Selective Enhancer Dependencies in MYC-Intact and MYC-Rearranged Germinal Center B-cell Diffuse Large B-cell Lymphoma Free

Diffuse large B-cell lymphoma (DLBCL) is a clinically heterogeneous cancer for which distinctive biological subgroups have been defined (1–3). The most aggressive subgroup, known as “dark zone signature” (DZsig+; refs. 4, 5) or “molecular high-grade” (6), is characterized by a germinal center (GC) B-cell phenotype and high expression of MYC and MYC target genes. The best-characterized mechanisms of MYC hyperactivity in DLBCL are long-distance rearrangements that juxtapose MYC to well-studied Ig locus enhancers (7–11), driving MYC expression above physiologic levels (12). MYC rearrangements to non-Ig loci are assumed to activate MYC by a similar “enhancer-hijacking” mechanism, but our functional understanding of these events is limited. Even less is known about how MYC is activated in the roughly half of DZsig/molecular high-grade–DLBCL that lack either detectable MYC locus rearrangements by FISH (5, 13) or cryptic insertions of heterologous enhancers into the MYC locus as identified by whole-genome sequencing (WGS) in a small proportion of additional cases (13). To address this, we performed high-throughput CRISPR-interference (CRISPRi) screens targeting candidate enhancers in the MYC locus and rearrangement partner (RP) loci, revealing regulatory principles of MYC non-Ig rearrangements and showing that a previously uncharacterized enhancer active in normal GC B cells serves as a common dependency of MYC-intact GC B-cell DLBCL (GCB-DLBCL).

We identified mature human BCL cell lines with MYC locus rearrangements to specific partner loci (14–16) or with no evidence of MYC locus aberrations (Supplementary Table S1). Consensus positions of Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) peaks across five mantle cell lymphoma (MCL) and 18 DLBCL cell lines were used to define enhancer nucleosome-free regions (NFR) genome-wide. ATAC-seq and H3K27ac chromatin immunoprecipitation sequencing (ChIP-seq) signals were then used to assign activity scores to each NFR-centered enhancer module (17), and NFRs with low activity scores in all cell lines were removed from the set of candidate enhancers. We next designed a custom single-guide RNA (sgRNA) library (“NFR CRISPRi library,” Fig. 1A; Supplementary Table S2) targeting candidate enhancer NFRs in the MYC locus and in MYC RP loci identified in five cell lines (one MCL: JeKo-1 and four GCB-DLBCL: OCI-Ly1, SU-DHL-4, WSU-DLCL2, and DB) that we had validated as amenable to high-throughput CRISPRi profiling. This library also included CRISPRi-optimized promoter-targeting sgRNAs (18) for transcription factor (TF) genes expressed in B cells, along with positive and negative essentiality controls and extended coverage of the MYC and BCL6 promoter regions.

Because our NFR library targets only the functional core of B-cell enhancers, it is significantly smaller than the sgRNA library previously used for MYC enhancer CRISPRi screens in myeloid leukemia (19) and multiple myeloma (20) cell lines (“tiling CRISRPi library,” which covers broad continuous regions of the MYC locus) despite the fact that the NFR library targets several additional loci. To ensure similar performance of the NFRs and tiling strategies for enhancers within the MYC locus, we selected mature BCL cell lines without MYC rearrangements for initial CRISPRi profiling, using both libraries. One NFR and two tiling CRISPRi screens in MCL cell lines confirmed dependency on previously identified 5′ Notch/RBPJ-activated MYC enhancers (Fig. 1B, Supplementary Tables S2 and S3; refs. 21, 22). In contrast, tiling- and NFR-focused CRISPRi screens in the GCB-DLBCL cell line Karpas-422 identified essential enhancers in the 3′ region of the MYC locus. None of our tiling screens showed evidence for essential enhancers in the “Blood ENhancer Cluster” that lies at the far 3′ end of the MYC topologically associated domain (TAD) and is essential for hematopoietic stem cells, myeloid leukemia, B-cell progenitors, and some B-lymphoblastic leukemias (23–25) but lacks acetylation in mature B cells and lymphomas (Supplementary Fig. S1A).

We performed CRISPRi screens with the NFR-focused library in two MCL cell lines (one with MYC rearrangement) and six GCB-DLBCL cell lines (four with MYC rearrangement). Essential gene promoter–targeting sgRNAs showed significant depletion in all eight screens (Supplementary Fig. S1B). We used the Model-based Analysis of Genome-wide CRISPR/Cas9 Knockout robust ranking aggregation (MAGeCK RRA) algorithm to score the significance and log₂ fold change of all NFRs across all eight screens. In all five MYC-rearranged cell lines, the most essential NFRs were within the corresponding MYC RP loci, whereas MCL and DLBCL cell lines lacking MYC rearrangements were most dependent on the previously mentioned 5′ enhancers and previously uncharacterized 3′ MYC locus enhancers, respectively (Fig. 1C and D; Supplementary Table S2).

The MCL cell line JeKo-1 and the GCB-DLBCL cell line OCI-Ly1 showed intrachromosomal rearrangements between the MYC locus and regions of chromosome 8 outside of the MYC TAD. In JeKo-1, a chromosomal fusion linked the 3′ side of MYC to 8p23.1, whereas input chromatin indicated a copy-number gain of the fused regions (Fig. 2A). CRISPRi revealed strong dependency on a single enhancer within this partner region, NFR 108, that showed no fitness effect in any other cell line. Other strongly acetylated elements in the partner locus such as the adjacent NFR 107 showed minimal fitness effect. The GCB-DLBCL cell line OCI-Ly1 bears a more complex MYC rearrangement involving two regions of chromosome 8q outside the MYC TAD, with subsequent chromosomal amplification (Fig. 2B). One tightly packed cluster of four NFRs (NFRs 138–141) was essential in the first partner locus, with weaker and more diffuse essentiality effects seen across several elements in the other.

Low-throughput CRISPRi studies confirmed that repression of NFR 108 lead to decreased MYC transcript levels in JeKo-1, but not OCI-Ly1 (Fig. 2C; Supplementary Fig. S2A), confirming NFR 108 as an enhancer-hijacking driver of MYC expression in JeKo-1. However, the equivalent assay in OCI-Ly1 with single sgRNAs targeting either of the most essential NFRs (NFR 139 and NFR 213) in each OCI-Ly1 RP region failed to repress MYC levels, suggesting that these multimodular enhancers show functional redundancy (26, 27) in this short-term validation assay. Indeed, we found that flow-sorted OCI-Ly1 populations expressing sgRNAs targeted to both NFRs simultaneously (Supplementary Fig. S2B) gave substantially stronger MYC repression upon KRAB-dCas9 induction than repression of either enhancer alone (Fig. 2D).

Because oncogenic MYC activation in JeKo-1 seemed to be selectively driven by a single enhancer (NFR 108) that lacks an essential role in other cell lines, we wondered whether JeKo-1 might show increased dependency on trans-regulators of that enhancer. Because MCL- and Epstein–Barr virus–transformed lymphoblastoid cell lines share a non-GC mature B-cell state, we reviewed ENCODE ChIP-seq data from the lymphoblastoid cell line GM12878 and found that EBF1 was among the factors binding most strongly to NFR 108 (Fig. 2E; Supplementary Fig. S2C). NFR 108 contains a centrally located EBF1-binding motif and also shows evidence for EBF1 binding in ChIP-seq data from the MCL cell line CCMCL1 (Supplementary Fig. S2D; ref. 28). Whereas EBF1 is a key component of the mature B-cell gene regulatory network and EBF1-targeting sgRNAs were depleted in all cell lines, they showed the strongest depletion in JeKo-1 (Fig. 2F). The other screened MCL cell line, SP-49, showed unique dependency on RBPJ, whose product is required for activation of the Notch-dependent 5′ MYC enhancers active in that line (21). Indeed, knockdown of EBF1 lead to reproducibly decreased MYC transcripts in JeKo-1 but not SP-49, whereas RBPJ knockdown showed the opposite pattern (Fig. 2G). These findings suggest that MYC rearrangements to specific partners may shape global trans-factor dependencies.

The t(3;8) (q27;q24) rearrangement that fuses the MYC locus to that of BCL6, another key lymphoma oncogene, is recurrent in GCB-DLBCL (29). A well-characterized, evolutionarily conserved, and GC-specific distal super-enhancer cluster known as the BCL6 locus control region (BCL6-LCR) lies 153 kbp upstream of the BCL6 gene (15, 30–32). Prior topological evidence indicated that this rearrangement activates MYC via enhancer-hijacking of BCL6-LCR elements (15, 16), but the specific elements responsible for MYC activation have not been identified. We performed CRISPRi screens on three lymphoma cell lines, WSU-DLCL2, DB, and SU-DHL-4, that bear structurally distinct variants of the MYC::BCL6-LCR rearrangement. We used previously identified genomic breakpoints (14), 4C-seq, and optical mapping data to generate parsimonious models for the structure of these rearrangements (Fig. 3A; Supplementary Fig. S3A and S3B), which vary from a simple fusion in WSU-DLCL2 to a complex event including multiple copies of the BCL6-LCR and additional segments of chromosomes 11 and 1 in SU-DHL-4. Topological interactions between the MYC promoter and BCL6-LCR were identified in all three cell lines [Supplementary Fig. S3C and previously shown for WSU-DLCL2 (16)]. All partner loci were included in our NFR-targeting CRISPRi library, thus providing an opportunity to evaluate how complexity affects the function of a recurrent enhancer-hijacking rearrangement.

CRISPRi profiling results in all three MYC::BCL6-LCR+ cell lines were remarkably similar (Figs. 1D, 3B and C). The most essential enhancer in all three cell lines was NFR 31 in the BCL6-LCR, with other BCL6-LCR enhancers also showing significant effects. No strongly essential enhancer was identified in the MYC locus in these cell lines, nor in the additional partner loci in SU-DHL-4 (Fig. 1D; Supplementary Fig. S3D). Essentiality effects in all three cell lines were dominated by NFRs selectively bound by factors implicated in enhancer-dependent BCL6 activation (15, 31), including the ternary complex members MEF2B, OCT-2, and OCA-B and the acetyltransferase p300 (Fig. 3B and C; Supplementary Fig. S3E). BCL6 is also an essential gene in many GCB-DLBCL cell lines, as confirmed by depletion of sgRNAs across the BCL6 promoter region (Supplementary Fig. S3F), but the fitness effects of sgRNAs targeting the most critical BCL6-LCR NFRs were stronger in all three MYC::BCL6-LCR+ GCB-DLBCL cell lines than in the other three GCB-DLBCL cell lines. The heightened dependency on BCL6-LCR elements in MYC::BCL6-LCR+ DLBCL was unlikely to be due to differences in TF binding, as ChIP-seq binding patterns were similar in MYC::BCL6-LCR+ DLBCL and DLBCL without MYC locus rearrangement (Supplementary Fig. S3E).

Assays with single sgRNAs targeting essential BCL6-LCR NFRs 31 or 41 showed only modest effects on MYC expression in MYC::BCL6-LCR+ GCB-DLBCL cell lines (Supplementary Fig. S3G), again suggesting short-term compensatory effects of the multimodular enhancer, but the combination of sgRNAs targeting both NFR 31 and 41 simultaneously resulted in significantly decreased cell growth (Fig. 3D) and decreased expression of both MYC and BCL6 (Fig. 3E; Supplementary Fig. S3H) in both WSU-DLCL2 and SU-DHL-4. CRISPRi knockdown of MEF2B, POU2F2 (OCT2), and POU2AF1 (OCA-B) encoding BCL6-LCR subunit–activating factors (31) led to similar decreases in both MYC and BCL6 expression in MYC::BCL6-LCR+ cell lines (Fig. 3F). Using a previously published model of de novo BCL6 distal enhancer activation (15), we found that CRISPRi repression of NFR 31 and/or NFR 41 were able to block the increase in BCL6 transcript levels caused by MEF2B overexpression in a non-GC lymphoma cell line (Supplementary Fig. S3I), confirming that these distal enhancers mediate MEF2B-dependent BCL6 activation in nonrearranged B cells. These results show that the t(3;8) (q27;q24) rearrangement and variants result in lymphomas for which MYC expression and ongoing cell growth are dependent on cis- and trans-regulatory mechanisms that normally regulate GC-specific expression of BCL6.

CRISPRi screens in non–MYC-rearranged GCB-DLBCL cell lines Karpas-422 and SU-DHL-5 both identified NFR 195, a site 437 kb downstream of the MYC promoter, as the most essential distal element, with sgRNAs targeting the adjacent NFR 196 showing lesser effects (Fig. 4A and B). Two other 3′ enhancers, NFR 181 and NFR 202, showed weaker essentiality signal in Karpas-422 CRISPRi screens, and all three essential enhancers showed increased topological interactions with the MYC promoter by 4C-seq (Supplementary Fig. S4A). NFR 195 was strongly acetylated in two other non–MYC-rearranged GCB-DLBCL cell lines, HT and BJAB, and in five of six primary GCB-DLBCL biopsies without known MYC rearrangements (Fig. 4A and B; Supplementary Fig. S4B and S4C). In contrast, this region lacked acetylation in all MYC-rearranged GCB-DLBCL cell lines and patient biopsies, analogous to the lack of acetylation seen at the 5′ enhancers in MYC-rearranged MCL (21). CRISPRi assays confirmed that NFR 195 repression reduced fitness (Fig. 4C) and lowered MYC transcript (Fig. 4D) and protein (Supplementary Fig. S4D) levels in four non–MYC-rearranged GCB-DLBCL cell lines but not in MYC-rearranged cell lines (Fig. 4C and D; Supplementary Figs S3G, S4D and S4E).

H3K27ac ChIP-seq data from sorted normal human B-lineage subpopulations (33) and B-lineage pseudobulk single-cell ATAC-seq data from normal human tonsils (34) showed selective acetylation and accessibility at NFR 195/196 in normal GC B cells (Fig. 4B and E; Supplementary Fig S4B, S4C, and S4F). Sequence elements in NFR 195 show strong evolutionary conservation, including in mice, and the orthologous mouse element also shows GC B cell–specific chromatin accessibility (Immgen project; Fig. 4F and Supplementary Fig. S4G; ref. 35). Furthermore, analysis of data from a recent high-resolution single-cell ATAC-seq and RNA-seq atlas of human tonsil cells (36) showed that chromatin accessibility of NFR 195 closely parallels the known expression pattern of MYC in GC B-cell subpopulations (Fig. 4G). NFR 195 shows modest accessibility in most GC DZ B cells, in which MYC expression is minimal, and the greatest accessibility in rare GC B-cell populations with transcriptional features indicative of commitment to reentry from the light zone (LZ) to the DZ, the stage at which MYC transcript levels are sharply increased (37, 38). Together, these findings indicate that NFR 195 contains a physiologic B-cell MYC enhancer, hereafter ‘GC B-cell MYC enhancer 1′ (GME-1), whose function promotes the fitness of many non–MYC-rearranged GCB-DLBCLs.

Examination of ATAC-seq and sequence conservation data indicated four potential TF-binding subunits within GME-1, NFR 195A/B and 196A/B, which were further supported by ChIP-seq binding patterns for the ternary complex factors MEF2B, OCT2, OCA-B, and acetyltransferase p300 in Karpas-422 (Fig. 5A). A published biotin–ChIP-seq dataset (39) also showed OCT2 binding to GME-1 in the BJAB cell line (Fig. 5A). Ternary factor binding was largely absent from the GME-1 subunits in MYC::BCL6-LCR+ cell lines (Supplementary Fig. S5A and S5B). Knockdown of POU2F2, POU2AF1, and (less consistently) MEF2B by CRISPRi led to decreased MYC transcript levels in GME-1–dependent cell lines (Fig. 5B; Supplementary Fig. S5C), and CRISPRi screen data confirmed that POU2F2, POU2AF1, and MEF2B were among the most essential TF genes in both MYC-intact and MYC::BCL6-LCR+ GCB-DLBCL cell lines (Supplementary Fig. S5D), suggesting that these factors contribute to MYC activation in both groups of cell lines via different regulatory elements. We used luciferase reporter assays to compare the transcription-activating function of GME-1 subunits and two ternary factor–regulated BCL6-LCR subunits (NFRs 31A and 41A) in MYC-intact and MYC::BCL6-LCR+ cell lines (Fig. 5C; Supplementary Fig. S5B, S5E–S5H). GME-1 subunit 195B showed strong activity in MYC-intact cell lines, comparable with that of the BCL6 enhancers, whereas MYC::BCL6-LCR+ cell lines showed stronger activity of BCL6 enhancers and minimal activity of 195B. Activity of the subunit 195B reporter was significantly reduced by deletion of an evolutionarily conserved OCT2-binding motif (Fig. 5D) and by CRISPRi knockdown of POU2F2 and POU2AF1 (Fig. 5E), which also reduced acetylation of the endogenous GME-1 (Supplementary Fig. S5I). Thus, our findings indicate that ternary complex factors contribute to GME-1–dependent activation of MYC but are not sufficient, with other trans-regulators presumably accounting for divergent activity of this element in MYC-intact versus MYC::BCL6-LCR+ cell lines.

We next wondered whether genetic alterations of MYC locus distal regulatory elements might contribute to MYC activation in some DLBCLs. We analyzed published WGS data for DLBCL cell lines used in this study to look for structural variants (SV; refs. 40, 41). Importantly, SU-DHL-5 showed a 176 bp tandem duplication within the middle portion of the NFR 195B element (Fig. 5A) that does not correspond to a known germline polymorphism in the NCBI Short Genetic Variation database (dbSNP) and is therefore most likely of somatic origin. Inverted read-pairs supporting this duplication were also present in our ATAC-seq data, and PCR confirmed that both wild-type (WT) and duplicated alleles are present in our SU-DHL-5 cells (Supplementary Fig. S5J and S5K). A reporter assay showed that the duplication-containing 195B allele drives stronger transcription than the WT 195B allele in three MYC-intact cell lines (Fig. 5F). An apparent somatic point mutation (hg38 chr8:128,172,267 A>T) was present within the duplicated region in WGS reads and a subset of NFR195B clones for SU-DHL-5, but this alteration did not increase reporter activity for either the WT or duplicated enhancer (Supplementary Fig. S5L). For the BJAB cell line, published WGS data and our chromatin data identified different SVs internal to the MYC locus that both maintain juxtaposition of GME-1 to the MYC gene (Supplementary Fig. S5M). Importantly, SVs were not necessary for GME-1 dependency, as two of the four GME-1–dependent cell lines showed no evidence of MYC locus SVs. Acetylation of GME-1 seemed to be biallelic in all four cell lines, as H3K27ac ChIP-seq reads contained presumed germline nucleotide variants (known common polymorphisms) at allelic ratios near 50% (Supplementary Fig. S5N).

We examined published WGS datasets from primary DLBCL biopsies to look for SVs affecting GME-1 (13, 40–42). Analysis of 144 DLBCL WGS tumor–normal pairs revealed four biopsies with focal MYC locus copy gains of a genomic region containing GME-1, but not the MYC gene itself (Fig. 6A; Supplementary Fig. S6A), including 2 of 48 biopsies from a cohort of recurrent or treatment-refractory DLBCL (42) and 2 of 96 unselected DLBCL biopsies (40, 41). These datasets showed no evidence for long-distance MYC locus rearrangements. A focal GME-1 copy gain was also present in 1 of 48 genomic microarray datasets from DLBCL biopsies profiled by The Cancer Genome Atlas (Supplementary Fig. S6B; ref. 43). The Ig loci and recurrent non-Ig MYC RP loci such as BCL6 and PAX5/ZCCHC7 (44) contain regulatory elements subject to increased somatic hypermutation in DLBCL. However, published DLBCL WGS analyses (45, 46, 41) did not identify increased somatic hypermutation clusters in the vicinity of GME-1 or other candidate 3′ native MYC enhancers (Supplementary Fig. S6C). Only 1 of 144 DLBCL WGS tumor–normal pairs we analyzed showed a high-confidence somatic mutation in NFR 195B, and this mutant allele did not show increased reporter activity in MYC-intact GCB-DLBCL cell lines (Supplementary Fig. S6D). Thus, focal tandem gain of the GME-1 MYC enhancer is an uncommon but recurrent event in primary DLBCL, but we see no evidence for oncogenic point mutations of GME-1.

Topological interactions with target gene promoters are key hallmarks of enhancer activity. We performed genome-wide topology mapping (Hi-C) on two biopsies of DLBCL with predicted MYC::BCL6-LCR fusions based on screening with a custom FISH probeset and two cases of GCB-DLBCL with ≥30% MYC expression by IHC but no MYC rearrangement by break-apart FISH (Fig. 6B; Supplementary Fig. S6E). Analysis with Hi-C Breakfinder (47) confirmed expected genomic rearrangements linking the MYC and BCL6 loci in both FISH-positive cases. For one biopsy (AGRR_003), a single unbalanced fusion linked the MYC gene to the BCL6-LCR in the same chromosomal orientation, whereas the second biopsy showed a fusion that linked MYC (chr8 + strand) to a site telomeric to the BCL6-LCR (chr3 − strand), thus juxtaposing the BCL6-LCR to MYC in the opposite genomic orientation. Despite these structural differences, both cases showed similar strong topological interactions between the BCL6-LCR and the MYC promoter, supporting a similar enhancer-hijacking mechanism. Virtual 4C analysis showed significant interactions between the GME-1 enhancer and the MYC promoter in the MYC-intact biopsies but not the two MYC::BCL6-LCR+ biopsies (Fig. 6C).

The promoter of the long noncoding RNA gene PVT1 (PVT1-P) has been shown to act as a negative cis-regulator of enhancer-dependent MYC activation in epithelial cancers (48). Focal deletion of PVT1-P was found to be a recurrent event in WGS datasets from DLBCL with the high-risk MYC-associated DZsig gene expression signature (13). Consistent with a potential role for PVT1-P in antagonizing enhancer function in DLBCL, our lymphoma biopsy Hi-C datasets indicated topological interactions between PVT1-P and GME-1 in MYC-intact DLBCL (Fig. 6B) and between PVT1-P and BCL6-LCR in MYC::BCL6-LCR+ DLBCL. WGS data from three DZsig+ lymphomas with PVT1-P deletions showed that one lacked additional SVs within the MYC TAD, whereas the other two cases showed multiple complex SVs that incorporated focal copy gains of native MYC enhancers identified in our CRISPRi screen (Fig. 7A) but also connected to short copy-gained segments of chromosome 8p outside of the MYC TAD, which could represent sources of heterologous regulatory elements (Supplementary Fig. S7A).

Our CRISPRi screens allowed us to directly measure the effects of PVT1-P silencing on lymphoma models driven by functionally characterized native and heterologous enhancers. Both our tiling and NFR-focused CRISPRi screens identified significant enrichment of sgRNAs targeting PVT1-P in GME-1–dependent cell lines (Fig. 7B and C) and in MCL cell lines driven by 5′ MYC enhancers. NFR-focused CRISPRi screens also showed a fitness-enhancing effect of PVT1-P silencing in the MYC::BCL6-LCR+ cell lines SU-DHL-4 and WSU-DLCL2, in which the MYC locus breakpoint is downstream of PVT1-P (Fig. 7B and C; Supplementary Fig. S7B). In contrast, we saw no growth-promoting effect of PVT1-P silencing in OCI-Ly1, Jeko-1, or DB, in which 3′ fusion breakpoints occur between the MYC gene and PVT1-P, excluding the latter from the derivative chromosome containing MYC. We confirmed that CRISPRi silencing of the PVT1 promoter increases MYC transcript levels in cell lines dependent on the 3′ GME-1 (Fig. 7D), with most PVT1-P–targeting sgRNAs also increasing MYC expression in cell lines dependent on 5′-enhancers or on a rearrangement placing the BCL6-LCR in cis with PVT1-P and MYC (Supplementary Fig. S7C). These findings confirm a role for PVT1-P in cis-repression of enhancer-dependent MYC activation in DLBCL and suggest that this mechanism drives selective pressure for PVT1-P deletion in lymphomas in which PVT1-P remains in cis with MYC-activating enhancers.

Transcriptional activation of MYC, a master regulator of anabolic metabolism and proliferation (49), is encoded by a massive diversity of tissue- and state-specific enhancers located throughout the 3 Mbp MYC topological domain (19, 21, 23, 25, 50, 51). Activation of MYC transcription in mature B cells and derived lymphomas has been previously linked to 5′ (upstream) Notch/RBPJ-activated enhancers (21, 52, 53), which are active in Epstein–Barr virus–transformed B cells (52), MCL (15, 21), and chronic lymphocytic leukemia (15, 21, 22). Initial growth of mitogen-stimulated naïve B cells requires NF-κB–dependent transcriptional activation of MYC (54, 55), which is associated with canonical NF-κB factor binding near the MYC promoter (56), although in vitro–activated B cells also show extensive interactions between the MYC promoter and an array of 3′ (downstream) candidate enhancers overlapping the PVT1 gene body (57, 58). In B cells within the GC LZ, stimulation of CD40 by cognate T follicular helper cells also activates MYC in an NF-κB–dependent fashion (59). In DLBCL, canonical NF-κB activation downstream of chronic active BCR and MYD88 signaling is associated with activated B cell–DLBCL but not GCB-DLBCL (60–62). Thus, these mechanisms seem unlikely to explain MYC activation in high-risk DZSig lymphomas, which show low expression of NF-κB and Notch target genes (5). Importantly, the GME-1–dependent GCB-DLBCL cell lines studied here show low expression of NF-κB target genes (Supplementary Fig. S7D) and low nuclear levels of canonical NF-κB factors linked to MYC activation (Supplementary Fig. S7E).

Our identification of the 3′ GME-1 enhancer as a common dependency of MYC-intact GCB-DLBCL cell lines suggests mechanisms that may drive increased MYC expression in aggressive MYC-intact lymphomas (Fig. 7E). GME-1 is acetylated and shows specific topological interactions with the MYC promoter in patient biopsies from MYC-intact, but not MYC-rearranged, GCB-DLBCL. We identified rare but recurrent focal tandem copy gains of GME-1 in DLBCL WGS data, including high-risk DLBCL cohorts, suggesting selection pressure for increased GME-1 cis-regulatory activity in DLBCL evolution. PVT1-P was originally shown to inhibit MYC activation in breast cancer cells that rely on enhancers within the PVT1 gene body (48), but we showed that PVT1-P also antagonizes MYC in DLBCL cell lines in which MYC is activated by the intergenic 3′ enhancer GME-1, possibly explaining the increased frequency of PVT1-P deletions in DZsig+ DLBCL lacking MYC rearrangement to recurrent intergenic partner loci (13).

We showed that GME-1 is a direct target of OCT2, OCA-B, and MEF2B, a triad of factors implicated in GC-specific transcriptional activation of BCL6 and other genes. It is intriguing to note that three of the four GME-1–dependent cell lines in this study bear genomic copy gains and correspondingly increased transcript and protein levels for POU2F2 [BJAB (39)] or POU2AF1 (Karpas-422 and HT; Supplementary Fig. S7F–S7I), although these cell lines did not show consistently increased levels of all three factors compared with other DLBCL models. Future efforts are warranted to uncover additional trans-regulators of GME-1 that could explain why its activity diverges from that of the ternary complex–regulated BCL6 enhancers and to determine the extent to which this element is regulated by upstream cell-autonomous or microenvironment-derived signals in vivo.

Our study also provided new insights into mechanisms of MYC activation via rearrangement to nonimmunoglobulin loci, clarifying that complex intrachromosomal MYC rearrangements can also function as enhancer-hijacking events and showing that the position of 3′ rearrangement breakpoints could affect negative regulation of MYC by PVT1-P. Our CRISPRi data showed that the OCI-Ly1 cell line relies on enhancers scattered across two heterologous regions of chromosome 8 that are rearranged to MYC in a manner that eliminates PVT1-P from its position in cis with MYC. Loss of PVT1-P suppression might allow for strong activation of MYC by either native 3′ enhancers or by heterologous enhancers with weaker intrinsic activity than those in common recurrent partner loci. We speculate that both mechanisms may apply in the case of DZsig+ lymphomas with concurrent PVT1-P deletion, native MYC enhancer copy gains, and incorporation of heterologous fragments from other chromosome 8p loci.

The BCL6 locus is a recurrent non-IG MYC RP, resulting in “pseudo-double-hit” rearrangements that can be misinterpreted as separate rearrangements of MYC and BCL6 on clinical FISH assays (15, 29). Our functional studies implicated specific OCT2/OCA-B/MEF2B-regulated BCL6-LCR elements as drivers of MYC activation in these cases regardless of rearrangement complexity. Hi-C data from DLBCL biopsies indicate that enhancer-hijacking fusion of the BCL6-LCR to MYC can occur with the enhancer in either orientation, as the fusion in biopsy AGRR_003 occurred telomeric to the BCL6-LCR in the BCL6 locus “alternative breakpoint region” (63) but resulted in similar topological interactions between MYC and the BCL6-LCR as a biopsy (AGRR_002) with the breakpoint on the centromeric side of the BCL6-LCR. Although the GME-1 MYC enhancer and BCL6 enhancers that are essential in MYC::BCL6-LCR–rearranged lymphomas are both regulated by the GC ternary complex TFs, the dynamic regulation of these elements seems to differ in normal GC B cells, with BCL6 enhancers uniformly accessible throughout in the GC cycle whereas GME-1 elements show strongest accessibility in LZ–DZ reentrant populations, paralleling expression of MYC. These findings suggest that enhancer-hijacking rearrangements in GCB-DLBCL may dysregulate the dynamics of MYC transcription rather than simply increasing transcript levels. Further investigations are warranted to identify other TFs and upstream pathways that contribute to its activation.

Precise regulation of MYC is critical in normal GC B cells, in which transient MYC upregulation facilitates reentry into the GC proliferation cycle (37, 64, 65), thus regulating dynamics of clonal selection (38, 66). Further investigation is warranted into the role of the evolutionarily conserved GME-1 enhancer in normal B-cell biology.

All lymphoma cell lines were grown at 37°C and 5% CO₂ in RPMI 1640 medium with Glutamax (Gibco #61870036), supplemented with 10% FBS (Gibco #16000044 or similar), 1× minimum essential medium nonessential amino acid solution (Gibco #11140050), 1 mmol/L sodium pyruvate (Gibco #11360070), 1× penicillin–streptomycin (Gibco #15140122), and 55 μmol/L 2-mercaptoethanol (Gibco #21985023). See Supplementary Table S1 for details of cell line source and short tandem repeat validation testing. Short tandem repeat testing (University of Illinois Urbana-Champaign Tumor Phenotyping Shared Resource) was repeated to confirm identity of cell line derivatives used for multiple experiments, such as KRAB-dCas9–expressing populations. Cell lines were tested to verify the absence of Mycoplasma (MycoAlert, Lonza #LT07-318) at the time of receipt and annually thereafter during active use.

Coded excess human lymphoma tissue samples obtained from a clinically indicated procedure were procured from the University of Michigan Hematopathology Repository (frozen tissue) or from the pathology archives of the University of Michigan [formalin-fixed paraffin-embedded (FFPE) tissue], along with limited demographic, clinical, and diagnostic information. Studies on coded/deidentified human tissue were performed in accordance with the policies of the University of Michigan Institutional Review Board and the U.S. Common Rule.

GCB-DLBCL samples used for chromatin studies were selected on the basis of the Hans IHC algorithm [(67); all samples were CD10⁺ and BCL6⁺] and tumor cellularity in the relevant sample of at least 70%. See Supplementary Table S4 for patient demographics, biopsy site, and the results of clinically performed IHC and FISH.

For virus production in 6-well plates, 6 × 10⁵ 293T cells were seeded 24 hours prior to transfection in 2 mL DMEM +10% FBS. Transfection with TransIT-LT1 (Thermo Fisher Scientific #MIR2300) was performed per the manufacturer’s instructions with standard second-generation lentiviral packaging plasmids pVSV-G, pGag-Pol-Tat-Rev, and the lentiviral transfer plasmid in a 1:10:10 ratio. Media were changed 18 hours later to 2.6 mL of DMEM + 40% heat-inactivated FBS (viral harvest media). The lentiviral supernatant was harvested 24 hours later and replaced, and a second harvest was performed at 48 hours Production was scaled-up as needed for larger experiments. For some experiments, lentivirus was concentrated with Lenti-X (Takara #631232) per the manufacturer’s instructions to increase titer.

For cell line H3K27ac ChIP-seq, 5 million cells were cross-linked in PBS + 1% formaldehyde (Thermo Fisher Scientific #28908) at room temperature for 10 minutes with periodic inversion, quenched for 5 minutes with glycine (Sigma #G7126) added to a final concentration of 125 mmol/L, and washed twice in PBS plus protease inhibitors (PI; Roche cOmplete #11836170001), and pellets were frozen at −80°C. Pellets were thawed, and nuclei were isolated via a cytoplasmic lysis buffer (20 mmol/L Tris-HCl, pH 8.0 (Thermo Fisher Scientific #NC9246955), 85 mmol/L KCl (Thermo Fisher Scientific #AM9640G), and 0.5% NP 40 (Sigma #I8896) + PI) and centrifugated at 400 g for 5 minutes at 4°C. Nuclei were resuspended in cold SDS lysis buffer (0.3% SDS, 10 mmol/L EDTA, and 50 mmol/L Tris-HCl, pH 8.1, + PI) and homogenized by passage through a 27-g needle. Sonication was performed using a Q800R2 Sonifier (QSonica) set to amplitude 70 alternating between 45 seconds on and 15 seconds off for a total of 8 minutes 30 seconds on. Samples were then diluted 1:3 in ChIP dilution buffer [0.01% SDS (Sigma #71736), 1.1% Triton X-100 (Sigma #T8787), 1.2 mmol/L EDTA (Boston Bioproducts #NC1584107), 16.7 mmol/L Tris-HCl, pH 8.1, and 167 mmol/L NaCl +PI] and rotated at 4°C overnight with 2 μg of antibody (H3K27ac, Active Motif, Cat. # 39133, RRID: AB_2561016). Chromatin complexes were bound for 4 hours on Protein G Dynabeads (Thermo Fisher Scientific #10-003-D). Subsequent washing, DNA elution, purification, and Illumina library preparation steps were performed as previously described (68).

For TF ChIP-seq, the protocol used was previously detailed (15). In brief, this protocol differs from that above as follows: 20 million cells were used per assay, cells were crosslinked in media + 1% formaldehyde at 37°C, nuclei were resuspended in higher SDS buffer (1% SDS, 10 mmol/L EDTA, and 50 mmol/L Tris-HCl, pH 8.1, + PI), and a Branson probe sonifier was used for chromatin shearing. IP was performed with 2 to 5 μg of antibodies against Oct2 (Santa Cruz Biotechnology, Cat. # sc-233, RRID: AB_2167205), OCA-B, (Santa Cruz Biotechnology, Cat. # sc-955, RRID: AB_2166917), or p300 (Bethyl, Cat. # A300-358A, RRID: AB_185565). MEF2B ChIP-seq datasets were previously published (15).

For DLBCL biopsy ChIP-seq, frozen excess surgical tissue from diagnostic GCB-DLBCL biopsies was provided by the University of Michigan Hematopathology Repository. OCT-embedded tissue was cryosectioned into 25 μmol/L scrolls after removal of areas with less than 70% estimated tumor cellularity as assessed by a board-certified hematopathologist (R.J.H. Ryan). A measure of 20 to 50 mg of tissue was resuspended in cold PBS + PI, and 10 mmol/L sodium butyrate, broken up by serial pipetting and passage through a 21-gauge syringe, was crosslinked in 1% formaldehyde at room temperature for 15 minutes, quenched in 125 mmol/L glycine, washed in PBS + PI, and resuspended in SDS lysis buffer. Sonication and subsequent steps were performed as described above.

For ChIP-PCR, sonicated chromatin was divided equally for IP with H3K27ac antibody or IgG control, IP, washing, and elution were performed as for ChIP-seq. Eluted IP and pre-IP input DNA were used for SYBR green qPCR with primers designed within H3K27ac signal peaks adjacent to MYC and BCL6 NFRs or control region, and % of input recovered by IP was calculated by the relative standard curve method.

ATAC-seq was performed as previously described (68). Nuclei were isolated from 50,000 cells for each sample using Nuclei EZ prep-Nuclei Isolation Kit (Sigma-Aldrich). The transposition reaction mix [25 μL of 2 × TD buffer, 2.5 μL of Tn5 transposase (Illumina), 15 μL of PBS, and 7.05 μL of nuclease-free water] was added to nuclei and incubated at 37°C for 1 hour in an orbital shaker at 300 RPM. 50 μL Qiagen buffer PB was added to each sample to stop the reaction, and DNA was isolated with AMPure XP beads (Beckman Coulter #A63881). Fifteen cycles of PCR were performed with transposed DNA using the dual-index primers and NEBNext PCR Master Mix (NEB #M0541), followed by AMPure XP purification. After quantification and fragment size analysis, libraries were sequenced on Illumina NextSeq.

Paired-end ChIP-seq and ATAC-seq reads were aligned to hg19 and hg38 using BWA-ALN (v 0.7.17) and filtered to remove PCR duplicates (Picard MarkDuplicates) and read-pairs mapping to >2 sites genome-wide. Display files were generated with deepTools bamCoverage and visualized with IGV. Single-end ChIP-seq datasets were processed similarly, except that individual reads mapping to >2 sites genome-wide were removed, and display file generation was performed with “igvtools count.” Scaling for all newly generated ChIP-seq and ATAC-seq tracks in figures is equal to local paired-end fragment coverage × (1,000,000/totalCount). ATAC-seq dataset quality (fragment length distribution and TSS enrichment) was confirmed with ataqv (69). For ATAC-seq datasets, peak summits were identified with MACS2 (v 2.2.7.1; ref. 70) using default parameters. ChIP-seq peak calling was performed with HOMER findPeaks (71) using the “factor” style and default parameters. All peak sets were post-filtered against hg38 blacklist regions (available at https://github.com/Boyle-Lab/Blacklist/blob/master/lists/hg38-blacklist.v2.bed.gz). Our complete processing pipeline is available at https://github.com/russell-ryan-lab/ChIPseq_ATACseq_pipelines.

To generate doxycycline (dox)-inducible KRAB-dCas9–expressing cell lines (19), cells were cotransduced with TRE-KRAB-dCas9-IRES-GFP (RRID: Addgene_85556) and EF1a_TetOn3G (Clontech) lentivirus and serially sorted to derive populations that were uniformly GFP-negative before and GFP-positive after doxy (Sigma #D9891) induction. The functionality of each KRAB-dCas9–expressing cell line for CRISPRi studies was validated by showing loss of viability when transduced with sgRNAs targeting essential genes (RPL34 and RPL8) versus nontargeting controls upon dox induction. For the cell line DB, we used a constitutive KRAB-dCas9 vector, UCOE-SFFV-KRAB-dCas9-P2A-mCherry (72).

To design the NFR-focused sgRNA library, we defined target regions within the MYC locus to include the MYC/PVT1 promoter region (chr8_MYC_genic) as well as 5′ (chr8_MYC_5′enh) and 3′ (chr8_MYC_3′enh) regions that include all strongly acetylated elements in the subject cell lines and interact topologically with the MYC promoter [see MYC promoter 4C-seq data from the Karpas-422 cell line in this study and published 4C-seq data from REC-1 cell line (21)]. We covered the entire TAD upstream of BCL6 (chr3_BCL6_5′enh). We also included two additional RP regions linked to the MYC and BCL6 genes in the SU-DHL-4 cell line based on optical mapping and 4C-seq (chr11_RP1_SU-DHL-4 and chr1_RP2_SU-DHL-4) and regions of chr8 rearranged to the MYC locus in the JeKo-1 (chr8_RP_JeKo1) and OCI-Ly1 (chr8_RP1_OCI-Ly1 and chr8_RP2_OCI-Ly1) based on PEAR-ChIP. Within these regions, ATAC-seq peaks from 26 DLBCL and MCL cell lines were used to define candidate NFRs, which were extended to 500 bp from the peak center. For each peak in each cell line, an enhancer “activity score” (17) was calculated as the geometric mean of normalized H3K27ac and ATAC-seq reads, and peaks with activity <15 were discarded. The remaining peaks across the 26 cell lines were merged into a consensus set of NFRs with high predicted enhancer activity in at least one cell line; overlapping peaks were merged.

All candidate sgRNAs within the target region NFRs were identified and scored with FlashFry (73), and sgRNAs that met the following scoring criteria were retained: Doench2014OnTarget >0.1, Hsu2013 > 50, JostCRISPRi_specificityscore >0.1, dangerous_GC = = “NONE,” dangerous_polyT = = “NONE,” dangerous_in_genome = = “IN_GENOME = 1,” and otCount <500. The remaining sgRNAs in each interval were downsampled to not more than 35 per 1 kb.

The final libraries also contained 646 sgRNAs covering extended tiling regions over the MYC and BCL6 promoters and proximal gene bodies and 392 guides targeting nonenhancer control regions in the MYC and BCL6 loci. We also included CRISPRi-optimized sgRNAs (five per gene) from the Dolcetto library (18) targeting the promoters of 50 pan-essential genes, 38 genes selectively essential in BCL based on DepMap CRISPR knockout screens (74), and 575 TF genes expressed in B-cell cancer cell lines. A total of 866 nontargeting sgRNAs were included as negative controls.

Pools of oligos with the sequence 5′-TATCTTGTGGAAAGGACGAAACACCG-{20-mer-seed}- GTTTAAGAGCTATGCTGGAAACAGCATAG-3′ were ordered from CustomArray or Twist Biosciences. Twenty-four nanograms of the oligo pool were PCR-amplified using NEBNext Master Mix (NEB #M0541) and 1 μM each of sgRNA_Library_Fwd/Rev primers. PCR products were purified with 1.5× AMPureXP beads (Beckman-Coulter #A63881) and cloned into BsmBI-digested sgOpti by Gibson assembly with NEBuilder HiFi DNA Assembly Master Mix (NEB #E2621) per the manufacturer’s protocol and a 5-fold molar excess insert to vector ratio. One μL of assembly reaction was electroporated with a Bio-Rad Gene Pulser Xcell system into Endura competent cells (Lucigen # 60242), and culture outgrowth, determination of transfection efficiency, and plasmid purification were performed as described (75).

Two strategies were used for functional screening of cis-elements by CRISPRi: “NFR” screens (cell lines SP-49, SU-DHL-5, Karpas-422, JeKo-1, OCI-Ly1, SU-DHL-4, WSU-DLCL2, and DB) using the new library described above and “tiling screens” (cell lines SP-49, Granta-519, and Karpas-422) conducted with a published sgRNA library (19) designed to cover broad regions of the MYC locus. CRISPRi screening was performed as previously described (19) in dox-inducible KRAB-dCas9–expressing cell lines with two to three biological replicates per tiling screen and three to four biological replicates per NFR-focused screen. Briefly, sgRNA lentiviral supernatant was produced in 293T cells, titered by small-scale transduction (24-well plate) with a range of viral supernatant concentrations (0%–50%) followed by 1 μg/mL puromycin treatment at 24 hours and cell counting at 96 hours after transduction. Scaled-up viral transduction was then performed in 6-well plates at a multiplicity of infection of 0.4 with a total cell number calculated to yield at least 500× coverage of transduced cells per library sgRNA per replicate (e.g., about 32 million cells for a 25k sgRNA library). Cells were treated with 1 μg/mL puromycin at 24 hours after transduction and expanded for 72 hours, and then initial samples were collected from each replicate (T0). Each replicate pool was then passaged in media containing 100 ng/μL dox, maintaining a population of >500 cells per library sgRNA at all passages, for a total of 14 population doublings until endpoint harvest (T14). For the screen conducted in the DB cell line with noninducible KRAB-dCas9, equal numbers of the parental cell line were transduced in parallel and harvested following selection as “T0” replicates, whereas the dCas9-KRAB–expressing replicates were grown for 14 doublings after puromycin selection. QIAmp DNA Midi or Maxi Kit (Qiagen #51183 or #51192) was used for genomic DNA extraction from T0 and T14 cell pellets. For each replicate and timepoint, genomic DNA corresponding to 500 cells per library sgRNA were amplified for 26 cycles across multiple 50 μL PCR reactions (6 μg DNA per reaction) with barcoded p5 and p7 Illumina primers (Supplementary Table S1) and ExTaq polymerase (Takara #RR001). PCR products for each replicate were pooled, and 25 μL of pooled PCR product were purified on one lane of E-Gel SizeSelect II (Thermo Fisher Scientific #G661012). Purified amplicons were pooled and submitted for Illumina sequencing.

Tiling sgRNA screens were analyzed as previously described (19). For NFR-focused screens, statistical analysis of sgRNA depletion/enrichment was performed with MAGeCK-RRA (RRID: SCR_025016; ref. 76), applied separately to NFR-targeting and gene-targeting sgRNA sets. Option “–norm-method control” was used to normalize log₂fold changes to the nontargeting sgRNAs.

For expression of sgRNAs, 20-mer seed sequences with a prepended “G” were cloned into BsmBI-digested sgOpti (RRID: Addgene_85681). For fitness assays, dox-inducible KRAB-dCas9–expressing cells were transduced in triplicate with sgOpti lentivirus and divided at 24 hours into media with 1 μg/mL puromycin with or without 500 ng/mL dox. Viable cellularity was quantified at day 7 with CellTiter-Glo (Promega #G7527), and the ratio of viable cellularity between induced and uninduced samples of each replicate was calculated, normalized to the mean value for control sgRNAs. For gene expression studies, 1 million cells per replicate were transduced with sgRNA lentivirus, transferred at 24 hours to media with 1 μg/mL puromycin (Thermo Fisher Scientific #A1113803) and 500 ng/mL dox (Sigma #D9891), and harvested 3 or 5 days after transduction for RNA extraction and qRT-PCR. Dual sgRNA studies used derivatives of sgOpti bearing tagBFP (pMW-tagBFP) or tagRFP (pMW-tagRFP) reporters. Inducible KRAB-dCas9–bearing cell lines were transduced with two sgRNA lentiviral supernatants bearing different fluorescent protein reporters, sorted for cells with dual reporter expression, and then induced with 100 to 500 ng/mL dox for 3 days (gene expression) or divided into triplicate dox-induced and triplicate uninduced sub-pools, with the normalized viable cellularity ratio quantified at day 7 with CellTiter-Glo as described above.

RNeasy columns (Qiagen #74104) were used for RNA isolation followed by spectrophotometric quantification. High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific #4368814) was used for cDNA generation from 500 ng RNA per sample in 20 μL reactions. Following the reverse transcription reaction, cDNA was diluted 1:10. qPCR was performed on a Thermo QuantStudio 5 384-well Real-Time PCR system using Power SYBR Green PCR Master Mix (Thermo Fisher Scientific # 4367659) with 2 μL cDNA per test sample. All independent biological samples were run with two to five technical PCR replicates. Primers used for qRT-PCR are listed in Supplementary Table S1A. Quantification was by the relative standard curve method with expression of each test gene normalized to EEF1A primers run on the same plate. Biological replicate values are the mean of corresponding technical replicates. For sgRNA experiments, values are expressed relative to the mean of control (nontargeting and control region–targeting) sgRNAs.

Fresh-frozen cell line pellets were sent to Bionano Genomics for high-molecular weight DNA extraction (Bionano Prep SP DNA Isolation Kit, Bionano #80030), direct labeling with DLE-1 and DNA backbone counterstaining (Direct Label and Stain-G2 Kit, Bionano #80046), and imaging on a Saphyr chip (RRID: SCR_017992). Optical map images were converted to BNX format after which de novo assembly and SV detection were performed using Bionano Genomics pipelines. SV read maps involving the MYC, BCL6, and partner loci were visualized with BioNano Access software v1.7.2.

4C-seq analysis was performed on Karpas-422, DB, and SU-DHL-4 cells as previously described (21, 77, 78), with the 4-cutter restriction enzymes DpnII (NEB #R0543; primary) and CviQI (NEB #R0639; secondary). Viewpoint primers for the MYC promoter and the BCL6-LCR element NFR 31 are listed in Supplementary Table S1. Heatmaps of 4C interaction frequencies across a range of resolutions in genomic loci of interest were generated with 4C-seq pipe (78). To visualize interactions across the MYC::BCL6-LCR fusion in DB and SU-DHL-4, 4C-seq pipe was run with a custom DpnII-CviQI–digested hg19 reference containing a ch8/chr3 fusion chromosome (chr8 position 1-128,760,000, followed by chr3 position 187,480,000 to terminus).

JeKo-1 populations engineered to express dox-inducible KRAB-dCas9 were transduced with lentivirus generated from different combinations of one tagBFP-expressing (pMW-tagBFP) and one tagRFP-expressing (pMW-tagRFP) sgRNA vector, followed by sorting for dual expressors. Each sorted population was then divided and transduced with lentivirus generated from pINDUCER-20 bearing either a MEF2B-HA or control (GFP) transgene and selected in G418 sulfate (Thermo Fisher Scientific #10131035) for 10 days. Cells were treated with 500 ng/mL dox for 72 hours to induce expression of both the pINDUCER transgene and KRAB-dCas9, followed by RNA extraction for qRT-PCR.

Seurat objects and ATAC-seq fragment files from a published source (36) were downloaded from Zenodo (https://zenodo.org/records/8373756) and visualized with Seurat 5.1.0 and Signac 1.13.0. For ATAC-seq data visualization in Fig. 4G, cells were assigned to published subclusters (“annotation_20230508”). As indicated by asterisks in Fig. 4G, some B-cell clusters were combined for simplicity as follows (merged cluster = original clusters): “memory B*” = “ncsMBC,” “csMBC”; “DZ S-phase*” = “DZ early S-phase,” “DZ late S-phase”; “DZ G2M phase*” = “DZ early G2M phase,” “DZ late G2M phase”; “plasma cell precursor*” = “IgG+ PC precursor,” “IgM+ PC precursor,” “IgD PC precursor”; “plasma cell*” = “premature IgG+ PC,” “premature IgM+ PC,” “mature IgG+ PC,” “MBC-derived IgG+ PC,” “mature IgA+ PC,” and “MBC-derived IgA+ PC.” Normalized pseudobulk ATAC-seq signal profiles for selected populations in regions of interest were visualized with Signac CoveragePlot (assay.scale = common). Pseudo-bulk expression of MYC and BCL6 in cells assigned to these same populations was plotted with Seurat AverageExpression.

To systematically identify conserved intervals in mm10 that correspond to hg19 NFRs, we used a UCSC hg19–mm10 genomic alignment net file to identify the mm10 position corresponding to the centermost base in each hg19 NFR (merged ATAC-seq peaks as described above) that is mappable to mm10. Mapped intervals were then extended to 200 bp for visualization. To extract conserved sequences from multiple vertebrate species corresponding to the essential MYC and BCL6 locus NFRs, evolutionarily conserved subregions were manually defined based on clusters of predicted evolutionarily conserved elements in hg38 (Phastcons vertebrate 100-way). We then extracted ungapped sequences aligned to the hg38 intervals consisting of the longest mappable region with base matching ratio >0.1 for each of 70 vertebrate species for which pairwise MULTIZ chain files to hg38 were available. HOMER findMotifs was then used to annotate the syntenic interval for each species with all significant matches to the HOMER known motif library. Specific motifs and motif families of interest were then selected for visualization.

Enhancers were PCR-cloned from genomic DNA into the STARR-seq firefly luciferase validation vector ORI_empty (RRID: Addgene_99297). Vectors ORI_SV40 (RRID: Addgene_99309) and ORI_neg.cont (RRID: Addgene_99315) were used as positive and negative controls. We modified the NFR195B clone via site-directed mutagenesis (Q5 SDM kit E0554S; NEB) to delete the indicated OCT2-binding motif (“TTTGCAT”) and to generate the patient-derived mutant allele from DLBCL 14-35632 (hg38 chr8:128172453_T>C/chr8:128172455_G>GA). Cell lines were electroporated in quadruplicate (Neon, Thermo Fisher Scientific) with firefly luciferase and Renilla control [pRL-SV40 (RRID: Addgene_27163)] plasmids in 1:10 ratio, plated in growth media for 48 hours. The ratio of firefly to Renilla luciferase activity was measured using Dual-Glo Luciferase Kit (Promega; E2940). Neon electroporator tips were cleaned with DNAase I solution and ethanol-sterilized for reuse as described (79).

Dox-inducible KRAB-dCas9–expressing HT cells were transduced with lentivirus (sgOpti) encoding CRISPRi-optimized sgRNA against the indicated target gene promoter or control. Twenty-four hours later, the cells were treated with 1 μg/mL puromycin. A measure of 500 ng/mL dox was added 72 hours later. Each transduced population was divided into subsamples and electroporated in triplicate with the indicated luciferase reporter vector and Renilla control (pRL-SV40) plasmids in 1:10 ratio 24 hours after dox induction. Luminescence was read out using Dual-Glo Luciferase Kit 48 hours after reporter plasmid electroporation (Promega; E2940).

Sequence variants in the GME-1 enhancer (hg19-chr8:129,180,000-129,190,000) were identified with bcftools mpileup in DLBCL H3K27ac ChIP-seq BAM files. The ratio of reference to alternate bases was calculated for loci represented in at least 10 reads.

Public data from sources listed in Supplementary Table S1C were used to assess GME-1 amplification events in human DLBCL samples. The raw FASTQ files were downloaded from the Sequence Read Archive databases for phs000328.v3.p1, phs003023.v1.p1, and phs000532, whereas aligned BAM files were downloaded from the European Genome-phenome Archive database for EGAD00001004142 and EGAD00001006087. Processing for WGS was performed through in-house genomic pipelines as previously described. Briefly, raw FASTQ files were first trimmed by overlap using bbduk which is part of the bbmap suite, followed by alignment using Burrows-Wheeler Aligner–MEM to the GRCh38 standard reference (GRCh38DH+alt+hla). After alignment, deduplication was performed using Picard rules, and duplicate reads were removed from all subsequent analyses. The overall pipeline matches the functional equivalence requirements as proposed by Regier and colleagues (80). For samples that were originally aligned to hg19 reference, unalignment using biobambam2 was first performed to extract raw reads from BAM files, followed by realignment using GRCh38 reference using the same methods and settings as previously stated.

SVs were identified using Manta (81), SVABA (82), and TNscope (bioRxiv 250647v1) on canonical hg38 chromosomes with default record-level sample-level filters for WGS data. SVs were kept in cases in which sample site depth was less than 3× the median chromosomal depth near one or both breakends (variants in which depth exceeded this threshold were manually reviewed), and paired and split-read evidence was required to improve accuracy of SV calls. All SV calls within the MYC topologically associating domain were additionally manually reviewed to remove recurrent artifacts and false-positive calls. For samples without normals, only Manta was performed, and more stringent manual curation was performed on the resulting calls to remove likely germline variants and alignment artifacts.

Copy-number analysis for WGS data was performed with CNVEX (https://github.com/mctp/cnvex) to estimate coverage within fixed genomic intervals. Whole-genome variant calls were used to compute B-allele frequencies (BAF) at variable positions observed in GNOMAD v3.1 (we used DNAScope germline calls). Coverages were computed in 10-kb bins, and the resulting log coverage ratios between tumor and normal samples were adjusted for GC bias using weighted LOESS smoothing across mappable and non-blacklisted genomic intervals within the GC range or 0.3 to 0.7, with a span of 0.5 (the target, blacklist, and configuration files are provided with CNVEX). The adjusted log coverage ratios (LR) and BAFs were jointly segmented by custom algorithm based on circular binary segmentation (CBS) and iterative pruning. For the CBS-based algorithm, first LRs and mirrored BAFs were independently segmented using CBS (parameters alpha = 0.01, trim = 0.025), and all candidate breakpoints were collected. The resulting segmentation track was iteratively “pruned” by merging segments that had similar LRs, BAFs, and short lengths.

Small somatic variants were called in DLBCL tumor–normal WGS datasets using TNscope, which is a modified version of GATK3 MuTect2 algorithm. The following settings were used “max_fisher_pv_active 0.05,” “min_tumor_allele_frac 0.0075,” “min_init_tumor_lod 2.5,” “assemble_mode 4,” “trim_soft_clip,” and a germline resource based on the Single Nucleotide Polymorphism Database 138 for GRCh38. The resulting variants were subsequently filtered using the following evidence: variant allele fraction in the tumor sample >0.05 and variant allele fraction in the normal sample <0.01. In addition, algorithm-specific thresholds for tumor genotype quality (GQ) >30, log-odds that the variant is not present in the normal sample NLOD >6, and log-odds that the variant is not present in the normal sample given the allele fraction in the tumor sample NLODF >3 were used.

Genomic microarray–derived copy-number segmentation data from The Cancer Genome Atlas PanCancer Atlas were downloaded from CBioPortal (https://www.cbioportal.org).

Tissue microarrays of clinically obtained FFPE lymphoma tissue biopsies were screened by FISH to determine rearrangement status for MYC (Vysis LSI MYC Dual Color Break-Apart Rearrangement Probe, Abbott, catalog #05J91-001) and to identify MYC::BCL6-LCRfusions. MYC::BCL6-LCRfusions were identified by a single-fusion FISH strategy utilizing a commercial probe for the MYC gene (Vysis LSI MYC SpectrumGreen Probe, Abbott, catalog #04N36-020) and a custom probe generated from bacterial artificial chromosomes RP11-964B17 and RP11-4679A8, labeled with 5-ROX dUTP (Empire Genomics). Five-micron sections (n = 5–10) were cut from FFPE lymphoma biopsies. Sections were macrodissected (if needed) to retain only areas with >75% lymphoma cellularity as assessed by a board-certified hematopathologist (RJHR). Tissue was sent to Arima Genomics and processed for Hi-C library construction with the Arima-HiC+ FFPE service (catalog #A201090) as described by the manufacturer. Libraries were sequenced on an Illumina NovaSeq 6000 Sequencing System (RRID: SCR_016387).

Hi-C sequencing data were processed with the Arima-SV pipeline (A101060, version v01, 02/2022). Briefly, HiCUP (83) was used for read alignment followed by hic_breakfinder (47) for identification of SVs. Juicer (84) was used for generation of the hg38 genome-wide contact matrix file (.hic), which was visualized with JuiceBox. Virtual 4C analysis was performed by extracting interactions from the genome-wide contact matrix at 10-kb resolution to obtain an interaction matrix. Interaction frequencies between the MYC promoter bin (chr8: 127,730,000–127,740,000) and other genomic bins were plotted as a histogram. To perform loop calling, hic contact matrix files were first converted to cool format using the HiCExplorer (85) hicConvertFormat command at a resolution of 10kb. Distal sites with significant MYC promoter interactions were identified using the hicDetectLoops command with the following parameters: P < 0.05, windowSize 6, and peakWidth 10. Plotgardener (86) version 1.6.1 was used for visualization.

Nuclear and cytoplasmic protein fractions were extracted from pellets of 10 to 20 million cells using NE-PER Nuclear and Cytoplasmic Extraction Reagents Kit (Thermo Fisher Scientific #78835) according to the manufacturer’s protocol. Lysates were supplemented with PI cocktail (Complete Mini, Sigma #11836153001) and with phosphatase inhibitor (Halt, Thermo Fisher Scientific #PI78420) for phospho-epitope blots. 6× Laemmli sample buffer (Thermo Fisher Scientific #AAJ61337AC) was added to equal amounts of protein for matched samples, which were then denatured in a 95°C heat block for 5 minutes. 20 μg of protein per lane were loaded on NuPAGE 4% to 12% Bis-Tris Midi 20-well Gel (Invitrogen #WG1402) in a SureLock Tandem Midi Gel Tank (Invitrogen #STM1001). After electrophoresis (120V, 75–90 minutes), proteins were transferred to a 0.45 µm nitrocellulose membrane (Thermo Fisher Scientific # 88018) with a Bio-Rad Trans Blot Turbo system (RRID: SCR_023156) system (60 minutes, 2.5 A, up to 25 V). The membrane was blocked for 1 hour in 5% nonfat dry milk in TBS +0.1% Tween 20 (TBST), and the primary antibody was added at 1:250 to 1:1,000 overnight at 4°C. After washing 3× in TBST, blots were probed with secondary antibody [horseradish peroxidase (HRP)-linked anti-mouse IgG, Cell Signaling Technology #7076, RRID: AB_330924; HRP-linked anti-rabbit IgG, Cell Signaling Technology #7074, RRID: AB_2099233] at a dilution of 1:1,000-1:2000 in TBST. Blots were washed 5× in TBST and then probed with enhanced chemiluminescent HRP substrate (SuperSignal West Pico PLUS Chemiluminescent Substrate Thermo Fisher Scientific #34579; SuperSignal West Femto Maximum Sensitivity Substrate Thermo Fisher Scientific #PI34096; SuperSignal West Atto Ultimate Sensitivity Substrate # PIA38554) imaged with Bio-Rad ChemiDoc Imaging System (RRID: SCR_019684), and images were processed with Image Lab Software (RRID: SCR_014210). The following primary antibodies were used: MEF2B (Atlas Antibodies #HPA004734, RRID: AB_10963939), EBF (Santa Cruz Biotechnology #sc-137065, RRID: AB_2246405), E2A (Santa Cruz Biotechnology #sc-349, RRID: AB_675504), Bob1 (Santa Cruz Biotechnology #sc-955, RRID: AB_2166917), Oct-2 (Santa Cruz Biotechnology #sc-233, RRID: AB_2167205), Bcl-6 (Santa Cruz Biotechnology #sc-7388, RRID: AB_2063455), c-Myc (Cell Signaling Technology #5605, RRID: AB_1903938), CTCF (Cell Signaling Technology #3418, RRID: AB_2086791), c-Rel (Cell Signaling Technology #12707, RRID: AB_2721030), phospho-Ser536-RelA (Cell Signaling Technology #3033, RRID: AB_331284), RelA (Cell Signaling Technology #8242, RRID: AB_10859369), NF-κB1/p50 (Cell Signaling Technology #12540, RRID: AB_2687614), NF-κB2/p52 (Cell Signaling Technology #3017, RRID: AB_10697356), and TBP (Abcam #ab818, RRID: AB_306337).

Bar plots were plotted in GraphPad Prism 9.0 (GraphPad Software). Error bars denote the SD unless otherwise indicated in the figure legend. P values were calculated in GraphPad Prism or Excel, using the indicated statistical test, and are indicated in figures with asterisks as follows: *, P < 0.05; **, P < 0.01; ***, P < 0.001; ns, not significant; P ≥ 0.05.

Raw and processed high-throughput sequencing datasets have been deposited in Gene Expression Omnibus (GSE230396). Bionano optical genome mapping data has been deposited as an NCBI BioProject (PRJNA1006630).

K. Sikkink reports personal fees from Arima Genomics outside the submitted work. C.P. Fulco reports a patent for “CRISPR methods and enhancer mapping” (US20200143907A1) pending and is employed by and may hold stock in Sanofi. A.D. Schmitt reports other support from University of Michigan during the conduct of the study and other support from Arima Genomics outside the submitted work; in addition, A.D. Schmitt has a patent for WO2020106776A2 pending and issued. J.M. Engreitz reports grants from the NIH during the conduct of the study and nonfinancial support from 10x Genomics, personal fees from GSK plc, and personal fees from Roche Genentech outside the submitted work; in addition, J.M. Engreitz has a patent for U.S. Patent App. 16/337,846 issued, licensed, and with royalties paid from related to CRISPR technologies used in this work. R.J.H. Ryan reports grants from the NIH/NCI, V Foundation for Cancer Research, and American Society of Hematology and nonfinancial support from Arima Genomics and BioNano Genomics during the conduct of the study, as well as grants from the Hyundai Hope On Wheels Foundation outside the submitted work. No disclosures were reported by the other authors.

A.R. Iyer: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft. A. Gurumurthy: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft. S.-C.A. Chu: Formal analysis, visualization, writing–original draft. R. Kodgule: Investigation. A.R. Aguilar: Investigation. T. Saari: Data curation, software, formal analysis. A. Ramzan: Software. J. Rosa: Formal analysis, investigation. J. Gupta: Investigation. A. Emmanuel: Investigation. C.N. Hall: Investigation. J.S. Runge: Investigation. A.B. Owczarczyk: Investigation. J.W. Cho: Investigation. M.B. Weiss: Investigation. R. Anyoha: Investigation. K. Sikkink: Investigation. S. Gemus: Formal analysis. C.P. Fulco: Software, investigation. A.M. Perry: Resources, investigation. A.D. Schmitt: Formal analysis, supervision. J.M. Engreitz: Software, methodology, writing–review and editing. N.A. Brown: Resources, investigation. M.P. Cieslik: Formal analysis, supervision, writing–review and editing. R.J.H. Ryan: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

R.J.H. Ryan was supported by grants from the NCI (K08-CA-208013 and R01-CA-245059), American Society of Hematology (Scholar Award), and V Foundation for Cancer Research (V2018-001). J.M. Engreitz was supported by the NIH NHGRI (K99HG009917 and R00HG009917), the Harvard Society of Fellows, Gordon and Betty Moore, and the BASE Research Initiative at the Lucile Packard Children’s Hospital at Stanford University. The authors wish to acknowledge Thomas Giordano, Deborah Postiff, Farah Keyoumarsi, and other staff of the University of Michigan Tissue Procurement Core Facility and their funding support (NIH P30 CA04659229), Yuanyuan Chang and Joyce Lee at Bionano Genomics for assistance with optical mapping data generation and analysis, Laura Hilton for sharing unpublished analyses of SVs in DLBCL WGS datasets, and Bradley E. Bernstein, Mark Y. Chiang, and Sami N. Malek for helpful feedback on the manuscript.