The CLL-1100 Project: Towards Complete Genomic Characterization and Improved Prognostics for CLL

Although chronic lymphocytic leukemia (CLL) has been consistently at the forefront of genomic discovery, complete characterization of its genomic landscape has been limited by sample size and cohort diversity. To address this challenge, we assembled and generated new genomic data from CLL samples from ~1100 patients ('CLL-1100'). In total, we analyzed 984 whole-exome sequences (WES), 177 whole-genome sequences (WGS), RNA-sequencing of 717 cases, and 758 methylome profiles.

With our large dataset, we had increased sensitivity to detect candidate drivers even with frequency less than 1%. By applying MutSig2CV to the WES data, we identified 89 putative cancer driver genes (q<0.1), 46 of which were novel. New candidate driver genes highlighted cellular pathways including metabolism (e.g. GPS2, CHKB, CD36) and protein synthesis/stability (e.g. RPS16, EEF1A1,USP8), thus expanding our knowledge of the processes contributing to CLL leukemogenesis. Analysis of somatic copy number alterations (CNAs) using GISTIC2 confirmed both high frequency and rarely reported arm-level events, as well as identified 6 novel focal amplifications and 54 deletions (35 new). Many of these genomic regions primarily contained known CLL drivers but also novel drivers found based on somatic single nucleotide variants (e.g. DDX5), providing further evidence that these likely contribute to pathogenesis. Thus, by approximately doubling the previously reported number of driver genes in CLL, we are able to assign a putative driving event to >92% of CLL samples within the cohort.

The gain in power was most evident in our characterization of the two major molecular subtypes of CLL, those with mutated (M-CLL) or unmutated (U-CLL) IGHV. Separate WES analyses of 513 M-CLLs and 459 U-CLLs revealed numerous differences between these two subtypes: (i) Mutation analysis revealed 24 and 59 candidate driver genes in M-CLL and U-CLL, respectively (q<0.1). Only a minority of genes were significant in both subgroups (n=13; e.g. TP53, SF3B1, NOTCH1), while most were significant in either M-CLL (n=11; e.g. MYD88, KLHL6, ITPKB) or U-CLL (n=46; e.g. XPO1, BCOR, KRAS). Moreover, IGHV subtype-specific analyses enabled further sensitivity to identify 13 novel putative drivers that were not identified in the pan-CLL analysis (e.g. DIS3 in M-CLL; CHKA in U-CLL); (ii) We found 37 and 46 putative focal CNA drivers specific to M-CLL and U-CLL; (iii) By evaluating mutation clustering in 3-D protein structures (using CLUMPS), we identified additional drivers enriched in M-CLL (e.g. DICER1) or U-CLL (e.g. RPS23, RAF1, MAP2K2) (q<0.1); and finally (iv) inference of timing and order of mutation acquisition (by PhylogicNDT) also revealed distinct evolutionary trajectories between subtypes. Altogether, these results highlight the divergent genomic landscape of the IGHV subtypes.

To further understand disease biology and to develop improved prognostic models for this heterogeneous disease, we performed transcriptomic analysis of 610 treatment-naive CLLs after correction for known and inferred covariates (e.g. PEER factors). We identified 8 expression clusters (ECs) that represent subgroups of U-CLL (n=2), M-CLL (n=5) or an intermediate methylation epigenetic subtype (n=1). The ECs were distinguishable based on association (q<0.1) with genomic drivers (e.g. tri(12), SF3B1, XPO1), biological processes (e.g. B-cell differentiation, TNF-𝜶 signaling, oxphos, migration, metabolism) and EC-defining marker genes (e.g. LPL, CTLA4, HCK, BCL7A, TOX2). Multivariable analysis including clinical features and IGHV subtype, revealed distinct outcomes, demonstrating their prognostic potential (OS p=0.013). Of note, ~10% of M-CLLs had U-CLL-like expression profiles and vice-versa. The known difference in clinical outcome between the IGHV subtypes was not observed within these non-canonical cases (OS log-rank p>0.05; p<0.01 for reduced OS difference in comparison to canonical M-CLL vs. U-CLL). Finally, we used machine-learning to robustly classify new samples into these ECs to demonstrate the potential for future clinical utility.

Altogether, the CLL-1100 cohort facilitates novel genomic discovery with multiomic insights and cross-validation, showing the distinct molecular spectrum of a diverse patient population. This sets the stage for the development of comprehensive and more precise genome-based prognostic tools.