Identification of two epitypes of IGHV unmutated chronic lymphocytic leukemia with distinct B‐cell‐related epigenetic imprinting and genetic alterations

Chronic lymphocytic leukemia (CLL), the most frequent leukemia in the Western world, is characterized by extensive biological and clinical heterogeneity.1 Immunogenetic, genetic, and epigenetic features have been described as major molecular traits strongly associated with clinical impact in CLL.2 The levels of somatic hypermutation (SHM) of the immunoglobulin heavy chain variable (IGHV) region segregate cases into IGHV unmutated and mutated CLLs (U-CLL and M-CLL, respectively), being the former clinically more aggressive.1 This dichotomous classification was further refined using DNA methylation imprints of normal B-cell maturation, leading to the identification of three distinct epigenetic subtypes or epitypes, including naive-like/low-programmed CLL (n-CLL), mostly corresponding to U-CLL, intermediate CLL and memory-like/high-programmed CLL, both of which belonging to M-CLL.3, 4 In spite of this classification into three categories, a model in which CLLs derive from a continuum of mature B-cell maturation stages was proposed, suggesting that CLL could be potentially stratified into additional epitypes.4 Apart from this link with normal B-cell maturation, the DNA methylome of CLL also reflects the past proliferative history and presents specific DNA methylation signatures related to its pathogenesis.5, 6 In addition to IGHV subtypes and epitypes, specific genetic lesions and the presence of complex karyotypes (CKs) carry prognostic1, 2, 7, 8 and predictive value.9 Such CK has been frequently but not always associated with U-CLL and alterations affecting the TP53 and ATM genes.10, 11 To our knowledge, studies analyzing the potential link between CK and epigenetic features have not been reported. Therefore, we have here analyzed the DNA methylome of a U-CLL cohort (n = 52, Supporting Information S2: Table 1, all belonging to the n-CLL epitype according to a previously reported classifier5) that has been carefully selected based on distinct CK levels and the presence of ATM deletion (ATMdel) and/or TP53 deletion or mutation (TP53del/mut) status (Figure 1A). We used Illumina EPIC V1 arrays to generate DNA methylation profiles, which were analyzed as previously described5 (Supporting Information S9: Supplementary Material). An unsupervised principal component analysis (PCA) showed that CK and non-CK U-CLLs are intermingled (Figure 1B), and the same finding was made when analyzing additional principal components (Supporting Information S1: Figure 1A). In line with this observation, a differential methylation analysis revealed a small and heterogeneous signature of 59 differentially methylated (DM) CpGs between CK and non-CK U-CLLs (Figure 1C, Supporting Information S3: Table 2). Comparing non-CK only with cases with high CK (≥5 aberrations), we obtained a still relatively small signature of 315 CpGs (Supporting Information S1: Figure 1B, Supporting Information S4: Table 3). Overall, these results suggest that despite the biological and clinical relevance of CK and DNA methylation in CLL, they do not seem to be strongly related to each other in our series. Beyond the poor association between CK and DNA methylation, a careful inspection of the results pointed to an unexpected finding. We observed that the DNA methylome segregated U-CLLs into two groups that were statistically validated using a consensus clustering approach (Supporting Information S1: Figure 2A,B and Supporting Information S9: Supplementary Methods). Most interestingly, one cluster (U-CLL C1, n = 25) was strongly enriched in ATMdel cases (P = 0.015), whereas the other (U-CLL C2, n = 24) contained all the TP53del/mut cases (P = 0.004) (Figure 1D,E). Three cases could not be unequivocally classified due to low clustering probability (unclassified, UC) and were excluded from all downstream analyses. Further genetic characterization of the two clusters showed that U-CLL C1 was also enriched in ATM mutations (P = 0.017) and del(13q) (P = 0.007), whereas U-CLL C2 was also related to trisomy 12 (tri12) (P = 0.004) (Figure 1F), suggesting that the clusters were associated with particular genetic alterations beyond those involving ATM and TP53. Of note, the two clusters were unrelated to CK (P = 0.15, Figure 1F). We next compared the DNA methylome of U-CLL C1 versus C2 and identified 4333 DM CpGs (Figure 2A,B, Supporting Information S5: Table 4, and Supporting Information S9: Supplementary Methods). To interpret this differential signature, we divided the CpGs into those whose methylation levels change or remain stable during B-cell differentiation, which we termed B-cell-related and B-cell-independent, respectively (Supporting Information S9: Supplementary Methods). A total of 2183 CpGs were B-cell independent (50.4%, Figure 2A), whereas 2150 were B-cell-related (49.6%, Figure 2B and Supporting Information S5: Table 4). The B-cell-independent signature was mainly associated with a de novo methylation loss in U-CLL C2 (n = 1679 CpGs, Figure 2C, Supporting Information S6: Table 5, and Supporting Information S1: Figure 3A). In contrast, U-CLL C1 only showed 118 de novo hypomethylated and 53 de novo hypermethylated CpGs (Figure 2C, Supporting Information S7: Table 6, and Supporting Information S1: Figure 3B). The hypomethylation signatures in both clusters were located in gene bodies and in regions with low CpG content, and were enriched in regulatory elements such as enhancers (Supporting Information S1: Figure 3C–H). On the other hand, the U-CLL C1 de novo hypermethylation was preferentially located in CpG-rich regions and poised promoters (Supporting Information S1: Figure 3I–K). A transcription factor binding site analysis of both B-cell-related and independent differential CpGs did not reveal clear enrichments, with the exception of a highly significant association between the de novo hypomethylation signature of U-CLLs C2 and the NFAT TF (Supporting Information S1: Figure 3L), which has been commonly linked to epigenetic signatures in CLL.4, 5, 12 Remarkably, when we analyzed B-cell-related differential CpGs, we noticed that the higher methylation levels in U-CLL C1 resembled pre-germinal center (GC) B cells, while methylation loss in U-CLL C2 was shared with GC-experienced B cells (Figure 2B). Congruent with this observation, Component 1 of a PCA using these differential CpGs clustered U-CLL C1 cases together with naïve B cells and U-CLL C2 cases together with GC B, memory B, and plasma cell samples (Figure 2D). In fact, a more stringent analysis with a methylation difference of 0.25 revealed 1243 CpGs with similar methylation levels between U-CLL C1 and naïve B cells and between U-CLL C2 and post-GC B cells (Supporting Information S1: Figure 4A and Supporting Information S8: Table 7). Of note, IGHV SHM was remarkably low in both groups, as they mostly contained cases with 100% IGHV homology (Supporting Information S1: Figure 4B). Overall, these findings suggest that these two U-CLL epitypes may derive from GC-independent B cells lacking detectable SHM but showing markedly different levels of epigenetic programming related to B-cell maturation. These results are aligned with the model proposed by Oakes and colleagues, in which CLLs may derive from a continuum of B-cell maturation stages that can be segregated into discrete epitypes.4 Supporting this notion, a phylogenetic analysis of our U-CLL cohort together with other CLL samples13 showed a continuum that can be segregated into four instead of the three previously reported epitypes2-4 (Figure 2E). We show that low-programmed CLLs (i.e., U-CLLs) can be separated into two epitypes that are, respectively, enriched in ATM alterations/del(13q), and TP53 alterations/tri12 (Figure 1F). Furthermore, U-CLL C1 presents a significantly lower proliferative history measured by the epiCMIT clock (P = 0.012; Figure 2F), which further strengthens the concept that they are derived from B cells with lower levels of epigenetic programming as compared to U-CLL C2. From the clinical perspective, these two epitypes did not show significant differences in overall survival (OS) time (Supporting Information S1: Figure 4C) or time to first treatment (TTT) (Supporting Information S1: Figure 4D). However, in the group with more variable epiCMIT levels (i.e., U-CLL C2), a higher epiCMIT score was significantly associated with a shorter TTT (Supporting Information S1: Figure 4E,F), as previously reported.5 We recognize that the initial U-CLL series had a selection bias towards CK with TP53 and ATM alterations. Therefore, to evaluate whether the identification of two new U-CLL/n-CLL epitypes could be generalized beyond this initial cohort, we used two previously published population-based U-CLL/n-CLL cohorts with 159 (ICGC,14 profiled with 450k arrays) and 87 cases (59 profiled with 450k arrays and 28 with EPIC V1 arrays13). For the first cohort, we could also identify two epitypes using 1744 CpGs present in the 450k array that overlap with the 4333 differential CpGs detected with the EPIC V1 array in the initial series (Supporting Information S1: Figure 5A,B), and 13 samples had lower cluster probabilities and were left UC. The two identified clusters (U-CLL C1, n = 86; U-CLL C2, n = 60) showed similar methylation patterns to those previously identified (Figure 2G). In addition, a phylogenetic analysis widely recapitulated our previous findings, where U-CLL C1 cases overall showed less B-cell epigenetic programming than U-CLL C2 (Figure 2H). We also found that U-CLL C1 was enriched in del(13q) (P = 0.004) and U-CLL C2 in tri12 (P < 0.001) as well as other particular genetic alterations in both clusters, including SF3B1 mutations (P = 0.004) and gain(2p) (P = 0.004) in U-CLL C1, and NOTCH1 mutations (P = 0.004) and del(14q) (P = 0.01) in U-CLL C2 (Figure 2I, Supporting Information S1: Figure 5C). As in the discovery series, IGHV SHM was low in both groups, and the noncanonical activation-induced cytidine deaminase (AID) mutational signature (SBS9) identified by whole-genome sequencing (WGS) was likewise low (Supporting Information S1: Figure 5D,E). The epiCMIT score was also found to be significantly lower in U-CLL C1 (P = 0.005; Supporting Information S1: Figure 5F). Finally, the two clusters did not show differences in OS or TTT in this validation series either (Supporting Information S1: Figure 5G,H). For the second cohort, we analyzed the 1749/4333 DM CpGs that were present in the normalized matrix of 450k array data and also identified two clusters with methylation patterns similar to those in the discovery cohort (63 C1 and 24 C2 U-CLLs, Supporting Information S1: Figure 6A). The corresponding phylogenetic analysis also pointed to a lower maturation level in U-CLL C1 (Supporting Information S1: Figure 6B) and a lower epiCMIT score (P = 0.003; Supporting Information S1: Figure 6C). Finally, U-CLL C1 was enriched in ATMdel (P = 0.01) and del(13q) (P = 0.006) whereas U-CLL C2 was linked to NOTCH1 mutation (P = 0.02) and tri12 (P < 0.001) (Supporting Information S1: Figure 6D). Overall, these findings corroborate our observations in the initial and first validation cohort. In summary, although this project was initiated with the intention of exploring the link between CK and DNA methylation, their association was minor in the studied U-CLL cohort. Instead, we serendipitously detected two U-CLL epitypes, which were validated in two additional U-CLL cohorts. Although these two epitypes did not further stratify U-CLL into distinct clinical behaviors, they showed markedly different biological features. These include different levels of B-cell-related epigenetic imprinting and enrichment in distinct genetic alterations. Such enrichment suggests that specific B-cell maturation stages are more prone to acquire particular genetic alterations, such as, for example, ATMdel in U-CLL C1, TP53del/mut/tri12 in U-CLL C2, SF3B1/IGLV3-21R110 mutations in M-CLL of the intermediate-programmed epitype,15 and MYD88 in the M-CLL/high-programmed.14 This study further highlights the power of integrative (epi)genetic profiling to identify, within a continuum of maturation stages, CLL epitypes linked to specific genetic alterations. Stella Charalampopoulou: Investigation; methodology; visualization; software; data curation; validation; writing—original draft. Silvia Ramos-Campoy: Investigation; methodology; writing—review and editing; data curation; resources. Marti Duran-Ferrer: Investigation; writing—review and editing; methodology; validation; data curation; software; visualization. Anna Puiggros: Investigation; writing—review and editing; methodology; data curation. Joanna Kamaso: Resources; writing—review and editing; data curation. Florence Nguyen-Khac: Data curation; resources; writing—review and editing. Gian Matteo Rigolin: Resources; data curation; writing—review and editing. Antonio Cuneo: Data curation; resources; writing—review and editing. Rosa Collado: Data curation; resources; writing—review and editing. Rocío García-Serra: Data curation; resources; writing—review and editing. Claudia Haferlach: Data curation; resources; writing—review and editing. Margarita Ortega: Data curation; resources; writing—review and editing. Pau Abrisqueta: Data curation; resources; writing—review and editing. Francesc Bosch: Data curation; resources; writing—review and editing. Thorsten Zenz: Resources; writing—review and editing; validation. María Laura Blanco: Data curation; resources; writing—review and editing. Rocío Salgado: Data curation; resources; writing—review and editing. Ma Dolores García-Malo: Data curation; Resources; Writing—review and editing. Eva Gimeno: Data curation; resources; writing—review and editing. Armando Lopez-Guillermo: Resources; validation; writing—review and editing. Elias Campo: Resources; validation; writing—review and editing. Laurence Etter: Data curation; resources; writing—review and editing. Jacqueline Schoumans: Data curation; resources; writing—review and editing. Blanca Espinet: Conceptualization; investigation; funding acquisition; writing—original draft; methodology; project administration; data curation; supervision; resources. Jose I. Martin-Subero: Conceptualization; investigation; funding acquisition; writing—original draft; methodology; validation; visualization; project administration; data curation; supervision; resources. The authors declare no conflicts of interest. This work was partly supported by grants from Generalitat de Catalunya (2021-SGR-00025), Gilead Sciences Fellowship (GLD17/00282), Ministerio de Universidades of Spain (FPU17/00361), Fundación Española de Hematología y Hemoterapia (FEHH)-AstraZeneca, European Research Council (ERC) under the European Union's Horizon 2020 Research and Innovation Programme (grant 810287), Generalitat de Catalunya Suport Grups de Recerca AGAUR (2021-SGR-01343 and 2021-SGR-01172), and CIBERONC (CB16/12/00225, CB16/12/00334, and CB24/12/00009). This work was partially developed at the Centre Esther Koplowitz (Barcelona, Spain). The data that support the findings of this study are openly available in Gene Expression Omnibus at https://www.ncbi.nlm.nih.gov/geo/, reference number GSE289415. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.