SPEN loss drives extra-follicular diffuse large B cell lymphoma with female-specific lethality and therapeutic vulnerabilities

Abstract

Diffuse large B-cell lymphomas (DLBCL) are genetically and phenotypically heterogeneous, making diagnosis and treatment challenging. Current models suggest DLBCL derive from follicular B cells engaged in adaptive immune responses. By studying co-occurring truncating mutations in SPEN and NOTCH2 in the BN2-DLBCL subtype, our data suggest a previously unrecognized extra-follicular trajectory. Using animal models and human specimens, we find this cooperative mutational axis supports expansion of putative clonal precursors with features of marginal zone, memory and a distinct, autoimmune B-cell-like state. This trajectory is associated with sex-biased outcomes: female patients and mice exhibit reduced survival compared to males in our cohorts. Further analysis links this disparity to enhanced X-chromosome-linked expression and functionality of toll-like receptor signaling. We show that IRAK inhibition represents a potential sex-specific therapeutic strategy in preclinical models. These findings support a distinct developmental origin for BN2-DLBCL and identify a high-risk female population with actionable targets for precision therapy.

INTRODUCTION

Diffuse large B-cell lymphomas (DLBCL) has the highest incidence among non-Hodgkin’s lymphomas in adults(1). Current frontline treatment consists of anthracycline-based immunotherapy (e.g., R-CHOP, Pola-R-CHP), however, these regimens fail to achieve durable remissions in ~30–40% of patients(1). Disparity in outcomes can be partly attributed to disease heterogeneity. Defining molecular subtypes gave rise to the cell-of-origin (COO) classification(2), which categorized tumors based on their transcriptional proximity to either germinal center (GC) B-cell-like (GCB) and non-GC B-cell-like (non-GCB) DLBCL, the latter resembling activated B cells or an “unclassifiable” group(2,3). Still, COO-adapted therapies have largely failed to improve outcomes(4). Therefore, more recent efforts sought to subgroup DLBCL based on recurrent genetic alterations(3,5,6). Some of these groupings identified distinct biological dependencies and therapeutic vulnerabilities(7). Despite these advances, origin and trajectory of DLBCL remains poorly understood. Most DLBCL are believed to originate from GC-derived cells(8). GCs are transient immune structures where B cells clonally expand, diversify their antigen receptors (i.e., B cell receptor (BCR)) through somatic hypermutation (SHM), and compete for T-cell directed positive selection, ultimately giving rise to high-affinity antibody-secreting cells and immune memory(8). Many DLBCL retain phenotypic/epigenetic GCB features even after transformation, while GC-transit is inferred in others based on the presence of SHM(5). In this regard, a genetically-defined subset of DLBCL designated as “BN2” or “cluster 1 (C1)” lacks this mutational signature, suggesting the existence of an alternative, GC-independent, path of lymphomagenesis(5,9). BN2-DLBCL are genetically characterized by frequent BCL6 translocations and truncating mutations of NOTCH2 (N2^TRUNC). Notably, under physiological conditions, NOTCH2 activation shunts B cells away from the GC fate, and towards the marginal zone B cell (MZB) pool(10). MZBs are innate-like B cells that respond rapidly to antigen by differentiating into plasma cells (PC), and secreting low-affinity antibodies(11). BN2-DLBCLs also often harbor truncating mutations of the transcriptional repressor SPEN (S^TRUNC)(3,5,6). Even though mutations in SPEN are also found in other non-hematological malignancies(12), their role in malignant transformation remains obscure. Functionally, SPEN serves as a switch between transcriptional activation and repression, depending on context and co-factors bound(13,14). While it has been described as being crucial for X chromosome inactivation (XCI) (15,16), its main function in the context of B cells is thought to be linked to repression of NOTCH2 signaling and regulation of the MZB fate choice(17). In particular, SPEN and NOTCH2 compete for binding to the transcription factor RBP-J(17), antagonizing each other. Therefore, it has been proposed that S^TRUNC mutations phenocopy N2^TRUNC, by inducing similar downstream effects(3,17). The co-occurrence of these two mutations in BN2-DLBCL, however, could also indicate alternative routes for NOTCH2 signaling as it has been shown that SPEN KO led to formation of new RBP-J binding sites(18) suggesting rewiring of NOTCH2 signaling upon SPEN deletion. Therefore, we embarked on an effort to understand the role of SPEN mutations in malignant transformation.

RESULTS

S^TRUNC mutations with loss of function effects are enriched in DLBCL.

To define the spectrum of SPEN mutations in DLBCL, we integrated genetic profiles from 4,680 patients from 14 independent DLBCL patient cohorts(3,5,6,19–28, bioRxiv2023.11.21.567983), contrasting with SPEN mutations in a pan-cancer cohort(29) (n=24,950 patients). Our analysis showed that S^TRUNC were significantly enriched in DLBCL vs other cancers (X²=85.04; p=0.0001)(Fig. 1A). S^TRUNC mutations in DLBCL were found throughout the coding region without distinct hotspots. However, a common theme was the deletion of the SPOC (Spen paralog and ortholog C-terminal) repressor domain (Fig. 1B), one of the main functional domains of SPEN, suggesting a loss-of-function (LOF) phenotype. Accordingly, SPEN protein abundance in B cell non-Hodkin’s lymphoma (B-NHL) cell lines with endogenous S^TRUNC mutations(30) was significantly reduced compared to S^WT B-NHL cell lines (Fig. S1A). To exclude that S^TRUNC give rise to stable truncated protein variants, we engineered S^TRUNC in the DLBCL cell line U2932 (Fig. S1B). Mass-spectrometry did not detect a truncated SPEN protein, even when the peptide used to identify SPEN was located towards the N-terminus of the engineered mutation (Fig. S1C). Finally, studying the relative distribution of S^TRUNC in DLBCL across genetic subtypes, we found these alterations disproportionately enriched in BN2-DLBCL (~60% of S^TRUNC cases, p=0.0001, Fig. 1C). In contrast, non-truncating SPEN (S^NON-TRUNC) mutations were not enriched in any particular genetic subtype (Fig. 1C). Taken together, S^TRUNC mutations are enriched in BN2-DLBCL, and most likely produce a LOF phenotype.

Fig. 1– S^TRUNC mutations are associated with unfavorable outcomes among DLBCL patients.

A. Percentage of S^TRUNC and S^NON-TRUNC mutations in DLBCL (n=432/4,722 patients) and solid cancers (n=1,095/24,950 patients). Chi-squared test.

B. Lollipop plot of SPEN mutations in DLBCL (upper panel) and solid tumors (lower panel). Green and red boxes: RNA recognition motif; blue: SPOC domain.

C. Subtype-association of S^TRUNC and S^NON-TRUNC in DLBCL, excluding subtype-composite cases. S^TRUNC=70 in BN2 subtype; S^TRUNC=47 in other subtypes; S^NON-TRUNC=10 in BN2 subtype; S^NON-TRUNC=101 in other subtypes. Fisher’s exact test comparing S^TRUNC vs all other DLBCL.

D. Mutations which are either co-occurring with or mutually exclusive to S^TRUNC mutations using the cohort from(3). Odds ratio (OR) of mutations in >7.5% patients; OR > 2 co-occurring (blue), OR < 0.5 mutually exclusive (purple), OR 0.5–2 neither (grey/white).

E. Percentage of N2^TRUNC and N2^NON-TRUNC mutations in DLBCL (n=344/4,722 patients) and solid cancers(31) (n=605/24,950 patients). Chi-squared test.

F. Lollipop plot of NOTCH2 mutation in DLBCL (upper panel) and solid tumors (lower panel). Green and red boxes: EGF-like motif; blue box: Lin12/Notch repeat domain; yellow and purple: Notch protein domain; orange and pink: ankyrin repeat; burgundy: PEST domain.

G. OS of non-GCB DLBCL S^TRUNC compared to non-GCB DLBCL S^WT patients. Log-rank test.

H. OS of non-GCB DLBCL N2^TRUNC compared to non-GCB DLBCL N2^WT patients. Log-rank test.

I. OS of non-GCB DLBCL SN2^TRUNC compared to non-GCB DLBCL SN2^WT patients. Log-rank test.

J. Mean IPI scores of non BN2-DLBCL SN2^WT (ØBN2), BN2-DLBCL SN2^WT (BN2) and SN2^TRUNC (SN2) patients. Bar plot showing mean and standard deviation (SD). Kruskal-Wallis test with Dunn’s multiple comparison test.

K. Mean IPI subcategory ECOG scores of non BN2-DLBCL SN2^WT (ØBN2), BN2-DLBCL SN2^WT (BN2) and SN2^TRUNC (SN2) patients. Bar plot showing mean and standard deviation (SD). Fisher’s exact test corrected for multiple testing.

S^TRUNC mutations frequently co-occur with N2^TRUNC mutations in BN2-DLBCL.

To understand the genetic context of S^TRUNC mutations in BN2-DLBCL, we investigated co-occurring genetic lesions and found that NOTCH2 mutations where the most significantly associated genetic lesions (adj. p=0.02; Fig. 1D). N2^TRUNC were also significantly enriched in DLBCL vs other tumors (X²=219.82; p=0.0001, Fig. 1E), and mostly deleted the PEST domain (Fig. 1F), suggesting protein-stabilization and gain-of-function (GOF) effect (31,32). Similar to SPEN, N2^TRUNC mutations were highly enriched among BN2-DLBCL, whereas non-truncating lesions (N2^NON-TRUNC) were not enriched in any particular subtype (Fig. S1D).

S^TRUNC mutations are associated with unfavorable outcomes among DLBCL patients.

To gain insight into potential phenotypic effects of SPEN^TRUNC mutations, we explored their impact on clinical outcomes. In DLBCL, patients with a non-GCB disease experience inferior overall survival (OS), and BN2 tumors largely fall within this category(3). Focusing on non-GCB cases, S^TRUNC mutations showed inferior OS compared to S^WT cases (HR:1.61, CI95:1.20–2.17, p=0.001, Fig. 1G), which held true when extending to the entire cohort (p=0.02, Fig. S1E), whereas there was only a modest trend for S^NON-TRUNC cases (p=0.1, Fig. S1E).

Co-occurrence of S^TRUNC and N2^TRUNC mutations is linked to poor outcomes among DLBCL patients.

Examining N2^TRUNC cases among non-GCB DLBCL showed borderline significantly inferior OS (HR:1.23, CI95:1.00–1.75, p=0.052, Fig. 1H), which also reached significance when evaluating the entire DLBCL cohort (HR:1.30, CI95:1.04–1.61, p=0.02, Fig. S1F). Notably, non-GCB patients with SPEN^TRUNC and NOTCH2^TRUNC mutations (SN2^TRUNC) exhibited further inferior OS compared to WT than S^TRUNC and N2^TRUNC mutations alone (HR:3.13, CI95:1.79–5.26, p<0.0001, Fig. 1I). Inferior OS was still observed when including all DLBCL (HR:2.04, CI95:1.3–3.23, p=0.002, Fig. S1G), even after stratifying for R-CHOP-treated patients only (all DLBCL patients: HR:2.33, CI95:1.21–4.55, p=0.001, Fig. S1H and non-GCB only: HR:3.33, CI95:1.47–7.69, p=0.002, Fig. S1I). Notably, SN2^TRUNC-DLBCL also featured significantly inferior survival when compared to S^TRUNC or N2^TRUNC cases (Fig. S1J) (p=0.02 and p=0.005 respectively). Significantly inferior survival in the truncating mutation groups was retained even when including non-truncating mutations in the group of “other” DLBCLs (Fig. S1K-M).

BN2-DLBCL were initially associated with relatively favorable outcomes (3). However, a recent larger study showed BN2-DLBCLs had intermediate outcomes(33), in line with our analysis (Fig. S1N). Altogether, our data identified an aggressive course in DLBCL patients with S^TRUNC mutations, which is further exacerbated when accompanied by N2^TRUNC mutations.

Further probing these adverse survival findings, we observed that SN2^TRUNC-DLBCL patients were older (mean: 71.3yrs) compared to DLBCL patients with other mutations (mean: 62.3yrs), but not to those with isolated S^TRUNC (mean: 64.4yrs) or N2^TRUNC mutations (mean: 64.8yrs) (Fig. S1O). When restricting to BN2-DLBCL patients only, the age difference was more pronounced (SN2^TRUNC: 73.4yrs vs SN2^WT:60.4yrs, Fig. S1P). However, even when adjusting for higher age, the inferior outcomes observed between non-GCB SN2^TRUNC and non-GCB SN2^WT DLBCL remained highly significant (HR:2.70, CI95:1.28–5.88, p=0.009). Furthermore, SN2^TRUNC patients had higher ‘International prognostic index’ (IPI) scores compared to non-BN2-DLBCL but not compared to other BN2-DLBCL (Fig. 1J). As this score is compiled from age, high lactate dehydrogenase, ECOG performance status, stage and extranodal disease(34), we evaluated each of these factors individually. Among these, only ECOG performance status was significantly worse when comparing SN2^TRUNC against all DLBCL and showed a worse trend compared to BN2-DLBCL (Fig. 1K & S1Q). This is particularly relevant, as ECOG performance status is predictive of shortened OS in R-CHOP-treated DLBCL patients and is associated with increased risk of chemotherapy-related toxicities(35). Multivariate analysis for IPI risk indicated a trend for worse OS in SN2^TRUNC vs SN2^WT DLBCL patients (all DLBCL patients: HR:1.85, CI95:0.87–3.85, p=0.11; and non-GCB patients: HR:2.04, CI95:0.76–5.56, p=0.16). However, a significant proportion of inferior survival was attributable to high IPI scores in SN2^TRUNC patients. Interestingly, high IPI (>4) non-GCB SN2^TRUNC patients still showed a trend towards inferior OS compared to high IPI non-GCB SN2^WT patients (HR:2.78, CI95:0.86–9.1, p=0.08, Fig. S1R) Taken together, these data highlight the unfavorable clinical presentation of co-occurring SN2^TRUNC mutations in DLBCL patients and point to a pressing need for precision therapy approaches.

Spen^LOF does not enhance Notch2^GOF skewing of B cells towards MZB expansion.

To assess the biological functions of SPEN and NOTCH2 mutations, we utilized Spen conditional knockout mice (Spen^{tm1a(EUCOMM)Hmgu})(36), mimicking the expected LOF phenotype of SPEN^TRUNC mutations (Spen^LOF), and NOTCH2 intracellular domain (ICN2) knock in mice (Rosa26-Icn2^-IRES-YFP)(37), to represent the expected GOF phenotype of human NOTCH2^TRUNC mutations (Notch2^GOF). As these truncating mutations are mostly heterozygous in DLBCL patients, we focused our studies on Spen^fl/wt and Icn2^+/− animals, using a Cd19^Cre allele(38) to specifically induce these mutations in B cells (Fig. S2A&B). We chose this model so as to assess both GC and extra-follicular developmental trajectories, given that the cell of origin of BN2-DLBCLs are unknown. We generated the following cohorts of mice: Cd19^Cre/+(WT), Cd19^Cre/+;Spen^fl/wt (Spen^LOF;S), Cd19^Cre/+;Icn2^+/−(Notch2^GOF;N2), and Cd19^Cre/+;Icn2^+/−;Spen^fl/wt(Spen^LOF/Notch2^GOF;SN2). Profiling spleen tissue in ~2–3 months old mice yielded no significant differences in the total fraction of splenic B cell among S and SN2 mice, as compared to age- and sex-matched WT controls, although there was a slight reduction in N2 animals as previously described(39) (Fig. S2C). Consistent with prior reports, Spen homozygous deletion induced expansion of MZBs(17), however, this was substantially attenuated in Spen heterozygous mice (Fig. S2D). We and others have reported that conditional expression of ICN2 in B-cells skews B cell development to the MZB fate at the expense of the follicular B cell (FOB) fate(10,40). Our results herein recapitulate these effects (Fig. 2A). Notably, the addition of Spen^LOF to Notch2^GOF did not further enhance this MZB-skewing phenotype observed with Notch2^GOF alone.

Fig. 2– Spen^LOF and Notch2^GOF mutations skew mature B cells towards the MZB compartment.

A. Flow cytometry analysis of splenic non-germinal center B cells differentiating between MZB (CD21⁺CD23^-) and FOB (CD21^intCD23⁺) of 2–3 months old, sex-matched, unimmunized mice (8 replicates/group). Gating strategy as described (Supplementary Methods). Bar plot of MZB and FOB showing mean and SD. One-way ANOVA & Tukey’s multiple comparison test.

B. Schematic representation of B cell maturation from Immature B cells egressing the bone marrow, maturating into Transitional B cells (T1B, T2B and T3B) and/or Marginal Zone B (MZB) and Follicular B (FOB) cells.

C. Flow cytometry analysis of splenic Transitional B cells differentiating between T1B (IgM⁺CD23^-), T2B (IgM⁺CD23⁺) and T3B (IgM^-CD23⁺) of 2–3 months old, sex-matched, unimmunized mice (8 replicates/group). Bar plot of T1B, T2B and T3B showing mean and SD. One-way ANOVA & Tukey’s multiple comparison test.

D. Geometric mean fluorescent intensity (gMFI) for CD21 expression on T1B cells.

E. Flow cytometry analysis of splenic PC and PB for differentiating between PC (B220^-CD138⁺) and PB (B220⁺CD138⁺) of 6 months old, sex-matched, LPS (1mg/kg body weight) immunized mice (6–8 replicates/group), euthanized on day 8. Bar plot of PC and PB showing mean and SD. One-way ANOVA & Tukey’s multiple comparison test.

Representative figures. All experiments have been performed at least twice. One-way ANOVA & Tukey’s multiple comparison test.

To further dissect the observed shifts in cell fate among mature B cells, we examined the preceding transitional B (TB) phases. Physiologically, mature B cells enter lymphoid organs in a transitional state and undergo stepwise transitions from T1B to T2B to T3B cells. T1B, and to a lesser extent T2B cells, form MZB cells, whereas FOB arise mostly from T2B and T3B (Fig. 2B)(41,42). The abundance of total TB cells was comparable among genotypes (Fig. S2E). Further analysis revealed a significant increase in T1B cells and a reciprocal depletion of T2B and T3B cells across all three mutant genotypes, compared to WT (Fig. 2C). Interestingly, S mice showed a significant increase in T1B but only a moderate increase in MZB cells, suggesting that heterozygous loss of Spen primes but does not fully drive MZB cell commitment. Along these lines, T1B cells in N2 and SN2 animals, but not in S mice, exhibited substantially higher expression of CD21, a characteristic MZB phenotypic marker, suggesting a strong drive toward MZB fate (Fig. 2D).

Notch2^GOF but not Spen^LOF mutations block MZB-to-plasma cell differentiation.

Differentiation blockade is acknowledged as an important mechanism of lymphomagenesis(8). The default cell-fate of activated MZBs is differentiation into plasmablasts (PB) and plasma cells (PC). Under basal conditions, PB and PC to MZB ratios were reduced across all mutant genotypes, with progressively stronger effects in S, N2 and SN2 mice (Fig. S2F). MZB activation is primarily induced by innate immune signals like LPS, which triggers Toll-like receptors (TLR) and, ultimately, differentiation to PC(43). Thus, the mutational effects may only appear during activation. We therefore injected mice with LPS and sacrificed them eight days later, when most activated MZBs would have converted to PC(44). Under these conditions, there was an even more pronounced reduction in the PC/MZB ratio in both N2 and SN2 mice (Fig. 2E). In contrast, LPS treatment restored PB ratios close to normal in N and SN2 mice (Fig. 2E). This suggests a potent effect of N2 in blocking terminal PC differentiation in activated MZBs, while the effect was independent of SPEN functions. In contrast to our PC findings, MZB and FOB fractions remained unchanged compared to baseline conditions (Fig. S2G), suggesting that MZB commitment is already fully engaged in the presence of N2.

Combined Spen^LOF/Notch2^GOF mutations cooperate to drive an increase in aged/autoimmune B cells.

The relatively little additional impact of SPEN^LOF on NOTCH2^GOF with regard to MZB and PC differentiation prompted us to explore other relevant B cell subsets. Several lines of evidence suggest that aberrant memory B cells (MBC) serve as precursors for certain B-cell lymphomas(45–47). The abundance of the canonical MBC pool (B220⁺CD19⁺CD138^-CD38⁺FAS⁻CD21^intCD23⁺IgD⁻) was unchanged in N2 and SN2 animals, whereas S mice exhibited a reduction compared to WT controls (Fig. S3A). However, MBCs are a phenotypically and functionally diverse population(48). Among them, aged/autoimmune memory B cells (AiBC) have drawn interest for their pathogenic role in autoimmune disorders like systemic lupus erythematosus (SLE)(49). AiBCs were also proposed as putative COO for the aggressive subtypes of DLBCL(46,50,51). They express intermediate-to-high levels of the transcription factor T-BET (encoded by: TBX21), along with the surface integrin CD11c (encoded by: ITGAX)(52,53). Measuring the abundance of AiBCs (B220⁺CD19⁺CD138^-CD38⁺FAS^-CD21^-CD23^-T-bet⁺CD11c⁺) we observed a significant expansion in SN2 animals, a modest increase in N2 mice, and a non-significant change in S mice, compared to WT (Fig. 3A). Similar findings were obtained when defining AiBCs by CD11c⁺CD11b⁺ (Fig. S3B). Interestingly, this expansion was even more pronounced in aged SN2 animals (Fig. S3C). These data suggest a cooperative effect of Spen^LOF and Notch2^GOF mutations in skewing B cell fates away from canonical MBC and towards aberrant AiBCs.

Fig. 3– Combined SpenLOF and Notch2GOF mutations drive an increase in AiBCs.

A. Flow cytometry analysis gated on splenic DN cells gating for AiBCs (CD11c+T-bet+) in 2–3 months old, sex-matched, unimmunized mice (7–8 replicates/group). Gating strategy as described (Supplementary Methods). Bar plots of AiBCs as % of CD21-CD23- cells showing mean and SD.

B. Flow cytometry analysis of splenic mature B cells differentiating between spontaneous GCB (CD38^- FAS⁺) and non-GCB (CD38⁺FAS^-) in 2–3 months old, sex-matched, NP-KLH immunized mice (8 replicates/group). Euthanized after 8d. Bar plot of GCB as % B cells showing mean and SD.

C. Flow cytometry analysis on splenic DN cells gating for AiBCs (CD11c⁺T-bet⁺). Pool of two independent experiments with sex-matched, unimmunized mice (5–8 replicates/group). Bar plots of AiBCs as % of DN cells showing mean and SD.

D. BCR-Seq of flow sorted AiBC of WT and SN2 animals. Representative pie charts of clonal diversity (one color per clone) of one WT (left) and one SN2 (right) animal. Bar plot of size of the top 20 IgH-clones as % of all clones showing mean and SD. Two-sided, Student’s t-test.

Representative figures. All experiments have been performed at least twice. One-way ANOVA & Tukey’s multiple comparison test in Fig. 3A-C.

Spen^LOF/ Notch2^GOF AiBCs arise in a GC independent manner and show clonal expansion.

Notably, AiBC can arise from either GCBs or FOBs(49,50), a distinction highly relevant considering the assumed GC origin of most DLBCL, and the possible GC-independent origin of BN2-DLBCL(3). As ICN2 expression blocks GC development(43), we examined GC response after immunization with NP-KLH. At peak of the reaction (day 8), GCBs were significantly reduced in S mice, and nearly absent in N2 and SN2 animals (Fig. 3B). The loss of GCBs prompted us to ask whether SN2-induced AiBCs required GC transit. Even though AiBCs were detectable in our models without immunization (Fig. 3A), mice housed in barrier facilities develop spontaneous GCs(50). To address this, we crossed our mouse models to Bcl6 conditional knockout mice(54), which abrogates GC formation(55). We generated Cd19^Cre/+;Icn2^+/−;Spen^fl/wt;Bcl6^fl/fl(SN2B) and Cd19^Cre/+;Bcl6^fl/fl(WTB) animals, in which AiBCs should originate independently of the GC. Notably, SN2B animals were able to form AiBCs at the same level as SN2 (i.e., Bcl6^wt/wt) mice (Fig. 3C). Given that neither WTB nor SN2B generated spontaneous GCs (Fig. S3D-F) these findings point to an extra-GC origin for SN2 AiBCs. Finally, to test whether SN2 AiBCs could be plausible lymphoma clonal precursor cell, we performed BCR-Seq of WT and SN2 AiBCs. Notably, AiBCs showed significantly higher clonality in SN2 compared to WT mice (Fig. 3D). Collectively, these data provide experimental basis for indicating an extra-follicular origin of AiBCs as the basis for this subset of BN2-DLBCLs.

Combined Spen^LOF/Notch2^GOF mutations induce formation of non-canonical AiBCs with MZB features.

To more precisely track how our genotypes might reprogram cell fate decisions we sorted splenic AiBC, MZBs, and FOBs, and performed combined single-cell RNA-Seq (scRNA-seq) and ATAC-Seq (scATAC-seq)(Fig. 4A). Sorted B cell subsets were identified through mRNA expression of characteristic markers (Fig. S4A), further validated by projection of published MZB, FOB, and AiBC gene signatures onto the multiome UMAP(56–58) (Fig. 4B&C). By performing Louvain cluster analysis incorporating scRNA-seq and scATAC-seq nearest neighbors, 10 clusters were identified (C0-C9) (Fig. 4B, Supplementary Table 1). As there was little difference among genotypes in FOB or MZB populations, we focused on AiBC clusters (C0, C3, C7-C9). Cluster C0, similarly distributed among genotypes, most likely captured canonical AiBCs expressing higher levels of Tbx21/T-bet and Itgax/Cd11c (Fig. 4D-F). In contrast, SN2 cells were relatively depleted in C3, enriched in C9, and largely dominated clusters C7 and C8, the latter being nearly exclusively composed by SN2 cells (Fig. 4D&E). Of note, AiBC cluster C8 had the highest expression of Myc (Fig. 4F&S4B), small changes of which can confer significant fitness advantage to B cells(59). C8 also expressed genes linked to MBCs (e.g., Zbtb20, Fcrl5, Cd80), TLR signaling (e.g., Tlr4, Tlr8) and MZBs (e.g., Cd9, Dtx1) suggestive of phenotypic plasticity. C9 was likely composed of proliferating cells, given their strong cell cycle signature (Fig. 4C), while C7 showed upregulation of genes associated with early plasma cell differentiation (e.g., Tigit, Xbp1) (Fig. 4C&F).

Fig. 4– Combined SpenLOF and Notch2GOF mutations induce formation of a non-canonical AiBC with MZB features.

A. Schematic representation of experimental procedure: MZBs, FOBs and AiBCs were sorted from 5– 6 months old, female mice, unimmunized, 2 replicates/genotype.

B. UMAP of total 8 samples from all genotypes generated by Louvain cluster analysis incorporating both RNA-Seq and ATAC-Seq nearest neighbors.

C. Projection of a FOB, MZB, AiBC and proliferation signature onto the multiome UMAP. Z-score.

D. Cluster composition of AiBC clusters. Percentage of WT, S, N2 and SN2 cells.

E. Deconvoluted UMAP of Fig. 4B per WT, S, N2 and SN2 genotype.

F. Bubble plot of marker genes in AiBCs clusters.

G. DynaVelo analysis projecting cell velocity (see Methods) representative graph of one replicate per genotypes.

H. Experimental setup of stimulation of sorted mouse splenic FOB and MZB.

I. Flow cytometry analysis of splenic, sorted and stimulated cells gated on CD11c and T-bet for identification of AiBCs (CD11c⁺T-bet⁺) at d2. 5–6 months old, sex-matched, unimmunized mice (2 replicates/group with 3 technical replicates). Bar plots of AiBCs as % of live cells showing mean and SD. One-way ANOVA & Tukey’s multiple comparison test.

J. Experimental setup of stimulation of sorted human tonsillar FOB and MZB-like.

K. Flow cytometry analysis of cells gated on CD11c and T-bet for identification of AiBCs (CD11c⁺T- bet⁺) at d3. (2–3 biological replicates/group with 3 technical replicates). Bar plots of AiBCs as % of live cells showing mean and SD. One-way ANOVA & Tukey’s multiple comparison test.

Representative figures. AiBC stimulation experiments have been repeated at least 3 times.

To assess dynamic relationships among AiBC clusters, we used DynaVelo (bioRxiv 2025.04.24.650458 ), a neural ordinary differential equation (ODE)-based trajectory model integrating scRNA-seq and scATAC-seq. This deep learning approach learns a latent time for all cells using a ChromVar-based analysis and placed the origin (time 0) of AiBC trajectory within C9, defined by high proliferation, branching either into a plasmacytoid phenotype (C7) or towards a canonical AiBC (C3 -> C0) (Fig. 4G&S4C). This trajectory was not significantly perturbed in S animals. However, in N2 mice, major influx from C9 to C7 was suggestive of AiBCs initiating but not fully differentiating into PCs, a canonical cell fate for these cells. Most strikingly, SN2 cells formed a unique trajectory reflecting an aberrant cell fate decision towards C8 (Fig. 4G&S4C). Collectively, these data suggest that combined SN2 mutations uniquely result in formation of aberrant AiBC-like cells with less defined cell lineage determination.

Combined Spen^LOF/Notch2^GOF mutations facilitate and illustrate the existence of a MZB cell fate decision resulting into AiBC differentiation.

The expansion of MZBs and AiBCs in SN2 mice prompted us to ask whether AiBCs could arise from MZBs. To address this, we sorted SN2 or WT FOBs and MZBs, and exposed them to cytokines inducing AiBC formation(60) (Fig. 4H, Supplementary Methods). After 48h of treatment, ~80% of SN2 MZBs showed expression of AiBC markers T-bet and CD11c, while only ~30% of WT MZBs acquired this phenotype (Fig. 4I). A similar effect was observed among FOBs (Fig. 4I). This suggests that while a MZB-to-AiBC transition is accessible to WT cells, it is more kinetically favorable in the SN2 mutant setting (Fig. 4I&S4D). Unstimulated MZB and FOB did not show expression of AiBC-specific markers (Fig. S4E). Human MZB cells are less clearly defined phenotypically (61). Therefore, we sought to validate whether a similar MZB-to-AiBC trajectory would be applicable to humans. We sorted FOBs (CD20⁺CD19⁺CD38^-CD27^-IgD⁺CD21⁺CD11c^-) and MZB-like cells (CD20⁺CD19⁺CD38^-CD27⁺IgG^-IgA^-CD21⁺CD11c^-) from human tonsils and cultured them with AiBC-inducing cytokines (Fig. 4J, Supplementary Methods). Under these conditions, ~60% of FOB and MZB acquired an AiBC-like phenotype, as defined by upregulation of CD11C and T-BET (Fig. 4K), suggesting a comparable trajectory in human and murine B cells. A recent study identified two distinct human MZB populations: MZB1 & MZB2, the latter with a strong NOTCH2 signature and low SHM(62). Mapping these populations to the human scTonsil atlas(63) we observed that the MZB2 signature was enriched among MBCs expressing FCRL4 and FCRL5 (Fig. 5A-B&S5A), a known feature of human AiBC(64), while the MZB1 signature was not enriched in any particular cluster (Fig. S5B&C). Indeed, differential gene expression analysis between the MZB2-high cluster and all other MBCs showed upregulation of canonical AiBC-associated genes (e.g., ITGAX, ITGAM, TBX21), in addition to MBC genes (e.g., CD1C, CCR5) (Fig. 5C). Taken together, these data suggest that human and murine MZBs are capable of differentiating into AiBCs, a trajectory that is specifically enhanced in the presence of SN2 genotypes.

Fig. 5– Combined SpenLOF and Notch2GOF mutations lead to increased fitness of female AiBC.

A. UMAP of non-germinal center B cells from the scTonsil atlas(68) – NBC/MBC spectrum. Color scheme representing different clusters defined by scRNA-seq. Abbreviations: naive B cell (NBC); early activated naive B cell (NBC early activation); naive B cells activated by interferon (NBC IFN-activated); CD229 expressing naive B cells (NBC CD229+); naive B cells prone for GC-entry expressing CD69 (Early GC- committed NBC); naive B cells committed to GC entry expressing CCND2 (GC-committed NBC); early GCB expressing MEF2B but not BCL6 or AICDA (pre-GC); proliferative naive B cell (proliferative NBC); non-proliferative GC with Dark-Zone phenotype (GC DZ nonproliferating); memory B cell expressing CD27 and MEF2B (Early MBC); non-class switched memory B cell expressing high IgD and IgM (ncsMBC); non-class switched memory B cell expressing FCRL4 and 5 (ncsMBC FCRL4/5+); class switched memory B cell (csMBC); class switched memory B cell expressing FCRL4 and 5 (csMBC FCRL4/5⁺); and memory B cells expressing FCRL5 but not FCRL4 (MBC FCRL5⁺).

B. UMAP of non-germinal center B cells from the scTonsil atlas with projection of MZB2(69) signature.

C. Differentially expressed genes (DEG) of MZB2-enriching cells in the NBC/MBC spectrum compared to cells not enriching in MZB2 signature. pAdj < 0.01. Wilcoxon rank sum test.

D. Flow cytometry analysis of splenic AiBCs gated on KI-67 and T-bet for identification of proliferating cells in 2–3 months old, sex-matched, unimmunized mice (7–8 replicates/group, two independent experiments). Bar plots of KI-67+ as % of AiBCs showing mean and SD. Two-sided, student’s t-test, corrected for multiple testing. Pool of two independent experiments.

E. Experimental set up for competitive male:female BMT in WT and SN2 mice, with male, female and female ovariectomized recipients.

F. Flow cytometry analysis of splenic AiBCs gated on CD45.1 and CD45.2 for identification of male (CD451/2) and female (CD452/2) in 2–3 months post BMT, unimmunized mice (4 replicates/group). Bar plots of female AiBC/B cell ratios showing mean and SD. Two-sided, Student’s t-test, corrected for multiple testing.

Combined Spen^LOF/Notch2^GOF mutations lead to selectively increased fitness of female but not male AiBCs.

While analyzing the tonsil Atlas, we observed that MZB2/AiBC-enriching cells tended to be more abundant in females than males(Fig. S5D). While relative abundance of AiBC in our mouse model did not differ among sexes (Fig. S5E), absolute numbers of AiBCs were significantly higher in female SN2 animals (Fig. S5F), in line with higher Ki67 expression in female AiBCs of SN2 mice, vs. the other genotypes (Fig. 5D). In contrast there was no difference in KI67 expression in MZBs between genotypes (Fig. S5G) further underlining the relevance of these AiBCs. To examine if this reflects greater cell fitness, we performed mixed chimera bone marrow transplantation (BMT) assays, comparing female vs. male cells transplanted at 1:1 ratio in WT or SN2 scenarios. As recipients, we used lethally irradiated male, female, or ovariectomized female mice (Fig. 5E), to exclude hormonal influences. Following engraftment, spleens were collected and analyzed by flow cytometry. AiBC abundance was normalized to the total B cell population derived from each donor in a given animal. Strikingly, female SN2 AiBCs displayed a clear competitive advantage over male SN2 AiBCs, an effect not observed in WT animals (Fig. 5F&S5H). This suggests that AiBCs manifested a significant competitive advantage, and hence greater fitness, over male SN2 cells in the same animal, regardless of the sex or hormone status of the host (Fig. 5F&S5H). SN2 chimeras showed increased proportions of male derived cells across lineages, perhaps reflecting an initial imbalance in transplanted cell proportions (Fig. S5I&J). Taken together, these data suggest SN2 mutations exert sex-biased cell-autonomous effects.

Spen^LOF/ Notch2^GOF mutant lymphomas are more aggressive in females.

To assess whether these sex-based differences had an impact on lymphomagenesis, we established a cohort of sex-balanced WT, S, N2 and SN2 mice, and tracked their survival. Unlike N2 and SN2 animals, S mice did not develop lymphomas in the studied timeframe (Fig. 6A&S6A). Strikingly, female SN2 mice showed significantly shortened median OS (mOS) compared to male SN2 mice (mOS female SN2 316d vs. male SN2 NR; p=0.001, Fig. 6A). This sex-difference was not observed among N2 animals (mOS female N2 vs. male N2: 258d vs. 275d; p=0.19, Fig. S6A). In line with these observations, R-CHOP-treated human female vs male SN2 patients with BN2-DLBCL manifested significantly inferior survival, which was not observed among N2 cases (SN2 female vs male: HR: 4.72; CI95: 0.99–22.4, p=0.04; N2 female vs male: HR: 1.49; CI95: 0.61–3.7, p=0.38, Fig. 6B&S6B). These data support that SN2 (but not N2) BN2-DLBCL are more lethal in females than males.

Fig. 6– Lymphomas with combined SpenLOF and Notch2GOF mutations exhibit an AiBC-like phenotype, which is attenuated in males.

A. Overall survival (all-cause mortality) of male (purple) and female (pink) mice of SN2 genotype compared to WT (black) of all-cause mortality. 21–22 mice/genotype, unimmunized. Log-rank Test.

B. Overall survival of male (purple, n=6) and female (pink, n=5) human SN2 DLBCL patients within the BN2 subtype, treated with R-CHOP. Log-rank Test.

C. H&E staining of lymphoma-involved mouse spleen, showing small cells (indolent) and large cells (aggressive). 40x magnification.

D. H&E staining of lymphoma-involved mouse lung. 40x magnification. Female specimen (left) and male specimen (right).

E. Flow cytometry analysis of splenic non-germinal center B cells gated on CD21 and CD23 for identification of DN cells in sex-matched, not age matched due to lymphoma-endpoint. 4–6 mice/group. Bar plots of DN cells as % of non-GCB showing mean and SD. Two-sided, Student’s t-test.

F. Flow cytometry analysis of splenic DN cells gated on CD11c and T-bet for identification of AiBCs(CD11c⁺T-bet⁺) in sex-matched, not age matched due to lymphoma-endpoint. 4–6 mice/group. Bar plots of AiBCs as % of B cells showing mean and SD. Two-sided, Student’s t-test.

G. Heatmap comparing expression of AiBC-specific genes determined by RNA-Seq of WT MZB andmale and female SN2 lymphomas enriched for CD19+ cells.

H. Gene Set Enrichment Analysis in SN2 lymphomas using an AiBC-lymphoma signature(54).

I. ScRNA-Seq analysis of a human SN2-DLBCL with expression of the AiBC-lymphoma signature (54) enriched in lymphoma cells (yellow/green). Violin plots of mean enrichment score, Wilcoxon-Mann-Withney test.

J. ScRNA-Seq analysis of a human SN2-DLBCL with expression of the SN2 lymphoma signature (this paper) enriched in lymphoma cells (yellow/green). Violin plots of mean enrichment score, Wilcoxon-Mann-Withney test.

K. Scatterplot showing higher expression of AiBC genes skewed to female (pink) vs male (purple) SN2 lymphomas.

L. MSigDB/STRING-DB pathway analysis using gene overlap calculations including genes of the AiBC signature showing higher expression in females over males. Full pathway terms can be found in Supplementary Table 3.

Spen^LOF/ Notch2^GOF mutant lymphomas exhibit an AiBC-like phenotype, which is more evident in females.

Histological evaluation of murine SN2 and N2 lymphomas showed destruction of splenic architecture by malignant B cells (Fig. 6C&S6C). Individually, different proportions of indolent and aggressive lymphoma features were present, resembling human marginal zone lymphoma (MZL) and DLBCL (Fig. 6C&S6C). Some animals showed significant pulmonary infiltration, with destruction of tissue architecture and intravascular occlusion (Fig. 6D&S6C). Immunophenotyping showed that lymphoma cells from SN2 and N2 tumors exhibited a CD21⁻CD23⁻ phenotype (Fig. 6E&S6D), which has been shown to encompass the AiBC population(51). Notably, approximately half of SN2 DLBCL cells displayed a T-bet⁺CD11c⁺ AiBC-like phenotype (Fig. 6F), whereas, AiBC-like cells comprised only ~20% of N2-DLBCL cells (Fig. S6E). Overall, CD21⁻CD23⁻ cells expressed significantly higher levels of T-bet in SN2 lymphomas (Fig. S6F), while CD11c was expressed at similar levels between SN2 and N2 lymphomas (Fig. S6G).

To gain further insights into the biology of these lymphomas, we conducted profiling of SN2 tumors by RNA-Seq, using WT MZBs as a reference population. In line with our flow cytometry data, SN2 lymphomas showed increased expression of genes commonly found in AiBCs (e.g. Irf4, Fcrl5) (Fig. 6G) and significant enrichment for a recently published AiBC-derived lymphoma signature(51) (NES: 1.32; FDR: 0.003, Fig. 6H). In contrast, SN2 murine lymphomas were not enriched for MZB signatures(10), although a subset of MZB-associated genes (e.g., Cd36, S1pr3) where still expressed (Fig. S6H-I). These findings suggest that these tumors retain plasticity reflective of their putative cells of origin with hybrid AiBC/MZB phenotypes.

Even though it is often challenging to map signatures from engineered murine DLBCLs to human patients, we observed an enrichment of the murine SN2 signature among human BN2-DLBCL (SN2: p=0.06, Fig. S6J). To validate this further, we performed scRNA-seq in a human SN2^TRUNC-DLBCL. Strikingly, lymphoma cells showed an increased mean enrichment score for the aforementioned AiBC-signature (Fig. 6I) compared to physiological B cell populations (p<0.001). Furthermore, the lymphoma cell populations also enriched for our SN2 mouse lymphoma signature (p<0.001, Fig. 6J), thus validating the fidelity of our mouse model. The striking sex-specific difference in OS prompted us to assess whether male and female SN2 lymphomas differ in enrichment for the AiBC phenotype. Indeed, comparing the top 500 genes from the AiBC lymphoma signature as well as genes reported to be relevant for AiBC development (e.g., Itgam, Itgax, Supplemental Table 2) revealed a subset of these genes were expressed higher in female SN2 lymphomas (Fig. 6K). We observed stronger and more significant enrichment for AiBC-related gene networks among transcripts more highly expressed in females (2.31, p=4.4e⁻¹⁶) compared to those more highly expressed in males (0.46, p=0.042). These included pathways such as Toll-like-receptor signaling and other AiBC signatures (Fig. 6L). MYC- and BCR-signaling pathway genes were expressed at similar levels in females and males including transcripts relevant for cell cycle (e.g., Cdk4, Npm1 and Ywhae), metabolism (e.g. Erh, Srm and Nme1) and RNA-binding processes (e.g. Apex1). (Fig. S6K, Supplemental Table 3).

Previous studies showed that much of the transforming activity of mutant NOTCH is due to induction of MYC expression(65). We therefore tested, if SN2, N2 lymphomas and human BN2-DLBCL showed enrichment for NOTCH signatures, which was not the case(Fig. S6L), consistent with a recent report suggesting absence of active NOTCH signaling in BN2-DLBCL(66). Collectively, these data suggest that whereas SN2 lymphomas generally express BCR- and MYC-associated pathway genes, female SN2 lymphomas exhibit an additional sex-specific signature related to TLR-signaling.

Spen^LOF/ Notch2^GOF mutant lymphomas show altered X-linked gene expression regulation with TLR7 overactivation.

TLR7, an X-linked gene, plays a central role in AiBC activation and mediates pathogenic effects of these cells in auto-immune diseases(49). Indeed, female murine SN2 lymphomas showed significantly increased levels of TLR7 protein (Fig. 7A), compared to their male counterparts, a trend which we also observed in female SN2-DLBCL, vs males (Fig. S7A). In contrast Tlr7 expression did not show female sex-bias in N2 murine lymphomas (Fig. S7B). TLR7 on the inactive X (Xi) can become dysregulated after SPEN loss due to impaired XCI(15). As Xi promoters are CpG-methylated, we investigated methylation in female SN2 lymphomas. Using SMRT-Seq in CD19⁺ cells from female N2 and SN2 murine lymphomas, we observed lower XChr promoter methylation in female SN2 over N2 lymphomas, whereas the opposite was observed in autosomes (Fig. 7B). Promoter hypomethylation was specifically observed at the TLR7 locus in SN2 but not N2 lymphoma cells (Fig. 7C). Disrupted XCI is often associated with dispersed Xist RNA to the Xi(67–69). Performing RNA FISH in LPS-stimulated MZBs of WT and SN2 mice to image localization of Xist(67,68), did not reveal significant differences in Xist RNA localization to the Xi. (Fig. S7C-E). As SPEN recruits PRC2 for deposition of H3K27me3 on the Xi, we also investigated H3K27me3 enrichment on the Xi. Again, we did not observe significant changes of co-localization of H3K27me3 with Xist on Xi (Fig. S7C-E). These data suggest that the combination of heterozygous Spen^LOF with Notch2^GOF disrupts the regulatory state of X-linked promoters without affecting global XCI maintenance.

Fig. 7– Combined SpenLOF and Notch2GOF mutant lymphomas show signs of XChr dysregulation and TLR pathway dependencies.

A. Contour flow plot of CD21^-CD23^- (DN) cells gated on TLR7 expression. 2–6 mice/group, sex- matched, not age matched due to lymphoma-endpoint. Bar plots showing female/male ratio of TLR7+ cells as % of DN cells showing mean and SD. Two-sided, Student’s t-test.

B. SMART-Seq of CD19+ sorted mouse female N2 (n=4) and SN2 (n=2) lymphoma samples compared to WT (n=4) B cells. Methylation difference of the Translation Start Site (TSS) +/−500bp stratified for the XChr and autosomes.

C. Average methylation of CpGs within the Tlr7 gene locus from analysis performed in Fig. 7B.

D. Flow cytometry analysis of TLR7 levels in the SN2F and SN2M cell line. Bar plots of gMFI for TLR7- PE showing mean and SD. Two-sided, Student’s t-test.

E. Phospho-flow cytometry analysis of SN2F and SN2M for pNFKB. Bar plots of gMFI for pNFkB showing mean and SD. Two-sided, Student’s t-test.

F. SN2F and SN2M were incubated IRAK1/4 inhibition AZ1495 over 4 days and growth inhibition was measured by CellTiter-Glo assay. Drug was reapplied once after 48h. Two-way ANOVA.

G. SN2F was transplanted s. c. into NSG mice. After engraftment mice were treated with AZ1495 for 7 days with tumor measurement by caliper at timepoints indicated in the diagram. Outcome was reported as per mouse area under the curve (AUC) with subsequent Mann-Whitney test.

H. Flow cytometry analysis of cleaved caspase 3 (CC3) in specimen from Fig. 7G. Two-sided, Student’s t-test.

I. A female SN2 PDX was transplanted into NSG recipients. After engraftment mice were treated with AZ1495 for 21 days with tumor measurement by caliper at timepoints indicated in the diagram. Outcome was reported as per mouse area under the curve (AUC) with subsequent Mann-Whitney test.

J. Median time to progression (mTTP) defined as ≥2× baseline tumor volume; animals not reaching endpoint were censored at last measurement. Log-rank test.

Spen^LOF/ Notch2^GOF mutant lymphomas show sex-specific vulnerabilities

Preferential dysregulation of TLR-associated programs in female vs. male SN2 DLBCLs prompted us to test whether there was also functional evidence for increased TLR signaling. To enable ex vivo studies and overcome asynchronous lymphomagenesis in autochthonous mice, we generated a transplantable lymphoma model(70). We subjected cells from spontaneous tumors of female SN2 or male SN2 donors to sequential passages in Rag1^KO recipient mice (Fig. S7F). This yielded transplantable female (SN2F) and male (SN2M) lymphoma cell lines, also amenable to ex vivo culture. Karyotyping verified the sex of the lines, though the predominant clone in the SN2F line showed XChr monosomy (Fig. S7G), again pointing towards an XCI-independent mechanism. Similar to what was observed in primary SN2 tumors, SN2F cells showed high expression of TLR7 (Fig. 7D) but also of its downstream mediator IRAK1 (Fig. S7H). TLR signaling induces NFkB-activity, associated with degradation of IkBa and phosphorylation of p65-NFkB. Notably, SN2F cells showed higher basal p65 phosphorylation than SN2M, although their IkBa levels were similar (Fig. 7E&S7I). Notably, stimulating SN2F and SN2M lymphoma cells with the TLR7 agonist R848 led to significantly greater reduction in IkBa and increased phosphorylation levels of p65 in female vs male cells (Fig. 7E&S7I). Transcriptional footprinting analysis based on ATAC-seq of primary mouse SN2 lymphomas further showed increased NFκB2 and RelA binding in females versus males (Fig. S7J), which was even more evident when comparing to sex matched WT B cell controls (Fig. S7K). Increased activity of NFκB2 and other related transcription factors e.g. IRF4 (Fig. S7L) was confirmed through a regulatory potential analysis in our ATAC-seq data, further supporting a higher relevance of NFkB signaling in female SN2 lymphomas.

As this might represent a targetable vulnerability, we treated SN2M and SN2F cells with AZ1495, an IRAK1/4 inhibitor that acts downstream of TLR, which led to significantly greater growth inhibition in SN2F cells across concentrations (p=0.0005, Fig. 7F). Validating these findings in vivo, we transplanted SN2F or SN2M cells into sex-matched immunodeficient mice and treated them with AZ1495 after engraftment. Again, only tumors derived from SN2F cells showed a significant reduction in volume in response to IRAK1/4 inhibition (Fig. 7G&S7M). This effect was partly due to increased apoptosis as demonstrated by higher levels of cleaved-caspase 3 (Fig. 7H). To cross validate these findings in primary human DLBCL cells, we obtained a female SN2 PDX model which, similar to our murine models, also showed X monosomy (Fig. S7N). Tumor cells were transplanted into female immunodeficient recipient mice and treated with AZ1495 for 21 days, as per a published protocol(71). Tumor burden was significantly lower in AZ1495-treated mice compared to their vehicle-treated counterparts (Fig. 7I), resulting in a nearly two-fold increase of median time to progression (mTTP) (Fig. 7J). Moreover, AZ1495 enhanced the effect of doxorubicin against these lymphomas (Fig. S7O), which is important given their resistant phenotype and limitations in administering full dose of chemotherapy to ECOG-high patients as is often the case in SN2-DLBCL. Taken together, these data point to a sex-specific vulnerability of SN2 lymphomas, based on TLR-signaling inhibition (Fig. S7P). This scenario raises the possibility of developing sex-based precision therapy for female DLBCL patients with these co-occurring mutations and, otherwise, highly unfavorable clinical outcomes.

DISCUSSION

Co-occurring mutations can co-operate in several ways. It was known, that S^TRUNC and N2^TRUNC mutations skew B cells towards a MZB fate(10,17), however, the addition of SPEN^LOF did not further enhance this “classical” NOTCH2 effect. Hence, the combination of both is most likely leading to the creation of a new SN2^TRUNC co-dependent pathway. Indeed, our data suggests that SPEN and NOTCH2 may have slightly different roles in driving this process, since SPEN^LOF caused expansion of T1B cells, which give rise to MZBs, to a similar extent as NOTCH2^GOF, but with almost no impact on MZB expansion. We speculate that this points to a more nuanced picture whereby SPEN may help to prime the binding distribution of NOTCH2 towards genes specifically required for MZB cell formation. Consistent with this notion, homozygous SPEN loss led to new RBP-J binding sites in HeLa cells(18).

The specific expansion of AiBCs is notable, given their emerging role as putative cell-of-origin in DLBCLs(50). AiBCs possess remarkable plasticity, transitioning into GCBs, PC or MBCs, and highly reactive to inflammatory stimuli. Recent studies have highlighted how mutations in MYD88 or TBL1XR1, despite their disparate roles, both facilitate a switch from GCBs to an AiBC phenotype(46,50). Additionally, AiBCs naturally accumulate with age and have been implicated in age-associated lymphomas(51). Our findings describe a distinct trajectory wherein SPEN^TRUNC/NOTCH2^TRUNC mutations drive the expansion of a different subset of AiBCs, potentially originating from MZBs, diverging from the FOB-derived pathways observed in prior models. These mutant AiBCs exhibit a hybrid phenotype, co-expressing MZB and AiBC markers, and appear distinct from classical GC-origin AiBCs. Our data support a model of AiBC formation from MZB precursors (72). The heterogeneity of AiBCs, whether arising in aging, infection (e.g., HIV), or autoimmunity (e.g., SLE), suggests convergent differentiation from multiple B cell lineages(73). We demonstrate that both FOB and MZB can give rise to AiBCs, with MZB-derived conversion accelerated and distorted by SPEN^TRUNC/NOTCH2^TRUNC mutations. GC-directed Cre/loxP systems also were reported to skew B cells harboring N2^TRUNC mutations towards MZBs (43), supporting that extra-follicular pathway bias is not an artifact of CD19-directed recombination.

The identification of an SN2-specific aberrant AiBC trajectory culminating in a transcriptionally distinct population provides mechanistic insights into cellular plasticity in lymphomagenesis. That these cells show a composite expression profile encompassing MZB, MBC, AiBC, and plasmacytoid features suggests profound phenotypic plasticity, a trait increasingly recognized as a facilitator of oncogenic adaptability(50). Such plasticity may reflect an underlying failure of lineage commitment, positioning these cells at a poised and unstable juncture, susceptible to malignant transformation(72,74).

It is notable that we did not observe evidence of enhanced canonical or non-canonical NOTCH2 signaling; in fact, NOTCH2 activity appeared suppressed, which is in line with recent reports(66). This paradox may reflect either (1) a transient requirement for NOTCH2 during the initial phases of transformation or (2) the operation of non-canonical, NOTCH-independent RBP-J-mediated signaling pathways(75). Given that RBP-J regulates key genes involved in B cell activation, this shift may reflect an epigenetic imprint that sustains clonal expansion despite the absence of upstream NOTCH2 activity. Interestingly, the aberrant AiBC cluster observed in SN2-mutant mice displayed elevated MYC expression, a critical effector downstream of NOTCH signaling that is thought to be necessary, and potentially sufficient, for its oncogenic effects(65). Together with the aforementioned lineage ambiguity, this enhanced fitness supports a model in which this SN2-specific population represents a candidate for a clonal precursor cell.

As noted earlier, BN2-DLBCLs have been proposed to represent a transformed form of MZL. This hypothesis is supported by the shared mutational profiles of both entities, most notably heterozygous truncating mutations in NOTCH2 and SPEN (76). Additionally, both lack a strong AICDA-associated somatic hypermutation signature, suggesting that these DLBCL may arise uniquely from a non-GC origin(5,9). Supporting this extra-follicular origin, we found that the expansion of aberrant, clonally enriched AiBCs occurred independently of GC formation. Therefore, our findings support an MZL-like ontogeny for BN2-DLBCL driven by SPEN^TRUNC and NOTCH2^TRUNC mutations subsequently resulting in an AiBC-like phenotype. We are unable to determine whether the aggressive lymphomas observed in our mouse models truly reflect MZL transformation as the presence of both indolent and aggressive components in anatomical proximity also raises the possibility of a parallel, rather than linear, evolutionary trajectory. While documented transformation rate of MZL in humans is relatively low, about 5% over 10 years(77), SN2 DLBCL patients tend to be significantly older than those with other DLBCL subtypes. This may reflect a prolonged latency period required for MZL transformation. Furthermore, due to their indolent nature, MZL may be underdiagnosed(78), leading to an antecedent MZL may never have been clinically recognized until DLBCL diagnosis. Future studies will be needed to determine whether other BN2-DLBCL-associated oncogenes can induce similar phenotypes, which would help clarify the generalizability of this model. In this context, it is noteworthy that lymphomas arising in mice with a Bcl6 gain-of-function allele, mimicking BCL6 translocations seen in BN2-DLBCL, develop an MZL-like phenotype when crossed with Aicda knockout mice, failing to produce GC-derived lymphomas(79).

One of the most striking aspects of our findings is the sex-dependent phenotype conferred by SN2 mutations. This phenotype appears to arise from disease originating in aberrant AiBCs, which are more abundant in females, a trait associated with their heightened susceptibility to autoimmune disorders. Given the fact that diseases as SLE have been linked to changes in X-linked gene expression, e.g., TLR7(17) and considering SPEN’s role in transcriptional repression of the Xi through XCI(17,18), it is plausible that a dose reduction of SPEN might lead to XChr dysregulation. Indeed, methylation of the XChr was significantly impaired in SN2 lymphomas, including TLR7 promoter hypomethylation which was associated with higher expression. This is in line with reports of patients with germline SPEN haploinsufficiency, which also showed an altered, XChr-specific methylation signature(80). However, global XIST localization to the Xi was not significantly affected in SN2 MZBs which suggests XChr dysregulation without overt XCI failure. It should be noted that allele-specific silencing could not be assessed in C57BL/6J mice used herein, due to their inbred status.

Although estrogen signaling has been associated with more robust B-cell immune responses in females(81), sex hormones are unlikely to explain our observations, as the female-specific fitness advantage persisted in ovariectomized mice.

As female lymphoma cells showed higher response to TLR7 agonists, we hypothesized that this might indicate a targetable vulnerability. However, as selective TLR7 inhibitors are still under development(82), we targeted the downstream kinases IRAK1/4, for which inhibitors already reached clinical trials(83). Our syngeneic female SN2 mutant lymphoma cell line was more vulnerable to IRAK1/4 inhibition than their male counterparts both ex vivo and in vivo, further validated by a female human SN2 mutant DLBCL PDX model. Intriguingly, both the human PDX and mouse models showed X monosomy, which is often found in cancer, but counterintuitive in this context(84) as one might expect lower dosage of TLR7 and subsequently less vulnerability towards TLR signaling inhibition. However, it is known that patients with Turner-syndrome, harboring a XO karyotype, are prone to develop autoimmune diseases driven by TLR7, like primary Sjogren’s Syndrome(85). Therefore, it is possible that the female SN2 lymphoma phenotype is driven by XChr dysregulation and subsequently persists through other transcriptional activation mechanisms, maintaining addiction to TLR signaling and hence vulnerability to TLR inhibition independently of XCI and XChr copy number. Exactly how this happens remains unclear and points to the need for additional research exploring such mechanisms in the future. Our finding of sex-dependent outcomes in DLBCL may warrant consideration in clinical trial design. Even though progress has been made in sex-equity in terms of trial inclusion, especially in the cancer field, women are still often underrepresented within study populations(86). These data highlight the importance of incorporating biological sex as a variable in future trial design. Overall, these findings suggest TLR inhibition may represent a sex-dependent vulnerability, which might reduce the need for high dose chemotherapy with its associated severe toxicities.

Materials & Methods

Patient survival and mutational data analysis

The following studies, all of which performed reported SPEN and NOTCH2 mutation status based on DNA-sequencing, were included in the analysis: HMRN(6), BCCA(24), NCI(3), DUKE(bioRxiv 2023.11.21.567983), DFCI(5), CHIBA(26), CHICAGO(25) BELLINZONA(22), GUANGZHOU(27), SHANXI(28), HOUSTON(19), BEJING(23), MSKCC(20). Overall survival was converted into months where necessary with log-rank test for estimation of survival differences. Due to sparsening of sample numbers over time, a the analysis was censored at 60 months(6,25). Lollipop plots were generated using the MutationMapper tool on cBioPortal(87). Even though patients have been recorded to receive R-CHOP-like regimens, we were not able to distinguish if this included less intensive treatments like R-CVP or Rmini-CHOP.

Media & Buffers

Mouse cell culture media: RPMI 1640 with L-glutamine (VWR 45000–396) supplemented with 1x GlutaMAX (Life Technologies, #35050–061), 10% fetal bovine serum, 10μM HEPES (Life Technologies, #15630–080), 1x NEAA (Life Technologies, #11140–050), 1μM Sodium Pyruvate (Life Technologies, #11360–070), 50,000 units of Penicillin-Streptomycin, and 33μM 2-Mercaptoethanol (Life Technologies, #21985–023).

Mouse Models

The conditional Spen^fl/fl (Spen^{tm1a(EUCOMM)Hmgu}) mouse model was developed by the EUCOMM consortium Cryopreserved embryos were purchased and derived at the at Mouse Genetics Core Facility Memorial Sloan Kettering Cancer Center and crossed with B6.Cg-Tg(ACTFLPe)9205Dym/J (stock 005703) for Flp-recombination. The conditional Notch2-overexpressing Icn2^+/+ (ROSA26-ICN2^-IRES-YFP) was kindly provided by Prof. Iannis Aifantis. The conditional Bcl6^fl/fl model was described in a previous study from our lab(37,88). Other strains included in the study were purchased from The Jackson Laboratory: C57BL/6J (stock 000664), B6.SJL-PtprcaPepcb/Boy (CD45.1, stock 002014), B6.129P2(C)-Cd19tm1(cre)Cgn/J (CD19^Cre, stock 006785), B6.129S7-Rag1tm1Mom/J (Rag1^KO, stock 002216), NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ (NSG, stock 005557), ovariectomized C57BL/6J (stock 000664). Use of littermate controls will be prioritized, but sex/age-matched controls from the same colony may be used as the second-best option. All animal work was approved by the Cornell Institutional Animal Care and Use Committee (IACUC).

Bone marrow transplantation

Recipient mice were lethally irradiated with a total of 900 rad (RS 2000 Biological Research X-ray Irradiator, Rad Source Technologies) in two equal doses (2× 450rad) on two consecutive days, the second day being the day of injection of donor bone marrow cells. Bone marrow cells from the indicated genotype donors were isolated from tibia and femurs, by flushing the bones with mouse cell culture media. Red cell lysis was performed using RBC lysis buffer (Qiagen, #158106). Cells were washed three times with PBS and injected i. v. at a density of 1 × 10⁶ cells/100μL per recipient. Mice used for survival analysis were set up as a congruent BMT cohort. Additional tumors arising in our colony were used for subsequent studies but were not included in survival analyses to avoid batch effects.

Immunizations

For the induction of GC, 2–3 months old sex- and age-matched mice were immunized i. p. with 4-Hydroxy-3-nitrophenylacetyl (NP)-keyhole limpet hemocyanin (KLH) (80μg/recipient, Thermo Fisher Scientific #NC1125442) in alum (Thermo Fisher Scientific, #77161) 1:1. For stimulation of MZB we immunized with LPS 1mg/kg body weight. For both, mice were euthanized on day 8 after immunization for flow cytometric analysis.

Generation of murine lymphoma cell lines

The generation of primary mouse lymphoma cell lines has been described elsewhere.(70) In brief, primary mouse lymphoma cells were harvested under sterile conditions. 1 × 10⁷ cells were injected i. p. into Rag1^KO recipients, which were then monitored closely for disease onset. Spleens were harvested and dissociated into single cell suspensions, from which 1 × 10⁷ cells were injected i. p. into a second Rag1^KO recipient. This process was repeated for various rounds, until cells showed capacity to survive and grow in vitro.

Growth inhibition assay of mouse lymphoma cell lines

Lymphoma cells were diluted to 2.22 × 10 ^ 5 cells/mL with medium, so that loading 45 μL contained 10,000 cells. Drug dilutions were prepared by serial titration. Working concentrations for AZ1495 (MedChemExpress, #HY-111101) were referenced from Scott et al.(71), reconstituted in DMSO, which was used as vehicle control. 5 μL of the respective drug was loaded onto the plate in quadruplicates and briefly spun, followed by 10,000 cells in 45 μL medium. For blanks, three wells were filled with medium. The loaded plate was incubated at 37°C for a total of 96 hours. At the 48th hour, a second dose, following the same concentration and serial dilution for the drugs and vehicle, was loaded onto the 384-well plate. At 96 hours, 50 μL of CellTiter-Glo (Promega, #G7572) was added to the loaded 384-well plate, followed by rocking at 30 rpm for 2 minutes, and incubating for 10 minutes at room temperature in the dark. Luminescence was measured using a BioTek Microplate Reader (Agilent). Drug-combination treatment was performed with a similar set up as above using and were treated for 72hrs without secondary drug-loading. Dose-reduction-index was calculated as described earlier (89). Inhibition curves were calculated using 4-parameter logistic (variable slope) regression.

Mouse-derived lymphoma cell line transfer and treatment

For the murine cell line xenograft experiments, 8–10 weeks old female or male NSG mice were subcutaneously injected with 1 × 10⁷ SN2F or SN2M cell lines resuspended with PBS containing 50% Matrigel (Corning, #354234), in the right flank. After engraftment, mice were randomized in control and treatment groups, and treatment was started when the tumor size reached 150~200mm³. AZ1495 was prepared in corn oil (Sigma, #C8267) with 10% (v/v) DMSO (Sigma, #673439), and orally administered at 12.5 mg/kg, once per day, for 7 days. Tumor size was measured with a digital caliper and calculated with the formula: smallest diameter² × largest diameter × 0.5.

Patient-derived xenograft (PDX) sequencing, transfer and treatment

NSG mice, 6–8 weeks of age, were subcutaneously implanted in the left flank with female SN2 DLBCL PDX (kindly provided by Michael Green, MD Anderson Cancer Center, Houston, USA). Mice were monitored regularly for tumor growth. Thirty-six animals were chosen based on similar tumor burden and randomly assigned into two groups (n=18). Animals received once daily oral gavage treatment of AZ1495 (12.5mg/kg), or control vehicle, for 21 consecutive days. Body weight, body condition and caliper measurements of tumor burden were assessed three times per week. During the study, animals in both groups showed body condition decline and weight loss, so the original formulation of the vehicle, 9:1 corn oil:DMSO (Sigma- Aldrich, #C8267 & Sigma-Aldrich, #673439) was changed to 6:1 beginning on the sixth day of dosing to help mitigate volume effects of corn oil. Animal body condition continued to decline and on the tenth day of the dosing regimen animals were switched to a new vehicle of 6:1, 20% SBE-Beta-CD/Saline:DMSO (Sigma-Aldrich, #142022). Tumor burden over time was summarized by per-animal AUC using observed measurements without imputation; time-to-progression (TTP) was defined as the first day tumor volume reached ≥2× baseline, with animals not reaching this threshold censored at the day of last measurement. Low-pass whole-genome sequencing and CNV analysis was performed as described earlier(90).

Human tonsillar cells

Left-over tonsils after surgery were collected after written informed consent. Single cell suspensions were established by mincing tonsils finely and transferring pieces on a 70μm strainer. A syringe plunger was then used to mash cells through the strainer, which was then washed with RPMI multiple times. Subsequently, ficoll density gradient centrifugation was performed and cells were frozen down in serum-free media (BamBanker, #BB05). Human specimen where obtained and de-identified in concordance with the Institutional Review Board protocols of Weill Cornell Medicine (New York, NY; #0804009762).

Flow Cytometry and Fluorescence-activated cell sorting (FACS)

Spleens were harvested from mice, cut in multiple pieces and placed onto a 70μM strainer in a 6 well-plate with mouse cell culture media(70). Spleen pieces were dissociated using a syringe plunger, and the resulting single cell suspension was filtered once more through a 70μM strainer. For ficoll density gradient-centrifugation or red cell lysis we followed the manufacturers protocol. Cells were resuspended in 1mL PBS containing Fc-block (1:1000, BD Pharmingen 553142) for 10 min, followed by incubation with live/dead staining with a Zombie NIR dye (1:2000, BioLegend 423106) or ViaDye (1:1000, CyTek SKU R7–690009) for 15 min, followed by surface-staining for 30 min on ice in the dark. If the mix contained a biotinylated antibody, this was followed by a washing step and another incubation with a secondary antibody for 30 min on ice in the dark. For intra-cellular staining, cells were fixed with the FoxP3-TF staining Kit (eBioscience, #00–5523-00), following manufacturer’s instructions. Cells were washed and resuspended in a permeabilization buffer containing intra-cellular stains and incubated overnight at 4 °C in the dark. Before analysis, cells were spun down at 800rcf for 5 min and resuspended in permeabilization buffer. Data was acquired on CyTek Aurora (CyTek Bioscience) and analyzed using FlowJo (FlowJo LLC) software. Gating strategies are provided in the Supplementary Methods. Following fluorescent-labelled antibodies have been used: anti-mouse: CD93–BUV395 (740275; 1:400), CD38–BUV563 (741271; 1:400), CD45–BUV615 (752418; 1:250), IgM–BUV661 (750660; 1:200), CD138-BUV737 (custom made, 1:500), FAS/CD95-BUV805 (741968; 1:200), B220-BV786 (563894; 1:200), IgD–BV510 (563110; 1:250), KI67-BV711 (563755; 1:500), CD21-BV750(746922. 1:1000), Streptavidin-BUV496 (612961; 1:200), CXCR4-Biotin (551968; 1:200) – all by BD Bioscience; IgD–BV711 (405731; 1:250), CD19-SN685 (115568; 1:200), TLR7-PE (160003; 1:150), T-bet-PE-Cy7 (644824; 1:400), T-bet-APC (644814; 1:200), CD86-BV605 (105037; 1:125) – all by BioLegend; CD23-PerCP-eF710 (46–0232-80; 1:200), CD11c-PE-Cy5.5 (35–0114-80; 1:250) – all by Invitrogen. Anti-human: CD21-BUV496 (750614; 1:100) – by BD Bioscience; IgA-VioGreen (130–114-007; 1:200) – by Miltenyi. IgD-PerCP-Cy5.5 (348208; 1:200), CD27-PE (356406; 1:100), CD11c-PE-Dazzle594 (337228; 1:400), CD38-SN684 (303552; 1:100), CD19-APC-Cy7 (302218; 1:50), IgG-BV711 (410739; 1:200), CD20-BV785 (302356; 1:100) – all by BioLegend.

Enrichment for MZBs, FOBs, AiBCs and CD19⁺ cells

Mice were euthanized and spleens were processed as described above. For enrichment of B cell subpopulations (MZBs, FOBs, and AiBCs), we used the MZ and FO B isoliation kit (Miltenyi, #130–100-366) which leads to two fractions: (1) FOBs (CD23⁺) and (2) MZBs + AiBCs (CD23^-). Fractions were incubated for red cell lysis following the manufacturer’s instructions. Afterwards, both fractions were stained with: CD23-PerCP-eF710 (46–0232-80; 1:200), CD11b-PE-Cy7 (25–0112-82, 1:400) – both by eBioscience; CD19-PE (152408; 1:200), CD11c-APC-Cy7 (117324; 1:200), CD21-APC (123412, 1:200), IgD-BV711 (405731, 1:200) all by BioLegend; FAS-BV421 (562633, 1:200), CD38-BUV395 (740245, 1:200), and CD138-BUV737 (custom made, 1:500), Cleaved Caspase 3-PE (561011. 1:200) – all by BD Biosciences and DAPI (LifeTechnologies, #D3571) as described above and sorted for MZB (CD19⁺CD138^-CD38⁺FAS^-CD21⁺CD23^-), FOBs (CD19⁺CD138^-CD38⁺FAS^-CD21^intCD23⁺IgD⁺) and AiBCs (CD19⁺CD138^-CD38⁺Fas^-CD21^-CD23^-CD11b⁺CD11c⁺) (Supplementary Methods) with a FACSAria II cell sorter (BD Bioscience). For the enrichment of CD19⁺ cells, we either sorted as described above or enriched by magnetic bead isolation using CD19 MicroBeads (Miltenyi, #130–121-301) with subsequent enrichment by column (Miltenyi, #130–042-401).

AiBC stimulation

Stimulation of mouse and human MZB and FOB was performed as described elsewhere(60). In brief, cells were sorted as described above and stimulated with IL21 (mouse: 0.05μg/mL; Miltenyi, 130–108-949; human: 0.05μg/mL; PeproTec, 200–21), R848 (mouse: 0.5μg/mL; human: 0.5μg/mL; Selleck, S8133), anti-CD40 (mouse: 1μg/mL; BioLegend, 102802; human: 10μg/mL; BioXCell, BE0189), IgM (mouse: 1μg/mL, Jackson, 115–006-020; human: 10μg/mL; Jackson, 109–006-129) and IFNg (mouse: 0.001μg/mL; BioLegend, 575302; human: 0.01μg/mL; PeproTec, 300–02). After 48 and 72 hrs (mouse) or 72 hrs (human) cells were stained for subsequent flow cytometry analysis.

XIST RNA-FISH & H3K27me3 immunofluorescence

The protocol for XIST RNA-FISH in B cells was published elsewhere(91). In brief, sorted primary mouse B cells were stimulated with 5mg/mL LPS for 24hrs. Cells were applied onto glass slides through coating with fibronectin and fixed. Dehydration was performed through washes in ascending ethanol concentrations and air dried. Freshly prepared FISH probes were applied and hybridized for 18hrs at 37°C. Slides are washed multiple times in 25% formamide (Sigma-Aldrich, #47671) and SSC (Sigma-Aldrich, #S6639) buffer and imaged afterwards.

Sequential XIST RNA fluorescence in situ hybridization (FISH) and H3K27me3 immunofluorescence staining (IF) was performed as previously described(74). In brief, we used two mouse Xist oligo probes (20 nucleotides in length), which recognize repeat regions across Xist gene, labeled with one Cy3 molecule at the 5’ end (IDT) for XIST RNA FISH. Slides were imaged, locations recorded, then used for IF. For IF staining, we blocked slides with 0.2% PBS-Tween 0.5% BSA and used a 1:100 dilution of H3K27me3 primary antibody (Active Motif Cat., #39055). After washing the slides, we imaged them using a Nikon Eclipse Microscope, using the same locations imaged for Xist RNA FISH. We quantified the type of XIST RNA localization pattern (patterns I-IV) in at least 100 nuclei per slide using established descriptions(68). Next, we determined the percentage of nuclei that contain a colocalized H3K27me3 and Xist RNA focus, or either exclusive Xist RNA localization or heterochromatin localization, and numbers of nuclei that exhibit neither Xist RNA nor H3K27me3 localization to the inactive X-chromosome, by comparing Xist RNA FISH images with IF images using the DAPI staining as a reference point and dividing the number counted for each category divided by 100.

CRISPR-based genome editing

Procedures were performed following IDTDNA’s instructions. In brief, ribonucleoprotein (RNP) was generated by adding crRNA (IDTDNA, custom) and tracrRNA (IDTDNA, # 1072532 ) 1:1 to form gRNA with subsequent addition of 125pmol Cas9 (Alt-R^™ S.p. Cas9 Nuclease V3, IDTDNA, #1081058,) enzyme per reaction. Cells were washed and resuspended in SF solution (Lonza, V4XC-2032) adding the RNP complex as well as 100μM electroporation enhancer (IDTDNA, #1075915) and 100μM HDR Donor Oligos (IDTDNA, custom) and electroporated immediately with a Nucleofector system (Lonza, #AAF-1003B, program: CM-137). Subsequently, cells were transferred in pre-warmed media containing HDR Enhancer V2 (IDTDNA, #10007910) for 24 hours. Media was exchanged and cells diluted and plated at a concentration of one cell per well. Sequences of crRNA and HDR Donor Oligos can be found in the Methods. The mutation was verified by Sanger sequencing (Genewiz, South Plainfield, New Jersey, USA). Cell lines were acquired by ATCC or generated within this project and have been validated by short tandem repeat sequencing and tested on Mycoplasma through MycoAlert (Lonza) regularly.

Histology

Tissues were fixed in 4% formaldehyde, followed by 70% ethanol. Organs were embedded in paraffin blocks, and cut, and deparaffinized slides were stained by the Laboratory for Comparative Pathology, Memorial Sloan Kettering Cancer Center, New York City, USA, using a Hematoxylin/Eosin.

DNA extraction & PCR for genotyping

DNA was extracted from cells using the DNeasy Blood & Tissue Kit (Qiagen, #69504), following the manufacturer’s instructions. Purified DNA was resuspended in nuclease-free water and analyzed on an Implen nanophotometer N120 (Implen) and Qubit 3 fluorometer (Life Technologies, Carlsbad, USA) for concentration quantification. The Promega GoTaq Green Master Mix was used for PCR reactions (Promega Corporation M7123), following the manufacturer’s instructions, in an Eppendorf Mastercycler Pro S Thermal Cycler (Eppendorf) Primers were obtained from Integrated DNA Technologies (Coralville, USA). Primers utilized are referenced in the Supplemental Table 4.

Bulk RNA-seq

RNA was extracted using the RNeasy Mini Kit (Qiagen, #74104), following the manufacturer’s instructions. Purified RNA was resuspended in nuclease-free water and analyzed on an Implen Nanophotometer N120 (Implen GmbH) for quality and Qubit 3 Fluorometer for concentration quantification. RNA-Seq was performed as described before(70). In brief, extracted RNA was sent to Medgenome or Admera Health for library preparation and sequencing. Reads were aligned to mm10 using STAR pipeline (STAR_2.7.10b)(92) and annotated to RefSeq through Rsubread package (Rsubread_2.16.1)(93). For centered and unscaled log-transformed transcript per million (TPM) values were used (r-4.3.1; R Core Team (2018). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/). For GSVA scores, human-mouse orthologs were used and enriched as maximum distances from the Kolmogorov-Smirnov random walk statistic using the GSVA package (GSVA_1.26.0)(94). Gene set enrichment was assessed using the GSEA algorithm, a computational method based on the Kolmogorov-Smirnov test.

Bulk ATAC-seq

Bulk ATAC-Seq on sorted CD19⁺ lymphoma cells was performed as described earlier (45). Transcription factor binding activity and differential binding between conditions were inferred from ATAC-seq footprints using TOBIAS with JASPAR motif annotations(95).

Multiome single-cell RNA and ATAC sequencing

Single-cell RNA sequencing (scRNA-seq) and single-cell ATAC sequencing (scATAC-seq) libraries were prepared using the 10X Genomics Single Cell Multiome assay kit (10X Genomics; #1000230, #1000283, #1000494, #1000215, #1000212), following the manufacturer’s instructions. Briefly, primary B cells were sorted from splenocytes, rinsed in 0.04% BSA in PBS, and subjected to the nuclei preparation protocol provided by the manufacturer, employing a combination of the regular and low-input protocols. Approximately 10,000 nuclei/sample were loaded into a 10x Chromium X instrument (10x Genomics). Subsequent library preparation steps were performed according to the manufacturer’s protocol. Library quality was assessed using an Agilent Bioanalyzer. Sequencing was performed on an Illumina NovaSeqPlus platform by the Epigenomics Core at Weill Cornell Medicine. Multiome data was processed as previously described(45,96). Briefly, Multiome fastq fils were aligned and processed with CellRanger Arc v2.0.2 (10x Genomics). CellRanger outputs were then processed individually using Signac v1.5.0(97). MACS2 was used to call peaks using the default settings(98). Samples were combined together with Signac, and a combined peak set was generated from all samples. Peaks were filtered to be less than 10kb and greater than 20 bases in size. Signac FindMultiModalNeighbors was used to generate multimodal nearest neighbors upon which the UMAP and clustering was based. Gene signature module scores for expression were calculated using the AddModuleScore function, with a control value of 5. Expression differences between Seurat(99) clusters was determined by FindAllMarkers, using the default settings.

BCR-Sequencing

Mice, 3–4 months old, were euthanized and spleens processed as described above. Antibody staining and sorting were performed as described above. RNA extraction was performed using Invitrogen TRIzol Reagent (Invitrogen, #15596026). RNA concentration was normalized across all samples then prepared for BCR-sequencing by cDNA transcription (Verso cDNA Synthesis Kit, Thermo Fisher, #Q32855). To analyze B-cell clonality of samples, BCR heavy and light chain genes were amplified from cDNA using homemade primer sequences. Primers were designed based on previous literature, targeting full length Ig heavy and Ig light chain variable region gene transcripts through a two-step, nested PCR(100). Experimental steps were modeled after the Amplicon’s guide (Illumina) for library preparation. As per this protocol, PCR products were size-selected using SpriSelect Bead-Based Reagent (Beckman Coulter, #NC0406406). Illumina adapter sequences were added to the nested primers to enable sample indexing using the Nextera XT Index Kit (Illumina, #FC-131–1002). Amplicons were sequenced using the Illumina MiSeq Platform.

The BCR sequence data was preprocessed using pRESTO v0.7.2(101) and analyzed using Python v3.11.9 and R v4.4.0 packages. Sequences were assembled from paired-end reads by initially attempting de novo assembly, then reference guided assembly. Phred quality scores were used for initial quality control, and all reads with a quality score < 20 were removed. To ensure proper orientation in the V(D)J reading frame, V- and C-region PCR primers were aligned with the reads using fixed position scoring and then removed. Duplicate BCRs were collapsed to a single unique sequence while allowing for interior N-valued positions, and these were filtered to those with at least 2 duplicates. BCR sequences were aligned with IgBlast v1.17.0(102) to the IMGT GENE-DB mouse reference database(103) to provide V and J gene annotations.

B cell clones, which are sequences that descend from a common V(D)J rearrangement, were identified using single linkage hierarchical clustering among sequences with the same V gene and J gene annotation, same junction length, and at most 10% nucleotide difference in the junction using the Change-O v1.3.0(104) package. Data was summarized using dplyr v1.1.4 (RRID:SCR_016708).

DynaVelo analysis

DynaVelo(bioRxiv 2025.04.24.650458) is a dynamic deep learning method to predict RNA and motif velocities using single cell multiome data with joint RNA and ATAC readouts. It uses transcription factor (TF) motif accessibility scores (from chromVAR(105)) and log-normalized RNA expression of a subset of genes as input and map them by encoders to a joint lower dimensional latent space, where it formulates the latent velocity function by a neural ODE and then maps them back to the observational spaces of RNA expression and TF motif accessibility.

Mass spectrometry

DTT (Thermo Fisher; Cat#: J15397–03) in water (1.5 μL, 100 mM) was added to cell lysates and incubated at 95 °C for 15 min. Iodoacetamide (Sigma-Aldrich; Cat#: I6125–5g) in water (1.5 μL, 200 mM) was added and incubated in the dark at room temperature for 30 min. Samples were transferred to a KingFisher 96 deep-well plate (Plate 1). SpeedBead magnetic carboxylate beads (Cytiva, Cat #: 65152105050250 and Cytiva, Cat #: 45152105050250 at 1:1, washed twice with water) were added followed by MeCN (to 50% final volume), and the plate was gently mixed by manual agitation. After 15 min, beads were washed using a KingFisher Flex 96 type 710 magnetic purification system (Ref. 5400500) with KingFisher 96 deep-well plates loaded with: EtOH 80% (Plate 2; 180 μL·well⁻¹), EtOH 80% (Plate 3; 180 μL·well⁻¹), and MeCN (Plate 4; 250 μL·well⁻¹). The following steps were implemented: (a) Plate 1 – collect beads (premix) (b) Plate 2 – wash (release beads, 30 sec, medium; wash beads, 1 min, medium), wash (release beads, 30 sec, fast dual mix; wash beads, 1 min, fast dual mix), wash (release beads, 30 sec, medium; wash beads, 1 min, medium) (c) Plate 3 – wash (release beads, 30 sec, medium; wash beads, 1 min, medium), wash (release beads, 30 sec, fast dual mix; wash beads, 1 min, fast dual mix), wash (release beads, 30 sec, medium; wash beads, 1 min, medium) (d) Plate 4 – wash (release beads, 30 sec, medium; wash beads, 1 min, medium), wash (release beads, 30 sec, fast dual mix; wash beads, 1 min, fast dual mix), wash (release beads, 30 sec, medium; wash beads, 1 min, medium). Beads were released in a KingFisher 96 deep-well plate loaded with trypsin (Pierce, Cat #: 90058; 0.2 μg·well⁻¹) in NH₄HCO₃ (Sigma-Aldrich Cat #: 098730, plate 5, 40 μL·well⁻¹, 5 ng·μL⁻¹ in 50 mM NH₄HCO₃). The following steps were implemented: (a) Plate 4 – collect beads (premix) (b) Plate 5 – wash (release beads, 30 sec, medium; wash beads, 30 sec, medium), wash (release beads, 30 sec, fast dual mix; wash beads, 30 sec, fast dual mix), wash (release beads, 30 sec, medium; wash beads, 30 sec, medium). Samples from plate 5 were carefully transferred to a full-skirted 96-well PCR plate, capped, and incubated at 37 °C overnight while inverted using a rotisserie apparatus. The plate was quickly centrifuged (200 × g, 20 °C, 1 min) to ensure all content sat at the bottom. Samples were place on a DynaMag^™ 96 Side Skirted Magnet (Thermo Scientific, 5 min incubation) and transferred to a new full-skirted 96-well PCR plate, containing formic acid (4.5 μL·well⁻¹, 5% v/v in water). An Oasis HLB 96-well plate (Waters; #186001828BA) was conditioned by washing once with MeCN + 0.1% formic acid (300 μL·well⁻¹, elution at 200 × g for 1 min) and twice with H₂O + 0.1% formic acid (300 μL·well⁻¹, elution at 200 × g for 1 min). The tryptic peptide solutions from on-bead digestion were loaded and eluted twice at 200 × g for 1 min. Adsorbed tryptic peptides were washed twice with H₂O + 0.1% v/v formic acid (300 μL·well⁻¹, elution at 200 × g for 1 min), then eluted once with 80% MeCN + 0.1% v/v formic acid (100 μL·well⁻¹, elution at 200 × g for 1 min). Volatiles were removed by vacuum concentration, and the samples reconstituted in 40 μL 2% MeCN + 0.1% formic acid. Samples were analyzed using a Bruker nanoElute 2 / timsTOF Pro 2 nano-UHPLC-IM/MS/MS system using a two-column nano-UHPLC separation method and data independent analysis (DIA) MS method. Acidified and digested samples (1 μL) were loaded on a trap column (Thermo-Fisher, Cat. No. 174500, PepMap Neo C18, 5 μm particle size, 300 μm ID, 5 mm length) using 12 volume equivalents of H₂O/0.1% FA. Flow through the trap was then reversed, and peptides eluted through a separation column (Bruker, Cat. No. 1893472, Bruker 10 C18, 1.9 μm particle size, 75 μm ID, 10 cm length) using a 2–35% gradient (H₂O/0.1% FA – MeCN/0.1% FA) at a 500 nL/min flow rate. A 20 μm ID CaptiveSpray emitter was used at 1400 V capillary voltage. MS analysis was performed using the instrument default short gradient DIA method in HyStar v. 6.2 without alteration.

Mass Spectrometry Data Analysis

Data was analyzed using DIA-NN 1.8(106). A Human FASTA sequence database from UniProt (accession date: 12–05-2022) was used. A trypsin-digested in silico spectral library generated in DIA-NN was used for analysis. DIA-NN settings: MS1 accuracy and mass accuracy tolerance: ±12.5 ppm; missed cleavages: 1; maximum number of variable modifications: 1; peptide length: 7–30; precursor charge range: 1–4; precursor m/z range: 300–1800; fragment ion m/z range: 200–1800; modifications: C carbamidomethylation, N-term M excision, Ox(M), Ac(N-term); Precursor FDR (%) = 1.0; match between runs enabled, neural network classifier: double-pass mode; protein inference: genes; quantification strategy: robust LC (high precision); cross-run normalization: off; library-generation: smart profiling. Processing of DIANN output files (report.pg_matrix.tsv, report.pr_matrix.tsv) was done in R (4.4.1) using the MSnbase package. Proteomics data was subjected to variance stabilization normalization (‘vsn’) across all experimental groups (uniform normalization), imputation using MinProb in MSnbase, and the number of peptides used for each protein inference determined and added as metadata. Once complete, individually processed MSnSet datasets were combined into a single table. Differential expression analysis was performed using linear models (‘limma’) with Benjamini-Hochberg FDR adjustment to identify significantly enriched proteins. A volcano plot comparing the log2 fold change and −log10(p-value) was generated(107–109).

SMRT-seq and methylation analysis

Female N2 and SN2 lymphoma cells as well as WT MZBs were sorted for CD19⁺ cells and high-molecular weight DNA was extracted using the Nanobind PanDNA Kit (PacBio, #103–260-300) following the manufacturers’ instructions. Preparation of libraries and subsequent sequencing were performed in the Genomics Core of Weill Cornell Medicine. The fastq files of QC-passing reads were processed according to the PacBio recommended settings, aligning by pbmm2 to mm10 with the HiFi preset (RRID:SCR_025549). The CpG methylation levels were extracted by the pb-CpG-tools using the pileup calling model (v1). [cited 2026 April 8]. Available from: https://github.com/PacificBiosciences/pb-CpG-tools. Transcription start site methylation was calculated by taking the median methylation within TSS +− 500bp. Differential promoter methylation was calculated by taking the difference of each CpG site within the promoter window and taking the median of the differences between conditions.

Human scRNA data

We used the HCATonsilData v1.4.0 package to access the scRNA-seq data for naïve and memory B cells of the human tonsil atlas(63). We projected signatures for MZB1/2(62). We used these signatures as input to AddModuleScore (Seurat(99), default parameters) to calculate per-cell scores. The code can be found in the Supplementary Methods. Raw snMultiome data were processed using Cell Ranger ARC (v2.0.2; 10x Genomics) for cellular barcode demultiplexing, alignment to the GRCh38 reference genome, and generation of RNA gene–barcode and ATAC fragment–barcode count matrices. Cell-containing droplets were identified, and ambient RNA contamination was corrected using CellBender (v0.3.0). Doublets predicted by scDblFinder (v1.16.0) were removed. All downstream analyses were performed using Scanpy (v1.10.4). Cells were filtered based on multiple quality-control criteria, excluding those with low or high gene counts (<500 or >7,500 genes) or high mitochondrial gene percentages (>25%). RNA counts were normalized to total counts per cell, scaled to 10,000, and log-transformed with a pseudocount of one. For dimensionality reduction, the top 2,000 highly variable genes were selected using scanpy.pp.highly_variable_genes (Seurat v3 flavor). These genes were scaled to mean 0 and standard deviation 1 and used as input for PCA. The top 50 PCs were used to construct a neighborhood graph, perform Leiden clustering, and generate UMAP visualizations. Batch correction of PCA embeddings was performed using Harmony with sample ID as the batch variable as implemented in scanpy.external.pp.harmony_integrate. Batch correction was applied only to embeddings excluding B cells to avoid spurious integration of tumor and rLN B cells. Marker genes were identified using Wilcoxon rank-sum tests implemented in rank_genes_groups, and cell-type identities were assigned based on established canonical markers. For signature enrichment in human SN2 DLBCL, differentially expressed genes were identified using scanpy.tl.rank_genes_groups with method=“wilcoxon”. Where applicable, mouse-to-human gene conversion was performed using pybiomart.Dataset with name=“mmusculus_gene_ensembl” and host=“http://apr2020.archive.ensembl.org“. Signature scores were calculated using scanpy.tl.score_genes.

Network analysis and Gene Overlap calculations

Network analysis was performed through STRING-DB(110), gene overlap was performed through STRING-DB and MSigDB(111). Enrichment strength was calculated the following formula:

$$enrichmentstrength=log10\left(\frac{k/n}{K/N}\right)$$

with k being overlapping genes, n being the number of genes within the gene set queried, K the number of genes within the gene set tested and N being the total number of genes

Cartoon

The cartoon in Fig. 2B was generated by adapting graphics from the SERVIER MEDICAL ART repository (https://smart.servier.com/). Licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).

Reagents and Resources

Other reagents and resources can be found in Supplementary Table 4.

Resource availability:

Lead contact

Requests for additional information and material should be directed to Ari M. Melnick (amm2014@med.cornell.edu).

Materials availability

The cell lines generated in the study are available through the lead contact upon reasonable request.

STATEMENT OF SIGNIFICANCE.

The findings in this manuscript support a distinct developmental origin for BN2-DLBCL and identify a high-risk female population with actionable targets for precision therapy.

Data and code availability

RNA-Seq, ATAC-Seq, SMRT-Seq and multiome-data have been deposited on the NCBI Gene Expression Omnibus via GSE296145 (token: wdczqkesjzijvwz). Human scRNA generated in this manuscript is available through accession number EGAS50000001594.