Dual biological role and clinical impact of de novo chromatin activation in chronic lymphocytic leukemia Free

Chronic lymphocytic leukemia (CLL) is the most prevalent leukemia in Western countries and is characterized by extensive genetic diversity and a wide spectrum of clinical outcomes.^1-3 The mutational load within the immunoglobulin heavy chain variable (IGHV) genes is the most recognized and validated prognostic marker, classifying patients into unmutated CLL (U-CLL), which lacks or shows low IGHV somatic hypermutation, and CLLs with mutated IGHV (M-CLL). The former clearly shows a worse clinical behavior than the latter.^4,5 Furthermore, TP53 alterations are widely described as markers of poor prognosis.³ However, despite the identification of genetic driver alterations and transcriptional programs,^6-10 the precise molecular mechanisms underlying CLL development and progression are not fully understood.

Over the last decades, epigenetics has emerged as a pivotal feature to understand cancer, including pathogenesis, evolution, clinical behavior, and therapy of several tumor types.¹¹ In CLL, DNA methylation signatures with clinical impact have been linked with the cell of origin, proliferative history, evolutionary dynamics, and deregulated pathways.^12-17 In particular, 3 epigenetic subtypes were identified based on DNA methylation imprints of normal B cells at different maturation states (from less to more aggressive; memory-like/high-programmed, intermediate, and naïve-like/low-programmed CLLs).^12,13 More recently, the regulatory chromatin landscape of CLL and mature B cells has been described,^18-21 identifying a signature of regulatory regions activated exclusively in CLL cells.¹⁸ Therefore, this signature was interpreted as a molecular feature related to all CLL cases, which contrasts with the overall biological heterogeneity of the disease. However, the possible link between the level of chromatin activation and clinical outcome in CLL has not been explored in any of the published chromatin profiling studies.

In this study, we evaluated whether the magnitude of de novo chromatin activation, based on chromatin immunoprecipitation with massively parallel DNA sequencing (ChIP-seq) profiles of histone 3 lysine 27 acetylation (H3K27ac), was linked with the clinical features of patients with CLL. Our findings unveil an unexpected dual role of chromatin activation, promoting either progressive disease (PD) or indolent disease (ID) clinical course. These signatures were integrated into a single balance score with strong clinical impact that was orthogonally validated with transcriptome and proteome data in independent series.

An extended version is provided in the supplemental Methods, available on the Blood website.

We studied a total of 650 patients with CLL from 7 independent cohorts, spanning 3 data modalities (chromatin activation, gene and protein expression, and generated on sorted CLL cells). These cohorts, named ES^6,18,22-24 (n = 280) for the Spanish cohort, DE1¹⁰ (n = 140) and DE2²⁵ (n = 45) for the German cohorts 1 and 2, US1²⁶ (n = 28), US2²⁷ (n = 21) and US3⁸ (n = 91) for the United States cohorts 1, 2 and 3, and UK²⁸ (n = 45) for the United Kingdom cohort, are detailed in supplemental Tables 1 and 2. Most of the data were mined from the source references. To validate CLL prognosis, we focused on 439 patients with available clinical information. All patients provided written informed consent, and the study was performed in accordance with the Declaration of Helsinki and was approved by the local ethics review boards.

H3K27ac ChIP-seq profiles for CLL and B cells were obtained from Beekman et al¹⁸ and reanalyzed (supplemental Methods). CLL raw FASTQ files or gene-level counts data from bulk and single-cell RNA sequencing (RNA-seq) were obtained from previous publications^{6,8,10,18,22-28} and mined using state-of-the-art computational tools (supplemental Methods).

Using normalized log-transformed data of chromatin activity, gene expression, or protein expression, we initially estimated 2 values per sample: the PD and ID scores from the PD and ID signatures, respectively. Specifically, we calculated the fold change of each chromatin region/gene relative to the median of the cohort (ie, in log-transformed data, the fold change is estimated by subtraction). We next computed the median of the fold changes of all the regions/genes per signature to derive the PD and ID scores, respectively. Then, we calculated the balance score by subtracting the ID score from the PD score, which, because the data are log-transformed, effectively represents the PD-to-ID ratio.²⁹ Finally, we calculated the z score of the balance score to study clinical associations and enable comparability across Cox regression models, irrespective of the data set (supplemental Figure 1).

The primary end point for survival analysis was time to first treatment (TTFT). We included all cases with available clinical data across cohorts with sufficient treatment-naïve samples (n > 30), sourced from 5 previous studies.^6,8,10,18,25 H3K27ac peak signal association with TTFT was determined using Cox scores with the SAM function from the samr³⁰ R package (v3.0). Log-rank tests were used to compare Kaplan-Meier curves. Likelihood ratio tests evaluated the fit of Cox regression models. Prognostic variables were included in the analysis only if they had a minimum of 5 events. Permutation tests were used to test the association of the balance score with TTFT in the discovery series (supplemental Methods).

We initially mined H3K27ac ChIP-seq data from 104 CLLs and 15 mature B-cell subpopulations.¹⁸ Within the 498 predefined de novo active regulatory regions,¹⁸ we identified 434 individual H3K27ac peaks de novo acquired in CLL (Figure 1A; supplemental Table 3). Although these regions seem to represent a signature of CLLness, we observed differences in the mean H3K27ac signal and the percentage of individual peaks detected for each sample (Figure 1A), indicating that the level of de novo chromatin activation varies among individual CLL samples.

Next, we applied unsupervised hierarchical clustering and detected 2 main sets of de novo active chromatin regions: signature S₁ comprising 130 peaks and signature S₂ comprising 304 peaks (Figure 1A; supplemental Table 3). Interestingly, this analysis showed that the H3K27ac signal in S₁ is clearly related to the IGHV mutational status (higher in U-CLL than in M-CLL; P < .0001; Figure 1B). In contrast, the regions of S₂ did not show global differences between IGHV groups (P = .3), although a subtle clustering of cases based on IGHV status could be observed in the heat map (Figure 1C). Because some of the de novo active chromatin regions undergo a higher level of activation in U-CLL than in M-CLL, we postulated that chromatin activation levels may correlate with clinical behavior. Therefore, we associated each region with TTFT as a surrogate for CLL progression³² using univariate Cox scores³³ (Figure 1D; supplemental Figure 2). As expected, due to the higher levels in U-CLL, the magnitude of chromatin activation in regions from S₁ was linked to worse clinical behavior (66/130 regions; hazard ratio >1; Figure 1E). In the case of S₂, we unexpectedly observe that, in spite of not being significantly related to the IGHV status (Figure 1C; right), higher H3K27ac signal was associated with a more indolent clinical behavior (139/304 regions, hazard ratio <1; Figure 1F). Interestingly, compared with S₁, regions of the S₂ signature were enriched in enhancer elements distant from promoters (P < .001; supplemental Figure 3).

These initial analyses suggest a dual role for de novo chromatin activation in CLL. In addition to the expected link between increased chromatin alterations and worse clinical behavior, we identify that indolent clinical behavior also seems to be an active process linked to de novo programming of the chromatin landscape.

The 2 chromatin signatures associated with shorter or longer TTFT exhibit a gradient of chromatin activation levels (Figure 2A-B). Consequently, we generated 2 scores that summarize the clinically relevant chromatin activation level in each sample and signature: S_PD for PD (n = 66 regions; Figure 2A; supplemental Table 3) and S_ID for ID (n = 139 regions; Figure 2B; supplemental Table 3). The S_PD score was clearly increased in U-CLL compared with M-CLL (P < .0001; Figure 2A). Although the initial S₂ signature was unrelated to the IGHV status, the refined S_ID, selecting clinically relevant regions, showed a significant increase in M-CLL, consistent with its better clinical behavior (P = .008; Figure 2B). However, the S_PD score showed a much stronger association with the IGHV status, as determined by the magnitude of the Cohen d statistic between U-CLL and M-CLL (1.64 for S_PD and –0.56 for S_ID). In the context of the 3 CLL epigenetic subtypes, the magnitude of the PD and ID scores was overall aligned with their reported clinical behavior (supplemental Figure 4). Interestingly, multivariate Cox analyses demonstrated that each score, taken as a continuous variable, independently predicted TTFT in opposite directions (Figure 2C).

We next postulated that the balance between these 2 forces³⁴ could be a better determinant of overall CLL behavior. Thus, we defined the balance score (S_B) as the S_PD-to-S_ID ratio (ie, S_PD minus S_ID in log-transformed values; Figure 2D). A permutation test revealed a strong association between the balance score and TTFT (P < .0001; Figure 2E). Notably, as a quantitative variable, the balance score outperformed the clinical impact of the 2 individual scores and seemed to encapsulate all prognostic information provided by each of them (Figure 2F). Moreover, a permutation test for TTFT demonstrated the independent prognostic value of the balance score compared with IGHV (P < .001).

Considering the well-established correlation between H3K27ac and gene expression,³⁵ we next selected 66 CLL samples from the H3K27ac ChIP-seq ES series with available RNA-seq data to identify a surrogate expression signature of chromatin activation levels. Of the potential target genes of the de novo active regions,¹⁸ we selected those with strong positive correlation (false discovery rate [FDR] 10^–5; Figure 3A). This analysis led to the identification of 30 and 38 genes associated with PD and ID, respectively (Figure 3B-C; supplemental Tables 4-5). Interestingly, these PD and ID signatures contain several genes (ie, FMOD, KSR2, CLNK, FILIP1L, and CTLA4) whose messenger RNA (mRNA) seems to be overexpressed in CLL compared with other lymphoproliferative disorders.³⁶ 

Subsequently, we generated 2 expression scores for each sample, that is S_PD and S_ID (Figure 3B-C), which were highly correlated with the analog chromatin scores (r = 0.87 for PD; r = 0.90 for ID; both P < .0001; supplemental Figure 5A-B). As expected, we observed higher levels of S_PD in U-CLL (P < .0001; Figure 3B) and higher levels of S_ID in M-CLL (P < .001; Figure 3C), with S_PD showing a more pronounced difference between the 2 IGHV subtypes than S_ID (Figure 3B-C; Cohen d = 1.82 for S_PD; Cohen d = –0.56 for S_ID). Classifying cases according to 3 CLL epigenetic subtypes also led to the expected magnitudes of the scores (supplemental Figure 5C-D). Multivariate Cox analysis on independent cases of the ES cohort (n = 110) confirmed the independent prognostic value of the S_PD and S_ID as continuous variables (Figure 3D). Stratification by maxstat (maximally selected rank statistics) further revealed a risk gradient by score combinations (supplemental Figure 6); comparisons of Cox regression models with and without scores interaction (likelihood ratio tests; P = .58) underscored their independence as prognostic factors of TTFT. Furthermore, using RNA-seq data from 2 independent cohorts,^18,28 we observed that healthy mature B cells displayed lower S_PD and S_ID than CLL, consistent with their lack of active regulatory chromatin enhancing target gene expression (P < .001; Figure 3E).

We next calculated the balance score S_B as the S_PD-to-S_ID ratio (calculated by subtraction in log-transformed data), which was strongly correlated with the original chromatin-based S_B (r = 0.86; P < .0001; Figure 3F). Importantly, the clinical impact of the S_B was confirmed on a set of independent cases (Figure 3G). Similar to the chromatin-based score, the S_B displays, as a continuous variable, a stronger association with TTFT than S_PD or S_ID individually (supplemental Table 6). These results were validated in a multivariate model with IGHV status, restricting the analysis to cases in Binet A (Figure 3H). Interestingly, we observed that spontaneously regressing CLLs described by Kwok et al ²⁸ had the lowest S_B, followed by indolent M-CLL, progressive M-CLL, and U-CLLs (Figure 3I).

Finally, as an additional layer of orthogonal validation, we mined proteome data of an independent series of 45 CLLs, of which 39 cases were also profiled by RNA-seq.²⁵ We identified 11 and 20 proteins whose expression levels positively correlated (1-tailed test; P < .05) with those of the associated coding genes of the PD and ID signatures, respectively (Figure 4A-B). The subsequent S_PD, S_ID, and final S_B scores showed the same clinical associations as those observed for chromatin and gene expression data (Figure 4A-D).

To gain insight into the potential mechanisms underlying the antagonistic prognostic impact of the ID and PD scores, we initially explored the functions of individual genes within the signatures. We identified that 7 intragenic H3K27ac peaks of the ID signature were associated with ATXN1 expression levels (Figure 5A; supplemental Table 5). This gene has been reported to inhibit NOTCH signaling,³⁷ a known oncogene in CLL.^38,39 Therefore, we further analyzed the association between ATXN1 and NOTCH1 signaling in lymph node (LNs) samples of CLL, because it has been reported that NOTCH1 signaling is operative in the LN niche and becomes downregulated in blood.^38,40-43 We observed a negative correlation of the mRNA abundance between ATXN1 and most of the canonical NOTCH1 signaling target genes in 2 series of CLLs sampled from unpurified and purified LNs (Figure 5B; supplemental Figure 7A). Moreover, comparing the expression of cases with ID high (ATXN1 high) vs ID low (ATXN1 low) in CLL cells from peripheral blood (PB), we observed a strong enrichment in NOTCH signaling (supplemental Figure 7B-C). Another region of interest of the ID signature comprised 9 H3K27ac peaks and 3 genes in 1q32. These genes included PRELP, a gene uniquely expressed in CLL, which reportedly binds and inhibits NF-κB activity⁴⁴; ATP2B4 (PMCA4), which plays a role in B-cell receptor (BCR)-induced calcium efflux⁴⁵; and FMOD, whose specific role in CLL remains elusive, but it has been recognized as a CLL-specific gene^36,46 (supplemental Figure 8A; supplemental Tables 4-5). Additionally, another gene present in the region, LAX1, still exhibited a positive association (although weaker than the other genes) with H3K27ac peak activation (FDR, 0.01). Noticeably, LAX1 has been identified as a negative regulator in lymphocyte signaling in B cells,⁴⁷ and its protein has been found to be upregulated in CLL.⁴⁸ 

Within the PD signature, we highlight a H3K27ac peak in 1p36 associated with the expression of TNFRSF1B, a tumor necrosis factor alpha (TNF-α) receptor⁴⁹ (supplemental Figure 8B; supplemental Tables 4-5); and 2 peaks in 1p22 targeting TGFBR3, a TGF-β receptor previously linked to CLL^50-52 (supplemental Figure 8C; supplemental Tables 4-5). Notably, TGFBR3 expression exhibits the highest correlation with the PD score and shows prognostic value for TTFT independently of IGHV mutational status (supplemental Figure 8D-E).

Next, beyond the individual ID and PD signatures, we investigated pathways and transcription factors (TFs) correlated with the magnitude of the balance score in U-CLL and M-CLL individually (supplemental Figure 9). To increase robustness, we explored 2 independent cohorts^10,18 and detected that the balance score was enriched in 2 main pathways (Figure 5C-D). First is a gene set composed of genes regulated by NF-κB in response to TNF-α signaling, which is supported by both the gene set enrichment analysis (GSEA) and the enrichment of NFKB1^53-55 and RELA^54,56 regulons. Second is mTOR signaling, recently linked to a proliferative drive in CLL,¹⁰ which could be further supported by enrichment in the ATF6,^57,58,HIF1A,^59-61,MYC⁶² and ATF4⁶³ regulons.

Finally, we evaluated the relationship between the balance score and genetic drivers in 170 CLLs.⁶ The results revealed positive associations with 22 drivers (Figure 5E). Although most of these associations were related to IGHV status, we identified specific associations within each CLL subtype. In M-CLL, we observed a strong association between higher balance score and trisomy 12, IGLV3-21^R110 mutation, deletion of 11q22.3, and mutations in SF3B1 and MYD88. In U-CLL, mutations in NOTCH1 and ATM were associated with a higher balance score. Notably, trisomy 12 and genetic alterations in ATM have been previously implicated in TNF-α signaling via NF-κB and/or mTOR signaling.^64,65 Additionally, we studied in detail the magnitude of the PD, ID, and balance scores in cases with the IGLV3-21^R110 mutation, which is related to autonomous BCR signaling and aggressive clinical behavior in M-CLL with moderate IGHV mutation levels and mostly overlaps with the intermediate CLL epigenetic subtype.^12,24,66 M-CLLs and intermediate CLLs with the IGLV3-21^R110 mutation showed higher PD, lower ID, and higher balance score than cases lacking this mutation (supplemental Figure 10A-B). Intermediate CLLs with IGLV3-21^R110 mutation showed similar scores to naïve-like CLLs, whereas intermediate cases without the mutation were similar to memory-like CLLs, which is expected based on their clinical behavior²⁴ (supplemental Figure 10C-E).

CLL cells recirculate between PB and lymphoid tissues, with LNs being a niche where CLL cells are stimulated to proliferate.^39,42,67-70 The LN imprint can be detected in circulating CLL cells, in which 2 major subpopulations are found: the "proliferative" fraction, comprising lymphoid tissue emigrants enriched in recently divided cells, and the "resting" fraction, containing CLL cells primed to home back to the LN.^69,71 Given this context, we explored whether our signatures are modulated in LNs using 4 cohorts; that is, paired LN and PB samples,²⁶ PB samples with sorted proliferative and resting fractions,²⁷ and 2 single-cell RNA-seq experiments^22,23 (supplemental Table 7), in which CXCR4 and CD27⁷¹ levels were used to identify proliferative and resting fractions (refer to supplemental Methods; supplemental Figure 11A-B).

We observed that the PD score significantly increased both in the LN, compared with PB, and in the proliferative fraction, compared with the resting fraction, using bulk and single-cell data (P ≤ .002; Figure 6A-C). In contrast, the ID remained relatively constant (P ≥ .15), with only a few cases showing changes in either direction between LN/proliferative fraction and PB/resting fraction. Consequently, the balance score tended to be higher in the LN or proliferative fraction (P ≤ .056), reflecting the strong increase in the PD score.

Beyond the comparison between proliferative and resting conditions, the data shown in Figure 6 also point to different degrees of PD, ID, and balance scores in different cases. Therefore, we next explored the association of the PD, ID, and balance scores with TTFT in proliferative and resting fractions from 21 cases²⁷ (Figure 6D). The balance score evidenced a stronger association with disease progression in the proliferative fraction (P = .012), while showing a trend in the resting fraction (P = .065). In line with our previous observations (Figures 2E and 3G), the balance score presented a stronger association with TTFT than the PD and ID scores in each cellular fraction.

We next studied the relationship between the balance score and the proliferative potential of CLL cells. We mined CLL samples treated with cytosine-phosphorothioate-guanine oligodeoxynucleotides (CpG-ODN),¹⁰ a class of Toll-like receptor 9⁷² agonists known to induce a proliferative response in CLL cells and whose magnitude has been suggested to be linked to clinical outcomes.⁷³ Our analysis revealed the expected positive and inverse correlation between the PD and ID scores and the percentage of Ki-67⁺ cells, respectively (supplemental Figure 12). Notably, the balance score exhibited a more significant association (ρ = 0.88; P < .0001) than the individual scores.

We conducted comprehensive multivariate analyses encompassing 439 CLL cases across independent cohorts and data modalities, including gene expression (n = 399 across 3 cohorts) and protein expression (n = 40). Whenever available and significant at the univariate level, we evaluated the balance score alongside other clinically relevant variables such as IGHV mutational status, TP53 alterations, clinical stage, lymphocyte count, lymphadenopathy, age, and sex (supplemental Figure 13).

Across all studied cohorts, the balance score maintained a strong prognostic value when analyzed alongside key clinical variables. Specifically, in validation cases from the ES cohort profiled with gene expression arrays, patient stratification by balance score quartiles was highly significant (P < .0001; Figure 7A). In line with this finding, the balance score as a quantitative variable showed prognostic power when adjusted for Binet stage, IGHV status, age, and sex (n = 165; Figure 7B). These data were further validated using RNA-seq in the same ES cohort (n = 81; Figure 7C). In this ES cohort, the balance score was also independent of lymphocyte count and lymphadenopathy (supplemental Figure 14). In the DE1 cohort profiled with RNA-seq, the score was an independent prognostic marker, alongside IGHV status and TP53 genetic alterations (n = 140; Figure 7D). These results were further replicated in the US3 cohort (RNA-seq, n = 91; Figure 7E) and the DE2 cohort (proteomics, n = 40; Figure 7F). Finally, we integrated the balance score from the previous individual cohorts and technologies into a single Cox regression model, confirming its role as an IGHV-independent prognostic marker (supplemental Figure 15).

Notably, although IGHV status is a well-established prognostic factor in CLL,^4,5,74 the balance score consistently outperformed it. We further explored this unexpected finding by comparing univariate Cox regression models incorporating the IGHV mutational status or the balance score and a multivariate Cox regression model including both variables in all the cohorts (supplemental Table 8). Consistent with our previous findings, these comparisons confirmed that including the balance score, but not the IGHV status, improved the model fit. Additionally, epigenetic subtypes also reduce their prognostic power when analyzed alongside the balance score (supplemental Figure 16).

This study provides evidence suggesting that the biological behavior of CLL cells and, consequently, the natural progression of the disease seem to be associated with the balance between 2 antagonistic molecular forces mediated by the same epigenetic mechanism of de novo chromatin activation. A similar concept was proposed in 2014 by Packham et al³⁴ from the University of Southampton, but it centered on the balance between proliferation or anergy triggered by the BCR. The target genes involved in our chromatin signatures not only provide links to the BCR but also to other functions. Genes of the ID score such as ATP2B4 and LAX1 may negatively modulate BCR signaling, but other genes such as ATXN1 and PRELP seem to inhibit NOTCH and NF-κB activity, respectively,^44,45,47,75 which altogether seem to induce cellular quiescence through downregulating the response to proliferative stimuli. The PD score contains genes such as TNFRSF1B, which is a receptor involved in TNF-α signaling, a pathway that can be linked with a more proliferative phenotype in CLL.⁷⁶ Most remarkably, the magnitude of the balance between PD and ID seems to be related to transcriptional states, leading to higher proliferation, as demonstrated by the correlation with the percentage of Ki67⁺ cells and specific pathways such as TNF-α via NF-κB and mTOR signaling. Previous studies associated higher levels of TNF-α with a poorer prognosis,⁷⁷ and TNF-α via NF-κB was identified as a driver of CLL.⁷⁶ Additionally, mTOR signaling has been linked to a proliferative drive in CLL.¹⁰ Interestingly, CLLs with trisomy 12, which have a higher balance score, show a specific enrichment of these 2 pathways.^65,78 These findings suggest TNF-α via NF-κB and mTOR signaling as potential therapeutical targets in CLL.

Our examination of the enrichment of the signatures in relation to CLL trafficking suggests that the PD can be extrinsically modulated in the LN niche, and such imprint gradually decreases in circulating CLL cells. In contrast, the ID is not consistently affected by CLL trafficking and seems to be independent of microenvironmental stimuli.

From the translational perspective, we validated the independent prognostic impact of the balance score using various modalities (chromatin, RNA-seq, microarrays, and proteomics) across several independent cohorts. We observed that the balance score surpasses the prognostic power of the IGHV mutational status, which seems to lose any independent prognostic value in multivariate models. Considering that the balance score is constructed as the ratio of the PD, which is strongly influenced by the IGHV status, and the ID, which is only slightly related to IGHV mutational status, it contains both IGHV-related and independent information.

From the translational perspective, we are aware that using ChIP-seq or RNA-seq–based assays may still be impractical. Therefore, we envision that the balance score could be implemented in the clinical setting through the development of dedicated gene expression assays based on standardized technologies, such as nanoString nCounter.⁷⁹ Furthermore, additional studies shall also evaluate the potential of the S_B as a predictive biomarker of response and overall survival in the context of current treatments.

Overall, we believe that this study provides compelling evidence that the balance score reflects CLL proliferative capacity through the integration of multiple elements involved in CLL pathogenesis. Due to its strong independent prognostic value, the incorporation of the balance score into current clinical models could serve as a new valuable asset in the clinical management of CLL.

This research was primarily funded by an accelerator award Cancer Research UK/Associazione Italiana per la Ricerca sul Cancro (AIRC)/Asociación Española contra el Cáncer (AECC) joint funder-partnership (J.I.M.-S.; grant C355/A26819), the Fundación Científica AECC (FC AECC) (J.I.M.-S.; grant PRYGN234766MART), and the European Research Council under the European Union’s Horizon 2020 research and innovation programme (J.I.M.-S. and E.C.; grant 810287). Additionally, the authors obtained funding for this project from La Caixa Foundation (CLLEvolution LCF/PR/HR17/52150017 [grant HR17-00221LCF] and CLLSYSTEMS-LCF/PR/HR22/52420015 [grant HR22-00172] Health Research 2017 and 2022 Programs to E.C.), Agència de Gestió d'Ajuts Universitaris i de Recerca (AGAUR) of the Generalitat de Catalunya (grant 2021-SGR-01343 to J.I.M.-S. and grant 2021-SGR-01172 to E.C.), and CIBERONC (group numbers CB24/12/00009, CB16/12/00225 and CB16/12/00334). This work was partially developed at the Centre Esther Koplowitz (Barcelona, Spain).

Contribution: V.C. performed H3K27ac ChIP-seq data analysis; V.C. and G.C. conducted statistical analyses; N.R., F.N., and A.V. contributed to sample collection; N.R. and F.N. contributed to bulk transcriptome data generation; V.C., F.N., and R.R. conducted bulk transcriptome data analysis; V.C. and R.M.-B. conducted single-cell RNA-seq data analysis; F.N., T.Z., J.L., H.H., S.D., S.A.H., D.J.B., and J.C.S. contributed with essential data resources; F.N., P.M., M.D.-F., J.A.P., N.V., A.L.-G., J.D., E.C., S.D., and S.A.H. contributed to biological and/or clinical annotation; V.C., J.I.M.-S., A.M.-D., J.M.-J., G.C., F.N., M.D.-F., L.L.-C., S.C., and E.C. participated in data interpretation; V.C. and J.I.M.-S. designed the study and wrote the manuscript; J.I.M.-S. directed the research; and all the authors read, commented on, and approved the manuscript.

Conflict-of-interest disclosure: V.C. and J.I.M.-S. hold the author rights of the prognostic index CLL Balance Score described in this study. The remaining authors declare no competing financial interests.

Correspondence: Jose I. Martin-Subero, Institut d'Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS), Carrer del Rosselló 149-153, Barcelona 08036, Spain; email: imartins@recerca.clinic.cat.