staining showed reduced lipid droplets accumulation after metformin treatment ( Figure 5E ).

In the LY10/CD5 and PBMC coculture system, after the addition of metformin (200 μM) for 72 h, the expressions of PD-1 on both CD4 T cells (from 7.0% ± 2.6% to 2.8% ± 0.4%, p = 0.048) and CD8 T cells (from 8.8% ± 2.6% to 4.4% ± 0.4%, p = 0.044) decreased significantly. Meanwhile, TIM3 expression decreased from 5.4% ± 2.6% to 0.8% ± 0.5% on CD8 T cells (p = 0.038) and decreased from 5.0% ± 2.3% to 1.1% ± 0.5% (p = 0.046) on CD4 T cells ( Figure 5F ).

Consistent with the in vitro results that metformin targeted immunosuppressive TME, CD5+ non-DE patients with metformin maintenance tended to have longer PFS and OS than those without metformin maintenance (5-year PFS, 87.5% vs. 50.3%, p = 0.096, and 5-year OS, 87.5% vs. 69.3%, p = 0.329; Figure 5G ).

To further explore the impact of CD36 inhibition on lipid metabolism in the coculture system, irreversible CD36 inhibitor SSO (100 μM) was applied in PBMC cocultured with LY-10/CD5 or LY-10/vector for 72 h; flow cytometry was used for the quantification of the M2 proportion and surface CD36 proportion. After SSO treatment, M2 proportion was significantly decreased from 13.0% ± 3.3% to 4.4% ± 0.5% (p = 0.012) in the LY-10/CD5 coculture system and decreased from 6.9% ± 1.7% to 3.8% ± 0.4% (p = 0.035) in the LY-10/vector coculture system ( Figure 6A ), as well as the reduction of surface CD36 proportion on M2 (p < 0.001) ( Figure 6B ). However, SSO treatment did not affect the mRNA expression level of CD36 but significantly decreased the mRNA level of FABP5 (p = 0.014) and FASN (p = 0.002) in M2 ( Figure 6C ). Consequently, the accumulation of lipid droplets was significantly inhibited in M2 upon SSO treatment ( Figure 6D ).

Comparable with the results of metformin treatment, blockage of CD36 with SSO in the coculture system (LY10/CD5 and PBMC) significantly decreased the immune checkpoint expression on CD4 T cells (PD-1: from 7.0% ± 2.6% to 0.7% ± 0.4%, p = 0.013; TIM3: from 5.0% ± 2.3% to 0.4% ± 0.2%, p = 0.026) and CD8 T cells (PD-1: from 8.8% ± 2.6% to 1.3% ± 0.7%, p = 0.009; TIM3: from 5.4% ± 2.6% to 0.8% ± 0.6%, p = 0.037) ( Figure 6E ).