B-cell Receptor Silencing Reveals the Origin and Dependencies of High-Grade B-cell Lymphomas with MYC and BCL2 Rearrangements Open Access

The B-cell receptor (BCR) serves as a signaling hub governing the survival and responsiveness of mature B cells to adaptive and innate signals (1). In mature B cell–derived malignancies, the BCR signaling platform is commonly hijacked and otherwise rewired to ensure chronic activation of downstream pathways either autonomously or in response to environmental signals (2, 3) Tonic signals from surface (s) BCR also support growth of malignant B cells, improving competitive fitness (4, 5). Selection for sustained sBCR expression and signaling in activated B cell–like diffuse large B-cell lymphoma (DLBCL) is commonly achieved via acquisition of somatic mutations targeting the ITAM motif of the CD79B signaling subunit (4, 6, 7). Follicular lymphoma (FL) treated with anti-idiotype antibodies directed against the malignant BCR mutate the target site to escape killing while preserving surface antigen receptor expression (8). In Burkitt lymphoma, the translocation t(8;14) (q24;q32) preferentially selects the nonproductive immunoglobulin heavy chain (IGH) locus for IGH::MYC fusions preserving BCR expression (9). Conversely, studies in MYC-driven human and murine B-cell lymphomas (BCL) indicate that malignant cells can acutely adapt to and chronically bypass BCR extinction (5, 10, 11). Similar conclusions can be reached interpreting sIG-null phenotypes previously reported for different types of mature B-cell neoplasms (12–14). Nonetheless, the concept that mature BCLs can evolve or eventually arise under conditions of BCR silencing has largely gone unrecognized. The relevance of assessing the influence of BCR silencing on lymphoma genesis and evolution stems from the property of B cells to repeatedly undergo, during their lifetime, antigen receptor downregulation upon participation in the germinal center (GC) reaction, or as a consequence of persistent engagement with (self-) antigens. The surface IG^low state of GC dark zone (DZ) centroblasts (or self-reactive B cells) entails the existence of BCR surrogate signals supporting the persistence of these cells. Whether some mature BCLs exploit such mechanisms to gain BCR independence remains unknown. In this study, we report recurrent IGH silencing in mature BCLs with DLBCL morphology after a comprehensive phenotypic screening of more than 300 real-world cases. Lymphomas with undetectable IGH (IGH^UND) protein were significantly enriched among GC B cell–like (GCB) DLBCL and predominantly belonged to high-grade BCLs (HGBCL) featuring MYC and BCL2 rearrangements [(“double-hit” (DH); HGBCL-DH-BCL2)]. Investigations into the mechanisms and consequences of IGH silencing in HGBCL-DH-BCL2 reveal a causal role for antigen receptor silencing in tumor onset and evolution, with potential clinical implications.

We conducted a screening by IHC for IGH chains IGM/D/G/A in 258 consecutive mature BCLs with DLBCL morphology (Fig. 1A and B; Supplementary Fig. S1A–S1C; Supplementary Table S1A). In two thirds of the cases (172/258; 67%), malignant B cells predominantly expressed one IGH chain isotype, primarily IGM, in more than 90% of the cells. In the remaining cases (86/258; 33%), IGH immunoreactivity was undetectable in at least 10% of lymphoma B cells (Fig. 1A and B; Supplementary Fig. S1A; Supplementary Table S1A). Among these cases, 57/86 (66%) showed >90% of atypical B cells with undetectable IGM/D/G/A, hereafter called IGH^UND tumors, whereas the remaining cases (29/86; 34%) consisted of a mosaic of IGH^UND (11%–89% of cells) and IGH-expressing tumoral cells, representing the IGH^UND/+ mixed group (Fig. 1B; Supplementary Fig. S1B; Supplementary Table S1A). For representative cases (n = 34), including a subset belonging to a recently published cohort (15), IGH IHC was matched with flow cytometric determination of surface (s) IG-κ (IGK) and IG-λ (IGL) light chains and CD79B protein, establishing a link between the IGH^UND phenotype and sBCR silencing (Fig. 1C; Supplementary Table S1B). IGH^UND cases exhibited a significant enrichment for GCB-type DLBCL as compared with non-GCB counterparts, classified according to the Hans algorithm (Fig. 1D; Supplementary Table S1A). Among GCB DLBCL, we observed a stronger association between the IGH^UND phenotype and cases later diagnosed as HGBCLs with MYC and BCL2 rearrangements (DH), hereafter called HGBCL-DH-BCL2. These tumors included a subset carrying also BCL6 rearrangements (HGBCL-DH-BCL2-BCL6; Fig. 1E; Supplementary Table S1A). Therefore, we extended the IGH screening to n = 104 HGBCL-DH-BCL2(-BCL6), scoring 65% with undetectable IGH immunoreactivity, mostly consisting of IGH^UND cases (Fig. 1F–H; Supplementary Tables S1C and S1D). The high frequency of IGH^UND HGBCL-DH-BCL2 cases was not solely linked to the common widespread expression of the MYC and BCL2 proteins. Indeed, HGBCL-DH-BCL2(-BCL6) showed a significantly stronger enrichment for IGH^UND cases when compared with MYC and BCL2 (MB2) dual protein expressor (DE) GCB DLBCL missing corresponding chromosomal rearrangements (n = 63; Fig. 1I; Supplementary Table S1C). The enrichment for IGH^UND cases and the limited understanding of the role of the BCR in HGBCL-DH-BCL2 ontogeny and evolution prompted us to investigate the mechanisms and biological implications of IGH silencing in these aggressive lymphomas.

We conducted a bulk transcriptomics comparison of IGH⁺ and IGH^UND HGBCL-DH-BCL2(-BCL6) cases (n = 14), incorporating also MB2 DE GCB DLBCL (n = 8) into the analysis (Supplementary Tables S2A and S2B). Unsupervised clustering according to the full transcriptome, excluding IGH transcripts, mostly separated IGH^UND lymphomas from IGH⁺ counterparts (Fig. 2A), suggestive of distinct biological entities. IGH^UND/+ HGBCL-DH-BCL2(-BCL6) tumors fell within the IGH^UND cluster and were thus incorporated into the latter group for downstream analyses. IGH^UND tumors differed from their IGH⁺ counterparts for an almost equal number of up- (n = 468) and down-regulated (n = 485) genes [log₂ fold change (FC) > 0.58; P-adj value < 0.05; Fig. 2B; Supplementary Table S2C], defining the IGH^UND dual-expressor (IGH^UND DE) molecular signature. IGH^UND tumors showed significantly (P-adj value < 0.05) higher expression of MYC (Fig. 2C), cell-cycle regulators, and factors involved in DNA replication and repair, mitosis, chromatin remodeling, telomere maintenance, mitochondrial respiration, and glucose, nucleotide, and amino acid metabolism (Fig. 2D; Supplementary Table S2D). IGH^UND lymphomas also expressed higher transcripts for components of the unfolded protein stress response and the SKP1-cullin 1(CUL1)-F-box E3 ligase complex (Fig. 2D; Supplementary Table S2D), suggesting a tight control over protein turnover. IGH⁺ lymphomas showed stronger expression of IFNγ-, IRF8- and NF-κB–regulated genes and immunomodulatory factors such as IL10, TGFB3, IDO1, IDO2, and CCL22 (Fig. 2D; Supplementary Table S2D). The IGH^UND DE signature was used to interrogate the transcriptome profile of an independent DLBCL cohort (16), focusing on cases assigned to the EZB (EZH2 mutated/BCL2 translocated) genetic subtype (n = 43), given similar genetic makeup and cell-of-origin (COO) with HGBCL-DH-BCL2 (17, 18). EZB-type DLBCL clustered into two major groups, with the majority (n = 31) exhibiting closer transcriptional similarity to IGH^UND HGBCL-DH-BCL2, including all EZB-MYC⁺ cases (n = 17) characterized by MYC deregulation (Supplementary Fig. S2A and S2B; Supplementary Table S2E; ref. 17). To interrogate the HGBCL-DH-BCL2 tumor ecosystem at higher resolution, distinct regions of interest (ROI) of representative IGH⁺ and IGH^UND cases were profiled in situ by spatial transcriptomics for 1825 cancer- and immune-relevant genes (Fig. 2E; Supplementary Fig. S2C; Supplementary Table S2F). IGH^UND HGBCL-DH-BCL2 ROI showed significant up- (n = 234) and down-modulation (n = 260) of selected transcripts compared with IGH⁺ counterparts (Fig. 2E; Supplementary Table S2F; P-adj value <0.05). In IGH⁺ HGBCL-DH-BCL2 ROI, higher expression of several chemokines, adhesion molecules, and extracellular matrix proteins suggested a richer representation of immune, inflammatory, and stromal/vascular cell types (Fig. 2E; Supplementary Table S2G). IGH⁺ HGBCL-DH-BCL2 tumor niches showed higher expression of transcripts associated with follicular dendritic cells and NK cells (Supplementary Table S2G), in line with cell-type deconvolution analyses (Supplementary Fig. S2D). Transcripts for T cell receptor complex components, T-follicular helper, T_H17 cells, and T-regulatory cells were also more represented in IGH⁺ HGBCL-DH-BCL2 ROI (Supplementary Table S2F). IHC and multiplex immunofluorescence (IF) analyses confirmed a richer infiltration of CD3⁺ T cells (both CD4- and CD8A-positive cells) and CD57⁺ NK cells in IGH⁺ HGBCL-DH-BCL2(-BCL6), as compared with the IGH^UND counterparts, whereas CD68⁺ macrophages seemed similarly represented in ROIs of the two tumors (Fig. 2F and G; Supplementary Table S2H). Stronger FCGR2B gene (Fig. 2E) and protein (Supplementary Fig. S2E) expression in IGH⁺ ROI may reflect attenuation of BCR signaling in the malignant B cells (19). Conversely, IGH^UND HGBCL-DH-BCL2 ROI showed stronger expression of MYC transcripts (Fig. 2E) and protein (Supplementary Fig. S2E) and factors involved in redox regulation, mitochondrial respiration, cell-cycle progression, and DNA replication and repair (Fig. 2E; Supplementary Table S2G), in agreement with bulk transcriptomics data. Collectively, IGH^UND and IGH⁺ HGBCL-DH-BCL2 establish distinct tumor ecosystems, with higher MYC expression and preferential T-cell exclusion in IGH-silenced cases.

We recently extrapolated GC DZ and LZ spatial gene expression signatures from human polarized reactive GCs of the tonsil (Supplementary Table S2I; https://pmc.ncbi.nlm.nih.gov/articles/PMC10984086/). Unsupervised interrogation of RNA sequencing (RNA-seq) data for GC spatial signatures identified two main clusters of GCB-type MB2 DE lymphomas (Fig. 2H). The first (n = 9; 41%), mostly comprising IGH^UND cases (7/9, 78%), more closely resembled DZ-resident cells (hereafter called “DZ-like”). The second cluster (n = 13; 59%), consisting primarily of IGH⁺ lymphomas (8/13; 61%), showed closer transcriptional relationship to LZ-resident cells (“LZ-like”; Fig. 2H). DZ-like HGBCL-DH-BCL2(-BCL6) were entirely included within the IGH^UND subset, whereas LZ-like counterparts grouped all IGH⁺ tumors and a fraction of IGH^UND cases (Fig. 2H). Assignment of HGBCL-DH-BCL2(-BCL6) cases to DZ- or LZ-like clusters was confirmed interrogating independent GC DZ- and LZ B-cell signatures (Supplementary Fig. S2F; ref. 20). Spatial transcriptomics of HGBCL-DH-BCL2 indicated preferential expression of DZ-associated transcripts in the ROI of the IGH^UND subset, whereas IGH⁺ counterparts expressed preferentially LZ-enriched mRNAs (Supplementary Fig. S2G). Interrogation of bulk transcriptome data for the DH/DZ molecular signature [DHIT/DZsig (21, 22)] identified a DHIT/DZsig⁺ group mostly consisting of IGH^UND HGBCL-DH-BCL2(-BCL6; 8/9; 89%). The remaining cases, including both IGH⁺ (n = 3) and IGH^UND (n = 2) HGBCL-DH-BCL2(-BCL6) showed a DHIT/DZ “indeterminate” signature with concurrent expression of DHITsig up- and down-regulated genes (Fig. 2I). In line with previous evidence (21), DHIT/DZsig-negative MB2 DE lymphomas mostly lacked BCL2 rearrangements (6/8 cases; 75%) while conserving IGH expression (7/8 cases; 87%; Fig. 2I).

We analyzed the mutational landscape of IGH^UND and IGH⁺ HGBCL-DH-BCL2(-BCL6), performing whole exome (and for few cases, whole genome) sequencing (Supplementary Table S3A). The analysis was limited to nonsynonymous single-nucleotide variants (nsSNV) of genes recurrently mutated in DLBCL, FL, HGBCL-DH-BCL2, and Burkitt lymphoma (16, 17, 23–27) to mitigate the lack of matching germline DNA (Fig. 3A; Supplementary Table S3A). IGH⁺ and IGH^UND HGBCL-DH-BCL2(-BCL6) shared nsSNV in genes coding for chromatin remodelers (CREBBP, KMT2C/D, and EP300/400), linker histone H1 D-E variants, and BCL7A (Fig. 3A; Supplementary Table S3A). Nonsilent mutations in genes relevant for GC B-cell biology (TNFRSF14, GNA13, KLHL14, and SOCS1) were also in common between IGH⁺ and IGH^UND HGBCL-DH-BCL2(-BCL6), along with HLA class-1 (HLA-A, HLA-B, and HLA-C) inactivating or missense gene mutations (Fig. 3A; Supplementary Table S3A). Irrespective of IGH status, HGBCL-DH-BCL2(-BCL6) carried nsSNVs in apoptosis and cell-cycle genes, including TP53, BCL2, BIRC6, CCND3, and CDKN2A (Fig. 3A; Supplementary Table S3A). Identification of nsSNVs in DNA damage checkpoint kinase genes ATM and ATR and in genes involved in DNA mismatch (MSH2, MSH4, MSH6, MSH5, and MLH1), nucleotide excision (ERCC2, ERCC3, ERCC4, ERCC5, and ERCC6), and homologous recombination repair (BRCA1 and BRCA2), and in DNA polymerases (POLB and POLE/E2) suggest a general status of genome instability and DNA hypermutation in both IGH⁺ and IGH^UND tumor subsets (Fig. 3A; Supplementary Table S3A). On such common mutational background, IGH⁺ and IGH^UND HGBCL-DH-BCL2(-BCL6) differed for a distinct set of gene mutations. IGH^UND lymphomas favored mutations in genes controlling MYC protein levels (i.e., MYC^T58A and FBXW7; Fig. 3A; Supplementary Table S3A; ref. 28). Similarly to EZB-DLBCL (16, 17), gain-of-function mutations in EZH2 (Y646N and A692V) were primarily identified in IGH^UND HGBCL-DH-BCL2(-BCL6), along with mutations in genes regulating the GC DZ B-cell program (BCL6, MEF2B, IRF8, and KLHL6; Fig. 3A; Supplementary Table S3A). FOXO1 gain-of-function mutations (M1L, R21H, T24I, and C23W), prevalently observed in EZB-MYC⁺ DLBCL (17), were exclusively found in IGH^UND HGBCL-DH-BCL2 (Fig. 3A; Supplementary Table S3A). N-terminus SGK1 truncating mutations and a BRAF^G469A gain-of-function (29) mutation suggest heightened PI3K signaling (30) and constitutive RAS/MAPK activation in individual IGH^UND HGBCL-DH-BCL2(-BCL6) cases, respectively (Fig. 3A; Supplementary Table S3A). Whereas mutations in the JAK/STAT signaling genes SOCS1 and TYK2 were shared between IGH⁺ and IGH^UND HGBCL-DH-BCL2(-BCL6), STAT6 mutations were restricted to IGH⁺ tumors (Fig. 3A; Supplementary Table S3A). Biallelic TET2 nsSNVs were identified in one IGH⁺ HGBCL-DH-BCL2 case, suggesting interference with DNA methylation. Mutations in effectors of the TLR4/NF-κB signaling pathway (TLR4, RIPK3, and NFKBIB) were observed across several IGH⁺ tumors (Fig. 3A; Supplementary Table S3A). Keeping with the low number of cases and the restricted selection of profiled genes, the mutational analyses identify a set of recurrent genetic alterations associated to HGBCL-DH-BCL2 outgrowth, with IGH⁺ and IGH^UND subsets diverging in the disease’s evolution through the preferential selection of nsSNVs in genes influencing GC B-cell zone identity and cross-talk with the tumor microenvironment.

High-throughput sequencing of IGHV gene libraries identified single-clone rearrangements in 22 HGBCL-DH-BCL2(-BCL6; Supplementary Table S3B). Together with IGH⁺ tumors, all IGH^UND HGBCL-DH-BCL2(-BCL6) cases carried potentially productive, hypermutated, clonal IGHV rearrangements (Supplementary Table S3B). We compared the mutational pattern of IGHV rearrangements from IGH^UND HGBCL-DH-BCL2(-BCL6) with that of GC-experienced B cell clones from reactive lymph nodes (rLN) and peripheral blood (PB) of healthy individuals as reference. For tumor IGHV analyses, we focused on mutations acquired after B-cell transformation, excluding mutations on lineage tree trunks. For healthy B-cell clones, we included trunk mutations to capture antigen-driven selection features, reaching similar results restricting analyses to trunk-less counterparts. Malignant GC B cells in IGH^UND HGBCL-DH-BCL2(-BCL6) exhibited significantly more mutations per clone than healthy counterparts (Fig. 3B), compatible with prolonged exposure to AID-initiated IGV mutagenesis (Supplementary Fig. S3A; Supplementary Table S3C). The minimal number of mutations per individual IGHV sequence, however, was lower in IGH^UND lymphomas compared with healthy GC/post-GC B-cell clones, excluding expansion of particular subclonal variants (Fig. 3C). This was consistent with lineage tree shapes of IGH^UND HGBCL-DH-BCL2 clones, indicating weaker vertical growth compared with healthy GCB counterparts (Fig. 3D). Lineage tree topology analysis suggested loosened antigen-driven selection in IGH^UND HGBCL-DH-BCL2(-BCL6), reflected in significantly more leaves (i.e., cells without descendants) and higher average number of children per node compared with reactive LN B cells (Fig. 3E and F). We applied the BASELINe algorithm to estimate IGHV-driven selection strength in HGBCL-DH-BCL2(-BCL6), separately analyzing CDR and FWR for patterns of mutations. IGH^UND HGBCL-DH-BCL2(-BCL6) preserved selection against amino acid replacement (R) mutations within FWRs (Fig. 3G), with scores for CDRs remaining neutral to weakly negative. IGHV rearrangements from circulating IG-mutated antigen-selected B-cell clones of healthy controls showed the expected selection for R mutations in CDR and against R mutations in FWR. Such selection was missing in B-cell clones bearing out-of-frame rearrangements from the nonfuntional IGH locus, confirming unbiased accumulation of R mutations (Fig. 3G). Finally, we interrogated IGHV genes of HGBCL-DH-BCL2 for AID-dependent acquisition of N-linked glycosylation motifs (N-X-S/T, in which X represents any amino acid except proline), hypothesizing a role for N-glycans in the selection of malignant B cells (31). Similarly to their IGH⁺ counterparts (n = 7), the majority of IGH^UND (n = 14) HGBCL-DH-BCL2(-BCL6) acquired N-glycosylation motifs within IGHV CDR and/or FWR across the entire tumor population (Fig. 3H; Supplementary Table S3D), pointing to a selected event for lymphoma genesis. Altogether, IGH^UND HGBCL-DH-BCL2(-BCL6) preserve structural integrity of IGHV domains, and recurrently acquire N-glycosylation motifs, suggestive of conserved dependence on a functional IGH chain.

To investigate whether IGH silencing is associated with changes in IGH gene expression, we analyzed RNA-seq data from HGBCL-DH-BCL2 cases, quantifying transcript levels for the most prevalent IGH constant region gene, assuming its malignant cell origin and normalizing for tumor cell content. IGH mRNA levels were significantly lower in the IGH^UND subset compared with their IGH⁺ counterparts (Fig. 4A). To validate these findings, we examined an independent cohort of DLBCL cases (16), ranking cases based on IGH transcript levels into IGH^Hi, IGH^Int, and IGH^Lo tertile groups and assigning them to genetic subtypes according to the LymphGen algorithm (17). Cases with low IGH transcript levels were significantly enriched among EZB-DLBCL, with a predominant occurrence in the EZB-MYC⁺ subset, which largely overlaps with HGBCL-DH-BCL2 (Supplementary Fig. S4A and S4B; ref. 17). These results suggest that IGH silencing in HGBCL-DH-BCL2 is contributed by reduced steady-state IGH gene transcripts.

We tested whether IGH shutdown was linked to any particular IGH class selected by the HGBCL-DH-BCL2(-BCL6) precursor cell. Analysis of RNA-seq data of IGH⁺ HGBCL-DH-BCL2(-BCL6) cases (n = 4) indicated predominant expression of IGHM constant region transcripts, concordant with protein data (Fig. 4B; Supplementary Table S4A). In IGH^UND HGBCL-DH-BCL2(-BCL6), low/undetectable IGHM was replaced by dominant isotype-switched IGHG1 (n = 2), IGHG2 (n = 2), IGHG4 (n = 2), IGHE (n = 1), or IGHA1/2 (n = 3) transcripts (Fig. 4B; Supplementary Table S4A). To exclude interference from tumor-infiltrating plasma cells expressing high IGH mRNA levels, we tracked in situ IGHM/D/G/A transcripts by RNA-scope technology (n = 34, including n = 12 analyzed by RNA-seq). In accordance with RNA-seq data, 13/14 IGH⁺ HGBCL-DH-BCL2(-BCL6) expressed IGHM (occasionally with IGHD) transcripts. In contrast, IGH^UND HGBCL-DH-BCL2(-BCL6) (n = 22) expressed either IGHG (n = 20) or IGHA (n = 2) mRNAs at the expense of IGHM/D transcripts, which were mostly undetectable (Fig. 4C and D; Supplementary Table S4B). The strong signal amplification inherent to RNA-scope technology prevented quantitative assessment of IGH transcripts. Collectively, IGH transcript analyses reveal a dominance of IGH-switched cases among HGBCL-DH-BCL2(-BCL6) (Fig. 4D), in line with (32), with 95% showing an IGH^UND phenotype (Fig. 4E). Conversely, 93% of IGH⁺ cases conserve IGM expression (Fig. 4E). A significant enrichment of cases expressing IGH-switched transcripts was shared with the GCB DLBCL ST2 and EZB-DLBCL subtypes, aligning with previous findings (Supplementary Fig. S4C; ref. 17). Notably, IGH-switched EZB-MYC⁺ DLBCL cases expressed significantly lower IGH transcripts compared with their IGHM⁺ counterparts, consistent with our observations in HGBCL-DH-BCL2 (Supplementary Fig. S4D). In contrast, no significant difference in IGH transcript levels was observed between IGM⁺ and IGH-switched cases within the EZB-MYC–negative subset (Supplementary Fig. S4D). To confirm IGH class switching in HGBCL-DH-BCL2(-BCL6), a quantitative genomic PCR was performed on representative cases to detect a segment of the IGHM gene, which is lost upon isotype switching. Differently from healthy IGM⁺ B cells carrying two IGHM loci, in 7/8 (88%) IGH^UND HGBCL-DH-BCL2(-BCL6), IGHM gene copy number (GCN) dropped below the unit, consistent with bi-allelic IG class switch recombination (CSR), with residual IGHM alleles likely contributed by tumor-infiltrating normal IGM⁺ B/plasma cells (Fig. 4F; Supplementary Table S4C). IGM⁺ HGBCL-DH-BCL2(-BCL6) carried on average one IGHM GCN (Fig. 4F; Supplementary Table S4C), compatible with monoallelic IG CSR targeting the nonproductive IGH chromosome (33). In summary, whereas IGH^UND HGBCL-DH-BCL2(-BCL6) tumors originate from IGG/A-switched B cells, most IGH⁺ tumors preserve IGM expression.

In the endoplasmic reticulum (ER) of mature B cells, IGH and IG light chains assemble with CD79A/B proteins to form the BCR complex, which gets transported to the cell surface after glycosylation of its subunits in the Golgi network (11). We asked whether cellular distribution and expression of CD79B were affected by IGH silencing in HGBCL-DH-BCL2(-BCL6). In IGH⁺ HGBCL-DH-BCL2(-BCL6) (n = 19), CD79B IHC marked the plasma membrane of the malignant cells combined with an intracellular localization (Fig. 4G; Supplementary Table S4D). In contrast, IGH^UND cases (n = 30) exhibited preferential (24/30) intracellular, often weaker, CD79B signals (Fig. 4G and H; Supplementary Table S4D), recalling the pattern observed in DZ-resident GC B cells (Supplementary Fig. S4E and S4F). We performed in situ proximity ligation assays (PLA) to assess whether in IGH^UND HGBCL-DH-BCL2(-BCL6), CD79A and CD79B proteins assembled into BCR complexes with residual traces of IGH protein. IGH^UND HGBCL-DH-BCL2(-BCL6) expressing IGHG mRNAs showed a significant reduction in the frequency of IGG/CD79B or IGG/CD79A complexes in the malignant B cells (Fig. 4I; Supplementary Table S4E). Conversely, IGG⁺ (or IGM⁺) DLBCL and B cells residing in reactive GC LZ areas showed conspicuous IGG (or IGM)/CD79A/B PLA⁺ signals (Fig. 4I; Supplementary Fig. S4G and S4H). IGH^UND tumors also showed marked contraction in the percentage of malignant cells forming CD79A/CD79B heterodimers, as compared with IGH⁺ counterparts (Supplementary Fig. S4I and S4J; Supplementary Table S4E). Collectively, IGH silencing in HGBCL-DH-BCL2(-BCL6) impairs the ability of the malignant B cells to incorporate CD79A/B proteins into BCR complexes, leading to their preferential intracellular retention.

To assess timing of IGH silencing in HGBCL-DH-BCL2(-BCL6), we analyzed three cases anticipated by a history of classic FL. Clonal relationship between metachronous tumors was confirmed by sequencing IGH::BCL2 and/or BCL6::IGH (in HGBCL-DH-BCL2-BCL6) translocation breakpoints and IGHV rearrangements (Supplementary Table S4F). In the IGH⁺ HGBCL-DH-BCL2-BCL6 tumor, IGM expression was shared with the preceding FL (Fig. 4J). Similarly, in two IGH^UND HGBCL-DH-BCL2 tumors, the IGH^UND phenotype and IGHG transcript pattern corresponded to those of the preceding FL (Fig. 4J). Matching of IGH expression status and class choice was confirmed for 10 additional FL-HGBCL-DH-BCL2(-BCL6) synchronous/metachronous tumors, except for a mixed IGM⁺/G⁺ FL, which evolved into an IGG-switched IGH^UND HGBCL-DH-BCL2 (Fig. 4K; Supplementary Table S4G). To assess whether metachronous HGBCL-DH-BCL2(-BCL6) cases arose through transformation of the preceding FL, or from a FL/HGBCL-DH-BCL2 common precursor cell (CPC; refs. 34, 35), we compared the mutational pattern of clonally related IGHV rearrangements of the two consecutive tumors of three patients. In the first two cases, diagnosed with IGH⁺ HGBCL-DH-BCL2-BCL6 (#19) and IGH^UND HGBCL-DH-BCL2 (#297), the FL and HGBCL malignant clones shared several amino acid substitutions within IGHV but also exhibited private mutations, suggesting divergent evolution from a CPC (Fig. 5A). The third patient suffered from an IGH^UND HGBCL-DH-BCL2-BCL6 (#245) whose malignant clone displayed a unique set of IGHV amino acid substitutions which added up to those acquired by the FL counterpart (Fig. 5A). A notable exception was a mutation causing an amino acid substitution that introduced an N-glycosylation motif within IGHV FWR1 which was restricted to the FL clone (Fig. 5A). These results support a model of branched evolution giving rise to the two tumors from a CPC, in which the selection of an N-glycosylation motif likely contributed to FL transformation while remaining dispensable for HGBCL-DH-BCL2 outgrowth. In summary, IGH silencing precedes HGBCL-DH-BCL2 onset, occurring in a precursor cell that can also give rise to FL following a distinct evolutionary trajectory.

Disruption of both IGH alleles through BCL2 and MYC (and/or BCL6) rearrangements occurring in trans may interfere with normal IGH gene expression in HGBCL-DH-BCL2, contributing to IGH silencing. To assess this, we conducted IGH break apart (BAP) DNA FISH analysis for 22 HGBCL-DH-BCL2(-BCL6) cases (Fig. 5B; Supplementary Fig. S5A and S5B; Supplementary Table S5A). Whereas most IGH⁺ HGBCL-DH-BCL2(-BCL6; 7/8; 87.5%) conserved one intact IGH locus, 43% (6/14) of IGH^UND tumors carried two disrupted IGH alleles (Fig. 5B; Supplementary Table S5A). IGH::BCL2 dual-fusion FISH in IGH^UND HGBCL-DH-BCL2(-BCL6) with two disrupted IGH loci scored positive in every case (n = 6; Fig. 5C; Supplementary Fig. S5C; Supplementary Table S5A). Instead, IGH::MYC fusions scored positive only in 2/6 of these tumors (Fig. 5C; Supplementary Table S5A). All eight IGH^UND HGBCL-DH-BCL2(-BCL6) tumors with single IGH locus breakage carried an IGH::BCL2 fusion, including two cases (25%) with an IGH::MYC rearrangement occurring in cis (Fig. 5C; Supplementary Table S5A). IGH⁺ HGBCL-DH-BCL2(-BCL6) showed predominant monoallelic IGH locus disruption (7/8 cases; 87.5%), mostly (6/7 cases) consisting of IGH::BCL2 fusions, whereas IGH::MYC rearrangements remained uncommon (2/8; 25%; Fig. 5C, Supplementary Table S5A). Collectively, IGH^UND HGBCL-DH-BCL2(-BCL6) undergo recurrent biallelic IGH locus disruption, which, yet, infrequently intercepts MYC.

The infrequent (∼25%) occurrence of t(8;14) in HGBCL-DH-BCL2 as compared with the high incidence detected in Burkitt lymphoma (>70%) postulates mechanism(s) alternative to AID-induced DNA breaks at IGH switch/intronic enhancer regions as causative of MYC rearrangements (32). To shed light on such process(es), we leveraged a pair of IGH^UND FL and HGBCL-DH-BCL2-BCL6 metachronous specimens (FL #1 and HGBCL-DH-BCL2-BCL6 #245). Most FL cells presented an IGH^UND phenotype, with residual cells expressing IGG (Fig. 5D). RNA in situ analyses confirmed expression of IGHG (and some IGHA) transcripts replacing IGHM (Fig. 5D). FISH analyses and whole genome sequencing (WGS) revealed biallelic disruption of IGH loci in FL cells through IGH::BCL2 and IGH::BCL6 rearrangements occurring in trans (Supplementary Fig. S5D–S5F; Supplementary Table S4F). The IGH::BCL6 rearrangement targeted the functional IGH allele intercepting the IGHG2 switch acceptor region, located downstream of the IGHG1 constant region expressed by the malignant B cells (Supplementary Fig. S5D). Clonal relationship between the HGBCL-DH-BCL2-BCL6 tumor and the preceding FL was supported by the sharing of IGH/KV gene rearrangements, IGH::BCL2 and IGH::BCL6 translocation breakpoints, IGH^UND phenotype, and IGHG expression (Fig. 5D; Supplementary Table S4F). WGS indicated the acquisition by IGH^UND HGBCL-DH-BCL2-BCL6 cells of a t(8;22) (q24;q11), confirmed by Sanger sequencing of the translocation breakpoint (Supplementary Table S4F). The t(8;22) juxtaposed a region downstream of MYC with the IGL Joining-2 (IGLJ2) gene segment, adjunct to its 12-recombination signal sequence (RSS; Supplementary Table S5B). Translocation breakpoints falling within (or 1–2 nucleotides away from) the IGLJ 12-RSS were also found in a subset of t(8;22)⁺ HGBCL-DH-BCL2 cases described in (32) and in the IGH^UND WILL-3 HGBCL-DH-BCL2 cell line (Supplementary Table S5B; ref. 36). Notably, in two t(8;22)⁺ HGBCL-DH-BCL2, cryptic 23-RSSs were also identified at the translocation breakpoint on chromosome (chr)-8, supporting a causal role of the RAG1/2 machinery in promoting DNA double-strand breaks (DSB) on both chromosomes involved in the translocation (Supplementary Table S5B). RNA-scope and RNA-seq analyses confirmed widespread RAG1/2 expression in IGH^UND HGBCL-DH-BCL2-BCL6 case #245 (Fig. 5E; Supplementary Fig. S5G; Supplementary Table S6A). A following screening of HGBCL-DH-BCL2(-BCL6) (n = 37) for RAG1/2 expression extended the fraction of recombinase positive cases to 32% (12/37), including both IGH^UND and IGH⁺ tumors (Fig. 5F; Supplementary Table S6B). RAG1/2 transcripts were also detected in multiple HGBCL-DH-BCL2 cell line models (DOGUM, DoGKiT, WILL-2, and WILL-3), often carrying t(8;22) and mostly lacking sBCR (Supplementary Fig. S5H and S5I). Together, these findings establish recurrent RAG1/2 expression in HGBCL-DH-BCL2(-BCL6), contributing to IGL::MYC rearrangements.

The recurrent involvement of IGLJ RSS sequences in t(8;22) translocations of HGBCL-DH-BCL2(-BCL6) could reflect an attempt of the lymphoma ancestor cell to revise antigen receptor specificity through IG light chain editing. Because BCR-edited B cells predominantly express IGL light chains, we screened HGBCL-DH-BCL2(-BCL6) (n = 49) for IGKC/IGLC transcripts in situ (Fig. 5G; Supplementary Table S6C). Ten of 14 (71%) IGH⁺ HGBCL-DH-BCL2(-BCL6) were IGKC⁺, whereas remaining tumors transcribed IGLC mRNAs, in line with the ∼2:1 IGK/L ratio seen in healthy mature B cells (Fig. 5G; Supplementary Table S6C). In sharp contrast, 81% of IGH^UND HGBCL-DH-BCL2(-BCL6; 21/26 cases) tumors expressed IGLC transcripts (Fig. 5G; Supplementary Table S6C). Nine additional tumors, including six IGH^UND HGBCL-DH-BCL2(-BCL6), expressed both IGKC and IGLC transcripts (Supplementary Fig. S5J; Supplementary Table S6C) and proteins (Supplementary Fig. S5J), suggestive of IG light chain isotype inclusion. Predominant IGL usage was similarly observed in IGH isotype-switched HGBCL-DH-BCL2(-BCL6) of an independent cohort (Supplementary Fig. S5K; Supplementary Table S6D; ref. 32). Differential IG light chain isotype usage between IGH⁺ and IGH^UND HGBCL-DH-BCL2(-BCL6) was validated profiling representative cases for IGK/LV rearrangements using 5′ rapid amplification of cDNA ends (5′RACE) technology (Supplementary Table S6E). Notably, in the RAG1/2⁺ IGH^UND HGBCL-DH-BCL2-BCL6 #245, several productive and nonproductive IGKV and IGLV rearrangements were concomitantly captured in combination with a single productive, hypermutated, IGHV rearrangement, consistent with ongoing IGV light chain editing (Supplementary Tables S3B and S6E) in the malignant clone. Similar conclusions were reached profiling IG light chain rearrangements of RAG1/2⁺ HGBCL-DH-BCL2 cell line models, extending the observation to a subset of DLBCL and Burkitt lymphoma counterparts (Supplementary Table S6F). Tumor lines recurrently displayed a sBCR/CD79B^UND phenotype or preferentially expressed IGL⁺ BCRs (Supplementary Fig. S5I; Supplementary Table S6F). In summary, IGH^UND HGBCL-DH-BCL2(-BCL6) exhibit hallmarks of IG light chain V gene editing, often associated with constitutive RAG1/2 expression.

To study the biological implications of RAG1/2 expression in HGBCL-DH-BCL2(-BCL6), we analyzed early-passage COH-DHL1 and COH-THL1 HGBCL-DH-BCL2(-BCL6) cell line models recently established from patients with progressive disease (37). COH-DHL1 cells exhibited IGH::BCL2 and IGL::MYC translocations (Supplementary Fig. S6A), whereas COH-THL1 cells disrupted both IGH loci through IGH::BCL2 and IGH::MYC rearrangements (Supplementary Fig. S6A and S6B). COH-THL1 cells carried an additional BCL6 rearrangement intercepting the chr-8 derivative [der(8)t(8;14;3)] (Supplementary Fig. S6A and S6B; ref. 37)]. COH-DHL1 cells showed surface IGH/K/L/CD79B extinction, extending previous evidence (Fig. 6A; Supplementary Fig. S6C; ref. 37). Similarly to IGH^UND HGBCL-DH-BCL2, COH-DHL1 cells transcribed a single, productive, somatically mutated, isotype-switched (IGHG1) IGHV rearrangement (Supplementary Table S6G). COH-DHL1 cells inactivated both IGKC loci through RAG-dependent κ-deleting element (KDE) rearrangements, followed by acquisition of an out-of-frame clonal IGLV rearrangement, reconciling with sBCR extinction (Fig. 6B; Supplementary Fig. S6D; Supplementary Table S6G). The COH-THL1 cell line was established from the PB of a treatment-relapsed patient with HGBCL-DH-BCL2-BCL6 (37). The primary tumor evolved shifting from IGK to IGL expression, which was linked to RAG1/2 expression (Fig. 6C) co-occurring with initial IGM detection. COH-THL1 cells consisted of a mixture of CD79B^UND/lo/IGH^UND and CD79B⁺/IGM⁺ cells, with the latter expressing either sIGK or sIGL chains (Fig. 6D; Supplementary Fig. S6D). FACS-purified IGH^UND COH-THL1 cells reconstituted over time distinct pools of sIGK⁺ or sIGL⁺ tumor cells (Fig. 6E), unmasking the contribution of IGL chain editing to the IGH^UND-to-IGH⁺ phenotypic switch. BCR⁺ COH-THL1 cells expressed several productive IGKV or IGLV rearrangements while sharing the same somatically mutated productive IGHV rearrangement (Supplementary Table S6G). Surface IGL⁺ COH-THL1 cells mostly carried oligoclonal out-of-frame IGKV rearrangements, supporting V-J recombination progressing from IGK to IGL loci (Supplementary Table S6G). Conversely, IGH^UND COH-THL1 cells transcribed a diverse set of out-of-frame IGKV and IGLV rearrangements, reconciling with the BCR-less phenotype. The same pool included cells which carried in-frame IGV light chain rearrangements fueling the IGH^UND-to-IGH⁺ switch (Supplementary Table S6G). An oligo/monoclonal representation of mostly out-of-frame IGLV gene rearrangements combined with biallelic IGKC deletion and KDE rearrangements was also observed in the WILL-2, WILL-3, and SC-1 HGBCL-DH-BCL2(-BCL6) cell line models (Supplementary Table S6F). Longitudinal follow-up of COH-THL1 primary cultures revealed a progressive erosion of IGK and IGL V gene rearrangements through ongoing IG light chain editing, favoring ultimately the expansion of an IGH^UND tumor population (Fig. 6F). IGH^UND COH-DHL1 and COH-THL1 cells expressed RAG1/2 transcripts at levels comparable with or higher than 697 pre-B leukemia cells serving as a positive control (Fig. 6G). Importantly, RAG1/2 expression was substantially downregulated in COH-THL1 cells undergoing spontaneous IGH^UND-to-IGH⁺ conversion, implicating a negative regulation imposed by sBCR expression on V-J recombination (Fig. 6H). We investigated whether the BCR/PI3K/AKT/FOXO1 axis modulated RAG1/2 expression in HGBCL-DH-BCL2(-BCL6), similarly to autoreactive B cells (38). Consistent with an inhibitory role exerted by the PI3K/AKT pathway on RAG1/2 expression, treatment with the PI3Kδ/γ dual inhibitor duvelisib induced a marked increase in RAG1/2 transcripts in five HGBCL-DH-BCL2(-BCL6) cell lines (DoGKiT, COH-THL1, COH-DHL1, WILL-2, and DOGUM), particularly in the IGH⁺ subset (Fig. 6I). Conversely, treatment with a small-molecule inhibitor (AS1842856) interfering with FOXO1 transcriptional activity consistently reduced RAG transcripts in HGBCL-DH-BCL2(-BCL6) cell lines, with a more pronounced effect on RAG1 (Fig. 6I). Coherent with these results, inducible IGH extinction in λ-MYC; B1-8f murine BCLs (Supplementary Fig. S6E; ref. 5) significantly increased Rag1/2 expression (Supplementary Fig. S6F), which restored in vivo sBCR expression (Supplementary Fig. S6G) in a fraction of IghB1-8^KO cells through productive V gene rearrangements on the second Igh allele (Supplementary Fig. S6H–S6J). Collectively, results support a model whereby weakening of the BCR/PI3K/FOXO1 axis in IGH^UND HGBCL-DH-BCL2(-BCL6) cells favors RAG1/2 reexpression, promoting IGL chain editing.

We investigated the mechanism(s) responsible for the downregulation of IGH protein in the IGH^UND HGBCL-DH-BCL2 COH-DHL1 model. A twofold reduction in IGHG1 transcript levels, compared with sIGG1-expressing NUDUL1 and DOGUM cells, could not explain the significant decrease in IGG1 protein levels observed in BCR^null COH-DHL1 cells (Supplementary Fig. S7A). In the absence of pairing IGL chains, nascent IGH chains may translocate from the ER to the cytoplasm for continuous degradation through the proteosome machinery. Supporting this scenario, treatment of COH-DHL1 cells with the proteosome inhibitor bortezomib led to a two-to-three fold increase in intracellular (IC) IGH protein levels, as assessed by flow cytometry and immunoblotting analyses (Fig. 7A). To confirm that the failure to pair with IG light chains accelerated IGH degradation, COH-DHL1 cells were individually complemented with IGK/IGL expression vectors. Random IGK/IGL chain specificities were chosen due to the lack of information of the original light chain expressed by COH-DHL1 cells. Complementation with five different IG light chains (either IGK or IGL) caused a substantial increase in IGG1 protein levels, favoring robust restoration of sBCR expression (Fig. 7B). Similar results were observed for the IGH^UND HGBCL-DH-BCL2 WILL-3 cell line (Supplementary Fig. S7B). For IGH^UND HGBCL-DH-BCL2 with potentially productive IGH/K/L V gene rearrangements, faulty IGH/L chain folding and/or pairing may account for the rapid degradation of these proteins. To assess this, we reconstituted HEK293T (non-B) cells with expression vectors for IGH/IGK (or IGL) pairs cloned from three primary IGH^UND HGBCL-DH-BCL2(-BCL6) cases. Flow cytometric analyses indicated robust expression of matched IGH and IG-light chains derived from two IGH^UND lymphomas, with a staining pattern compatible with correct IGH/L pairing (Supplementary Fig. S7C). In contrast, IGH/K pairs from HGBCL-DH-BCL2-BCL6 #245 showed poor expression and pairing. Notably, such defects were normalized converting IGHV/K sequences to their unmutated versions (Supplementary Fig. S7C). These results indicate a prevalent posttranscriptional mechanism enforcing continuous clearance of nascent IGH chains in IGH^UND HGBCL-DH-BCL2(-BCL6), in particular when cells lose IG light chain expression through consecutive rounds of nonproductive IG light chain V gene editing events or upon accumulation of IGV somatic mutations precluding correct expression, folding, and/or pairing of the antibody chains.

Selection for productive IGHV rearrangements in IGH^UND HGBCL-DH-BCL2 predicts conserved dependence of the malignant cells on residual traces of IGH chain protein. To test this, we disrupted the IGHG1 gene in COH-DHL1 cells by CRISPR/Cas9 gene editing using two independent guide RNAs (gRNA). Because flow cytometric detection of lowly expressed IGG1 was challenging, we instead evaluated the impact of the knockout (KO) by performing high-throughput sequencing of IGHG1 alleles and transcripts of the bulk culture of edited cells. Approximately two thirds of the edited IGHG1 genes contained insertions or deletions (indel) disrupting the coding frame (Fig. 7C). In sharp contrast, IGHG1 out-of-frame mutations were largely replaced by in-frame indels when the analysis was restricted to the sequence of the functional full-length IGHG1 transcript (Fig. 7C; Supplementary Tables S6H and S6I). These findings indicate that BCR^null COH-DHL1 cells remain dependent on residual IGH protein for continuous growth. Next, we investigated whether COH-DHL1 cells required the BCR signaling subunit CD79B for optimal growth. We disrupted the CD79B gene in COH-DHL1 cells by gene editing and tracked mutant cells over time by IC flow cytometry. Despite varying editing efficiencies (5%–60%) achieved with three different gRNAs (Fig. 7D), CD79B^null cells were consistently outcompeted by their CD79B-proficient counterparts (Fig. 7D). In summary, despite the irreversible loss of IG light chain expression, IGH^UND COH-DHL1 cells remain dependent on residual IGH chain and CD79B expression, which confers optimal competitive fitness.

Preferential intracellular CD79B distribution associated with IGH silencing may hinder the response of HGBCL-DH-BCL2(-BCL6) to polatuzumab vedotin (PV), an anti-CD79B–drug conjugate approved for first-line treatment of DLBCL not otherwise specified (NOS) and HGBCL (39). To investigate this, mixed cultures of IGH⁺ and IGH^UND COH-THL1 cells, differing in surface CD79B expression, were exposed to PV treatment. Whereas a single dose of PV killed most sCD79B-expressing cells, IGH^UND/CD79B^lo counterparts largely resisted the treatment, overtaking the culture (Fig. 7E). Reduced sensitivity to PV was also observed in the IGH^UND COH-DHL1 and DoGKiT HGBCL-DH-BCL2(-BCL6) cell lines, which behaved similarly to sCD79B-negative HT DLBCL cells (Fig. 7F and G). Consistent with residual sCD79B expression, the sBCR/IG light chain–null WILL-2 and SC-1 HGBCL-DH-BCL2(-BCL6) cell lines retained PV sensitivity, similar to their IGH⁺ counterpart, DOGUM (Fig. 7F and G). PV-mediated killing of HGBCL-DH-BCL2(-BCL6) required BCR-dependent internalization of the drug conjugate, as malignant B cells were largely unaffected by exposure to the unconjugated antibody, irrespective of sCD79B status (Fig. 7F). To investigate the response to BCR signaling inhibitors, sBCR^null HGBCL-DH-BCL2 models were treated with drugs targeting BCR-associated kinases SYK (R406), BTK (ibrutinib), and PI3Kδ/γ (duvelisib). BCR^null WILL-3, DoGKiT, COH-DHL1, and SC-1 HGBCL-DH-BCL2 cells failed to respond to all three inhibitors, with IC₅₀ values remaining unreached (>10 μmol/L; Supplementary Table S6J). Ibrutinib-resistant WILL-2 lymphoma cells displayed residual sensitivity to PI3Kγ/δ (Supplementary Table S6J). Consistent with constitutive activation of the RAS/MAPK pathway linked to a KRAS^G12V gain-of-function mutation (37), COH-DHL1 cells were highly sensitive to pharmacologic MEK inhibition (Fig. 7H). Next, we performed a targeted small-drug screening in IGH^UND COH-THL1 and COH-DHL1 cells to identify vulnerabilities. HGBCL-DH-BCL2(-BCL6) models showed high sensitivity to inhibitors of BCL2, mTOR, regulators of proteostasis, DNA damage response regulators (PARP1/2, CHK1/2, ATR, and WEE1), cell-cycle progression (CDK4/6, CDK1/2/5/9, and PLK), BET proteins, HDAC class-1, and nuclear export regulators (Fig. 7H). The most potent response was observed with the nicotinamide phosphoribosyltransferase–specific inhibitor (NAMPTi) OT-82, targeting the NAD⁺ salvage pathway, with IC₅₀ values reaching the picomolar range (Fig. 7H). Another NAMPTi, KPT9274, also demonstrated potency in the single and double digit nanomolar range in IGH^UND COH-DHL1 and COH-THL1, respectively. Altogether, sBCR silencing in HGBCL-DH-BCL2 models reduces efficacy of CD79B-targeting antibody–drug conjugates as well as inhibitors of BCR-proximal signaling while preserving vulnerability to inhibition of selected modulators of apoptosis, cell-cycle progression, DNA damage response, chromatin remodeling, and particularly to NAMPTi.

IHC screening of 258 consecutive DLBCL cases revealed that approximately 22% exhibit undetectable IGH immunoreactivity across the tumor bulk. Flow cytometric assessment of surface IG light chain expression and PLA determination of IGH/CD79 complexes indicated BCR silencing in IGH^UND lymphomas. The significant enrichment for cases with BCR shutdown among GCB-type DLBCL (47%) suggests a link between the tumor COO and the IGH^UND state, extending previous conclusions (17). IGH^UND lymphomas may inherit mechanisms of BCR downregulation from their GC B-cell precursors (40–42), with transformation events trapping malignant B cells in a DZ-like state adapted to thrive with limited-to-undetectable IGH/BCR expression (43). Among GCB-type lymphomas with DLBCL morphology, HGBCL-DH-BCL2(-BCL6) emerged with more than 65% displaying IGH-silencing. This result reconciles several earlier findings (32, 44–47). The concordance across IHC, flow cytometry, and PLA assays establishes the robustness of the IGH^UND phenotypic screen, warranting additional investigations to explain the recently reported lower frequency of IGH^UND HGBCL-DH-BCL2 cases (32).

Transcriptome and gene mutational analyses underscore the relevant impact of IGH silencing on the biology of HGBCL-DH-BCL2(-BCL6). IGH^UND tumors preferentially exhibit a GC DZ-like molecular program characterized by increased expression of cell-cycle regulators and upregulation of metabolic and stress-response pathways in association with heightened MYC expression (22, 48, 49). These features may represent adaptive responses selected by malignant cells to resist BCR shutdown (5). Some IGH^UND tumors retain GC LZ-like transcriptional traits, likely driven by BCR surrogate signals or residual sBCR. These tumors display an indeterminate molecular high-grade signature (21) and transcriptional similarities to IGH⁺ cases, suggesting an origin from GC B cells preserving DZ/LZ interconversion. LZ-like IGH⁺ HGBCL-DH-BCL2(-BCL6) exhibited richer interactions with the TME, linked to stronger recruitment of T cells, whose activity appears dampened by the local expression of immunomodulatory factors. Conversely, the silencing of BCR in IGH^UND HGBCL-DH-BCL2 may represent a selected trait reproducing the immune cell evasive microenvironment of the GC DZ area (https://pmc.ncbi.nlm.nih.gov/articles/PMC10984086/).

The preservation of potentially productive IGHV rearrangements in IGH^UND HGBCL-DH-BCL2(-BCL6) indicates a selective advantage conferred to the malignant B cells by residual IGH molecules. Functional experiments reveal that disruption of the IGHG1 gene on the productive IGH allele in the BCR^null (i.e., IG light chain–deficient) COH-DHL1 HGBCL-DH-BCL2 cell line model is incompatible with tumor B-cell persistence in a competitive setting. Similar results are observed disrupting CD79B, pointing to the requirement for an intact IGH/CD79B signaling axis to sustain the growth of BCR^null HGBCL-DH-BCL2 cells. Future studies will determine whether residual IGH supports HGBCL-DH-BCL2 fitness by sustaining ER-dependent mitochondrial function in a BCR/SYK-independent fashion, as observed in Burkitt lymphoma models and primary B cells (https://www.biorxiv.org/content/10.1101/2022.03.11.483940v1; ref. 50). Studies in IGH^UND HGBCL-DH-BCL2(-BCL6) models reveal the importance of the proteosome machinery for continuous clearance of nascent IGH chains, unless complemented with pairing IG light chains. These data, along with the preserved folding and pairing of exogenously expressed IGH/L chains from two representative IGH^UND HGBCL-DH-BCL2, propose a posttranslational mechanism primarily responsible for IGH shutdown in these tumors. Reduced IGH transcription and/or mRNA stability may represent an additional layer of regulation contributing to IGH silencing in HGBCL-DH-BCL2, consistent with transcriptome analyses of EZB-MYC⁺ DLBCL (17, 18).

Predominant IGHM expression in IGH⁺ HGBCL-DH-BCL2(-BCL6) contrasted with exclusive transcription of class-switched IGH constant region genes in the IGH^UND counterparts. IGM⁺ tumors expressing AID may escape constitutive IGH CSR through the selection of monoallelic intraswitch-μ deletions in cis (51) while not sparing the nonproductive IGH allele disrupted by the BCL2 translocation (33). In contrast, IGH^UND HGBCL-DH-BCL2(-BCL6) completed biallelic IG CSR, establishing an origin from GC-experienced IGG⁺/A⁺ B cells.

In healthy B cells, IGM BCRs strictly depend on the CD79 subunits for signaling, whereas IGG receptors can bypass such requirement through own signaling-competent cytoplasmic domains (52–54). Moreover, surface BCR levels are differentially controlled between IGM⁺ and IGG⁺ GC B cells, with the IGG intracellular domain favoring stronger BCR internalization (55). These properties, along with evidence of lower transcript levels for BCR components in isotype-switched FL (56) and HGBCL-DH-BCL2(-BCL6) (this study), as well as weaker signaling from switched BCRs in DLBCL (https://www.biorxiv.org/content/10.1101/2024.04.12.585865v2), may explain the stronger propensity of IGG B cells to transform into IGH^UND HGBCL-DH-BCL2. Surface BCR shutdown may contribute to the transformation process protecting IGH class-switched HGBCL-DH-BCL2(-BCL6) precursor cells from antigen-driven terminal differentiation (53, 57, 58). The identification of an IGA-expressing HGBCL-DH-BCL2-BCL6 harboring a heterozygous CD79A^Y188* truncating mutation, which weakens BCR signaling (4), supports the notion that isotype-switched tumors are selected for reduced BCR signaling strength. Conversely, occasional sIGG1⁺ HGBCL-DH-BCL2 may select truncating mutations within the IGG1 intracellular domain to sustain surface BCR expression (https://www.biorxiv.org/content/10.1101/2024.04.12.585865v2).

Studies of metachronous FL-HGBCL-DH-BCL2(-BCL6) pairs indicate that IGH expression and class choice are established early during HGBCL-DH-BCL2 genesis, likely within the CPC, and potentially as early as the stage of in situ follicular neoplasia (34, 35, 59). IGM⁺ FL cases frequently acquire N-linked glycosylation motifs within IGV domains through IG somatic hypermutation mostly preserving them throughout tumor evolution (60–62). The incorporation of oligomannose-rich glycans within these domains facilitates interactions between FL BCRs and host or bacterial mannose-binding lectins, delivering signals that support antigen-independent growth of malignant B cells (31, 63–66). The recurrent acquisition of N-glycosylation motifs within IGHV genes of IGH⁺ HGBCL-DH-BCL2 suggests a similar reliance on mannose-rich lectin binding for tumor persistence. IGH-switched FL cases more often express autoreactive BCRs than their IGM⁺ counterparts (65, 67, 68). Whereas the self/autoreactive nature of BCRs expressed by the IGH^UND HGBCL-DH-BCL2 precursor cell remains undetermined, mutational analyses indicate that autoreactive B cells can select over time gene mutations and structural variants in common with DLBCL (69, 70). Given the unrestricted entry of autoreactive B cells into the long-lived IG-switched memory B-cell pool (71), we postulate that t(14;18)⁺ IG-switched autoreactive B cells serve as potential reservoir for IGH^UND HGBCL-DH-BCL2 precursor cells, protected from self-antigen–induced terminal differentiation by sBCR downregulation. Interestingly, the acquisition of N-glycosylation motifs within IGHV genes was conserved in the bulk tumor population of most IGH^UND HGBCL-DH-BCL2(-BCL6) tumors. Residual IGH molecules in IGH^UND tumors may continue to interact with local (intracellular?) lectins while escaping from the requirement for (self-) antigen recognition. Alternatively starting from their precursors, IGH^UND malignant B cells may select N-glycosylation motifs within IGV domains to tune down self-reactivity, similarly to IGG⁺ DLBCL and memory B cells (72, 73).

Similarly to self-reactive B cells that undergo IG light chain editing in an attempt to revise BCR specificity (74), IGH^UND HGBCL-DH-BCL2(-BCL6) exhibited a strong bias toward IGL light chain usage. In line with observations in normal B cells and FL, IGH shutdown likely acts as a trigger for IG light chain revision in HGBCL-DH-BCL2 (67, 75). Longitudinal tracking of IGK/LV gene rearrangements provided formal proof of ongoing IG light chain editing in a RAG-expressing HGBCL-DH-BCL2-BCL6 primary tumor, whereas its COH-THL1 cell line derivative (37) enabled real-time monitoring of the impact of IG light chain revision on lymphoma evolution. Low-passage COH-THL1 cultures showed remarkable intratumor BCR diversification, promoting the coexistence of sIGK⁺, sIGL⁺, and sIGK⁺/L⁺ double-producer tumor cells alongside a population of IGH^UND cells. Such heterogeneity underscores the potential of HGBCL-DH-BCL2(-BCL6) cells to bidirectionally switch between IGH⁺ and IGH^UND states, fueled by secondary IG light chain rearrangements. The IGH^UND/+ mixed phenotype described in a subset of HGBCL-DH-BCL2(-BCL6) may represent a snapshot of this dynamic process.

Whereas the COH-THL1 model captures the reversible nature of BCR silencing in HGBCL-DH-BCL2, the COH-DHL1 model represents an evolutionary endpoint of the disease. In this trajectory, a GC-derived IGG1/K⁺ HGBCL-DH-BCL2 cell (or its precursor) irreversibly extinguished BCR expression through consecutive rounds of nonproductive RAG-mediated IGK/LV rearrangements while retaining IGHV productivity. A similar dynamic can explain BCR extinction in several additional HGBCL-DH-BCL2(-BCL6) cell lines described in this study.

Our data support the concept that BCR silencing creates a permissive state for RAG1/2 reexpression in HGBCL-DH-BCL2. This phenomenon, supported by studies in cell line models, finds confirmation in murine λ-MYC BCLs, in murine and human immature and GC B cells (76, 77), and in DLBCL (78). Whereas RAG1/2 (re)expression is mostly expected in IGH^UND HGBCL-DH-BCL2, it is also observed in the IGH⁺ counterparts. For the latter cases, particular gene mutations and/or microenvironmental factors may explain RAG1/2 reexpression, with BCR signaling counteracting such effect in proportion to its strength. Preliminary evidence assigns to the PI3K/AKT/FOXO1 axis a key function in RAG1/2 gene regulation in HGBCL-DH-BCL2, consistent with recent work in autoreactive murine B cells (38).

The low frequency of t(8;14) in HGBCL-DH-BCL2(-BCL6) supports the hypothesis that malignant B cells require the preservation of a functional IGH transcriptional unit once the second IGH allele is disrupted by the t(14;18) translocation (32). At the same time, we report an unexpectedly high prevalence of biallelic disruption of IGH loci in IGH^UND HGBCL-DH-BCL2. Unlike IGH⁺ tumors, IGH^UND lymphomas are characterized by a BCR-silenced state, potentially allowing for tolerance to, or even selection for, bi-allelic IGH translocations. BCR silencing in the HGBCL-DH-BCL2 precursor could relieve the selective pressure against translocations disrupting both IGH alleles. In turn, these rearrangements may contribute to reduce IGH gene expression by interfering with long-range cis interactions between the functional IGH transcriptional unit and the 3′ IGH regulatory regions (79).

A case study consisting of IGH^UND FL-HGBCL-DH-BCL2-BCL6 metachronous specimens causally linked the acquisition of t(8;22)(q24;q11) to the activity of the RAG1/2 recombinases. Sequencing of the translocation breakpoint revealed the juxtaposition of MYC to IGJL2, replacing the 12-RSS sequence, consistent with illegitimate V-J recombination. Similar observations in an independent series of HGBCL-DH-BCL2 (32) identified additional cases with genetic footprints of aberrant VJ recombination as driver of t(8;22). The causal relationship between RAG1/2 expression and IGL::MYC translocations driving IGH^UND lymphomas from IGH-switched GC B cells recalls earlier observations in compound mutant mice lacking Tp53 and the DNA DSB repair protein XRCC4 in mature B cells (80, 81). Consistent with the latter findings, our data support a model in which IGH-switched t(14;18)⁺ lymphoma precursor GC B cells activate IGLV gene editing in response to chronic sBCR downregulation (or weakening of PI3K/AKT signaling). This process, driven by RAG1/2 reexpression, exposes IGLJ loci to frequent DNA DSBs at RSS sites, which can occasionally coincide with similar breaks targeting the MYC locus downstream region (32), thereby promoting t(8;22) translocations and onset of HGBCL-DH-BCL2. In line with this model, we identified cryptic 23-RSS adjunct to the translocation breakpoint on chr-8 in a fraction of t(8;22)⁺ cases. Aberrant V-J recombination may account for t(8;22) rearrangements in several previously reported BCLs with MYC and BCL2 rearrangements, including cases with a previous history of FL (82–84). RAG-induced DNA breaks at cryptic RSS may also contribute to non-IG::MYC rearrangements, which frequently occur in HGBCL-DH-BCL2 (32), particularly in cases with impaired DNA damage repair (85, 86). The conserved expression of RAG1/2 in a subset of Burkitt lymphoma (26) and DLBCL (87) anticipates a broader contribution of the recombinases to the acquisition of structural variants in multiple aggressive forms of GC B cell–derived lymphomas.

Recent clinical studies have highlighted reduced efficacy of chemo-immunotherapies based on the CD79B-targeting agent PV in GCB-type DLBCL and HGBCL-DH-BCL2 (88, 89). IGH silencing in HGBCL-DH-BCL2, causing intracellular CD79B retention through BCR downregulation, aligns with the limited sensitivity to PV observed in IGH^UND HGBCL-DH-BCL2 cell line models, offering an explanation for the clinical findings (88, 90). We report two sBCR^null HGBCL-DH-BCL2 cell lines, SC-1 and WILL-2, that remain sensitive to PV as a result of preserved surface CD79B expression. Exploiting mechanisms favoring BCR-independent surface CD79B delivery in malignant B cells (10) are expected to improve the responsiveness of patients with GCB-DLBCL and HGBCL-DH-BCL2 to CD79B-targeting agents (42).

Inhibitors of the BCR proximal kinases SYK, BTK, and PI3Kδ/γ were mostly ineffective in BCR^null HGBCL-DH-BCL2 cell lines. Conversely, malignant B cells showed remarkable sensitivity to NAMPTi, consistent with recent studies in BCL2-rearranged GCB-DLBCL cell line models (91). These findings identify the NAD⁺ salvage pathway as a critical metabolic vulnerability in IGH^UND HGBCL-DH-BCL2(-BCL6), representing a promising therapeutic target for the treatment of these aggressive lymphomas.

A large body of experimental evidence has consolidated the notion that the BCR critically contributes to the onset and persistence of several mature B-cell neoplasms, rendering it an effective target of therapy (3). In this study, we present first-time evidence that chronic BCR silencing shapes the biology and evolution of HGBCL-DH-BCL2(-BCL6) from its putative t(14;18)⁺ GC DZ B-cell precursor. By inducing IG light chain editing, IGH silencing actively contributes to lymphomagenesis by facilitating t(8;22) translocations and fueling disease progression through intraclonal BCR diversification, which can lead to irreversible antigen receptor extinction. The conserved dependence of IGH^UND HGBCL-DH-BCL2 cells on residual IGH chains, particularly in cases in which IG light chain expression is irreversibly extinguished by receptor revision, reveals a BCR-independent protumoral role for IGH in these, and possibly other, BCL subtypes.

Our findings call for careful evaluation of CD79B-targeted therapies and highlight the potential of NAMPTi to fight IGH^UND HGBCL-DH-BCL2 and more broadly GCB-type DLBCL NOS (91). Our data provide a strong rationale for routine IGH IHC monitoring to guide treatment decisions and improve outcomes in aggressive GC-derived BCLs.

HGBCL-DH-BCL2(-BCL6) tumors represent infrequent B-cell malignancies, rendering biological studies challenging, and therefore identification of improved therapies remains an unmet medical need. We tackled this, collecting through a multicenter study more than 100 HGBCL-DH-BCL2(-BCL6) specimens, ensuring the largest in situ study on IGH expression in this aggressive lymphoma subtype to date. Despite the limited number of HGBCL-DH-BCL2(-BCL6) cases clustered for IGH expression and profiled for bulk and spatial transcriptomics, we extrapolated information on tumor-infiltrating T cells, which we successfully validated by IHC/IF analyses in a largely independent cohort. The absence of reference germline DNA and the limited number of profiled HGBCL-DH-BCL2(-BCL6) cases have constrained the power of the genomics information, highlighting the need to validate and extend these findings in a larger cohort stratified for IGH expression. The analysis of an initial HGBCL-DH-BCL2-BCL6 case study has unraveled the connection between IGH silencing and RAG1/2 reexpression. This finding has been confirmed through a comprehensive analysis of additional HGBCL-DH-BCL2(-BCL6) cases, gaining mechanistic insights using a combination of patient-derived and mouse lymphoma models. The clinical impact of the IGH^UND phenotype on DLBCL and HGBCL-DH-BCL2 patient response to CD79B-targeted and more widely used therapies such as R-CHOP awaits controlled trials.

The study was approved by institutional ethical committee boards of participating sites: NP3719-FF-BCR2 and NP4296-CoronAId (ASST Spedali Civili, Brescia), INT 218/21 (Fondazione IRCCS Istituto Nazionale Tumori, Milano), n. 707_2020 (Fondazione IRCCS Ca’ Granda Ospedale Maggiore Policlinico, Milano), MAD-2022-12376707_ID_5444_Prot_003275/23 (Fondazione Policlinico Universitario A. Gemelli, Roma), and Wakayama Medical University Approval Number 4333, complying with ethical regulations and in accordance with the Declaration of Helsinki, at the respective collection centers: Azienda Socio-Sanitaria Territoriale degli Spedali Civili di Brescia (Italy), San Raffaele Hospital (Milan, Italy), Istituto Nazionale Tumori Milano (Italy), Azienda Ospedaliero Universitaria Padova (Italy), and Fondazione Policlinico Universitario A. Gemelli, Wakayama Medical Center, Japan. Written informed consent was obtained from all patients enrolled in the study. A total of n = 368 patients with a diagnosis of DLBCL, including GCB or non-GCB subtypes, or HGBCL with MYC and BCL2 rearrangements (with and without BCL6 rearrangements; “DH”, HGBCL-DH-BCL2) were included in the study. Two main cohorts of patients were enrolled. The first consisted of 258 consecutive DLBCL and HGBCL-DH-BCL2(-BCL6) cases for which residual tumor material was available (Supplementary Table S1A). These included: (i) 156 formalin-fixed paraffin-embedded (FFPE) tumor biopsies (# 1–156, Supplementary Table S1A) diagnosed between January 2016 and December 2020 at the Pathology unit of Spedali Civili, Brescia (Italy); (ii) 75 tissue microarray (TMA) biopsies (# 157–231) diagnosed between January 2011 and December 2016 at Azienda Ospedaliero-Universitaria di Padova (Italy); and (iii) 27 consecutive DLBCL and HGBCL (# 331–352) specimens prospectively collected between October 2020 and December 2023 at Spedali Civili, Brescia (Italy). Low-quality samples with extensive fibrosis and/or necrosis and/or with insufficient material (i.e., biopsies smaller than 15 mm) were excluded from analyses. Cases with both nodal (n = 177) or extranodal (n = 81) biopsies were included in the study. The second cohort was restricted to GCB MYC and BCL2 dual expressor lymphomas (MB2 DE) only, comprising 57 MB2 DE non-HGBCL-DH-BCL2(-BCL6) and HGBCL-DH-BCL2(-BCL6) from the consecutive cohort, and 109 additional cases, (Supplementary Table S1C), the latter obtained through a multicenter collection from Spedali Civili, Brescia (Italy), San Raffaele Hospital, Milan (Italy), Ospedale di Cremona (Italy), and Istituto Nazionale Tumori, Milan (Italy). Two biopsies from one additional HGBCL-DH-BCL2(-BCL6) case (#329A and B) were provided by the City of Hope Comprehensive Cancer Center. A total of 104 HGBCL-DH-BCL2(-BCL6) cases were included in this cohort (Supplementary Table S1D). Reactive tonsil and LN samples from healthy donors were used as controls.

The DLBCL COO was established according to Hans’ algorithm (CD10, BCL6, and MUM-1 expression), together with MYC and BCL2 protein expression (cutoff for positivity, ≥40% and ≥50%, respectively). IGH immunostainings were centralized and systematically evaluated by two pathologists (Spedali Civili, Brescia), selecting tissue areas devoid of diffuse fibrous tissue, numerous apoptotic cells, diffuse necrosis, or with diffuse intercellular tissue stain. As further confirmation, immunostained digitalized images (Aperio Scanscope CS from Leica Microsystems) were evaluated by three additional pathologists (San Raffaele Hospital, Milan; Spedali Civili, Brescia; and the University of Padova, Padova), reaching a concordance of 0.99 (Cohen’s κ). Any nonconcordant diagnoses among pathologists were rereviewed, and resolution was achieved by performing independent immunostainings in a second center.

FFPE tissue sections were immunostained for IGH chains IGM, IGD, IGG, IGA, and CD79B(cl.128-4C5), CD20, MYC, and CD3 proteins using the Bond Max/Bond-III (Leica Microsystems) or the BenchMark Ultra (Ventana Medical Systems) autostainers. Stainings with anti-AICDA, CD79B (cl. EPR6861), and CD32B antibodies were performed manually. Antibodies used for analyses are listed in Supplementary Table S7. Deparaffinized slides were subjected to antigen unmasking procedure using a water bath at 98°C for 40′ in buffer Tris-EDTA, pH 9.0 (AICDA and CD32B), or citrate, pH 6.0 (CD79B), and then incubated for 1 hour at room temperature (RT) with the antibody diluted 1:1,000 and 1:250 and 1:200, respectively. Novolink Polymer was used as a detection system, and chromogen reaction was developed using 3,3′diaminobenzidine. Immunostained sections were photographed with an Olympus DP-73 WDR digital camera mounted on an Olympus BX60 microscope. CellSens standard 1.17 was used as acquisition software. Digitalization through Aperio ScanScope CS (Leica Microsystems) was also used to take images at low magnification. Positivity was determined by the presence of complete membranous and/or cytoplasmic reactivity in tumor cells. Cases were classified according to the percentage of IGH-expressing cells. Specifically: cases with more than 90% of tumor cells exhibiting IGH immunoreactivity were categorized as IGH-positive (IGH⁺). Conversely, when IGH reactivity was undetectable in more than 90% of cells, the tumor was classified as IGH-undetectable (IGH^UND). Cases showing IGH expression in a range between 10% and 90% of atypical cells were classified as IgH^UND/+. Plasma cells acted as internal IGH staining control. AICDA stain was scored in HGBCL-DH-BCL2(-BCL6) cases (n = 26) and defined as positive when reactivity was detected in more than 25% of tumor cells. CD79B expression in HGBCL-DH-BCL2(-BCL6; n = 49) was scored as cytoplasmic versus membrane-bound, regardless of intensity. Quantification (number of cells/mm²) of CD3⁺ infiltrating lymphocytes was performed on digitalized slides from IGH⁺ and IGH^UND DHL/THL cases (n = 55) using the Aperio IHC Nuclear Image analysis algorithm in ImageScope v.12.3.2.8013 (Leica Microsystems). Cell counting was performed in 1 mm² tumor-rich areas.

Multiplex IF analysis on FFPE sections of HGBCL-DH-BCL2(-BCL6) was performed using the MACSima platform (Miltenyi), with antibodies listed in Supplementary Table S7. Three-micrometer sections were deparaffinized, rehydrated, and antigen-unmasked using TEC (Trizma base, EDTA, and sodium citrate tribasic dihydrate) buffer (pH 9) at 85°C, followed by heating at 98°C for 20 minutes and cooling to 85°C. After washing, sections were mounted on a MACSwell Imaging Frame, and DAPI (4′,6-diamidino-2-phenylindole) prestaining was performed before imaging. The MACSima system automated staining cycles, including IF staining, washing, multifield imaging, and signal erasure via photobleaching. Twelve fields of view were analyzed for IGH^UND and nine for IGH⁺ HGBCL-DH-BCL2, respectively. Imaging was conducted using an epifluorescence setup with 20× and 2× objectives, custom LED excitation, and an sCMOS camera. Autofocusing utilized hardware and image-based optimization. Image processing corrected optical artifacts and subtracted autofluorescence, with datasets analyzed using magnetic-activated cell sorting (MACS) iQ View Software.

Human BCL cell lines SU-DHL-4 (M, 38y), SU-DHL-6 (M, 43y), DoGKIT (M, 56y), WILL-2 (F, 63y), WILL-3 (F, 60y), SC-1 (M, 67y), NU-DUL-1 (M, 43y), NU-DHL-1 (M, 73y), 697 (M, 12y), DOGUM (F, 49y), HT (M, 70y), and RAMOS (M, 3y) were purchased from DMSZ and cultured according to instructions reported by the biorepository. COH-DHL1 (M, 58y) and COH-THL1 (M, 78y) cell lines were cultured in RPMI-1640 medium (Euroclone) supplemented with 10% heat-inactivated FBS (Sigma-Aldrich), 2 mmol/L L-glutamine, and 10mmol/L HEPES (Euroclone) at a density of 0.5−1.5 × 10⁶ cells/mL. λ-MYC;B1-8f mouse lymphoma lines (5) were cultured at 2 × 10⁵ cells/mL in high-glucose DMEM supplemented with 10% heat-inactivated FBS, 0.1 mmol/L nonessential amino acids, 1 mmol/L sodium pyruvate, 50 μmol/L β-mercaptoethanol (Thermo Fisher Scientific), and 2 mmol/L L-glutamine (Euroclone). Cell lines were periodically tested for Mycoplasma (MycoAlert Mycoplasma Detection Kit, Lonza) and grown at 37°C in a humidified atmosphere with 5% CO₂. Cell line authentication was performed using the GenePrint 10 System (10-Locus STR System for Cell Line Authentication, Promega).

IGG1⁺ SU-DHL-4 and IGM⁺ SU-DHL-6 lymphoma cell lines acted as controls for IGH isotype-specific IHC stainings (Supplementary Fig. S1C). Prewarmed (40°C) low-melting 1% agarose (Euroclone GellyPhor LM) solution (in PBS) was mixed with 1 × 10⁷ lymphoma cell suspensions and solidified in a conical stamp for at least 30 minutes. Following agarose embedding, blocks were removed from the stamp, fixed with 4% paraformaldehyde for 4 hours, washed with 70% ethanol, and embedded as paraffin blocks. The same procedure was used to establish IGH/L status in COH-DHL1 and COH-THL1 HGBCL-DH-BCL2(-BCL6) cell line models.

Anti-human CD79B mouse monoclonal Ab (mAb; cl.128-4C5) was established by B.F. at the OncoHematologic Research Center (CREO), Perugia University. The mAb was raised against a recombinant peptide corresponding to the CD79B protein’s amino terminal extracellular portion. Specificity was confirmed by immunoblotting and flow cytometry using CD79B-positive (normal B cells and transformed B cell lines SU-DHL16 and RIVA) and negative (T lymphocytes and the acute myeloid leukemia cell line KG-1) cells, matching the results with those obtained with a commercial anti-CD79B antibody. Immunoreactivity against HEK293T cells complemented with CD79B cDNA or flow cytometric and immunoblotting analyses on CD79B gene-edited/KO SU-DHL-16 cells were included as additional 128-4C5 specificity tests. Anti-CD79B staining on FFPE tissue sections was performed on BOND-MAX autostainer (Leica Biosystems) using the EDTA 15 antigen retrieval protocol and incubating the antibody (dilution 1:4) for 30 minutes.

Primary BCL suspensions were prepared from freshly collected nodal biopsies. After surgical resection, tumor specimens were placed in isotonic PBS buffer. A 5 to 10 mm³ tissue fragment was transferred in dissociation medium (RPMI-1640 supplemented with L-glutamine and penicillin/streptomycin), finely minced, and subjected to tissue dissociation into single-cell suspension using a Gentle MACS dissociator (Miltenyi Biotec). Cell suspensions were filtered through a 70-μm cell strainer to remove aggregates and centrifuged at 1,200 RPM for 5 minutes at 4°C. Residual red blood cells were lysed incubating cell suspension for 3 minutes at 4°C with 1× red blood cell lysis buffer (BioLegend). Cell suspensions were washed with PBS, counted, and used for downstream analysis or cryopreserved in freezing solution (90% FBS and 10% dimethylsulfoxide) in liquid nitrogen.

B lymphoma cells from primary tumors or cell lines (0.5 - 1 x 10⁶ cells) were washed in PBS and stained in FACS Buffer (1% BSA, 2 mmol/L EDTA buffer, and 0.01% sodium azide in PBS) with combinations of fluorescently labeled antibodies (listed in Supplementary Table S7) for 20 minutes in the dark at 4°C. Stained cells were washed twice with FACS buffer and analyzed on a BD FACSCanto II cell analyzer (BD Biosciences) or on a Cytek Aurora spectral analyzer (Cytek Biosciences). Alternatively, cells were purified in sorting medium (DMEM supplemented with 30% FBS, 2 mmol/L L-glutamine, and 1× penicillin/streptomycin solution) using a FACS Aria cell sorter (BD Biosciences). Data analysis was performed using FACSDiva (BD Biosciences), SpectroFlo (Cytek Biosciences), and FlowJo v. 10.9.0 (BD Biosciences) software.

FISH analysis was performed on FFPE tissue sections using the IQFISH break apart probes for MYC (8q24), BCL2 (18q21), BCL6 (3q27; Agilent Technologies), and IGH (14q32.33) loci (ZytoLight SPEC IGH Dual Color Break Apart Probe from ZytoVision); IGH/MYC/CPP8 Tri-color Fusion/Translocation FISH Probe Kit (CytoTest), ZytoLight SPEC BCL2/IGH Dual Color Dual Fusion Probe (ZytoVision), and IGL/MYC Dual Fusion/Translocation LR FISH Probe Kit (CytoTest) were used to identify partners of IGH and IGL rearrangements. The FlexISH BCL2/BCL6 DistinguISH Probe (ZytoVision) was used to check if BCL2 and BCL6 rearrangements occurred in cis. FISH was carried out according to the manufacturer’s guidelines. At least n = 100 cells were evaluated, and FISH images were captured at × 100 magnification using the Nikon Eclipse 90i microscope and elaborated using Genikon software 3.7.17 (Nikon) or at × 63 magnification using the Leica DM6000B automated microscope and elaborated using CytoVision MB8 software (Leica Microsystems).

RNA in situ hybridization was performed on FFPE sections using Advanced Cell Diagnostics probes complementary to IGHM, IGHD,IGHG, and IGHA, as well as IGKC and IGLC, following the manufacturer’s instructions. Probes complementary to UBC and DAPB were used to control for RNA quality before hybridization with target gene-specific probes. Briefly, freshly cut 4-μm FFPE sections were deparaffinized in xylene, treated with RNAscope hydrogen peroxide solution (Advanced Cell Diagnostics) for 10 minutes at RT, and subjected to retrieval for 15 minutes at 98°C in 1 × target retrieval reagent (Advanced Cell Diagnostics) and protease treatment with RNAscope Protease Plus (Advanced Cell Diagnostics) at 40°C for 30 minutes. Hybridization was performed for 2 hours at 40°C, and signal was revealed using RNAscope 2.5 HD Detection Reagent FAST RED (Advanced Cell Diagnostics). Only cytoplasmic signals were scored. Tissue-infiltrated plasma cells expressing high levels of immunoglobulin transcripts acted as internal positive controls.

Human RAG1 and RAG2 probes (Advanced Cell Diagnostics) were hybridized using RNAscope 2.5 HD Duplex Reagent Kit (Advanced Cell Diagnostics), adopting an extended 1-hour incubation in Amplifier reagent 5 and 30-minute incubation in Amplifier reagent 6 for RAG1/RAG2 hybridization. Slide digitalization was achieved using an Aperio CS2 digital scanner (Leica Biosystems) with ImageScope software (ImageScope version 12.3.2.8013, Leica Biosystems). Quantitative in situ mRNA analysis was performed on whole sections using HALO software (v3.2.1851.229, Indica Labs). A threshold of 20% of cells expressing RAG1 and/or RAG2 was used as a cutoff for positivity.

PLA was performed using NaveniBright HRP Kit (Navinci Diagnostics) according to the manufacturer’s instructions. Polyclonal rabbit anti-human IGM (Leica Biosystems) or rabbit anti-human IGG (Leica Biosystems) was matched with mouse anti-human CD79A (Leica Biosystems) or mouse anti-human CD79B (cl. 128-4C5), respectively. Rabbit anti-human CD79A (Abclonal) was matched with mouse anti-human CD79B. Stainings with single primary antibodies were used as negative controls. PLA signals were quantified in 10 nonoverlapping GC areas of human tonsils and 5 non-overlapping fields of view for IGH⁺ and IGH^UND cases at × 200 magnification using HALO software (v3.2.1851.229; Indica Labs).

DNA was extracted from FFPE tissue specimens using QIAamp DNA FFPE Tissue Kit according to the manufacturer’s instructions (QIAGEN). Amplification of different segments of the ALB gene differing in size (300 bp, 484 bp, 600 bp, and 800 bp, respectively), served upfront to assess the extent of DNA fragmentation. Concomitant DNA and RNA extraction (AllPrep DNA/RNA Mini Kit; QIAGEN) was achieved from frozen tissue biopsies upon homogenization with stainless steel beads (3–7 mm mean diameter) using TissueLyser II (QIAGEN; 20” at 30 Hz, for two cycles) according to the manufacturer’s instructions. DNA/RNA concentration and integrity were assessed using Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific), NanoDrop 2000/2000c spectrophotometers, and Agilent 2100 Bioanalyzer system.

Total RNA was extracted from 22 lymphoma specimens, including 14 HGBCL-DH-BCL2(-BCL6) cases (AllPrep DMBackspaceNA/RN A Mini Kit; QIAGEN). RNA quality and concentration were determined using RNA 6000 NanoChip on 2100 Bioanalyzer (Agilent Technologies). Indexed polyadenylated RNA-seq libraries were prepared with 500 ng of high-quality (RNA integrity number: >6.5) total RNA input using a stranded mRNA ligation kit (Illumina) according to the manufacturer’s instructions. Single-end sequencing (1 × 75 bp reads, 36 million reads/sample) was used to sequence pooled libraries on a NextSeq 500/550 Illumina sequencer. Raw reads were aligned to the human genome (NCBI build 38, hg38) with STAR v2.7 with the parameter –quantMode GeneCounts to count gene reads. Ensemble IDs were changed into gene symbols using BioMArt. The Bioconductor R package DESeq2 was used to normalize gene count and to calculate the log₂ FC between comparisons. The Benjamini–Hochberg method was applied to correct P values for multiple comparisons. Significantly up/downregulated genes were identified using an adjusted P value < 0.05 and an absolute log₂ FC > 0.58. Heatmaps for specific gene signatures were generated with the pheatmap function of the pheatmap R package on DESeq2 normalized counts. IGH isotype expressed by lymphoma cells was inferred from bulk RNA-seq data according to the most expressed IGH constant region transcript. FPKM or log₂-normalized counts were used for comparisons of gene expression levels, as indicated. Unsupervised interrogation of RNA-seq data from MB2 DE GCB lymphomas for (i) DZ/LZ B cell–associated gene expression signatures (20), (ii) DH/DZ molecular signature [DHIT/DZsig, (21)], and (iii) a 370-gene DZ/LZ spatial signature https://pmc.ncbi.nlm.nih.gov/articles/PMC10984086/) was used to cluster lymphoma samples according to the COO. After z-score normalization, hierarchical clustering based on the gene signatures listed above was applied to identify clusters reproducing GC heterogeneity. Hierarchical clustering analysis was performed using the Ward.D2 method and considering the Euclidean distances. The Rand index measure was used to quantify the similarity of the two clustering results. Normalized gene expression profiles were subjected to the CIBERSORTx algorithm to estimate the immune cell subset composition of each lymphoma sample. The LM22 matrix was considered as a reference for immune cell deconvolution. Bulk-mode batch correction (B-mode) was applied for cell fraction estimation. The Wilcoxon–Mann–Whitney test was used to compare the CIBERSORTx fractions among conditions.

IGH IHC staining on 4-μmol/L-thick tissue sections from selected FFPE HGBCL-DH-BCL2 samples was combined with ROI definition and segmentation, and in situ analysis of 1825 curated cancer- and immune-associated transcripts using the GeoMx Digital Spatial Profiler platform (NanoString) according to the manufacturer’s instructions. mRNA-binding DNA probes (35–50 nt in size) conjugated with UV photocleavable indexing oligos were hybridized to the tissue as previously reported (92). The UV photocleavable probes were released from each ROI according to custom masks for UV illumination and digitally counted using the NanoString nCounter Analysis System. For nCounter data analysis, digital counts from barcodes corresponding to mRNA probes were normalized to internal spike-in controls. Moreover, a set of six internal housekeeping genes was included in the Human Immuno Oncology panel to control for system variation, including ROI size and cellularity (92, 93). Seven negative control probes were adopted to evaluate and filter ROIs with a high degree of nonspecific binding, as previously reported (48). Differential expression analyses between IGH⁺ and IGH^UND lymphoma ROIs were carried out by applying the moderated t test using the Limma package. The Benjamini–Hochberg method was applied to adjust P values and identify significant differentially expressed genes (adjusted P value < 0.05). The SpatialDecon algorithm (https://bioconductor.org/packages/SpatialDecon, RRID: SCR_026836) was used to estimate immunodeconvolution fractions (the safeTME profile matrix was considered as a reference). The Wilcoxon–Mann–Whitney test was used to compare the SpatialDecon fractions among conditions. The Fisher exact test was applied to assess the significance of the overlap between the gene sets.

WGS was performed on genomic DNA extracted from FL-HGBCL-DH-BCL2(-BCL6) metachronous cases. Isolated genomic DNA was quantified using Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific), and quality was assessed by agarose gel. Library preparation was performed using KAPA Hyper Prep Kit (Roche) as per the manufacturer’s recommendations. Briefly, gDNA was sheared to 500 bp using Covaris LE220-plus, adapters were ligated, and DNA fragments were amplified with minimal PCR cycles. Library quantity and quality were assessed using Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific), Tapestation High Sensitivity D1000 Assay (Agilent Technologies), and QuantStudio 5 System (Applied Biosystems). Illumina 8-nt dual indices were used. Equimolar pooling of libraries was performed based on quality control (QC) values and sequenced on an Illumina NovaSeq S4 (Illumina) with a read length configuration of 150 paired-end for 600 mol/L paired-end reads (300 mol/L in each direction) at Biodiversa s.r.l. (Treviso, Italy). Reads were aligned to reference genome (GRCh38) with the Burrows-Wheeler Aligner (BWA-MEM) algorithm [arXiv:1303.3997v2 (q-bio.GN)]. Duplicates were removed using the Picard MarkDuplicates tool. To identify IGHV rearrangements and translocations targeting IGH/K/L loci, IgCaller (94) was run on each sample using the bam aligned files after duplicate removal and the hg38 reference file. A custom script in R was run to identify IGV rearrangements with the same junction in metachronous samples. Genomic coordinates of the breakpoints captured using IgCaller were validated by PCR amplification followed by Sanger sequencing.

Genomic DNA was extracted from five IGH⁺ and seven IGH^UND lymphoma cases using either QIAamp DNA FFPE Tissue Kit (QIAGEN; FFPE tissues) or AllPrep DNA/RNA Mini Kit (QIAGEN; frozen tissues) according to the manufacturer’s instructions. Isolated genomic DNA was quantified using Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific), and quality was assessed by Tapestation genomic DNA Assay (Agilent Technologies).

Library preparation was performed using SureSelectXT Reagent Kit (Agilent Technologies) according to manufacturer’s recommendations. Exome capture was performed using SureSelect Human All Exon V7 Kit (Agilent Technologies). Library quality and quantity were assessed using Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific), Tapestation High Sensitivity D1000 Assay (Agilent Technologies), and QuantStudio 5 System (Applied Biosystems). Illumina 8-nt dual indices were used. Equimolar pooling of libraries was performed based on QC values and sequenced on an Illumina NovaSeq 6000 with a read length configuration of 150 paired-end for output required at Biodiversa. Read preprocessing was performed using the Genome Analysis Toolkit (GATK4) Best Practices Workflow. Reads were aligned to the reference genome (GRCh38) with the BWA-MEM algorithm (https://doi.org/10.48550/arXiv.1303.3997). Removal of duplicates was performed with the MarkDuplicates tool. Base quality score recalibration was achieved by applying the BaseRecalibrator tool with the list of 1,000G phase1 high-confidence SNPs as known polymorphic sites, followed by the ApplyBQSR tool, which recalibrated the base qualities of the input reads based on the recalibration table produced by the BaseRecalibrator tool. The HaplotypeCaller command of GATK4 was run to call SNVs and indels. Given the low number of samples, a hard-filtering approach was chosen to filter variants according to quality parameters (“QD < 0.2 || MQ < 40.0 || FS > 60.0 || SOR >3.0 || MQRankSum < −12.5 || ReadPosRankSum < −8.0” for SNPs and “QD < 0.2 || FS > 200.0 || SOR >3.0 || MQRankSum < −12.5||ReadPosRankSum < −20.0” for indels). Functional annotation of variants was performed with the FUNCtional annOTATOR (Funcotator) tool of GATK4 using the germline hg38 database (v1.4.20180829g) for gene annotation, variant classification, and DBSNP frequency. The SnpSift tool was applied for annotation of SNPs present in the Single Nucleotide Polymorphism database using as a reference file the All_20180418.vcf file downloaded from ftp://ftp.ncbi.nih.gov/snp/organisms/human_9606/VCF/.

The Annovar software was utilized to infer the pathogenic score predictions using the ljb26 database. Variant filtering was performed in R after the merging of Funcotator, SnpSift, and Annovar outputs. Variants with read depth <10 in all the samples or with allele frequency <0.05 in all samples were removed from the analysis. Variants annotated in dbSNPs with a minor allele frequency >5% in more than one population also were filtered out.

IGHV rearrangements were PCR-amplified from genomic DNA extracted from FFPE tissue sections or frozen biopsies of selected lymphoma cases, followed by next-generation sequencing on the Illumina MiSeq platform, using a modified version of the two-step PCR-based BIOMED-2 protocol (95). As first-step PCR amplification, a set of forward primers annealing to human IGHV genes was coupled with reverse primers annealing to human IGHJ segments. All primers were adapted for compatibility with the Illumina Indexing set (NextEra XT Index Kit – 96 indexes, Illumina; primer sequences are listed in Supplementary Table S7). First-round PCR products were purified using AMPure XP Beads (Beckman Coulter) before the second-round indexing PCR and library preparation steps. IGHV rearrangements were also profiled using the LymphoTrack Dx IGHV Leader Somatic Hypermutation Assay for Illumina MiSeq (Invivoscribe) according to the manufacturer’s instructions. The concentration and quality of libraries were determined using Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific) and the 2100 Bioanalyzer platform (Agilent), respectively. Libraries were sequenced using MiSeq Reagent Kit v3, 2.300bp (Illumina) or MiSeq Reagent Kit v2, 2.300bp (Illumina) (2 × 250 paired-end reads) on the MiSeq Illumina platform. Illumina PhiX spike-in was added to low-complexity libraries (Illumina). IGHV sequencing reads were analyzed using the Immcantation framework (https://immcantation.readthedocs.io/en/stable/). Assembly of the paired reads, quality filter, and primer mask were performed using the pRESTO toolkit (96). Each set of paired-end reads was first assembled into a full-length IGV sequence using the subcommand AssemblePairs.py of pRESTO. The FilterSeq.py command was then used to remove low-quality reads with a Phred quality score of less than 20, and the MaskPrimers.py was applied to identify and remove PCR primers annealing to V- and C-regions for both reads. The CollapseSeq.py command was used with the “-n 20 –inner –uf CPRIMER –cf VPRIMER –act set” parameters to identify unique sequences. Following duplicate removal, unique sequences with at least two representative reads were selected by using the command SplitSeq.py on the count field (-f DUPCOUNT) and a threshold of 2 (–num 2). Reads were subsequently aligned to the IMGT human IG V, D, and J reference sequences with IgBlast using the igblastn command line. The ParseDb.py command of Change-O (97) was run with the option “select” to subset heavy chain reads. Clonality was determined with the DefineClones.py command of Change-O that was run with the following parameters: “–act set –model ham –norm len –gf d_call –mode allele –maxmiss 6 –link average” and using an appropriate threshold determined with the distToNearest function of the SHazaM package that calculates the distance between each sequence in the data and its nearest neighbor. The latter function was also called with model = “ham” to use a single nucleotide Hamming distance matrix with gaps assigned zero distance and normalize = “len” to normalize the distance by the length of the sequence group. The union of ambiguous gene assignments was used to group all sequences with any overlapping gene calls.

IGV lineage tree and selection analyses were performed on selected IGH⁺ and IGH^UND cases. High-throughput sequencing of IGHV reads was preprocessed using pRESTO version 0.5.13 (96). The preprocessing included assembly of paired ends, quality filtering by trimming low-quality edges, filtering out reads with an average Phred score lower than 25, and masking bases with Phred scores lower than 20. Sequences with more than 10 masked or missing bases were discarded. FWR1-region and J-region primers (95) were masked (replaced by N’s) to preserve gene length. Next, identical sequences were collapsed, and only sequences with two copies or more were selected for analysis; the selected sequences were processed using Change-O version 0.4.6 (97) and in-house custom scripts. The processing included sequence annotation with IMGT/GENE-DB (98); (germline segment sequences were downloaded from IMGT on July 1, 2021), clonal assignment, and assessments of sampling depth and clonal size distributions. Putative germline sequences for each clone were created based on the same IMGT/GENE-DB database and clonal consensus in junction regions. The largest (dominant) clone identified in each repertoire was assumed to be the tumoral clone. Samples with polyclonal profiles were omitted from the analysis. Clones containing more than two unique sequences from the dominant, nondominant, or healthy control groups were analyzed using IgTreeZ (99, 100).

To exclude the pretransformation mutational history of each lymphoma clone from the analysis, trunks were removed from the IGH^UND and IGH⁺ trees, and the first split node of each “trunkless” tree was assigned to be the new root node. Trees that originally had no trunks were removed from the trunkless analysis to avoid bias due to trees not having enough information about their diversification history.

Visual representations of lineage tree shapes were created using the graph description language DOT, as implemented in Graphviz (https://graphviz.org/). Node (sequence) names were omitted for better tree visualization. Selection analysis was performed using the program BASELINe, as implemented in the SHazaM package (97, 101). The BASELINe algorithm was used to (i) calculate the expected mutation frequency, (ii) estimate the selection strength for each clonal lineage tree, and (iii) compare the selection scores of the repertoires in each group of clones. IGHV repertoires from PB mononuclear cells isolated from healthy donors (102) were used as controls.

Comparisons between lymphoma lineage tree characteristics with those of healthy repertoires, which included more than 50,000 trees, were performed based on patient average to overcome the bias of the control dataset being so much larger. Comparisons against clones from rLN, which included 599 trees from two patients, were done based on trees, as two samples are insufficient for statistical inference. For each comparison, the assumptions of normal data distribution and variance homogeneity were tested using the Shapiro test and the Levene test, respectively. If the data were normally distributed and had homogenous variances, the Student t test was used. Otherwise, the nonparametric Mann–Whitney U-test was used. To correct for multiple comparisons, the Benjamini–Hochberg’s FDR method was applied.

Genomic DNA was extracted using AllPrep DNA/RNA Mini Kit (QIAGEN) according to the manufacturer’s protocol. Human IGHM constant region gene sequence was obtained from GenBank (https://www.ncbi.nlm.nih.gov/genbank). Primers annealing to IGHM were designed using Primer3plus tool and are listed in Supplementary Table S7. Quantitative genomic RT-PCR reaction was performed using TB Green Premix Ex Taq TM II (Tli RNaseH Plus; Takara) following standard procedures on the Light Cycler 480 II (Roche Diagnostics). GCN was determined for HGBCL-DH-BCL2(-BCL6) samples after normalization for DNA input, testing a segment of CD3G as a reference, using the comparative Ct method (2^−ΔΔCt method). FACS-sorted tonsillar IGM⁺ human B cells from healthy donors were used as a reference.

Total RNA was extracted from human and mouse lymphoma and leukemia cell lines using RNeasy Mini Extraction Kit (QIAGEN) following the manufacturer’s instructions. Total RNA was reverse transcribed into cDNA using Superscript-III Reverse Transcriptase (Thermo Fisher Scientific) according to the manufacturer’s instructions. qRT-PCR reactions were performed using TB Green Premix Ex Taq TM II (Tli RNaseH Plus; Takara) on a Light Cycler 480 II (Roche Diagnostics). GAPDH gene served as control for cDNA input. Human and murine RAG1-, RAG2-, and GAPDH-specific primers are listed in Supplementary Table S7.

The 5’RACE protocol was adapted from Vazquez Bernat and colleagues (103). Briefly, total RNA was extracted using AllPrep DNA/RNA Mini Kit (QIAGEN), and first-strand cDNA synthesis reaction was performed using 200 ng of total RNA with 10 μmol/L oligo dT and Superscript II Reverse Transcriptase (Thermo Fisher Scientific) at 42°C for 1 hour. Template switching was performed by adding to the reaction mix a template switch oligonucleotide (Read1_TS Supplementary Table S7) composed of a five nucleotide poly-guanosine (poly-G) stretch, a 12 nucleotide unique molecular identified, and an Illumina’s universal amplification sequence (Read1). The template switch reaction was carried out at 42°C for 1 hour. Complementary DNA was purified using the Wizard SV Gel and PCR clean-up system (Promega) according to the manufacturer’s instructions. 2X KAPA HiFi Hot Start Ready Mix (Roche) was used to amplify the purified cDNA template through a 2-step semi-nested PCR using forward Read1-specific primers (Read1-U) and two sets of class-specific IGH and lGK/L constant region gene-specific reverse primers (with second-round primers, including Illumina Read2 sequence). Primer sequences are listed in Supplementary Table S7. A purification step was conducted after each PCR round using AMPure XP Bead-Based Reagent (Beckman Coulter Life Sciences) according to the manufacturer’s instructions using a 0.65× beads-to-sample ratio. 10 ng of purified PCR products were indexed using 2 × KAPA HiFi Hot Start Ready Mix, with forward (P5_R1) and reverse (P7_R2) indexing primers, in a 13-cycle index PCR reaction. Indexed libraries were purified using AMPure XP Beads, quality controlled using Agilent 2100 Bioanalyzer system, and sequenced (2 × 250 bp read length) on a MiSeq Illumina sequencer using MiSeq Reagent Kit v2, 2.300bp (Illumina).

Fastq files were merged with usearch with standard parameters (v.11.0.667, -fastq_minmergelen 15, -fastq_pctid 90). 3′ primers were removed with Cutadapt, and sequences not containing 3′ primers were discarded (v.3.4, –overlap 20, -e 0.05, –discard-untrimmed). Primer-removed sequences were collapsed with an in-house Python package; sequences observed only once were discarded. Fasta sequences were aligned with Igblast (v.1.21.0) against the IMGT database. Further analyses were performed with the Immcantation change-o package [v.1.3.0 (97)]. To avoid wrong V-gene assignment during clonotypes identification due to an intrinsic heterogeneity in RACE’s amplicon size, a Python-based clonotype assigner was used to integrate differences in V-gene assignment, weighting them through hamming distance calculation. Sequences with identical CDR3 length and D-J-gene assignment were grouped to perform CDR3 clustering analysis. CDR3 nucleotide distances were calculated among grouped sequences, and precomputed V-gene weights were added depending on the IgBlast V-gene assignment. Agglomerative clustering was performed to cluster the same clonotype sequences (sklearn package AgglomerativeClustering, metric = “precomputed”, n_clusters = None, linkage = “average”, distance_threshold = cutoff). A dynamic distance cutoff based on CDR3 length (cutoff = 0.15 if cdr3 length ≤30, cutoff = 0.09 if cdr3 length ≤51, cutoff = 0.08 if cdr3 length >51) was applied.

Genomic DNA was extracted from HeLa, PBMC, COH-DHL1, COH-THL1 IGK⁺, COH-THL1 IGL⁺, and COH-THL1 IGH^UND cell lines using AllPrep DNA/RNA Mini Kit (QIAGEN) following the manufacturer’s instructions. IGKV family–specific forward primers, together with an oligonucleotide annealing to the genomic region between IGKJ and IGKC segments, were combined with a KDE reverse primer to amplify KDE rearrangements (104). PCR reactions were carried out using GoTaq Flexi Polymerase Kit (Promega) using the SimpliAmp Thermal Cycler (Thermo Fisher Scientific–Applied Biosystem). Primers amplifying a segment of the IGKC gene were designed to assess biallelic KDE rearrangements by genomic PCR. Amplification of a segment of the RPLPO gene served to control DNA input. Primer sequences are listed in Supplementary Table S7.

IGHG1- and CD79B-targeting gRNAs (Supplementary Table S7) were cloned into lentiCRISPRv2-Puro lentiviral vector (Addgene #98290). Viral particles were produced transfecting lentiCRISPRv2-Puro containing gRNAs into HEK293T packaging cell line together with psPAX2 (gift from Dr Didier Trono, Addgene plasmid # 12260) and phCMV-GALV-MTR (Addgene #163612). Viral supernatants were collected at 48 and 96 hours after transfection and used to perform two consecutive rounds of spin infections (60 minutes at 1,800 rpm at 32°C) of COH-DHL1. Forty-eight hours after infection, cells were washed, and puromycin (0.2 μg/mL) was added to the culture medium. Infected cells were washed every 48 hours and refreshed with puromycin-containing complete culture medium for 7 to 10 days. After selection, puromycin-resistant cells were expanded in culture medium without puromycin. Cells were collected 2 to 3 weeks after antibiotic washout and analyzed by flow cytometry for CD79B expression or collected for genotyping of IGHG1 alleles and transcripts. IGHG1-targeted regions were PCR-amplified, and amplicons were purified (Wizard SV Gel and PCR Clean-Up System - Promega), quantified using Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific), and subjected to end-repair and A-tailing, followed by ligation with Illumina-compatible adapters using the Watchmaker DNA Library Prep (Watchmaker Genomics). Unligated adapters were removed using AMPure XP beads, and library amplification was carried out using the Twist Universal Adapter System compatible with TruSeq (Twist Bioscience) following the manufacturer’s recommendations. After cleanup and quantification, libraries were normalized to 4 nmol/L before pooling and sequencing on the Illumina MiSeq platform using MiSeq Reagent Kit v2 (2 × 251, Illumina).

Exponentially growing COH-DHL1 cells were diluted at 5 × 10⁵ cells/mL and treated for 8 or 24 hours with 5 or 10 nmol/L bortezomib (Cell Signaling Technology) resuspended in DMSO. Vehicle-only–treated cells acted as controls. Treated cells were collected and either analyzed by IC flow cytometry or immunoblotting analyses. Flow cytometric analysis was conducted after fixation and permeabilization of the cells (BD Cytofix/Cytoperm, BD Biosciences) and staining with a fluorescently labeled anti-IGG antibody (Jackson ImmunoResearch, Supplementary Table S7). Stained cells were acquired on a FACSCanto instrument, and data were analyzed using FlowJo software (BD Biosciences). For immunoblotting analyses, bortezomib/vehicle-treated cells were counted, and equal number of cells were lysed with urea buffer (8 mol/L urea, 50 mmol/L TRIS 7.5, and 150 mmol/L NaCl, containing protease inhibitor cocktail). Protein extracts corresponding to 5 × 10⁵ cells were subjected to PAGE, transferred onto polyvinylidene difluoride membranes, and incubated with anti-human IGG (Abcam, Supplementary Table S7) or anti-human CD79B (Cell Signaling Technology, Supplementary Table S7) antibodies. Proteins were visualized staining the membrane with secondary antibodies conjugated to horseradish peroxidase.

For IG-light chain complementation of COH-DHL1 cells, we randomly selected three IGKV and two IGLV rearrangements from a library of antibodies cloned from single SARS-CoV2 RBD-specific circulating CD27⁺ memory B cells of Covid19 convalescent patients (to be described elsewhere). Full-length IGK/L V gene sequences (listed in Supplementary Table S7) were cloned into AbVec1.1-IGKC (Addgene #80796) and AbVec1.1-IGLC2 (Addgene #99575) expression vectors (105), kindly provided by Dr. Hedda Wardemann through Dr. Karlsson Hedestam. IG expression vectors were electroporated into COH-DHL1 cells using the NEPA electroporator (NEPA21, Nepa Gene; poring pulse: 200V, 5 ms length, 50 ms interval, D. rate 10%, polarity +, 2 pulses; transfer pulse: 20V, 50 ms length, 50 ms interval, D. rate 40%, and polarity ±, 5 pulses) in serum-free Opti-MEM medium. Electroporated cells were collected at 8 and 24 hours following transfections and analyzed for IC and surface IGG and IGK/L expression by flow cytometry. Complemented cells were stained with anti-IGG and IGK/L-specific antibodies (Supplementary Table S7) both on the surface and intracellularly. Stained cells were acquired on a FACSCanto instrument and data analyzed using FlowJo software (BD Biosciences). IGG mean fluorescence intensity values were measured in control and complemented cells, gating on total live and sIGK/L-expressing cells, respectively.

Full-length IGH and IGK/L chain V gene rearrangements recovered from the sequencing of 5’RACE amplicons of three IGH^UND HGBCL-DH-BCL2(-BCL6) were cloned upstream of the human IGG1, IGKC or IGLC2 constant region genes into the AbVec2.0-IGHG1 (Addgene #80795; gift from Dr. Hedda Wardemann), AbVec1.1-IGKC, and AbVec1.1-IGLC2 expression vectors. Expression vectors were transiently transfected into HEK293T cells. Cells were collected at 8 and 24 hours after transfection and analyzed by IC flow cytometry. Transfected cells were fixed, permeabilized (BD Cytofix/Cytoperm, BD Biosciences), and stained intracellularly at RT in the dark for 20’ with fluorescently labeled anti-IGG and IGK/L antibodies (Supplementary Table S7). Stained cells were acquired on a FACSCanto instrument, and data were analyzed using the FlowJo software (BD Biosciences).

Human HT (DLBCL), Burkitt lymphoma RAMOS, and HGBCL-DH-BCL2(-BCL6) DoGKIT, COH-THL1-IGH⁺, COH-THL1-IGH^UND, COH-DHL1, WILL-3, WILL-2, and SC-1 cell lines were treated with PV or polatuzumab (MedChemExpress). PV was dissolved in PBS and stored in aliquots at −80°C. Cells were seeded in a 6-well plate at a density of 5 × 10⁵ cells/mL and treated with PV or polatuzumab at a concentration of 1.25 μg/mL. All experiments were performed at least in triplicate. Vehicle-treated controls were included in each experiment. Cell viability was assessed 72 hours after treatment using LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (Thermo Fisher Scientific).

HGBCL-DH-BCL2(-BCL6) DoGKIT, COH-DHL1, WILL-3, WILL-2, and SC-1 cell lines, as well as NU-DUL-1, TMD8, SU-DHL-4, and RAMOS acting as positive controls, were seeded at a density of 8 × 10⁴ cells/mL (2,000 cells/well in 25 μL), in a white, flat-bottom 384-well microplate (Greiner). The next day, cells were treated with 25 μL of serial dilutions of inhibitors targeting BTK (ibrutinib, 50 μmol/L – 5 nmol/L), SYK (R406, 20 μmol/L – 250 nmol/L), and PI3Kγ/δ (duvelisib 50 μmol/L – 80 nmol/L) in culture medium and incubated for 48 hours. At the end of the treatment, 50 μL CellTiter-Glo 2.0 Cell Viability Assay substrate (CTG; Promega) diluted 1:1 in PBS was added to the cells, followed by 15 minutes of incubation. Luminescence signal was measured using the Infinite F200 plate reader (Tecan). All treatments were performed in triplicate in at least three experiments.

HGBCL-DH-BCL2(-BCL6) DoGKIT, COH-THL1-IGH⁺, COH-THL1-IGH^UND, COH-DHL1, WILL-3, WILL-2, and SC-1 were seeded at a density of 2.5 × 10⁵ cells/mL and treated for 48 hours with PI3Kγ/δ (duvelisib; 10 μmol/L, in DMSO) or FOXO1 (AS1842856; 100 nmol/L, in DMSO) inhibitors. Vehicle-treated cells acted as controls. After the treatment, cells were collected and total RNA was extracted (RNAeasy, QIAGEN) according to the manufacturer’s instructions, for downstream gene expression analyses. Experiments were reproduced at least three times.

A selected panel of inhibitors was prepared by performing for each compound a ten-point dose–response curve in DMSO (VWR Chemicals, 23500.260) with 1:4 step-dilution starting from 10 mmol/L in V-bottom 384-well polypropylene plates (Thermo Fisher Scientific, 4312). Compound plates were stored at 4°C and diluted 1:25 in RPMI medium before adding to cell plates. COH-DHL1 and COH-THL1 cells were seeded (3,500 and 2,500 cells/well in 35 μL, respectively) in flat-bottom sterile white 384-well plates (Greiner, 781080). The next day, cells were treated with 5 μL of drug serial dilutions in RPMI medium and incubated for 48 hours. In all test plates, 0.5% DMSO and 100 nmol/L panobinostat (SelleckChemical) were used as negative and positive controls, respectively (100% and 0% cell vitality). A dose–response curve for panobinostat (final concentration 2.5 μmol/L to 0.01 nmol/L) was also included as internal compound reference. At the end of the treatment, 40 μL of CellTiter-Glo 2.0 reagent (CTG; Promega, G9243) diluted 1:1 in water was added to the cells, followed by incubation for 15 minutes. Luminescence signal was measured using a Tecan Infinite F200 plate reader. All treatments were performed in triplicate. The steps of cell seeding, treatment, and CTG addition were performed with a Hamilton MicroLab Star M liquid handler. Plate-wise data were preprocessed to correct for potential systematic spatial effects using a surface determined by regression through maximum likelihood estimation to interpolate the negative controls sample data. Data from all plates that passed the QC (Z-factors >0.5 between negative and positive controls) were subjected to intraplate normalization to calculate viability percentages (where 100% was defined according to negative controls and 0% was defined according to positive controls). IC₅₀ values were calculated from logarithmic plots of normalized data in MatLab (MathWorks).

BCR-positive and BCR-negative λ-MYC;B1-8f lymphoma cells were isolated from two independent primary tumors (# 2646 and # 2567) from 12- to 20-week-old male and female λ-MYC;B1-8f mice on a C57BL/6J × BALB/c mixed genetic background, as previously described (5). 8- to 12-week-old male immunocompetent syngeneic CB6F1/J mice were used as recipients for tumor transfer experiments. All animals were housed following institutional and national guidelines and regulations. Animal procedures and studies were performed in compliance with Italian national and European Union (EU) directives (2010/63/EU) for animal research with protocols approved by the local Ethics Committee and the Italian Ministry of Health (# 561/2021-PR).

BCR-negative murine λ-MYC;B1-8f lymphoma cells were in vitro-derived from their BCR⁺ counterparts by TAT-Cre transduction starting from two independent primary tumors (# 2646 and # 2567), as previously described (5). Briefly, BCR⁺ lymphoma B cells were transduced with TAT-Cre (37.5 μg/mL) protein in serum-free media (Hyclone) at 5 × 10⁶ cells/mL for 45 minutes, at 37°C. Transduced cells were washed three times and cultured using complete B-cell medium (high-glucose DMEM supplemented with 10% heat-inactivated FBS, 0.1 mmol/L nonessential amino acids, 1 mmol/L sodium pyruvate, 50 μmol/L β-mercaptoethanol, and 2 mmol/L L-glutamine). BCR⁻ lymphoma cells were isolated by negative selection through MACS using biotin-labeled anti-mouse IgM (cl. R33.24.12; gift from Dr. K. Rajewsky), followed by anti-biotin microbeads (Miltenyi Biotech), or by cell sorting using the FACS Aria instrument (BD Pharmingen) upon staining with a monovalent fluorescently labeled anti-mouse IgM (Fab, Jackson Immunoresearch). A total of 1.5 million purified BCR⁺ or BCR⁻ lymphoma cells were transferred intravenously into 8- to 12-week-old male immunocompetent syngeneic CB6F1/J recipients. Transplanted animals were sacrificed 14 (BCR⁺) or 21 (BCR⁻) days after transfer. Flow cytometry analysis was used to assess the infiltration of IgM⁺ or IgM^- lymphoma B cells (FSC^hi, CD19⁺, and MYC⁺) in primary and secondary lymphoid organs. FSC^hi/CD19⁺ BCR⁺ or BCR-less lymphomas were FACS-purified using a monovalent (Fab) anti-IgM antibody to prevent antigen receptor crosslinking. DNA from BCR⁺ and BCR⁻ex-vivo purified cells was extracted using genomic DNeasy Tissue Kit (QIAGEN). Genomic PCR amplification of the B1-8f allele (present in BCR⁺ while lacking in BCR⁻ derivatives) was used to BCR genotype lymphomas cells. Mouse VDJ profiling was performed as previously described (106). Briefly, DNA amplification was carried out in two rounds of nested PCR approach using GoTaq G2 DNA Polymerase (Promega). Two degenerate forward primers annealing to framework region-1 of IGHV genes and a degenerate reverse primer annealing to IGHJ segments were used for the first-round PCR amplification. In the second PCR reaction, 1 μL of the first-round amplicon was reamplified with forward primers annealing to IGHV family–specific primers (J558, GAM3, 36-60, S107, 71835, X-24, and Q52) in combination with an IGHJ4-specific reverse primer. PCR amplicons were cloned into pGEM-Teasy vector, transformed in DH5α bacteria, and plasmid DNA subjected to Sanger sequencing. IGHV analysis was performed using IGBLAST.

No statistical methods were used to predetermine the sample size. GraphPad Prism software v.10.1.1 was used to graphically represent data and perform statistical analyses. Unpaired Student t test, nonparametric Mann–Whitney test, Fisher test, and χ² test were used to analyze data, as indicated. The results section, figures, and figure legends report information on sample size, mean and median values, and the statistical test used. For all analyses, P values < 0.05 were considered significant.

Further information and requests for resources, reagents, or materials is available from the lead contact upon request.

Data generated as part of this study have been deposited with restricted access due to patient privacy requirements in the European Genome–phenome Archive (EGA, EGAS50000001047) which is hosted by the European Bioinformatics Institute (EBI) and the Centre for Genomic Regulation (CRG). This study does not report original code. Other data supporting this study and scripts to reproduce the analyses are available from the corresponding author upon reasonable request.

R. Siebert reports grants from the German Ministry of Science and Education (BMBF) during the conduct of the study and other support from AstraZeneca and Rhythm outside the submitted work. G. Pruneri reports grants from Novartis and personal fees from Lilly, Illumina, and Thermo Fisher Scientific outside the submitted work. S. Hohaus reports personal fees from Roche, BeiGene, Takeda, Incyte, Gilead, NovartisLilly, Abbvie, Bristol Myers Squibb outside the submitted work. No disclosures were reported by the other authors.

G. Varano: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. S. Lonardi: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft. P. Sindaco: Data curation, formal analysis, validation, investigation, visualization, methodology. I. Pietrini: Formal analysis, validation, investigation, visualization, methodology. G. Morello: Formal analysis, validation, investigation, visualization, methodology. P. Balzarini: Formal analysis, validation, investigation, visualization, methodology. F. Vit: Data curation, software, formal analysis, investigation. H. Neuman: Data curation, software, formal analysis, investigation, visualization. G. Bertolazzi: Data curation, software, formal analysis. S. Brambillasca: Formal analysis, investigation. N.C. Parr: Formal analysis, investigation. M. Chiarini: Formal analysis, investigation. S. Bellesi: Formal analysis, investigation. E. Maiolo: Formal analysis, investigation. S. Giampaolo: Formal analysis, investigation. F. Mainoldi: Formal analysis, investigation. V. Selvarasa: Investigation. H. Arima: Investigation. V. Pellegrini: Investigation. C. Pagani: Investigation. M. Bugatti: Investigation, methodology. C. Ranise: Investigation. T.M. Taddei: Investigation. T. Sonoki: Resources. H. Thanasi: Resources. E. Morlacchi: Investigation. D. Segura-Garzon: Investigation. E. Albertini: Investigation. R. Daffini: Investigation. A. Sivacegaram: Investigation. H. Yang: Supervision. Y. Li: Data curation, formal analysis, validation, investigation. V. Cancila: Formal analysis, investigation. G. Cicio: Formal analysis. M. Robusto: Investigation. B. Leuzzi: Investigation. A. Andronache: Investigation. P. Trifiro: Investigation. M. Riboni: Investigation. S.P. Minardi: Supervision. R.J.P. Bonnal: Formal analysis. C.L. Gonzalez: Visualization. E. Visco: Investigation. P. Capaccio: Resources. S. Torretta: Resources. L. Pignataro: Resources. C. Almici: Resources. M. Varasi: Supervision. L.M. Larocca: Resources. R. Siebert: Supervision. B. Falini: Resources. A.J.M. Ferreri: Resources. A. Tucci: Resources. D. Lorenzini: Resources, formal analysis. A.D. Cabras: Resources, formal analysis. G. Pruneri: Resources, formal analysis. A. Di Napoli: Resources, formal analysis. M. Ungari: Resources. M. Pizzi: Resources, formal analysis. S. Hohaus: Supervision. C. Mercurio: Formal analysis, supervision. J.Y. Song: Resources, formal analysis. W.C. Chan: Resources, formal analysis, validation. L. Lorenzi: Formal analysis. R. Bomben: Formal analysis, supervision, investigation. M. Ponzoni: Formal analysis, supervision, investigation. R. Mehr: Formal analysis, supervision, investigation, writing–original draft. C. Tripodo: Formal analysis, supervision, investigation, visualization, methodology, writing–original draft. F. Facchetti: Resources, formal analysis, supervision, validation, visualization, methodology, writing–original draft, project administration. S. Casola: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This work was supported by grants from the Italian Association for Cancer Research (AIRC; IG #23747 to S. Casola and #22145 to C. Tripodo), the Italian Ministry of Education, University and Research (PRIN #20175L9H7H to F. Facchetti, M. Ponzoni, M. Pizzi, and PRIN #2017K7FSYB to C. Tripodo), the Cariplo Foundation (CoronAId to S. Casola), Fondazione Spedali Civili di Brescia (CoronAId iperimmuni to S. Casola and C. Almici), and the United States–Israel Binational Science Foundation (grant 2013432 to R. Mehr). S. Hohaus acknowledges research support by PNRR-MAD-2022-12376707. Through S. Casola, this work has been supported by the project SCALE UP – Department of Excellence 2023–2027, funded by the Italian Ministry of University and Research to the Department of Molecular Biotechnology and Translational Medicine, University of Milan. S. Casola acknowledges funding from the SMART-FL project by the NRRP Next Generation EU – Mission 4, Component 2, Investment 1.4 – Project CN00000041 “National Center for Gene Therapy and Drugs based on RNA Technology” - CUP B83C22002870006. P. Sindaco was supported by post-doctoral fellowships sponsored by the Italian Association for Cancer Research (AIRC) AND PRIN #20175L9H7H. S. Lonardi and L. Lorenzi were supported by Fondazione Beretta per la Ricerca sul Cancro. G. Varano received support from an American Society of Hematology Global Research Award (AGRA2023-4) and the Marie Skłodowska-Curie postdoctoral training program (H2020-MSCA-IF-2019 #895887). S. Giampaolo is supported by a Marie Skłodowska-Curie postdoctoral fellowship (H2022-MSCA-IF-2022-PF-01 # 101111183) and the CARIPLO Foundation (Progetto Giovani Ricercatori). D. Segura-Garzon received support from the Marie Skłodowska-Curie Innovative Training Network (H2020-MSCA-ITN - 765158-COSMIC). H. Neuman was supported by a Bar-Ilan University President’s scholarship. We wish to thank Drs. D.W. Scott and L.K. Hilton for sharing results and for discussions. We thank Drs. S. Cenci and M. Resnati for sharing the bortezomib reagent and protocols. We acknowledge the support of the IFOM Imaging (M.G. Totaro and D. Parazzoli) and Cellular & Preclinical Models (I. Rancati, S. Lavore, and G. Ossolengo) technological development units and the COGENTECH Histopathology (F. Pisati) and DNA Sanger sequencing (S. Volorio, S. Fortuzzi, D. Sardella) units for support in flow cytometry, cell sorting, cell line biobanking, molecular profiling, and histopathology. Thanks to Dr. A. Tironi, Dr.ssa R. Rinaldi, Dr.ssa R. Marchione, Dr. D. Ghisulfino, and Dr. V. Agosti for contributing biopsy specimens and S. Zini, A. Valzelli, M. Tomaselli, P. Bossini, A. de Zorzi, L. Fappani, L. Fontana, T. Gulotta, S. Rosola, M. Benedetti, F. Filippini, and L. Zavaglio for technical and administrative support.