Development of a Novel Chromatin Activation Score As a Powerful Independent Prognostic Factor in Chronic Lymphocytic Leukemia

Abstract

Background & Aim: In a previous study, we identified a signature of 498 regulatory regions with de novo increased H3K27ac levels (a mark of chromatin activation) in patients with chronic lymphocytic leukemia (CLL), which seemed to be present in all cases studied (Beekman et al., Nat Med 2018). In this abstract, we show that such de novo chromatin activation is not stable among CLL patients but heterogeneous, and therefore, we aim at exploring whether such heterogeneity has a biological and clinical impact. Results: The signature of 498 de novo active regulatory regions was related to 434 specific H3K27ac peaks based on chromatin immunoprecipitation sequencing (ChIPseq) data from 104 CLLs. Analyzing the signal variability among patients, we observed a high degree of heterogeneity. An unsupervised clustering revealed two different sets of regions whose mean H3K27ac signal was associated with time to first treatment (TTFT) using a univariate Cox model. In detail, a higher mean H3K27ac intensity was associated with a shorter TTFT in one of the sets (P=0.001 and hazard ratio (HR)=3.5) while, intriguingly, showed a trend towards longer TTFT (P=0.09 HR=0.6) in the other set. These findings suggest that not only disease progression but also disease stability seem to be active processes in CLL involving de novo chromatin changes. Then, we argued that the clinical behavior of CLL patients could be better captured by the balance between the activation levels of these two sets of regions. Indeed, the ratio of both means was statistically more significant (P=1.4x10-9 HR=15.4). Next, we further refined these groups by selecting individual regions associated with shorter (HR>1, 77 peaks) or longer (HR<1, 144 peaks) TTFT (FDR 0.2). We subsequently defined the score SPP as the mean signal in the pro-progressive (PP) group (HR>1), the score SPS as the mean signal in the pro-stable (PS) disease group (HR<1), and SB as their ratio. Ultimately, this led to an improvement of the statistical significance of the SB with TTFT (P=1.4x10-14 HR=8.6). From a biological perspective, we explored the validity of the SB performing ChIPseq for H3K27ac in 10 primary CLLs before and after a co-culture system inducing strong proliferation (Mangolini et al., Nat Comms in press). In these experiments, we observed an increased SB (P<10-3) that was mostly associated with a decreased SPS while maintaining the SPP levels. To further validate the biological implications of our findings, we mined RNAseq data in 64 cases with concurrent ChIPseq experimentation. We identified target genes whose expression levels correlated with H3K27ac levels (43 genes associated with the PP regions and 62 target genes associated with the PS regions). Both lists included genes previously described in CLL (e.g. CTLA4, FMOD or KSR2). Analogously to H3K27ac, we defined three new scores based on expression levels: SPP-exp, SPS-exp and SB-exp. These scores highly correlated with the H3K27ac-based scores (Pearson r between 0.82 to 0.88). Interestingly, cases with higher SB-exp were enriched in TNF-α and NF-κB pathways and in RELB, FOS, JUN, JUNB and EGR1 transcription factors regulons. We also identified 21 genetic driver alterations related to a higher SB-exp including trisomy 12, 11q loss and mutations in IGLV3-21R110, ATM, SF3B1, NOTCH1 and U1. Finally, we evaluated the potential of SB and SB-exp as independent prognostic factors in multivariate models. In the ChIPseq series (n=104), SB was significant independently of the IGHV mutational status, Binet stage and number of genetic driver alterations. Similarly, using three independent CLL series of gene expression data (n=110, 174 and 140), we observed that the SB-exp was a highly significant prognostic factor independent from IGHV status, Binet stage and number of drivers. To favour potential clinical translation, we generated a reduced SB-exp with only 12 genes which maintained the clinical impact. Conclusions: This study revealed an unexpected heterogeneity within de novo active chromatin regions in CLL, and that such heterogeneity has a dual clinical impact on CLL. A balance score between regions related to the two tendencies (i.e. more progressive vs more stable disease) identified groups of patients with different biological features. Finally, using both chromatin and gene expression data as a surrogate, the balance score was a highly significant independent prognostic factor in independent CLL series.