Towards Response Prediction Using Integrated Genomics in Chronic Lymphocytic Leukaemia: Results on 250 First-Line FCR Treated Patients from UK Clinical Trials

Background: Major progress has been made in understanding disease biology and therapeutic options for patients with chronic lymphocytic leukaemia (CLL). Recurrent mutations have been discovered using next generation sequencing, but with the exception of TP53 disruption their potential impact on response to treatment is unknown. In order to address this question, we characterised the genomic landscape of 250 first-line chemo-immunotherapy treated CLL patients within UK clinical trials using targeted resequencing and whole-genome SNP array.

Methods: We studied patients from two UK-based Phase II randomised controlled trials (AdMIRe and ARCTIC) receiving FCR-based treatment in a first-line treatment setting. A TruSeq Custom Amplicon panel (TSCA, Illumina) was designed targeting 10 genes recurrently mutated in CLL based on recent publications.Average sequencing depth was 2260X. The cumulated length of targets sequenced was 7.87 kb from 330 amplicons covering 160 exons. Alignment and variant calling included a combination of three pipelines to confidently detect SNVs, indels and low level frequency mutations. SNP array testing was performed using HumanOmni2.5-8 BeadChips, (Illumina) and data analysed using Nexus 6.1 Discovery Edition, Biodiscovery. We performed targeted resequencing and genome-wide SNP arrays using selected samples’ germline material to confirm somatic mutations (n=40). Univariate and multivariate analyses using minimal residual disease (MRD) as the outcome measure were performed for 220 of the 250 patients.

Results: Pathogenic mutations were identified in 165 (66%) patients, totalling 268 mutations in 10 genes. ATM was the most frequently mutated gene affecting 67 patients (29%) followed by SF3B1 (n=56, 24%), NOTCH1 (n= 32, 14%), TP53 (n= 21, 9%), BIRC3 (n= 17, 7%) and XPO1 (n=14, 6%). Less frequently recurrent mutations were seen in SAMHD1 (n=8, 3%), MYD88 (n= 4, 2%), MED12 (n=7, 3%) and ZFPM2 (n=5, 2%). Integrating sequencing and array results increased the patients with one or more CLL driver mutation from 66% to 94%. As previously reported del17p and TP53 mutations are co-occurring and associate with MRD positivity in all cases (n=15, p=0.0002). We report on minor TP53 subclones in 11 patients (VAF 1-5%), 8 of whom have MRD data available and were also associated with MRD positivity. Deletions of 11q were present in 44 patients. These lesions always included ATM but not always BIRC3. Bialleleic disruption was present in ATM for 27 patients (significantly associated with MRD positivity) and in BIRC3 for 4 patients. Rather surprisingly, trisomy 12 (n=33) and NOTCH1 mutations (n=28) were associated with MRD negativity (p=0.006 and 0.097, respectively). Analysing clonal and subclonal mutations per gene revealed the majority of mutations in SF3B1 and BIRC3 were subclonal (65% and 87% respectively). In contrast almost all SAMHD1 and MYD88 mutations were clonally distributed. There was an association between NOTCH1 subclonal mutations and MRD negativity, compared to clonal mutations, but this difference was not seen in the remaining mutated genes. From our copy number data, the presence of subclones was associated with MRD positivity (p=0.05). Combining important lesions in a multiple logistic regression analysis to predict MRD positivity, bialleleic ATM disruption, together with TP53 disruption, were the strongest predictors, followed by SAMHD1, whereas BIRC3 monoalleleic mutations were a medium predictor for MRD negativity.

Conclusion: This is the first integrated genome-wide analysis of the distribution and associations of CLL drivers, using targeted deep resequencing and whole genome SNP arrays in an FCR-based first-line treatment setting. We have shown subclonal and clonal mutation profiles in all patients. For patients with two or more CLL-associated mutations we have begun to unravel clonal hierarchies. We have developed a comprehensive model using MRD as an outcome measure and have found bialleleic ATM mutations and SAMHD1 disruption to strongly predict for MRD positivity. Using MRD status as a robust proxy for PFS not only enables us to confirm results of previous studies, but is advantageous also in considerably reducing the timeframe for results. Indeed, we suggest that MRD status should be assessed routinely in future studies to complement modern integrated genomics approaches.