Mutational landscape of high-grade B-cell lymphoma with MYC-, BCL2 and/or BCL6 rearrangements characterized by whole-exome sequencing.

High-grade B-cell lymphoma accompanied with double/triple-hit MYC and BCL2 and/or BCL6 rearrangements (HGBLDH/ TH) poses a cytogenetically-defined provisional entity among aggressive B-cell lymphomas that is traditionally associated with unfavorable prognosis. In order to better understand the mutational and molecular landscape of HGBLDH/ TH we here performed whole-exome sequencing and deep panel next-generation sequencing of 47 clinically annotated cases. Oncogenic drivers, mutational signatures and perturbed pathways were compared with data from follicular lymphoma (FL), diffuse large B-cell lymphoma (DLBCL) and Burkitt lymphoma (BL). We find an accumulation of oncogenic mutations in NOTCH, IL6/JAK/STAT and NFκB signaling pathways and delineate the mutational relationship within the continuum between FL/DLBCL, HGBL-DH/TH and BL. Further, we provide evidence of a molecular divergence between BCL2 and BCL6 rearranged HGBL-DH. Beyond a significant congruency with the C3/EZB DLBCL cluster in BCL2 rearranged cases on an exome-wide level, we observe an enrichment of the SBS6 mutation signature in BCL6 rearranged cases. Differential gene set enrichment and subsequent network propagation analysis according to cytogenetically defined subgroups revealed an impairment of TP53 and MYC pathway signaling in BCL2 rearranged cases, whereas BCL6 rearranged cases lacked this enrichment, but instead showed impairment of E2F targets. Intriguingly, HGBL-TH displayed intermediate mutational features considering all three aspects. This study elucidates a recurrent pattern of mutational events driving FL into MYC-driven BCL2-rearranged HGBL, unveiling the mutational pathogenesis of this provisional entity. Through this refinement of the molecular taxonomy for aggressive, germinal center-derived B-cell lymphomas, this calls into question the current World Health Organization classification system, especially regarding the status of MYC/BCL6- rearranged HGBL.

Introduction

High-grade B-cell lymphoma (HGBL) with MYC-, BCL2 and/or BCL6 rearrangements poses a novel, yet provisional, cytogenetically-defined entity within the current World Health Organization classification of lymphoid tumors. It is allocated in the pathobiological continuum between diffuse large B-cell (DLBCL) and Burkitt lymphoma (BL).[1] The t(8;14)(q24;q32) IgH/MYC rearrangement constitutes the molecular hallmark of BL. This or further derivative chromosomal rearrangements that juxtapose MYC to a genomic enhancer, occur in approximately 10% of DLBCL and have been shown to correlate with inferior clinical outcome.[2] The rearrangement is a driver of oncogenesis that is accompanied in approximately 50% of cases by additional rearrangements involving BCL2 and/or BCL6 and referred to as double-hit (DH) or triple-hit (TH) lympho-mas.[3-11]

While the clinical outcome in double/triple-hit HGBL (HGBL-DH/TH) patients is generally poor, recent studies have hinted at a significant impact of MYC translocation partners and defined MYC/Ig rearrangements to be the most reliable predictors of adverse outcome.[2,7]

In a prior study, we discovered an elevated frequency of TP53 impairment in MYC-driven DH/TH, whose presence was subsequently demonstrated for a subset of patients with a single-hit MYC translocation as well, indicating inferior outcome.[12,13]

By conventional cytogenetics HGBL-DH/TH were shown to recurrently harbor a complex karyotype.[4] Data on the genetic basis of this entity, however, remains elusive. Several preliminary studies, predominantly focusing on HGBL-DH/TH with DLBCL morphology, have employed a panelbased next-generation sequencing (NGS) approach.[14-16] The insights from these studies were all restricted by gene panel design and the associated clinicopathological data, even though their central assertions included a significant enrichment in mutations affecting CREBBP, BCL2 and KMT2D alongside an overall reflection of the phenotypical gray zone between DLBCL and BL. Most recently, Cucco et al. elucidated significant aspects of the molecular signature of HGBL in a panel-based sequencing and gene expression study, employing a 70-gene HaloPlex panel and an array-based gene expression approach. The authors restricted their study to samples with DLBCL morphology that stemmed from a clinical trial and the UK's population-based Hematological Malignancy Research Network.[17] A comprehensive, exome-wide assessment of oncogenic driver mutations in HGBL-DH/TH, including cases with BL-like morphology is, however, still warranted and of vital importance to the refinement of the pathogenetic understanding of this clinically challenging entity ultimately enabling targeted therapeutic approaches.

We therefore conducted a whole-exome sequencing (WES) study on a large cohort of HGBL-DH/TH, validated by panel-based NGS and supplemented these data with a comprehensive clinicopathological assessment of the study group. Here, we report on oncogenic drivers, somatic copy number alterations (SCNA) and putative pathway perturbations, thus refining the molecular taxonomy of MYC-driven germinal center-derived aggressive lymphomas.

Methods

Case selection and clinicopathological characteristics

In a retrospective approach, we reviewed our institutional database to identify HGBL patients whose primary diagnostic biopsy specimen had been referred to the Reference Center for Hematopathology, University Hospital Schleswig Holstein, Campus Lübeck and Hämatopatholo-gie Lübeck for centralized histopathological panel evaluation between January 2007 and December 2019. For additional Information on clinicopathological work-up, please see the Online Supplementary Appendix and the Online Supplementary Table S1.

This retrospective study was approved by the ethics committee of the University of Lübeck (reference number 18-356) and conducted in accordance with the Declaration of Helsinki. Patients had given written informed consent regarding routine diagnostic and academic assessment of their biopsy specimen including molecular studies at the Reference Center for Hematopathology and transfer of their clinical data.

Whole-exome and targeted amplicon-based sequencing

WES of n=47 HGBL-DH/TH samples was performed by a hybrid capture approach with the Agilent SureSelect Human All Exon V6 library preparation kit (Agilent Technologies) followed by Illumina short read sequencing on a NovaSeq platform (Illumina) to an average depth of 304x (standard deviation ±195x; median 234x; sequencing depth was estimated using mosdepth v0.3.2)18 by Novogene (UK) Co., Ltd (Online Supplementary Table S2). Seeking to validate the initial delineation of the exome sequencing-derived mutational landscape in HGBL we employed our in-house custom AmpliSeq panel (Thermo Fisher Scientific, Waltham, MA, USA) for targeted amplicon sequencing (tNGS), encompassing all coding exons of 43 genes (see Online Supplementary Table S3) in 21 cases. DNA preparation for validation experiments was extracted from the same sample but in an independent approach from deeper tissue sections. Raw paired-end data (fastq format) was trimmed and quality filtered using FASTP (v0.20.0; minimum length 50 bp, maximum unqualified bases 30%, trim tail set to 1)[19] and trimmed reads were mapped to GRCh37/hg19 using BWA MEM (v0.7.15).[20] Resulting alignment files in SAM format were cleaned, sorted, and converted into BAM format using PICARD TOOLS (v2.18.4). Single nucleotide variants (SNV), as well as short insertions and deletions (InDels) were identified following the best practices for somatic mutations calling provided by GATK.[21] Somatic copy number aberrations (SCNA) were identified by CONTROL-FREEC (v11.4).[22] For further details on nucleic acid extraction, panel sequencing, single nucleotide and copy number variant calling please see the Online Supplementary Appendix.

Mutational deleteriousness and significance, network propagation, gene set enrichment and mutational cluster analysis

The MUTSIGCV algorithm was employed on WES data to delineate significantly mutated genes within the study cohort while deleteriousness was assessed via the CADD v1.3. The acquired genomic data were then processed through the LymphGen algorithm and underwent manual screening for an enrichment in overlapping aberrations with the molecular clusters proposed by Chapuy et al. followed by validation through a logistic regression frame-work.[23,24] Cytogenetically defined subgroups (HGBL with MYC and BCL2 aberrations, HGBL with MYC and BCL6 aberrations, and HGBL-TH) underwent differential downstream analysis by a network propagation approach simulating a protein-protein interaction network. Subsequently gene set variation analysis was performed against HALLMARK gene sets.

For details including statistical approaches correlating molecular and clinicopathological findings please see the Online Supplementary Appendix.

Results

Clinicopathological characteristics of the study group

We collected 47 cases of HGBL-DH/TH at diagnosis with sufficient formalin-fixed, paraffin-embedded (FFPE) tissue samples for molecular studies (median age 71 years; range, 35–89 years) all of which were included in the final analysis, following successful library preparation for WES. There was insufficient clinical follow-up in nine of 47 (19%) cases. An underlying HIV infection was clinically excluded in all cases. The majority of patients in our study were male (25/47; 53%) and presented with advanced stage disease (24/38 stage III/IV; 63%) and an adverse prognostic constellation (24/38 [63%], revised International Prognostic Index [R-IPI] >2). Most patients received an intensive cyclophosphamide, doxorubicin hydrochloride, vincristine sulfate, and prednisone (CHOP)-like therapeutic frontline approach (25/38; 66%). The overall response rate after first line (immuno-) chemotherapy was 76% resembling a general therapeutic response in 29 of 38 cases.

Table 1 summarizes the baseline characteristics of all HGBL-DH/TH cases included in the current study. Histo-logically, the predominant morphology was that of DLBCL (NOS) (32/47), however, Burkitt-like morphology and im-munophenotype was present in 15 of 47 patients. Cyto-genetically 21 of 47 cases presented with MYC/BCL2, 17 of 47 presented with MYC/BCL6 DH constellation and nine TH lymphomas were included. MYC translocation partner revealed MYC-Ig rearrangement in eight of 17 cases. The treatment outcome in our cohort was unfavorable yet in keeping with previous data reported by Rosenwald and colleagues.[2] For confirmatory purposes, we included four cases, which were assessed for TP53 mutation status in a previous study and were able to validate both cytogenetic as well as molecular observations.[12]

Table 1.

Clinical characteristics of the study group.

The mutational landscape of double-/triple-hit high-grade B-cell lymphoma identified by whole-exome sequencing

In order to characterize the mutational landscape in an extensive cohort of HGBL-DH/TH cases, we successfully performed WES in 47 patient-derived tumor biopsies and matched constitutional DNA in seven cases. We further applied the analytical framework outlined above to analyze WES data in the absence of paired germline DNA in the majority of cases. Following the primary identification of SNV and Indels in individual samples and subsequent filtering to correct for FFPE-derived artefacts and spurious mutations, we applied the MutSig2CV algorithm and thereby identified 22 significant candidate driver genes (P<0.001; 13 genes with q<0.1; Online Supplementary Table S4).[25]

All HGBL-DH/TH cases carried mutations in genes of on-cogenic relevance according to our bioinformatic annotations. In total, we described 10,092 presumably harmful somatic mutations (cut-off see materials and methods) involving 5,521 genes after variant filtering. Of these, SNV and InDels represented 74.1% of the mutations (7,479 SNV). Among them, missense mutations were the most frequent alterations (85.2%), followed by nonsense (5.7%) and InDels (5.6%), while splice site mutations posed 3.3% of somatic mutations (Figure 1A). Displaying an overall intermediate tumor mutational burden (median 3.974; range, 1.065–18.234 mutations/Mbase; Figure 1B), HGBL-DH/TH revealed no evidence of MSI-related hypermuta-tions, which is in keeping with observations in DLBCL (0.3%), even though it differs from other aggressive lym-phomas (e.g., primary mediastinal B-cell lymphoma).[26]

Figure 1.

Variant classification and mutations per sample. Panel (A) shows the number of variants stratified by variant classification while panel (B) delineates the number of mutations per sample with a median of 153 mutations per sample. Ins: insertion.

Upon comparative analysis of WES and targeted rese-quencing data, we were able to demonstrate a concordance rate of 92.0% (46/50 in 18 matched samples) of mutational calls, prompting high confidence in mutational calls derived from WES, even in non-germline-paired cases. A comprehensive description of all variants described by WES as well as panel based NGS is provided in the Online Supplementary Tables S5 and S6. Nevertheless, we observed a significant enrichment of non-germline matched samples in non-synonymous SNV, which prompted us to include significantly mutated genes according to the MUTSIGCV analysis, only. As an exception to this rule, we also included MYC mutations below the statistical significance level due to their previously established clinical and functional relevance.

Recurrent copy number alterations in double-/triple-hit high-grade B-cell lymphoma

We investigated our HGBL-DH/TH cohort for SCNA employing the CONTROL-FREEC [22] algorithm in tumor-normal and tumor-only mode, respectively, followed by GISTIC2.0 [27] analysis. The analysis excluded chromosomes X and Y as well as common benign copy number variants defined by the Broad Institute’s panel of normals. Upon cross-referencing our findings with genomic loci of known onco-genes, tumor-suppressors and elements of significant signaling pathways we identified recurrent copy number gains in oncogenes such as MEF2B and CSF1R, which have previously been implicated in the pathogenesis of malignant lymphomas.[28,29] Further, copy number losses in tumor suppressors like NPM1 were recurrently identified (Figure 2A and B). No significant differences were detected for genes affected by copy number alterations between the three cytogenetically defined subgroups (Fisher exact test P>0.05 after Bonferroni correction for multiple testing). Common CNA, as defined by the above referenced panel of normal were encountered at the expected fre-quencies.[30,31]

Figure 2.

Genomic and mutational landscape in double-/triple-hit high-grade B-cell lymphoma. The location of somatic copy number alterations (SCNA) along the genome is shown in (A) (red bars denote gains; blue bars denote losses; gene names refer to affected oncogenes according to OncoKB). (B) Display of oncogenes from (A) and SCNA status (red refers to gain, blue refers to loss). Additionally, BCL2 and BCL6 status are shown for each case. Co-oncoplot for genes identified as significant driver genes by MUTSIGCV (P<0.001; n=22) in our cohort stratified by cytogenetical subtypes is shown in (C); different types of mutations are colour coded and additional covariates are shown below the plot for each sample. DLBCL: diffuse large B-cell lymphoma; Ins: insertion; Fol: folicular.

Significantly mutated candidate driver genes and mutational signatures

Putative candidate driver genes comprised several genes previously implicated in HGBL-DH/TH pathogenesis, such as KMT2D, CREBBP and TP53 alongside several further mutated genes such as CDKN2A, LNP1 or SI.[14] Established mutational drivers known from other B-cell lymphopro-liferative disorders (e.g., FL, DLBCL, BL) were recurrently encountered (e.g., CCND3, ARID1A) (Figures 2C and 3A).[32,33] Following SNV and InDel evaluation with MUTSIGCV, a network propagation approach (Figure 3B and C) was employed on significantly mutated genes to delineate the functional implications of significant genetic events on neighboring genes. Additionally, we investigated MUT-SIGCV genes using our HGBL-DH/TH (D/THL) cohort, the cytogenetical subgroups (BCL2, BCL6, THL), as well as cohorts of ABC-type DLBCL (n= 67),[33] germinal center B-cell like DLBCL (GCB-type DLBCL) (n=45)[22] BL (n=108)[38] and FL (n=199)[38] (all retrieved via cBioPortal) and overlapping genes between the five lymphoma subtypes and the three cytogenetical subtypes of HGBL-DH/TH (Figure 3D and E). In our limited cohort, distinctions regarding subtype-specific mutational signatures were found to be marginal among BCL2/BCL6 status or Burkitt-like versus non-Bur-kitt-like morphology. However, we found CCND3 mutations, previously reported as driver mutations in BL pathogenesis, to be significantly enriched in HGBL-DH/TH patients with Burkitt-like morphology (9/15 vs. 3/28). This observation hints at partially similar molecular paths of pathogenesis between BL and HGBL with Burkitt-like morphology. Further, an enrichment of mutations affecting CREBBP in HGBL-DH/TH patients with BCL2 rearrangement was observed, which is well in keeping with its proposed fundamental role in FL pathogenesis. Distribution of mutations within selected, significantly mutated genes is depicted in the Online Supplementary Figure S1. Additional profiling of mutational signatures driving HGBL-DH/TH revealed a predominance of the SBS5 (implicated in aging, potential FFPE artifacts and tobacco exposure) signature across all subtypes alongside the emphasized occurrence of the SBS6 signature (implicated in defective DNA mismatch repair [MMR]) in patients with BCL6 rearrangements (Online Supplementary Figure S2; Online Supplementary Table S7).

Figure 3.

Mutational analysis for significantly mutated genes, network propagation and mutational overlap with related entities. (A) Significance levels for significantly mutated genes in our double-/triple-hit high-grade B-cell lymphoma (HGBL-DH/TH) cohort, regardless of subgroup (MUTSIGCV P<0.001; gene names in orange indicate q<0.1); (B) and (C) show pathway enrichment analysis results for network propagation analysis (see the Online Supplementary Appendix for details) of significant MUTSIGCV genes (MUTSIGCV P<0.001) for MYC/BCL2 subgroup (MYC/BCL2 genes included: POU2AF1, HVCN1, B2M, TP53, AP3S1, LNP1, CREBBP, GULP1, TNFRSF14) and MYC/BCL6 subgroup (MYC/BCL6 genes included: CDKN2A, CD78B, LNP1, KRTAP13-1, UBE2A, CCND3) against HALLMARK gene sets and NFκB pathway. UpSet plot (D) showing the overlap of MUTSIGCV genes using our HGBL-DH/TH (D/THL) cohort, the cytogenetical subgroups (BCL2, BCL6, THL), as well as cohorts of ABC-type diffuse large B-cell lymphoma (DLBCL) (n=67)[34], germinal-center B-cell lymphoma (GCB-type DLBCL) (n=45)[23], Burkitt lymphoma (BL) (n=108)[34] and follicular lymphoma (FL) (n=199)[34] (all retrieved via cBioPortal); set size refers to the number of genes per cohort and intersection size shows the number of overlapping genes per comparison. Comparisons are denoted by black points and black connecting lines; (E) shows the overlapping genes between the 5 lymphoma subtypes and the 3 cytogenetical subtypes of HGBL-DH/TH; grey denotes mutated genes.

Comparative analysis of mutational landscape in double-/triple-hit high-grade B-cell lymphoma, related entities and molecular clusters in diffuse large B-cell lymphoma

Next, we sought to refine the genomic taxonomy of aggressive GC-derived BCL and to investigate the mutational commonalities and differences between HGBL-DT/TH and other related pathological entities. We subsequently selected cBioPortal cohorts from several entities for their similar or divergent features of B-cell differentiation (FL, GCB-type DLBCL and BL vs. ABC-type DLBCL). A comparative analysis of candidate mutational drivers in HGBL-DH/TH (as described earlier) and cohorts of ABC-type DLCBL (n=67), GCB-type DLBCL (n=45)[23], BL (n=108)[34] and FL (n=199)[34] was conducted, screening for shared as well as mutually exclusive putative driver mutations. Interestingly, we identified one overlapping candidate driver common to all entities (KMT2D). Additionally, CREBBP was found in all entities except ABC-type DLCBL. Mutations affecting EZH2, IRF8 and TNFRSF14 were, however, specifically occured in HGBL-DT/TH and FL/GCB-type DLBCL, while CCND3 mutations appeared to be a pathogenetic feature shared between BL and HGBL-DT/TH. Additionally, TP53 mutations posed a predominant feature of aggressive lymphomas present in all HGBL-DH/TH, GCB-type DLBCL and BL types and therefore most likely acquired during high-grade transformation (Figure 3D and E). In basic accordance with previous studies, our data suggest a common origin especially for BCL2-rearranged HGBL-DH/TH and FL/GCB-type DLBCL.[14,17]

Upon comparative investigation of our current data and mutational clusters, previously described in DLBCL, we observed a striking predominance of C3/EZB cluster cases in the BCL2-rearranged subgroup according to the inte-grative molecular classification proposed by Chapuy et al. and Wright et al., respectively. This is in keeping with a significant enrichment of these cases with DLBCL morphology in terms of MYC rearrangement status (Figure 4; Online Supplementary Figure S3).[23,24] Complementary to our analysis, employing the LymphGen algorithm (cf. Online Supplementary Table S8), a logistic regression indicated a significantly different number of mutated genes in C3 between the HGBL subtypes. HGBL harboring only BCL2 were shown to exhibit the highest number of mutated C3 genes, while HGBL with BCL6 alterations had the lowest number of mutated C3 genes (BCL2/6 cases: P=6.059*10-5, adjusted R2=0.3108). In contrast to TH cases, HGBL with MYC and an isolated additional BCL6 rearrangement showed a significant decrease in the number of mutated C3 genes (P=2.00*10-5, estimate: -1.7589) (Online Supplementary Figure S4; Online Supplementary Table S9). Within the subgroup of BCL6 rearranged cases, the BN2 cluster was more prominent than the EZB cluster. In keeping with their strong affinity towards the C3/EZB cluster, BCL2-rearranged cases exhibited an enrichment for mutations in CREBBP and KMT2D, while BCL6-re-arranged cases were, in contrast, enriched for mutations in ARID1A. The vast majority of TH cases was also classified within the EZB cluster.

Mutational impairment of NOTCH, RTK-RAS and TP53 signaling in double-/triple-hit high-grade B-cell lymphoma

Cumulatively, we detected genetic lesions, putatively impairing NOTCH signaling in 74% of HGBL-DH/TH patients. Expanding on previously reported CREBBP, EP300 and DTX1 mutations in HGBL we further identified recurrent mutations affecting NCOR1 and others (Online Supplementary Figure S5).[14,17] NOTCH signaling was thereby the predominant target of somatic mutation in HGBL-DH/TH, albeit with a quite heterogeneous mutational pattern affecting 35 of 47 patients with lesions in 28 of 71 genes (Online Supplementary Figure S6A). Most of these genomic aberrations had been previously reported to be gain-of-function mutations putatively resulting in constitutive NOTCH pathway activation in various types of predominantly GCB-type DLBCL. Several of these mutational hits including NCOR1 and DTX1 have been shown to herald adverse clinical outcome.[35,36]

This remained the case when undertaking a differential downstream analysis within the cytogenetically defined subgroups (HGBL with MYC and BCL2 aberrations, HGBL with MYC and BCL6 aberrations and HGBL-TH), which was prompted by their significantly divergent distribution onto molecular clusters. Through this analysis a mutational signature became apparent that is additionally dominated by impairment of TP53 and MYC signaling in BCL2 rearranged cases. BCL6 rearranged cases lacked this enrichment, while HGBL-TH cases revealed intermediate mutational features. As another recurrent feature across all subgroups we observed alterations, putatively affecting IL6/JAK/STAT signaling in 74% of patients (Online Supplementary Figure S5). Mutations in PIM1 and SOCS1 were most frequently encountered in our case series and have been previously implicated in HGBL-DH/TH pathogenesis. These genes failed, however, to reach the predefined level of statistical significance within the scope of our MUT-SIGCV analysis and required further investigation in a more comprehensive dataset.[17] These candidate driver genes are supplemented by mutations in LT B and STAT3 (both previously identified in HIV-associated plasma-blastic lymphoma) among others.[37] Beyond this, we observed a relatively dispersed mutational pattern with putative driver events affecting 28 genes within the NOTCH-pathway (Online Supplementary Figure S6A).

In accordance with previous studies, we found mutations directly impacting NF-kB signaling in 62% of cases (Online Supplementary Figure S5).[38] While this was, among the predominant pathways, identified through our network propagation approach, mutations affecting the pathway were narrowly detectable with only BCL2 harboring mutations in more than three patients (34%) followed by recurrent SNV and Indels in PARP1 (6%) and BIRC3 (4%) and CARD11 (4%).

The network propagation approach further underscored the mutational impairment of the aforementioned pathways alongside WNT and PI3K signaling. These observations are in accordance with preliminary impressions derived from targeted sequencing studies, employing panel-based approaches.[14,17] In addition to the divergent results from our mutational pathway analysis, we identified an enrichment of E2F targets impacted by significantly mutated genes in both DH and TH cases affected by BCL6 rearrangements. Of further interest, we report on highly recurrent mutations in known activation-induced cytidine deaminase (AID-) and somatic hypermutation (SHM) targets such as PIM1, SOCS1 and others.

Survival analysis

Following integrated analysis of molecular and clinical data we investigated genomic alterations present in >15% of patients for their impact on overall survival (OS) and progression-free survival (PFS). Hereby we identified ARID1A mutations to be predictive of worse clinical outcome in our cohort (OS: P=0.0049; PFS: P=0.025). Subsequent Bonferroni correction for multiple testing was performed. Thus, we identified a significant impact of mutations affecting ARID1A which was maintained regarding OS when correction for multiple testing was applied while its primarily significant effect on PFS was reduced to a trend of borderline statistical significance (Figure 5). A Cox proportional hazard model revealed this effect to be independent of the established clinical International Prognostic Index (IPI) prognosticators (age, lactate dehydrogenase [LDH], extra nodal manifestations, stage, and performance status; OS: P<0.001; hazard ratio [HR]: 13.989; 95% confidence interval [CI]: 3.362–58.205; PFS: P=0.001; HR: 6.648; 95% CI: 2.098–21.061). Within the cytogenetically defined subgroups, we identified no alterations with independent impact on clinical outcome. However, a trend of borderline statistical significance towards inferior outcome in MYC/BCL2 rearranged cases harboring FOXO1 mutations was observed (Online Supplementary Figure S7).

Figure 5.

Survival curves according to ARID1A mutational status. (A) Overall survival (OS) and (B) progression-free survival (PFS) according to ARID1A mutational status. Numbers at risk alongside hazard ratios and P-values according to log-rank testing are provided. Subsequent Bonferroni correction for multiple testing (all genes with a mutational frequency >15% were investigated) identified a significant impact of ARID1A mutations regarding overall survival OS while its primarily significant effect on PFS was reduced to a trend of borderline statistical significance. Fol: folicular; DLBCL: diffuse large B-cell lymphoma.

Discussion

Here we report on WES data from an extensive cohort of HGBL-DH/TH tumors, which is to the best of our knowledge the hitherto largest cohort and most extensive molecular data set for this entity. Previous reports on HGBL-DH/TH were limited by low sample numbers and/or targeted sequencing approaches. Contrary to this, WES here allowed to systematically define recurrent mutations, predominant mutational signatures and SCNV in their respective clinicopathological context from which we report three central observations.

Firstly, being the first exome-wide mutational investigation for this rare subtype of lymphoma, we identify a significant overlap of mutational drivers between HGBL-DH/TH and FL as well as GCB-type DLBCL (e.g., TNFRSF14, EZH2 and IRF4) as its high-grade counterpart. Aggressive transformation was associated with the acquisition of mutations in TP53. Moreover, shared features, including CCND3 and CDKN2A mutations underscore a close molecular relation between HGBL-DH/TH and BL.[34,39] This is additionally reflected in the enrichment of HGBL-DH/TH patients with Burkitt-like morphology for CCND3 mutations. Further, we identify a number of significant mutational drivers not captured by previous, panel-based sequencing studies. Most frequently among these, we find SI mutations that have been previously implicated in CLL progression aa well as mutations in POU2AF1, which has been recently found to be an augmented target of mutations during aggressive transformation of FL to DLBCL.[40,41] Although MYC did not meet the predefined MUTSIGCV significance level in our study, we still observed mutations in 19% of cohort samples (Online Supplementary Figure S8), which is in agreement with previous panel-based studies.[17]

Secondly, upon screening the mutational landscape in HGBL-DH/TH in comparison to the molecular clusters of DLBCL, proposed by Chapuy et al. and the LymphGen algorithm proposed by Wright et al., we unveil a striking overlap of BCL2-rearranged cases with the C3/EZB cluster, which was previously shown to be enriched for MYC rearrangements and oncogenic drivers implicated in FL pathogenesis.[23,24] We argue that a predominant subset of HGBL-DH/TH most likely corresponds to these transformed FL. This offers a potential explanation for the inferior clinical outcome of C3 DLBCL patients, despite their GCB-phenotype, through an enrichment for MYC-rearranged HGBL-DH/TH cases. Of note, we find the pre-dominant impairment of TP53 and to a lesser extent MYC signaling in BCL2 rearranged cases to be in keeping with a previous study on an independent set of HGBL-DH/TH, in which we found TP53 mutations to be a recurrent feature of HGBL-DH with BCL2, but not BCL6 rearrangements.[12] Intriguingly, we further observed two molecular subtypes in MYC/BCL6 only rearranged cases. While selected cases were categorized within the EZB cluster, several cases revealed an association with the BN2 cluster, potentially hinting at a MYC-driven high-grade transformation of a precursor lesion with an origin within the marginal zone, as previously described.[24] In addition to these observations, we found TH cases to reflect EZB lymphomas in the vast majority of cases, potentially hinting at BCL6 rearrangements as late and non-defining events in HGBL-TH lymphomagenesis. Supporting this assumption, Pedrosa et al. have shown DLBCL with BCL2 and BCL6, but without MYC rearrangements to be exclusively associated with the EZB cluster.[42] Considering the significantly mutated genes, our observations underscore previous assumptions regarding a molecular divergence between BCL2- and BCL6- rearranged HGBL-DH.[14,17,43] The predominant mutational distinction between these groups was the presumably FL-derived enrichment for CREBBP mutations in the BCL2-rearranged subgroup. On an exome-wide level we observed an enrichment of the SBS6 signature (implicated in defective DNA mismatch repair) and a significantly diminished congruency with the C3/EZB DLBCL cluster in the BCL6-rearranged subgroup. A pronounced SBS6 signature in BCL6-rearranged cases is in keeping with previous reports by Gu et al. who described genomic instability as a result of defective MMR and thereby a shorter latency to the development of BCL6-driven DLBCL in a murine model.[44] Of note, these findings fundamentally dispute the combined characterization in the current World Health Organization classification, despite several shared clinical aspects common to all subtypes of HGBL-DH/TH.[1,2] Beyond de novo DLBCL with BCL6 rearrangement, potential alternative explanations for this phenomenon include both clonal evolution and subsequent aggressive transformation from rare cases of BCL6-rearranged marginal zone lymphomas alongside BCL2 non-rearranged/BCL6-rearranged FL, which were previously shown to be characterized by a heterogenous mutational landscape.[45,46] From our data, we further deduce an intermediate role for HGBL-TH, which may indicate two divergent paths of clonal evolution originating from a BCL2- or a BCL6-driven disease with subsequent acquisition of the alternative rearrangement. The predominance of the SBS5 signature across all cytogen-etic subtypes is most likely attributable to none-filtered FFPE-artifacts and advanced patient age, as was recently described.[47], [4] [8]

Allocation of double-/triple-hit high-grade B-cell lymphoma samples unto the molecular subgroups/clusters of diffuse large B-cell lymphoma, according to LymphGen based on their mutational signature. Additionally, covariates are shown above the plot for each sample. Row names refer to chromosomal alterations, genes and fusions and are background coloured by their specific subtype (light blue = BN2, light orange = EZB).

Lastly, we describe a pronounced mutational impairment of NOTCH, IL6/JAK/STAT and NFκB signaling pathways and recurrent oncogenetically relevant genes affected by SCNV (including MEF2B, which was previously shown to be enriched in mutations/aberrations within the C3 DLBCL cluster) thereby systematically characterize the oncogen-etic footprint of this subgroup of lymphoma. This is further combined with the identification of novel putative mutational drivers (e.g., NCOR1, DTX1, LT B and STAT3) alongside several previously established mutational hotspots in HGBL-DH/TH. Of note, and in keeping with previous observations by Zhang et al., who described an increased AID activity in DH lymphomas, we observe a significant accumulation of mutations in known AID and SHM-targets such as PIM1, SOCS1 and others.[49-51] Moreover, among these significantly mutated genes we describe ARID1A which emerges as a potential prognosticator of treatment response and outcome from our correlative assessment of clinical and molecular features of our present cohort, which was found to be independent from previously established clinical prognostic factors.

We acknowledge the shortcomings inherent to the retrospective design of the study alongside the limited availability of germline DNA for matched pair analysis. The latter aspect is reflected in a significantly elevated number of mutations in non-matched samples and an uneven distribution of controlled cases unto the cytogenetic subtypes. This prompted us to limit our subsequent analysis to significantly mutated genes (except for MYC and BCL2 mutations, which were additionally included based on their proven relevance in prior studies)[14,17,23,52] and thereby equalizing the above-mentioned effect. A minor divergence in mutational calls between WES and ampli-con sequencing may be attributable to a diverse clonal architecture with mutationally different subclones as DNA samples for WES and Panel-NGS were isolated from the same biopsies but different tissue sections. Additionally, on average between 78.35% and 99.17% (first quantile 91.67%, third quantile 96.59%) of the exome targets were covered with at least 40x coverage per sample, while only variants with a minimum coverage of 40x were considered present, which might have led to the exclusion of variants on a low percentage of occaisons due to too low WES sequencing coverage. However, this trade-off reduces the number of false positive variant calls and thereby enhances confidence in our calls.

Pairing of our WES-results with RNA sequencing data, preferably in an extended, clinically annotated cohort, which was beyond the scope of the present study, would further deepen our molecular understanding of HGBL-DH/TH, especially regarding cases with prominent Burkitt or Burkitt-like morphology.

In summary, our identification of distinct mutational landscapes among HGBL-DH/TH, derived from an exome-wide sequencing approach shows both overlapping and distinctive features compared with GC-derived lympho-mas such as GCB-type DLBCL and low-grade FL as well as BL. Our work further underscores the developing notion of a recurrent pattern of mutational events driving a potentially unidentified preexisting FL into MYC-driven HGBL-DH/TH, offering insight into the molecular patho-genesis of this provisional entity. By refining the molecular taxonomy for aggressive, GC-derived B-cell lymphomas, these results call into question the current World Health Organization classification system, especially regarding the status of MYC/BCL6-rearranged HGBL.